xref: /freebsd/sys/kern/sysv_shm.c (revision cbe953461557af3eb3243cf64fe8f12bbea6c743)
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/lock.h>
72 #include <sys/sysctl.h>
73 #include <sys/shm.h>
74 #include <sys/proc.h>
75 #include <sys/malloc.h>
76 #include <sys/mman.h>
77 #include <sys/module.h>
78 #include <sys/mutex.h>
79 #include <sys/resourcevar.h>
80 #include <sys/stat.h>
81 #include <sys/syscall.h>
82 #include <sys/syscallsubr.h>
83 #include <sys/sysent.h>
84 #include <sys/sysproto.h>
85 #include <sys/jail.h>
86 
87 #include <security/mac/mac_framework.h>
88 
89 #include <vm/vm.h>
90 #include <vm/vm_param.h>
91 #include <vm/pmap.h>
92 #include <vm/vm_object.h>
93 #include <vm/vm_map.h>
94 #include <vm/vm_page.h>
95 #include <vm/vm_pager.h>
96 
97 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
98 
99 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
100 struct oshmctl_args;
101 static int oshmctl(struct thread *td, struct oshmctl_args *uap);
102 #endif
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 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
110 /* XXX casting to (sy_call_t *) is bogus, as usual. */
111 static sy_call_t *shmcalls[] = {
112 	(sy_call_t *)shmat, (sy_call_t *)oshmctl,
113 	(sy_call_t *)shmdt, (sy_call_t *)shmget,
114 	(sy_call_t *)shmctl
115 };
116 #endif
117 
118 #define	SHMSEG_FREE     	0x0200
119 #define	SHMSEG_REMOVED  	0x0400
120 #define	SHMSEG_ALLOCATED	0x0800
121 #define	SHMSEG_WANTED		0x1000
122 
123 static int shm_last_free, shm_nused, shmalloced;
124 vm_size_t shm_committed;
125 static struct shmid_kernel	*shmsegs;
126 
127 struct shmmap_state {
128 	vm_offset_t va;
129 	int shmid;
130 };
131 
132 static void shm_deallocate_segment(struct shmid_kernel *);
133 static int shm_find_segment_by_key(key_t);
134 static struct shmid_kernel *shm_find_segment_by_shmid(int);
135 static struct shmid_kernel *shm_find_segment_by_shmidx(int);
136 static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *);
137 static void shmrealloc(void);
138 static void shminit(void);
139 static int sysvshm_modload(struct module *, int, void *);
140 static int shmunload(void);
141 static void shmexit_myhook(struct vmspace *vm);
142 static void shmfork_myhook(struct proc *p1, struct proc *p2);
143 static int sysctl_shmsegs(SYSCTL_HANDLER_ARGS);
144 
145 /*
146  * Tuneable values.
147  */
148 #ifndef SHMMAXPGS
149 #define	SHMMAXPGS	8192	/* Note: sysv shared memory is swap backed. */
150 #endif
151 #ifndef SHMMAX
152 #define	SHMMAX	(SHMMAXPGS*PAGE_SIZE)
153 #endif
154 #ifndef SHMMIN
155 #define	SHMMIN	1
156 #endif
157 #ifndef SHMMNI
158 #define	SHMMNI	192
159 #endif
160 #ifndef SHMSEG
161 #define	SHMSEG	128
162 #endif
163 #ifndef SHMALL
164 #define	SHMALL	(SHMMAXPGS)
165 #endif
166 
167 struct	shminfo shminfo = {
168 	SHMMAX,
169 	SHMMIN,
170 	SHMMNI,
171 	SHMSEG,
172 	SHMALL
173 };
174 
175 static int shm_use_phys;
176 static int shm_allow_removed;
177 
178 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0,
179     "Maximum shared memory segment size");
180 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0,
181     "Minimum shared memory segment size");
182 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RDTUN, &shminfo.shmmni, 0,
183     "Number of shared memory identifiers");
184 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RDTUN, &shminfo.shmseg, 0,
185     "Number of segments per process");
186 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0,
187     "Maximum number of pages available for shared memory");
188 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW,
189     &shm_use_phys, 0, "Enable/Disable locking of shared memory pages in core");
190 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RW,
191     &shm_allow_removed, 0,
192     "Enable/Disable attachment to attached segments marked for removal");
193 SYSCTL_PROC(_kern_ipc, OID_AUTO, shmsegs, CTLFLAG_RD,
194     NULL, 0, sysctl_shmsegs, "",
195     "Current number of shared memory segments allocated");
196 
197 static int
198 shm_find_segment_by_key(key)
199 	key_t key;
200 {
201 	int i;
202 
203 	for (i = 0; i < shmalloced; i++)
204 		if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) &&
205 		    shmsegs[i].u.shm_perm.key == key)
206 			return (i);
207 	return (-1);
208 }
209 
210 static struct shmid_kernel *
211 shm_find_segment_by_shmid(int shmid)
212 {
213 	int segnum;
214 	struct shmid_kernel *shmseg;
215 
216 	segnum = IPCID_TO_IX(shmid);
217 	if (segnum < 0 || segnum >= shmalloced)
218 		return (NULL);
219 	shmseg = &shmsegs[segnum];
220 	if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
221 	    (!shm_allow_removed &&
222 	     (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0) ||
223 	    shmseg->u.shm_perm.seq != IPCID_TO_SEQ(shmid))
224 		return (NULL);
225 	return (shmseg);
226 }
227 
228 static struct shmid_kernel *
229 shm_find_segment_by_shmidx(int segnum)
230 {
231 	struct shmid_kernel *shmseg;
232 
233 	if (segnum < 0 || segnum >= shmalloced)
234 		return (NULL);
235 	shmseg = &shmsegs[segnum];
236 	if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
237 	    (!shm_allow_removed &&
238 	     (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0))
239 		return (NULL);
240 	return (shmseg);
241 }
242 
243 static void
244 shm_deallocate_segment(shmseg)
245 	struct shmid_kernel *shmseg;
246 {
247 	vm_size_t size;
248 
249 	GIANT_REQUIRED;
250 
251 	vm_object_deallocate(shmseg->u.shm_internal);
252 	shmseg->u.shm_internal = NULL;
253 	size = round_page(shmseg->shm_bsegsz);
254 	shm_committed -= btoc(size);
255 	shm_nused--;
256 	shmseg->u.shm_perm.mode = SHMSEG_FREE;
257 #ifdef MAC
258 	mac_sysvshm_cleanup(shmseg);
259 #endif
260 }
261 
262 static int
263 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
264 {
265 	struct shmid_kernel *shmseg;
266 	int segnum, result;
267 	vm_size_t size;
268 
269 	GIANT_REQUIRED;
270 
271 	segnum = IPCID_TO_IX(shmmap_s->shmid);
272 	shmseg = &shmsegs[segnum];
273 	size = round_page(shmseg->shm_bsegsz);
274 	result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
275 	if (result != KERN_SUCCESS)
276 		return (EINVAL);
277 	shmmap_s->shmid = -1;
278 	shmseg->u.shm_dtime = time_second;
279 	if ((--shmseg->u.shm_nattch <= 0) &&
280 	    (shmseg->u.shm_perm.mode & SHMSEG_REMOVED)) {
281 		shm_deallocate_segment(shmseg);
282 		shm_last_free = segnum;
283 	}
284 	return (0);
285 }
286 
287 #ifndef _SYS_SYSPROTO_H_
288 struct shmdt_args {
289 	const void *shmaddr;
290 };
291 #endif
292 int
293 shmdt(td, uap)
294 	struct thread *td;
295 	struct shmdt_args *uap;
296 {
297 	struct proc *p = td->td_proc;
298 	struct shmmap_state *shmmap_s;
299 #ifdef MAC
300 	struct shmid_kernel *shmsegptr;
301 #endif
302 	int i;
303 	int error = 0;
304 
305 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
306 		return (ENOSYS);
307 	mtx_lock(&Giant);
308 	shmmap_s = p->p_vmspace->vm_shm;
309  	if (shmmap_s == NULL) {
310 		error = EINVAL;
311 		goto done2;
312 	}
313 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
314 		if (shmmap_s->shmid != -1 &&
315 		    shmmap_s->va == (vm_offset_t)uap->shmaddr) {
316 			break;
317 		}
318 	}
319 	if (i == shminfo.shmseg) {
320 		error = EINVAL;
321 		goto done2;
322 	}
323 #ifdef MAC
324 	shmsegptr = &shmsegs[IPCID_TO_IX(shmmap_s->shmid)];
325 	error = mac_sysvshm_check_shmdt(td->td_ucred, shmsegptr);
326 	if (error != 0)
327 		goto done2;
328 #endif
329 	error = shm_delete_mapping(p->p_vmspace, shmmap_s);
330 done2:
331 	mtx_unlock(&Giant);
332 	return (error);
333 }
334 
335 #ifndef _SYS_SYSPROTO_H_
336 struct shmat_args {
337 	int shmid;
338 	const void *shmaddr;
339 	int shmflg;
340 };
341 #endif
342 int
343 kern_shmat(td, shmid, shmaddr, shmflg)
344 	struct thread *td;
345 	int shmid;
346 	const void *shmaddr;
347 	int shmflg;
348 {
349 	struct proc *p = td->td_proc;
350 	int i, flags;
351 	struct shmid_kernel *shmseg;
352 	struct shmmap_state *shmmap_s = NULL;
353 	vm_offset_t attach_va;
354 	vm_prot_t prot;
355 	vm_size_t size;
356 	int rv;
357 	int error = 0;
358 
359 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
360 		return (ENOSYS);
361 	mtx_lock(&Giant);
362 	shmmap_s = p->p_vmspace->vm_shm;
363 	if (shmmap_s == NULL) {
364 		shmmap_s = malloc(shminfo.shmseg * sizeof(struct shmmap_state),
365 		    M_SHM, M_WAITOK);
366 		for (i = 0; i < shminfo.shmseg; i++)
367 			shmmap_s[i].shmid = -1;
368 		p->p_vmspace->vm_shm = shmmap_s;
369 	}
370 	shmseg = shm_find_segment_by_shmid(shmid);
371 	if (shmseg == NULL) {
372 		error = EINVAL;
373 		goto done2;
374 	}
375 	error = ipcperm(td, &shmseg->u.shm_perm,
376 	    (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
377 	if (error)
378 		goto done2;
379 #ifdef MAC
380 	error = mac_sysvshm_check_shmat(td->td_ucred, shmseg, shmflg);
381 	if (error != 0)
382 		goto done2;
383 #endif
384 	for (i = 0; i < shminfo.shmseg; i++) {
385 		if (shmmap_s->shmid == -1)
386 			break;
387 		shmmap_s++;
388 	}
389 	if (i >= shminfo.shmseg) {
390 		error = EMFILE;
391 		goto done2;
392 	}
393 	size = round_page(shmseg->shm_bsegsz);
394 	prot = VM_PROT_READ;
395 	if ((shmflg & SHM_RDONLY) == 0)
396 		prot |= VM_PROT_WRITE;
397 	flags = MAP_ANON | MAP_SHARED;
398 	if (shmaddr) {
399 		flags |= MAP_FIXED;
400 		if (shmflg & SHM_RND) {
401 			attach_va = (vm_offset_t)shmaddr & ~(SHMLBA-1);
402 		} else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0) {
403 			attach_va = (vm_offset_t)shmaddr;
404 		} else {
405 			error = EINVAL;
406 			goto done2;
407 		}
408 	} else {
409 		/*
410 		 * This is just a hint to vm_map_find() about where to
411 		 * put it.
412 		 */
413 		PROC_LOCK(p);
414 		attach_va = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
415 		    lim_max(p, RLIMIT_DATA));
416 		PROC_UNLOCK(p);
417 	}
418 
419 	vm_object_reference(shmseg->u.shm_internal);
420 	rv = vm_map_find(&p->p_vmspace->vm_map, shmseg->u.shm_internal,
421 	    0, &attach_va, size, (flags & MAP_FIXED) ? VMFS_NO_SPACE :
422 	    VMFS_ANY_SPACE, prot, prot, 0);
423 	if (rv != KERN_SUCCESS) {
424 		vm_object_deallocate(shmseg->u.shm_internal);
425 		error = ENOMEM;
426 		goto done2;
427 	}
428 	vm_map_inherit(&p->p_vmspace->vm_map,
429 		attach_va, attach_va + size, VM_INHERIT_SHARE);
430 
431 	shmmap_s->va = attach_va;
432 	shmmap_s->shmid = shmid;
433 	shmseg->u.shm_lpid = p->p_pid;
434 	shmseg->u.shm_atime = time_second;
435 	shmseg->u.shm_nattch++;
436 	td->td_retval[0] = attach_va;
437 done2:
438 	mtx_unlock(&Giant);
439 	return (error);
440 }
441 
442 int
443 shmat(td, uap)
444 	struct thread *td;
445 	struct shmat_args *uap;
446 {
447 	return kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg);
448 }
449 
450 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
451 struct oshmid_ds {
452 	struct	ipc_perm shm_perm;	/* operation perms */
453 	int	shm_segsz;		/* size of segment (bytes) */
454 	u_short	shm_cpid;		/* pid, creator */
455 	u_short	shm_lpid;		/* pid, last operation */
456 	short	shm_nattch;		/* no. of current attaches */
457 	time_t	shm_atime;		/* last attach time */
458 	time_t	shm_dtime;		/* last detach time */
459 	time_t	shm_ctime;		/* last change time */
460 	void	*shm_handle;		/* internal handle for shm segment */
461 };
462 
463 struct oshmctl_args {
464 	int shmid;
465 	int cmd;
466 	struct oshmid_ds *ubuf;
467 };
468 static int
469 oshmctl(td, uap)
470 	struct thread *td;
471 	struct oshmctl_args *uap;
472 {
473 #ifdef COMPAT_43
474 	int error = 0;
475 	struct shmid_kernel *shmseg;
476 	struct oshmid_ds outbuf;
477 
478 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
479 		return (ENOSYS);
480 	mtx_lock(&Giant);
481 	shmseg = shm_find_segment_by_shmid(uap->shmid);
482 	if (shmseg == NULL) {
483 		error = EINVAL;
484 		goto done2;
485 	}
486 	switch (uap->cmd) {
487 	case IPC_STAT:
488 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
489 		if (error)
490 			goto done2;
491 #ifdef MAC
492 		error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, uap->cmd);
493 		if (error != 0)
494 			goto done2;
495 #endif
496 		outbuf.shm_perm = shmseg->u.shm_perm;
497 		outbuf.shm_segsz = shmseg->u.shm_segsz;
498 		outbuf.shm_cpid = shmseg->u.shm_cpid;
499 		outbuf.shm_lpid = shmseg->u.shm_lpid;
500 		outbuf.shm_nattch = shmseg->u.shm_nattch;
501 		outbuf.shm_atime = shmseg->u.shm_atime;
502 		outbuf.shm_dtime = shmseg->u.shm_dtime;
503 		outbuf.shm_ctime = shmseg->u.shm_ctime;
504 		outbuf.shm_handle = shmseg->u.shm_internal;
505 		error = copyout(&outbuf, uap->ubuf, sizeof(outbuf));
506 		if (error)
507 			goto done2;
508 		break;
509 	default:
510 		error = shmctl(td, (struct shmctl_args *)uap);
511 		break;
512 	}
513 done2:
514 	mtx_unlock(&Giant);
515 	return (error);
516 #else
517 	return (EINVAL);
518 #endif
519 }
520 #endif
521 
522 #ifndef _SYS_SYSPROTO_H_
523 struct shmctl_args {
524 	int shmid;
525 	int cmd;
526 	struct shmid_ds *buf;
527 };
528 #endif
529 int
530 kern_shmctl(td, shmid, cmd, buf, bufsz)
531 	struct thread *td;
532 	int shmid;
533 	int cmd;
534 	void *buf;
535 	size_t *bufsz;
536 {
537 	int error = 0;
538 	struct shmid_kernel *shmseg;
539 
540 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
541 		return (ENOSYS);
542 
543 	mtx_lock(&Giant);
544 	switch (cmd) {
545 	/*
546 	 * It is possible that kern_shmctl is being called from the Linux ABI
547 	 * layer, in which case, we will need to implement IPC_INFO.  It should
548 	 * be noted that other shmctl calls will be funneled through here for
549 	 * Linix binaries as well.
550 	 *
551 	 * NB: The Linux ABI layer will convert this data to structure(s) more
552 	 * consistent with the Linux ABI.
553 	 */
554 	case IPC_INFO:
555 		memcpy(buf, &shminfo, sizeof(shminfo));
556 		if (bufsz)
557 			*bufsz = sizeof(shminfo);
558 		td->td_retval[0] = shmalloced;
559 		goto done2;
560 	case SHM_INFO: {
561 		struct shm_info shm_info;
562 		shm_info.used_ids = shm_nused;
563 		shm_info.shm_rss = 0;	/*XXX where to get from ? */
564 		shm_info.shm_tot = 0;	/*XXX where to get from ? */
565 		shm_info.shm_swp = 0;	/*XXX where to get from ? */
566 		shm_info.swap_attempts = 0;	/*XXX where to get from ? */
567 		shm_info.swap_successes = 0;	/*XXX where to get from ? */
568 		memcpy(buf, &shm_info, sizeof(shm_info));
569 		if (bufsz)
570 			*bufsz = sizeof(shm_info);
571 		td->td_retval[0] = shmalloced;
572 		goto done2;
573 	}
574 	}
575 	if (cmd == SHM_STAT)
576 		shmseg = shm_find_segment_by_shmidx(shmid);
577 	else
578 		shmseg = shm_find_segment_by_shmid(shmid);
579 	if (shmseg == NULL) {
580 		error = EINVAL;
581 		goto done2;
582 	}
583 #ifdef MAC
584 	error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, cmd);
585 	if (error != 0)
586 		goto done2;
587 #endif
588 	switch (cmd) {
589 	case SHM_STAT:
590 	case IPC_STAT:
591 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
592 		if (error)
593 			goto done2;
594 		memcpy(buf, &shmseg->u, sizeof(struct shmid_ds));
595 		if (bufsz)
596 			*bufsz = sizeof(struct shmid_ds);
597 		if (cmd == SHM_STAT)
598 			td->td_retval[0] = IXSEQ_TO_IPCID(shmid, shmseg->u.shm_perm);
599 		break;
600 	case IPC_SET: {
601 		struct shmid_ds *shmid;
602 
603 		shmid = (struct shmid_ds *)buf;
604 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
605 		if (error)
606 			goto done2;
607 		shmseg->u.shm_perm.uid = shmid->shm_perm.uid;
608 		shmseg->u.shm_perm.gid = shmid->shm_perm.gid;
609 		shmseg->u.shm_perm.mode =
610 		    (shmseg->u.shm_perm.mode & ~ACCESSPERMS) |
611 		    (shmid->shm_perm.mode & ACCESSPERMS);
612 		shmseg->u.shm_ctime = time_second;
613 		break;
614 	}
615 	case IPC_RMID:
616 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
617 		if (error)
618 			goto done2;
619 		shmseg->u.shm_perm.key = IPC_PRIVATE;
620 		shmseg->u.shm_perm.mode |= SHMSEG_REMOVED;
621 		if (shmseg->u.shm_nattch <= 0) {
622 			shm_deallocate_segment(shmseg);
623 			shm_last_free = IPCID_TO_IX(shmid);
624 		}
625 		break;
626 #if 0
627 	case SHM_LOCK:
628 	case SHM_UNLOCK:
629 #endif
630 	default:
631 		error = EINVAL;
632 		break;
633 	}
634 done2:
635 	mtx_unlock(&Giant);
636 	return (error);
637 }
638 
639 int
640 shmctl(td, uap)
641 	struct thread *td;
642 	struct shmctl_args *uap;
643 {
644 	int error = 0;
645 	struct shmid_ds buf;
646 	size_t bufsz;
647 
648 	/*
649 	 * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
650 	 * Linux binaries.  If we see the call come through the FreeBSD ABI,
651 	 * return an error back to the user since we do not to support this.
652 	 */
653 	if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
654 	    uap->cmd == SHM_STAT)
655 		return (EINVAL);
656 
657 	/* IPC_SET needs to copyin the buffer before calling kern_shmctl */
658 	if (uap->cmd == IPC_SET) {
659 		if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds))))
660 			goto done;
661 	}
662 
663 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
664 	if (error)
665 		goto done;
666 
667 	/* Cases in which we need to copyout */
668 	switch (uap->cmd) {
669 	case IPC_STAT:
670 		error = copyout(&buf, uap->buf, bufsz);
671 		break;
672 	}
673 
674 done:
675 	if (error) {
676 		/* Invalidate the return value */
677 		td->td_retval[0] = -1;
678 	}
679 	return (error);
680 }
681 
682 
683 #ifndef _SYS_SYSPROTO_H_
684 struct shmget_args {
685 	key_t key;
686 	size_t size;
687 	int shmflg;
688 };
689 #endif
690 static int
691 shmget_existing(td, uap, mode, segnum)
692 	struct thread *td;
693 	struct shmget_args *uap;
694 	int mode;
695 	int segnum;
696 {
697 	struct shmid_kernel *shmseg;
698 	int error;
699 
700 	shmseg = &shmsegs[segnum];
701 	if (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) {
702 		/*
703 		 * This segment is in the process of being allocated.  Wait
704 		 * until it's done, and look the key up again (in case the
705 		 * allocation failed or it was freed).
706 		 */
707 		shmseg->u.shm_perm.mode |= SHMSEG_WANTED;
708 		error = tsleep(shmseg, PLOCK | PCATCH, "shmget", 0);
709 		if (error)
710 			return (error);
711 		return (EAGAIN);
712 	}
713 	if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
714 		return (EEXIST);
715 #ifdef MAC
716 	error = mac_sysvshm_check_shmget(td->td_ucred, shmseg, uap->shmflg);
717 	if (error != 0)
718 		return (error);
719 #endif
720 	if (uap->size != 0 && uap->size > shmseg->shm_bsegsz)
721 		return (EINVAL);
722 	td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
723 	return (0);
724 }
725 
726 static int
727 shmget_allocate_segment(td, uap, mode)
728 	struct thread *td;
729 	struct shmget_args *uap;
730 	int mode;
731 {
732 	int i, segnum, shmid;
733 	size_t size;
734 	struct ucred *cred = td->td_ucred;
735 	struct shmid_kernel *shmseg;
736 	vm_object_t shm_object;
737 
738 	GIANT_REQUIRED;
739 
740 	if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
741 		return (EINVAL);
742 	if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
743 		return (ENOSPC);
744 	size = round_page(uap->size);
745 	if (shm_committed + btoc(size) > shminfo.shmall)
746 		return (ENOMEM);
747 	if (shm_last_free < 0) {
748 		shmrealloc();	/* Maybe expand the shmsegs[] array. */
749 		for (i = 0; i < shmalloced; i++)
750 			if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE)
751 				break;
752 		if (i == shmalloced)
753 			return (ENOSPC);
754 		segnum = i;
755 	} else  {
756 		segnum = shm_last_free;
757 		shm_last_free = -1;
758 	}
759 	shmseg = &shmsegs[segnum];
760 	/*
761 	 * In case we sleep in malloc(), mark the segment present but deleted
762 	 * so that noone else tries to create the same key.
763 	 */
764 	shmseg->u.shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
765 	shmseg->u.shm_perm.key = uap->key;
766 	shmseg->u.shm_perm.seq = (shmseg->u.shm_perm.seq + 1) & 0x7fff;
767 	shmid = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
768 
769 	/*
770 	 * We make sure that we have allocated a pager before we need
771 	 * to.
772 	 */
773 	if (shm_use_phys) {
774 		shm_object =
775 		    vm_pager_allocate(OBJT_PHYS, 0, size, VM_PROT_DEFAULT, 0);
776 	} else {
777 		shm_object =
778 		    vm_pager_allocate(OBJT_SWAP, 0, size, VM_PROT_DEFAULT, 0);
779 	}
780 	VM_OBJECT_LOCK(shm_object);
781 	vm_object_clear_flag(shm_object, OBJ_ONEMAPPING);
782 	vm_object_set_flag(shm_object, OBJ_NOSPLIT);
783 	VM_OBJECT_UNLOCK(shm_object);
784 
785 	shmseg->u.shm_internal = shm_object;
786 	shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = cred->cr_uid;
787 	shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid;
788 	shmseg->u.shm_perm.mode = (shmseg->u.shm_perm.mode & SHMSEG_WANTED) |
789 	    (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
790 	shmseg->u.shm_segsz = uap->size;
791 	shmseg->shm_bsegsz = uap->size;
792 	shmseg->u.shm_cpid = td->td_proc->p_pid;
793 	shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0;
794 	shmseg->u.shm_atime = shmseg->u.shm_dtime = 0;
795 #ifdef MAC
796 	mac_sysvshm_create(cred, shmseg);
797 #endif
798 	shmseg->u.shm_ctime = time_second;
799 	shm_committed += btoc(size);
800 	shm_nused++;
801 	if (shmseg->u.shm_perm.mode & SHMSEG_WANTED) {
802 		/*
803 		 * Somebody else wanted this key while we were asleep.  Wake
804 		 * them up now.
805 		 */
806 		shmseg->u.shm_perm.mode &= ~SHMSEG_WANTED;
807 		wakeup(shmseg);
808 	}
809 	td->td_retval[0] = shmid;
810 	return (0);
811 }
812 
813 int
814 shmget(td, uap)
815 	struct thread *td;
816 	struct shmget_args *uap;
817 {
818 	int segnum, mode;
819 	int error;
820 
821 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
822 		return (ENOSYS);
823 	mtx_lock(&Giant);
824 	mode = uap->shmflg & ACCESSPERMS;
825 	if (uap->key != IPC_PRIVATE) {
826 	again:
827 		segnum = shm_find_segment_by_key(uap->key);
828 		if (segnum >= 0) {
829 			error = shmget_existing(td, uap, mode, segnum);
830 			if (error == EAGAIN)
831 				goto again;
832 			goto done2;
833 		}
834 		if ((uap->shmflg & IPC_CREAT) == 0) {
835 			error = ENOENT;
836 			goto done2;
837 		}
838 	}
839 	error = shmget_allocate_segment(td, uap, mode);
840 done2:
841 	mtx_unlock(&Giant);
842 	return (error);
843 }
844 
845 int
846 shmsys(td, uap)
847 	struct thread *td;
848 	/* XXX actually varargs. */
849 	struct shmsys_args /* {
850 		int	which;
851 		int	a2;
852 		int	a3;
853 		int	a4;
854 	} */ *uap;
855 {
856 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
857 	int error;
858 
859 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
860 		return (ENOSYS);
861 	if (uap->which < 0 ||
862 	    uap->which >= sizeof(shmcalls)/sizeof(shmcalls[0]))
863 		return (EINVAL);
864 	mtx_lock(&Giant);
865 	error = (*shmcalls[uap->which])(td, &uap->a2);
866 	mtx_unlock(&Giant);
867 	return (error);
868 #else
869 	return (nosys(td, NULL));
870 #endif
871 }
872 
873 static void
874 shmfork_myhook(p1, p2)
875 	struct proc *p1, *p2;
876 {
877 	struct shmmap_state *shmmap_s;
878 	size_t size;
879 	int i;
880 
881 	mtx_lock(&Giant);
882 	size = shminfo.shmseg * sizeof(struct shmmap_state);
883 	shmmap_s = malloc(size, M_SHM, M_WAITOK);
884 	bcopy(p1->p_vmspace->vm_shm, shmmap_s, size);
885 	p2->p_vmspace->vm_shm = shmmap_s;
886 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
887 		if (shmmap_s->shmid != -1)
888 			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++;
889 	mtx_unlock(&Giant);
890 }
891 
892 static void
893 shmexit_myhook(struct vmspace *vm)
894 {
895 	struct shmmap_state *base, *shm;
896 	int i;
897 
898 	if ((base = vm->vm_shm) != NULL) {
899 		vm->vm_shm = NULL;
900 		mtx_lock(&Giant);
901 		for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
902 			if (shm->shmid != -1)
903 				shm_delete_mapping(vm, shm);
904 		}
905 		mtx_unlock(&Giant);
906 		free(base, M_SHM);
907 	}
908 }
909 
910 static void
911 shmrealloc(void)
912 {
913 	int i;
914 	struct shmid_kernel *newsegs;
915 
916 	if (shmalloced >= shminfo.shmmni)
917 		return;
918 
919 	newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
920 	if (newsegs == NULL)
921 		return;
922 	for (i = 0; i < shmalloced; i++)
923 		bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
924 	for (; i < shminfo.shmmni; i++) {
925 		shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
926 		shmsegs[i].u.shm_perm.seq = 0;
927 #ifdef MAC
928 		mac_sysvshm_init(&shmsegs[i]);
929 #endif
930 	}
931 	free(shmsegs, M_SHM);
932 	shmsegs = newsegs;
933 	shmalloced = shminfo.shmmni;
934 }
935 
936 static void
937 shminit()
938 {
939 	int i;
940 
941 	TUNABLE_ULONG_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall);
942 	for (i = PAGE_SIZE; i > 0; i--) {
943 		shminfo.shmmax = shminfo.shmall * i;
944 		if (shminfo.shmmax >= shminfo.shmall)
945 			break;
946 	}
947 	TUNABLE_ULONG_FETCH("kern.ipc.shmmin", &shminfo.shmmin);
948 	TUNABLE_ULONG_FETCH("kern.ipc.shmmni", &shminfo.shmmni);
949 	TUNABLE_ULONG_FETCH("kern.ipc.shmseg", &shminfo.shmseg);
950 	TUNABLE_INT_FETCH("kern.ipc.shm_use_phys", &shm_use_phys);
951 
952 	shmalloced = shminfo.shmmni;
953 	shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
954 	if (shmsegs == NULL)
955 		panic("cannot allocate initial memory for sysvshm");
956 	for (i = 0; i < shmalloced; i++) {
957 		shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
958 		shmsegs[i].u.shm_perm.seq = 0;
959 #ifdef MAC
960 		mac_sysvshm_init(&shmsegs[i]);
961 #endif
962 	}
963 	shm_last_free = 0;
964 	shm_nused = 0;
965 	shm_committed = 0;
966 	shmexit_hook = &shmexit_myhook;
967 	shmfork_hook = &shmfork_myhook;
968 }
969 
970 static int
971 shmunload()
972 {
973 #ifdef MAC
974 	int i;
975 #endif
976 
977 	if (shm_nused > 0)
978 		return (EBUSY);
979 
980 #ifdef MAC
981 	for (i = 0; i < shmalloced; i++)
982 		mac_sysvshm_destroy(&shmsegs[i]);
983 #endif
984 	free(shmsegs, M_SHM);
985 	shmexit_hook = NULL;
986 	shmfork_hook = NULL;
987 	return (0);
988 }
989 
990 static int
991 sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
992 {
993 
994 	return (SYSCTL_OUT(req, shmsegs, shmalloced * sizeof(shmsegs[0])));
995 }
996 
997 static int
998 sysvshm_modload(struct module *module, int cmd, void *arg)
999 {
1000 	int error = 0;
1001 
1002 	switch (cmd) {
1003 	case MOD_LOAD:
1004 		shminit();
1005 		break;
1006 	case MOD_UNLOAD:
1007 		error = shmunload();
1008 		break;
1009 	case MOD_SHUTDOWN:
1010 		break;
1011 	default:
1012 		error = EINVAL;
1013 		break;
1014 	}
1015 	return (error);
1016 }
1017 
1018 static moduledata_t sysvshm_mod = {
1019 	"sysvshm",
1020 	&sysvshm_modload,
1021 	NULL
1022 };
1023 
1024 SYSCALL_MODULE_HELPER(shmsys);
1025 SYSCALL_MODULE_HELPER(shmat);
1026 SYSCALL_MODULE_HELPER(shmctl);
1027 SYSCALL_MODULE_HELPER(shmdt);
1028 SYSCALL_MODULE_HELPER(shmget);
1029 
1030 DECLARE_MODULE(sysvshm, sysvshm_mod,
1031 	SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
1032 MODULE_VERSION(sysvshm, 1);
1033