2 * Copyright (c) 2006, 2011 Robert N. M. Watson
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * Support for shared swap-backed anonymous memory objects via
29 * shm_open(2) and shm_unlink(2). While most of the implementation is
30 * here, vm_mmap.c contains mapping logic changes.
34 * (1) Need to export data to a userland tool via a sysctl. Should ipcs(1)
35 * and ipcrm(1) be expanded or should new tools to manage both POSIX
36 * kernel semaphores and POSIX shared memory be written?
38 * (2) Add support for this file type to fstat(1).
40 * (3) Resource limits? Does this need its own resource limits or are the
41 * existing limits in mmap(2) sufficient?
44 #include <sys/cdefs.h>
45 __FBSDID("$FreeBSD$");
47 #include "opt_capsicum.h"
49 #include <sys/param.h>
50 #include <sys/capability.h>
51 #include <sys/fcntl.h>
53 #include <sys/filedesc.h>
54 #include <sys/fnv_hash.h>
55 #include <sys/kernel.h>
57 #include <sys/malloc.h>
59 #include <sys/mutex.h>
62 #include <sys/refcount.h>
63 #include <sys/resourcevar.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysproto.h>
67 #include <sys/systm.h>
70 #include <sys/vnode.h>
72 #include <security/mac/mac_framework.h>
75 #include <vm/vm_param.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_pageout.h>
81 #include <vm/vm_pager.h>
82 #include <vm/swap_pager.h>
87 struct shmfd *sm_shmfd;
88 LIST_ENTRY(shm_mapping) sm_link;
91 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
92 static LIST_HEAD(, shm_mapping) *shm_dictionary;
93 static struct sx shm_dict_lock;
94 static struct mtx shm_timestamp_lock;
95 static u_long shm_hash;
97 #define SHM_HASH(fnv) (&shm_dictionary[(fnv) & shm_hash])
99 static int shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags);
100 static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode);
101 static void shm_dict_init(void *arg);
102 static void shm_drop(struct shmfd *shmfd);
103 static struct shmfd *shm_hold(struct shmfd *shmfd);
104 static void shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
105 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
106 static int shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
107 static int shm_dotruncate(struct shmfd *shmfd, off_t length);
109 static fo_rdwr_t shm_read;
110 static fo_rdwr_t shm_write;
111 static fo_truncate_t shm_truncate;
112 static fo_ioctl_t shm_ioctl;
113 static fo_poll_t shm_poll;
114 static fo_kqfilter_t shm_kqfilter;
115 static fo_stat_t shm_stat;
116 static fo_close_t shm_close;
117 static fo_chmod_t shm_chmod;
118 static fo_chown_t shm_chown;
120 /* File descriptor operations. */
121 static struct fileops shm_ops = {
123 .fo_write = shm_write,
124 .fo_truncate = shm_truncate,
125 .fo_ioctl = shm_ioctl,
127 .fo_kqfilter = shm_kqfilter,
129 .fo_close = shm_close,
130 .fo_chmod = shm_chmod,
131 .fo_chown = shm_chown,
132 .fo_flags = DFLAG_PASSABLE
135 FEATURE(posix_shm, "POSIX shared memory");
138 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
139 int flags, struct thread *td)
146 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
147 int flags, struct thread *td)
154 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
164 error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
168 return (shm_dotruncate(shmfd, length));
172 shm_ioctl(struct file *fp, u_long com, void *data,
173 struct ucred *active_cred, struct thread *td)
180 shm_poll(struct file *fp, int events, struct ucred *active_cred,
188 shm_kqfilter(struct file *fp, struct knote *kn)
195 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
206 error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
212 * Attempt to return sanish values for fstat() on a memory file
215 bzero(sb, sizeof(*sb));
216 sb->st_blksize = PAGE_SIZE;
217 sb->st_size = shmfd->shm_size;
218 sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize;
219 mtx_lock(&shm_timestamp_lock);
220 sb->st_atim = shmfd->shm_atime;
221 sb->st_ctim = shmfd->shm_ctime;
222 sb->st_mtim = shmfd->shm_mtime;
223 sb->st_birthtim = shmfd->shm_birthtime;
224 sb->st_mode = S_IFREG | shmfd->shm_mode; /* XXX */
225 sb->st_uid = shmfd->shm_uid;
226 sb->st_gid = shmfd->shm_gid;
227 mtx_unlock(&shm_timestamp_lock);
233 shm_close(struct file *fp, struct thread *td)
245 shm_dotruncate(struct shmfd *shmfd, off_t length)
249 vm_pindex_t idx, nobjsize;
253 object = shmfd->shm_object;
254 VM_OBJECT_LOCK(object);
255 if (length == shmfd->shm_size) {
256 VM_OBJECT_UNLOCK(object);
259 nobjsize = OFF_TO_IDX(length + PAGE_MASK);
261 /* Are we shrinking? If so, trim the end. */
262 if (length < shmfd->shm_size) {
265 * Zero the truncated part of the last page.
267 base = length & PAGE_MASK;
269 idx = OFF_TO_IDX(length);
271 m = vm_page_lookup(object, idx);
273 if ((m->oflags & VPO_BUSY) != 0 ||
275 vm_page_sleep(m, "shmtrc");
278 } else if (vm_pager_has_page(object, idx, NULL, NULL)) {
279 m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL);
281 VM_OBJECT_UNLOCK(object);
283 VM_OBJECT_LOCK(object);
285 } else if (m->valid != VM_PAGE_BITS_ALL) {
287 rv = vm_pager_get_pages(object, ma, 1,
289 m = vm_page_lookup(object, idx);
291 /* A cached page was reactivated. */
294 if (rv == VM_PAGER_OK) {
295 vm_page_deactivate(m);
301 VM_OBJECT_UNLOCK(object);
306 pmap_zero_page_area(m, base, PAGE_SIZE - base);
307 KASSERT(m->valid == VM_PAGE_BITS_ALL,
308 ("shm_dotruncate: page %p is invalid", m));
310 vm_pager_page_unswapped(m);
313 delta = ptoa(object->size - nobjsize);
315 /* Toss in memory pages. */
316 if (nobjsize < object->size)
317 vm_object_page_remove(object, nobjsize, object->size,
320 /* Toss pages from swap. */
321 if (object->type == OBJT_SWAP)
322 swap_pager_freespace(object, nobjsize, delta);
324 /* Free the swap accounted for shm */
325 swap_release_by_cred(delta, object->cred);
326 object->charge -= delta;
328 /* Attempt to reserve the swap */
329 delta = ptoa(nobjsize - object->size);
330 if (!swap_reserve_by_cred(delta, object->cred)) {
331 VM_OBJECT_UNLOCK(object);
334 object->charge += delta;
336 shmfd->shm_size = length;
337 mtx_lock(&shm_timestamp_lock);
338 vfs_timestamp(&shmfd->shm_ctime);
339 shmfd->shm_mtime = shmfd->shm_ctime;
340 mtx_unlock(&shm_timestamp_lock);
341 object->size = nobjsize;
342 VM_OBJECT_UNLOCK(object);
347 * shmfd object management including creation and reference counting
350 static struct shmfd *
351 shm_alloc(struct ucred *ucred, mode_t mode)
355 shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
357 shmfd->shm_uid = ucred->cr_uid;
358 shmfd->shm_gid = ucred->cr_gid;
359 shmfd->shm_mode = mode;
360 shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
361 shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
362 KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
363 VM_OBJECT_LOCK(shmfd->shm_object);
364 vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
365 vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
366 VM_OBJECT_UNLOCK(shmfd->shm_object);
367 vfs_timestamp(&shmfd->shm_birthtime);
368 shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
369 shmfd->shm_birthtime;
370 refcount_init(&shmfd->shm_refs, 1);
372 mac_posixshm_init(shmfd);
373 mac_posixshm_create(ucred, shmfd);
379 static struct shmfd *
380 shm_hold(struct shmfd *shmfd)
383 refcount_acquire(&shmfd->shm_refs);
388 shm_drop(struct shmfd *shmfd)
391 if (refcount_release(&shmfd->shm_refs)) {
393 mac_posixshm_destroy(shmfd);
395 vm_object_deallocate(shmfd->shm_object);
396 free(shmfd, M_SHMFD);
401 * Determine if the credentials have sufficient permissions for a
402 * specified combination of FREAD and FWRITE.
405 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
415 mtx_lock(&shm_timestamp_lock);
416 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
417 accmode, ucred, NULL);
418 mtx_unlock(&shm_timestamp_lock);
423 * Dictionary management. We maintain an in-kernel dictionary to map
424 * paths to shmfd objects. We use the FNV hash on the path to store
425 * the mappings in a hash table.
428 shm_dict_init(void *arg)
431 mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
432 sx_init(&shm_dict_lock, "shm dictionary");
433 shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
435 SYSINIT(shm_dict_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_dict_init, NULL);
437 static struct shmfd *
438 shm_lookup(char *path, Fnv32_t fnv)
440 struct shm_mapping *map;
442 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
443 if (map->sm_fnv != fnv)
445 if (strcmp(map->sm_path, path) == 0)
446 return (map->sm_shmfd);
453 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
455 struct shm_mapping *map;
457 map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
460 map->sm_shmfd = shm_hold(shmfd);
461 LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
465 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
467 struct shm_mapping *map;
470 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
471 if (map->sm_fnv != fnv)
473 if (strcmp(map->sm_path, path) == 0) {
475 error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
479 error = shm_access(map->sm_shmfd, ucred,
483 LIST_REMOVE(map, sm_link);
484 shm_drop(map->sm_shmfd);
485 free(map->sm_path, M_SHMFD);
496 sys_shm_open(struct thread *td, struct shm_open_args *uap)
498 struct filedesc *fdp;
506 #ifdef CAPABILITY_MODE
508 * shm_open(2) is only allowed for anonymous objects.
510 if (IN_CAPABILITY_MODE(td) && (uap->path != SHM_ANON))
514 if ((uap->flags & O_ACCMODE) != O_RDONLY &&
515 (uap->flags & O_ACCMODE) != O_RDWR)
518 if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0)
521 fdp = td->td_proc->p_fd;
522 cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
524 error = falloc(td, &fp, &fd, 0);
528 /* A SHM_ANON path pointer creates an anonymous object. */
529 if (uap->path == SHM_ANON) {
530 /* A read-only anonymous object is pointless. */
531 if ((uap->flags & O_ACCMODE) == O_RDONLY) {
532 fdclose(fdp, fp, fd, td);
536 shmfd = shm_alloc(td->td_ucred, cmode);
538 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
539 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
541 /* Require paths to start with a '/' character. */
542 if (error == 0 && path[0] != '/')
545 fdclose(fdp, fp, fd, td);
551 fnv = fnv_32_str(path, FNV1_32_INIT);
552 sx_xlock(&shm_dict_lock);
553 shmfd = shm_lookup(path, fnv);
555 /* Object does not yet exist, create it if requested. */
556 if (uap->flags & O_CREAT) {
558 error = mac_posixshm_check_create(td->td_ucred,
562 shmfd = shm_alloc(td->td_ucred, cmode);
563 shm_insert(path, fnv, shmfd);
573 * Object already exists, obtain a new
574 * reference if requested and permitted.
577 if ((uap->flags & (O_CREAT | O_EXCL)) ==
582 error = mac_posixshm_check_open(td->td_ucred,
583 shmfd, FFLAGS(uap->flags & O_ACCMODE));
586 error = shm_access(shmfd, td->td_ucred,
587 FFLAGS(uap->flags & O_ACCMODE));
591 * Truncate the file back to zero length if
592 * O_TRUNC was specified and the object was
593 * opened with read/write.
596 (uap->flags & (O_ACCMODE | O_TRUNC)) ==
597 (O_RDWR | O_TRUNC)) {
599 error = mac_posixshm_check_truncate(
600 td->td_ucred, fp->f_cred, shmfd);
603 shm_dotruncate(shmfd, 0);
608 sx_xunlock(&shm_dict_lock);
611 fdclose(fdp, fp, fd, td);
617 finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
620 if (fdp->fd_ofiles[fd] == fp)
621 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
622 FILEDESC_XUNLOCK(fdp);
623 td->td_retval[0] = fd;
630 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
636 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
637 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
643 fnv = fnv_32_str(path, FNV1_32_INIT);
644 sx_xlock(&shm_dict_lock);
645 error = shm_remove(path, fnv, td->td_ucred);
646 sx_xunlock(&shm_dict_lock);
653 * mmap() helper to validate mmap() requests against shm object state
654 * and give mmap() the vm_object to use for the mapping.
657 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
662 * XXXRW: This validation is probably insufficient, and subject to
663 * sign errors. It should be fixed.
665 if (foff >= shmfd->shm_size ||
666 foff + objsize > round_page(shmfd->shm_size))
669 mtx_lock(&shm_timestamp_lock);
670 vfs_timestamp(&shmfd->shm_atime);
671 mtx_unlock(&shm_timestamp_lock);
672 vm_object_reference(shmfd->shm_object);
673 *obj = shmfd->shm_object;
678 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
686 mtx_lock(&shm_timestamp_lock);
688 * SUSv4 says that x bits of permission need not be affected.
689 * Be consistent with our shm_open there.
692 error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
696 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
697 shmfd->shm_gid, VADMIN, active_cred, NULL);
700 shmfd->shm_mode = mode & ACCESSPERMS;
702 mtx_unlock(&shm_timestamp_lock);
707 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
715 mtx_lock(&shm_timestamp_lock);
717 error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
721 if (uid == (uid_t)-1)
722 uid = shmfd->shm_uid;
723 if (gid == (gid_t)-1)
724 gid = shmfd->shm_gid;
725 if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
726 (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
727 (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
729 shmfd->shm_uid = uid;
730 shmfd->shm_gid = gid;
732 mtx_unlock(&shm_timestamp_lock);