xref: /freebsd/sys/kern/sysv_shm.c (revision f4c5766baa461767ccb595252b1614f1ecc6f1a7)
1 /* $FreeBSD$ */
2 /*	$NetBSD: sysv_shm.c,v 1.23 1994/07/04 23:25:12 glass Exp $	*/
3 
4 /*
5  * Copyright (c) 1994 Adam Glass and Charles Hannum.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *	This product includes software developed by Adam Glass and Charles
18  *	Hannum.
19  * 4. The names of the authors may not be used to endorse or promote products
20  *    derived from this software without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25  * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include "opt_compat.h"
35 #include "opt_sysvipc.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/sysctl.h>
42 #include <sys/shm.h>
43 #include <sys/proc.h>
44 #include <sys/malloc.h>
45 #include <sys/mman.h>
46 #include <sys/mutex.h>
47 #include <sys/stat.h>
48 #include <sys/syscall.h>
49 #include <sys/syscallsubr.h>
50 #include <sys/sysent.h>
51 #include <sys/sysproto.h>
52 #include <sys/jail.h>
53 
54 #include <vm/vm.h>
55 #include <vm/vm_param.h>
56 #include <vm/pmap.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_map.h>
59 #include <vm/vm_page.h>
60 #include <vm/vm_pager.h>
61 
62 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
63 
64 struct oshmctl_args;
65 static int oshmctl(struct thread *td, struct oshmctl_args *uap);
66 
67 static int shmget_allocate_segment(struct thread *td,
68     struct shmget_args *uap, int mode);
69 static int shmget_existing(struct thread *td, struct shmget_args *uap,
70     int mode, int segnum);
71 
72 /* XXX casting to (sy_call_t *) is bogus, as usual. */
73 static sy_call_t *shmcalls[] = {
74 	(sy_call_t *)shmat, (sy_call_t *)oshmctl,
75 	(sy_call_t *)shmdt, (sy_call_t *)shmget,
76 	(sy_call_t *)shmctl
77 };
78 
79 #define	SHMSEG_FREE     	0x0200
80 #define	SHMSEG_REMOVED  	0x0400
81 #define	SHMSEG_ALLOCATED	0x0800
82 #define	SHMSEG_WANTED		0x1000
83 
84 static int shm_last_free, shm_nused, shm_committed, shmalloced;
85 static struct shmid_ds	*shmsegs;
86 
87 struct shm_handle {
88 	/* vm_offset_t kva; */
89 	vm_object_t shm_object;
90 };
91 
92 struct shmmap_state {
93 	vm_offset_t va;
94 	int shmid;
95 };
96 
97 static void shm_deallocate_segment(struct shmid_ds *);
98 static int shm_find_segment_by_key(key_t);
99 static struct shmid_ds *shm_find_segment_by_shmid(int, int);
100 static struct shmid_ds *shm_find_segment_by_shmidx(int, int);
101 static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *);
102 static void shmrealloc(void);
103 static void shminit(void);
104 static int sysvshm_modload(struct module *, int, void *);
105 static int shmunload(void);
106 static void shmexit_myhook(struct vmspace *vm);
107 static void shmfork_myhook(struct proc *p1, struct proc *p2);
108 static int sysctl_shmsegs(SYSCTL_HANDLER_ARGS);
109 
110 /*
111  * Tuneable values.
112  */
113 #ifndef SHMMAXPGS
114 #define	SHMMAXPGS	8192	/* Note: sysv shared memory is swap backed. */
115 #endif
116 #ifndef SHMMAX
117 #define	SHMMAX	(SHMMAXPGS*PAGE_SIZE)
118 #endif
119 #ifndef SHMMIN
120 #define	SHMMIN	1
121 #endif
122 #ifndef SHMMNI
123 #define	SHMMNI	192
124 #endif
125 #ifndef SHMSEG
126 #define	SHMSEG	128
127 #endif
128 #ifndef SHMALL
129 #define	SHMALL	(SHMMAXPGS)
130 #endif
131 
132 struct	shminfo shminfo = {
133 	SHMMAX,
134 	SHMMIN,
135 	SHMMNI,
136 	SHMSEG,
137 	SHMALL
138 };
139 
140 static int shm_use_phys;
141 
142 SYSCTL_DECL(_kern_ipc);
143 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0, "");
144 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0, "");
145 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RD, &shminfo.shmmni, 0, "");
146 SYSCTL_INT(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RD, &shminfo.shmseg, 0, "");
147 SYSCTL_INT(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0, "");
148 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW,
149     &shm_use_phys, 0, "");
150 SYSCTL_PROC(_kern_ipc, OID_AUTO, shmsegs, CTLFLAG_RD,
151     NULL, 0, sysctl_shmsegs, "", "");
152 
153 static int
154 shm_find_segment_by_key(key)
155 	key_t key;
156 {
157 	int i;
158 
159 	for (i = 0; i < shmalloced; i++)
160 		if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
161 		    shmsegs[i].shm_perm.key == key)
162 			return (i);
163 	return (-1);
164 }
165 
166 static struct shmid_ds *
167 shm_find_segment_by_shmid(int shmid, int wantrem)
168 {
169 	int segnum;
170 	struct shmid_ds *shmseg;
171 
172 	segnum = IPCID_TO_IX(shmid);
173 	if (segnum < 0 || segnum >= shmalloced)
174 		return (NULL);
175 	shmseg = &shmsegs[segnum];
176 	if (!((shmseg->shm_perm.mode & SHMSEG_ALLOCATED) ||
177   	    (wantrem && !(shmseg->shm_perm.mode & SHMSEG_REMOVED))) ||
178 	    shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid))
179 		return (NULL);
180 	return (shmseg);
181 }
182 
183 static struct shmid_ds *
184 shm_find_segment_by_shmidx(int segnum, int wantrem)
185 {
186 	struct shmid_ds *shmseg;
187 
188 	if (segnum < 0 || segnum >= shmalloced)
189 		return (NULL);
190 	shmseg = &shmsegs[segnum];
191 	if (!((shmseg->shm_perm.mode & SHMSEG_ALLOCATED) ||
192   	    (wantrem && !(shmseg->shm_perm.mode & SHMSEG_REMOVED))))
193 		return (NULL);
194 	return (shmseg);
195 }
196 
197 static void
198 shm_deallocate_segment(shmseg)
199 	struct shmid_ds *shmseg;
200 {
201 	struct shm_handle *shm_handle;
202 	size_t size;
203 
204 	GIANT_REQUIRED;
205 
206 	shm_handle = shmseg->shm_internal;
207 	vm_object_deallocate(shm_handle->shm_object);
208 	free(shm_handle, M_SHM);
209 	shmseg->shm_internal = NULL;
210 	size = round_page(shmseg->shm_segsz);
211 	shm_committed -= btoc(size);
212 	shm_nused--;
213 	shmseg->shm_perm.mode = SHMSEG_FREE;
214 }
215 
216 static int
217 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
218 {
219 	struct shmid_ds *shmseg;
220 	int segnum, result;
221 	size_t size;
222 
223 	GIANT_REQUIRED;
224 
225 	segnum = IPCID_TO_IX(shmmap_s->shmid);
226 	shmseg = &shmsegs[segnum];
227 	size = round_page(shmseg->shm_segsz);
228 	result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
229 	if (result != KERN_SUCCESS)
230 		return (EINVAL);
231 	shmmap_s->shmid = -1;
232 	shmseg->shm_dtime = time_second;
233 	if ((--shmseg->shm_nattch <= 0) &&
234 	    (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
235 		shm_deallocate_segment(shmseg);
236 		shm_last_free = segnum;
237 	}
238 	return (0);
239 }
240 
241 #ifndef _SYS_SYSPROTO_H_
242 struct shmdt_args {
243 	const void *shmaddr;
244 };
245 #endif
246 
247 /*
248  * MPSAFE
249  */
250 int
251 shmdt(td, uap)
252 	struct thread *td;
253 	struct shmdt_args *uap;
254 {
255 	struct proc *p = td->td_proc;
256 	struct shmmap_state *shmmap_s;
257 	int i;
258 	int error = 0;
259 
260 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
261 		return (ENOSYS);
262 	mtx_lock(&Giant);
263 	shmmap_s = p->p_vmspace->vm_shm;
264  	if (shmmap_s == NULL) {
265 		error = EINVAL;
266 		goto done2;
267 	}
268 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
269 		if (shmmap_s->shmid != -1 &&
270 		    shmmap_s->va == (vm_offset_t)uap->shmaddr) {
271 			break;
272 		}
273 	}
274 	if (i == shminfo.shmseg) {
275 		error = EINVAL;
276 		goto done2;
277 	}
278 	error = shm_delete_mapping(p->p_vmspace, shmmap_s);
279 done2:
280 	mtx_unlock(&Giant);
281 	return (error);
282 }
283 
284 #ifndef _SYS_SYSPROTO_H_
285 struct shmat_args {
286 	int shmid;
287 	const void *shmaddr;
288 	int shmflg;
289 };
290 #endif
291 
292 /*
293  * MPSAFE
294  */
295 int
296 kern_shmat(td, shmid, shmaddr, shmflg, wantrem)
297 	struct thread *td;
298 	int shmid;
299 	const void *shmaddr;
300 	int shmflg;
301 	int wantrem;
302 {
303 	struct proc *p = td->td_proc;
304 	int i, flags;
305 	struct shmid_ds *shmseg;
306 	struct shmmap_state *shmmap_s = NULL;
307 	struct shm_handle *shm_handle;
308 	vm_offset_t attach_va;
309 	vm_prot_t prot;
310 	vm_size_t size;
311 	int rv;
312 	int error = 0;
313 
314 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
315 		return (ENOSYS);
316 	mtx_lock(&Giant);
317 	shmmap_s = p->p_vmspace->vm_shm;
318 	if (shmmap_s == NULL) {
319 		size = shminfo.shmseg * sizeof(struct shmmap_state);
320 		shmmap_s = malloc(size, M_SHM, M_WAITOK);
321 		for (i = 0; i < shminfo.shmseg; i++)
322 			shmmap_s[i].shmid = -1;
323 		p->p_vmspace->vm_shm = shmmap_s;
324 	}
325 	shmseg = shm_find_segment_by_shmid(shmid, wantrem);
326 	if (shmseg == NULL) {
327 		error = EINVAL;
328 		goto done2;
329 	}
330 	error = ipcperm(td, &shmseg->shm_perm,
331 	    (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
332 	if (error)
333 		goto done2;
334 	for (i = 0; i < shminfo.shmseg; i++) {
335 		if (shmmap_s->shmid == -1)
336 			break;
337 		shmmap_s++;
338 	}
339 	if (i >= shminfo.shmseg) {
340 		error = EMFILE;
341 		goto done2;
342 	}
343 	size = round_page(shmseg->shm_segsz);
344 #ifdef VM_PROT_READ_IS_EXEC
345 	prot = VM_PROT_READ | VM_PROT_EXECUTE;
346 #else
347 	prot = VM_PROT_READ;
348 #endif
349 	if ((shmflg & SHM_RDONLY) == 0)
350 		prot |= VM_PROT_WRITE;
351 	flags = MAP_ANON | MAP_SHARED;
352 	if (shmaddr) {
353 		flags |= MAP_FIXED;
354 		if (shmflg & SHM_RND) {
355 			attach_va = (vm_offset_t)shmaddr & ~(SHMLBA-1);
356 		} else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0) {
357 			attach_va = (vm_offset_t)shmaddr;
358 		} else {
359 			error = EINVAL;
360 			goto done2;
361 		}
362 	} else {
363 		/*
364 		 * This is just a hint to vm_map_find() about where to
365 		 * put it.
366 		 */
367 		attach_va = round_page((vm_offset_t)p->p_vmspace->vm_taddr
368 		    + maxtsiz + maxdsiz);
369 	}
370 
371 	shm_handle = shmseg->shm_internal;
372 	vm_object_reference(shm_handle->shm_object);
373 	rv = vm_map_find(&p->p_vmspace->vm_map, shm_handle->shm_object,
374 		0, &attach_va, size, (flags & MAP_FIXED)?0:1, prot, prot, 0);
375 	if (rv != KERN_SUCCESS) {
376 		error = ENOMEM;
377 		goto done2;
378 	}
379 	vm_map_inherit(&p->p_vmspace->vm_map,
380 		attach_va, attach_va + size, VM_INHERIT_SHARE);
381 
382 	shmmap_s->va = attach_va;
383 	shmmap_s->shmid = shmid;
384 	shmseg->shm_lpid = p->p_pid;
385 	shmseg->shm_atime = time_second;
386 	shmseg->shm_nattch++;
387 	td->td_retval[0] = attach_va;
388 done2:
389 	mtx_unlock(&Giant);
390 	return (error);
391 }
392 
393 int
394 shmat(td, uap)
395 	struct thread *td;
396 	struct shmat_args *uap;
397 {
398 	return kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg, 0);
399 }
400 
401 struct oshmid_ds {
402 	struct	ipc_perm shm_perm;	/* operation perms */
403 	int	shm_segsz;		/* size of segment (bytes) */
404 	ushort	shm_cpid;		/* pid, creator */
405 	ushort	shm_lpid;		/* pid, last operation */
406 	short	shm_nattch;		/* no. of current attaches */
407 	time_t	shm_atime;		/* last attach time */
408 	time_t	shm_dtime;		/* last detach time */
409 	time_t	shm_ctime;		/* last change time */
410 	void	*shm_handle;		/* internal handle for shm segment */
411 };
412 
413 struct oshmctl_args {
414 	int shmid;
415 	int cmd;
416 	struct oshmid_ds *ubuf;
417 };
418 
419 /*
420  * MPSAFE
421  */
422 static int
423 oshmctl(td, uap)
424 	struct thread *td;
425 	struct oshmctl_args *uap;
426 {
427 #ifdef COMPAT_43
428 	int error = 0;
429 	struct shmid_ds *shmseg;
430 	struct oshmid_ds outbuf;
431 
432 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
433 		return (ENOSYS);
434 	mtx_lock(&Giant);
435 	shmseg = shm_find_segment_by_shmid(uap->shmid, 0);
436 	if (shmseg == NULL) {
437 		error = EINVAL;
438 		goto done2;
439 	}
440 	switch (uap->cmd) {
441 	case IPC_STAT:
442 		error = ipcperm(td, &shmseg->shm_perm, IPC_R);
443 		if (error)
444 			goto done2;
445 		outbuf.shm_perm = shmseg->shm_perm;
446 		outbuf.shm_segsz = shmseg->shm_segsz;
447 		outbuf.shm_cpid = shmseg->shm_cpid;
448 		outbuf.shm_lpid = shmseg->shm_lpid;
449 		outbuf.shm_nattch = shmseg->shm_nattch;
450 		outbuf.shm_atime = shmseg->shm_atime;
451 		outbuf.shm_dtime = shmseg->shm_dtime;
452 		outbuf.shm_ctime = shmseg->shm_ctime;
453 		outbuf.shm_handle = shmseg->shm_internal;
454 		error = copyout(&outbuf, uap->ubuf, sizeof(outbuf));
455 		if (error)
456 			goto done2;
457 		break;
458 	default:
459 		/* XXX casting to (sy_call_t *) is bogus, as usual. */
460 		error = ((sy_call_t *)shmctl)(td, uap);
461 		break;
462 	}
463 done2:
464 	mtx_unlock(&Giant);
465 	return (error);
466 #else
467 	return (EINVAL);
468 #endif
469 }
470 
471 #ifndef _SYS_SYSPROTO_H_
472 struct shmctl_args {
473 	int shmid;
474 	int cmd;
475 	struct shmid_ds *buf;
476 };
477 #endif
478 
479 /*
480  * MPSAFE
481  */
482 int
483 kern_shmctl(td, shmid, cmd, buf, bufsz, wantrem)
484 	struct thread *td;
485 	int shmid;
486 	int cmd;
487 	void *buf;
488 	size_t *bufsz;
489 	int wantrem;
490 {
491 	int error = 0;
492 	struct shmid_ds *shmseg;
493 
494 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
495 		return (ENOSYS);
496 
497 	mtx_lock(&Giant);
498 	switch (cmd) {
499 	case IPC_INFO:
500 		memcpy(buf, &shminfo, sizeof(shminfo));
501 		if (bufsz)
502 			*bufsz = sizeof(shminfo);
503 		td->td_retval[0] = shmalloced;
504 		goto done2;
505 	case SHM_INFO: {
506 		struct shm_info shm_info;
507 		shm_info.used_ids = shm_nused;
508 		shm_info.shm_rss = 0;	/*XXX where to get from ? */
509 		shm_info.shm_tot = 0;	/*XXX where to get from ? */
510 		shm_info.shm_swp = 0;	/*XXX where to get from ? */
511 		shm_info.swap_attempts = 0;	/*XXX where to get from ? */
512 		shm_info.swap_successes = 0;	/*XXX where to get from ? */
513 		memcpy(buf, &shm_info, sizeof(shm_info));
514 		if (bufsz)
515 			*bufsz = sizeof(shm_info);
516 		td->td_retval[0] = shmalloced;
517 		goto done2;
518 	}
519 	}
520 	if (cmd == SHM_STAT)
521 		shmseg = shm_find_segment_by_shmidx(shmid, wantrem);
522 	else
523 		shmseg = shm_find_segment_by_shmid(shmid, wantrem);
524 	if (shmseg == NULL) {
525 		error = EINVAL;
526 		goto done2;
527 	}
528 	switch (cmd) {
529 	case SHM_STAT:
530 	case IPC_STAT:
531 		error = ipcperm(td, &shmseg->shm_perm, IPC_R);
532 		if (error)
533 			goto done2;
534 		memcpy(buf, shmseg, sizeof(struct shmid_ds));
535 		if (bufsz)
536 			*bufsz = sizeof(struct shmid_ds);
537 		if (cmd == SHM_STAT)
538 			td->td_retval[0] = IXSEQ_TO_IPCID(shmid, shmseg->shm_perm);
539 		break;
540 	case IPC_SET: {
541 		struct shmid_ds *shmid;
542 
543 		shmid = (struct shmid_ds *)buf;
544 		error = ipcperm(td, &shmseg->shm_perm, IPC_M);
545 		if (error)
546 			goto done2;
547 		shmseg->shm_perm.uid = shmid->shm_perm.uid;
548 		shmseg->shm_perm.gid = shmid->shm_perm.gid;
549 		shmseg->shm_perm.mode =
550 		    (shmseg->shm_perm.mode & ~ACCESSPERMS) |
551 		    (shmid->shm_perm.mode & ACCESSPERMS);
552 		shmseg->shm_ctime = time_second;
553 		break;
554 	}
555 	case IPC_RMID:
556 		error = ipcperm(td, &shmseg->shm_perm, IPC_M);
557 		if (error)
558 			goto done2;
559 		shmseg->shm_perm.key = IPC_PRIVATE;
560 		shmseg->shm_perm.mode |= SHMSEG_REMOVED;
561 		if (shmseg->shm_nattch <= 0) {
562 			shm_deallocate_segment(shmseg);
563 			shm_last_free = IPCID_TO_IX(shmid);
564 		}
565 		break;
566 #if 0
567 	case SHM_LOCK:
568 	case SHM_UNLOCK:
569 #endif
570 	default:
571 		error = EINVAL;
572 		break;
573 	}
574 done2:
575 	mtx_unlock(&Giant);
576 	return (error);
577 }
578 
579 int
580 shmctl(td, uap)
581 	struct thread *td;
582 	struct shmctl_args *uap;
583 {
584 	int error = 0;
585 	struct shmid_ds buf;
586 	size_t bufsz;
587 
588 	/* IPC_SET needs to copyin the buffer before calling kern_shmctl */
589 	if (uap->cmd == IPC_SET) {
590 		if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds))))
591 			goto done;
592 	}
593 
594 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz, 0);
595 	if (error)
596 		goto done;
597 
598 	/* Cases in which we need to copyout */
599 	switch (uap->cmd) {
600 	case IPC_INFO:
601 	case SHM_INFO:
602 	case SHM_STAT:
603 	case IPC_STAT:
604 		error = copyout(&buf, uap->buf, bufsz);
605 		break;
606 	}
607 
608 done:
609 	if (error) {
610 		/* Invalidate the return value */
611 		td->td_retval[0] = -1;
612 	}
613 	return (error);
614 }
615 
616 
617 #ifndef _SYS_SYSPROTO_H_
618 struct shmget_args {
619 	key_t key;
620 	size_t size;
621 	int shmflg;
622 };
623 #endif
624 
625 static int
626 shmget_existing(td, uap, mode, segnum)
627 	struct thread *td;
628 	struct shmget_args *uap;
629 	int mode;
630 	int segnum;
631 {
632 	struct shmid_ds *shmseg;
633 	int error;
634 
635 	shmseg = &shmsegs[segnum];
636 	if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
637 		/*
638 		 * This segment is in the process of being allocated.  Wait
639 		 * until it's done, and look the key up again (in case the
640 		 * allocation failed or it was freed).
641 		 */
642 		shmseg->shm_perm.mode |= SHMSEG_WANTED;
643 		error = tsleep(shmseg, PLOCK | PCATCH, "shmget", 0);
644 		if (error)
645 			return (error);
646 		return (EAGAIN);
647 	}
648 	if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
649 		return (EEXIST);
650 	error = ipcperm(td, &shmseg->shm_perm, mode);
651 	if (error)
652 		return (error);
653 	if (uap->size && uap->size > shmseg->shm_segsz)
654 		return (EINVAL);
655 	td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
656 	return (0);
657 }
658 
659 static int
660 shmget_allocate_segment(td, uap, mode)
661 	struct thread *td;
662 	struct shmget_args *uap;
663 	int mode;
664 {
665 	int i, segnum, shmid, size;
666 	struct ucred *cred = td->td_ucred;
667 	struct shmid_ds *shmseg;
668 	struct shm_handle *shm_handle;
669 
670 	GIANT_REQUIRED;
671 
672 	if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
673 		return (EINVAL);
674 	if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
675 		return (ENOSPC);
676 	size = round_page(uap->size);
677 	if (shm_committed + btoc(size) > shminfo.shmall)
678 		return (ENOMEM);
679 	if (shm_last_free < 0) {
680 		shmrealloc();	/* Maybe expand the shmsegs[] array. */
681 		for (i = 0; i < shmalloced; i++)
682 			if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
683 				break;
684 		if (i == shmalloced)
685 			return (ENOSPC);
686 		segnum = i;
687 	} else  {
688 		segnum = shm_last_free;
689 		shm_last_free = -1;
690 	}
691 	shmseg = &shmsegs[segnum];
692 	/*
693 	 * In case we sleep in malloc(), mark the segment present but deleted
694 	 * so that noone else tries to create the same key.
695 	 */
696 	shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
697 	shmseg->shm_perm.key = uap->key;
698 	shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
699 	shm_handle = (struct shm_handle *)
700 	    malloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
701 	shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
702 
703 	/*
704 	 * We make sure that we have allocated a pager before we need
705 	 * to.
706 	 */
707 	if (shm_use_phys) {
708 		shm_handle->shm_object =
709 		    vm_pager_allocate(OBJT_PHYS, 0, size, VM_PROT_DEFAULT, 0);
710 	} else {
711 		shm_handle->shm_object =
712 		    vm_pager_allocate(OBJT_SWAP, 0, size, VM_PROT_DEFAULT, 0);
713 	}
714 	VM_OBJECT_LOCK(shm_handle->shm_object);
715 	vm_object_clear_flag(shm_handle->shm_object, OBJ_ONEMAPPING);
716 	vm_object_set_flag(shm_handle->shm_object, OBJ_NOSPLIT);
717 	VM_OBJECT_UNLOCK(shm_handle->shm_object);
718 
719 	shmseg->shm_internal = shm_handle;
720 	shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
721 	shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
722 	shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
723 	    (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
724 	shmseg->shm_segsz = uap->size;
725 	shmseg->shm_cpid = td->td_proc->p_pid;
726 	shmseg->shm_lpid = shmseg->shm_nattch = 0;
727 	shmseg->shm_atime = shmseg->shm_dtime = 0;
728 	shmseg->shm_ctime = time_second;
729 	shm_committed += btoc(size);
730 	shm_nused++;
731 	if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
732 		/*
733 		 * Somebody else wanted this key while we were asleep.  Wake
734 		 * them up now.
735 		 */
736 		shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
737 		wakeup(shmseg);
738 	}
739 	td->td_retval[0] = shmid;
740 	return (0);
741 }
742 
743 /*
744  * MPSAFE
745  */
746 int
747 shmget(td, uap)
748 	struct thread *td;
749 	struct shmget_args *uap;
750 {
751 	int segnum, mode;
752 	int error;
753 
754 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
755 		return (ENOSYS);
756 	mtx_lock(&Giant);
757 	mode = uap->shmflg & ACCESSPERMS;
758 	if (uap->key != IPC_PRIVATE) {
759 	again:
760 		segnum = shm_find_segment_by_key(uap->key);
761 		if (segnum >= 0) {
762 			error = shmget_existing(td, uap, mode, segnum);
763 			if (error == EAGAIN)
764 				goto again;
765 			goto done2;
766 		}
767 		if ((uap->shmflg & IPC_CREAT) == 0) {
768 			error = ENOENT;
769 			goto done2;
770 		}
771 	}
772 	error = shmget_allocate_segment(td, uap, mode);
773 done2:
774 	mtx_unlock(&Giant);
775 	return (error);
776 }
777 
778 /*
779  * MPSAFE
780  */
781 int
782 shmsys(td, uap)
783 	struct thread *td;
784 	/* XXX actually varargs. */
785 	struct shmsys_args /* {
786 		u_int	which;
787 		int	a2;
788 		int	a3;
789 		int	a4;
790 	} */ *uap;
791 {
792 	int error;
793 
794 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
795 		return (ENOSYS);
796 	if (uap->which >= sizeof(shmcalls)/sizeof(shmcalls[0]))
797 		return (EINVAL);
798 	mtx_lock(&Giant);
799 	error = (*shmcalls[uap->which])(td, &uap->a2);
800 	mtx_unlock(&Giant);
801 	return (error);
802 }
803 
804 static void
805 shmfork_myhook(p1, p2)
806 	struct proc *p1, *p2;
807 {
808 	struct shmmap_state *shmmap_s;
809 	size_t size;
810 	int i;
811 
812 	size = shminfo.shmseg * sizeof(struct shmmap_state);
813 	shmmap_s = malloc(size, M_SHM, M_WAITOK);
814 	bcopy(p1->p_vmspace->vm_shm, shmmap_s, size);
815 	p2->p_vmspace->vm_shm = shmmap_s;
816 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
817 		if (shmmap_s->shmid != -1)
818 			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
819 }
820 
821 static void
822 shmexit_myhook(struct vmspace *vm)
823 {
824 	struct shmmap_state *base, *shm;
825 	int i;
826 
827 	GIANT_REQUIRED;
828 
829 	if ((base = vm->vm_shm) != NULL) {
830 		vm->vm_shm = NULL;
831 		for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
832 			if (shm->shmid != -1)
833 				shm_delete_mapping(vm, shm);
834 		}
835 		free(base, M_SHM);
836 	}
837 }
838 
839 static void
840 shmrealloc(void)
841 {
842 	int i;
843 	struct shmid_ds *newsegs;
844 
845 	if (shmalloced >= shminfo.shmmni)
846 		return;
847 
848 	newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
849 	if (newsegs == NULL)
850 		return;
851 	for (i = 0; i < shmalloced; i++)
852 		bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
853 	for (; i < shminfo.shmmni; i++) {
854 		shmsegs[i].shm_perm.mode = SHMSEG_FREE;
855 		shmsegs[i].shm_perm.seq = 0;
856 	}
857 	free(shmsegs, M_SHM);
858 	shmsegs = newsegs;
859 	shmalloced = shminfo.shmmni;
860 }
861 
862 static void
863 shminit()
864 {
865 	int i;
866 
867 	TUNABLE_INT_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall);
868 	for (i = PAGE_SIZE; i > 0; i--) {
869 		shminfo.shmmax = shminfo.shmall * PAGE_SIZE;
870 		if (shminfo.shmmax >= shminfo.shmall)
871 			break;
872 	}
873 	TUNABLE_INT_FETCH("kern.ipc.shmmin", &shminfo.shmmin);
874 	TUNABLE_INT_FETCH("kern.ipc.shmmni", &shminfo.shmmni);
875 	TUNABLE_INT_FETCH("kern.ipc.shmseg", &shminfo.shmseg);
876 	TUNABLE_INT_FETCH("kern.ipc.shm_use_phys", &shm_use_phys);
877 
878 	shmalloced = shminfo.shmmni;
879 	shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
880 	if (shmsegs == NULL)
881 		panic("cannot allocate initial memory for sysvshm");
882 	for (i = 0; i < shmalloced; i++) {
883 		shmsegs[i].shm_perm.mode = SHMSEG_FREE;
884 		shmsegs[i].shm_perm.seq = 0;
885 	}
886 	shm_last_free = 0;
887 	shm_nused = 0;
888 	shm_committed = 0;
889 	shmexit_hook = &shmexit_myhook;
890 	shmfork_hook = &shmfork_myhook;
891 }
892 
893 static int
894 shmunload()
895 {
896 
897 	if (shm_nused > 0)
898 		return (EBUSY);
899 
900 	free(shmsegs, M_SHM);
901 	shmexit_hook = NULL;
902 	shmfork_hook = NULL;
903 	return (0);
904 }
905 
906 static int
907 sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
908 {
909 
910 	return (SYSCTL_OUT(req, shmsegs, shmalloced * sizeof(shmsegs[0])));
911 }
912 
913 static int
914 sysvshm_modload(struct module *module, int cmd, void *arg)
915 {
916 	int error = 0;
917 
918 	switch (cmd) {
919 	case MOD_LOAD:
920 		shminit();
921 		break;
922 	case MOD_UNLOAD:
923 		error = shmunload();
924 		break;
925 	case MOD_SHUTDOWN:
926 		break;
927 	default:
928 		error = EINVAL;
929 		break;
930 	}
931 	return (error);
932 }
933 
934 static moduledata_t sysvshm_mod = {
935 	"sysvshm",
936 	&sysvshm_modload,
937 	NULL
938 };
939 
940 SYSCALL_MODULE_HELPER(shmsys);
941 SYSCALL_MODULE_HELPER(shmat);
942 SYSCALL_MODULE_HELPER(shmctl);
943 SYSCALL_MODULE_HELPER(shmdt);
944 SYSCALL_MODULE_HELPER(shmget);
945 
946 DECLARE_MODULE(sysvshm, sysvshm_mod,
947 	SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
948 MODULE_VERSION(sysvshm, 1);
949