1 /* $FreeBSD: src/sys/kern/sysv_shm.c,v 1.45.2.6 2002/10/22 20:45:03 fjoe Exp $ */
2 /* $DragonFly: src/sys/kern/sysv_shm.c,v 1.19 2006/09/11 20:25:01 dillon Exp $ */
3 /* $NetBSD: sysv_shm.c,v 1.23 1994/07/04 23:25:12 glass Exp $ */
6 * Copyright (c) 1994 Adam Glass and Charles Hannum. All rights reserved.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by Adam Glass and Charles
20 * 4. The names of the authors may not be used to endorse or promote products
21 * derived from this software without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
24 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26 * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
27 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
28 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
32 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 #include "opt_compat.h"
36 #include "opt_sysvipc.h"
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/sysproto.h>
41 #include <sys/kernel.h>
42 #include <sys/sysctl.h>
45 #include <sys/malloc.h>
48 #include <sys/sysent.h>
52 #include <vm/vm_param.h>
55 #include <vm/vm_object.h>
56 #include <vm/vm_map.h>
57 #include <vm/vm_page.h>
58 #include <vm/vm_pager.h>
60 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
63 static int sys_oshmctl (struct proc *p, struct oshmctl_args *uap);
65 static int shmget_allocate_segment (struct proc *p, struct shmget_args *uap, int mode);
66 static int shmget_existing (struct proc *p, struct shmget_args *uap, int mode, int segnum);
68 /* XXX casting to (sy_call_t *) is bogus, as usual. */
69 static sy_call_t *shmcalls[] = {
70 (sy_call_t *)sys_shmat, (sy_call_t *)sys_oshmctl,
71 (sy_call_t *)sys_shmdt, (sy_call_t *)sys_shmget,
72 (sy_call_t *)sys_shmctl
75 #define SHMSEG_FREE 0x0200
76 #define SHMSEG_REMOVED 0x0400
77 #define SHMSEG_ALLOCATED 0x0800
78 #define SHMSEG_WANTED 0x1000
80 static int shm_last_free, shm_nused, shm_committed, shmalloced;
81 static struct shmid_ds *shmsegs;
84 /* vm_offset_t kva; */
85 vm_object_t shm_object;
93 static void shm_deallocate_segment (struct shmid_ds *);
94 static int shm_find_segment_by_key (key_t);
95 static struct shmid_ds *shm_find_segment_by_shmid (int);
96 static int shm_delete_mapping (struct vmspace *vm, struct shmmap_state *);
97 static void shmrealloc (void);
98 static void shminit (void *);
104 #define SHMMAXPGS 8192 /* note: sysv shared memory is swap backed */
107 #define SHMMAX (SHMMAXPGS*PAGE_SIZE)
119 #define SHMALL (SHMMAXPGS)
122 struct shminfo shminfo = {
130 static int shm_use_phys;
132 TUNABLE_INT("kern.ipc.shmmin", &shminfo.shmmin);
133 TUNABLE_INT("kern.ipc.shmmni", &shminfo.shmmni);
134 TUNABLE_INT("kern.ipc.shmseg", &shminfo.shmseg);
135 TUNABLE_INT("kern.ipc.shmmaxpgs", &shminfo.shmall);
136 TUNABLE_INT("kern.ipc.shm_use_phys", &shm_use_phys);
138 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0, "");
139 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0, "");
140 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RD, &shminfo.shmmni, 0, "");
141 SYSCTL_INT(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RW, &shminfo.shmseg, 0, "");
142 SYSCTL_INT(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0, "");
143 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW, &shm_use_phys, 0, "");
146 shm_find_segment_by_key(key)
151 for (i = 0; i < shmalloced; i++)
152 if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
153 shmsegs[i].shm_perm.key == key)
158 static struct shmid_ds *
159 shm_find_segment_by_shmid(shmid)
163 struct shmid_ds *shmseg;
165 segnum = IPCID_TO_IX(shmid);
166 if (segnum < 0 || segnum >= shmalloced)
168 shmseg = &shmsegs[segnum];
169 if ((shmseg->shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
170 != SHMSEG_ALLOCATED ||
171 shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid))
177 shm_deallocate_segment(shmseg)
178 struct shmid_ds *shmseg;
180 struct shm_handle *shm_handle;
183 shm_handle = shmseg->shm_internal;
184 vm_object_deallocate(shm_handle->shm_object);
185 kfree((caddr_t)shm_handle, M_SHM);
186 shmseg->shm_internal = NULL;
187 size = round_page(shmseg->shm_segsz);
188 shm_committed -= btoc(size);
190 shmseg->shm_perm.mode = SHMSEG_FREE;
194 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
196 struct shmid_ds *shmseg;
200 segnum = IPCID_TO_IX(shmmap_s->shmid);
201 shmseg = &shmsegs[segnum];
202 size = round_page(shmseg->shm_segsz);
203 result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
204 if (result != KERN_SUCCESS)
206 shmmap_s->shmid = -1;
207 shmseg->shm_dtime = time_second;
208 if ((--shmseg->shm_nattch <= 0) &&
209 (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
210 shm_deallocate_segment(shmseg);
211 shm_last_free = segnum;
217 sys_shmdt(struct shmdt_args *uap)
219 struct proc *p = curproc;
220 struct shmmap_state *shmmap_s;
223 if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
226 shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
227 if (shmmap_s == NULL)
229 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
230 if (shmmap_s->shmid != -1 &&
231 shmmap_s->va == (vm_offset_t)uap->shmaddr)
233 if (i == shminfo.shmseg)
235 return shm_delete_mapping(p->p_vmspace, shmmap_s);
239 sys_shmat(struct shmat_args *uap)
241 struct proc *p = curproc;
243 struct shmid_ds *shmseg;
244 struct shmmap_state *shmmap_s = NULL;
245 struct shm_handle *shm_handle;
246 vm_offset_t attach_va;
251 if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
254 shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
255 if (shmmap_s == NULL) {
256 size = shminfo.shmseg * sizeof(struct shmmap_state);
257 shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
258 for (i = 0; i < shminfo.shmseg; i++)
259 shmmap_s[i].shmid = -1;
260 p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
262 shmseg = shm_find_segment_by_shmid(uap->shmid);
265 error = ipcperm(p, &shmseg->shm_perm,
266 (uap->shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
269 for (i = 0; i < shminfo.shmseg; i++) {
270 if (shmmap_s->shmid == -1)
274 if (i >= shminfo.shmseg)
276 size = round_page(shmseg->shm_segsz);
277 #ifdef VM_PROT_READ_IS_EXEC
278 prot = VM_PROT_READ | VM_PROT_EXECUTE;
282 if ((uap->shmflg & SHM_RDONLY) == 0)
283 prot |= VM_PROT_WRITE;
284 flags = MAP_ANON | MAP_SHARED;
287 if (uap->shmflg & SHM_RND)
288 attach_va = (vm_offset_t)uap->shmaddr & ~(SHMLBA-1);
289 else if (((vm_offset_t)uap->shmaddr & (SHMLBA-1)) == 0)
290 attach_va = (vm_offset_t)uap->shmaddr;
294 /* This is just a hint to vm_map_find() about where to put it. */
295 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_taddr + maxtsiz + maxdsiz);
298 shm_handle = shmseg->shm_internal;
299 vm_object_reference(shm_handle->shm_object);
300 rv = vm_map_find(&p->p_vmspace->vm_map,
301 shm_handle->shm_object, 0,
303 ((flags & MAP_FIXED) ? 0 : 1),
307 if (rv != KERN_SUCCESS) {
308 vm_object_deallocate(shm_handle->shm_object);
311 vm_map_inherit(&p->p_vmspace->vm_map,
312 attach_va, attach_va + size, VM_INHERIT_SHARE);
314 shmmap_s->va = attach_va;
315 shmmap_s->shmid = uap->shmid;
316 shmseg->shm_lpid = p->p_pid;
317 shmseg->shm_atime = time_second;
318 shmseg->shm_nattch++;
319 uap->sysmsg_result = attach_va;
324 struct ipc_perm shm_perm; /* operation perms */
325 int shm_segsz; /* size of segment (bytes) */
326 ushort shm_cpid; /* pid, creator */
327 ushort shm_lpid; /* pid, last operation */
328 short shm_nattch; /* no. of current attaches */
329 time_t shm_atime; /* last attach time */
330 time_t shm_dtime; /* last detach time */
331 time_t shm_ctime; /* last change time */
332 void *shm_handle; /* internal handle for shm segment */
335 struct oshmctl_args {
336 struct sysmsg sysmsg;
339 struct oshmid_ds *ubuf;
345 struct oshmctl_args *uap;
349 struct shmid_ds *shmseg;
350 struct oshmid_ds outbuf;
352 if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
355 shmseg = shm_find_segment_by_shmid(uap->shmid);
360 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
363 outbuf.shm_perm = shmseg->shm_perm;
364 outbuf.shm_segsz = shmseg->shm_segsz;
365 outbuf.shm_cpid = shmseg->shm_cpid;
366 outbuf.shm_lpid = shmseg->shm_lpid;
367 outbuf.shm_nattch = shmseg->shm_nattch;
368 outbuf.shm_atime = shmseg->shm_atime;
369 outbuf.shm_dtime = shmseg->shm_dtime;
370 outbuf.shm_ctime = shmseg->shm_ctime;
371 outbuf.shm_handle = shmseg->shm_internal;
372 error = copyout((caddr_t)&outbuf, uap->ubuf, sizeof(outbuf));
377 /* XXX casting to (sy_call_t *) is bogus, as usual. */
378 return (sys_shmctl((struct shmctl_args *)uap));
387 sys_shmctl(struct shmctl_args *uap)
389 struct proc *p = curproc;
391 struct shmid_ds inbuf;
392 struct shmid_ds *shmseg;
394 if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
397 shmseg = shm_find_segment_by_shmid(uap->shmid);
402 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
405 error = copyout((caddr_t)shmseg, uap->buf, sizeof(inbuf));
410 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
413 error = copyin(uap->buf, (caddr_t)&inbuf, sizeof(inbuf));
416 shmseg->shm_perm.uid = inbuf.shm_perm.uid;
417 shmseg->shm_perm.gid = inbuf.shm_perm.gid;
418 shmseg->shm_perm.mode =
419 (shmseg->shm_perm.mode & ~ACCESSPERMS) |
420 (inbuf.shm_perm.mode & ACCESSPERMS);
421 shmseg->shm_ctime = time_second;
424 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
427 shmseg->shm_perm.key = IPC_PRIVATE;
428 shmseg->shm_perm.mode |= SHMSEG_REMOVED;
429 if (shmseg->shm_nattch <= 0) {
430 shm_deallocate_segment(shmseg);
431 shm_last_free = IPCID_TO_IX(uap->shmid);
445 shmget_existing(p, uap, mode, segnum)
447 struct shmget_args *uap;
451 struct shmid_ds *shmseg;
454 shmseg = &shmsegs[segnum];
455 if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
457 * This segment is in the process of being allocated. Wait
458 * until it's done, and look the key up again (in case the
459 * allocation failed or it was freed).
461 shmseg->shm_perm.mode |= SHMSEG_WANTED;
462 error = tsleep((caddr_t)shmseg, PCATCH, "shmget", 0);
467 if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
469 error = ipcperm(p, &shmseg->shm_perm, mode);
472 if (uap->size && uap->size > shmseg->shm_segsz)
474 uap->sysmsg_result = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
479 shmget_allocate_segment(p, uap, mode)
481 struct shmget_args *uap;
484 int i, segnum, shmid, size;
485 struct ucred *cred = p->p_ucred;
486 struct shmid_ds *shmseg;
487 struct shm_handle *shm_handle;
489 if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
491 if (shm_nused >= shminfo.shmmni) /* any shmids left? */
493 size = round_page(uap->size);
494 if (shm_committed + btoc(size) > shminfo.shmall)
496 if (shm_last_free < 0) {
497 shmrealloc(); /* maybe expand the shmsegs[] array */
498 for (i = 0; i < shmalloced; i++)
499 if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
505 segnum = shm_last_free;
508 shmseg = &shmsegs[segnum];
510 * In case we sleep in malloc(), mark the segment present but deleted
511 * so that noone else tries to create the same key.
513 shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
514 shmseg->shm_perm.key = uap->key;
515 shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
516 shm_handle = kmalloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
517 shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
520 * We make sure that we have allocated a pager before we need
524 shm_handle->shm_object =
525 vm_pager_allocate(OBJT_PHYS, 0, size, VM_PROT_DEFAULT, 0);
527 shm_handle->shm_object =
528 vm_pager_allocate(OBJT_SWAP, 0, size, VM_PROT_DEFAULT, 0);
530 vm_object_clear_flag(shm_handle->shm_object, OBJ_ONEMAPPING);
531 vm_object_set_flag(shm_handle->shm_object, OBJ_NOSPLIT);
533 shmseg->shm_internal = shm_handle;
534 shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
535 shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
536 shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
537 (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
538 shmseg->shm_segsz = uap->size;
539 shmseg->shm_cpid = p->p_pid;
540 shmseg->shm_lpid = shmseg->shm_nattch = 0;
541 shmseg->shm_atime = shmseg->shm_dtime = 0;
542 shmseg->shm_ctime = time_second;
543 shm_committed += btoc(size);
545 if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
547 * Somebody else wanted this key while we were asleep. Wake
550 shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
551 wakeup((caddr_t)shmseg);
553 uap->sysmsg_result = shmid;
558 sys_shmget(struct shmget_args *uap)
560 struct proc *p = curproc;
561 int segnum, mode, error;
563 if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
566 mode = uap->shmflg & ACCESSPERMS;
567 if (uap->key != IPC_PRIVATE) {
569 segnum = shm_find_segment_by_key(uap->key);
571 error = shmget_existing(p, uap, mode, segnum);
576 if ((uap->shmflg & IPC_CREAT) == 0)
579 return shmget_allocate_segment(p, uap, mode);
583 * shmsys_args(int which, int a2, ...) (VARARGS)
586 sys_shmsys(struct shmsys_args *uap)
588 struct proc *p = curproc;
589 unsigned int which = (unsigned int)uap->which;
592 if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
595 if (which >= sizeof(shmcalls)/sizeof(shmcalls[0]))
597 bcopy(&uap->a2, &uap->which,
598 sizeof(struct shmsys_args) - offsetof(struct shmsys_args, a2));
599 error = ((*shmcalls[which])(uap));
605 struct proc *p1, *p2;
607 struct shmmap_state *shmmap_s;
611 size = shminfo.shmseg * sizeof(struct shmmap_state);
612 shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
613 bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmmap_s, size);
614 p2->p_vmspace->vm_shm = (caddr_t)shmmap_s;
615 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
616 if (shmmap_s->shmid != -1)
617 shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
621 shmexit(struct vmspace *vm)
623 struct shmmap_state *base, *shm;
626 if ((base = (struct shmmap_state *)vm->vm_shm) != NULL) {
628 for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
629 if (shm->shmid != -1)
630 shm_delete_mapping(vm, shm);
640 struct shmid_ds *newsegs;
642 if (shmalloced >= shminfo.shmmni)
645 newsegs = kmalloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
648 for (i = 0; i < shmalloced; i++)
649 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
650 for (; i < shminfo.shmmni; i++) {
651 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
652 shmsegs[i].shm_perm.seq = 0;
654 kfree(shmsegs, M_SHM);
656 shmalloced = shminfo.shmmni;
665 shminfo.shmmax = shminfo.shmall * PAGE_SIZE;
666 shmalloced = shminfo.shmmni;
667 shmsegs = kmalloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
669 panic("cannot allocate initial memory for sysvshm");
670 for (i = 0; i < shmalloced; i++) {
671 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
672 shmsegs[i].shm_perm.seq = 0;
678 SYSINIT(sysv_shm, SI_SUB_SYSV_SHM, SI_ORDER_FIRST, shminit, NULL);