xref: /netbsd-src/sys/kern/sysv_shm.c (revision 3b01aba77a7a698587faaae455bbfe740923c1f5)
1 /*	$NetBSD: sysv_shm.c,v 1.61 2001/03/15 06:10:56 chs Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the NetBSD
22  *	Foundation, Inc. and its contributors.
23  * 4. Neither the name of The NetBSD Foundation nor the names of its
24  *    contributors may be used to endorse or promote products derived
25  *    from this software without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37  * POSSIBILITY OF SUCH DAMAGE.
38  */
39 
40 /*
41  * Copyright (c) 1994 Adam Glass and Charles M. Hannum.  All rights reserved.
42  *
43  * Redistribution and use in source and binary forms, with or without
44  * modification, are permitted provided that the following conditions
45  * are met:
46  * 1. Redistributions of source code must retain the above copyright
47  *    notice, this list of conditions and the following disclaimer.
48  * 2. Redistributions in binary form must reproduce the above copyright
49  *    notice, this list of conditions and the following disclaimer in the
50  *    documentation and/or other materials provided with the distribution.
51  * 3. All advertising materials mentioning features or use of this software
52  *    must display the following acknowledgement:
53  *	This product includes software developed by Adam Glass and Charles M.
54  *	Hannum.
55  * 4. The names of the authors may not be used to endorse or promote products
56  *    derived from this software without specific prior written permission.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
59  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
60  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
61  * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
62  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
63  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
64  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
65  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
66  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
67  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
68  */
69 
70 #define SYSVSHM
71 
72 #include <sys/types.h>
73 #include <sys/param.h>
74 #include <sys/kernel.h>
75 #include <sys/shm.h>
76 #include <sys/malloc.h>
77 #include <sys/mman.h>
78 #include <sys/stat.h>
79 #include <sys/sysctl.h>
80 #include <sys/mount.h>		/* XXX for <sys/syscallargs.h> */
81 #include <sys/syscallargs.h>
82 
83 #include <uvm/uvm_extern.h>
84 
85 struct shmid_ds *shm_find_segment_by_shmid __P((int));
86 
87 /*
88  * Provides the following externally accessible functions:
89  *
90  * shminit(void);		                 initialization
91  * shmexit(struct vmspace *)                     cleanup
92  * shmfork(struct vmspace *, struct vmspace *)   fork handling
93  *
94  * Structures:
95  * shmsegs (an array of 'struct shmid_ds')
96  * per proc array of 'struct shmmap_state'
97  */
98 
99 #define	SHMSEG_FREE     	0x0200
100 #define	SHMSEG_REMOVED  	0x0400
101 #define	SHMSEG_ALLOCATED	0x0800
102 #define	SHMSEG_WANTED		0x1000
103 
104 int	shm_last_free, shm_nused, shm_committed;
105 struct	shmid_ds *shmsegs;
106 
107 struct shm_handle {
108 	struct uvm_object *shm_object;
109 };
110 
111 struct shmmap_state {
112 	vaddr_t va;
113 	int shmid;
114 };
115 
116 static int shm_find_segment_by_key __P((key_t));
117 static void shm_deallocate_segment __P((struct shmid_ds *));
118 static void shm_delete_mapping __P((struct vmspace *, struct shmmap_state *));
119 static int shmget_existing __P((struct proc *, struct sys_shmget_args *,
120 				int, int, register_t *));
121 static int shmget_allocate_segment __P((struct proc *, struct sys_shmget_args *,
122 					int, register_t *));
123 
124 static int
125 shm_find_segment_by_key(key)
126 	key_t key;
127 {
128 	int i;
129 
130 	for (i = 0; i < shminfo.shmmni; i++)
131 		if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
132 		    shmsegs[i].shm_perm._key == key)
133 			return i;
134 	return -1;
135 }
136 
137 struct shmid_ds *
138 shm_find_segment_by_shmid(shmid)
139 	int shmid;
140 {
141 	int segnum;
142 	struct shmid_ds *shmseg;
143 
144 	segnum = IPCID_TO_IX(shmid);
145 	if (segnum < 0 || segnum >= shminfo.shmmni)
146 		return NULL;
147 	shmseg = &shmsegs[segnum];
148 	if ((shmseg->shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
149 	    != SHMSEG_ALLOCATED ||
150 	    shmseg->shm_perm._seq != IPCID_TO_SEQ(shmid))
151 		return NULL;
152 	return shmseg;
153 }
154 
155 static void
156 shm_deallocate_segment(shmseg)
157 	struct shmid_ds *shmseg;
158 {
159 	struct shm_handle *shm_handle;
160 	size_t size;
161 
162 	shm_handle = shmseg->_shm_internal;
163 	size = (shmseg->shm_segsz + PGOFSET) & ~PGOFSET;
164 	uao_detach(shm_handle->shm_object);
165 	free((caddr_t)shm_handle, M_SHM);
166 	shmseg->_shm_internal = NULL;
167 	shm_committed -= btoc(size);
168 	shmseg->shm_perm.mode = SHMSEG_FREE;
169 	shm_nused--;
170 }
171 
172 static void
173 shm_delete_mapping(vm, shmmap_s)
174 	struct vmspace *vm;
175 	struct shmmap_state *shmmap_s;
176 {
177 	struct shmid_ds *shmseg;
178 	int segnum;
179 	size_t size;
180 
181 	segnum = IPCID_TO_IX(shmmap_s->shmid);
182 	shmseg = &shmsegs[segnum];
183 	size = (shmseg->shm_segsz + PGOFSET) & ~PGOFSET;
184 	uvm_deallocate(&vm->vm_map, shmmap_s->va, size);
185 	shmmap_s->shmid = -1;
186 	shmseg->shm_dtime = time.tv_sec;
187 	if ((--shmseg->shm_nattch <= 0) &&
188 	    (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
189 		shm_deallocate_segment(shmseg);
190 		shm_last_free = segnum;
191 	}
192 }
193 
194 int
195 sys_shmdt(p, v, retval)
196 	struct proc *p;
197 	void *v;
198 	register_t *retval;
199 {
200 	struct sys_shmdt_args /* {
201 		syscallarg(const void *) shmaddr;
202 	} */ *uap = v;
203 	struct shmmap_state *shmmap_s;
204 	int i;
205 
206 	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
207 	if (shmmap_s == NULL)
208 		return EINVAL;
209 
210 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
211 		if (shmmap_s->shmid != -1 &&
212 		    shmmap_s->va == (vaddr_t)SCARG(uap, shmaddr))
213 			break;
214 	if (i == shminfo.shmseg)
215 		return EINVAL;
216 	shm_delete_mapping(p->p_vmspace, shmmap_s);
217 	return 0;
218 }
219 
220 int
221 sys_shmat(p, v, retval)
222 	struct proc *p;
223 	void *v;
224 	register_t *retval;
225 {
226 	struct sys_shmat_args /* {
227 		syscallarg(int) shmid;
228 		syscallarg(const void *) shmaddr;
229 		syscallarg(int) shmflg;
230 	} */ *uap = v;
231 	int error, i, flags;
232 	struct ucred *cred = p->p_ucred;
233 	struct shmid_ds *shmseg;
234 	struct shmmap_state *shmmap_s = NULL;
235 	struct shm_handle *shm_handle;
236 	vaddr_t attach_va;
237 	vm_prot_t prot;
238 	vsize_t size;
239 
240 	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
241 	if (shmmap_s == NULL) {
242 		size = shminfo.shmseg * sizeof(struct shmmap_state);
243 		shmmap_s = malloc(size, M_SHM, M_WAITOK);
244 		for (i = 0; i < shminfo.shmseg; i++)
245 			shmmap_s[i].shmid = -1;
246 		p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
247 	}
248 	shmseg = shm_find_segment_by_shmid(SCARG(uap, shmid));
249 	if (shmseg == NULL)
250 		return EINVAL;
251 	error = ipcperm(cred, &shmseg->shm_perm,
252 		    (SCARG(uap, shmflg) & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
253 	if (error)
254 		return error;
255 	for (i = 0; i < shminfo.shmseg; i++) {
256 		if (shmmap_s->shmid == -1)
257 			break;
258 		shmmap_s++;
259 	}
260 	if (i >= shminfo.shmseg)
261 		return EMFILE;
262 	size = (shmseg->shm_segsz + PGOFSET) & ~PGOFSET;
263 	prot = VM_PROT_READ;
264 	if ((SCARG(uap, shmflg) & SHM_RDONLY) == 0)
265 		prot |= VM_PROT_WRITE;
266 	flags = MAP_ANON | MAP_SHARED;
267 	if (SCARG(uap, shmaddr)) {
268 		flags |= MAP_FIXED;
269 		if (SCARG(uap, shmflg) & SHM_RND)
270 			attach_va =
271 			    (vaddr_t)SCARG(uap, shmaddr) & ~(SHMLBA-1);
272 		else if (((vaddr_t)SCARG(uap, shmaddr) & (SHMLBA-1)) == 0)
273 			attach_va = (vaddr_t)SCARG(uap, shmaddr);
274 		else
275 			return EINVAL;
276 	} else {
277 		/* This is just a hint to vm_mmap() about where to put it. */
278 		attach_va =
279 		    round_page((vaddr_t)p->p_vmspace->vm_taddr +
280 			MAXTSIZ + MAXDSIZ);
281 	}
282 	shm_handle = shmseg->_shm_internal;
283 	uao_reference(shm_handle->shm_object);
284 	error = uvm_map(&p->p_vmspace->vm_map, &attach_va, size,
285 	    shm_handle->shm_object, 0, 0,
286 	    UVM_MAPFLAG(prot, prot, UVM_INH_SHARE, UVM_ADV_RANDOM, 0));
287 	if (error) {
288 		return error;
289 	}
290 	shmmap_s->va = attach_va;
291 	shmmap_s->shmid = SCARG(uap, shmid);
292 	shmseg->shm_lpid = p->p_pid;
293 	shmseg->shm_atime = time.tv_sec;
294 	shmseg->shm_nattch++;
295 	*retval = attach_va;
296 	return 0;
297 }
298 
299 int
300 sys___shmctl13(p, v, retval)
301 	struct proc *p;
302 	void *v;
303 	register_t *retval;
304 {
305 	struct sys___shmctl13_args /* {
306 		syscallarg(int) shmid;
307 		syscallarg(int) cmd;
308 		syscallarg(struct shmid_ds *) buf;
309 	} */ *uap = v;
310 	struct shmid_ds shmbuf;
311 	int cmd, error;
312 
313 	cmd = SCARG(uap, cmd);
314 
315 	if (cmd == IPC_SET) {
316 		error = copyin(SCARG(uap, buf), &shmbuf, sizeof(shmbuf));
317 		if (error)
318 			return (error);
319 	}
320 
321 	error = shmctl1(p, SCARG(uap, shmid), cmd,
322 	    (cmd == IPC_SET || cmd == IPC_STAT) ? &shmbuf : NULL);
323 
324 	if (error == 0 && cmd == IPC_STAT)
325 		error = copyout(&shmbuf, SCARG(uap, buf), sizeof(shmbuf));
326 
327 	return (error);
328 }
329 
330 int
331 shmctl1(p, shmid, cmd, shmbuf)
332 	struct proc *p;
333 	int shmid;
334 	int cmd;
335 	struct shmid_ds *shmbuf;
336 {
337 	struct ucred *cred = p->p_ucred;
338 	struct shmid_ds *shmseg;
339 	int error = 0;
340 
341 	shmseg = shm_find_segment_by_shmid(shmid);
342 	if (shmseg == NULL)
343 		return EINVAL;
344 	switch (cmd) {
345 	case IPC_STAT:
346 		if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_R)) != 0)
347 			return error;
348 		memcpy(shmbuf, shmseg, sizeof(struct shmid_ds));
349 		break;
350 	case IPC_SET:
351 		if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_M)) != 0)
352 			return error;
353 		shmseg->shm_perm.uid = shmbuf->shm_perm.uid;
354 		shmseg->shm_perm.gid = shmbuf->shm_perm.gid;
355 		shmseg->shm_perm.mode =
356 		    (shmseg->shm_perm.mode & ~ACCESSPERMS) |
357 		    (shmbuf->shm_perm.mode & ACCESSPERMS);
358 		shmseg->shm_ctime = time.tv_sec;
359 		break;
360 	case IPC_RMID:
361 		if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_M)) != 0)
362 			return error;
363 		shmseg->shm_perm._key = IPC_PRIVATE;
364 		shmseg->shm_perm.mode |= SHMSEG_REMOVED;
365 		if (shmseg->shm_nattch <= 0) {
366 			shm_deallocate_segment(shmseg);
367 			shm_last_free = IPCID_TO_IX(shmid);
368 		}
369 		break;
370 	case SHM_LOCK:
371 	case SHM_UNLOCK:
372 	default:
373 		return EINVAL;
374 	}
375 	return 0;
376 }
377 
378 static int
379 shmget_existing(p, uap, mode, segnum, retval)
380 	struct proc *p;
381 	struct sys_shmget_args /* {
382 		syscallarg(key_t) key;
383 		syscallarg(size_t) size;
384 		syscallarg(int) shmflg;
385 	} */ *uap;
386 	int mode;
387 	int segnum;
388 	register_t *retval;
389 {
390 	struct shmid_ds *shmseg;
391 	struct ucred *cred = p->p_ucred;
392 	int error;
393 
394 	shmseg = &shmsegs[segnum];
395 	if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
396 		/*
397 		 * This segment is in the process of being allocated.  Wait
398 		 * until it's done, and look the key up again (in case the
399 		 * allocation failed or it was freed).
400 		 */
401 		shmseg->shm_perm.mode |= SHMSEG_WANTED;
402 		error = tsleep((caddr_t)shmseg, PLOCK | PCATCH, "shmget", 0);
403 		if (error)
404 			return error;
405 		return EAGAIN;
406 	}
407 	if ((error = ipcperm(cred, &shmseg->shm_perm, mode)) != 0)
408 		return error;
409 	if (SCARG(uap, size) && SCARG(uap, size) > shmseg->shm_segsz)
410 		return EINVAL;
411 	if ((SCARG(uap, shmflg) & (IPC_CREAT | IPC_EXCL)) ==
412 	    (IPC_CREAT | IPC_EXCL))
413 		return EEXIST;
414 	*retval = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
415 	return 0;
416 }
417 
418 static int
419 shmget_allocate_segment(p, uap, mode, retval)
420 	struct proc *p;
421 	struct sys_shmget_args /* {
422 		syscallarg(key_t) key;
423 		syscallarg(size_t) size;
424 		syscallarg(int) shmflg;
425 	} */ *uap;
426 	int mode;
427 	register_t *retval;
428 {
429 	int i, segnum, shmid, size;
430 	struct ucred *cred = p->p_ucred;
431 	struct shmid_ds *shmseg;
432 	struct shm_handle *shm_handle;
433 	int error = 0;
434 
435 	if (SCARG(uap, size) < shminfo.shmmin ||
436 	    SCARG(uap, size) > shminfo.shmmax)
437 		return EINVAL;
438 	if (shm_nused >= shminfo.shmmni) /* any shmids left? */
439 		return ENOSPC;
440 	size = (SCARG(uap, size) + PGOFSET) & ~PGOFSET;
441 	if (shm_committed + btoc(size) > shminfo.shmall)
442 		return ENOMEM;
443 	if (shm_last_free < 0) {
444 		for (i = 0; i < shminfo.shmmni; i++)
445 			if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
446 				break;
447 		if (i == shminfo.shmmni)
448 			panic("shmseg free count inconsistent");
449 		segnum = i;
450 	} else  {
451 		segnum = shm_last_free;
452 		shm_last_free = -1;
453 	}
454 	shmseg = &shmsegs[segnum];
455 	/*
456 	 * In case we sleep in malloc(), mark the segment present but deleted
457 	 * so that noone else tries to create the same key.
458 	 */
459 	shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
460 	shmseg->shm_perm._key = SCARG(uap, key);
461 	shmseg->shm_perm._seq = (shmseg->shm_perm._seq + 1) & 0x7fff;
462 	shm_handle = (struct shm_handle *)
463 	    malloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
464 	shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
465 
466 	shm_handle->shm_object = uao_create(size, 0);
467 
468 	shmseg->_shm_internal = shm_handle;
469 	shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
470 	shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
471 	shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
472 	    (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
473 	shmseg->shm_segsz = SCARG(uap, size);
474 	shmseg->shm_cpid = p->p_pid;
475 	shmseg->shm_lpid = shmseg->shm_nattch = 0;
476 	shmseg->shm_atime = shmseg->shm_dtime = 0;
477 	shmseg->shm_ctime = time.tv_sec;
478 	shm_committed += btoc(size);
479 	shm_nused++;
480 
481 	*retval = shmid;
482 	if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
483 		/*
484 		 * Somebody else wanted this key while we were asleep.  Wake
485 		 * them up now.
486 		 */
487 		shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
488 		wakeup((caddr_t)shmseg);
489 	}
490 	return error;
491 }
492 
493 int
494 sys_shmget(p, v, retval)
495 	struct proc *p;
496 	void *v;
497 	register_t *retval;
498 {
499 	struct sys_shmget_args /* {
500 		syscallarg(key_t) key;
501 		syscallarg(int) size;
502 		syscallarg(int) shmflg;
503 	} */ *uap = v;
504 	int segnum, mode, error;
505 
506 	mode = SCARG(uap, shmflg) & ACCESSPERMS;
507 	if (SCARG(uap, key) != IPC_PRIVATE) {
508 	again:
509 		segnum = shm_find_segment_by_key(SCARG(uap, key));
510 		if (segnum >= 0) {
511 			error = shmget_existing(p, uap, mode, segnum, retval);
512 			if (error == EAGAIN)
513 				goto again;
514 			return error;
515 		}
516 		if ((SCARG(uap, shmflg) & IPC_CREAT) == 0)
517 			return ENOENT;
518 	}
519 	return shmget_allocate_segment(p, uap, mode, retval);
520 }
521 
522 void
523 shmfork(vm1, vm2)
524 	struct vmspace *vm1, *vm2;
525 {
526 	struct shmmap_state *shmmap_s;
527 	size_t size;
528 	int i;
529 
530 	if (vm1->vm_shm == NULL) {
531 		vm2->vm_shm = NULL;
532 		return;
533 	}
534 
535 	size = shminfo.shmseg * sizeof(struct shmmap_state);
536 	shmmap_s = malloc(size, M_SHM, M_WAITOK);
537 	memcpy(shmmap_s, vm1->vm_shm, size);
538 	vm2->vm_shm = (caddr_t)shmmap_s;
539 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
540 		if (shmmap_s->shmid != -1)
541 			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
542 }
543 
544 void
545 shmexit(vm)
546 	struct vmspace *vm;
547 {
548 	struct shmmap_state *shmmap_s;
549 	int i;
550 
551 	shmmap_s = (struct shmmap_state *)vm->vm_shm;
552 	if (shmmap_s == NULL)
553 		return;
554 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
555 		if (shmmap_s->shmid != -1)
556 			shm_delete_mapping(vm, shmmap_s);
557 	free(vm->vm_shm, M_SHM);
558 	vm->vm_shm = NULL;
559 }
560 
561 void
562 shminit()
563 {
564 	int i;
565 
566 	shminfo.shmmax *= PAGE_SIZE;
567 
568 	for (i = 0; i < shminfo.shmmni; i++) {
569 		shmsegs[i].shm_perm.mode = SHMSEG_FREE;
570 		shmsegs[i].shm_perm._seq = 0;
571 	}
572 	shm_last_free = 0;
573 	shm_nused = 0;
574 	shm_committed = 0;
575 }
576