FreeBSD/Linux Kernel Cross Reference
sys/kern/sysv_shm.c
1 /* $NetBSD: sysv_shm.c,v 1.93 2006/11/28 20:35:16 ad 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 #include <sys/cdefs.h>
71 __KERNEL_RCSID(0, "$NetBSD: sysv_shm.c,v 1.93 2006/11/28 20:35:16 ad Exp $");
72
73 #define SYSVSHM
74
75 #include <sys/param.h>
76 #include <sys/kernel.h>
77 #include <sys/shm.h>
78 #include <sys/lock.h>
79 #include <sys/malloc.h>
80 #include <sys/mman.h>
81 #include <sys/stat.h>
82 #include <sys/sysctl.h>
83 #include <sys/mount.h> /* XXX for <sys/syscallargs.h> */
84 #include <sys/sa.h>
85 #include <sys/syscallargs.h>
86 #include <sys/queue.h>
87 #include <sys/pool.h>
88 #include <sys/kauth.h>
89
90 #include <uvm/uvm_extern.h>
91 #include <uvm/uvm_object.h>
92
93 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
94
95 /*
96 * Provides the following externally accessible functions:
97 *
98 * shminit(void); initialization
99 * shmexit(struct vmspace *) cleanup
100 * shmfork(struct vmspace *, struct vmspace *) fork handling
101 *
102 * Structures:
103 * shmsegs (an array of 'struct shmid_ds')
104 * per proc array of 'struct shmmap_state'
105 */
106
107 int shm_nused;
108 struct shmid_ds *shmsegs;
109
110 struct shmmap_entry {
111 SLIST_ENTRY(shmmap_entry) next;
112 vaddr_t va;
113 int shmid;
114 };
115
116 struct lock shm_lock;
117 static int shm_last_free, shm_committed, shm_use_phys;
118
119 static POOL_INIT(shmmap_entry_pool, sizeof(struct shmmap_entry), 0, 0, 0,
120 "shmmp", &pool_allocator_nointr);
121
122 struct shmmap_state {
123 unsigned int nitems;
124 unsigned int nrefs;
125 SLIST_HEAD(, shmmap_entry) entries;
126 };
127
128 static int shm_find_segment_by_key(key_t);
129 static void shm_deallocate_segment(struct shmid_ds *);
130 static void shm_delete_mapping(struct vmspace *, struct shmmap_state *,
131 struct shmmap_entry *);
132 static int shmget_existing(struct lwp *, struct sys_shmget_args *,
133 int, int, register_t *);
134 static int shmget_allocate_segment(struct lwp *, struct sys_shmget_args *,
135 int, register_t *);
136 static struct shmmap_state *shmmap_getprivate(struct proc *);
137 static struct shmmap_entry *shm_find_mapping(struct shmmap_state *, vaddr_t);
138 static int shmrealloc(int);
139
140 static int
141 shm_find_segment_by_key(key_t key)
142 {
143 int i;
144
145 for (i = 0; i < shminfo.shmmni; i++)
146 if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
147 shmsegs[i].shm_perm._key == key)
148 return i;
149 return -1;
150 }
151
152 static struct shmid_ds *
153 shm_find_segment_by_shmid(int shmid)
154 {
155 int segnum;
156 struct shmid_ds *shmseg;
157
158 segnum = IPCID_TO_IX(shmid);
159 if (segnum < 0 || segnum >= shminfo.shmmni)
160 return NULL;
161 shmseg = &shmsegs[segnum];
162 if ((shmseg->shm_perm.mode & SHMSEG_ALLOCATED) == 0)
163 return NULL;
164 if ((shmseg->shm_perm.mode & (SHMSEG_REMOVED|SHMSEG_RMLINGER)) == SHMSEG_REMOVED)
165 return NULL;
166 if (shmseg->shm_perm._seq != IPCID_TO_SEQ(shmid))
167 return NULL;
168 return shmseg;
169 }
170
171 static void
172 shm_deallocate_segment(struct shmid_ds *shmseg)
173 {
174 struct uvm_object *uobj = shmseg->_shm_internal;
175 size_t size = (shmseg->shm_segsz + PGOFSET) & ~PGOFSET;
176
177 #ifdef SHMDEBUG
178 printf("shm freeing key 0x%lx seq 0x%x\n",
179 shmseg->shm_perm._key, shmseg->shm_perm._seq);
180 #endif
181
182 (*uobj->pgops->pgo_detach)(uobj);
183 shmseg->_shm_internal = NULL;
184 shm_committed -= btoc(size);
185 shmseg->shm_perm.mode = SHMSEG_FREE;
186 shm_nused--;
187 }
188
189 static void
190 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s,
191 struct shmmap_entry *shmmap_se)
192 {
193 struct shmid_ds *shmseg;
194 int segnum;
195 size_t size;
196
197 segnum = IPCID_TO_IX(shmmap_se->shmid);
198 #ifdef DEBUG
199 if (segnum < 0 || segnum >= shminfo.shmmni)
200 panic("shm_delete_mapping: vmspace %p state %p entry %p - "
201 "entry segment ID bad (%d)",
202 vm, shmmap_s, shmmap_se, segnum);
203 #endif
204 shmseg = &shmsegs[segnum];
205 size = (shmseg->shm_segsz + PGOFSET) & ~PGOFSET;
206 uvm_deallocate(&vm->vm_map, shmmap_se->va, size);
207 SLIST_REMOVE(&shmmap_s->entries, shmmap_se, shmmap_entry, next);
208 shmmap_s->nitems--;
209 pool_put(&shmmap_entry_pool, shmmap_se);
210 shmseg->shm_dtime = time_second;
211 if ((--shmseg->shm_nattch <= 0) &&
212 (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
213 shm_deallocate_segment(shmseg);
214 shm_last_free = segnum;
215 }
216 }
217
218 /*
219 * Get a non-shared shm map for that vmspace.
220 * 3 cases:
221 * - no shm map present: create a fresh one
222 * - a shm map with refcount=1, just used by ourselves: fine
223 * - a shared shm map: copy to a fresh one and adjust refcounts
224 */
225 static struct shmmap_state *
226 shmmap_getprivate(struct proc *p)
227 {
228 struct shmmap_state *oshmmap_s, *shmmap_s;
229 struct shmmap_entry *oshmmap_se, *shmmap_se;
230
231 oshmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
232 if (oshmmap_s && oshmmap_s->nrefs == 1)
233 return (oshmmap_s);
234
235 shmmap_s = malloc(sizeof(struct shmmap_state), M_SHM, M_WAITOK);
236 memset(shmmap_s, 0, sizeof(struct shmmap_state));
237 shmmap_s->nrefs = 1;
238 SLIST_INIT(&shmmap_s->entries);
239 p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
240
241 if (!oshmmap_s)
242 return (shmmap_s);
243
244 #ifdef SHMDEBUG
245 printf("shmmap_getprivate: vm %p split (%d entries), was used by %d\n",
246 p->p_vmspace, oshmmap_s->nitems, oshmmap_s->nrefs);
247 #endif
248 SLIST_FOREACH(oshmmap_se, &oshmmap_s->entries, next) {
249 shmmap_se = pool_get(&shmmap_entry_pool, PR_WAITOK);
250 shmmap_se->va = oshmmap_se->va;
251 shmmap_se->shmid = oshmmap_se->shmid;
252 SLIST_INSERT_HEAD(&shmmap_s->entries, shmmap_se, next);
253 }
254 shmmap_s->nitems = oshmmap_s->nitems;
255 oshmmap_s->nrefs--;
256 return (shmmap_s);
257 }
258
259 static struct shmmap_entry *
260 shm_find_mapping(struct shmmap_state *map, vaddr_t va)
261 {
262 struct shmmap_entry *shmmap_se;
263
264 SLIST_FOREACH(shmmap_se, &map->entries, next) {
265 if (shmmap_se->va == va)
266 return shmmap_se;
267 }
268 return 0;
269 }
270
271 int
272 sys_shmdt(struct lwp *l, void *v, register_t *retval)
273 {
274 struct sys_shmdt_args /* {
275 syscallarg(const void *) shmaddr;
276 } */ *uap = v;
277 struct proc *p = l->l_proc;
278 struct shmmap_state *shmmap_s, *shmmap_s1;
279 struct shmmap_entry *shmmap_se;
280
281 shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
282 if (shmmap_s == NULL)
283 return EINVAL;
284
285 shmmap_se = shm_find_mapping(shmmap_s, (vaddr_t)SCARG(uap, shmaddr));
286 if (!shmmap_se)
287 return EINVAL;
288
289 shmmap_s1 = shmmap_getprivate(p);
290 if (shmmap_s1 != shmmap_s) {
291 /* map has been copied, lookup entry in new map */
292 shmmap_se = shm_find_mapping(shmmap_s1,
293 (vaddr_t)SCARG(uap, shmaddr));
294 KASSERT(shmmap_se != NULL);
295 }
296 #ifdef SHMDEBUG
297 printf("shmdt: vm %p: remove %d @%lx\n",
298 p->p_vmspace, shmmap_se->shmid, shmmap_se->va);
299 #endif
300 shm_delete_mapping(p->p_vmspace, shmmap_s1, shmmap_se);
301 return 0;
302 }
303
304 int
305 sys_shmat(struct lwp *l, void *v, register_t *retval)
306 {
307 struct sys_shmat_args /* {
308 syscallarg(int) shmid;
309 syscallarg(const void *) shmaddr;
310 syscallarg(int) shmflg;
311 } */ *uap = v;
312 int error, flags;
313 struct proc *p = l->l_proc;
314 kauth_cred_t cred = l->l_cred;
315 struct shmid_ds *shmseg;
316 struct shmmap_state *shmmap_s;
317 struct uvm_object *uobj;
318 vaddr_t attach_va;
319 vm_prot_t prot;
320 vsize_t size;
321 struct shmmap_entry *shmmap_se;
322
323 shmseg = shm_find_segment_by_shmid(SCARG(uap, shmid));
324 if (shmseg == NULL)
325 return EINVAL;
326 error = ipcperm(cred, &shmseg->shm_perm,
327 (SCARG(uap, shmflg) & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
328 if (error)
329 return error;
330
331 shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
332 if (shmmap_s && shmmap_s->nitems >= shminfo.shmseg)
333 return EMFILE;
334
335 size = (shmseg->shm_segsz + PGOFSET) & ~PGOFSET;
336 prot = VM_PROT_READ;
337 if ((SCARG(uap, shmflg) & SHM_RDONLY) == 0)
338 prot |= VM_PROT_WRITE;
339 flags = MAP_ANON | MAP_SHARED;
340 if (SCARG(uap, shmaddr)) {
341 flags |= MAP_FIXED;
342 if (SCARG(uap, shmflg) & SHM_RND)
343 attach_va =
344 (vaddr_t)SCARG(uap, shmaddr) & ~(SHMLBA-1);
345 else if (((vaddr_t)SCARG(uap, shmaddr) & (SHMLBA-1)) == 0)
346 attach_va = (vaddr_t)SCARG(uap, shmaddr);
347 else
348 return EINVAL;
349 } else {
350 /* This is just a hint to uvm_mmap() about where to put it. */
351 attach_va = p->p_emul->e_vm_default_addr(p,
352 (vaddr_t)p->p_vmspace->vm_daddr, size);
353 }
354 uobj = shmseg->_shm_internal;
355 (*uobj->pgops->pgo_reference)(uobj);
356 error = uvm_map(&p->p_vmspace->vm_map, &attach_va, size,
357 uobj, 0, 0,
358 UVM_MAPFLAG(prot, prot, UVM_INH_SHARE, UVM_ADV_RANDOM, 0));
359 if (error)
360 goto out;
361 /* Lock the memory */
362 if (shm_use_phys || (shmseg->shm_perm.mode & SHMSEG_WIRED)) {
363 /* Wire the map */
364 error = uvm_map_pageable(&p->p_vmspace->vm_map, attach_va,
365 attach_va + size, FALSE, 0);
366 if (error) {
367 if (error == EFAULT)
368 error = ENOMEM;
369 goto out;
370 }
371 }
372
373 shmmap_se = pool_get(&shmmap_entry_pool, PR_WAITOK);
374 shmmap_se->va = attach_va;
375 shmmap_se->shmid = SCARG(uap, shmid);
376 shmmap_s = shmmap_getprivate(p);
377 #ifdef SHMDEBUG
378 printf("shmat: vm %p: add %d @%lx\n", p->p_vmspace, shmmap_se->shmid, attach_va);
379 #endif
380 SLIST_INSERT_HEAD(&shmmap_s->entries, shmmap_se, next);
381 shmmap_s->nitems++;
382 shmseg->shm_lpid = p->p_pid;
383 shmseg->shm_atime = time_second;
384 shmseg->shm_nattch++;
385
386 retval[0] = attach_va;
387 return 0;
388 out:
389 (*uobj->pgops->pgo_detach)(uobj);
390 return error;
391 }
392
393 int
394 sys___shmctl13(struct lwp *l, void *v, register_t *retval)
395 {
396 struct sys___shmctl13_args /* {
397 syscallarg(int) shmid;
398 syscallarg(int) cmd;
399 syscallarg(struct shmid_ds *) buf;
400 } */ *uap = v;
401 struct shmid_ds shmbuf;
402 int cmd, error;
403
404 cmd = SCARG(uap, cmd);
405
406 if (cmd == IPC_SET) {
407 error = copyin(SCARG(uap, buf), &shmbuf, sizeof(shmbuf));
408 if (error)
409 return (error);
410 }
411
412 error = shmctl1(l, SCARG(uap, shmid), cmd,
413 (cmd == IPC_SET || cmd == IPC_STAT) ? &shmbuf : NULL);
414
415 if (error == 0 && cmd == IPC_STAT)
416 error = copyout(&shmbuf, SCARG(uap, buf), sizeof(shmbuf));
417
418 return (error);
419 }
420
421 int
422 shmctl1(struct lwp *l, int shmid, int cmd, struct shmid_ds *shmbuf)
423 {
424 kauth_cred_t cred = l->l_cred;
425 struct proc *p = l->l_proc;
426 struct shmid_ds *shmseg;
427 struct shmmap_entry *shmmap_se;
428 struct shmmap_state *shmmap_s;
429 int error = 0;
430 size_t size;
431
432 shmseg = shm_find_segment_by_shmid(shmid);
433 if (shmseg == NULL)
434 return EINVAL;
435
436 switch (cmd) {
437 case IPC_STAT:
438 if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_R)) != 0)
439 return error;
440 memcpy(shmbuf, shmseg, sizeof(struct shmid_ds));
441 break;
442 case IPC_SET:
443 if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_M)) != 0)
444 return error;
445 shmseg->shm_perm.uid = shmbuf->shm_perm.uid;
446 shmseg->shm_perm.gid = shmbuf->shm_perm.gid;
447 shmseg->shm_perm.mode =
448 (shmseg->shm_perm.mode & ~ACCESSPERMS) |
449 (shmbuf->shm_perm.mode & ACCESSPERMS);
450 shmseg->shm_ctime = time_second;
451 break;
452 case IPC_RMID:
453 if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_M)) != 0)
454 return error;
455 shmseg->shm_perm._key = IPC_PRIVATE;
456 shmseg->shm_perm.mode |= SHMSEG_REMOVED;
457 if (shmseg->shm_nattch <= 0) {
458 shm_deallocate_segment(shmseg);
459 shm_last_free = IPCID_TO_IX(shmid);
460 }
461 break;
462 case SHM_LOCK:
463 case SHM_UNLOCK:
464 if ((error = kauth_authorize_generic(cred,
465 KAUTH_GENERIC_ISSUSER, NULL)) != 0)
466 return error;
467 shmmap_s = shmmap_getprivate(p);
468 /* Find our shared memory address by shmid */
469 SLIST_FOREACH(shmmap_se, &shmmap_s->entries, next) {
470 if (shmmap_se->shmid != shmid)
471 continue;
472
473 size = (shmseg->shm_segsz + PGOFSET) & ~PGOFSET;
474
475 if (cmd == SHM_LOCK &&
476 !(shmseg->shm_perm.mode & SHMSEG_WIRED)) {
477 /* Wire the entire object */
478 error = uobj_wirepages(shmseg->_shm_internal, 0,
479 round_page(shmseg->shm_segsz));
480 if (error)
481 return EIO;
482 /* Wire the map */
483 error = uvm_map_pageable(&p->p_vmspace->vm_map,
484 shmmap_se->va, shmmap_se->va + size, FALSE,
485 0);
486 if (error) {
487 uobj_unwirepages(shmseg->_shm_internal,
488 0, round_page(shmseg->shm_segsz));
489 if (error == EFAULT)
490 error = ENOMEM;
491 return error;
492 }
493 /* Tag as wired */
494 shmseg->shm_perm.mode |= SHMSEG_WIRED;
495
496 } else if (cmd == SHM_UNLOCK &&
497 (shmseg->shm_perm.mode & SHMSEG_WIRED)) {
498 /* Unwire the object */
499 uobj_unwirepages(shmseg->_shm_internal, 0,
500 round_page(shmseg->shm_segsz));
501 error = uvm_map_pageable(&p->p_vmspace->vm_map,
502 shmmap_se->va, shmmap_se->va + size, TRUE,
503 0);
504 if (error) {
505 /*
506 * In fact, uvm_map_pageable could fail
507 * only if arguments are invalid,
508 * otherwise it should allways return 0.
509 */
510 return EIO;
511 }
512 /* Tag as unwired */
513 shmseg->shm_perm.mode &= ~SHMSEG_WIRED;
514 }
515 }
516 break;
517 default:
518 return EINVAL;
519 }
520 return 0;
521 }
522
523 static int
524 shmget_existing(struct lwp *l, struct sys_shmget_args *uap, int mode,
525 int segnum, register_t *retval)
526 {
527 struct shmid_ds *shmseg;
528 kauth_cred_t cred = l->l_cred;
529 int error;
530
531 shmseg = &shmsegs[segnum];
532 if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
533 /*
534 * This segment is in the process of being allocated. Wait
535 * until it's done, and look the key up again (in case the
536 * allocation failed or it was freed).
537 */
538 shmseg->shm_perm.mode |= SHMSEG_WANTED;
539 error = tsleep((caddr_t)shmseg, PLOCK | PCATCH, "shmget", 0);
540 if (error)
541 return error;
542 return EAGAIN;
543 }
544 if ((error = ipcperm(cred, &shmseg->shm_perm, mode)) != 0)
545 return error;
546 if (SCARG(uap, size) && SCARG(uap, size) > shmseg->shm_segsz)
547 return EINVAL;
548 if ((SCARG(uap, shmflg) & (IPC_CREAT | IPC_EXCL)) ==
549 (IPC_CREAT | IPC_EXCL))
550 return EEXIST;
551 *retval = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
552 return 0;
553 }
554
555 static int
556 shmget_allocate_segment(struct lwp *l, struct sys_shmget_args *uap, int mode,
557 register_t *retval)
558 {
559 int i, segnum, shmid, size;
560 kauth_cred_t cred = l->l_cred;
561 struct shmid_ds *shmseg;
562 int error = 0;
563
564 if (SCARG(uap, size) < shminfo.shmmin ||
565 SCARG(uap, size) > shminfo.shmmax)
566 return EINVAL;
567 if (shm_nused >= shminfo.shmmni) /* any shmids left? */
568 return ENOSPC;
569 size = (SCARG(uap, size) + PGOFSET) & ~PGOFSET;
570 if (shm_committed + btoc(size) > shminfo.shmall)
571 return ENOMEM;
572 if (shm_last_free < 0) {
573 for (i = 0; i < shminfo.shmmni; i++)
574 if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
575 break;
576 if (i == shminfo.shmmni)
577 panic("shmseg free count inconsistent");
578 segnum = i;
579 } else {
580 segnum = shm_last_free;
581 shm_last_free = -1;
582 }
583 shmseg = &shmsegs[segnum];
584 /*
585 * In case we sleep in malloc(), mark the segment present but deleted
586 * so that noone else tries to create the same key.
587 */
588 shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
589 shmseg->shm_perm._key = SCARG(uap, key);
590 shmseg->shm_perm._seq = (shmseg->shm_perm._seq + 1) & 0x7fff;
591 shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
592
593 shmseg->_shm_internal = uao_create(size, 0);
594
595 shmseg->shm_perm.cuid = shmseg->shm_perm.uid = kauth_cred_geteuid(cred);
596 shmseg->shm_perm.cgid = shmseg->shm_perm.gid = kauth_cred_getegid(cred);
597 shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
598 (mode & (ACCESSPERMS|SHMSEG_RMLINGER)) | SHMSEG_ALLOCATED;
599 shmseg->shm_segsz = SCARG(uap, size);
600 shmseg->shm_cpid = l->l_proc->p_pid;
601 shmseg->shm_lpid = shmseg->shm_nattch = 0;
602 shmseg->shm_atime = shmseg->shm_dtime = 0;
603 shmseg->shm_ctime = time_second;
604 shm_committed += btoc(size);
605 shm_nused++;
606
607 *retval = shmid;
608 if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
609 /*
610 * Somebody else wanted this key while we were asleep. Wake
611 * them up now.
612 */
613 shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
614 wakeup((caddr_t)shmseg);
615 }
616
617 /* Lock the memory */
618 if (shm_use_phys) {
619 /* Wire the entire object */
620 error = uobj_wirepages(shmseg->_shm_internal, 0,
621 round_page(shmseg->shm_segsz));
622 if (error) {
623 shm_deallocate_segment(shmseg);
624 } else {
625 /* Tag as wired */
626 shmseg->shm_perm.mode |= SHMSEG_WIRED;
627 }
628 }
629
630 return error;
631 }
632
633 int
634 sys_shmget(struct lwp *l, void *v, register_t *retval)
635 {
636 struct sys_shmget_args /* {
637 syscallarg(key_t) key;
638 syscallarg(int) size;
639 syscallarg(int) shmflg;
640 } */ *uap = v;
641 int segnum, mode, error;
642
643 mode = SCARG(uap, shmflg) & ACCESSPERMS;
644 if (SCARG(uap, shmflg) & _SHM_RMLINGER)
645 mode |= SHMSEG_RMLINGER;
646
647 #ifdef SHMDEBUG
648 printf("shmget: key 0x%lx size 0x%x shmflg 0x%x mode 0x%x\n",
649 SCARG(uap, key), SCARG(uap, size), SCARG(uap, shmflg), mode);
650 #endif
651
652 if (SCARG(uap, key) != IPC_PRIVATE) {
653 again:
654 segnum = shm_find_segment_by_key(SCARG(uap, key));
655 if (segnum >= 0) {
656 error = shmget_existing(l, uap, mode, segnum, retval);
657 if (error == EAGAIN)
658 goto again;
659 return error;
660 }
661 if ((SCARG(uap, shmflg) & IPC_CREAT) == 0)
662 return ENOENT;
663 }
664 return shmget_allocate_segment(l, uap, mode, retval);
665 }
666
667 void
668 shmfork(struct vmspace *vm1, struct vmspace *vm2)
669 {
670 struct shmmap_state *shmmap_s;
671 struct shmmap_entry *shmmap_se;
672
673 vm2->vm_shm = vm1->vm_shm;
674
675 if (vm1->vm_shm == NULL)
676 return;
677
678 #ifdef SHMDEBUG
679 printf("shmfork %p->%p\n", vm1, vm2);
680 #endif
681
682 shmmap_s = (struct shmmap_state *)vm1->vm_shm;
683
684 SLIST_FOREACH(shmmap_se, &shmmap_s->entries, next)
685 shmsegs[IPCID_TO_IX(shmmap_se->shmid)].shm_nattch++;
686 shmmap_s->nrefs++;
687 }
688
689 void
690 shmexit(struct vmspace *vm)
691 {
692 struct shmmap_state *shmmap_s;
693 struct shmmap_entry *shmmap_se;
694
695 shmmap_s = (struct shmmap_state *)vm->vm_shm;
696 if (shmmap_s == NULL)
697 return;
698
699 vm->vm_shm = NULL;
700
701 if (--shmmap_s->nrefs > 0) {
702 #ifdef SHMDEBUG
703 printf("shmexit: vm %p drop ref (%d entries), now used by %d\n",
704 vm, shmmap_s->nitems, shmmap_s->nrefs);
705 #endif
706 SLIST_FOREACH(shmmap_se, &shmmap_s->entries, next)
707 shmsegs[IPCID_TO_IX(shmmap_se->shmid)].shm_nattch--;
708 return;
709 }
710
711 #ifdef SHMDEBUG
712 printf("shmexit: vm %p cleanup (%d entries)\n", vm, shmmap_s->nitems);
713 #endif
714 while (!SLIST_EMPTY(&shmmap_s->entries)) {
715 shmmap_se = SLIST_FIRST(&shmmap_s->entries);
716 shm_delete_mapping(vm, shmmap_s, shmmap_se);
717 }
718 KASSERT(shmmap_s->nitems == 0);
719 free(shmmap_s, M_SHM);
720 }
721
722 static int
723 shmrealloc(int newshmni)
724 {
725 int i, sz;
726 vaddr_t v;
727 struct shmid_ds *newshmsegs;
728
729 /* XXX: Would be good to have a upper limit */
730 if (newshmni < 1)
731 return EINVAL;
732
733 /* We can't reallocate lesser memory than we use */
734 if (shm_nused > newshmni)
735 return EPERM;
736
737 /* Allocate new memory area */
738 sz = newshmni * sizeof(struct shmid_ds);
739 v = uvm_km_alloc(kernel_map, round_page(sz), 0, UVM_KMF_WIRED);
740 if (v == 0)
741 return ENOMEM;
742
743 newshmsegs = (void *)v;
744
745 /* Copy all memory to the new area */
746 for (i = 0; i < shm_nused; i++)
747 (void)memcpy(&newshmsegs[i], &shmsegs[i],
748 sizeof(newshmsegs[0]));
749
750 /* Mark as free all new segments, if there is any */
751 for (; i < newshmni; i++) {
752 newshmsegs[i].shm_perm.mode = SHMSEG_FREE;
753 newshmsegs[i].shm_perm._seq = 0;
754 }
755
756 sz = shminfo.shmmni * sizeof(struct shmid_ds);
757 uvm_km_free(kernel_map, (vaddr_t)shmsegs, sz, UVM_KMF_WIRED);
758 shmsegs = newshmsegs;
759
760 return 0;
761 }
762
763 void
764 shminit(void)
765 {
766 int i, sz;
767 vaddr_t v;
768
769 lockinit(&shm_lock, PWAIT, "shmlk", 0, 0);
770
771 /* Allocate pageable memory for our structures */
772 sz = shminfo.shmmni * sizeof(struct shmid_ds);
773 v = uvm_km_alloc(kernel_map, round_page(sz), 0, UVM_KMF_WIRED);
774 if (v == 0)
775 panic("sysv_shm: cannot allocate memory");
776 shmsegs = (void *)v;
777
778 shminfo.shmmax *= PAGE_SIZE;
779
780 for (i = 0; i < shminfo.shmmni; i++) {
781 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
782 shmsegs[i].shm_perm._seq = 0;
783 }
784 shm_last_free = 0;
785 shm_nused = 0;
786 shm_committed = 0;
787 }
788
789 static int
790 sysctl_ipc_shmmni(SYSCTLFN_ARGS)
791 {
792 int newsize, error;
793 struct sysctlnode node;
794 node = *rnode;
795 node.sysctl_data = &newsize;
796
797 newsize = shminfo.shmmni;
798 error = sysctl_lookup(SYSCTLFN_CALL(&node));
799 if (error || newp == NULL)
800 return error;
801
802 lockmgr(&shm_lock, LK_EXCLUSIVE, NULL);
803 error = shmrealloc(newsize);
804 if (error == 0)
805 shminfo.shmmni = newsize;
806 lockmgr(&shm_lock, LK_RELEASE, NULL);
807
808 return error;
809 }
810
811 static int
812 sysctl_ipc_shmmaxpgs(SYSCTLFN_ARGS)
813 {
814 int newsize, error;
815 struct sysctlnode node;
816 node = *rnode;
817 node.sysctl_data = &newsize;
818 newsize = shminfo.shmall;
819 error = sysctl_lookup(SYSCTLFN_CALL(&node));
820 if (error || newp == NULL)
821 return error;
822
823 /* XXX: Would be good to have a upper limit */
824 if (newsize < 1)
825 return EINVAL;
826
827 shminfo.shmall = newsize;
828 shminfo.shmmax = shminfo.shmall * PAGE_SIZE;
829
830 return 0;
831 }
832
833 SYSCTL_SETUP(sysctl_ipc_shm_setup, "sysctl kern.ipc subtree setup")
834 {
835 sysctl_createv(clog, 0, NULL, NULL,
836 CTLFLAG_PERMANENT,
837 CTLTYPE_NODE, "kern", NULL,
838 NULL, 0, NULL, 0,
839 CTL_KERN, CTL_EOL);
840
841 sysctl_createv(clog, 0, NULL, NULL,
842 CTLFLAG_PERMANENT,
843 CTLTYPE_NODE, "ipc",
844 SYSCTL_DESCR("SysV IPC options"),
845 NULL, 0, NULL, 0,
846 CTL_KERN, KERN_SYSVIPC, CTL_EOL);
847
848 sysctl_createv(clog, 0, NULL, NULL,
849 CTLFLAG_PERMANENT | CTLFLAG_READONLY,
850 CTLTYPE_INT, "shmmax",
851 SYSCTL_DESCR("Max shared memory segment size in bytes"),
852 NULL, 0, &shminfo.shmmax, 0,
853 CTL_KERN, KERN_SYSVIPC, KERN_SYSVIPC_SHMMAX, CTL_EOL);
854
855 sysctl_createv(clog, 0, NULL, NULL,
856 CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
857 CTLTYPE_INT, "shmmni",
858 SYSCTL_DESCR("Max number of shared memory identifiers"),
859 sysctl_ipc_shmmni, 0, &shminfo.shmmni, 0,
860 CTL_KERN, KERN_SYSVIPC, KERN_SYSVIPC_SHMMNI, CTL_EOL);
861
862 sysctl_createv(clog, 0, NULL, NULL,
863 CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
864 CTLTYPE_INT, "shmseg",
865 SYSCTL_DESCR("Max shared memory segments per process"),
866 NULL, 0, &shminfo.shmseg, 0,
867 CTL_KERN, KERN_SYSVIPC, KERN_SYSVIPC_SHMSEG, CTL_EOL);
868
869 sysctl_createv(clog, 0, NULL, NULL,
870 CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
871 CTLTYPE_INT, "shmmaxpgs",
872 SYSCTL_DESCR("Max amount of shared memory in pages"),
873 sysctl_ipc_shmmaxpgs, 0, &shminfo.shmall, 0,
874 CTL_KERN, KERN_SYSVIPC, KERN_SYSVIPC_SHMMAXPGS, CTL_EOL);
875
876 sysctl_createv(clog, 0, NULL, NULL,
877 CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
878 CTLTYPE_INT, "shm_use_phys",
879 SYSCTL_DESCR("Enable/disable locking of shared memory in "
880 "physical memory"), NULL, 0, &shm_use_phys, 0,
881 CTL_KERN, KERN_SYSVIPC, KERN_SYSVIPC_SHMUSEPHYS, CTL_EOL);
882 }
Cache object: 8ea7220355b3a3ce0403af5890ccee8c
|