xref: /freebsd/sys/kern/sysv_shm.c (revision 6b806d21d144c25f4fad714e1c0cf780f5e27d7e)
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 #include <sys/mac.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 #ifdef MAC_DEBUG
98 #define MPRINTF(a)      printf a
99 #else
100 #define MPRINTF(a)
101 #endif
102 
103 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
104 
105 struct oshmctl_args;
106 static int oshmctl(struct thread *td, struct oshmctl_args *uap);
107 
108 static int shmget_allocate_segment(struct thread *td,
109     struct shmget_args *uap, int mode);
110 static int shmget_existing(struct thread *td, struct shmget_args *uap,
111     int mode, int segnum);
112 
113 /* XXX casting to (sy_call_t *) is bogus, as usual. */
114 static sy_call_t *shmcalls[] = {
115 	(sy_call_t *)shmat, (sy_call_t *)oshmctl,
116 	(sy_call_t *)shmdt, (sy_call_t *)shmget,
117 	(sy_call_t *)shmctl
118 };
119 
120 #define	SHMSEG_FREE     	0x0200
121 #define	SHMSEG_REMOVED  	0x0400
122 #define	SHMSEG_ALLOCATED	0x0800
123 #define	SHMSEG_WANTED		0x1000
124 
125 static int shm_last_free, shm_nused, shm_committed, shmalloced;
126 static struct shmid_kernel	*shmsegs;
127 
128 struct shmmap_state {
129 	vm_offset_t va;
130 	int shmid;
131 };
132 
133 static void shm_deallocate_segment(struct shmid_kernel *);
134 static int shm_find_segment_by_key(key_t);
135 static struct shmid_kernel *shm_find_segment_by_shmid(int);
136 static struct shmid_kernel *shm_find_segment_by_shmidx(int);
137 static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *);
138 static void shmrealloc(void);
139 static void shminit(void);
140 static int sysvshm_modload(struct module *, int, void *);
141 static int shmunload(void);
142 static void shmexit_myhook(struct vmspace *vm);
143 static void shmfork_myhook(struct proc *p1, struct proc *p2);
144 static int sysctl_shmsegs(SYSCTL_HANDLER_ARGS);
145 
146 /*
147  * Tuneable values.
148  */
149 #ifndef SHMMAXPGS
150 #define	SHMMAXPGS	8192	/* Note: sysv shared memory is swap backed. */
151 #endif
152 #ifndef SHMMAX
153 #define	SHMMAX	(SHMMAXPGS*PAGE_SIZE)
154 #endif
155 #ifndef SHMMIN
156 #define	SHMMIN	1
157 #endif
158 #ifndef SHMMNI
159 #define	SHMMNI	192
160 #endif
161 #ifndef SHMSEG
162 #define	SHMSEG	128
163 #endif
164 #ifndef SHMALL
165 #define	SHMALL	(SHMMAXPGS)
166 #endif
167 
168 struct	shminfo shminfo = {
169 	SHMMAX,
170 	SHMMIN,
171 	SHMMNI,
172 	SHMSEG,
173 	SHMALL
174 };
175 
176 static int shm_use_phys;
177 static int shm_allow_removed;
178 
179 SYSCTL_DECL(_kern_ipc);
180 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0, "");
181 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0, "");
182 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RDTUN, &shminfo.shmmni, 0, "");
183 SYSCTL_INT(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RDTUN, &shminfo.shmseg, 0, "");
184 SYSCTL_INT(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0, "");
185 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW,
186     &shm_use_phys, 0, "");
187 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RW,
188     &shm_allow_removed, 0, "");
189 SYSCTL_PROC(_kern_ipc, OID_AUTO, shmsegs, CTLFLAG_RD,
190     NULL, 0, sysctl_shmsegs, "", "");
191 
192 static int
193 shm_find_segment_by_key(key)
194 	key_t key;
195 {
196 	int i;
197 
198 	for (i = 0; i < shmalloced; i++)
199 		if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) &&
200 		    shmsegs[i].u.shm_perm.key == key)
201 			return (i);
202 	return (-1);
203 }
204 
205 static struct shmid_kernel *
206 shm_find_segment_by_shmid(int shmid)
207 {
208 	int segnum;
209 	struct shmid_kernel *shmseg;
210 
211 	segnum = IPCID_TO_IX(shmid);
212 	if (segnum < 0 || segnum >= shmalloced)
213 		return (NULL);
214 	shmseg = &shmsegs[segnum];
215 	if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
216 	    (!shm_allow_removed &&
217 	     (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0) ||
218 	    shmseg->u.shm_perm.seq != IPCID_TO_SEQ(shmid))
219 		return (NULL);
220 	return (shmseg);
221 }
222 
223 static struct shmid_kernel *
224 shm_find_segment_by_shmidx(int segnum)
225 {
226 	struct shmid_kernel *shmseg;
227 
228 	if (segnum < 0 || segnum >= shmalloced)
229 		return (NULL);
230 	shmseg = &shmsegs[segnum];
231 	if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
232 	    (!shm_allow_removed &&
233 	     (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0))
234 		return (NULL);
235 	return (shmseg);
236 }
237 
238 static void
239 shm_deallocate_segment(shmseg)
240 	struct shmid_kernel *shmseg;
241 {
242 	size_t size;
243 
244 	GIANT_REQUIRED;
245 
246 	vm_object_deallocate(shmseg->u.shm_internal);
247 	shmseg->u.shm_internal = NULL;
248 	size = round_page(shmseg->u.shm_segsz);
249 	shm_committed -= btoc(size);
250 	shm_nused--;
251 	shmseg->u.shm_perm.mode = SHMSEG_FREE;
252 #ifdef MAC
253 	mac_cleanup_sysv_shm(shmseg);
254 #endif
255 }
256 
257 static int
258 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
259 {
260 	struct shmid_kernel *shmseg;
261 	int segnum, result;
262 	size_t size;
263 
264 	GIANT_REQUIRED;
265 
266 	segnum = IPCID_TO_IX(shmmap_s->shmid);
267 	shmseg = &shmsegs[segnum];
268 	size = round_page(shmseg->u.shm_segsz);
269 	result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
270 	if (result != KERN_SUCCESS)
271 		return (EINVAL);
272 	shmmap_s->shmid = -1;
273 	shmseg->u.shm_dtime = time_second;
274 	if ((--shmseg->u.shm_nattch <= 0) &&
275 	    (shmseg->u.shm_perm.mode & SHMSEG_REMOVED)) {
276 		shm_deallocate_segment(shmseg);
277 		shm_last_free = segnum;
278 	}
279 	return (0);
280 }
281 
282 #ifndef _SYS_SYSPROTO_H_
283 struct shmdt_args {
284 	const void *shmaddr;
285 };
286 #endif
287 
288 /*
289  * MPSAFE
290  */
291 int
292 shmdt(td, uap)
293 	struct thread *td;
294 	struct shmdt_args *uap;
295 {
296 	struct proc *p = td->td_proc;
297 	struct shmmap_state *shmmap_s;
298 #ifdef MAC
299 	struct shmid_kernel *shmsegptr;
300 #endif
301 	int i;
302 	int error = 0;
303 
304 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
305 		return (ENOSYS);
306 	mtx_lock(&Giant);
307 	shmmap_s = p->p_vmspace->vm_shm;
308  	if (shmmap_s == NULL) {
309 		error = EINVAL;
310 		goto done2;
311 	}
312 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
313 		if (shmmap_s->shmid != -1 &&
314 		    shmmap_s->va == (vm_offset_t)uap->shmaddr) {
315 			break;
316 		}
317 	}
318 	if (i == shminfo.shmseg) {
319 		error = EINVAL;
320 		goto done2;
321 	}
322 #ifdef MAC
323 	shmsegptr = &shmsegs[IPCID_TO_IX(shmmap_s->shmid)];
324 	error = mac_check_sysv_shmdt(td->td_ucred, shmsegptr);
325 	if (error != 0) {
326 		MPRINTF(("mac_check_sysv_shmdt returned %d\n", error));
327 		goto done2;
328 	}
329 #endif
330 	error = shm_delete_mapping(p->p_vmspace, shmmap_s);
331 done2:
332 	mtx_unlock(&Giant);
333 	return (error);
334 }
335 
336 #ifndef _SYS_SYSPROTO_H_
337 struct shmat_args {
338 	int shmid;
339 	const void *shmaddr;
340 	int shmflg;
341 };
342 #endif
343 
344 /*
345  * MPSAFE
346  */
347 int
348 kern_shmat(td, shmid, shmaddr, shmflg)
349 	struct thread *td;
350 	int shmid;
351 	const void *shmaddr;
352 	int shmflg;
353 {
354 	struct proc *p = td->td_proc;
355 	int i, flags;
356 	struct shmid_kernel *shmseg;
357 	struct shmmap_state *shmmap_s = NULL;
358 	vm_offset_t attach_va;
359 	vm_prot_t prot;
360 	vm_size_t size;
361 	int rv;
362 	int error = 0;
363 
364 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
365 		return (ENOSYS);
366 	mtx_lock(&Giant);
367 	shmmap_s = p->p_vmspace->vm_shm;
368 	if (shmmap_s == NULL) {
369 		size = shminfo.shmseg * sizeof(struct shmmap_state);
370 		shmmap_s = malloc(size, M_SHM, M_WAITOK);
371 		for (i = 0; i < shminfo.shmseg; i++)
372 			shmmap_s[i].shmid = -1;
373 		p->p_vmspace->vm_shm = shmmap_s;
374 	}
375 	shmseg = shm_find_segment_by_shmid(shmid);
376 	if (shmseg == NULL) {
377 		error = EINVAL;
378 		goto done2;
379 	}
380 	error = ipcperm(td, &shmseg->u.shm_perm,
381 	    (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
382 	if (error)
383 		goto done2;
384 #ifdef MAC
385 	error = mac_check_sysv_shmat(td->td_ucred, shmseg, shmflg);
386 	if (error != 0) {
387 	 	MPRINTF(("mac_check_sysv_shmat returned %d\n", error));
388 		goto done2;
389 	}
390 #endif
391 	for (i = 0; i < shminfo.shmseg; i++) {
392 		if (shmmap_s->shmid == -1)
393 			break;
394 		shmmap_s++;
395 	}
396 	if (i >= shminfo.shmseg) {
397 		error = EMFILE;
398 		goto done2;
399 	}
400 	size = round_page(shmseg->u.shm_segsz);
401 #ifdef VM_PROT_READ_IS_EXEC
402 	prot = VM_PROT_READ | VM_PROT_EXECUTE;
403 #else
404 	prot = VM_PROT_READ;
405 #endif
406 	if ((shmflg & SHM_RDONLY) == 0)
407 		prot |= VM_PROT_WRITE;
408 	flags = MAP_ANON | MAP_SHARED;
409 	if (shmaddr) {
410 		flags |= MAP_FIXED;
411 		if (shmflg & SHM_RND) {
412 			attach_va = (vm_offset_t)shmaddr & ~(SHMLBA-1);
413 		} else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0) {
414 			attach_va = (vm_offset_t)shmaddr;
415 		} else {
416 			error = EINVAL;
417 			goto done2;
418 		}
419 	} else {
420 		/*
421 		 * This is just a hint to vm_map_find() about where to
422 		 * put it.
423 		 */
424 		PROC_LOCK(p);
425 		attach_va = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
426 		    lim_max(p, RLIMIT_DATA));
427 		PROC_UNLOCK(p);
428 	}
429 
430 	vm_object_reference(shmseg->u.shm_internal);
431 	rv = vm_map_find(&p->p_vmspace->vm_map, shmseg->u.shm_internal,
432 		0, &attach_va, size, (flags & MAP_FIXED)?0:1, prot, prot, 0);
433 	if (rv != KERN_SUCCESS) {
434 		vm_object_deallocate(shmseg->u.shm_internal);
435 		error = ENOMEM;
436 		goto done2;
437 	}
438 	vm_map_inherit(&p->p_vmspace->vm_map,
439 		attach_va, attach_va + size, VM_INHERIT_SHARE);
440 
441 	shmmap_s->va = attach_va;
442 	shmmap_s->shmid = shmid;
443 	shmseg->u.shm_lpid = p->p_pid;
444 	shmseg->u.shm_atime = time_second;
445 	shmseg->u.shm_nattch++;
446 	td->td_retval[0] = attach_va;
447 done2:
448 	mtx_unlock(&Giant);
449 	return (error);
450 }
451 
452 int
453 shmat(td, uap)
454 	struct thread *td;
455 	struct shmat_args *uap;
456 {
457 	return kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg);
458 }
459 
460 struct oshmid_ds {
461 	struct	ipc_perm shm_perm;	/* operation perms */
462 	int	shm_segsz;		/* size of segment (bytes) */
463 	u_short	shm_cpid;		/* pid, creator */
464 	u_short	shm_lpid;		/* pid, last operation */
465 	short	shm_nattch;		/* no. of current attaches */
466 	time_t	shm_atime;		/* last attach time */
467 	time_t	shm_dtime;		/* last detach time */
468 	time_t	shm_ctime;		/* last change time */
469 	void	*shm_handle;		/* internal handle for shm segment */
470 };
471 
472 struct oshmctl_args {
473 	int shmid;
474 	int cmd;
475 	struct oshmid_ds *ubuf;
476 };
477 
478 /*
479  * MPSAFE
480  */
481 static int
482 oshmctl(td, uap)
483 	struct thread *td;
484 	struct oshmctl_args *uap;
485 {
486 #ifdef COMPAT_43
487 	int error = 0;
488 	struct shmid_kernel *shmseg;
489 	struct oshmid_ds outbuf;
490 
491 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
492 		return (ENOSYS);
493 	mtx_lock(&Giant);
494 	shmseg = shm_find_segment_by_shmid(uap->shmid);
495 	if (shmseg == NULL) {
496 		error = EINVAL;
497 		goto done2;
498 	}
499 	switch (uap->cmd) {
500 	case IPC_STAT:
501 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
502 		if (error)
503 			goto done2;
504 #ifdef MAC
505 		error = mac_check_sysv_shmctl(td->td_ucred, shmseg, uap->cmd);
506 		if (error != 0) {
507 			MPRINTF(("mac_check_sysv_shmctl returned %d\n",
508 			    error));
509 			goto done2;
510 		}
511 #endif
512 		outbuf.shm_perm = shmseg->u.shm_perm;
513 		outbuf.shm_segsz = shmseg->u.shm_segsz;
514 		outbuf.shm_cpid = shmseg->u.shm_cpid;
515 		outbuf.shm_lpid = shmseg->u.shm_lpid;
516 		outbuf.shm_nattch = shmseg->u.shm_nattch;
517 		outbuf.shm_atime = shmseg->u.shm_atime;
518 		outbuf.shm_dtime = shmseg->u.shm_dtime;
519 		outbuf.shm_ctime = shmseg->u.shm_ctime;
520 		outbuf.shm_handle = shmseg->u.shm_internal;
521 		error = copyout(&outbuf, uap->ubuf, sizeof(outbuf));
522 		if (error)
523 			goto done2;
524 		break;
525 	default:
526 		error = shmctl(td, (struct shmctl_args *)uap);
527 		break;
528 	}
529 done2:
530 	mtx_unlock(&Giant);
531 	return (error);
532 #else
533 	return (EINVAL);
534 #endif
535 }
536 
537 #ifndef _SYS_SYSPROTO_H_
538 struct shmctl_args {
539 	int shmid;
540 	int cmd;
541 	struct shmid_ds *buf;
542 };
543 #endif
544 
545 /*
546  * MPSAFE
547  */
548 int
549 kern_shmctl(td, shmid, cmd, buf, bufsz)
550 	struct thread *td;
551 	int shmid;
552 	int cmd;
553 	void *buf;
554 	size_t *bufsz;
555 {
556 	int error = 0;
557 	struct shmid_kernel *shmseg;
558 
559 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
560 		return (ENOSYS);
561 
562 	mtx_lock(&Giant);
563 	switch (cmd) {
564 	case IPC_INFO:
565 		memcpy(buf, &shminfo, sizeof(shminfo));
566 		if (bufsz)
567 			*bufsz = sizeof(shminfo);
568 		td->td_retval[0] = shmalloced;
569 		goto done2;
570 	case SHM_INFO: {
571 		struct shm_info shm_info;
572 		shm_info.used_ids = shm_nused;
573 		shm_info.shm_rss = 0;	/*XXX where to get from ? */
574 		shm_info.shm_tot = 0;	/*XXX where to get from ? */
575 		shm_info.shm_swp = 0;	/*XXX where to get from ? */
576 		shm_info.swap_attempts = 0;	/*XXX where to get from ? */
577 		shm_info.swap_successes = 0;	/*XXX where to get from ? */
578 		memcpy(buf, &shm_info, sizeof(shm_info));
579 		if (bufsz)
580 			*bufsz = sizeof(shm_info);
581 		td->td_retval[0] = shmalloced;
582 		goto done2;
583 	}
584 	}
585 	if (cmd == SHM_STAT)
586 		shmseg = shm_find_segment_by_shmidx(shmid);
587 	else
588 		shmseg = shm_find_segment_by_shmid(shmid);
589 	if (shmseg == NULL) {
590 		error = EINVAL;
591 		goto done2;
592 	}
593 #ifdef MAC
594 	error = mac_check_sysv_shmctl(td->td_ucred, shmseg, cmd);
595 	if (error != 0) {
596 		MPRINTF(("mac_check_sysv_shmctl returned %d\n", error));
597 		goto done2;
598 	}
599 #endif
600 	switch (cmd) {
601 	case SHM_STAT:
602 	case IPC_STAT:
603 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
604 		if (error)
605 			goto done2;
606 		memcpy(buf, &shmseg->u, sizeof(struct shmid_ds));
607 		if (bufsz)
608 			*bufsz = sizeof(struct shmid_ds);
609 		if (cmd == SHM_STAT)
610 			td->td_retval[0] = IXSEQ_TO_IPCID(shmid, shmseg->u.shm_perm);
611 		break;
612 	case IPC_SET: {
613 		struct shmid_ds *shmid;
614 
615 		shmid = (struct shmid_ds *)buf;
616 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
617 		if (error)
618 			goto done2;
619 		shmseg->u.shm_perm.uid = shmid->shm_perm.uid;
620 		shmseg->u.shm_perm.gid = shmid->shm_perm.gid;
621 		shmseg->u.shm_perm.mode =
622 		    (shmseg->u.shm_perm.mode & ~ACCESSPERMS) |
623 		    (shmid->shm_perm.mode & ACCESSPERMS);
624 		shmseg->u.shm_ctime = time_second;
625 		break;
626 	}
627 	case IPC_RMID:
628 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
629 		if (error)
630 			goto done2;
631 		shmseg->u.shm_perm.key = IPC_PRIVATE;
632 		shmseg->u.shm_perm.mode |= SHMSEG_REMOVED;
633 		if (shmseg->u.shm_nattch <= 0) {
634 			shm_deallocate_segment(shmseg);
635 			shm_last_free = IPCID_TO_IX(shmid);
636 		}
637 		break;
638 #if 0
639 	case SHM_LOCK:
640 	case SHM_UNLOCK:
641 #endif
642 	default:
643 		error = EINVAL;
644 		break;
645 	}
646 done2:
647 	mtx_unlock(&Giant);
648 	return (error);
649 }
650 
651 int
652 shmctl(td, uap)
653 	struct thread *td;
654 	struct shmctl_args *uap;
655 {
656 	int error = 0;
657 	struct shmid_ds buf;
658 	size_t bufsz;
659 
660 	/* IPC_SET needs to copyin the buffer before calling kern_shmctl */
661 	if (uap->cmd == IPC_SET) {
662 		if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds))))
663 			goto done;
664 	}
665 
666 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
667 	if (error)
668 		goto done;
669 
670 	/* Cases in which we need to copyout */
671 	switch (uap->cmd) {
672 	case IPC_INFO:
673 	case SHM_INFO:
674 	case SHM_STAT:
675 	case IPC_STAT:
676 		error = copyout(&buf, uap->buf, bufsz);
677 		break;
678 	}
679 
680 done:
681 	if (error) {
682 		/* Invalidate the return value */
683 		td->td_retval[0] = -1;
684 	}
685 	return (error);
686 }
687 
688 
689 #ifndef _SYS_SYSPROTO_H_
690 struct shmget_args {
691 	key_t key;
692 	size_t size;
693 	int shmflg;
694 };
695 #endif
696 
697 static int
698 shmget_existing(td, uap, mode, segnum)
699 	struct thread *td;
700 	struct shmget_args *uap;
701 	int mode;
702 	int segnum;
703 {
704 	struct shmid_kernel *shmseg;
705 	int error;
706 
707 	shmseg = &shmsegs[segnum];
708 	if (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) {
709 		/*
710 		 * This segment is in the process of being allocated.  Wait
711 		 * until it's done, and look the key up again (in case the
712 		 * allocation failed or it was freed).
713 		 */
714 		shmseg->u.shm_perm.mode |= SHMSEG_WANTED;
715 		error = tsleep(shmseg, PLOCK | PCATCH, "shmget", 0);
716 		if (error)
717 			return (error);
718 		return (EAGAIN);
719 	}
720 	if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
721 		return (EEXIST);
722 	error = ipcperm(td, &shmseg->u.shm_perm, mode);
723 #ifdef MAC
724 	error = mac_check_sysv_shmget(td->td_ucred, shmseg, uap->shmflg);
725 	if (error != 0)
726 		MPRINTF(("mac_check_sysv_shmget returned %d\n", error));
727 #endif
728 	if (error)
729 		return (error);
730 	if (uap->size && uap->size > shmseg->u.shm_segsz)
731 		return (EINVAL);
732 	td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
733 	return (0);
734 }
735 
736 static int
737 shmget_allocate_segment(td, uap, mode)
738 	struct thread *td;
739 	struct shmget_args *uap;
740 	int mode;
741 {
742 	int i, segnum, shmid, size;
743 	struct ucred *cred = td->td_ucred;
744 	struct shmid_kernel *shmseg;
745 	vm_object_t shm_object;
746 
747 	GIANT_REQUIRED;
748 
749 	if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
750 		return (EINVAL);
751 	if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
752 		return (ENOSPC);
753 	size = round_page(uap->size);
754 	if (shm_committed + btoc(size) > shminfo.shmall)
755 		return (ENOMEM);
756 	if (shm_last_free < 0) {
757 		shmrealloc();	/* Maybe expand the shmsegs[] array. */
758 		for (i = 0; i < shmalloced; i++)
759 			if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE)
760 				break;
761 		if (i == shmalloced)
762 			return (ENOSPC);
763 		segnum = i;
764 	} else  {
765 		segnum = shm_last_free;
766 		shm_last_free = -1;
767 	}
768 	shmseg = &shmsegs[segnum];
769 	/*
770 	 * In case we sleep in malloc(), mark the segment present but deleted
771 	 * so that noone else tries to create the same key.
772 	 */
773 	shmseg->u.shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
774 	shmseg->u.shm_perm.key = uap->key;
775 	shmseg->u.shm_perm.seq = (shmseg->u.shm_perm.seq + 1) & 0x7fff;
776 	shmid = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
777 
778 	/*
779 	 * We make sure that we have allocated a pager before we need
780 	 * to.
781 	 */
782 	if (shm_use_phys) {
783 		shm_object =
784 		    vm_pager_allocate(OBJT_PHYS, 0, size, VM_PROT_DEFAULT, 0);
785 	} else {
786 		shm_object =
787 		    vm_pager_allocate(OBJT_SWAP, 0, size, VM_PROT_DEFAULT, 0);
788 	}
789 	VM_OBJECT_LOCK(shm_object);
790 	vm_object_clear_flag(shm_object, OBJ_ONEMAPPING);
791 	vm_object_set_flag(shm_object, OBJ_NOSPLIT);
792 	VM_OBJECT_UNLOCK(shm_object);
793 
794 	shmseg->u.shm_internal = shm_object;
795 	shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = cred->cr_uid;
796 	shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid;
797 	shmseg->u.shm_perm.mode = (shmseg->u.shm_perm.mode & SHMSEG_WANTED) |
798 	    (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
799 	shmseg->u.shm_segsz = uap->size;
800 	shmseg->u.shm_cpid = td->td_proc->p_pid;
801 	shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0;
802 	shmseg->u.shm_atime = shmseg->u.shm_dtime = 0;
803 #ifdef MAC
804 	mac_create_sysv_shm(cred, shmseg);
805 #endif
806 	shmseg->u.shm_ctime = time_second;
807 	shm_committed += btoc(size);
808 	shm_nused++;
809 	if (shmseg->u.shm_perm.mode & SHMSEG_WANTED) {
810 		/*
811 		 * Somebody else wanted this key while we were asleep.  Wake
812 		 * them up now.
813 		 */
814 		shmseg->u.shm_perm.mode &= ~SHMSEG_WANTED;
815 		wakeup(shmseg);
816 	}
817 	td->td_retval[0] = shmid;
818 	return (0);
819 }
820 
821 /*
822  * MPSAFE
823  */
824 int
825 shmget(td, uap)
826 	struct thread *td;
827 	struct shmget_args *uap;
828 {
829 	int segnum, mode;
830 	int error;
831 
832 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
833 		return (ENOSYS);
834 	mtx_lock(&Giant);
835 	mode = uap->shmflg & ACCESSPERMS;
836 	if (uap->key != IPC_PRIVATE) {
837 	again:
838 		segnum = shm_find_segment_by_key(uap->key);
839 		if (segnum >= 0) {
840 			error = shmget_existing(td, uap, mode, segnum);
841 			if (error == EAGAIN)
842 				goto again;
843 			goto done2;
844 		}
845 		if ((uap->shmflg & IPC_CREAT) == 0) {
846 			error = ENOENT;
847 			goto done2;
848 		}
849 	}
850 	error = shmget_allocate_segment(td, uap, mode);
851 done2:
852 	mtx_unlock(&Giant);
853 	return (error);
854 }
855 
856 /*
857  * MPSAFE
858  */
859 int
860 shmsys(td, uap)
861 	struct thread *td;
862 	/* XXX actually varargs. */
863 	struct shmsys_args /* {
864 		int	which;
865 		int	a2;
866 		int	a3;
867 		int	a4;
868 	} */ *uap;
869 {
870 	int error;
871 
872 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
873 		return (ENOSYS);
874 	if (uap->which < 0 ||
875 	    uap->which >= sizeof(shmcalls)/sizeof(shmcalls[0]))
876 		return (EINVAL);
877 	mtx_lock(&Giant);
878 	error = (*shmcalls[uap->which])(td, &uap->a2);
879 	mtx_unlock(&Giant);
880 	return (error);
881 }
882 
883 static void
884 shmfork_myhook(p1, p2)
885 	struct proc *p1, *p2;
886 {
887 	struct shmmap_state *shmmap_s;
888 	size_t size;
889 	int i;
890 
891 	mtx_lock(&Giant);
892 	size = shminfo.shmseg * sizeof(struct shmmap_state);
893 	shmmap_s = malloc(size, M_SHM, M_WAITOK);
894 	bcopy(p1->p_vmspace->vm_shm, shmmap_s, size);
895 	p2->p_vmspace->vm_shm = shmmap_s;
896 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
897 		if (shmmap_s->shmid != -1)
898 			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++;
899 	mtx_unlock(&Giant);
900 }
901 
902 static void
903 shmexit_myhook(struct vmspace *vm)
904 {
905 	struct shmmap_state *base, *shm;
906 	int i;
907 
908 	if ((base = vm->vm_shm) != NULL) {
909 		vm->vm_shm = NULL;
910 		mtx_lock(&Giant);
911 		for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
912 			if (shm->shmid != -1)
913 				shm_delete_mapping(vm, shm);
914 		}
915 		mtx_unlock(&Giant);
916 		free(base, M_SHM);
917 	}
918 }
919 
920 static void
921 shmrealloc(void)
922 {
923 	int i;
924 	struct shmid_kernel *newsegs;
925 
926 	if (shmalloced >= shminfo.shmmni)
927 		return;
928 
929 	newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
930 	if (newsegs == NULL)
931 		return;
932 	for (i = 0; i < shmalloced; i++)
933 		bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
934 	for (; i < shminfo.shmmni; i++) {
935 		shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
936 		shmsegs[i].u.shm_perm.seq = 0;
937 #ifdef MAC
938 		mac_init_sysv_shm(&shmsegs[i]);
939 #endif
940 	}
941 	free(shmsegs, M_SHM);
942 	shmsegs = newsegs;
943 	shmalloced = shminfo.shmmni;
944 }
945 
946 static void
947 shminit()
948 {
949 	int i;
950 
951 	TUNABLE_INT_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall);
952 	for (i = PAGE_SIZE; i > 0; i--) {
953 		shminfo.shmmax = shminfo.shmall * PAGE_SIZE;
954 		if (shminfo.shmmax >= shminfo.shmall)
955 			break;
956 	}
957 	TUNABLE_INT_FETCH("kern.ipc.shmmin", &shminfo.shmmin);
958 	TUNABLE_INT_FETCH("kern.ipc.shmmni", &shminfo.shmmni);
959 	TUNABLE_INT_FETCH("kern.ipc.shmseg", &shminfo.shmseg);
960 	TUNABLE_INT_FETCH("kern.ipc.shm_use_phys", &shm_use_phys);
961 
962 	shmalloced = shminfo.shmmni;
963 	shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
964 	if (shmsegs == NULL)
965 		panic("cannot allocate initial memory for sysvshm");
966 	for (i = 0; i < shmalloced; i++) {
967 		shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
968 		shmsegs[i].u.shm_perm.seq = 0;
969 #ifdef MAC
970 		mac_init_sysv_shm(&shmsegs[i]);
971 #endif
972 	}
973 	shm_last_free = 0;
974 	shm_nused = 0;
975 	shm_committed = 0;
976 	shmexit_hook = &shmexit_myhook;
977 	shmfork_hook = &shmfork_myhook;
978 }
979 
980 static int
981 shmunload()
982 {
983 #ifdef MAC
984 	int i;
985 #endif
986 
987 	if (shm_nused > 0)
988 		return (EBUSY);
989 
990 #ifdef MAC
991 	for (i = 0; i < shmalloced; i++)
992 		mac_destroy_sysv_shm(&shmsegs[i]);
993 #endif
994 	free(shmsegs, M_SHM);
995 	shmexit_hook = NULL;
996 	shmfork_hook = NULL;
997 	return (0);
998 }
999 
1000 static int
1001 sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
1002 {
1003 
1004 	return (SYSCTL_OUT(req, shmsegs, shmalloced * sizeof(shmsegs[0])));
1005 }
1006 
1007 static int
1008 sysvshm_modload(struct module *module, int cmd, void *arg)
1009 {
1010 	int error = 0;
1011 
1012 	switch (cmd) {
1013 	case MOD_LOAD:
1014 		shminit();
1015 		break;
1016 	case MOD_UNLOAD:
1017 		error = shmunload();
1018 		break;
1019 	case MOD_SHUTDOWN:
1020 		break;
1021 	default:
1022 		error = EINVAL;
1023 		break;
1024 	}
1025 	return (error);
1026 }
1027 
1028 static moduledata_t sysvshm_mod = {
1029 	"sysvshm",
1030 	&sysvshm_modload,
1031 	NULL
1032 };
1033 
1034 SYSCALL_MODULE_HELPER(shmsys);
1035 SYSCALL_MODULE_HELPER(shmat);
1036 SYSCALL_MODULE_HELPER(shmctl);
1037 SYSCALL_MODULE_HELPER(shmdt);
1038 SYSCALL_MODULE_HELPER(shmget);
1039 
1040 DECLARE_MODULE(sysvshm, sysvshm_mod,
1041 	SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
1042 MODULE_VERSION(sysvshm, 1);
1043