xref: /freebsd/sys/kern/sysv_shm.c (revision 1e413cf93298b5b97441a21d9a50fdcd0ee9945e)
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 #include "opt_mac.h"
68 
69 #include <sys/param.h>
70 #include <sys/systm.h>
71 #include <sys/kernel.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 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
101 struct oshmctl_args;
102 static int oshmctl(struct thread *td, struct oshmctl_args *uap);
103 #endif
104 
105 static int shmget_allocate_segment(struct thread *td,
106     struct shmget_args *uap, int mode);
107 static int shmget_existing(struct thread *td, struct shmget_args *uap,
108     int mode, int segnum);
109 
110 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
111 /* XXX casting to (sy_call_t *) is bogus, as usual. */
112 static sy_call_t *shmcalls[] = {
113 	(sy_call_t *)shmat, (sy_call_t *)oshmctl,
114 	(sy_call_t *)shmdt, (sy_call_t *)shmget,
115 	(sy_call_t *)shmctl
116 };
117 #endif
118 
119 #define	SHMSEG_FREE     	0x0200
120 #define	SHMSEG_REMOVED  	0x0400
121 #define	SHMSEG_ALLOCATED	0x0800
122 #define	SHMSEG_WANTED		0x1000
123 
124 static int shm_last_free, shm_nused, shm_committed, shmalloced;
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 	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->u.shm_segsz);
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 	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->u.shm_segsz);
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 (!jail_sysvipc_allowed && jailed(td->td_ucred))
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 (!jail_sysvipc_allowed && jailed(td->td_ucred))
360 		return (ENOSYS);
361 	mtx_lock(&Giant);
362 	shmmap_s = p->p_vmspace->vm_shm;
363 	if (shmmap_s == NULL) {
364 		size = shminfo.shmseg * sizeof(struct shmmap_state);
365 		shmmap_s = malloc(size, 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->u.shm_segsz);
394 #ifdef VM_PROT_READ_IS_EXEC
395 	prot = VM_PROT_READ | VM_PROT_EXECUTE;
396 #else
397 	prot = VM_PROT_READ;
398 #endif
399 	if ((shmflg & SHM_RDONLY) == 0)
400 		prot |= VM_PROT_WRITE;
401 	flags = MAP_ANON | MAP_SHARED;
402 	if (shmaddr) {
403 		flags |= MAP_FIXED;
404 		if (shmflg & SHM_RND) {
405 			attach_va = (vm_offset_t)shmaddr & ~(SHMLBA-1);
406 		} else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0) {
407 			attach_va = (vm_offset_t)shmaddr;
408 		} else {
409 			error = EINVAL;
410 			goto done2;
411 		}
412 	} else {
413 		/*
414 		 * This is just a hint to vm_map_find() about where to
415 		 * put it.
416 		 */
417 		PROC_LOCK(p);
418 		attach_va = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
419 		    lim_max(p, RLIMIT_DATA));
420 		PROC_UNLOCK(p);
421 	}
422 
423 	vm_object_reference(shmseg->u.shm_internal);
424 	rv = vm_map_find(&p->p_vmspace->vm_map, shmseg->u.shm_internal,
425 		0, &attach_va, size, (flags & MAP_FIXED)?0:1, prot, prot, 0);
426 	if (rv != KERN_SUCCESS) {
427 		vm_object_deallocate(shmseg->u.shm_internal);
428 		error = ENOMEM;
429 		goto done2;
430 	}
431 	vm_map_inherit(&p->p_vmspace->vm_map,
432 		attach_va, attach_va + size, VM_INHERIT_SHARE);
433 
434 	shmmap_s->va = attach_va;
435 	shmmap_s->shmid = shmid;
436 	shmseg->u.shm_lpid = p->p_pid;
437 	shmseg->u.shm_atime = time_second;
438 	shmseg->u.shm_nattch++;
439 	td->td_retval[0] = attach_va;
440 done2:
441 	mtx_unlock(&Giant);
442 	return (error);
443 }
444 
445 int
446 shmat(td, uap)
447 	struct thread *td;
448 	struct shmat_args *uap;
449 {
450 	return kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg);
451 }
452 
453 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
454 struct oshmid_ds {
455 	struct	ipc_perm shm_perm;	/* operation perms */
456 	int	shm_segsz;		/* size of segment (bytes) */
457 	u_short	shm_cpid;		/* pid, creator */
458 	u_short	shm_lpid;		/* pid, last operation */
459 	short	shm_nattch;		/* no. of current attaches */
460 	time_t	shm_atime;		/* last attach time */
461 	time_t	shm_dtime;		/* last detach time */
462 	time_t	shm_ctime;		/* last change time */
463 	void	*shm_handle;		/* internal handle for shm segment */
464 };
465 
466 struct oshmctl_args {
467 	int shmid;
468 	int cmd;
469 	struct oshmid_ds *ubuf;
470 };
471 static int
472 oshmctl(td, uap)
473 	struct thread *td;
474 	struct oshmctl_args *uap;
475 {
476 #ifdef COMPAT_43
477 	int error = 0;
478 	struct shmid_kernel *shmseg;
479 	struct oshmid_ds outbuf;
480 
481 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
482 		return (ENOSYS);
483 	mtx_lock(&Giant);
484 	shmseg = shm_find_segment_by_shmid(uap->shmid);
485 	if (shmseg == NULL) {
486 		error = EINVAL;
487 		goto done2;
488 	}
489 	switch (uap->cmd) {
490 	case IPC_STAT:
491 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
492 		if (error)
493 			goto done2;
494 #ifdef MAC
495 		error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, uap->cmd);
496 		if (error != 0)
497 			goto done2;
498 #endif
499 		outbuf.shm_perm = shmseg->u.shm_perm;
500 		outbuf.shm_segsz = shmseg->u.shm_segsz;
501 		outbuf.shm_cpid = shmseg->u.shm_cpid;
502 		outbuf.shm_lpid = shmseg->u.shm_lpid;
503 		outbuf.shm_nattch = shmseg->u.shm_nattch;
504 		outbuf.shm_atime = shmseg->u.shm_atime;
505 		outbuf.shm_dtime = shmseg->u.shm_dtime;
506 		outbuf.shm_ctime = shmseg->u.shm_ctime;
507 		outbuf.shm_handle = shmseg->u.shm_internal;
508 		error = copyout(&outbuf, uap->ubuf, sizeof(outbuf));
509 		if (error)
510 			goto done2;
511 		break;
512 	default:
513 		error = shmctl(td, (struct shmctl_args *)uap);
514 		break;
515 	}
516 done2:
517 	mtx_unlock(&Giant);
518 	return (error);
519 #else
520 	return (EINVAL);
521 #endif
522 }
523 #endif
524 
525 #ifndef _SYS_SYSPROTO_H_
526 struct shmctl_args {
527 	int shmid;
528 	int cmd;
529 	struct shmid_ds *buf;
530 };
531 #endif
532 int
533 kern_shmctl(td, shmid, cmd, buf, bufsz)
534 	struct thread *td;
535 	int shmid;
536 	int cmd;
537 	void *buf;
538 	size_t *bufsz;
539 {
540 	int error = 0;
541 	struct shmid_kernel *shmseg;
542 
543 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
544 		return (ENOSYS);
545 
546 	mtx_lock(&Giant);
547 	switch (cmd) {
548 	case IPC_INFO:
549 		memcpy(buf, &shminfo, sizeof(shminfo));
550 		if (bufsz)
551 			*bufsz = sizeof(shminfo);
552 		td->td_retval[0] = shmalloced;
553 		goto done2;
554 	case SHM_INFO: {
555 		struct shm_info shm_info;
556 		shm_info.used_ids = shm_nused;
557 		shm_info.shm_rss = 0;	/*XXX where to get from ? */
558 		shm_info.shm_tot = 0;	/*XXX where to get from ? */
559 		shm_info.shm_swp = 0;	/*XXX where to get from ? */
560 		shm_info.swap_attempts = 0;	/*XXX where to get from ? */
561 		shm_info.swap_successes = 0;	/*XXX where to get from ? */
562 		memcpy(buf, &shm_info, sizeof(shm_info));
563 		if (bufsz)
564 			*bufsz = sizeof(shm_info);
565 		td->td_retval[0] = shmalloced;
566 		goto done2;
567 	}
568 	}
569 	if (cmd == SHM_STAT)
570 		shmseg = shm_find_segment_by_shmidx(shmid);
571 	else
572 		shmseg = shm_find_segment_by_shmid(shmid);
573 	if (shmseg == NULL) {
574 		error = EINVAL;
575 		goto done2;
576 	}
577 #ifdef MAC
578 	error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, cmd);
579 	if (error != 0)
580 		goto done2;
581 #endif
582 	switch (cmd) {
583 	case SHM_STAT:
584 	case IPC_STAT:
585 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
586 		if (error)
587 			goto done2;
588 		memcpy(buf, &shmseg->u, sizeof(struct shmid_ds));
589 		if (bufsz)
590 			*bufsz = sizeof(struct shmid_ds);
591 		if (cmd == SHM_STAT)
592 			td->td_retval[0] = IXSEQ_TO_IPCID(shmid, shmseg->u.shm_perm);
593 		break;
594 	case IPC_SET: {
595 		struct shmid_ds *shmid;
596 
597 		shmid = (struct shmid_ds *)buf;
598 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
599 		if (error)
600 			goto done2;
601 		shmseg->u.shm_perm.uid = shmid->shm_perm.uid;
602 		shmseg->u.shm_perm.gid = shmid->shm_perm.gid;
603 		shmseg->u.shm_perm.mode =
604 		    (shmseg->u.shm_perm.mode & ~ACCESSPERMS) |
605 		    (shmid->shm_perm.mode & ACCESSPERMS);
606 		shmseg->u.shm_ctime = time_second;
607 		break;
608 	}
609 	case IPC_RMID:
610 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
611 		if (error)
612 			goto done2;
613 		shmseg->u.shm_perm.key = IPC_PRIVATE;
614 		shmseg->u.shm_perm.mode |= SHMSEG_REMOVED;
615 		if (shmseg->u.shm_nattch <= 0) {
616 			shm_deallocate_segment(shmseg);
617 			shm_last_free = IPCID_TO_IX(shmid);
618 		}
619 		break;
620 #if 0
621 	case SHM_LOCK:
622 	case SHM_UNLOCK:
623 #endif
624 	default:
625 		error = EINVAL;
626 		break;
627 	}
628 done2:
629 	mtx_unlock(&Giant);
630 	return (error);
631 }
632 
633 int
634 shmctl(td, uap)
635 	struct thread *td;
636 	struct shmctl_args *uap;
637 {
638 	int error = 0;
639 	struct shmid_ds buf;
640 	size_t bufsz;
641 
642 	/* IPC_SET needs to copyin the buffer before calling kern_shmctl */
643 	if (uap->cmd == IPC_SET) {
644 		if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds))))
645 			goto done;
646 	}
647 
648 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
649 	if (error)
650 		goto done;
651 
652 	/* Cases in which we need to copyout */
653 	switch (uap->cmd) {
654 	case IPC_INFO:
655 	case SHM_INFO:
656 	case SHM_STAT:
657 	case IPC_STAT:
658 		error = copyout(&buf, uap->buf, bufsz);
659 		break;
660 	}
661 
662 done:
663 	if (error) {
664 		/* Invalidate the return value */
665 		td->td_retval[0] = -1;
666 	}
667 	return (error);
668 }
669 
670 
671 #ifndef _SYS_SYSPROTO_H_
672 struct shmget_args {
673 	key_t key;
674 	size_t size;
675 	int shmflg;
676 };
677 #endif
678 static int
679 shmget_existing(td, uap, mode, segnum)
680 	struct thread *td;
681 	struct shmget_args *uap;
682 	int mode;
683 	int segnum;
684 {
685 	struct shmid_kernel *shmseg;
686 	int error;
687 
688 	shmseg = &shmsegs[segnum];
689 	if (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) {
690 		/*
691 		 * This segment is in the process of being allocated.  Wait
692 		 * until it's done, and look the key up again (in case the
693 		 * allocation failed or it was freed).
694 		 */
695 		shmseg->u.shm_perm.mode |= SHMSEG_WANTED;
696 		error = tsleep(shmseg, PLOCK | PCATCH, "shmget", 0);
697 		if (error)
698 			return (error);
699 		return (EAGAIN);
700 	}
701 	if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
702 		return (EEXIST);
703 #ifdef MAC
704 	error = mac_sysvshm_check_shmget(td->td_ucred, shmseg, uap->shmflg);
705 	if (error != 0)
706 		return (error);
707 #endif
708 	if (uap->size && uap->size > shmseg->u.shm_segsz)
709 		return (EINVAL);
710 	td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
711 	return (0);
712 }
713 
714 static int
715 shmget_allocate_segment(td, uap, mode)
716 	struct thread *td;
717 	struct shmget_args *uap;
718 	int mode;
719 {
720 	int i, segnum, shmid, size;
721 	struct ucred *cred = td->td_ucred;
722 	struct shmid_kernel *shmseg;
723 	vm_object_t shm_object;
724 
725 	GIANT_REQUIRED;
726 
727 	if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
728 		return (EINVAL);
729 	if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
730 		return (ENOSPC);
731 	size = round_page(uap->size);
732 	if (shm_committed + btoc(size) > shminfo.shmall)
733 		return (ENOMEM);
734 	if (shm_last_free < 0) {
735 		shmrealloc();	/* Maybe expand the shmsegs[] array. */
736 		for (i = 0; i < shmalloced; i++)
737 			if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE)
738 				break;
739 		if (i == shmalloced)
740 			return (ENOSPC);
741 		segnum = i;
742 	} else  {
743 		segnum = shm_last_free;
744 		shm_last_free = -1;
745 	}
746 	shmseg = &shmsegs[segnum];
747 	/*
748 	 * In case we sleep in malloc(), mark the segment present but deleted
749 	 * so that noone else tries to create the same key.
750 	 */
751 	shmseg->u.shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
752 	shmseg->u.shm_perm.key = uap->key;
753 	shmseg->u.shm_perm.seq = (shmseg->u.shm_perm.seq + 1) & 0x7fff;
754 	shmid = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
755 
756 	/*
757 	 * We make sure that we have allocated a pager before we need
758 	 * to.
759 	 */
760 	if (shm_use_phys) {
761 		shm_object =
762 		    vm_pager_allocate(OBJT_PHYS, 0, size, VM_PROT_DEFAULT, 0);
763 	} else {
764 		shm_object =
765 		    vm_pager_allocate(OBJT_SWAP, 0, size, VM_PROT_DEFAULT, 0);
766 	}
767 	VM_OBJECT_LOCK(shm_object);
768 	vm_object_clear_flag(shm_object, OBJ_ONEMAPPING);
769 	vm_object_set_flag(shm_object, OBJ_NOSPLIT);
770 	VM_OBJECT_UNLOCK(shm_object);
771 
772 	shmseg->u.shm_internal = shm_object;
773 	shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = cred->cr_uid;
774 	shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid;
775 	shmseg->u.shm_perm.mode = (shmseg->u.shm_perm.mode & SHMSEG_WANTED) |
776 	    (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
777 	shmseg->u.shm_segsz = uap->size;
778 	shmseg->u.shm_cpid = td->td_proc->p_pid;
779 	shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0;
780 	shmseg->u.shm_atime = shmseg->u.shm_dtime = 0;
781 #ifdef MAC
782 	mac_sysvshm_create(cred, shmseg);
783 #endif
784 	shmseg->u.shm_ctime = time_second;
785 	shm_committed += btoc(size);
786 	shm_nused++;
787 	if (shmseg->u.shm_perm.mode & SHMSEG_WANTED) {
788 		/*
789 		 * Somebody else wanted this key while we were asleep.  Wake
790 		 * them up now.
791 		 */
792 		shmseg->u.shm_perm.mode &= ~SHMSEG_WANTED;
793 		wakeup(shmseg);
794 	}
795 	td->td_retval[0] = shmid;
796 	return (0);
797 }
798 
799 int
800 shmget(td, uap)
801 	struct thread *td;
802 	struct shmget_args *uap;
803 {
804 	int segnum, mode;
805 	int error;
806 
807 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
808 		return (ENOSYS);
809 	mtx_lock(&Giant);
810 	mode = uap->shmflg & ACCESSPERMS;
811 	if (uap->key != IPC_PRIVATE) {
812 	again:
813 		segnum = shm_find_segment_by_key(uap->key);
814 		if (segnum >= 0) {
815 			error = shmget_existing(td, uap, mode, segnum);
816 			if (error == EAGAIN)
817 				goto again;
818 			goto done2;
819 		}
820 		if ((uap->shmflg & IPC_CREAT) == 0) {
821 			error = ENOENT;
822 			goto done2;
823 		}
824 	}
825 	error = shmget_allocate_segment(td, uap, mode);
826 done2:
827 	mtx_unlock(&Giant);
828 	return (error);
829 }
830 
831 int
832 shmsys(td, uap)
833 	struct thread *td;
834 	/* XXX actually varargs. */
835 	struct shmsys_args /* {
836 		int	which;
837 		int	a2;
838 		int	a3;
839 		int	a4;
840 	} */ *uap;
841 {
842 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
843 	int error;
844 
845 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
846 		return (ENOSYS);
847 	if (uap->which < 0 ||
848 	    uap->which >= sizeof(shmcalls)/sizeof(shmcalls[0]))
849 		return (EINVAL);
850 	mtx_lock(&Giant);
851 	error = (*shmcalls[uap->which])(td, &uap->a2);
852 	mtx_unlock(&Giant);
853 	return (error);
854 #else
855 	return (nosys(td, NULL));
856 #endif
857 }
858 
859 static void
860 shmfork_myhook(p1, p2)
861 	struct proc *p1, *p2;
862 {
863 	struct shmmap_state *shmmap_s;
864 	size_t size;
865 	int i;
866 
867 	mtx_lock(&Giant);
868 	size = shminfo.shmseg * sizeof(struct shmmap_state);
869 	shmmap_s = malloc(size, M_SHM, M_WAITOK);
870 	bcopy(p1->p_vmspace->vm_shm, shmmap_s, size);
871 	p2->p_vmspace->vm_shm = shmmap_s;
872 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
873 		if (shmmap_s->shmid != -1)
874 			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++;
875 	mtx_unlock(&Giant);
876 }
877 
878 static void
879 shmexit_myhook(struct vmspace *vm)
880 {
881 	struct shmmap_state *base, *shm;
882 	int i;
883 
884 	if ((base = vm->vm_shm) != NULL) {
885 		vm->vm_shm = NULL;
886 		mtx_lock(&Giant);
887 		for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
888 			if (shm->shmid != -1)
889 				shm_delete_mapping(vm, shm);
890 		}
891 		mtx_unlock(&Giant);
892 		free(base, M_SHM);
893 	}
894 }
895 
896 static void
897 shmrealloc(void)
898 {
899 	int i;
900 	struct shmid_kernel *newsegs;
901 
902 	if (shmalloced >= shminfo.shmmni)
903 		return;
904 
905 	newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
906 	if (newsegs == NULL)
907 		return;
908 	for (i = 0; i < shmalloced; i++)
909 		bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
910 	for (; i < shminfo.shmmni; i++) {
911 		shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
912 		shmsegs[i].u.shm_perm.seq = 0;
913 #ifdef MAC
914 		mac_sysvshm_init(&shmsegs[i]);
915 #endif
916 	}
917 	free(shmsegs, M_SHM);
918 	shmsegs = newsegs;
919 	shmalloced = shminfo.shmmni;
920 }
921 
922 static void
923 shminit()
924 {
925 	int i;
926 
927 	TUNABLE_ULONG_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall);
928 	for (i = PAGE_SIZE; i > 0; i--) {
929 		shminfo.shmmax = shminfo.shmall * i;
930 		if (shminfo.shmmax >= shminfo.shmall)
931 			break;
932 	}
933 	TUNABLE_ULONG_FETCH("kern.ipc.shmmin", &shminfo.shmmin);
934 	TUNABLE_ULONG_FETCH("kern.ipc.shmmni", &shminfo.shmmni);
935 	TUNABLE_ULONG_FETCH("kern.ipc.shmseg", &shminfo.shmseg);
936 	TUNABLE_INT_FETCH("kern.ipc.shm_use_phys", &shm_use_phys);
937 
938 	shmalloced = shminfo.shmmni;
939 	shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
940 	if (shmsegs == NULL)
941 		panic("cannot allocate initial memory for sysvshm");
942 	for (i = 0; i < shmalloced; i++) {
943 		shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
944 		shmsegs[i].u.shm_perm.seq = 0;
945 #ifdef MAC
946 		mac_sysvshm_init(&shmsegs[i]);
947 #endif
948 	}
949 	shm_last_free = 0;
950 	shm_nused = 0;
951 	shm_committed = 0;
952 	shmexit_hook = &shmexit_myhook;
953 	shmfork_hook = &shmfork_myhook;
954 }
955 
956 static int
957 shmunload()
958 {
959 #ifdef MAC
960 	int i;
961 #endif
962 
963 	if (shm_nused > 0)
964 		return (EBUSY);
965 
966 #ifdef MAC
967 	for (i = 0; i < shmalloced; i++)
968 		mac_sysvshm_destroy(&shmsegs[i]);
969 #endif
970 	free(shmsegs, M_SHM);
971 	shmexit_hook = NULL;
972 	shmfork_hook = NULL;
973 	return (0);
974 }
975 
976 static int
977 sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
978 {
979 
980 	return (SYSCTL_OUT(req, shmsegs, shmalloced * sizeof(shmsegs[0])));
981 }
982 
983 static int
984 sysvshm_modload(struct module *module, int cmd, void *arg)
985 {
986 	int error = 0;
987 
988 	switch (cmd) {
989 	case MOD_LOAD:
990 		shminit();
991 		break;
992 	case MOD_UNLOAD:
993 		error = shmunload();
994 		break;
995 	case MOD_SHUTDOWN:
996 		break;
997 	default:
998 		error = EINVAL;
999 		break;
1000 	}
1001 	return (error);
1002 }
1003 
1004 static moduledata_t sysvshm_mod = {
1005 	"sysvshm",
1006 	&sysvshm_modload,
1007 	NULL
1008 };
1009 
1010 SYSCALL_MODULE_HELPER(shmsys);
1011 SYSCALL_MODULE_HELPER(shmat);
1012 SYSCALL_MODULE_HELPER(shmctl);
1013 SYSCALL_MODULE_HELPER(shmdt);
1014 SYSCALL_MODULE_HELPER(shmget);
1015 
1016 DECLARE_MODULE(sysvshm, sysvshm_mod,
1017 	SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
1018 MODULE_VERSION(sysvshm, 1);
1019