xref: /netbsd-src/sys/kern/sysv_shm.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*	$NetBSD: sysv_shm.c,v 1.82 2004/10/17 11:24:42 jdolecek Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the NetBSD
22  *	Foundation, Inc. and its contributors.
23  * 4. Neither the name of The NetBSD Foundation nor the names of its
24  *    contributors may be used to endorse or promote products derived
25  *    from this software without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37  * POSSIBILITY OF SUCH DAMAGE.
38  */
39 
40 /*
41  * Copyright (c) 1994 Adam Glass and Charles M. Hannum.  All rights reserved.
42  *
43  * Redistribution and use in source and binary forms, with or without
44  * modification, are permitted provided that the following conditions
45  * are met:
46  * 1. Redistributions of source code must retain the above copyright
47  *    notice, this list of conditions and the following disclaimer.
48  * 2. Redistributions in binary form must reproduce the above copyright
49  *    notice, this list of conditions and the following disclaimer in the
50  *    documentation and/or other materials provided with the distribution.
51  * 3. All advertising materials mentioning features or use of this software
52  *    must display the following acknowledgement:
53  *	This product includes software developed by Adam Glass and Charles M.
54  *	Hannum.
55  * 4. The names of the authors may not be used to endorse or promote products
56  *    derived from this software without specific prior written permission.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
59  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
60  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
61  * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
62  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
63  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
64  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
65  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
66  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
67  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
68  */
69 
70 #include <sys/cdefs.h>
71 __KERNEL_RCSID(0, "$NetBSD: sysv_shm.c,v 1.82 2004/10/17 11:24:42 jdolecek Exp $");
72 
73 #define SYSVSHM
74 
75 #include <sys/param.h>
76 #include <sys/kernel.h>
77 #include <sys/shm.h>
78 #include <sys/malloc.h>
79 #include <sys/mman.h>
80 #include <sys/stat.h>
81 #include <sys/sysctl.h>
82 #include <sys/mount.h>		/* XXX for <sys/syscallargs.h> */
83 #include <sys/sa.h>
84 #include <sys/syscallargs.h>
85 #include <sys/queue.h>
86 #include <sys/pool.h>
87 
88 #include <uvm/uvm_extern.h>
89 #include <uvm/uvm_object.h>
90 
91 struct shmid_ds *shm_find_segment_by_shmid(int);
92 
93 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
94 
95 /*
96  * Provides the following externally accessible functions:
97  *
98  * shminit(void);		                 initialization
99  * shmexit(struct vmspace *)                     cleanup
100  * shmfork(struct vmspace *, struct vmspace *)   fork handling
101  *
102  * Structures:
103  * shmsegs (an array of 'struct shmid_ds')
104  * per proc array of 'struct shmmap_state'
105  */
106 
107 #define	SHMSEG_FREE     	0x0200
108 #define	SHMSEG_REMOVED  	0x0400
109 #define	SHMSEG_ALLOCATED	0x0800
110 #define	SHMSEG_WANTED		0x1000
111 #define	SHMSEG_RMLINGER		0x2000
112 
113 static int	shm_last_free, shm_nused, shm_committed;
114 struct	shmid_ds *shmsegs;
115 
116 struct shmmap_entry {
117 	SLIST_ENTRY(shmmap_entry) next;
118 	vaddr_t va;
119 	int shmid;
120 };
121 
122 static POOL_INIT(shmmap_entry_pool, sizeof(struct shmmap_entry), 0, 0, 0,
123     "shmmp", &pool_allocator_nointr);
124 
125 struct shmmap_state {
126 	unsigned int nitems;
127 	unsigned int nrefs;
128 	SLIST_HEAD(, shmmap_entry) entries;
129 };
130 
131 static int shm_find_segment_by_key(key_t);
132 static void shm_deallocate_segment(struct shmid_ds *);
133 static void shm_delete_mapping(struct vmspace *, struct shmmap_state *,
134 			       struct shmmap_entry *);
135 static int shmget_existing(struct proc *, struct sys_shmget_args *,
136 			   int, int, register_t *);
137 static int shmget_allocate_segment(struct proc *, struct sys_shmget_args *,
138 				   int, register_t *);
139 static struct shmmap_state *shmmap_getprivate(struct proc *);
140 static struct shmmap_entry *shm_find_mapping(struct shmmap_state *, vaddr_t);
141 
142 static int
143 shm_find_segment_by_key(key)
144 	key_t key;
145 {
146 	int i;
147 
148 	for (i = 0; i < shminfo.shmmni; i++)
149 		if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
150 		    shmsegs[i].shm_perm._key == key)
151 			return i;
152 	return -1;
153 }
154 
155 struct shmid_ds *
156 shm_find_segment_by_shmid(shmid)
157 	int shmid;
158 {
159 	int segnum;
160 	struct shmid_ds *shmseg;
161 
162 	segnum = IPCID_TO_IX(shmid);
163 	if (segnum < 0 || segnum >= shminfo.shmmni)
164 		return NULL;
165 	shmseg = &shmsegs[segnum];
166 	if ((shmseg->shm_perm.mode & SHMSEG_ALLOCATED) == 0)
167 		return NULL;
168 	if ((shmseg->shm_perm.mode & (SHMSEG_REMOVED|SHMSEG_RMLINGER)) == SHMSEG_REMOVED)
169 		return NULL;
170 	if (shmseg->shm_perm._seq != IPCID_TO_SEQ(shmid))
171 		return NULL;
172 	return shmseg;
173 }
174 
175 static void
176 shm_deallocate_segment(shmseg)
177 	struct shmid_ds *shmseg;
178 {
179 	struct uvm_object *uobj = shmseg->_shm_internal;
180 	size_t size = (shmseg->shm_segsz + PGOFSET) & ~PGOFSET;
181 
182 	(*uobj->pgops->pgo_detach)(uobj);
183 	shmseg->_shm_internal = NULL;
184 	shm_committed -= btoc(size);
185 	shmseg->shm_perm.mode = SHMSEG_FREE;
186 	shm_nused--;
187 }
188 
189 static void
190 shm_delete_mapping(vm, shmmap_s, shmmap_se)
191 	struct vmspace *vm;
192 	struct shmmap_state *shmmap_s;
193 	struct shmmap_entry *shmmap_se;
194 {
195 	struct shmid_ds *shmseg;
196 	int segnum;
197 	size_t size;
198 
199 	segnum = IPCID_TO_IX(shmmap_se->shmid);
200 #ifdef DEBUG
201 	if (segnum < 0 || segnum >= shminfo.shmmni)
202 		panic("shm_delete_mapping: vmspace %p state %p entry %p - "
203 		    "entry segment ID bad (%d)",
204 		    vm, shmmap_s, shmmap_se, segnum);
205 #endif
206 	shmseg = &shmsegs[segnum];
207 	size = (shmseg->shm_segsz + PGOFSET) & ~PGOFSET;
208 	uvm_deallocate(&vm->vm_map, shmmap_se->va, size);
209 	SLIST_REMOVE(&shmmap_s->entries, shmmap_se, shmmap_entry, next);
210 	shmmap_s->nitems--;
211 	pool_put(&shmmap_entry_pool, shmmap_se);
212 	shmseg->shm_dtime = time.tv_sec;
213 	if ((--shmseg->shm_nattch <= 0) &&
214 	    (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
215 		shm_deallocate_segment(shmseg);
216 		shm_last_free = segnum;
217 	}
218 }
219 
220 /*
221  * Get a non-shared shm map for that vmspace.
222  * 3 cases:
223  *   - no shm map present: create a fresh one
224  *   - a shm map with refcount=1, just used by ourselves: fine
225  *   - a shared shm map: copy to a fresh one and adjust refcounts
226  */
227 static struct shmmap_state *
228 shmmap_getprivate(struct proc *p)
229 {
230 	struct shmmap_state *oshmmap_s, *shmmap_s;
231 	struct shmmap_entry *oshmmap_se, *shmmap_se;
232 
233 	oshmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
234 	if (oshmmap_s && oshmmap_s->nrefs == 1)
235 		return (oshmmap_s);
236 
237 	shmmap_s = malloc(sizeof(struct shmmap_state), M_SHM, M_WAITOK);
238 	memset(shmmap_s, 0, sizeof(struct shmmap_state));
239 	shmmap_s->nrefs = 1;
240 	SLIST_INIT(&shmmap_s->entries);
241 	p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
242 
243 	if (!oshmmap_s)
244 		return (shmmap_s);
245 
246 #ifdef SHMDEBUG
247 	printf("shmmap_getprivate: vm %p split (%d entries), was used by %d\n",
248 	       p->p_vmspace, oshmmap_s->nitems, oshmmap_s->nrefs);
249 #endif
250 	SLIST_FOREACH(oshmmap_se, &oshmmap_s->entries, next) {
251 		shmmap_se = pool_get(&shmmap_entry_pool, PR_WAITOK);
252 		shmmap_se->va = oshmmap_se->va;
253 		shmmap_se->shmid = oshmmap_se->shmid;
254 		SLIST_INSERT_HEAD(&shmmap_s->entries, shmmap_se, next);
255 	}
256 	shmmap_s->nitems = oshmmap_s->nitems;
257 	oshmmap_s->nrefs--;
258 	return (shmmap_s);
259 }
260 
261 static struct shmmap_entry *
262 shm_find_mapping(map, va)
263 	struct shmmap_state *map;
264 	vaddr_t va;
265 {
266 	struct shmmap_entry *shmmap_se;
267 
268 	SLIST_FOREACH(shmmap_se, &map->entries, next) {
269 		if (shmmap_se->va == va)
270 			return shmmap_se;
271 	}
272 	return 0;
273 }
274 
275 int
276 sys_shmdt(l, v, retval)
277 	struct lwp *l;
278 	void *v;
279 	register_t *retval;
280 {
281 	struct sys_shmdt_args /* {
282 		syscallarg(const void *) shmaddr;
283 	} */ *uap = v;
284 	struct proc *p = l->l_proc;
285 	struct shmmap_state *shmmap_s, *shmmap_s1;
286 	struct shmmap_entry *shmmap_se;
287 
288 	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
289 	if (shmmap_s == NULL)
290 		return EINVAL;
291 
292 	shmmap_se = shm_find_mapping(shmmap_s, (vaddr_t)SCARG(uap, shmaddr));
293 	if (!shmmap_se)
294 		return EINVAL;
295 
296 	shmmap_s1 = shmmap_getprivate(p);
297 	if (shmmap_s1 != shmmap_s) {
298 		/* map has been copied, lookup entry in new map */
299 		shmmap_se = shm_find_mapping(shmmap_s1,
300 					     (vaddr_t)SCARG(uap, shmaddr));
301 		KASSERT(shmmap_se != NULL);
302 	}
303 #ifdef SHMDEBUG
304 	printf("shmdt: vm %p: remove %d @%lx\n",
305 	       p->p_vmspace, shmmap_se->shmid, shmmap_se->va);
306 #endif
307 	shm_delete_mapping(p->p_vmspace, shmmap_s1, shmmap_se);
308 	return 0;
309 }
310 
311 int
312 sys_shmat(l, v, retval)
313 	struct lwp *l;
314 	void *v;
315 	register_t *retval;
316 {
317 	struct sys_shmat_args /* {
318 		syscallarg(int) shmid;
319 		syscallarg(const void *) shmaddr;
320 		syscallarg(int) shmflg;
321 	} */ *uap = v;
322 	int error, flags;
323 	struct proc *p = l->l_proc;
324 	struct ucred *cred = p->p_ucred;
325 	struct shmid_ds *shmseg;
326 	struct shmmap_state *shmmap_s;
327 	struct uvm_object *uobj;
328 	vaddr_t attach_va;
329 	vm_prot_t prot;
330 	vsize_t size;
331 	struct shmmap_entry *shmmap_se;
332 
333 	shmseg = shm_find_segment_by_shmid(SCARG(uap, shmid));
334 	if (shmseg == NULL)
335 		return EINVAL;
336 	error = ipcperm(cred, &shmseg->shm_perm,
337 		    (SCARG(uap, shmflg) & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
338 	if (error)
339 		return error;
340 
341 	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
342 	if (shmmap_s && shmmap_s->nitems >= shminfo.shmseg)
343 		return EMFILE;
344 
345 	size = (shmseg->shm_segsz + PGOFSET) & ~PGOFSET;
346 	prot = VM_PROT_READ;
347 	if ((SCARG(uap, shmflg) & SHM_RDONLY) == 0)
348 		prot |= VM_PROT_WRITE;
349 	flags = MAP_ANON | MAP_SHARED;
350 	if (SCARG(uap, shmaddr)) {
351 		flags |= MAP_FIXED;
352 		if (SCARG(uap, shmflg) & SHM_RND)
353 			attach_va =
354 			    (vaddr_t)SCARG(uap, shmaddr) & ~(SHMLBA-1);
355 		else if (((vaddr_t)SCARG(uap, shmaddr) & (SHMLBA-1)) == 0)
356 			attach_va = (vaddr_t)SCARG(uap, shmaddr);
357 		else
358 			return EINVAL;
359 	} else {
360 		/* This is just a hint to uvm_mmap() about where to put it. */
361 		attach_va = VM_DEFAULT_ADDRESS(p->p_vmspace->vm_daddr, size);
362 	}
363 	uobj = shmseg->_shm_internal;
364 	(*uobj->pgops->pgo_reference)(uobj);
365 	error = uvm_map(&p->p_vmspace->vm_map, &attach_va, size,
366 	    uobj, 0, 0,
367 	    UVM_MAPFLAG(prot, prot, UVM_INH_SHARE, UVM_ADV_RANDOM, 0));
368 	if (error) {
369 		(*uobj->pgops->pgo_detach)(uobj);
370 		return error;
371 	}
372 	shmmap_se = pool_get(&shmmap_entry_pool, PR_WAITOK);
373 	shmmap_se->va = attach_va;
374 	shmmap_se->shmid = SCARG(uap, shmid);
375 	shmmap_s = shmmap_getprivate(p);
376 #ifdef SHMDEBUG
377 	printf("shmat: vm %p: add %d @%lx\n", p->p_vmspace, shmid, attach_va);
378 #endif
379 	SLIST_INSERT_HEAD(&shmmap_s->entries, shmmap_se, next);
380 	shmmap_s->nitems++;
381 	shmseg->shm_lpid = p->p_pid;
382 	shmseg->shm_atime = time.tv_sec;
383 	shmseg->shm_nattch++;
384 
385 	retval[0] = attach_va;
386 	return 0;
387 }
388 
389 int
390 sys___shmctl13(l, v, retval)
391 	struct lwp *l;
392 	void *v;
393 	register_t *retval;
394 {
395 	struct sys___shmctl13_args /* {
396 		syscallarg(int) shmid;
397 		syscallarg(int) cmd;
398 		syscallarg(struct shmid_ds *) buf;
399 	} */ *uap = v;
400 	struct proc *p = l->l_proc;
401 	struct shmid_ds shmbuf;
402 	int cmd, error;
403 
404 	cmd = SCARG(uap, cmd);
405 
406 	if (cmd == IPC_SET) {
407 		error = copyin(SCARG(uap, buf), &shmbuf, sizeof(shmbuf));
408 		if (error)
409 			return (error);
410 	}
411 
412 	error = shmctl1(p, SCARG(uap, shmid), cmd,
413 	    (cmd == IPC_SET || cmd == IPC_STAT) ? &shmbuf : NULL);
414 
415 	if (error == 0 && cmd == IPC_STAT)
416 		error = copyout(&shmbuf, SCARG(uap, buf), sizeof(shmbuf));
417 
418 	return (error);
419 }
420 
421 int
422 shmctl1(p, shmid, cmd, shmbuf)
423 	struct proc *p;
424 	int shmid;
425 	int cmd;
426 	struct shmid_ds *shmbuf;
427 {
428 	struct ucred *cred = p->p_ucred;
429 	struct shmid_ds *shmseg;
430 	int error = 0;
431 
432 	shmseg = shm_find_segment_by_shmid(shmid);
433 	if (shmseg == NULL)
434 		return EINVAL;
435 	switch (cmd) {
436 	case IPC_STAT:
437 		if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_R)) != 0)
438 			return error;
439 		memcpy(shmbuf, shmseg, sizeof(struct shmid_ds));
440 		break;
441 	case IPC_SET:
442 		if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_M)) != 0)
443 			return error;
444 		shmseg->shm_perm.uid = shmbuf->shm_perm.uid;
445 		shmseg->shm_perm.gid = shmbuf->shm_perm.gid;
446 		shmseg->shm_perm.mode =
447 		    (shmseg->shm_perm.mode & ~ACCESSPERMS) |
448 		    (shmbuf->shm_perm.mode & ACCESSPERMS);
449 		shmseg->shm_ctime = time.tv_sec;
450 		break;
451 	case IPC_RMID:
452 		if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_M)) != 0)
453 			return error;
454 		shmseg->shm_perm._key = IPC_PRIVATE;
455 		shmseg->shm_perm.mode |= SHMSEG_REMOVED;
456 		if (shmseg->shm_nattch <= 0) {
457 			shm_deallocate_segment(shmseg);
458 			shm_last_free = IPCID_TO_IX(shmid);
459 		}
460 		break;
461 	case SHM_LOCK:
462 	case SHM_UNLOCK:
463 	default:
464 		return EINVAL;
465 	}
466 	return 0;
467 }
468 
469 static int
470 shmget_existing(p, uap, mode, segnum, retval)
471 	struct proc *p;
472 	struct sys_shmget_args /* {
473 		syscallarg(key_t) key;
474 		syscallarg(size_t) size;
475 		syscallarg(int) shmflg;
476 	} */ *uap;
477 	int mode;
478 	int segnum;
479 	register_t *retval;
480 {
481 	struct shmid_ds *shmseg;
482 	struct ucred *cred = p->p_ucred;
483 	int error;
484 
485 	shmseg = &shmsegs[segnum];
486 	if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
487 		/*
488 		 * This segment is in the process of being allocated.  Wait
489 		 * until it's done, and look the key up again (in case the
490 		 * allocation failed or it was freed).
491 		 */
492 		shmseg->shm_perm.mode |= SHMSEG_WANTED;
493 		error = tsleep((caddr_t)shmseg, PLOCK | PCATCH, "shmget", 0);
494 		if (error)
495 			return error;
496 		return EAGAIN;
497 	}
498 	if ((error = ipcperm(cred, &shmseg->shm_perm, mode)) != 0)
499 		return error;
500 	if (SCARG(uap, size) && SCARG(uap, size) > shmseg->shm_segsz)
501 		return EINVAL;
502 	if ((SCARG(uap, shmflg) & (IPC_CREAT | IPC_EXCL)) ==
503 	    (IPC_CREAT | IPC_EXCL))
504 		return EEXIST;
505 	*retval = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
506 	return 0;
507 }
508 
509 static int
510 shmget_allocate_segment(p, uap, mode, retval)
511 	struct proc *p;
512 	struct sys_shmget_args /* {
513 		syscallarg(key_t) key;
514 		syscallarg(size_t) size;
515 		syscallarg(int) shmflg;
516 	} */ *uap;
517 	int mode;
518 	register_t *retval;
519 {
520 	int i, segnum, shmid, size;
521 	struct ucred *cred = p->p_ucred;
522 	struct shmid_ds *shmseg;
523 	int error = 0;
524 
525 	if (SCARG(uap, size) < shminfo.shmmin ||
526 	    SCARG(uap, size) > shminfo.shmmax)
527 		return EINVAL;
528 	if (shm_nused >= shminfo.shmmni) /* any shmids left? */
529 		return ENOSPC;
530 	size = (SCARG(uap, size) + PGOFSET) & ~PGOFSET;
531 	if (shm_committed + btoc(size) > shminfo.shmall)
532 		return ENOMEM;
533 	if (shm_last_free < 0) {
534 		for (i = 0; i < shminfo.shmmni; i++)
535 			if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
536 				break;
537 		if (i == shminfo.shmmni)
538 			panic("shmseg free count inconsistent");
539 		segnum = i;
540 	} else  {
541 		segnum = shm_last_free;
542 		shm_last_free = -1;
543 	}
544 	shmseg = &shmsegs[segnum];
545 	/*
546 	 * In case we sleep in malloc(), mark the segment present but deleted
547 	 * so that noone else tries to create the same key.
548 	 */
549 	shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
550 	shmseg->shm_perm._key = SCARG(uap, key);
551 	shmseg->shm_perm._seq = (shmseg->shm_perm._seq + 1) & 0x7fff;
552 	shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
553 
554 	shmseg->_shm_internal = uao_create(size, 0);
555 
556 	shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
557 	shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
558 	shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
559 	    (mode & (ACCESSPERMS|SHMSEG_RMLINGER)) | SHMSEG_ALLOCATED;
560 	shmseg->shm_segsz = SCARG(uap, size);
561 	shmseg->shm_cpid = p->p_pid;
562 	shmseg->shm_lpid = shmseg->shm_nattch = 0;
563 	shmseg->shm_atime = shmseg->shm_dtime = 0;
564 	shmseg->shm_ctime = time.tv_sec;
565 	shm_committed += btoc(size);
566 	shm_nused++;
567 
568 	*retval = shmid;
569 	if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
570 		/*
571 		 * Somebody else wanted this key while we were asleep.  Wake
572 		 * them up now.
573 		 */
574 		shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
575 		wakeup((caddr_t)shmseg);
576 	}
577 	return error;
578 }
579 
580 int
581 sys_shmget(l, v, retval)
582 	struct lwp *l;
583 	void *v;
584 	register_t *retval;
585 {
586 	struct sys_shmget_args /* {
587 		syscallarg(key_t) key;
588 		syscallarg(int) size;
589 		syscallarg(int) shmflg;
590 	} */ *uap = v;
591 	struct proc *p = l->l_proc;
592 	int segnum, mode, error;
593 
594 	mode = SCARG(uap, shmflg) & ACCESSPERMS;
595 	if (SCARG(uap, shmflg) & _SHM_RMLINGER)
596 		mode |= SHMSEG_RMLINGER;
597 
598 	if (SCARG(uap, key) != IPC_PRIVATE) {
599 	again:
600 		segnum = shm_find_segment_by_key(SCARG(uap, key));
601 		if (segnum >= 0) {
602 			error = shmget_existing(p, uap, mode, segnum, retval);
603 			if (error == EAGAIN)
604 				goto again;
605 			return error;
606 		}
607 		if ((SCARG(uap, shmflg) & IPC_CREAT) == 0)
608 			return ENOENT;
609 	}
610 	return shmget_allocate_segment(p, uap, mode, retval);
611 }
612 
613 void
614 shmfork(vm1, vm2)
615 	struct vmspace *vm1, *vm2;
616 {
617 	struct shmmap_state *shmmap_s;
618 	struct shmmap_entry *shmmap_se;
619 
620 	vm2->vm_shm = vm1->vm_shm;
621 
622 	if (vm1->vm_shm == NULL)
623 		return;
624 
625 #ifdef SHMDEBUG
626 	printf("shmfork %p->%p\n", vm1, vm2);
627 #endif
628 
629 	shmmap_s = (struct shmmap_state *)vm1->vm_shm;
630 
631 	SLIST_FOREACH(shmmap_se, &shmmap_s->entries, next)
632 		shmsegs[IPCID_TO_IX(shmmap_se->shmid)].shm_nattch++;
633 	shmmap_s->nrefs++;
634 }
635 
636 void
637 shmexit(vm)
638 	struct vmspace *vm;
639 {
640 	struct shmmap_state *shmmap_s;
641 	struct shmmap_entry *shmmap_se;
642 
643 	shmmap_s = (struct shmmap_state *)vm->vm_shm;
644 	if (shmmap_s == NULL)
645 		return;
646 
647 	vm->vm_shm = NULL;
648 
649 	if (--shmmap_s->nrefs > 0) {
650 #ifdef SHMDEBUG
651 		printf("shmexit: vm %p drop ref (%d entries), now used by %d\n",
652 		       vm, shmmap_s->nitems, shmmap_s->nrefs);
653 #endif
654 		SLIST_FOREACH(shmmap_se, &shmmap_s->entries, next)
655 			shmsegs[IPCID_TO_IX(shmmap_se->shmid)].shm_nattch--;
656 		return;
657 	}
658 
659 #ifdef SHMDEBUG
660 	printf("shmexit: vm %p cleanup (%d entries)\n", vm, shmmap_s->nitems);
661 #endif
662 	while (!SLIST_EMPTY(&shmmap_s->entries)) {
663 		shmmap_se = SLIST_FIRST(&shmmap_s->entries);
664 		shm_delete_mapping(vm, shmmap_s, shmmap_se);
665 	}
666 	KASSERT(shmmap_s->nitems == 0);
667 	free(shmmap_s, M_SHM);
668 }
669 
670 void
671 shminit()
672 {
673 	int i, sz;
674 	vaddr_t v;
675 
676 	/* Allocate pageable memory for our structures */
677 	sz = shminfo.shmmni * sizeof(struct shmid_ds);
678 	if ((v = uvm_km_alloc(kernel_map, round_page(sz))) == 0)
679 		panic("sysv_shm: cannot allocate memory");
680 	shmsegs = (void *)v;
681 
682 	shminfo.shmmax *= PAGE_SIZE;
683 
684 	for (i = 0; i < shminfo.shmmni; i++) {
685 		shmsegs[i].shm_perm.mode = SHMSEG_FREE;
686 		shmsegs[i].shm_perm._seq = 0;
687 	}
688 	shm_last_free = 0;
689 	shm_nused = 0;
690 	shm_committed = 0;
691 }
692