xref: /dragonfly/sys/kern/sysv_shm.c (revision 768af85b)
1 /* $FreeBSD: src/sys/kern/sysv_shm.c,v 1.45.2.6 2002/10/22 20:45:03 fjoe Exp $ */
2 /* $DragonFly: src/sys/kern/sysv_shm.c,v 1.21 2008/01/06 16:55:51 swildner Exp $ */
3 /*	$NetBSD: sysv_shm.c,v 1.23 1994/07/04 23:25:12 glass Exp $	*/
4 
5 /*
6  * Copyright (c) 1994 Adam Glass and Charles Hannum.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by Adam Glass and Charles
19  *	Hannum.
20  * 4. The names of the authors may not be used to endorse or promote products
21  *    derived from this software without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26  * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
27  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
28  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
32  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 #include "opt_compat.h"
36 #include "opt_sysvipc.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/sysproto.h>
41 #include <sys/kernel.h>
42 #include <sys/sysctl.h>
43 #include <sys/shm.h>
44 #include <sys/proc.h>
45 #include <sys/malloc.h>
46 #include <sys/mman.h>
47 #include <sys/stat.h>
48 #include <sys/sysent.h>
49 #include <sys/jail.h>
50 
51 #include <sys/mplock2.h>
52 
53 #include <vm/vm.h>
54 #include <vm/vm_param.h>
55 #include <sys/lock.h>
56 #include <vm/pmap.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_map.h>
59 #include <vm/vm_page.h>
60 #include <vm/vm_pager.h>
61 
62 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
63 
64 struct oshmctl_args;
65 static int sys_oshmctl (struct proc *p, struct oshmctl_args *uap);
66 
67 static int shmget_allocate_segment (struct proc *p, struct shmget_args *uap, int mode);
68 static int shmget_existing (struct proc *p, struct shmget_args *uap, int mode, int segnum);
69 
70 /* XXX casting to (sy_call_t *) is bogus, as usual. */
71 static sy_call_t *shmcalls[] = {
72 	(sy_call_t *)sys_shmat, (sy_call_t *)sys_oshmctl,
73 	(sy_call_t *)sys_shmdt, (sy_call_t *)sys_shmget,
74 	(sy_call_t *)sys_shmctl
75 };
76 
77 #define	SHMSEG_FREE     	0x0200
78 #define	SHMSEG_REMOVED  	0x0400
79 #define	SHMSEG_ALLOCATED	0x0800
80 #define	SHMSEG_WANTED		0x1000
81 
82 static int shm_last_free, shm_nused, shm_committed, shmalloced;
83 static struct shmid_ds	*shmsegs;
84 
85 struct shm_handle {
86 	/* vm_offset_t kva; */
87 	vm_object_t shm_object;
88 };
89 
90 struct shmmap_state {
91 	vm_offset_t va;
92 	int shmid;
93 };
94 
95 static void shm_deallocate_segment (struct shmid_ds *);
96 static int shm_find_segment_by_key (key_t);
97 static struct shmid_ds *shm_find_segment_by_shmid (int);
98 static int shm_delete_mapping (struct vmspace *vm, struct shmmap_state *);
99 static void shmrealloc (void);
100 static void shminit (void *);
101 
102 /*
103  * Tuneable values
104  */
105 #ifndef SHMMAXPGS
106 #define	SHMMAXPGS	8192	/* note: sysv shared memory is swap backed */
107 #endif
108 #ifndef SHMMAX
109 #define	SHMMAX	(SHMMAXPGS*PAGE_SIZE)
110 #endif
111 #ifndef SHMMIN
112 #define	SHMMIN	1
113 #endif
114 #ifndef SHMMNI
115 #define	SHMMNI	192
116 #endif
117 #ifndef SHMSEG
118 #define	SHMSEG	128
119 #endif
120 #ifndef SHMALL
121 #define	SHMALL	(SHMMAXPGS)
122 #endif
123 
124 struct	shminfo shminfo = {
125 	SHMMAX,
126 	SHMMIN,
127 	SHMMNI,
128 	SHMSEG,
129 	SHMALL
130 };
131 
132 static int shm_use_phys;
133 
134 TUNABLE_INT("kern.ipc.shmmin", &shminfo.shmmin);
135 TUNABLE_INT("kern.ipc.shmmni", &shminfo.shmmni);
136 TUNABLE_INT("kern.ipc.shmseg", &shminfo.shmseg);
137 TUNABLE_INT("kern.ipc.shmmaxpgs", &shminfo.shmall);
138 TUNABLE_INT("kern.ipc.shm_use_phys", &shm_use_phys);
139 
140 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0, "");
141 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0, "");
142 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RD, &shminfo.shmmni, 0, "");
143 SYSCTL_INT(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RW, &shminfo.shmseg, 0, "");
144 SYSCTL_INT(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0, "");
145 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW, &shm_use_phys, 0, "");
146 
147 static int
148 shm_find_segment_by_key(key_t key)
149 {
150 	int i;
151 
152 	for (i = 0; i < shmalloced; i++)
153 		if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
154 		    shmsegs[i].shm_perm.key == key)
155 			return i;
156 	return -1;
157 }
158 
159 static struct shmid_ds *
160 shm_find_segment_by_shmid(int shmid)
161 {
162 	int segnum;
163 	struct shmid_ds *shmseg;
164 
165 	segnum = IPCID_TO_IX(shmid);
166 	if (segnum < 0 || segnum >= shmalloced)
167 		return NULL;
168 	shmseg = &shmsegs[segnum];
169 	if ((shmseg->shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
170 	    != SHMSEG_ALLOCATED ||
171 	    shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid))
172 		return NULL;
173 	return shmseg;
174 }
175 
176 static void
177 shm_deallocate_segment(struct shmid_ds *shmseg)
178 {
179 	struct shm_handle *shm_handle;
180 	size_t size;
181 
182 	shm_handle = shmseg->shm_internal;
183 	vm_object_deallocate(shm_handle->shm_object);
184 	kfree((caddr_t)shm_handle, M_SHM);
185 	shmseg->shm_internal = NULL;
186 	size = round_page(shmseg->shm_segsz);
187 	shm_committed -= btoc(size);
188 	shm_nused--;
189 	shmseg->shm_perm.mode = SHMSEG_FREE;
190 }
191 
192 static int
193 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
194 {
195 	struct shmid_ds *shmseg;
196 	int segnum, result;
197 	size_t size;
198 
199 	segnum = IPCID_TO_IX(shmmap_s->shmid);
200 	shmseg = &shmsegs[segnum];
201 	size = round_page(shmseg->shm_segsz);
202 	result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
203 	if (result != KERN_SUCCESS)
204 		return EINVAL;
205 	shmmap_s->shmid = -1;
206 	shmseg->shm_dtime = time_second;
207 	if ((--shmseg->shm_nattch <= 0) &&
208 	    (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
209 		shm_deallocate_segment(shmseg);
210 		shm_last_free = segnum;
211 	}
212 	return 0;
213 }
214 
215 /*
216  * MPALMOSTSAFE
217  */
218 int
219 sys_shmdt(struct shmdt_args *uap)
220 {
221 	struct thread *td = curthread;
222 	struct proc *p = td->td_proc;
223 	struct shmmap_state *shmmap_s;
224 	int i;
225 	int error;
226 
227 	if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
228 		return (ENOSYS);
229 
230 	get_mplock();
231 	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
232 	if (shmmap_s == NULL) {
233 		error = EINVAL;
234 		goto done;
235 	}
236 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
237 		if (shmmap_s->shmid != -1 &&
238 		    shmmap_s->va == (vm_offset_t)uap->shmaddr)
239 			break;
240 	}
241 	if (i == shminfo.shmseg)
242 		error = EINVAL;
243 	else
244 		error = shm_delete_mapping(p->p_vmspace, shmmap_s);
245 done:
246 	rel_mplock();
247 	return (error);
248 }
249 
250 /*
251  * MPALMOSTSAFE
252  */
253 int
254 sys_shmat(struct shmat_args *uap)
255 {
256 	struct thread *td = curthread;
257 	struct proc *p = td->td_proc;
258 	int error, i, flags;
259 	struct shmid_ds *shmseg;
260 	struct shmmap_state *shmmap_s = NULL;
261 	struct shm_handle *shm_handle;
262 	vm_offset_t attach_va;
263 	vm_prot_t prot;
264 	vm_size_t size;
265 	int rv;
266 
267 	if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
268 		return (ENOSYS);
269 
270 	get_mplock();
271 again:
272 	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
273 	if (shmmap_s == NULL) {
274 		size = shminfo.shmseg * sizeof(struct shmmap_state);
275 		shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
276 		for (i = 0; i < shminfo.shmseg; i++)
277 			shmmap_s[i].shmid = -1;
278 		if (p->p_vmspace->vm_shm != NULL) {
279 			kfree(shmmap_s, M_SHM);
280 			goto again;
281 		}
282 		p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
283 	}
284 	shmseg = shm_find_segment_by_shmid(uap->shmid);
285 	if (shmseg == NULL) {
286 		error = EINVAL;
287 		goto done;
288 	}
289 	error = ipcperm(p, &shmseg->shm_perm,
290 			(uap->shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
291 	if (error)
292 		goto done;
293 	for (i = 0; i < shminfo.shmseg; i++) {
294 		if (shmmap_s->shmid == -1)
295 			break;
296 		shmmap_s++;
297 	}
298 	if (i >= shminfo.shmseg) {
299 		error = EMFILE;
300 		goto done;
301 	}
302 	size = round_page(shmseg->shm_segsz);
303 #ifdef VM_PROT_READ_IS_EXEC
304 	prot = VM_PROT_READ | VM_PROT_EXECUTE;
305 #else
306 	prot = VM_PROT_READ;
307 #endif
308 	if ((uap->shmflg & SHM_RDONLY) == 0)
309 		prot |= VM_PROT_WRITE;
310 	flags = MAP_ANON | MAP_SHARED;
311 	if (uap->shmaddr) {
312 		flags |= MAP_FIXED;
313 		if (uap->shmflg & SHM_RND) {
314 			attach_va = (vm_offset_t)uap->shmaddr & ~(SHMLBA-1);
315 		} else if (((vm_offset_t)uap->shmaddr & (SHMLBA-1)) == 0) {
316 			attach_va = (vm_offset_t)uap->shmaddr;
317 		} else {
318 			error = EINVAL;
319 			goto done;
320 		}
321 	} else {
322 		/*
323 		 * This is just a hint to vm_map_find() about where to put it.
324 		 */
325 		attach_va = round_page((vm_offset_t)p->p_vmspace->vm_taddr + maxtsiz + maxdsiz);
326 	}
327 
328 	shm_handle = shmseg->shm_internal;
329 	vm_object_reference(shm_handle->shm_object);
330 	rv = vm_map_find(&p->p_vmspace->vm_map,
331 			 shm_handle->shm_object, 0,
332 			 &attach_va, size,
333 			 ((flags & MAP_FIXED) ? 0 : 1),
334 			 VM_MAPTYPE_NORMAL,
335 			 prot, prot,
336 			 0);
337 	if (rv != KERN_SUCCESS) {
338                 vm_object_deallocate(shm_handle->shm_object);
339 		error = ENOMEM;
340 		goto done;
341 	}
342 	vm_map_inherit(&p->p_vmspace->vm_map,
343 		       attach_va, attach_va + size, VM_INHERIT_SHARE);
344 
345 	KKASSERT(shmmap_s->shmid == -1);
346 	shmmap_s->va = attach_va;
347 	shmmap_s->shmid = uap->shmid;
348 	shmseg->shm_lpid = p->p_pid;
349 	shmseg->shm_atime = time_second;
350 	shmseg->shm_nattch++;
351 	uap->sysmsg_result = attach_va;
352 	error = 0;
353 done:
354 	rel_mplock();
355 	return error;
356 }
357 
358 struct oshmid_ds {
359 	struct	ipc_perm shm_perm;	/* operation perms */
360 	int	shm_segsz;		/* size of segment (bytes) */
361 	ushort	shm_cpid;		/* pid, creator */
362 	ushort	shm_lpid;		/* pid, last operation */
363 	short	shm_nattch;		/* no. of current attaches */
364 	time_t	shm_atime;		/* last attach time */
365 	time_t	shm_dtime;		/* last detach time */
366 	time_t	shm_ctime;		/* last change time */
367 	void	*shm_handle;		/* internal handle for shm segment */
368 };
369 
370 struct oshmctl_args {
371 	struct sysmsg sysmsg;
372 	int shmid;
373 	int cmd;
374 	struct oshmid_ds *ubuf;
375 };
376 
377 /*
378  * MPALMOSTSAFE
379  */
380 static int
381 sys_oshmctl(struct proc *p, struct oshmctl_args *uap)
382 {
383 #ifdef COMPAT_43
384 	struct thread *td = curthread;
385 	struct shmid_ds *shmseg;
386 	struct oshmid_ds outbuf;
387 	int error;
388 
389 	if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
390 		return (ENOSYS);
391 
392 	get_mplock();
393 	shmseg = shm_find_segment_by_shmid(uap->shmid);
394 	if (shmseg == NULL) {
395 		error = EINVAL;
396 		goto done;
397 	}
398 
399 	switch (uap->cmd) {
400 	case IPC_STAT:
401 		error = ipcperm(p, &shmseg->shm_perm, IPC_R);
402 		if (error)
403 			break;
404 		outbuf.shm_perm = shmseg->shm_perm;
405 		outbuf.shm_segsz = shmseg->shm_segsz;
406 		outbuf.shm_cpid = shmseg->shm_cpid;
407 		outbuf.shm_lpid = shmseg->shm_lpid;
408 		outbuf.shm_nattch = shmseg->shm_nattch;
409 		outbuf.shm_atime = shmseg->shm_atime;
410 		outbuf.shm_dtime = shmseg->shm_dtime;
411 		outbuf.shm_ctime = shmseg->shm_ctime;
412 		outbuf.shm_handle = shmseg->shm_internal;
413 		error = copyout((caddr_t)&outbuf, uap->ubuf, sizeof(outbuf));
414 		break;
415 	default:
416 		/* XXX casting to (sy_call_t *) is bogus, as usual. */
417 		error = sys_shmctl((struct shmctl_args *)uap);
418 	}
419 done:
420 	rel_mplock();
421 	return error;
422 #else
423 	return EINVAL;
424 #endif
425 }
426 
427 /*
428  * MPALMOSTSAFE
429  */
430 int
431 sys_shmctl(struct shmctl_args *uap)
432 {
433 	struct thread *td = curthread;
434 	struct proc *p = td->td_proc;
435 	int error;
436 	struct shmid_ds inbuf;
437 	struct shmid_ds *shmseg;
438 
439 	if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
440 		return (ENOSYS);
441 
442 	get_mplock();
443 	shmseg = shm_find_segment_by_shmid(uap->shmid);
444 	if (shmseg == NULL) {
445 		error = EINVAL;
446 		goto done;
447 	}
448 
449 	switch (uap->cmd) {
450 	case IPC_STAT:
451 		error = ipcperm(p, &shmseg->shm_perm, IPC_R);
452 		if (error == 0)
453 			error = copyout(shmseg, uap->buf, sizeof(inbuf));
454 		break;
455 	case IPC_SET:
456 		error = ipcperm(p, &shmseg->shm_perm, IPC_M);
457 		if (error == 0)
458 			error = copyin(uap->buf, &inbuf, sizeof(inbuf));
459 		if (error == 0) {
460 			shmseg->shm_perm.uid = inbuf.shm_perm.uid;
461 			shmseg->shm_perm.gid = inbuf.shm_perm.gid;
462 			shmseg->shm_perm.mode =
463 			    (shmseg->shm_perm.mode & ~ACCESSPERMS) |
464 			    (inbuf.shm_perm.mode & ACCESSPERMS);
465 			shmseg->shm_ctime = time_second;
466 		}
467 		break;
468 	case IPC_RMID:
469 		error = ipcperm(p, &shmseg->shm_perm, IPC_M);
470 		if (error == 0) {
471 			shmseg->shm_perm.key = IPC_PRIVATE;
472 			shmseg->shm_perm.mode |= SHMSEG_REMOVED;
473 			if (shmseg->shm_nattch <= 0) {
474 				shm_deallocate_segment(shmseg);
475 				shm_last_free = IPCID_TO_IX(uap->shmid);
476 			}
477 		}
478 		break;
479 #if 0
480 	case SHM_LOCK:
481 	case SHM_UNLOCK:
482 #endif
483 	default:
484 		error = EINVAL;
485 		break;
486 	}
487 done:
488 	rel_mplock();
489 	return error;
490 }
491 
492 static int
493 shmget_existing(struct proc *p, struct shmget_args *uap, int mode, int segnum)
494 {
495 	struct shmid_ds *shmseg;
496 	int error;
497 
498 	shmseg = &shmsegs[segnum];
499 	if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
500 		/*
501 		 * This segment is in the process of being allocated.  Wait
502 		 * until it's done, and look the key up again (in case the
503 		 * allocation failed or it was freed).
504 		 */
505 		shmseg->shm_perm.mode |= SHMSEG_WANTED;
506 		error = tsleep((caddr_t)shmseg, PCATCH, "shmget", 0);
507 		if (error)
508 			return error;
509 		return EAGAIN;
510 	}
511 	if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
512 		return EEXIST;
513 	error = ipcperm(p, &shmseg->shm_perm, mode);
514 	if (error)
515 		return error;
516 	if (uap->size && uap->size > shmseg->shm_segsz)
517 		return EINVAL;
518 	uap->sysmsg_result = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
519 	return 0;
520 }
521 
522 static int
523 shmget_allocate_segment(struct proc *p, struct shmget_args *uap, int mode)
524 {
525 	int i, segnum, shmid, size;
526 	struct ucred *cred = p->p_ucred;
527 	struct shmid_ds *shmseg;
528 	struct shm_handle *shm_handle;
529 
530 	if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
531 		return EINVAL;
532 	if (shm_nused >= shminfo.shmmni) /* any shmids left? */
533 		return ENOSPC;
534 	size = round_page(uap->size);
535 	if (shm_committed + btoc(size) > shminfo.shmall)
536 		return ENOMEM;
537 	if (shm_last_free < 0) {
538 		shmrealloc();	/* maybe expand the shmsegs[] array */
539 		for (i = 0; i < shmalloced; i++)
540 			if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
541 				break;
542 		if (i == shmalloced)
543 			return ENOSPC;
544 		segnum = i;
545 	} else  {
546 		segnum = shm_last_free;
547 		shm_last_free = -1;
548 	}
549 	shmseg = &shmsegs[segnum];
550 	/*
551 	 * In case we sleep in malloc(), mark the segment present but deleted
552 	 * so that noone else tries to create the same key.
553 	 */
554 	shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
555 	shmseg->shm_perm.key = uap->key;
556 	shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
557 	shm_handle = kmalloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
558 	shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
559 
560 	/*
561 	 * We make sure that we have allocated a pager before we need
562 	 * to.
563 	 */
564 	if (shm_use_phys) {
565 		shm_handle->shm_object =
566 		   phys_pager_alloc(NULL, size, VM_PROT_DEFAULT, 0);
567 	} else {
568 		shm_handle->shm_object =
569 		   swap_pager_alloc(NULL, size, VM_PROT_DEFAULT, 0);
570 	}
571 	vm_object_clear_flag(shm_handle->shm_object, OBJ_ONEMAPPING);
572 	vm_object_set_flag(shm_handle->shm_object, OBJ_NOSPLIT);
573 
574 	shmseg->shm_internal = shm_handle;
575 	shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
576 	shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
577 	shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
578 	    (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
579 	shmseg->shm_segsz = uap->size;
580 	shmseg->shm_cpid = p->p_pid;
581 	shmseg->shm_lpid = shmseg->shm_nattch = 0;
582 	shmseg->shm_atime = shmseg->shm_dtime = 0;
583 	shmseg->shm_ctime = time_second;
584 	shm_committed += btoc(size);
585 	shm_nused++;
586 	if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
587 		/*
588 		 * Somebody else wanted this key while we were asleep.  Wake
589 		 * them up now.
590 		 */
591 		shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
592 		wakeup((caddr_t)shmseg);
593 	}
594 	uap->sysmsg_result = shmid;
595 	return 0;
596 }
597 
598 /*
599  * MPALMOSTSAFE
600  */
601 int
602 sys_shmget(struct shmget_args *uap)
603 {
604 	struct thread *td = curthread;
605 	struct proc *p = td->td_proc;
606 	int segnum, mode, error;
607 
608 	if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
609 		return (ENOSYS);
610 
611 	mode = uap->shmflg & ACCESSPERMS;
612 	get_mplock();
613 
614 	if (uap->key != IPC_PRIVATE) {
615 	again:
616 		segnum = shm_find_segment_by_key(uap->key);
617 		if (segnum >= 0) {
618 			error = shmget_existing(p, uap, mode, segnum);
619 			if (error == EAGAIN)
620 				goto again;
621 			goto done;
622 		}
623 		if ((uap->shmflg & IPC_CREAT) == 0) {
624 			error = ENOENT;
625 			goto done;
626 		}
627 	}
628 	error = shmget_allocate_segment(p, uap, mode);
629 done:
630 	rel_mplock();
631 	return (error);
632 }
633 
634 /*
635  * shmsys_args(int which, int a2, ...) (VARARGS)
636  *
637  * MPALMOSTSAFE
638  */
639 int
640 sys_shmsys(struct shmsys_args *uap)
641 {
642 	struct thread *td = curthread;
643 	unsigned int which = (unsigned int)uap->which;
644 	int error;
645 
646 	if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
647 		return (ENOSYS);
648 
649 	if (which >= sizeof(shmcalls)/sizeof(shmcalls[0]))
650 		return EINVAL;
651 	get_mplock();
652 	bcopy(&uap->a2, &uap->which,
653 		sizeof(struct shmsys_args) - offsetof(struct shmsys_args, a2));
654 	error = ((*shmcalls[which])(uap));
655 	rel_mplock();
656 
657 	return(error);
658 }
659 
660 void
661 shmfork(struct proc *p1, struct proc *p2)
662 {
663 	struct shmmap_state *shmmap_s;
664 	size_t size;
665 	int i;
666 
667 	size = shminfo.shmseg * sizeof(struct shmmap_state);
668 	shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
669 	bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmmap_s, size);
670 	p2->p_vmspace->vm_shm = (caddr_t)shmmap_s;
671 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
672 		if (shmmap_s->shmid != -1)
673 			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
674 }
675 
676 void
677 shmexit(struct vmspace *vm)
678 {
679 	struct shmmap_state *base, *shm;
680 	int i;
681 
682 	if ((base = (struct shmmap_state *)vm->vm_shm) != NULL) {
683 		vm->vm_shm = NULL;
684 		for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
685 			if (shm->shmid != -1)
686 				shm_delete_mapping(vm, shm);
687 		}
688 		kfree(base, M_SHM);
689 	}
690 }
691 
692 static void
693 shmrealloc(void)
694 {
695 	int i;
696 	struct shmid_ds *newsegs;
697 
698 	if (shmalloced >= shminfo.shmmni)
699 		return;
700 
701 	newsegs = kmalloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
702 	for (i = 0; i < shmalloced; i++)
703 		bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
704 	for (; i < shminfo.shmmni; i++) {
705 		shmsegs[i].shm_perm.mode = SHMSEG_FREE;
706 		shmsegs[i].shm_perm.seq = 0;
707 	}
708 	kfree(shmsegs, M_SHM);
709 	shmsegs = newsegs;
710 	shmalloced = shminfo.shmmni;
711 }
712 
713 static void
714 shminit(void *dummy)
715 {
716 	int i;
717 
718 	shminfo.shmmax = shminfo.shmall * PAGE_SIZE;
719 	shmalloced = shminfo.shmmni;
720 	shmsegs = kmalloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
721 	for (i = 0; i < shmalloced; i++) {
722 		shmsegs[i].shm_perm.mode = SHMSEG_FREE;
723 		shmsegs[i].shm_perm.seq = 0;
724 	}
725 	shm_last_free = 0;
726 	shm_nused = 0;
727 	shm_committed = 0;
728 }
729 SYSINIT(sysv_shm, SI_SUB_SYSV_SHM, SI_ORDER_FIRST, shminit, NULL);
730