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 1.13 2004/02/05 18:43:22 drhodus 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 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 *)shmat, (sy_call_t *)oshmctl,
71 (sy_call_t *)shmdt, (sy_call_t *)shmget,
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_DECL(_kern_ipc);
139 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0, "");
140 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0, "");
141 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RD, &shminfo.shmmni, 0, "");
142 SYSCTL_INT(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RW, &shminfo.shmseg, 0, "");
143 SYSCTL_INT(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0, "");
144 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW, &shm_use_phys, 0, "");
147 shm_find_segment_by_key(key)
152 for (i = 0; i < shmalloced; i++)
153 if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
154 shmsegs[i].shm_perm.key == key)
159 static struct shmid_ds *
160 shm_find_segment_by_shmid(shmid)
164 struct shmid_ds *shmseg;
166 segnum = IPCID_TO_IX(shmid);
167 if (segnum < 0 || segnum >= shmalloced)
169 shmseg = &shmsegs[segnum];
170 if ((shmseg->shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
171 != SHMSEG_ALLOCATED ||
172 shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid))
178 shm_deallocate_segment(shmseg)
179 struct shmid_ds *shmseg;
181 struct shm_handle *shm_handle;
184 shm_handle = shmseg->shm_internal;
185 vm_object_deallocate(shm_handle->shm_object);
186 free((caddr_t)shm_handle, M_SHM);
187 shmseg->shm_internal = NULL;
188 size = round_page(shmseg->shm_segsz);
189 shm_committed -= btoc(size);
191 shmseg->shm_perm.mode = SHMSEG_FREE;
195 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
197 struct shmid_ds *shmseg;
201 segnum = IPCID_TO_IX(shmmap_s->shmid);
202 shmseg = &shmsegs[segnum];
203 size = round_page(shmseg->shm_segsz);
204 result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
205 if (result != KERN_SUCCESS)
207 shmmap_s->shmid = -1;
208 shmseg->shm_dtime = time_second;
209 if ((--shmseg->shm_nattch <= 0) &&
210 (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
211 shm_deallocate_segment(shmseg);
212 shm_last_free = segnum;
218 shmdt(struct shmdt_args *uap)
220 struct proc *p = curproc;
221 struct shmmap_state *shmmap_s;
224 if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
227 shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
228 if (shmmap_s == NULL)
230 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
231 if (shmmap_s->shmid != -1 &&
232 shmmap_s->va == (vm_offset_t)uap->shmaddr)
234 if (i == shminfo.shmseg)
236 return shm_delete_mapping(p->p_vmspace, shmmap_s);
240 shmat(struct shmat_args *uap)
242 struct proc *p = curproc;
244 struct shmid_ds *shmseg;
245 struct shmmap_state *shmmap_s = NULL;
246 struct shm_handle *shm_handle;
247 vm_offset_t attach_va;
252 if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
255 shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
256 if (shmmap_s == NULL) {
257 size = shminfo.shmseg * sizeof(struct shmmap_state);
258 shmmap_s = malloc(size, M_SHM, M_WAITOK);
259 for (i = 0; i < shminfo.shmseg; i++)
260 shmmap_s[i].shmid = -1;
261 p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
263 shmseg = shm_find_segment_by_shmid(uap->shmid);
266 error = ipcperm(p, &shmseg->shm_perm,
267 (uap->shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
270 for (i = 0; i < shminfo.shmseg; i++) {
271 if (shmmap_s->shmid == -1)
275 if (i >= shminfo.shmseg)
277 size = round_page(shmseg->shm_segsz);
278 #ifdef VM_PROT_READ_IS_EXEC
279 prot = VM_PROT_READ | VM_PROT_EXECUTE;
283 if ((uap->shmflg & SHM_RDONLY) == 0)
284 prot |= VM_PROT_WRITE;
285 flags = MAP_ANON | MAP_SHARED;
288 if (uap->shmflg & SHM_RND)
289 attach_va = (vm_offset_t)uap->shmaddr & ~(SHMLBA-1);
290 else if (((vm_offset_t)uap->shmaddr & (SHMLBA-1)) == 0)
291 attach_va = (vm_offset_t)uap->shmaddr;
295 /* This is just a hint to vm_map_find() about where to put it. */
296 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_taddr + maxtsiz + maxdsiz);
299 shm_handle = shmseg->shm_internal;
300 vm_object_reference(shm_handle->shm_object);
301 rv = vm_map_find(&p->p_vmspace->vm_map, shm_handle->shm_object,
302 0, &attach_va, size, (flags & MAP_FIXED)?0:1, prot, prot, 0);
303 if (rv != KERN_SUCCESS) {
304 vm_object_deallocate(shm_handle->shm_object);
307 vm_map_inherit(&p->p_vmspace->vm_map,
308 attach_va, attach_va + size, VM_INHERIT_SHARE);
310 shmmap_s->va = attach_va;
311 shmmap_s->shmid = uap->shmid;
312 shmseg->shm_lpid = p->p_pid;
313 shmseg->shm_atime = time_second;
314 shmseg->shm_nattch++;
315 uap->sysmsg_result = attach_va;
320 struct ipc_perm shm_perm; /* operation perms */
321 int shm_segsz; /* size of segment (bytes) */
322 ushort shm_cpid; /* pid, creator */
323 ushort shm_lpid; /* pid, last operation */
324 short shm_nattch; /* no. of current attaches */
325 time_t shm_atime; /* last attach time */
326 time_t shm_dtime; /* last detach time */
327 time_t shm_ctime; /* last change time */
328 void *shm_handle; /* internal handle for shm segment */
331 struct oshmctl_args {
332 struct sysmsg sysmsg;
335 struct oshmid_ds *ubuf;
341 struct oshmctl_args *uap;
345 struct shmid_ds *shmseg;
346 struct oshmid_ds outbuf;
348 if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
351 shmseg = shm_find_segment_by_shmid(uap->shmid);
356 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
359 outbuf.shm_perm = shmseg->shm_perm;
360 outbuf.shm_segsz = shmseg->shm_segsz;
361 outbuf.shm_cpid = shmseg->shm_cpid;
362 outbuf.shm_lpid = shmseg->shm_lpid;
363 outbuf.shm_nattch = shmseg->shm_nattch;
364 outbuf.shm_atime = shmseg->shm_atime;
365 outbuf.shm_dtime = shmseg->shm_dtime;
366 outbuf.shm_ctime = shmseg->shm_ctime;
367 outbuf.shm_handle = shmseg->shm_internal;
368 error = copyout((caddr_t)&outbuf, uap->ubuf, sizeof(outbuf));
373 /* XXX casting to (sy_call_t *) is bogus, as usual. */
374 return ((sy_call_t *)shmctl)(uap);
383 shmctl(struct shmctl_args *uap)
385 struct proc *p = curproc;
387 struct shmid_ds inbuf;
388 struct shmid_ds *shmseg;
390 if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
393 shmseg = shm_find_segment_by_shmid(uap->shmid);
398 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
401 error = copyout((caddr_t)shmseg, uap->buf, sizeof(inbuf));
406 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
409 error = copyin(uap->buf, (caddr_t)&inbuf, sizeof(inbuf));
412 shmseg->shm_perm.uid = inbuf.shm_perm.uid;
413 shmseg->shm_perm.gid = inbuf.shm_perm.gid;
414 shmseg->shm_perm.mode =
415 (shmseg->shm_perm.mode & ~ACCESSPERMS) |
416 (inbuf.shm_perm.mode & ACCESSPERMS);
417 shmseg->shm_ctime = time_second;
420 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
423 shmseg->shm_perm.key = IPC_PRIVATE;
424 shmseg->shm_perm.mode |= SHMSEG_REMOVED;
425 if (shmseg->shm_nattch <= 0) {
426 shm_deallocate_segment(shmseg);
427 shm_last_free = IPCID_TO_IX(uap->shmid);
441 shmget_existing(p, uap, mode, segnum)
443 struct shmget_args *uap;
447 struct shmid_ds *shmseg;
450 shmseg = &shmsegs[segnum];
451 if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
453 * This segment is in the process of being allocated. Wait
454 * until it's done, and look the key up again (in case the
455 * allocation failed or it was freed).
457 shmseg->shm_perm.mode |= SHMSEG_WANTED;
458 error = tsleep((caddr_t)shmseg, PCATCH, "shmget", 0);
463 if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
465 error = ipcperm(p, &shmseg->shm_perm, mode);
468 if (uap->size && uap->size > shmseg->shm_segsz)
470 uap->sysmsg_result = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
475 shmget_allocate_segment(p, uap, mode)
477 struct shmget_args *uap;
480 int i, segnum, shmid, size;
481 struct ucred *cred = p->p_ucred;
482 struct shmid_ds *shmseg;
483 struct shm_handle *shm_handle;
485 if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
487 if (shm_nused >= shminfo.shmmni) /* any shmids left? */
489 size = round_page(uap->size);
490 if (shm_committed + btoc(size) > shminfo.shmall)
492 if (shm_last_free < 0) {
493 shmrealloc(); /* maybe expand the shmsegs[] array */
494 for (i = 0; i < shmalloced; i++)
495 if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
501 segnum = shm_last_free;
504 shmseg = &shmsegs[segnum];
506 * In case we sleep in malloc(), mark the segment present but deleted
507 * so that noone else tries to create the same key.
509 shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
510 shmseg->shm_perm.key = uap->key;
511 shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
512 shm_handle = (struct shm_handle *)
513 malloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
514 shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
517 * We make sure that we have allocated a pager before we need
521 shm_handle->shm_object =
522 vm_pager_allocate(OBJT_PHYS, 0, size, VM_PROT_DEFAULT, 0);
524 shm_handle->shm_object =
525 vm_pager_allocate(OBJT_SWAP, 0, size, VM_PROT_DEFAULT, 0);
527 vm_object_clear_flag(shm_handle->shm_object, OBJ_ONEMAPPING);
528 vm_object_set_flag(shm_handle->shm_object, OBJ_NOSPLIT);
530 shmseg->shm_internal = shm_handle;
531 shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
532 shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
533 shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
534 (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
535 shmseg->shm_segsz = uap->size;
536 shmseg->shm_cpid = p->p_pid;
537 shmseg->shm_lpid = shmseg->shm_nattch = 0;
538 shmseg->shm_atime = shmseg->shm_dtime = 0;
539 shmseg->shm_ctime = time_second;
540 shm_committed += btoc(size);
542 if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
544 * Somebody else wanted this key while we were asleep. Wake
547 shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
548 wakeup((caddr_t)shmseg);
550 uap->sysmsg_result = shmid;
555 shmget(struct shmget_args *uap)
557 struct proc *p = curproc;
558 int segnum, mode, error;
560 if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
563 mode = uap->shmflg & ACCESSPERMS;
564 if (uap->key != IPC_PRIVATE) {
566 segnum = shm_find_segment_by_key(uap->key);
568 error = shmget_existing(p, uap, mode, segnum);
573 if ((uap->shmflg & IPC_CREAT) == 0)
576 return shmget_allocate_segment(p, uap, mode);
580 * shmsys_args(int which, int a2, ...) (VARARGS)
583 shmsys(struct shmsys_args *uap)
585 struct proc *p = curproc;
586 unsigned int which = (unsigned int)uap->which;
589 if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
592 if (which >= sizeof(shmcalls)/sizeof(shmcalls[0]))
594 bcopy(&uap->a2, &uap->which,
595 sizeof(struct shmsys_args) - offsetof(struct shmsys_args, a2));
596 error = ((*shmcalls[which])(uap));
602 struct proc *p1, *p2;
604 struct shmmap_state *shmmap_s;
608 size = shminfo.shmseg * sizeof(struct shmmap_state);
609 shmmap_s = malloc(size, M_SHM, M_WAITOK);
610 bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmmap_s, size);
611 p2->p_vmspace->vm_shm = (caddr_t)shmmap_s;
612 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
613 if (shmmap_s->shmid != -1)
614 shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
618 shmexit(struct vmspace *vm)
620 struct shmmap_state *base, *shm;
623 if ((base = (struct shmmap_state *)vm->vm_shm) != NULL) {
625 for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
626 if (shm->shmid != -1)
627 shm_delete_mapping(vm, shm);
637 struct shmid_ds *newsegs;
639 if (shmalloced >= shminfo.shmmni)
642 newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
645 for (i = 0; i < shmalloced; i++)
646 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
647 for (; i < shminfo.shmmni; i++) {
648 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
649 shmsegs[i].shm_perm.seq = 0;
651 free(shmsegs, M_SHM);
653 shmalloced = shminfo.shmmni;
662 shminfo.shmmax = shminfo.shmall * PAGE_SIZE;
663 shmalloced = shminfo.shmmni;
664 shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
666 panic("cannot allocate initial memory for sysvshm");
667 for (i = 0; i < shmalloced; i++) {
668 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
669 shmsegs[i].shm_perm.seq = 0;
675 SYSINIT(sysv_shm, SI_SUB_SYSV_SHM, SI_ORDER_FIRST, shminit, NULL);