2 * Copyright (c) 1994 Adam Glass and Charles Hannum. All rights reserved.
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
7 * 1. Redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution.
12 * 3. All advertising materials mentioning features or use of this software
13 * must display the following acknowledgement:
14 * This product includes software developed by Adam Glass and Charles
16 * 4. The names of the authors may not be used to endorse or promote products
17 * derived from this software without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
20 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22 * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31 #include "opt_compat.h"
32 #include "opt_sysvipc.h"
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/sysproto.h>
37 #include <sys/kernel.h>
38 #include <sys/sysctl.h>
41 #include <sys/malloc.h>
44 #include <sys/sysent.h>
47 #include <sys/mplock2.h>
50 #include <vm/vm_param.h>
53 #include <vm/vm_object.h>
54 #include <vm/vm_map.h>
55 #include <vm/vm_page.h>
56 #include <vm/vm_pager.h>
58 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
61 static int sys_oshmctl (struct proc *p, struct oshmctl_args *uap);
63 static int shmget_allocate_segment (struct proc *p, struct shmget_args *uap, int mode);
64 static int shmget_existing (struct proc *p, struct shmget_args *uap, int mode, int segnum);
66 /* XXX casting to (sy_call_t *) is bogus, as usual. */
67 static sy_call_t *shmcalls[] = {
68 (sy_call_t *)sys_shmat, (sy_call_t *)sys_oshmctl,
69 (sy_call_t *)sys_shmdt, (sy_call_t *)sys_shmget,
70 (sy_call_t *)sys_shmctl
73 #define SHMSEG_FREE 0x0200
74 #define SHMSEG_REMOVED 0x0400
75 #define SHMSEG_ALLOCATED 0x0800
76 #define SHMSEG_WANTED 0x1000
78 static int shm_last_free, shm_committed, shmalloced;
80 static struct shmid_ds *shmsegs;
83 /* vm_offset_t kva; */
84 vm_object_t shm_object;
92 static void shm_deallocate_segment (struct shmid_ds *);
93 static int shm_find_segment_by_key (key_t);
94 static struct shmid_ds *shm_find_segment_by_shmid (int);
95 static int shm_delete_mapping (struct vmspace *vm, struct shmmap_state *);
96 static void shmrealloc (void);
97 static void shminit (void *);
112 struct shminfo shminfo = {
120 static int shm_use_phys = 1;
122 TUNABLE_LONG("kern.ipc.shmmin", &shminfo.shmmin);
123 TUNABLE_LONG("kern.ipc.shmmni", &shminfo.shmmni);
124 TUNABLE_LONG("kern.ipc.shmseg", &shminfo.shmseg);
125 TUNABLE_LONG("kern.ipc.shmmaxpgs", &shminfo.shmall);
126 TUNABLE_INT("kern.ipc.shm_use_phys", &shm_use_phys);
128 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0,
129 "Max shared memory segment size");
130 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0,
131 "Min shared memory segment size");
132 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RD, &shminfo.shmmni, 0,
133 "Max number of shared memory identifiers");
134 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RW, &shminfo.shmseg, 0,
135 "Max shared memory segments per process");
136 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0,
137 "Max pages of shared memory");
138 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW, &shm_use_phys, 0,
139 "Use phys pager allocation instead of swap pager allocation");
142 shm_find_segment_by_key(key_t key)
146 for (i = 0; i < shmalloced; i++) {
147 if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
148 shmsegs[i].shm_perm.key == key)
154 static struct shmid_ds *
155 shm_find_segment_by_shmid(int shmid)
158 struct shmid_ds *shmseg;
160 segnum = IPCID_TO_IX(shmid);
161 if (segnum < 0 || segnum >= shmalloced)
163 shmseg = &shmsegs[segnum];
164 if ((shmseg->shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
165 != SHMSEG_ALLOCATED ||
166 shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid)) {
173 shm_deallocate_segment(struct shmid_ds *shmseg)
175 struct shm_handle *shm_handle;
178 shm_handle = shmseg->shm_internal;
179 vm_object_deallocate(shm_handle->shm_object);
180 kfree((caddr_t)shm_handle, M_SHM);
181 shmseg->shm_internal = NULL;
182 size = round_page(shmseg->shm_segsz);
183 shm_committed -= btoc(size);
185 shmseg->shm_perm.mode = SHMSEG_FREE;
189 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
191 struct shmid_ds *shmseg;
195 segnum = IPCID_TO_IX(shmmap_s->shmid);
196 shmseg = &shmsegs[segnum];
197 size = round_page(shmseg->shm_segsz);
198 result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
199 if (result != KERN_SUCCESS)
201 shmmap_s->shmid = -1;
202 shmseg->shm_dtime = time_second;
203 if ((--shmseg->shm_nattch <= 0) &&
204 (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
205 shm_deallocate_segment(shmseg);
206 shm_last_free = segnum;
215 sys_shmdt(struct shmdt_args *uap)
217 struct thread *td = curthread;
218 struct proc *p = td->td_proc;
219 struct shmmap_state *shmmap_s;
223 if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
227 shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
228 if (shmmap_s == NULL) {
232 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
233 if (shmmap_s->shmid != -1 &&
234 shmmap_s->va == (vm_offset_t)uap->shmaddr)
237 if (i == shminfo.shmseg)
240 error = shm_delete_mapping(p->p_vmspace, shmmap_s);
250 sys_shmat(struct shmat_args *uap)
252 struct thread *td = curthread;
253 struct proc *p = td->td_proc;
256 struct shmid_ds *shmseg;
257 struct shmmap_state *shmmap_s = NULL;
258 struct shm_handle *shm_handle;
259 vm_offset_t attach_va;
265 if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
270 shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
271 if (shmmap_s == NULL) {
272 size = shminfo.shmseg * sizeof(struct shmmap_state);
273 shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
274 for (i = 0; i < shminfo.shmseg; i++)
275 shmmap_s[i].shmid = -1;
276 if (p->p_vmspace->vm_shm != NULL) {
277 kfree(shmmap_s, M_SHM);
280 p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
282 shmseg = shm_find_segment_by_shmid(uap->shmid);
283 if (shmseg == NULL) {
287 error = ipcperm(p, &shmseg->shm_perm,
288 (uap->shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
291 for (i = 0; i < shminfo.shmseg; i++) {
292 if (shmmap_s->shmid == -1)
296 if (i >= shminfo.shmseg) {
300 size = round_page(shmseg->shm_segsz);
301 #ifdef VM_PROT_READ_IS_EXEC
302 prot = VM_PROT_READ | VM_PROT_EXECUTE;
306 if ((uap->shmflg & SHM_RDONLY) == 0)
307 prot |= VM_PROT_WRITE;
308 flags = MAP_ANON | MAP_SHARED;
311 if (uap->shmflg & SHM_RND) {
312 attach_va = (vm_offset_t)uap->shmaddr & ~(SHMLBA-1);
313 } else if (((vm_offset_t)uap->shmaddr & (SHMLBA-1)) == 0) {
314 attach_va = (vm_offset_t)uap->shmaddr;
321 * This is just a hint to vm_map_find() about where to put it.
323 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_taddr +
328 * Handle alignment. For large memory maps it is possible
329 * that the MMU can optimize the page table so align anything
330 * that is a multiple of SEG_SIZE to SEG_SIZE.
332 if ((flags & MAP_FIXED) == 0 && (size & SEG_MASK) == 0)
337 shm_handle = shmseg->shm_internal;
338 vm_object_hold(shm_handle->shm_object);
339 vm_object_chain_wait(shm_handle->shm_object);
340 vm_object_reference_locked(shm_handle->shm_object);
341 rv = vm_map_find(&p->p_vmspace->vm_map,
342 shm_handle->shm_object, 0,
345 ((flags & MAP_FIXED) ? 0 : 1),
349 vm_object_drop(shm_handle->shm_object);
350 if (rv != KERN_SUCCESS) {
351 vm_object_deallocate(shm_handle->shm_object);
355 vm_map_inherit(&p->p_vmspace->vm_map,
356 attach_va, attach_va + size, VM_INHERIT_SHARE);
358 KKASSERT(shmmap_s->shmid == -1);
359 shmmap_s->va = attach_va;
360 shmmap_s->shmid = uap->shmid;
361 shmseg->shm_lpid = p->p_pid;
362 shmseg->shm_atime = time_second;
363 shmseg->shm_nattch++;
364 uap->sysmsg_resultp = (void *)attach_va;
372 struct ipc_perm shm_perm; /* operation perms */
373 int shm_segsz; /* size of segment (bytes) */
374 ushort shm_cpid; /* pid, creator */
375 ushort shm_lpid; /* pid, last operation */
376 short shm_nattch; /* no. of current attaches */
377 time_t shm_atime; /* last attach time */
378 time_t shm_dtime; /* last detach time */
379 time_t shm_ctime; /* last change time */
380 void *shm_handle; /* internal handle for shm segment */
383 struct oshmctl_args {
384 struct sysmsg sysmsg;
387 struct oshmid_ds *ubuf;
394 sys_oshmctl(struct proc *p, struct oshmctl_args *uap)
397 struct thread *td = curthread;
398 struct shmid_ds *shmseg;
399 struct oshmid_ds outbuf;
402 if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
406 shmseg = shm_find_segment_by_shmid(uap->shmid);
407 if (shmseg == NULL) {
414 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
417 outbuf.shm_perm = shmseg->shm_perm;
418 outbuf.shm_segsz = shmseg->shm_segsz;
419 outbuf.shm_cpid = shmseg->shm_cpid;
420 outbuf.shm_lpid = shmseg->shm_lpid;
421 outbuf.shm_nattch = shmseg->shm_nattch;
422 outbuf.shm_atime = shmseg->shm_atime;
423 outbuf.shm_dtime = shmseg->shm_dtime;
424 outbuf.shm_ctime = shmseg->shm_ctime;
425 outbuf.shm_handle = shmseg->shm_internal;
426 error = copyout((caddr_t)&outbuf, uap->ubuf, sizeof(outbuf));
429 /* XXX casting to (sy_call_t *) is bogus, as usual. */
430 error = sys_shmctl((struct shmctl_args *)uap);
444 sys_shmctl(struct shmctl_args *uap)
446 struct thread *td = curthread;
447 struct proc *p = td->td_proc;
449 struct shmid_ds inbuf;
450 struct shmid_ds *shmseg;
452 if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
456 shmseg = shm_find_segment_by_shmid(uap->shmid);
457 if (shmseg == NULL) {
464 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
466 error = copyout(shmseg, uap->buf, sizeof(inbuf));
469 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
471 error = copyin(uap->buf, &inbuf, sizeof(inbuf));
473 shmseg->shm_perm.uid = inbuf.shm_perm.uid;
474 shmseg->shm_perm.gid = inbuf.shm_perm.gid;
475 shmseg->shm_perm.mode =
476 (shmseg->shm_perm.mode & ~ACCESSPERMS) |
477 (inbuf.shm_perm.mode & ACCESSPERMS);
478 shmseg->shm_ctime = time_second;
482 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
484 shmseg->shm_perm.key = IPC_PRIVATE;
485 shmseg->shm_perm.mode |= SHMSEG_REMOVED;
486 if (shmseg->shm_nattch <= 0) {
487 shm_deallocate_segment(shmseg);
488 shm_last_free = IPCID_TO_IX(uap->shmid);
506 shmget_existing(struct proc *p, struct shmget_args *uap, int mode, int segnum)
508 struct shmid_ds *shmseg;
511 shmseg = &shmsegs[segnum];
512 if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
514 * This segment is in the process of being allocated. Wait
515 * until it's done, and look the key up again (in case the
516 * allocation failed or it was freed).
518 shmseg->shm_perm.mode |= SHMSEG_WANTED;
519 error = tsleep((caddr_t)shmseg, PCATCH, "shmget", 0);
524 if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
526 error = ipcperm(p, &shmseg->shm_perm, mode);
529 if (uap->size && uap->size > shmseg->shm_segsz)
531 uap->sysmsg_result = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
536 shmget_allocate_segment(struct proc *p, struct shmget_args *uap, int mode)
538 int i, segnum, shmid;
540 struct ucred *cred = p->p_ucred;
541 struct shmid_ds *shmseg;
542 struct shm_handle *shm_handle;
544 if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
546 if (shm_nused >= shminfo.shmmni) /* any shmids left? */
548 size = round_page(uap->size);
549 if (shm_committed + btoc(size) > shminfo.shmall)
551 if (shm_last_free < 0) {
552 shmrealloc(); /* maybe expand the shmsegs[] array */
553 for (i = 0; i < shmalloced; i++) {
554 if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
561 segnum = shm_last_free;
564 shmseg = &shmsegs[segnum];
566 * In case we sleep in malloc(), mark the segment present but deleted
567 * so that noone else tries to create the same key.
569 shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
570 shmseg->shm_perm.key = uap->key;
571 shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
572 shm_handle = kmalloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
573 shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
576 * We make sure that we have allocated a pager before we need
580 shm_handle->shm_object =
581 phys_pager_alloc(NULL, size, VM_PROT_DEFAULT, 0);
583 shm_handle->shm_object =
584 swap_pager_alloc(NULL, size, VM_PROT_DEFAULT, 0);
586 vm_object_clear_flag(shm_handle->shm_object, OBJ_ONEMAPPING);
587 vm_object_set_flag(shm_handle->shm_object, OBJ_NOSPLIT);
589 shmseg->shm_internal = shm_handle;
590 shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
591 shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
592 shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
593 (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
594 shmseg->shm_segsz = uap->size;
595 shmseg->shm_cpid = p->p_pid;
596 shmseg->shm_lpid = shmseg->shm_nattch = 0;
597 shmseg->shm_atime = shmseg->shm_dtime = 0;
598 shmseg->shm_ctime = time_second;
599 shm_committed += btoc(size);
603 * If a physical mapping is desired and we have a ton of free pages
604 * we pre-allocate the pages here in order to avoid on-the-fly
605 * allocation later. This has a big effect on database warm-up
606 * times since DFly supports concurrent page faults coming from the
607 * same VM object for pages which already exist.
609 * This can hang the kernel for a while so only do it if shm_use_phys
610 * is set to 2 or higher.
612 if (shm_use_phys > 1) {
613 vm_pindex_t pi, pmax;
616 pmax = round_page(shmseg->shm_segsz) >> PAGE_SHIFT;
617 vm_object_hold(shm_handle->shm_object);
618 if (pmax > vmstats.v_free_count)
619 pmax = vmstats.v_free_count;
620 for (pi = 0; pi < pmax; ++pi) {
621 m = vm_page_grab(shm_handle->shm_object, pi,
622 VM_ALLOC_SYSTEM | VM_ALLOC_NULL_OK |
626 vm_pager_get_page(shm_handle->shm_object, &m, 1);
631 vm_object_drop(shm_handle->shm_object);
634 if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
636 * Somebody else wanted this key while we were asleep. Wake
639 shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
640 wakeup((caddr_t)shmseg);
642 uap->sysmsg_result = shmid;
650 sys_shmget(struct shmget_args *uap)
652 struct thread *td = curthread;
653 struct proc *p = td->td_proc;
654 int segnum, mode, error;
656 if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
659 mode = uap->shmflg & ACCESSPERMS;
662 if (uap->key != IPC_PRIVATE) {
664 segnum = shm_find_segment_by_key(uap->key);
666 error = shmget_existing(p, uap, mode, segnum);
671 if ((uap->shmflg & IPC_CREAT) == 0) {
676 error = shmget_allocate_segment(p, uap, mode);
683 * shmsys_args(int which, int a2, ...) (VARARGS)
688 sys_shmsys(struct shmsys_args *uap)
690 struct thread *td = curthread;
691 unsigned int which = (unsigned int)uap->which;
694 if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
697 if (which >= NELEM(shmcalls))
700 bcopy(&uap->a2, &uap->which,
701 sizeof(struct shmsys_args) - offsetof(struct shmsys_args, a2));
702 error = ((*shmcalls[which])(uap));
709 shmfork(struct proc *p1, struct proc *p2)
711 struct shmmap_state *shmmap_s;
716 size = shminfo.shmseg * sizeof(struct shmmap_state);
717 shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
718 bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmmap_s, size);
719 p2->p_vmspace->vm_shm = (caddr_t)shmmap_s;
720 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
721 if (shmmap_s->shmid != -1)
722 shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
728 shmexit(struct vmspace *vm)
730 struct shmmap_state *base, *shm;
733 if ((base = (struct shmmap_state *)vm->vm_shm) != NULL) {
736 for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
737 if (shm->shmid != -1)
738 shm_delete_mapping(vm, shm);
749 struct shmid_ds *newsegs;
751 if (shmalloced >= shminfo.shmmni)
754 newsegs = kmalloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
755 for (i = 0; i < shmalloced; i++)
756 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
757 for (; i < shminfo.shmmni; i++) {
758 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
759 shmsegs[i].shm_perm.seq = 0;
761 kfree(shmsegs, M_SHM);
763 shmalloced = shminfo.shmmni;
772 * If not overridden by a tunable set the maximum shm to
773 * 2/3 of main memory.
775 if (shminfo.shmall == 0)
776 shminfo.shmall = (size_t)vmstats.v_page_count * 2 / 3;
778 shminfo.shmmax = shminfo.shmall * PAGE_SIZE;
779 shmalloced = shminfo.shmmni;
780 shmsegs = kmalloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
781 for (i = 0; i < shmalloced; i++) {
782 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
783 shmsegs[i].shm_perm.seq = 0;
789 SYSINIT(sysv_shm, SI_SUB_SYSV_SHM, SI_ORDER_FIRST, shminit, NULL);