xref: /freebsd/sys/kern/sysv_shm.c (revision b3a1f9373a31b644f8a65de1ba35929af3f6a9fe)
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 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
106 struct oshmctl_args;
107 static int oshmctl(struct thread *td, struct oshmctl_args *uap);
108 #endif
109 
110 static int shmget_allocate_segment(struct thread *td,
111     struct shmget_args *uap, int mode);
112 static int shmget_existing(struct thread *td, struct shmget_args *uap,
113     int mode, int segnum);
114 
115 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
116 /* XXX casting to (sy_call_t *) is bogus, as usual. */
117 static sy_call_t *shmcalls[] = {
118 	(sy_call_t *)shmat, (sy_call_t *)oshmctl,
119 	(sy_call_t *)shmdt, (sy_call_t *)shmget,
120 	(sy_call_t *)shmctl
121 };
122 #endif
123 
124 #define	SHMSEG_FREE     	0x0200
125 #define	SHMSEG_REMOVED  	0x0400
126 #define	SHMSEG_ALLOCATED	0x0800
127 #define	SHMSEG_WANTED		0x1000
128 
129 static int shm_last_free, shm_nused, shm_committed, shmalloced;
130 static struct shmid_kernel	*shmsegs;
131 
132 struct shmmap_state {
133 	vm_offset_t va;
134 	int shmid;
135 };
136 
137 static void shm_deallocate_segment(struct shmid_kernel *);
138 static int shm_find_segment_by_key(key_t);
139 static struct shmid_kernel *shm_find_segment_by_shmid(int);
140 static struct shmid_kernel *shm_find_segment_by_shmidx(int);
141 static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *);
142 static void shmrealloc(void);
143 static void shminit(void);
144 static int sysvshm_modload(struct module *, int, void *);
145 static int shmunload(void);
146 static void shmexit_myhook(struct vmspace *vm);
147 static void shmfork_myhook(struct proc *p1, struct proc *p2);
148 static int sysctl_shmsegs(SYSCTL_HANDLER_ARGS);
149 
150 /*
151  * Tuneable values.
152  */
153 #ifndef SHMMAXPGS
154 #define	SHMMAXPGS	8192	/* Note: sysv shared memory is swap backed. */
155 #endif
156 #ifndef SHMMAX
157 #define	SHMMAX	(SHMMAXPGS*PAGE_SIZE)
158 #endif
159 #ifndef SHMMIN
160 #define	SHMMIN	1
161 #endif
162 #ifndef SHMMNI
163 #define	SHMMNI	192
164 #endif
165 #ifndef SHMSEG
166 #define	SHMSEG	128
167 #endif
168 #ifndef SHMALL
169 #define	SHMALL	(SHMMAXPGS)
170 #endif
171 
172 struct	shminfo shminfo = {
173 	SHMMAX,
174 	SHMMIN,
175 	SHMMNI,
176 	SHMSEG,
177 	SHMALL
178 };
179 
180 static int shm_use_phys;
181 static int shm_allow_removed;
182 
183 SYSCTL_DECL(_kern_ipc);
184 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0,
185     "Maximum shared memory segment size");
186 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0,
187     "Minimum shared memory segment size");
188 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RDTUN, &shminfo.shmmni, 0,
189     "Number of shared memory identifiers");
190 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RDTUN, &shminfo.shmseg, 0,
191     "Number of segments per process");
192 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0,
193     "Maximum number of pages available for shared memory");
194 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW,
195     &shm_use_phys, 0, "Enable/Disable locking of shared memory pages in core");
196 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RW,
197     &shm_allow_removed, 0,
198     "Enable/Disable attachment to attached segments marked for removal");
199 SYSCTL_PROC(_kern_ipc, OID_AUTO, shmsegs, CTLFLAG_RD,
200     NULL, 0, sysctl_shmsegs, "",
201     "Current number of shared memory segments allocated");
202 
203 static int
204 shm_find_segment_by_key(key)
205 	key_t key;
206 {
207 	int i;
208 
209 	for (i = 0; i < shmalloced; i++)
210 		if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) &&
211 		    shmsegs[i].u.shm_perm.key == key)
212 			return (i);
213 	return (-1);
214 }
215 
216 static struct shmid_kernel *
217 shm_find_segment_by_shmid(int shmid)
218 {
219 	int segnum;
220 	struct shmid_kernel *shmseg;
221 
222 	segnum = IPCID_TO_IX(shmid);
223 	if (segnum < 0 || segnum >= shmalloced)
224 		return (NULL);
225 	shmseg = &shmsegs[segnum];
226 	if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
227 	    (!shm_allow_removed &&
228 	     (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0) ||
229 	    shmseg->u.shm_perm.seq != IPCID_TO_SEQ(shmid))
230 		return (NULL);
231 	return (shmseg);
232 }
233 
234 static struct shmid_kernel *
235 shm_find_segment_by_shmidx(int segnum)
236 {
237 	struct shmid_kernel *shmseg;
238 
239 	if (segnum < 0 || segnum >= shmalloced)
240 		return (NULL);
241 	shmseg = &shmsegs[segnum];
242 	if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
243 	    (!shm_allow_removed &&
244 	     (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0))
245 		return (NULL);
246 	return (shmseg);
247 }
248 
249 static void
250 shm_deallocate_segment(shmseg)
251 	struct shmid_kernel *shmseg;
252 {
253 	size_t size;
254 
255 	GIANT_REQUIRED;
256 
257 	vm_object_deallocate(shmseg->u.shm_internal);
258 	shmseg->u.shm_internal = NULL;
259 	size = round_page(shmseg->u.shm_segsz);
260 	shm_committed -= btoc(size);
261 	shm_nused--;
262 	shmseg->u.shm_perm.mode = SHMSEG_FREE;
263 #ifdef MAC
264 	mac_cleanup_sysv_shm(shmseg);
265 #endif
266 }
267 
268 static int
269 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
270 {
271 	struct shmid_kernel *shmseg;
272 	int segnum, result;
273 	size_t size;
274 
275 	GIANT_REQUIRED;
276 
277 	segnum = IPCID_TO_IX(shmmap_s->shmid);
278 	shmseg = &shmsegs[segnum];
279 	size = round_page(shmseg->u.shm_segsz);
280 	result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
281 	if (result != KERN_SUCCESS)
282 		return (EINVAL);
283 	shmmap_s->shmid = -1;
284 	shmseg->u.shm_dtime = time_second;
285 	if ((--shmseg->u.shm_nattch <= 0) &&
286 	    (shmseg->u.shm_perm.mode & SHMSEG_REMOVED)) {
287 		shm_deallocate_segment(shmseg);
288 		shm_last_free = segnum;
289 	}
290 	return (0);
291 }
292 
293 #ifndef _SYS_SYSPROTO_H_
294 struct shmdt_args {
295 	const void *shmaddr;
296 };
297 #endif
298 
299 /*
300  * MPSAFE
301  */
302 int
303 shmdt(td, uap)
304 	struct thread *td;
305 	struct shmdt_args *uap;
306 {
307 	struct proc *p = td->td_proc;
308 	struct shmmap_state *shmmap_s;
309 #ifdef MAC
310 	struct shmid_kernel *shmsegptr;
311 #endif
312 	int i;
313 	int error = 0;
314 
315 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
316 		return (ENOSYS);
317 	mtx_lock(&Giant);
318 	shmmap_s = p->p_vmspace->vm_shm;
319  	if (shmmap_s == NULL) {
320 		error = EINVAL;
321 		goto done2;
322 	}
323 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
324 		if (shmmap_s->shmid != -1 &&
325 		    shmmap_s->va == (vm_offset_t)uap->shmaddr) {
326 			break;
327 		}
328 	}
329 	if (i == shminfo.shmseg) {
330 		error = EINVAL;
331 		goto done2;
332 	}
333 #ifdef MAC
334 	shmsegptr = &shmsegs[IPCID_TO_IX(shmmap_s->shmid)];
335 	error = mac_check_sysv_shmdt(td->td_ucred, shmsegptr);
336 	if (error != 0) {
337 		MPRINTF(("mac_check_sysv_shmdt returned %d\n", error));
338 		goto done2;
339 	}
340 #endif
341 	error = shm_delete_mapping(p->p_vmspace, shmmap_s);
342 done2:
343 	mtx_unlock(&Giant);
344 	return (error);
345 }
346 
347 #ifndef _SYS_SYSPROTO_H_
348 struct shmat_args {
349 	int shmid;
350 	const void *shmaddr;
351 	int shmflg;
352 };
353 #endif
354 
355 /*
356  * MPSAFE
357  */
358 int
359 kern_shmat(td, shmid, shmaddr, shmflg)
360 	struct thread *td;
361 	int shmid;
362 	const void *shmaddr;
363 	int shmflg;
364 {
365 	struct proc *p = td->td_proc;
366 	int i, flags;
367 	struct shmid_kernel *shmseg;
368 	struct shmmap_state *shmmap_s = NULL;
369 	vm_offset_t attach_va;
370 	vm_prot_t prot;
371 	vm_size_t size;
372 	int rv;
373 	int error = 0;
374 
375 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
376 		return (ENOSYS);
377 	mtx_lock(&Giant);
378 	shmmap_s = p->p_vmspace->vm_shm;
379 	if (shmmap_s == NULL) {
380 		size = shminfo.shmseg * sizeof(struct shmmap_state);
381 		shmmap_s = malloc(size, M_SHM, M_WAITOK);
382 		for (i = 0; i < shminfo.shmseg; i++)
383 			shmmap_s[i].shmid = -1;
384 		p->p_vmspace->vm_shm = shmmap_s;
385 	}
386 	shmseg = shm_find_segment_by_shmid(shmid);
387 	if (shmseg == NULL) {
388 		error = EINVAL;
389 		goto done2;
390 	}
391 	error = ipcperm(td, &shmseg->u.shm_perm,
392 	    (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
393 	if (error)
394 		goto done2;
395 #ifdef MAC
396 	error = mac_check_sysv_shmat(td->td_ucred, shmseg, shmflg);
397 	if (error != 0) {
398 	 	MPRINTF(("mac_check_sysv_shmat returned %d\n", error));
399 		goto done2;
400 	}
401 #endif
402 	for (i = 0; i < shminfo.shmseg; i++) {
403 		if (shmmap_s->shmid == -1)
404 			break;
405 		shmmap_s++;
406 	}
407 	if (i >= shminfo.shmseg) {
408 		error = EMFILE;
409 		goto done2;
410 	}
411 	size = round_page(shmseg->u.shm_segsz);
412 #ifdef VM_PROT_READ_IS_EXEC
413 	prot = VM_PROT_READ | VM_PROT_EXECUTE;
414 #else
415 	prot = VM_PROT_READ;
416 #endif
417 	if ((shmflg & SHM_RDONLY) == 0)
418 		prot |= VM_PROT_WRITE;
419 	flags = MAP_ANON | MAP_SHARED;
420 	if (shmaddr) {
421 		flags |= MAP_FIXED;
422 		if (shmflg & SHM_RND) {
423 			attach_va = (vm_offset_t)shmaddr & ~(SHMLBA-1);
424 		} else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0) {
425 			attach_va = (vm_offset_t)shmaddr;
426 		} else {
427 			error = EINVAL;
428 			goto done2;
429 		}
430 	} else {
431 		/*
432 		 * This is just a hint to vm_map_find() about where to
433 		 * put it.
434 		 */
435 		PROC_LOCK(p);
436 		attach_va = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
437 		    lim_max(p, RLIMIT_DATA));
438 		PROC_UNLOCK(p);
439 	}
440 
441 	vm_object_reference(shmseg->u.shm_internal);
442 	rv = vm_map_find(&p->p_vmspace->vm_map, shmseg->u.shm_internal,
443 		0, &attach_va, size, (flags & MAP_FIXED)?0:1, prot, prot, 0);
444 	if (rv != KERN_SUCCESS) {
445 		vm_object_deallocate(shmseg->u.shm_internal);
446 		error = ENOMEM;
447 		goto done2;
448 	}
449 	vm_map_inherit(&p->p_vmspace->vm_map,
450 		attach_va, attach_va + size, VM_INHERIT_SHARE);
451 
452 	shmmap_s->va = attach_va;
453 	shmmap_s->shmid = shmid;
454 	shmseg->u.shm_lpid = p->p_pid;
455 	shmseg->u.shm_atime = time_second;
456 	shmseg->u.shm_nattch++;
457 	td->td_retval[0] = attach_va;
458 done2:
459 	mtx_unlock(&Giant);
460 	return (error);
461 }
462 
463 int
464 shmat(td, uap)
465 	struct thread *td;
466 	struct shmat_args *uap;
467 {
468 	return kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg);
469 }
470 
471 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
472 struct oshmid_ds {
473 	struct	ipc_perm shm_perm;	/* operation perms */
474 	int	shm_segsz;		/* size of segment (bytes) */
475 	u_short	shm_cpid;		/* pid, creator */
476 	u_short	shm_lpid;		/* pid, last operation */
477 	short	shm_nattch;		/* no. of current attaches */
478 	time_t	shm_atime;		/* last attach time */
479 	time_t	shm_dtime;		/* last detach time */
480 	time_t	shm_ctime;		/* last change time */
481 	void	*shm_handle;		/* internal handle for shm segment */
482 };
483 
484 struct oshmctl_args {
485 	int shmid;
486 	int cmd;
487 	struct oshmid_ds *ubuf;
488 };
489 
490 /*
491  * MPSAFE
492  */
493 static int
494 oshmctl(td, uap)
495 	struct thread *td;
496 	struct oshmctl_args *uap;
497 {
498 #ifdef COMPAT_43
499 	int error = 0;
500 	struct shmid_kernel *shmseg;
501 	struct oshmid_ds outbuf;
502 
503 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
504 		return (ENOSYS);
505 	mtx_lock(&Giant);
506 	shmseg = shm_find_segment_by_shmid(uap->shmid);
507 	if (shmseg == NULL) {
508 		error = EINVAL;
509 		goto done2;
510 	}
511 	switch (uap->cmd) {
512 	case IPC_STAT:
513 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
514 		if (error)
515 			goto done2;
516 #ifdef MAC
517 		error = mac_check_sysv_shmctl(td->td_ucred, shmseg, uap->cmd);
518 		if (error != 0) {
519 			MPRINTF(("mac_check_sysv_shmctl returned %d\n",
520 			    error));
521 			goto done2;
522 		}
523 #endif
524 		outbuf.shm_perm = shmseg->u.shm_perm;
525 		outbuf.shm_segsz = shmseg->u.shm_segsz;
526 		outbuf.shm_cpid = shmseg->u.shm_cpid;
527 		outbuf.shm_lpid = shmseg->u.shm_lpid;
528 		outbuf.shm_nattch = shmseg->u.shm_nattch;
529 		outbuf.shm_atime = shmseg->u.shm_atime;
530 		outbuf.shm_dtime = shmseg->u.shm_dtime;
531 		outbuf.shm_ctime = shmseg->u.shm_ctime;
532 		outbuf.shm_handle = shmseg->u.shm_internal;
533 		error = copyout(&outbuf, uap->ubuf, sizeof(outbuf));
534 		if (error)
535 			goto done2;
536 		break;
537 	default:
538 		error = shmctl(td, (struct shmctl_args *)uap);
539 		break;
540 	}
541 done2:
542 	mtx_unlock(&Giant);
543 	return (error);
544 #else
545 	return (EINVAL);
546 #endif
547 }
548 #endif
549 
550 #ifndef _SYS_SYSPROTO_H_
551 struct shmctl_args {
552 	int shmid;
553 	int cmd;
554 	struct shmid_ds *buf;
555 };
556 #endif
557 
558 /*
559  * MPSAFE
560  */
561 int
562 kern_shmctl(td, shmid, cmd, buf, bufsz)
563 	struct thread *td;
564 	int shmid;
565 	int cmd;
566 	void *buf;
567 	size_t *bufsz;
568 {
569 	int error = 0;
570 	struct shmid_kernel *shmseg;
571 
572 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
573 		return (ENOSYS);
574 
575 	mtx_lock(&Giant);
576 	switch (cmd) {
577 	case IPC_INFO:
578 		memcpy(buf, &shminfo, sizeof(shminfo));
579 		if (bufsz)
580 			*bufsz = sizeof(shminfo);
581 		td->td_retval[0] = shmalloced;
582 		goto done2;
583 	case SHM_INFO: {
584 		struct shm_info shm_info;
585 		shm_info.used_ids = shm_nused;
586 		shm_info.shm_rss = 0;	/*XXX where to get from ? */
587 		shm_info.shm_tot = 0;	/*XXX where to get from ? */
588 		shm_info.shm_swp = 0;	/*XXX where to get from ? */
589 		shm_info.swap_attempts = 0;	/*XXX where to get from ? */
590 		shm_info.swap_successes = 0;	/*XXX where to get from ? */
591 		memcpy(buf, &shm_info, sizeof(shm_info));
592 		if (bufsz)
593 			*bufsz = sizeof(shm_info);
594 		td->td_retval[0] = shmalloced;
595 		goto done2;
596 	}
597 	}
598 	if (cmd == SHM_STAT)
599 		shmseg = shm_find_segment_by_shmidx(shmid);
600 	else
601 		shmseg = shm_find_segment_by_shmid(shmid);
602 	if (shmseg == NULL) {
603 		error = EINVAL;
604 		goto done2;
605 	}
606 #ifdef MAC
607 	error = mac_check_sysv_shmctl(td->td_ucred, shmseg, cmd);
608 	if (error != 0) {
609 		MPRINTF(("mac_check_sysv_shmctl returned %d\n", error));
610 		goto done2;
611 	}
612 #endif
613 	switch (cmd) {
614 	case SHM_STAT:
615 	case IPC_STAT:
616 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
617 		if (error)
618 			goto done2;
619 		memcpy(buf, &shmseg->u, sizeof(struct shmid_ds));
620 		if (bufsz)
621 			*bufsz = sizeof(struct shmid_ds);
622 		if (cmd == SHM_STAT)
623 			td->td_retval[0] = IXSEQ_TO_IPCID(shmid, shmseg->u.shm_perm);
624 		break;
625 	case IPC_SET: {
626 		struct shmid_ds *shmid;
627 
628 		shmid = (struct shmid_ds *)buf;
629 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
630 		if (error)
631 			goto done2;
632 		shmseg->u.shm_perm.uid = shmid->shm_perm.uid;
633 		shmseg->u.shm_perm.gid = shmid->shm_perm.gid;
634 		shmseg->u.shm_perm.mode =
635 		    (shmseg->u.shm_perm.mode & ~ACCESSPERMS) |
636 		    (shmid->shm_perm.mode & ACCESSPERMS);
637 		shmseg->u.shm_ctime = time_second;
638 		break;
639 	}
640 	case IPC_RMID:
641 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
642 		if (error)
643 			goto done2;
644 		shmseg->u.shm_perm.key = IPC_PRIVATE;
645 		shmseg->u.shm_perm.mode |= SHMSEG_REMOVED;
646 		if (shmseg->u.shm_nattch <= 0) {
647 			shm_deallocate_segment(shmseg);
648 			shm_last_free = IPCID_TO_IX(shmid);
649 		}
650 		break;
651 #if 0
652 	case SHM_LOCK:
653 	case SHM_UNLOCK:
654 #endif
655 	default:
656 		error = EINVAL;
657 		break;
658 	}
659 done2:
660 	mtx_unlock(&Giant);
661 	return (error);
662 }
663 
664 int
665 shmctl(td, uap)
666 	struct thread *td;
667 	struct shmctl_args *uap;
668 {
669 	int error = 0;
670 	struct shmid_ds buf;
671 	size_t bufsz;
672 
673 	/* IPC_SET needs to copyin the buffer before calling kern_shmctl */
674 	if (uap->cmd == IPC_SET) {
675 		if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds))))
676 			goto done;
677 	}
678 
679 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
680 	if (error)
681 		goto done;
682 
683 	/* Cases in which we need to copyout */
684 	switch (uap->cmd) {
685 	case IPC_INFO:
686 	case SHM_INFO:
687 	case SHM_STAT:
688 	case IPC_STAT:
689 		error = copyout(&buf, uap->buf, bufsz);
690 		break;
691 	}
692 
693 done:
694 	if (error) {
695 		/* Invalidate the return value */
696 		td->td_retval[0] = -1;
697 	}
698 	return (error);
699 }
700 
701 
702 #ifndef _SYS_SYSPROTO_H_
703 struct shmget_args {
704 	key_t key;
705 	size_t size;
706 	int shmflg;
707 };
708 #endif
709 
710 static int
711 shmget_existing(td, uap, mode, segnum)
712 	struct thread *td;
713 	struct shmget_args *uap;
714 	int mode;
715 	int segnum;
716 {
717 	struct shmid_kernel *shmseg;
718 	int error;
719 
720 	shmseg = &shmsegs[segnum];
721 	if (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) {
722 		/*
723 		 * This segment is in the process of being allocated.  Wait
724 		 * until it's done, and look the key up again (in case the
725 		 * allocation failed or it was freed).
726 		 */
727 		shmseg->u.shm_perm.mode |= SHMSEG_WANTED;
728 		error = tsleep(shmseg, PLOCK | PCATCH, "shmget", 0);
729 		if (error)
730 			return (error);
731 		return (EAGAIN);
732 	}
733 	if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
734 		return (EEXIST);
735 #ifdef MAC
736 	error = mac_check_sysv_shmget(td->td_ucred, shmseg, uap->shmflg);
737 	if (error != 0) {
738 		MPRINTF(("mac_check_sysv_shmget returned %d\n", error));
739 		return (error);
740 	}
741 #endif
742 	error = ipcperm(td, &shmseg->u.shm_perm, mode);
743 	if (error)
744 		return (error);
745 	if (uap->size && uap->size > shmseg->u.shm_segsz)
746 		return (EINVAL);
747 	td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
748 	return (0);
749 }
750 
751 static int
752 shmget_allocate_segment(td, uap, mode)
753 	struct thread *td;
754 	struct shmget_args *uap;
755 	int mode;
756 {
757 	int i, segnum, shmid, size;
758 	struct ucred *cred = td->td_ucred;
759 	struct shmid_kernel *shmseg;
760 	vm_object_t shm_object;
761 
762 	GIANT_REQUIRED;
763 
764 	if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
765 		return (EINVAL);
766 	if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
767 		return (ENOSPC);
768 	size = round_page(uap->size);
769 	if (shm_committed + btoc(size) > shminfo.shmall)
770 		return (ENOMEM);
771 	if (shm_last_free < 0) {
772 		shmrealloc();	/* Maybe expand the shmsegs[] array. */
773 		for (i = 0; i < shmalloced; i++)
774 			if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE)
775 				break;
776 		if (i == shmalloced)
777 			return (ENOSPC);
778 		segnum = i;
779 	} else  {
780 		segnum = shm_last_free;
781 		shm_last_free = -1;
782 	}
783 	shmseg = &shmsegs[segnum];
784 	/*
785 	 * In case we sleep in malloc(), mark the segment present but deleted
786 	 * so that noone else tries to create the same key.
787 	 */
788 	shmseg->u.shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
789 	shmseg->u.shm_perm.key = uap->key;
790 	shmseg->u.shm_perm.seq = (shmseg->u.shm_perm.seq + 1) & 0x7fff;
791 	shmid = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
792 
793 	/*
794 	 * We make sure that we have allocated a pager before we need
795 	 * to.
796 	 */
797 	if (shm_use_phys) {
798 		shm_object =
799 		    vm_pager_allocate(OBJT_PHYS, 0, size, VM_PROT_DEFAULT, 0);
800 	} else {
801 		shm_object =
802 		    vm_pager_allocate(OBJT_SWAP, 0, size, VM_PROT_DEFAULT, 0);
803 	}
804 	VM_OBJECT_LOCK(shm_object);
805 	vm_object_clear_flag(shm_object, OBJ_ONEMAPPING);
806 	vm_object_set_flag(shm_object, OBJ_NOSPLIT);
807 	VM_OBJECT_UNLOCK(shm_object);
808 
809 	shmseg->u.shm_internal = shm_object;
810 	shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = cred->cr_uid;
811 	shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid;
812 	shmseg->u.shm_perm.mode = (shmseg->u.shm_perm.mode & SHMSEG_WANTED) |
813 	    (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
814 	shmseg->u.shm_segsz = uap->size;
815 	shmseg->u.shm_cpid = td->td_proc->p_pid;
816 	shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0;
817 	shmseg->u.shm_atime = shmseg->u.shm_dtime = 0;
818 #ifdef MAC
819 	mac_create_sysv_shm(cred, shmseg);
820 #endif
821 	shmseg->u.shm_ctime = time_second;
822 	shm_committed += btoc(size);
823 	shm_nused++;
824 	if (shmseg->u.shm_perm.mode & SHMSEG_WANTED) {
825 		/*
826 		 * Somebody else wanted this key while we were asleep.  Wake
827 		 * them up now.
828 		 */
829 		shmseg->u.shm_perm.mode &= ~SHMSEG_WANTED;
830 		wakeup(shmseg);
831 	}
832 	td->td_retval[0] = shmid;
833 	return (0);
834 }
835 
836 /*
837  * MPSAFE
838  */
839 int
840 shmget(td, uap)
841 	struct thread *td;
842 	struct shmget_args *uap;
843 {
844 	int segnum, mode;
845 	int error;
846 
847 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
848 		return (ENOSYS);
849 	mtx_lock(&Giant);
850 	mode = uap->shmflg & ACCESSPERMS;
851 	if (uap->key != IPC_PRIVATE) {
852 	again:
853 		segnum = shm_find_segment_by_key(uap->key);
854 		if (segnum >= 0) {
855 			error = shmget_existing(td, uap, mode, segnum);
856 			if (error == EAGAIN)
857 				goto again;
858 			goto done2;
859 		}
860 		if ((uap->shmflg & IPC_CREAT) == 0) {
861 			error = ENOENT;
862 			goto done2;
863 		}
864 	}
865 	error = shmget_allocate_segment(td, uap, mode);
866 done2:
867 	mtx_unlock(&Giant);
868 	return (error);
869 }
870 
871 /*
872  * MPSAFE
873  */
874 int
875 shmsys(td, uap)
876 	struct thread *td;
877 	/* XXX actually varargs. */
878 	struct shmsys_args /* {
879 		int	which;
880 		int	a2;
881 		int	a3;
882 		int	a4;
883 	} */ *uap;
884 {
885 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
886 	int error;
887 
888 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
889 		return (ENOSYS);
890 	if (uap->which < 0 ||
891 	    uap->which >= sizeof(shmcalls)/sizeof(shmcalls[0]))
892 		return (EINVAL);
893 	mtx_lock(&Giant);
894 	error = (*shmcalls[uap->which])(td, &uap->a2);
895 	mtx_unlock(&Giant);
896 	return (error);
897 #else
898 	return (nosys(td, NULL));
899 #endif
900 }
901 
902 static void
903 shmfork_myhook(p1, p2)
904 	struct proc *p1, *p2;
905 {
906 	struct shmmap_state *shmmap_s;
907 	size_t size;
908 	int i;
909 
910 	mtx_lock(&Giant);
911 	size = shminfo.shmseg * sizeof(struct shmmap_state);
912 	shmmap_s = malloc(size, M_SHM, M_WAITOK);
913 	bcopy(p1->p_vmspace->vm_shm, shmmap_s, size);
914 	p2->p_vmspace->vm_shm = shmmap_s;
915 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
916 		if (shmmap_s->shmid != -1)
917 			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++;
918 	mtx_unlock(&Giant);
919 }
920 
921 static void
922 shmexit_myhook(struct vmspace *vm)
923 {
924 	struct shmmap_state *base, *shm;
925 	int i;
926 
927 	if ((base = vm->vm_shm) != NULL) {
928 		vm->vm_shm = NULL;
929 		mtx_lock(&Giant);
930 		for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
931 			if (shm->shmid != -1)
932 				shm_delete_mapping(vm, shm);
933 		}
934 		mtx_unlock(&Giant);
935 		free(base, M_SHM);
936 	}
937 }
938 
939 static void
940 shmrealloc(void)
941 {
942 	int i;
943 	struct shmid_kernel *newsegs;
944 
945 	if (shmalloced >= shminfo.shmmni)
946 		return;
947 
948 	newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
949 	if (newsegs == NULL)
950 		return;
951 	for (i = 0; i < shmalloced; i++)
952 		bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
953 	for (; i < shminfo.shmmni; i++) {
954 		shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
955 		shmsegs[i].u.shm_perm.seq = 0;
956 #ifdef MAC
957 		mac_init_sysv_shm(&shmsegs[i]);
958 #endif
959 	}
960 	free(shmsegs, M_SHM);
961 	shmsegs = newsegs;
962 	shmalloced = shminfo.shmmni;
963 }
964 
965 static void
966 shminit()
967 {
968 	int i;
969 
970 	TUNABLE_ULONG_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall);
971 	for (i = PAGE_SIZE; i > 0; i--) {
972 		shminfo.shmmax = shminfo.shmall * i;
973 		if (shminfo.shmmax >= shminfo.shmall)
974 			break;
975 	}
976 	TUNABLE_ULONG_FETCH("kern.ipc.shmmin", &shminfo.shmmin);
977 	TUNABLE_ULONG_FETCH("kern.ipc.shmmni", &shminfo.shmmni);
978 	TUNABLE_ULONG_FETCH("kern.ipc.shmseg", &shminfo.shmseg);
979 	TUNABLE_INT_FETCH("kern.ipc.shm_use_phys", &shm_use_phys);
980 
981 	shmalloced = shminfo.shmmni;
982 	shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
983 	if (shmsegs == NULL)
984 		panic("cannot allocate initial memory for sysvshm");
985 	for (i = 0; i < shmalloced; i++) {
986 		shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
987 		shmsegs[i].u.shm_perm.seq = 0;
988 #ifdef MAC
989 		mac_init_sysv_shm(&shmsegs[i]);
990 #endif
991 	}
992 	shm_last_free = 0;
993 	shm_nused = 0;
994 	shm_committed = 0;
995 	shmexit_hook = &shmexit_myhook;
996 	shmfork_hook = &shmfork_myhook;
997 }
998 
999 static int
1000 shmunload()
1001 {
1002 #ifdef MAC
1003 	int i;
1004 #endif
1005 
1006 	if (shm_nused > 0)
1007 		return (EBUSY);
1008 
1009 #ifdef MAC
1010 	for (i = 0; i < shmalloced; i++)
1011 		mac_destroy_sysv_shm(&shmsegs[i]);
1012 #endif
1013 	free(shmsegs, M_SHM);
1014 	shmexit_hook = NULL;
1015 	shmfork_hook = NULL;
1016 	return (0);
1017 }
1018 
1019 static int
1020 sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
1021 {
1022 
1023 	return (SYSCTL_OUT(req, shmsegs, shmalloced * sizeof(shmsegs[0])));
1024 }
1025 
1026 static int
1027 sysvshm_modload(struct module *module, int cmd, void *arg)
1028 {
1029 	int error = 0;
1030 
1031 	switch (cmd) {
1032 	case MOD_LOAD:
1033 		shminit();
1034 		break;
1035 	case MOD_UNLOAD:
1036 		error = shmunload();
1037 		break;
1038 	case MOD_SHUTDOWN:
1039 		break;
1040 	default:
1041 		error = EINVAL;
1042 		break;
1043 	}
1044 	return (error);
1045 }
1046 
1047 static moduledata_t sysvshm_mod = {
1048 	"sysvshm",
1049 	&sysvshm_modload,
1050 	NULL
1051 };
1052 
1053 SYSCALL_MODULE_HELPER(shmsys);
1054 SYSCALL_MODULE_HELPER(shmat);
1055 SYSCALL_MODULE_HELPER(shmctl);
1056 SYSCALL_MODULE_HELPER(shmdt);
1057 SYSCALL_MODULE_HELPER(shmget);
1058 
1059 DECLARE_MODULE(sysvshm, sysvshm_mod,
1060 	SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
1061 MODULE_VERSION(sysvshm, 1);
1062