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_sysvipc.h"
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/sysproto.h>
36 #include <sys/kernel.h>
37 #include <sys/sysctl.h>
40 #include <sys/malloc.h>
43 #include <sys/sysent.h>
47 #include <vm/vm_param.h>
50 #include <vm/vm_object.h>
51 #include <vm/vm_map.h>
52 #include <vm/vm_page.h>
53 #include <vm/vm_pager.h>
55 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
57 static int shmget_allocate_segment (struct proc *p, struct shmget_args *uap, int mode);
58 static int shmget_existing (struct proc *p, struct shmget_args *uap, int mode, int segnum);
60 #define SHMSEG_FREE 0x0200
61 #define SHMSEG_REMOVED 0x0400
62 #define SHMSEG_ALLOCATED 0x0800
63 #define SHMSEG_WANTED 0x1000
65 static int shm_last_free, shm_committed, shmalloced;
67 static struct shmid_ds *shmsegs;
68 static struct lwkt_token shm_token = LWKT_TOKEN_INITIALIZER(shm_token);
71 /* vm_offset_t kva; */
72 vm_object_t shm_object;
80 static void shm_deallocate_segment (struct shmid_ds *);
81 static int shm_find_segment_by_key (key_t);
82 static struct shmid_ds *shm_find_segment_by_shmid (int);
83 static int shm_delete_mapping (struct vmspace *vm, struct shmmap_state *);
84 static void shmrealloc (void);
85 static void shminit (void *);
100 struct shminfo shminfo = {
109 * allow-removed Allow a shared memory segment to be attached by its shmid
110 * even after it has been deleted, as long as it was still
111 * being referenced by someone. This is a trick used by
112 * chrome and other applications to avoid leaving shm
113 * segments hanging around after the application is killed
114 * or seg-faults unexpectedly.
116 * use-phys Shared memory segments are to use physical memory by
117 * default, which allows the kernel to optimize (remove)
118 * pv_entry management structures for the related PTEs and
119 * prevents paging. This has distinctly different and
120 * usually desireable characteristics verses mmap()ing
123 static int shm_allow_removed = 1;
124 static int shm_use_phys = 1;
126 TUNABLE_LONG("kern.ipc.shmmin", &shminfo.shmmin);
127 TUNABLE_LONG("kern.ipc.shmmni", &shminfo.shmmni);
128 TUNABLE_LONG("kern.ipc.shmseg", &shminfo.shmseg);
129 TUNABLE_LONG("kern.ipc.shmmaxpgs", &shminfo.shmall);
130 TUNABLE_INT("kern.ipc.shm_use_phys", &shm_use_phys);
132 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0,
133 "Max shared memory segment size");
134 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0,
135 "Min shared memory segment size");
136 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RD, &shminfo.shmmni, 0,
137 "Max number of shared memory identifiers");
138 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RW, &shminfo.shmseg, 0,
139 "Max shared memory segments per process");
140 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0,
141 "Max pages of shared memory");
142 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW, &shm_use_phys, 0,
143 "Use phys pager allocation instead of swap pager allocation");
144 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RW,
145 &shm_allow_removed, 0,
146 "Enable/Disable attachment to attached segments marked for removal");
149 shm_find_segment_by_key(key_t key)
153 for (i = 0; i < shmalloced; i++) {
154 if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
155 shmsegs[i].shm_perm.key == key)
161 static struct shmid_ds *
162 shm_find_segment_by_shmid(int shmid)
165 struct shmid_ds *shmseg;
167 segnum = IPCID_TO_IX(shmid);
168 if (segnum < 0 || segnum >= shmalloced)
170 shmseg = &shmsegs[segnum];
171 if ((shmseg->shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
172 (!shm_allow_removed &&
173 (shmseg->shm_perm.mode & SHMSEG_REMOVED) != 0) ||
174 shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid)) {
181 shm_deallocate_segment(struct shmid_ds *shmseg)
183 struct shm_handle *shm_handle;
186 shm_handle = shmseg->shm_internal;
187 vm_object_deallocate(shm_handle->shm_object);
188 kfree((caddr_t)shm_handle, M_SHM);
189 shmseg->shm_internal = NULL;
190 size = round_page(shmseg->shm_segsz);
191 shm_committed -= btoc(size);
193 shmseg->shm_perm.mode = SHMSEG_FREE;
197 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
199 struct shmid_ds *shmseg;
203 segnum = IPCID_TO_IX(shmmap_s->shmid);
204 shmseg = &shmsegs[segnum];
205 size = round_page(shmseg->shm_segsz);
206 result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
207 if (result != KERN_SUCCESS)
209 shmmap_s->shmid = -1;
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;
223 sys_shmdt(struct shmdt_args *uap)
225 struct thread *td = curthread;
226 struct proc *p = td->td_proc;
227 struct shmmap_state *shmmap_s;
231 if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
234 lwkt_gettoken(&shm_token);
235 shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
236 if (shmmap_s == NULL) {
240 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
241 if (shmmap_s->shmid != -1 &&
242 shmmap_s->va == (vm_offset_t)uap->shmaddr)
245 if (i == shminfo.shmseg)
248 error = shm_delete_mapping(p->p_vmspace, shmmap_s);
250 lwkt_reltoken(&shm_token);
259 sys_shmat(struct shmat_args *uap)
261 struct thread *td = curthread;
262 struct proc *p = td->td_proc;
265 struct shmid_ds *shmseg;
266 struct shmmap_state *shmmap_s = NULL;
267 struct shm_handle *shm_handle;
268 vm_offset_t attach_va;
274 if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
277 lwkt_gettoken(&shm_token);
279 shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
280 if (shmmap_s == NULL) {
281 size = shminfo.shmseg * sizeof(struct shmmap_state);
282 shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
283 for (i = 0; i < shminfo.shmseg; i++)
284 shmmap_s[i].shmid = -1;
285 if (p->p_vmspace->vm_shm != NULL) {
286 kfree(shmmap_s, M_SHM);
289 p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
291 shmseg = shm_find_segment_by_shmid(uap->shmid);
292 if (shmseg == NULL) {
296 error = ipcperm(p, &shmseg->shm_perm,
297 (uap->shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
300 for (i = 0; i < shminfo.shmseg; i++) {
301 if (shmmap_s->shmid == -1)
305 if (i >= shminfo.shmseg) {
309 size = round_page(shmseg->shm_segsz);
310 #ifdef VM_PROT_READ_IS_EXEC
311 prot = VM_PROT_READ | VM_PROT_EXECUTE;
315 if ((uap->shmflg & SHM_RDONLY) == 0)
316 prot |= VM_PROT_WRITE;
317 flags = MAP_ANON | MAP_SHARED;
320 if (uap->shmflg & SHM_RND) {
321 attach_va = (vm_offset_t)uap->shmaddr & ~(SHMLBA-1);
322 } else if (((vm_offset_t)uap->shmaddr & (SHMLBA-1)) == 0) {
323 attach_va = (vm_offset_t)uap->shmaddr;
330 * This is just a hint to vm_map_find() about where to put it.
332 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_taddr +
337 * Handle alignment. For large memory maps it is possible
338 * that the MMU can optimize the page table so align anything
339 * that is a multiple of SEG_SIZE to SEG_SIZE.
341 if ((flags & MAP_FIXED) == 0 && (size & SEG_MASK) == 0)
346 shm_handle = shmseg->shm_internal;
347 vm_object_hold(shm_handle->shm_object);
348 vm_object_chain_wait(shm_handle->shm_object, 0);
349 vm_object_reference_locked(shm_handle->shm_object);
350 rv = vm_map_find(&p->p_vmspace->vm_map,
351 shm_handle->shm_object, NULL,
354 ((flags & MAP_FIXED) ? 0 : 1),
355 VM_MAPTYPE_NORMAL, VM_SUBSYS_SHMEM,
357 vm_object_drop(shm_handle->shm_object);
358 if (rv != KERN_SUCCESS) {
359 vm_object_deallocate(shm_handle->shm_object);
363 vm_map_inherit(&p->p_vmspace->vm_map,
364 attach_va, attach_va + size, VM_INHERIT_SHARE);
366 KKASSERT(shmmap_s->shmid == -1);
367 shmmap_s->va = attach_va;
368 shmmap_s->shmid = uap->shmid;
369 shmseg->shm_lpid = p->p_pid;
370 shmseg->shm_atime = time_second;
371 shmseg->shm_nattch++;
372 uap->sysmsg_resultp = (void *)attach_va;
375 lwkt_reltoken(&shm_token);
384 sys_shmctl(struct shmctl_args *uap)
386 struct thread *td = curthread;
387 struct proc *p = td->td_proc;
389 struct shmid_ds inbuf;
390 struct shmid_ds *shmseg;
392 if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
395 lwkt_gettoken(&shm_token);
396 shmseg = shm_find_segment_by_shmid(uap->shmid);
397 if (shmseg == NULL) {
404 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
406 error = copyout(shmseg, uap->buf, sizeof(inbuf));
409 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
411 error = copyin(uap->buf, &inbuf, sizeof(inbuf));
413 shmseg->shm_perm.uid = inbuf.shm_perm.uid;
414 shmseg->shm_perm.gid = inbuf.shm_perm.gid;
415 shmseg->shm_perm.mode =
416 (shmseg->shm_perm.mode & ~ACCESSPERMS) |
417 (inbuf.shm_perm.mode & ACCESSPERMS);
418 shmseg->shm_ctime = time_second;
422 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
424 shmseg->shm_perm.key = IPC_PRIVATE;
425 shmseg->shm_perm.mode |= SHMSEG_REMOVED;
426 if (shmseg->shm_nattch <= 0) {
427 shm_deallocate_segment(shmseg);
428 shm_last_free = IPCID_TO_IX(uap->shmid);
441 lwkt_reltoken(&shm_token);
447 shmget_existing(struct proc *p, struct shmget_args *uap, int mode, int segnum)
449 struct shmid_ds *shmseg;
452 shmseg = &shmsegs[segnum];
453 if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
455 * This segment is in the process of being allocated. Wait
456 * until it's done, and look the key up again (in case the
457 * allocation failed or it was freed).
459 shmseg->shm_perm.mode |= SHMSEG_WANTED;
460 error = tsleep((caddr_t)shmseg, PCATCH, "shmget", 0);
465 if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
467 error = ipcperm(p, &shmseg->shm_perm, mode);
470 if (uap->size && uap->size > shmseg->shm_segsz)
472 uap->sysmsg_result = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
477 shmget_allocate_segment(struct proc *p, struct shmget_args *uap, int mode)
479 int i, segnum, shmid;
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)
502 segnum = shm_last_free;
505 shmseg = &shmsegs[segnum];
507 * In case we sleep in malloc(), mark the segment present but deleted
508 * so that noone else tries to create the same key.
510 shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
511 shmseg->shm_perm.key = uap->key;
512 shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
513 shm_handle = kmalloc(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 phys_pager_alloc(NULL, size, VM_PROT_DEFAULT, 0);
524 shm_handle->shm_object =
525 swap_pager_alloc(NULL, 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);
544 * If a physical mapping is desired and we have a ton of free pages
545 * we pre-allocate the pages here in order to avoid on-the-fly
546 * allocation later. This has a big effect on database warm-up
547 * times since DFly supports concurrent page faults coming from the
548 * same VM object for pages which already exist.
550 * This can hang the kernel for a while so only do it if shm_use_phys
551 * is set to 2 or higher.
553 if (shm_use_phys > 1) {
554 vm_pindex_t pi, pmax;
557 pmax = round_page(shmseg->shm_segsz) >> PAGE_SHIFT;
558 vm_object_hold(shm_handle->shm_object);
559 if (pmax > vmstats.v_free_count)
560 pmax = vmstats.v_free_count;
561 for (pi = 0; pi < pmax; ++pi) {
562 m = vm_page_grab(shm_handle->shm_object, pi,
563 VM_ALLOC_SYSTEM | VM_ALLOC_NULL_OK |
567 vm_pager_get_page(shm_handle->shm_object, &m, 1);
572 vm_object_drop(shm_handle->shm_object);
575 if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
577 * Somebody else wanted this key while we were asleep. Wake
580 shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
581 wakeup((caddr_t)shmseg);
583 uap->sysmsg_result = shmid;
591 sys_shmget(struct shmget_args *uap)
593 struct thread *td = curthread;
594 struct proc *p = td->td_proc;
595 int segnum, mode, error;
597 if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
600 mode = uap->shmflg & ACCESSPERMS;
602 lwkt_gettoken(&shm_token);
604 if (uap->key != IPC_PRIVATE) {
606 segnum = shm_find_segment_by_key(uap->key);
608 error = shmget_existing(p, uap, mode, segnum);
613 if ((uap->shmflg & IPC_CREAT) == 0) {
618 error = shmget_allocate_segment(p, uap, mode);
620 lwkt_reltoken(&shm_token);
626 shmfork(struct proc *p1, struct proc *p2)
628 struct shmmap_state *shmmap_s;
632 lwkt_gettoken(&shm_token);
633 size = shminfo.shmseg * sizeof(struct shmmap_state);
634 shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
635 bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmmap_s, size);
636 p2->p_vmspace->vm_shm = (caddr_t)shmmap_s;
637 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
638 if (shmmap_s->shmid != -1)
639 shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
641 lwkt_reltoken(&shm_token);
645 shmexit(struct vmspace *vm)
647 struct shmmap_state *base, *shm;
650 if ((base = (struct shmmap_state *)vm->vm_shm) != NULL) {
652 lwkt_gettoken(&shm_token);
653 for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
654 if (shm->shmid != -1)
655 shm_delete_mapping(vm, shm);
658 lwkt_reltoken(&shm_token);
666 struct shmid_ds *newsegs;
668 if (shmalloced >= shminfo.shmmni)
671 newsegs = kmalloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
672 for (i = 0; i < shmalloced; i++)
673 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
674 for (; i < shminfo.shmmni; i++) {
675 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
676 shmsegs[i].shm_perm.seq = 0;
678 kfree(shmsegs, M_SHM);
680 shmalloced = shminfo.shmmni;
689 * If not overridden by a tunable set the maximum shm to
690 * 2/3 of main memory.
692 if (shminfo.shmall == 0)
693 shminfo.shmall = (size_t)vmstats.v_page_count * 2 / 3;
695 shminfo.shmmax = shminfo.shmall * PAGE_SIZE;
696 shmalloced = shminfo.shmmni;
697 shmsegs = kmalloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
698 for (i = 0; i < shmalloced; i++) {
699 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
700 shmsegs[i].shm_perm.seq = 0;
706 SYSINIT(sysv_shm, SI_SUB_SYSV_SHM, SI_ORDER_FIRST, shminit, NULL);