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 1.15 2004/07/23 14:07:46 joerg 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_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 free((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 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 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
= malloc(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
, shm_handle
->shm_object
,
301 0, &attach_va
, size
, (flags
& MAP_FIXED
)?0:1, prot
, prot
, 0);
302 if (rv
!= KERN_SUCCESS
) {
303 vm_object_deallocate(shm_handle
->shm_object
);
306 vm_map_inherit(&p
->p_vmspace
->vm_map
,
307 attach_va
, attach_va
+ size
, VM_INHERIT_SHARE
);
309 shmmap_s
->va
= attach_va
;
310 shmmap_s
->shmid
= uap
->shmid
;
311 shmseg
->shm_lpid
= p
->p_pid
;
312 shmseg
->shm_atime
= time_second
;
313 shmseg
->shm_nattch
++;
314 uap
->sysmsg_result
= attach_va
;
319 struct ipc_perm shm_perm
; /* operation perms */
320 int shm_segsz
; /* size of segment (bytes) */
321 ushort shm_cpid
; /* pid, creator */
322 ushort shm_lpid
; /* pid, last operation */
323 short shm_nattch
; /* no. of current attaches */
324 time_t shm_atime
; /* last attach time */
325 time_t shm_dtime
; /* last detach time */
326 time_t shm_ctime
; /* last change time */
327 void *shm_handle
; /* internal handle for shm segment */
330 struct oshmctl_args
{
331 struct sysmsg sysmsg
;
334 struct oshmid_ds
*ubuf
;
340 struct oshmctl_args
*uap
;
344 struct shmid_ds
*shmseg
;
345 struct oshmid_ds outbuf
;
347 if (!jail_sysvipc_allowed
&& p
->p_ucred
->cr_prison
!= NULL
)
350 shmseg
= shm_find_segment_by_shmid(uap
->shmid
);
355 error
= ipcperm(p
, &shmseg
->shm_perm
, IPC_R
);
358 outbuf
.shm_perm
= shmseg
->shm_perm
;
359 outbuf
.shm_segsz
= shmseg
->shm_segsz
;
360 outbuf
.shm_cpid
= shmseg
->shm_cpid
;
361 outbuf
.shm_lpid
= shmseg
->shm_lpid
;
362 outbuf
.shm_nattch
= shmseg
->shm_nattch
;
363 outbuf
.shm_atime
= shmseg
->shm_atime
;
364 outbuf
.shm_dtime
= shmseg
->shm_dtime
;
365 outbuf
.shm_ctime
= shmseg
->shm_ctime
;
366 outbuf
.shm_handle
= shmseg
->shm_internal
;
367 error
= copyout((caddr_t
)&outbuf
, uap
->ubuf
, sizeof(outbuf
));
372 /* XXX casting to (sy_call_t *) is bogus, as usual. */
373 return (shmctl((struct shmctl_args
*)uap
));
382 shmctl(struct shmctl_args
*uap
)
384 struct proc
*p
= curproc
;
386 struct shmid_ds inbuf
;
387 struct shmid_ds
*shmseg
;
389 if (!jail_sysvipc_allowed
&& p
->p_ucred
->cr_prison
!= NULL
)
392 shmseg
= shm_find_segment_by_shmid(uap
->shmid
);
397 error
= ipcperm(p
, &shmseg
->shm_perm
, IPC_R
);
400 error
= copyout((caddr_t
)shmseg
, uap
->buf
, sizeof(inbuf
));
405 error
= ipcperm(p
, &shmseg
->shm_perm
, IPC_M
);
408 error
= copyin(uap
->buf
, (caddr_t
)&inbuf
, sizeof(inbuf
));
411 shmseg
->shm_perm
.uid
= inbuf
.shm_perm
.uid
;
412 shmseg
->shm_perm
.gid
= inbuf
.shm_perm
.gid
;
413 shmseg
->shm_perm
.mode
=
414 (shmseg
->shm_perm
.mode
& ~ACCESSPERMS
) |
415 (inbuf
.shm_perm
.mode
& ACCESSPERMS
);
416 shmseg
->shm_ctime
= time_second
;
419 error
= ipcperm(p
, &shmseg
->shm_perm
, IPC_M
);
422 shmseg
->shm_perm
.key
= IPC_PRIVATE
;
423 shmseg
->shm_perm
.mode
|= SHMSEG_REMOVED
;
424 if (shmseg
->shm_nattch
<= 0) {
425 shm_deallocate_segment(shmseg
);
426 shm_last_free
= IPCID_TO_IX(uap
->shmid
);
440 shmget_existing(p
, uap
, mode
, segnum
)
442 struct shmget_args
*uap
;
446 struct shmid_ds
*shmseg
;
449 shmseg
= &shmsegs
[segnum
];
450 if (shmseg
->shm_perm
.mode
& SHMSEG_REMOVED
) {
452 * This segment is in the process of being allocated. Wait
453 * until it's done, and look the key up again (in case the
454 * allocation failed or it was freed).
456 shmseg
->shm_perm
.mode
|= SHMSEG_WANTED
;
457 error
= tsleep((caddr_t
)shmseg
, PCATCH
, "shmget", 0);
462 if ((uap
->shmflg
& (IPC_CREAT
| IPC_EXCL
)) == (IPC_CREAT
| IPC_EXCL
))
464 error
= ipcperm(p
, &shmseg
->shm_perm
, mode
);
467 if (uap
->size
&& uap
->size
> shmseg
->shm_segsz
)
469 uap
->sysmsg_result
= IXSEQ_TO_IPCID(segnum
, shmseg
->shm_perm
);
474 shmget_allocate_segment(p
, uap
, mode
)
476 struct shmget_args
*uap
;
479 int i
, segnum
, shmid
, size
;
480 struct ucred
*cred
= p
->p_ucred
;
481 struct shmid_ds
*shmseg
;
482 struct shm_handle
*shm_handle
;
484 if (uap
->size
< shminfo
.shmmin
|| uap
->size
> shminfo
.shmmax
)
486 if (shm_nused
>= shminfo
.shmmni
) /* any shmids left? */
488 size
= round_page(uap
->size
);
489 if (shm_committed
+ btoc(size
) > shminfo
.shmall
)
491 if (shm_last_free
< 0) {
492 shmrealloc(); /* maybe expand the shmsegs[] array */
493 for (i
= 0; i
< shmalloced
; i
++)
494 if (shmsegs
[i
].shm_perm
.mode
& SHMSEG_FREE
)
500 segnum
= shm_last_free
;
503 shmseg
= &shmsegs
[segnum
];
505 * In case we sleep in malloc(), mark the segment present but deleted
506 * so that noone else tries to create the same key.
508 shmseg
->shm_perm
.mode
= SHMSEG_ALLOCATED
| SHMSEG_REMOVED
;
509 shmseg
->shm_perm
.key
= uap
->key
;
510 shmseg
->shm_perm
.seq
= (shmseg
->shm_perm
.seq
+ 1) & 0x7fff;
511 shm_handle
= (struct shm_handle
*)
512 malloc(sizeof(struct shm_handle
), M_SHM
, M_WAITOK
);
513 shmid
= IXSEQ_TO_IPCID(segnum
, shmseg
->shm_perm
);
516 * We make sure that we have allocated a pager before we need
520 shm_handle
->shm_object
=
521 vm_pager_allocate(OBJT_PHYS
, 0, size
, VM_PROT_DEFAULT
, 0);
523 shm_handle
->shm_object
=
524 vm_pager_allocate(OBJT_SWAP
, 0, size
, VM_PROT_DEFAULT
, 0);
526 vm_object_clear_flag(shm_handle
->shm_object
, OBJ_ONEMAPPING
);
527 vm_object_set_flag(shm_handle
->shm_object
, OBJ_NOSPLIT
);
529 shmseg
->shm_internal
= shm_handle
;
530 shmseg
->shm_perm
.cuid
= shmseg
->shm_perm
.uid
= cred
->cr_uid
;
531 shmseg
->shm_perm
.cgid
= shmseg
->shm_perm
.gid
= cred
->cr_gid
;
532 shmseg
->shm_perm
.mode
= (shmseg
->shm_perm
.mode
& SHMSEG_WANTED
) |
533 (mode
& ACCESSPERMS
) | SHMSEG_ALLOCATED
;
534 shmseg
->shm_segsz
= uap
->size
;
535 shmseg
->shm_cpid
= p
->p_pid
;
536 shmseg
->shm_lpid
= shmseg
->shm_nattch
= 0;
537 shmseg
->shm_atime
= shmseg
->shm_dtime
= 0;
538 shmseg
->shm_ctime
= time_second
;
539 shm_committed
+= btoc(size
);
541 if (shmseg
->shm_perm
.mode
& SHMSEG_WANTED
) {
543 * Somebody else wanted this key while we were asleep. Wake
546 shmseg
->shm_perm
.mode
&= ~SHMSEG_WANTED
;
547 wakeup((caddr_t
)shmseg
);
549 uap
->sysmsg_result
= shmid
;
554 shmget(struct shmget_args
*uap
)
556 struct proc
*p
= curproc
;
557 int segnum
, mode
, error
;
559 if (!jail_sysvipc_allowed
&& p
->p_ucred
->cr_prison
!= NULL
)
562 mode
= uap
->shmflg
& ACCESSPERMS
;
563 if (uap
->key
!= IPC_PRIVATE
) {
565 segnum
= shm_find_segment_by_key(uap
->key
);
567 error
= shmget_existing(p
, uap
, mode
, segnum
);
572 if ((uap
->shmflg
& IPC_CREAT
) == 0)
575 return shmget_allocate_segment(p
, uap
, mode
);
579 * shmsys_args(int which, int a2, ...) (VARARGS)
582 shmsys(struct shmsys_args
*uap
)
584 struct proc
*p
= curproc
;
585 unsigned int which
= (unsigned int)uap
->which
;
588 if (!jail_sysvipc_allowed
&& p
->p_ucred
->cr_prison
!= NULL
)
591 if (which
>= sizeof(shmcalls
)/sizeof(shmcalls
[0]))
593 bcopy(&uap
->a2
, &uap
->which
,
594 sizeof(struct shmsys_args
) - offsetof(struct shmsys_args
, a2
));
595 error
= ((*shmcalls
[which
])(uap
));
601 struct proc
*p1
, *p2
;
603 struct shmmap_state
*shmmap_s
;
607 size
= shminfo
.shmseg
* sizeof(struct shmmap_state
);
608 shmmap_s
= malloc(size
, M_SHM
, M_WAITOK
);
609 bcopy((caddr_t
)p1
->p_vmspace
->vm_shm
, (caddr_t
)shmmap_s
, size
);
610 p2
->p_vmspace
->vm_shm
= (caddr_t
)shmmap_s
;
611 for (i
= 0; i
< shminfo
.shmseg
; i
++, shmmap_s
++)
612 if (shmmap_s
->shmid
!= -1)
613 shmsegs
[IPCID_TO_IX(shmmap_s
->shmid
)].shm_nattch
++;
617 shmexit(struct vmspace
*vm
)
619 struct shmmap_state
*base
, *shm
;
622 if ((base
= (struct shmmap_state
*)vm
->vm_shm
) != NULL
) {
624 for (i
= 0, shm
= base
; i
< shminfo
.shmseg
; i
++, shm
++) {
625 if (shm
->shmid
!= -1)
626 shm_delete_mapping(vm
, shm
);
636 struct shmid_ds
*newsegs
;
638 if (shmalloced
>= shminfo
.shmmni
)
641 newsegs
= malloc(shminfo
.shmmni
* sizeof(*newsegs
), M_SHM
, M_WAITOK
);
644 for (i
= 0; i
< shmalloced
; i
++)
645 bcopy(&shmsegs
[i
], &newsegs
[i
], sizeof(newsegs
[0]));
646 for (; i
< shminfo
.shmmni
; i
++) {
647 shmsegs
[i
].shm_perm
.mode
= SHMSEG_FREE
;
648 shmsegs
[i
].shm_perm
.seq
= 0;
650 free(shmsegs
, M_SHM
);
652 shmalloced
= shminfo
.shmmni
;
661 shminfo
.shmmax
= shminfo
.shmall
* PAGE_SIZE
;
662 shmalloced
= shminfo
.shmmni
;
663 shmsegs
= malloc(shmalloced
* sizeof(shmsegs
[0]), M_SHM
, M_WAITOK
);
665 panic("cannot allocate initial memory for sysvshm");
666 for (i
= 0; i
< shmalloced
; i
++) {
667 shmsegs
[i
].shm_perm
.mode
= SHMSEG_FREE
;
668 shmsegs
[i
].shm_perm
.seq
= 0;
674 SYSINIT(sysv_shm
, SI_SUB_SYSV_SHM
, SI_ORDER_FIRST
, shminit
, NULL
);