sysv_shm.c revision 1.56.2.2 1 /* $NetBSD: sysv_shm.c,v 1.56.2.2 2004/04/11 02:21:06 jmc 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
80 #include <vm/vm.h>
81 #include <uvm/uvm_extern.h>
82
83 struct uvm_faultinfo;
84 #include <uvm/uvm_pager.h>
85 #include <uvm/uvm_object.h>
86
87 #include <sys/sysctl.h>
88 #include <sys/mount.h> /* XXX for <sys/syscallargs.h> */
89 #include <sys/syscallargs.h>
90
91 struct shmid_ds *shm_find_segment_by_shmid __P((int));
92
93 /*
94 * Provides the following externally accessible functions:
95 *
96 * shminit(void); initialization
97 * shmexit(struct vmspace *) cleanup
98 * shmfork(struct vmspace *, struct vmspace *) fork handling
99 *
100 * Structures:
101 * shmsegs (an array of 'struct shmid_ds')
102 * per proc array of 'struct shmmap_state'
103 */
104
105 #define SHMSEG_FREE 0x0200
106 #define SHMSEG_REMOVED 0x0400
107 #define SHMSEG_ALLOCATED 0x0800
108 #define SHMSEG_WANTED 0x1000
109
110 int shm_last_free, shm_nused, shm_committed;
111 struct shmid_ds *shmsegs;
112
113 struct shm_handle {
114 struct uvm_object *shm_object;
115 };
116
117 struct shmmap_state {
118 vaddr_t va;
119 int shmid;
120 };
121
122 static int shm_find_segment_by_key __P((key_t));
123 static void shm_deallocate_segment __P((struct shmid_ds *));
124 static int shm_delete_mapping __P((struct vmspace *, struct shmmap_state *));
125 static int shmget_existing __P((struct proc *, struct sys_shmget_args *,
126 int, int, register_t *));
127 static int shmget_allocate_segment __P((struct proc *, struct sys_shmget_args *,
128 int, register_t *));
129
130 static int
131 shm_find_segment_by_key(key)
132 key_t key;
133 {
134 int i;
135
136 for (i = 0; i < shminfo.shmmni; i++)
137 if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
138 shmsegs[i].shm_perm._key == key)
139 return i;
140 return -1;
141 }
142
143 struct shmid_ds *
144 shm_find_segment_by_shmid(shmid)
145 int shmid;
146 {
147 int segnum;
148 struct shmid_ds *shmseg;
149
150 segnum = IPCID_TO_IX(shmid);
151 if (segnum < 0 || segnum >= shminfo.shmmni)
152 return NULL;
153 shmseg = &shmsegs[segnum];
154 if ((shmseg->shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
155 != SHMSEG_ALLOCATED ||
156 shmseg->shm_perm._seq != IPCID_TO_SEQ(shmid))
157 return NULL;
158 return shmseg;
159 }
160
161 static void
162 shm_deallocate_segment(shmseg)
163 struct shmid_ds *shmseg;
164 {
165 struct shm_handle *shm_handle = shmseg->_shm_internal;
166 struct uvm_object *uobj = shm_handle->shm_object;
167 size_t size = (shmseg->shm_segsz + PGOFSET) & ~PGOFSET;
168
169 (*uobj->pgops->pgo_detach)(uobj);
170 free((caddr_t)shm_handle, M_SHM);
171 shmseg->_shm_internal = NULL;
172 shm_committed -= btoc(size);
173 shmseg->shm_perm.mode = SHMSEG_FREE;
174 shm_nused--;
175 }
176
177 static int
178 shm_delete_mapping(vm, shmmap_s)
179 struct vmspace *vm;
180 struct shmmap_state *shmmap_s;
181 {
182 struct shmid_ds *shmseg;
183 int segnum, result;
184 size_t size;
185
186 segnum = IPCID_TO_IX(shmmap_s->shmid);
187 shmseg = &shmsegs[segnum];
188 size = (shmseg->shm_segsz + PGOFSET) & ~PGOFSET;
189 result = uvm_deallocate(&vm->vm_map, shmmap_s->va, size);
190 if (result != KERN_SUCCESS)
191 return EINVAL;
192 shmmap_s->shmid = -1;
193 shmseg->shm_dtime = time.tv_sec;
194 if ((--shmseg->shm_nattch <= 0) &&
195 (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
196 shm_deallocate_segment(shmseg);
197 shm_last_free = segnum;
198 }
199 return 0;
200 }
201
202 int
203 sys_shmdt(p, v, retval)
204 struct proc *p;
205 void *v;
206 register_t *retval;
207 {
208 struct sys_shmdt_args /* {
209 syscallarg(const void *) shmaddr;
210 } */ *uap = v;
211 struct shmmap_state *shmmap_s;
212 int i;
213
214 shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
215 if (shmmap_s == NULL)
216 return EINVAL;
217
218 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
219 if (shmmap_s->shmid != -1 &&
220 shmmap_s->va == (vaddr_t)SCARG(uap, shmaddr))
221 break;
222 if (i == shminfo.shmseg)
223 return EINVAL;
224 return shm_delete_mapping(p->p_vmspace, shmmap_s);
225 }
226
227 int
228 sys_shmat(p, v, retval)
229 struct proc *p;
230 void *v;
231 register_t *retval;
232 {
233 struct sys_shmat_args /* {
234 syscallarg(int) shmid;
235 syscallarg(const void *) shmaddr;
236 syscallarg(int) shmflg;
237 } */ *uap = v;
238 int error, i, flags;
239 struct ucred *cred = p->p_ucred;
240 struct shmid_ds *shmseg;
241 struct shmmap_state *shmmap_s = NULL;
242 struct uvm_object *uobj;
243 vaddr_t attach_va;
244 vm_prot_t prot;
245 vsize_t size;
246 int rv;
247
248 shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
249 if (shmmap_s == NULL) {
250 size = shminfo.shmseg * sizeof(struct shmmap_state);
251 shmmap_s = malloc(size, M_SHM, M_WAITOK);
252 for (i = 0; i < shminfo.shmseg; i++)
253 shmmap_s[i].shmid = -1;
254 p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
255 }
256 shmseg = shm_find_segment_by_shmid(SCARG(uap, shmid));
257 if (shmseg == NULL)
258 return EINVAL;
259 error = ipcperm(cred, &shmseg->shm_perm,
260 (SCARG(uap, shmflg) & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
261 if (error)
262 return error;
263 for (i = 0; i < shminfo.shmseg; i++) {
264 if (shmmap_s->shmid == -1)
265 break;
266 shmmap_s++;
267 }
268 if (i >= shminfo.shmseg)
269 return EMFILE;
270 size = (shmseg->shm_segsz + PGOFSET) & ~PGOFSET;
271 prot = VM_PROT_READ;
272 if ((SCARG(uap, shmflg) & SHM_RDONLY) == 0)
273 prot |= VM_PROT_WRITE;
274 flags = MAP_ANON | MAP_SHARED;
275 if (SCARG(uap, shmaddr)) {
276 flags |= MAP_FIXED;
277 if (SCARG(uap, shmflg) & SHM_RND)
278 attach_va =
279 (vaddr_t)SCARG(uap, shmaddr) & ~(SHMLBA-1);
280 else if (((vaddr_t)SCARG(uap, shmaddr) & (SHMLBA-1)) == 0)
281 attach_va = (vaddr_t)SCARG(uap, shmaddr);
282 else
283 return EINVAL;
284 } else {
285 /* This is just a hint to vm_mmap() about where to put it. */
286 attach_va =
287 round_page((vaddr_t)p->p_vmspace->vm_taddr +
288 MAXTSIZ + MAXDSIZ);
289 }
290 uobj = ((struct shm_handle *)shmseg->_shm_internal)->shm_object;
291 (*uobj->pgops->pgo_reference)(uobj);
292 rv = uvm_map(&p->p_vmspace->vm_map, &attach_va, size,
293 uobj, 0,
294 UVM_MAPFLAG(prot, prot, UVM_INH_SHARE,
295 UVM_ADV_RANDOM, 0));
296 if (rv != KERN_SUCCESS) {
297 (*uobj->pgops->pgo_detach)(uobj);
298 return ENOMEM;
299 }
300
301 shmmap_s->va = attach_va;
302 shmmap_s->shmid = SCARG(uap, shmid);
303 shmseg->shm_lpid = p->p_pid;
304 shmseg->shm_atime = time.tv_sec;
305 shmseg->shm_nattch++;
306 *retval = attach_va;
307 return 0;
308 }
309
310 int
311 sys___shmctl13(p, v, retval)
312 struct proc *p;
313 void *v;
314 register_t *retval;
315 {
316 struct sys___shmctl13_args /* {
317 syscallarg(int) shmid;
318 syscallarg(int) cmd;
319 syscallarg(struct shmid_ds *) buf;
320 } */ *uap = v;
321 struct shmid_ds shmbuf;
322 int cmd, error;
323
324 cmd = SCARG(uap, cmd);
325
326 if (cmd == IPC_SET) {
327 error = copyin(SCARG(uap, buf), &shmbuf, sizeof(shmbuf));
328 if (error)
329 return (error);
330 }
331
332 error = shmctl1(p, SCARG(uap, shmid), cmd,
333 (cmd == IPC_SET || cmd == IPC_STAT) ? &shmbuf : NULL);
334
335 if (error == 0 && cmd == IPC_STAT)
336 error = copyout(&shmbuf, SCARG(uap, buf), sizeof(shmbuf));
337
338 return (error);
339 }
340
341 int
342 shmctl1(p, shmid, cmd, shmbuf)
343 struct proc *p;
344 int shmid;
345 int cmd;
346 struct shmid_ds *shmbuf;
347 {
348 struct ucred *cred = p->p_ucred;
349 struct shmid_ds *shmseg;
350 int error = 0;
351
352 shmseg = shm_find_segment_by_shmid(shmid);
353 if (shmseg == NULL)
354 return EINVAL;
355 switch (cmd) {
356 case IPC_STAT:
357 if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_R)) != 0)
358 return error;
359 memcpy(shmbuf, shmseg, sizeof(struct shmid_ds));
360 break;
361 case IPC_SET:
362 if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_M)) != 0)
363 return error;
364 shmseg->shm_perm.uid = shmbuf->shm_perm.uid;
365 shmseg->shm_perm.gid = shmbuf->shm_perm.gid;
366 shmseg->shm_perm.mode =
367 (shmseg->shm_perm.mode & ~ACCESSPERMS) |
368 (shmbuf->shm_perm.mode & ACCESSPERMS);
369 shmseg->shm_ctime = time.tv_sec;
370 break;
371 case IPC_RMID:
372 if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_M)) != 0)
373 return error;
374 shmseg->shm_perm._key = IPC_PRIVATE;
375 shmseg->shm_perm.mode |= SHMSEG_REMOVED;
376 if (shmseg->shm_nattch <= 0) {
377 shm_deallocate_segment(shmseg);
378 shm_last_free = IPCID_TO_IX(shmid);
379 }
380 break;
381 case SHM_LOCK:
382 case SHM_UNLOCK:
383 default:
384 return EINVAL;
385 }
386 return 0;
387 }
388
389 static int
390 shmget_existing(p, uap, mode, segnum, retval)
391 struct proc *p;
392 struct sys_shmget_args /* {
393 syscallarg(key_t) key;
394 syscallarg(size_t) size;
395 syscallarg(int) shmflg;
396 } */ *uap;
397 int mode;
398 int segnum;
399 register_t *retval;
400 {
401 struct shmid_ds *shmseg;
402 struct ucred *cred = p->p_ucred;
403 int error;
404
405 shmseg = &shmsegs[segnum];
406 if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
407 /*
408 * This segment is in the process of being allocated. Wait
409 * until it's done, and look the key up again (in case the
410 * allocation failed or it was freed).
411 */
412 shmseg->shm_perm.mode |= SHMSEG_WANTED;
413 error = tsleep((caddr_t)shmseg, PLOCK | PCATCH, "shmget", 0);
414 if (error)
415 return error;
416 return EAGAIN;
417 }
418 if ((error = ipcperm(cred, &shmseg->shm_perm, mode)) != 0)
419 return error;
420 if (SCARG(uap, size) && SCARG(uap, size) > shmseg->shm_segsz)
421 return EINVAL;
422 if ((SCARG(uap, shmflg) & (IPC_CREAT | IPC_EXCL)) ==
423 (IPC_CREAT | IPC_EXCL))
424 return EEXIST;
425 *retval = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
426 return 0;
427 }
428
429 static int
430 shmget_allocate_segment(p, uap, mode, retval)
431 struct proc *p;
432 struct sys_shmget_args /* {
433 syscallarg(key_t) key;
434 syscallarg(size_t) size;
435 syscallarg(int) shmflg;
436 } */ *uap;
437 int mode;
438 register_t *retval;
439 {
440 int i, segnum, shmid, size;
441 struct ucred *cred = p->p_ucred;
442 struct shmid_ds *shmseg;
443 struct shm_handle *shm_handle;
444 int error = 0;
445
446 if (SCARG(uap, size) < shminfo.shmmin ||
447 SCARG(uap, size) > shminfo.shmmax)
448 return EINVAL;
449 if (shm_nused >= shminfo.shmmni) /* any shmids left? */
450 return ENOSPC;
451 size = (SCARG(uap, size) + PGOFSET) & ~PGOFSET;
452 if (shm_committed + btoc(size) > shminfo.shmall)
453 return ENOMEM;
454 if (shm_last_free < 0) {
455 for (i = 0; i < shminfo.shmmni; i++)
456 if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
457 break;
458 if (i == shminfo.shmmni)
459 panic("shmseg free count inconsistent");
460 segnum = i;
461 } else {
462 segnum = shm_last_free;
463 shm_last_free = -1;
464 }
465 shmseg = &shmsegs[segnum];
466 /*
467 * In case we sleep in malloc(), mark the segment present but deleted
468 * so that noone else tries to create the same key.
469 */
470 shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
471 shmseg->shm_perm._key = SCARG(uap, key);
472 shmseg->shm_perm._seq = (shmseg->shm_perm._seq + 1) & 0x7fff;
473 shm_handle = (struct shm_handle *)
474 malloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
475 shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
476
477 shm_handle->shm_object = uao_create(size, 0);
478
479 shmseg->_shm_internal = shm_handle;
480 shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
481 shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
482 shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
483 (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
484 shmseg->shm_segsz = SCARG(uap, size);
485 shmseg->shm_cpid = p->p_pid;
486 shmseg->shm_lpid = shmseg->shm_nattch = 0;
487 shmseg->shm_atime = shmseg->shm_dtime = 0;
488 shmseg->shm_ctime = time.tv_sec;
489 shm_committed += btoc(size);
490 shm_nused++;
491
492 *retval = shmid;
493 if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
494 /*
495 * Somebody else wanted this key while we were asleep. Wake
496 * them up now.
497 */
498 shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
499 wakeup((caddr_t)shmseg);
500 }
501 return error;
502 }
503
504 int
505 sys_shmget(p, v, retval)
506 struct proc *p;
507 void *v;
508 register_t *retval;
509 {
510 struct sys_shmget_args /* {
511 syscallarg(key_t) key;
512 syscallarg(int) size;
513 syscallarg(int) shmflg;
514 } */ *uap = v;
515 int segnum, mode, error;
516
517 mode = SCARG(uap, shmflg) & ACCESSPERMS;
518 if (SCARG(uap, key) != IPC_PRIVATE) {
519 again:
520 segnum = shm_find_segment_by_key(SCARG(uap, key));
521 if (segnum >= 0) {
522 error = shmget_existing(p, uap, mode, segnum, retval);
523 if (error == EAGAIN)
524 goto again;
525 return error;
526 }
527 if ((SCARG(uap, shmflg) & IPC_CREAT) == 0)
528 return ENOENT;
529 }
530 return shmget_allocate_segment(p, uap, mode, retval);
531 }
532
533 void
534 shmfork(vm1, vm2)
535 struct vmspace *vm1, *vm2;
536 {
537 struct shmmap_state *shmmap_s;
538 size_t size;
539 int i;
540
541 if (vm1->vm_shm == NULL) {
542 vm2->vm_shm = NULL;
543 return;
544 }
545
546 size = shminfo.shmseg * sizeof(struct shmmap_state);
547 shmmap_s = malloc(size, M_SHM, M_WAITOK);
548 memcpy(shmmap_s, vm1->vm_shm, size);
549 vm2->vm_shm = (caddr_t)shmmap_s;
550 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
551 if (shmmap_s->shmid != -1)
552 shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
553 }
554
555 void
556 shmexit(vm)
557 struct vmspace *vm;
558 {
559 struct shmmap_state *shmmap_s;
560 int i;
561
562 shmmap_s = (struct shmmap_state *)vm->vm_shm;
563 if (shmmap_s == NULL)
564 return;
565 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
566 if (shmmap_s->shmid != -1)
567 shm_delete_mapping(vm, shmmap_s);
568 free(vm->vm_shm, M_SHM);
569 vm->vm_shm = NULL;
570 }
571
572 void
573 shminit()
574 {
575 int i;
576
577 shminfo.shmmax *= NBPG;
578
579 for (i = 0; i < shminfo.shmmni; i++) {
580 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
581 shmsegs[i].shm_perm._seq = 0;
582 }
583 shm_last_free = 0;
584 shm_nused = 0;
585 shm_committed = 0;
586 }
587