xref: /freebsd/sys/kern/sysv_shm.c (revision 830940567b49bb0c08dfaed40418999e76616909)
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/resourcevar.h>
81 #include <sys/stat.h>
82 #include <sys/syscall.h>
83 #include <sys/syscallsubr.h>
84 #include <sys/sysent.h>
85 #include <sys/sysproto.h>
86 #include <sys/jail.h>
87 
88 #include <security/mac/mac_framework.h>
89 
90 #include <vm/vm.h>
91 #include <vm/vm_param.h>
92 #include <vm/pmap.h>
93 #include <vm/vm_object.h>
94 #include <vm/vm_map.h>
95 #include <vm/vm_page.h>
96 #include <vm/vm_pager.h>
97 
98 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
99 
100 static int shmget_allocate_segment(struct thread *td,
101     struct shmget_args *uap, int mode);
102 static int shmget_existing(struct thread *td, struct shmget_args *uap,
103     int mode, int segnum);
104 
105 #define	SHMSEG_FREE     	0x0200
106 #define	SHMSEG_REMOVED  	0x0400
107 #define	SHMSEG_ALLOCATED	0x0800
108 #define	SHMSEG_WANTED		0x1000
109 
110 static int shm_last_free, shm_nused, shmalloced;
111 vm_size_t shm_committed;
112 static struct shmid_kernel	*shmsegs;
113 
114 struct shmmap_state {
115 	vm_offset_t va;
116 	int shmid;
117 };
118 
119 static void shm_deallocate_segment(struct shmid_kernel *);
120 static int shm_find_segment_by_key(key_t);
121 static struct shmid_kernel *shm_find_segment_by_shmid(int);
122 static struct shmid_kernel *shm_find_segment_by_shmidx(int);
123 static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *);
124 static void shmrealloc(void);
125 static void shminit(void);
126 static int sysvshm_modload(struct module *, int, void *);
127 static int shmunload(void);
128 static void shmexit_myhook(struct vmspace *vm);
129 static void shmfork_myhook(struct proc *p1, struct proc *p2);
130 static int sysctl_shmsegs(SYSCTL_HANDLER_ARGS);
131 
132 /*
133  * Tuneable values.
134  */
135 #ifndef SHMMAXPGS
136 #define	SHMMAXPGS	8192	/* Note: sysv shared memory is swap backed. */
137 #endif
138 #ifndef SHMMAX
139 #define	SHMMAX	(SHMMAXPGS*PAGE_SIZE)
140 #endif
141 #ifndef SHMMIN
142 #define	SHMMIN	1
143 #endif
144 #ifndef SHMMNI
145 #define	SHMMNI	192
146 #endif
147 #ifndef SHMSEG
148 #define	SHMSEG	128
149 #endif
150 #ifndef SHMALL
151 #define	SHMALL	(SHMMAXPGS)
152 #endif
153 
154 struct	shminfo shminfo = {
155 	SHMMAX,
156 	SHMMIN,
157 	SHMMNI,
158 	SHMSEG,
159 	SHMALL
160 };
161 
162 static int shm_use_phys;
163 static int shm_allow_removed;
164 
165 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0,
166     "Maximum shared memory segment size");
167 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0,
168     "Minimum shared memory segment size");
169 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RDTUN, &shminfo.shmmni, 0,
170     "Number of shared memory identifiers");
171 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RDTUN, &shminfo.shmseg, 0,
172     "Number of segments per process");
173 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0,
174     "Maximum number of pages available for shared memory");
175 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW,
176     &shm_use_phys, 0, "Enable/Disable locking of shared memory pages in core");
177 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RW,
178     &shm_allow_removed, 0,
179     "Enable/Disable attachment to attached segments marked for removal");
180 SYSCTL_PROC(_kern_ipc, OID_AUTO, shmsegs, CTLFLAG_RD,
181     NULL, 0, sysctl_shmsegs, "",
182     "Current number of shared memory segments allocated");
183 
184 static int
185 shm_find_segment_by_key(key)
186 	key_t key;
187 {
188 	int i;
189 
190 	for (i = 0; i < shmalloced; i++)
191 		if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) &&
192 		    shmsegs[i].u.shm_perm.key == key)
193 			return (i);
194 	return (-1);
195 }
196 
197 static struct shmid_kernel *
198 shm_find_segment_by_shmid(int shmid)
199 {
200 	int segnum;
201 	struct shmid_kernel *shmseg;
202 
203 	segnum = IPCID_TO_IX(shmid);
204 	if (segnum < 0 || segnum >= shmalloced)
205 		return (NULL);
206 	shmseg = &shmsegs[segnum];
207 	if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
208 	    (!shm_allow_removed &&
209 	     (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0) ||
210 	    shmseg->u.shm_perm.seq != IPCID_TO_SEQ(shmid))
211 		return (NULL);
212 	return (shmseg);
213 }
214 
215 static struct shmid_kernel *
216 shm_find_segment_by_shmidx(int segnum)
217 {
218 	struct shmid_kernel *shmseg;
219 
220 	if (segnum < 0 || segnum >= shmalloced)
221 		return (NULL);
222 	shmseg = &shmsegs[segnum];
223 	if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
224 	    (!shm_allow_removed &&
225 	     (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0))
226 		return (NULL);
227 	return (shmseg);
228 }
229 
230 static void
231 shm_deallocate_segment(shmseg)
232 	struct shmid_kernel *shmseg;
233 {
234 	vm_size_t size;
235 
236 	GIANT_REQUIRED;
237 
238 	vm_object_deallocate(shmseg->object);
239 	shmseg->object = NULL;
240 	size = round_page(shmseg->u.shm_segsz);
241 	shm_committed -= btoc(size);
242 	shm_nused--;
243 	shmseg->u.shm_perm.mode = SHMSEG_FREE;
244 #ifdef MAC
245 	mac_sysvshm_cleanup(shmseg);
246 #endif
247 }
248 
249 static int
250 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
251 {
252 	struct shmid_kernel *shmseg;
253 	int segnum, result;
254 	vm_size_t size;
255 
256 	GIANT_REQUIRED;
257 
258 	segnum = IPCID_TO_IX(shmmap_s->shmid);
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 #ifndef _SYS_SYSPROTO_H_
275 struct shmdt_args {
276 	const void *shmaddr;
277 };
278 #endif
279 int
280 shmdt(td, uap)
281 	struct thread *td;
282 	struct shmdt_args *uap;
283 {
284 	struct proc *p = td->td_proc;
285 	struct shmmap_state *shmmap_s;
286 #ifdef MAC
287 	struct shmid_kernel *shmsegptr;
288 #endif
289 	int i;
290 	int error = 0;
291 
292 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
293 		return (ENOSYS);
294 	mtx_lock(&Giant);
295 	shmmap_s = p->p_vmspace->vm_shm;
296  	if (shmmap_s == NULL) {
297 		error = EINVAL;
298 		goto done2;
299 	}
300 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
301 		if (shmmap_s->shmid != -1 &&
302 		    shmmap_s->va == (vm_offset_t)uap->shmaddr) {
303 			break;
304 		}
305 	}
306 	if (i == shminfo.shmseg) {
307 		error = EINVAL;
308 		goto done2;
309 	}
310 #ifdef MAC
311 	shmsegptr = &shmsegs[IPCID_TO_IX(shmmap_s->shmid)];
312 	error = mac_sysvshm_check_shmdt(td->td_ucred, shmsegptr);
313 	if (error != 0)
314 		goto done2;
315 #endif
316 	error = shm_delete_mapping(p->p_vmspace, shmmap_s);
317 done2:
318 	mtx_unlock(&Giant);
319 	return (error);
320 }
321 
322 #ifndef _SYS_SYSPROTO_H_
323 struct shmat_args {
324 	int shmid;
325 	const void *shmaddr;
326 	int shmflg;
327 };
328 #endif
329 int
330 kern_shmat(td, shmid, shmaddr, shmflg)
331 	struct thread *td;
332 	int shmid;
333 	const void *shmaddr;
334 	int shmflg;
335 {
336 	struct proc *p = td->td_proc;
337 	int i, flags;
338 	struct shmid_kernel *shmseg;
339 	struct shmmap_state *shmmap_s = NULL;
340 	vm_offset_t attach_va;
341 	vm_prot_t prot;
342 	vm_size_t size;
343 	int rv;
344 	int error = 0;
345 
346 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
347 		return (ENOSYS);
348 	mtx_lock(&Giant);
349 	shmmap_s = p->p_vmspace->vm_shm;
350 	if (shmmap_s == NULL) {
351 		shmmap_s = malloc(shminfo.shmseg * sizeof(struct shmmap_state),
352 		    M_SHM, M_WAITOK);
353 		for (i = 0; i < shminfo.shmseg; i++)
354 			shmmap_s[i].shmid = -1;
355 		p->p_vmspace->vm_shm = shmmap_s;
356 	}
357 	shmseg = shm_find_segment_by_shmid(shmid);
358 	if (shmseg == NULL) {
359 		error = EINVAL;
360 		goto done2;
361 	}
362 	error = ipcperm(td, &shmseg->u.shm_perm,
363 	    (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
364 	if (error)
365 		goto done2;
366 #ifdef MAC
367 	error = mac_sysvshm_check_shmat(td->td_ucred, shmseg, shmflg);
368 	if (error != 0)
369 		goto done2;
370 #endif
371 	for (i = 0; i < shminfo.shmseg; i++) {
372 		if (shmmap_s->shmid == -1)
373 			break;
374 		shmmap_s++;
375 	}
376 	if (i >= shminfo.shmseg) {
377 		error = EMFILE;
378 		goto done2;
379 	}
380 	size = round_page(shmseg->u.shm_segsz);
381 	prot = VM_PROT_READ;
382 	if ((shmflg & SHM_RDONLY) == 0)
383 		prot |= VM_PROT_WRITE;
384 	flags = MAP_ANON | MAP_SHARED;
385 	if (shmaddr) {
386 		flags |= MAP_FIXED;
387 		if (shmflg & SHM_RND) {
388 			attach_va = (vm_offset_t)shmaddr & ~(SHMLBA-1);
389 		} else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0) {
390 			attach_va = (vm_offset_t)shmaddr;
391 		} else {
392 			error = EINVAL;
393 			goto done2;
394 		}
395 	} else {
396 		/*
397 		 * This is just a hint to vm_map_find() about where to
398 		 * put it.
399 		 */
400 		PROC_LOCK(p);
401 		attach_va = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
402 		    lim_max(p, RLIMIT_DATA));
403 		PROC_UNLOCK(p);
404 	}
405 
406 	vm_object_reference(shmseg->object);
407 	rv = vm_map_find(&p->p_vmspace->vm_map, shmseg->object,
408 	    0, &attach_va, size, (flags & MAP_FIXED) ? VMFS_NO_SPACE :
409 	    VMFS_ANY_SPACE, prot, prot, 0);
410 	if (rv != KERN_SUCCESS) {
411 		vm_object_deallocate(shmseg->object);
412 		error = ENOMEM;
413 		goto done2;
414 	}
415 	vm_map_inherit(&p->p_vmspace->vm_map,
416 		attach_va, attach_va + size, VM_INHERIT_SHARE);
417 
418 	shmmap_s->va = attach_va;
419 	shmmap_s->shmid = shmid;
420 	shmseg->u.shm_lpid = p->p_pid;
421 	shmseg->u.shm_atime = time_second;
422 	shmseg->u.shm_nattch++;
423 	td->td_retval[0] = attach_va;
424 done2:
425 	mtx_unlock(&Giant);
426 	return (error);
427 }
428 
429 int
430 shmat(td, uap)
431 	struct thread *td;
432 	struct shmat_args *uap;
433 {
434 	return kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg);
435 }
436 
437 int
438 kern_shmctl(td, shmid, cmd, buf, bufsz)
439 	struct thread *td;
440 	int shmid;
441 	int cmd;
442 	void *buf;
443 	size_t *bufsz;
444 {
445 	int error = 0;
446 	struct shmid_kernel *shmseg;
447 
448 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
449 		return (ENOSYS);
450 
451 	mtx_lock(&Giant);
452 	switch (cmd) {
453 	/*
454 	 * It is possible that kern_shmctl is being called from the Linux ABI
455 	 * layer, in which case, we will need to implement IPC_INFO.  It should
456 	 * be noted that other shmctl calls will be funneled through here for
457 	 * Linix binaries as well.
458 	 *
459 	 * NB: The Linux ABI layer will convert this data to structure(s) more
460 	 * consistent with the Linux ABI.
461 	 */
462 	case IPC_INFO:
463 		memcpy(buf, &shminfo, sizeof(shminfo));
464 		if (bufsz)
465 			*bufsz = sizeof(shminfo);
466 		td->td_retval[0] = shmalloced;
467 		goto done2;
468 	case SHM_INFO: {
469 		struct shm_info shm_info;
470 		shm_info.used_ids = shm_nused;
471 		shm_info.shm_rss = 0;	/*XXX where to get from ? */
472 		shm_info.shm_tot = 0;	/*XXX where to get from ? */
473 		shm_info.shm_swp = 0;	/*XXX where to get from ? */
474 		shm_info.swap_attempts = 0;	/*XXX where to get from ? */
475 		shm_info.swap_successes = 0;	/*XXX where to get from ? */
476 		memcpy(buf, &shm_info, sizeof(shm_info));
477 		if (bufsz)
478 			*bufsz = sizeof(shm_info);
479 		td->td_retval[0] = shmalloced;
480 		goto done2;
481 	}
482 	}
483 	if (cmd == SHM_STAT)
484 		shmseg = shm_find_segment_by_shmidx(shmid);
485 	else
486 		shmseg = shm_find_segment_by_shmid(shmid);
487 	if (shmseg == NULL) {
488 		error = EINVAL;
489 		goto done2;
490 	}
491 #ifdef MAC
492 	error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, cmd);
493 	if (error != 0)
494 		goto done2;
495 #endif
496 	switch (cmd) {
497 	case SHM_STAT:
498 	case IPC_STAT:
499 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
500 		if (error)
501 			goto done2;
502 		memcpy(buf, &shmseg->u, sizeof(struct shmid_ds));
503 		if (bufsz)
504 			*bufsz = sizeof(struct shmid_ds);
505 		if (cmd == SHM_STAT)
506 			td->td_retval[0] = IXSEQ_TO_IPCID(shmid, shmseg->u.shm_perm);
507 		break;
508 	case IPC_SET: {
509 		struct shmid_ds *shmid;
510 
511 		shmid = (struct shmid_ds *)buf;
512 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
513 		if (error)
514 			goto done2;
515 		shmseg->u.shm_perm.uid = shmid->shm_perm.uid;
516 		shmseg->u.shm_perm.gid = shmid->shm_perm.gid;
517 		shmseg->u.shm_perm.mode =
518 		    (shmseg->u.shm_perm.mode & ~ACCESSPERMS) |
519 		    (shmid->shm_perm.mode & ACCESSPERMS);
520 		shmseg->u.shm_ctime = time_second;
521 		break;
522 	}
523 	case IPC_RMID:
524 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
525 		if (error)
526 			goto done2;
527 		shmseg->u.shm_perm.key = IPC_PRIVATE;
528 		shmseg->u.shm_perm.mode |= SHMSEG_REMOVED;
529 		if (shmseg->u.shm_nattch <= 0) {
530 			shm_deallocate_segment(shmseg);
531 			shm_last_free = IPCID_TO_IX(shmid);
532 		}
533 		break;
534 #if 0
535 	case SHM_LOCK:
536 	case SHM_UNLOCK:
537 #endif
538 	default:
539 		error = EINVAL;
540 		break;
541 	}
542 done2:
543 	mtx_unlock(&Giant);
544 	return (error);
545 }
546 
547 #ifndef _SYS_SYSPROTO_H_
548 struct shmctl_args {
549 	int shmid;
550 	int cmd;
551 	struct shmid_ds *buf;
552 };
553 #endif
554 int
555 shmctl(td, uap)
556 	struct thread *td;
557 	struct shmctl_args *uap;
558 {
559 	int error = 0;
560 	struct shmid_ds buf;
561 	size_t bufsz;
562 
563 	/*
564 	 * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
565 	 * Linux binaries.  If we see the call come through the FreeBSD ABI,
566 	 * return an error back to the user since we do not to support this.
567 	 */
568 	if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
569 	    uap->cmd == SHM_STAT)
570 		return (EINVAL);
571 
572 	/* IPC_SET needs to copyin the buffer before calling kern_shmctl */
573 	if (uap->cmd == IPC_SET) {
574 		if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds))))
575 			goto done;
576 	}
577 
578 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
579 	if (error)
580 		goto done;
581 
582 	/* Cases in which we need to copyout */
583 	switch (uap->cmd) {
584 	case IPC_STAT:
585 		error = copyout(&buf, uap->buf, bufsz);
586 		break;
587 	}
588 
589 done:
590 	if (error) {
591 		/* Invalidate the return value */
592 		td->td_retval[0] = -1;
593 	}
594 	return (error);
595 }
596 
597 
598 static int
599 shmget_existing(td, uap, mode, segnum)
600 	struct thread *td;
601 	struct shmget_args *uap;
602 	int mode;
603 	int segnum;
604 {
605 	struct shmid_kernel *shmseg;
606 	int error;
607 
608 	shmseg = &shmsegs[segnum];
609 	if (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) {
610 		/*
611 		 * This segment is in the process of being allocated.  Wait
612 		 * until it's done, and look the key up again (in case the
613 		 * allocation failed or it was freed).
614 		 */
615 		shmseg->u.shm_perm.mode |= SHMSEG_WANTED;
616 		error = tsleep(shmseg, PLOCK | PCATCH, "shmget", 0);
617 		if (error)
618 			return (error);
619 		return (EAGAIN);
620 	}
621 	if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
622 		return (EEXIST);
623 #ifdef MAC
624 	error = mac_sysvshm_check_shmget(td->td_ucred, shmseg, uap->shmflg);
625 	if (error != 0)
626 		return (error);
627 #endif
628 	if (uap->size != 0 && uap->size > shmseg->u.shm_segsz)
629 		return (EINVAL);
630 	td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
631 	return (0);
632 }
633 
634 static int
635 shmget_allocate_segment(td, uap, mode)
636 	struct thread *td;
637 	struct shmget_args *uap;
638 	int mode;
639 {
640 	int i, segnum, shmid;
641 	size_t size;
642 	struct ucred *cred = td->td_ucred;
643 	struct shmid_kernel *shmseg;
644 	vm_object_t shm_object;
645 
646 	GIANT_REQUIRED;
647 
648 	if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
649 		return (EINVAL);
650 	if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
651 		return (ENOSPC);
652 	size = round_page(uap->size);
653 	if (shm_committed + btoc(size) > shminfo.shmall)
654 		return (ENOMEM);
655 	if (shm_last_free < 0) {
656 		shmrealloc();	/* Maybe expand the shmsegs[] array. */
657 		for (i = 0; i < shmalloced; i++)
658 			if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE)
659 				break;
660 		if (i == shmalloced)
661 			return (ENOSPC);
662 		segnum = i;
663 	} else  {
664 		segnum = shm_last_free;
665 		shm_last_free = -1;
666 	}
667 	shmseg = &shmsegs[segnum];
668 	/*
669 	 * In case we sleep in malloc(), mark the segment present but deleted
670 	 * so that noone else tries to create the same key.
671 	 */
672 	shmseg->u.shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
673 	shmseg->u.shm_perm.key = uap->key;
674 	shmseg->u.shm_perm.seq = (shmseg->u.shm_perm.seq + 1) & 0x7fff;
675 	shmid = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
676 
677 	/*
678 	 * We make sure that we have allocated a pager before we need
679 	 * to.
680 	 */
681 	shm_object = vm_pager_allocate(shm_use_phys ? OBJT_PHYS : OBJT_SWAP,
682 	    0, size, VM_PROT_DEFAULT, 0, cred);
683 	if (shm_object == NULL)
684 		return (ENOMEM);
685 	VM_OBJECT_LOCK(shm_object);
686 	vm_object_clear_flag(shm_object, OBJ_ONEMAPPING);
687 	vm_object_set_flag(shm_object, OBJ_NOSPLIT);
688 	VM_OBJECT_UNLOCK(shm_object);
689 
690 	shmseg->object = shm_object;
691 	shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = cred->cr_uid;
692 	shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid;
693 	shmseg->u.shm_perm.mode = (shmseg->u.shm_perm.mode & SHMSEG_WANTED) |
694 	    (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
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 	if (shmseg->u.shm_perm.mode & SHMSEG_WANTED) {
706 		/*
707 		 * Somebody else wanted this key while we were asleep.  Wake
708 		 * them up now.
709 		 */
710 		shmseg->u.shm_perm.mode &= ~SHMSEG_WANTED;
711 		wakeup(shmseg);
712 	}
713 	td->td_retval[0] = shmid;
714 	return (0);
715 }
716 
717 #ifndef _SYS_SYSPROTO_H_
718 struct shmget_args {
719 	key_t key;
720 	size_t size;
721 	int shmflg;
722 };
723 #endif
724 int
725 shmget(td, uap)
726 	struct thread *td;
727 	struct shmget_args *uap;
728 {
729 	int segnum, mode;
730 	int error;
731 
732 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
733 		return (ENOSYS);
734 	mtx_lock(&Giant);
735 	mode = uap->shmflg & ACCESSPERMS;
736 	if (uap->key != IPC_PRIVATE) {
737 	again:
738 		segnum = shm_find_segment_by_key(uap->key);
739 		if (segnum >= 0) {
740 			error = shmget_existing(td, uap, mode, segnum);
741 			if (error == EAGAIN)
742 				goto again;
743 			goto done2;
744 		}
745 		if ((uap->shmflg & IPC_CREAT) == 0) {
746 			error = ENOENT;
747 			goto done2;
748 		}
749 	}
750 	error = shmget_allocate_segment(td, uap, mode);
751 done2:
752 	mtx_unlock(&Giant);
753 	return (error);
754 }
755 
756 static void
757 shmfork_myhook(p1, p2)
758 	struct proc *p1, *p2;
759 {
760 	struct shmmap_state *shmmap_s;
761 	size_t size;
762 	int i;
763 
764 	mtx_lock(&Giant);
765 	size = shminfo.shmseg * sizeof(struct shmmap_state);
766 	shmmap_s = malloc(size, M_SHM, M_WAITOK);
767 	bcopy(p1->p_vmspace->vm_shm, shmmap_s, size);
768 	p2->p_vmspace->vm_shm = shmmap_s;
769 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
770 		if (shmmap_s->shmid != -1)
771 			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++;
772 	mtx_unlock(&Giant);
773 }
774 
775 static void
776 shmexit_myhook(struct vmspace *vm)
777 {
778 	struct shmmap_state *base, *shm;
779 	int i;
780 
781 	if ((base = vm->vm_shm) != NULL) {
782 		vm->vm_shm = NULL;
783 		mtx_lock(&Giant);
784 		for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
785 			if (shm->shmid != -1)
786 				shm_delete_mapping(vm, shm);
787 		}
788 		mtx_unlock(&Giant);
789 		free(base, M_SHM);
790 	}
791 }
792 
793 static void
794 shmrealloc(void)
795 {
796 	int i;
797 	struct shmid_kernel *newsegs;
798 
799 	if (shmalloced >= shminfo.shmmni)
800 		return;
801 
802 	newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
803 	if (newsegs == NULL)
804 		return;
805 	for (i = 0; i < shmalloced; i++)
806 		bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
807 	for (; i < shminfo.shmmni; i++) {
808 		shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
809 		shmsegs[i].u.shm_perm.seq = 0;
810 #ifdef MAC
811 		mac_sysvshm_init(&shmsegs[i]);
812 #endif
813 	}
814 	free(shmsegs, M_SHM);
815 	shmsegs = newsegs;
816 	shmalloced = shminfo.shmmni;
817 }
818 
819 static void
820 shminit()
821 {
822 	int i;
823 
824 	TUNABLE_ULONG_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall);
825 	for (i = PAGE_SIZE; i > 0; i--) {
826 		shminfo.shmmax = shminfo.shmall * i;
827 		if (shminfo.shmmax >= shminfo.shmall)
828 			break;
829 	}
830 	TUNABLE_ULONG_FETCH("kern.ipc.shmmin", &shminfo.shmmin);
831 	TUNABLE_ULONG_FETCH("kern.ipc.shmmni", &shminfo.shmmni);
832 	TUNABLE_ULONG_FETCH("kern.ipc.shmseg", &shminfo.shmseg);
833 	TUNABLE_INT_FETCH("kern.ipc.shm_use_phys", &shm_use_phys);
834 
835 	shmalloced = shminfo.shmmni;
836 	shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
837 	if (shmsegs == NULL)
838 		panic("cannot allocate initial memory for sysvshm");
839 	for (i = 0; i < shmalloced; i++) {
840 		shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
841 		shmsegs[i].u.shm_perm.seq = 0;
842 #ifdef MAC
843 		mac_sysvshm_init(&shmsegs[i]);
844 #endif
845 	}
846 	shm_last_free = 0;
847 	shm_nused = 0;
848 	shm_committed = 0;
849 	shmexit_hook = &shmexit_myhook;
850 	shmfork_hook = &shmfork_myhook;
851 }
852 
853 static int
854 shmunload()
855 {
856 #ifdef MAC
857 	int i;
858 #endif
859 
860 	if (shm_nused > 0)
861 		return (EBUSY);
862 
863 #ifdef MAC
864 	for (i = 0; i < shmalloced; i++)
865 		mac_sysvshm_destroy(&shmsegs[i]);
866 #endif
867 	free(shmsegs, M_SHM);
868 	shmexit_hook = NULL;
869 	shmfork_hook = NULL;
870 	return (0);
871 }
872 
873 static int
874 sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
875 {
876 
877 	return (SYSCTL_OUT(req, shmsegs, shmalloced * sizeof(shmsegs[0])));
878 }
879 
880 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
881 struct oshmid_ds {
882 	struct	ipc_perm_old shm_perm;	/* operation perms */
883 	int	shm_segsz;		/* size of segment (bytes) */
884 	u_short	shm_cpid;		/* pid, creator */
885 	u_short	shm_lpid;		/* pid, last operation */
886 	short	shm_nattch;		/* no. of current attaches */
887 	time_t	shm_atime;		/* last attach time */
888 	time_t	shm_dtime;		/* last detach time */
889 	time_t	shm_ctime;		/* last change time */
890 	void	*shm_handle;		/* internal handle for shm segment */
891 };
892 
893 struct oshmctl_args {
894 	int shmid;
895 	int cmd;
896 	struct oshmid_ds *ubuf;
897 };
898 
899 static int
900 oshmctl(struct thread *td, struct oshmctl_args *uap)
901 {
902 #ifdef COMPAT_43
903 	int error = 0;
904 	struct shmid_kernel *shmseg;
905 	struct oshmid_ds outbuf;
906 
907 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
908 		return (ENOSYS);
909 	mtx_lock(&Giant);
910 	shmseg = shm_find_segment_by_shmid(uap->shmid);
911 	if (shmseg == NULL) {
912 		error = EINVAL;
913 		goto done2;
914 	}
915 	switch (uap->cmd) {
916 	case IPC_STAT:
917 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
918 		if (error)
919 			goto done2;
920 #ifdef MAC
921 		error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, uap->cmd);
922 		if (error != 0)
923 			goto done2;
924 #endif
925 		ipcperm_new2old(&shmseg->u.shm_perm, &outbuf.shm_perm);
926 		outbuf.shm_segsz = shmseg->u.shm_segsz;
927 		outbuf.shm_cpid = shmseg->u.shm_cpid;
928 		outbuf.shm_lpid = shmseg->u.shm_lpid;
929 		outbuf.shm_nattch = shmseg->u.shm_nattch;
930 		outbuf.shm_atime = shmseg->u.shm_atime;
931 		outbuf.shm_dtime = shmseg->u.shm_dtime;
932 		outbuf.shm_ctime = shmseg->u.shm_ctime;
933 		outbuf.shm_handle = shmseg->object;
934 		error = copyout(&outbuf, uap->ubuf, sizeof(outbuf));
935 		if (error)
936 			goto done2;
937 		break;
938 	default:
939 		error = freebsd7_shmctl(td, (struct freebsd7_shmctl_args *)uap);
940 		break;
941 	}
942 done2:
943 	mtx_unlock(&Giant);
944 	return (error);
945 #else
946 	return (EINVAL);
947 #endif
948 }
949 
950 /* XXX casting to (sy_call_t *) is bogus, as usual. */
951 static sy_call_t *shmcalls[] = {
952 	(sy_call_t *)shmat, (sy_call_t *)oshmctl,
953 	(sy_call_t *)shmdt, (sy_call_t *)shmget,
954 	(sy_call_t *)freebsd7_shmctl
955 };
956 
957 int
958 shmsys(td, uap)
959 	struct thread *td;
960 	/* XXX actually varargs. */
961 	struct shmsys_args /* {
962 		int	which;
963 		int	a2;
964 		int	a3;
965 		int	a4;
966 	} */ *uap;
967 {
968 	int error;
969 
970 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
971 		return (ENOSYS);
972 	if (uap->which < 0 ||
973 	    uap->which >= sizeof(shmcalls)/sizeof(shmcalls[0]))
974 		return (EINVAL);
975 	mtx_lock(&Giant);
976 	error = (*shmcalls[uap->which])(td, &uap->a2);
977 	mtx_unlock(&Giant);
978 	return (error);
979 }
980 
981 SYSCALL_MODULE_HELPER(shmsys);
982 #endif	/* i386 && (COMPAT_FREEBSD4 || COMPAT_43) */
983 
984 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
985     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
986 
987 #define CP(src, dst, fld)	do { (dst).fld = (src).fld; } while (0)
988 
989 
990 #ifndef _SYS_SYSPROTO_H_
991 struct freebsd7_shmctl_args {
992 	int shmid;
993 	int cmd;
994 	struct shmid_ds_old *buf;
995 };
996 #endif
997 int
998 freebsd7_shmctl(td, uap)
999 	struct thread *td;
1000 	struct freebsd7_shmctl_args *uap;
1001 {
1002 	int error = 0;
1003 	struct shmid_ds_old old;
1004 	struct shmid_ds buf;
1005 	size_t bufsz;
1006 
1007 	/*
1008 	 * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
1009 	 * Linux binaries.  If we see the call come through the FreeBSD ABI,
1010 	 * return an error back to the user since we do not to support this.
1011 	 */
1012 	if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
1013 	    uap->cmd == SHM_STAT)
1014 		return (EINVAL);
1015 
1016 	/* IPC_SET needs to copyin the buffer before calling kern_shmctl */
1017 	if (uap->cmd == IPC_SET) {
1018 		if ((error = copyin(uap->buf, &old, sizeof(old))))
1019 			goto done;
1020 		ipcperm_old2new(&old.shm_perm, &buf.shm_perm);
1021 		CP(old, buf, shm_segsz);
1022 		CP(old, buf, shm_lpid);
1023 		CP(old, buf, shm_cpid);
1024 		CP(old, buf, shm_nattch);
1025 		CP(old, buf, shm_atime);
1026 		CP(old, buf, shm_dtime);
1027 		CP(old, buf, shm_ctime);
1028 	}
1029 
1030 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
1031 	if (error)
1032 		goto done;
1033 
1034 	/* Cases in which we need to copyout */
1035 	switch (uap->cmd) {
1036 	case IPC_STAT:
1037 		ipcperm_new2old(&buf.shm_perm, &old.shm_perm);
1038 		if (buf.shm_segsz > INT_MAX)
1039 			old.shm_segsz = INT_MAX;
1040 		else
1041 			CP(buf, old, shm_segsz);
1042 		CP(buf, old, shm_lpid);
1043 		CP(buf, old, shm_cpid);
1044 		if (buf.shm_nattch > SHRT_MAX)
1045 			old.shm_nattch = SHRT_MAX;
1046 		else
1047 			CP(buf, old, shm_nattch);
1048 		CP(buf, old, shm_atime);
1049 		CP(buf, old, shm_dtime);
1050 		CP(buf, old, shm_ctime);
1051 		old.shm_internal = NULL;
1052 		error = copyout(&old, uap->buf, sizeof(old));
1053 		break;
1054 	}
1055 
1056 done:
1057 	if (error) {
1058 		/* Invalidate the return value */
1059 		td->td_retval[0] = -1;
1060 	}
1061 	return (error);
1062 }
1063 
1064 SYSCALL_MODULE_HELPER(freebsd7_shmctl);
1065 
1066 #undef CP
1067 
1068 #endif	/* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 ||
1069 	   COMPAT_FREEBSD7 */
1070 
1071 static int
1072 sysvshm_modload(struct module *module, int cmd, void *arg)
1073 {
1074 	int error = 0;
1075 
1076 	switch (cmd) {
1077 	case MOD_LOAD:
1078 		shminit();
1079 		break;
1080 	case MOD_UNLOAD:
1081 		error = shmunload();
1082 		break;
1083 	case MOD_SHUTDOWN:
1084 		break;
1085 	default:
1086 		error = EINVAL;
1087 		break;
1088 	}
1089 	return (error);
1090 }
1091 
1092 static moduledata_t sysvshm_mod = {
1093 	"sysvshm",
1094 	&sysvshm_modload,
1095 	NULL
1096 };
1097 
1098 SYSCALL_MODULE_HELPER(shmat);
1099 SYSCALL_MODULE_HELPER(shmctl);
1100 SYSCALL_MODULE_HELPER(shmdt);
1101 SYSCALL_MODULE_HELPER(shmget);
1102 
1103 DECLARE_MODULE(sysvshm, sysvshm_mod, SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
1104 MODULE_VERSION(sysvshm, 1);
1105