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>
64 #include <sys/rwlock.h>
66 #include <sys/sysctl.h>
67 #include <sys/sysproto.h>
68 #include <sys/systm.h>
71 #include <sys/vnode.h>
72 #include <sys/unistd.h>
74 #include <security/mac/mac_framework.h>
77 #include <vm/vm_param.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_kern.h>
82 #include <vm/vm_object.h>
83 #include <vm/vm_page.h>
84 #include <vm/vm_pageout.h>
85 #include <vm/vm_pager.h>
86 #include <vm/swap_pager.h>
91 struct shmfd *sm_shmfd;
92 LIST_ENTRY(shm_mapping) sm_link;
95 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
96 static LIST_HEAD(, shm_mapping) *shm_dictionary;
97 static struct sx shm_dict_lock;
98 static struct mtx shm_timestamp_lock;
99 static u_long shm_hash;
101 #define SHM_HASH(fnv) (&shm_dictionary[(fnv) & shm_hash])
103 static int shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags);
104 static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode);
105 static void shm_dict_init(void *arg);
106 static void shm_drop(struct shmfd *shmfd);
107 static struct shmfd *shm_hold(struct shmfd *shmfd);
108 static void shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
109 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
110 static int shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
111 static int shm_dotruncate(struct shmfd *shmfd, off_t length);
113 static fo_rdwr_t shm_read;
114 static fo_rdwr_t shm_write;
115 static fo_truncate_t shm_truncate;
116 static fo_ioctl_t shm_ioctl;
117 static fo_poll_t shm_poll;
118 static fo_kqfilter_t shm_kqfilter;
119 static fo_stat_t shm_stat;
120 static fo_close_t shm_close;
121 static fo_chmod_t shm_chmod;
122 static fo_chown_t shm_chown;
123 static fo_seek_t shm_seek;
125 /* File descriptor operations. */
126 static struct fileops shm_ops = {
128 .fo_write = shm_write,
129 .fo_truncate = shm_truncate,
130 .fo_ioctl = shm_ioctl,
132 .fo_kqfilter = shm_kqfilter,
134 .fo_close = shm_close,
135 .fo_chmod = shm_chmod,
136 .fo_chown = shm_chown,
137 .fo_sendfile = vn_sendfile,
139 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
142 FEATURE(posix_shm, "POSIX shared memory");
145 uiomove_object_page(vm_object_t obj, size_t len, struct uio *uio)
150 int error, offset, rv;
152 idx = OFF_TO_IDX(uio->uio_offset);
153 offset = uio->uio_offset & PAGE_MASK;
154 tlen = MIN(PAGE_SIZE - offset, len);
156 VM_OBJECT_WLOCK(obj);
159 * Parallel reads of the page content from disk are prevented
162 * Although the tmpfs vnode lock is held here, it is
163 * nonetheless safe to sleep waiting for a free page. The
164 * pageout daemon does not need to acquire the tmpfs vnode
165 * lock to page out tobj's pages because tobj is a OBJT_SWAP
168 m = vm_page_grab(obj, idx, VM_ALLOC_NORMAL);
169 if (m->valid != VM_PAGE_BITS_ALL) {
170 if (vm_pager_has_page(obj, idx, NULL, NULL)) {
171 rv = vm_pager_get_pages(obj, &m, 1, 0);
172 m = vm_page_lookup(obj, idx);
175 "uiomove_object: vm_obj %p idx %jd null lookup rv %d\n",
177 VM_OBJECT_WUNLOCK(obj);
180 if (rv != VM_PAGER_OK) {
182 "uiomove_object: vm_obj %p idx %jd valid %x pager error %d\n",
183 obj, idx, m->valid, rv);
187 VM_OBJECT_WUNLOCK(obj);
191 vm_page_zero_invalid(m, TRUE);
197 VM_OBJECT_WUNLOCK(obj);
198 error = uiomove_fromphys(&m, offset, tlen, uio);
199 if (uio->uio_rw == UIO_WRITE && error == 0) {
200 VM_OBJECT_WLOCK(obj);
202 VM_OBJECT_WUNLOCK(obj);
206 if (m->queue == PQ_NONE) {
207 vm_page_deactivate(m);
209 /* Requeue to maintain LRU ordering. */
218 uiomove_object(vm_object_t obj, off_t obj_size, struct uio *uio)
225 while ((resid = uio->uio_resid) > 0) {
226 if (obj_size <= uio->uio_offset)
228 len = MIN(obj_size - uio->uio_offset, resid);
231 error = uiomove_object_page(obj, len, uio);
232 if (error != 0 || resid == uio->uio_resid)
239 shm_seek(struct file *fp, off_t offset, int whence, struct thread *td)
246 foffset = foffset_lock(fp, 0);
251 (offset > 0 && foffset > OFF_MAX - offset)) {
258 if (offset > 0 && shmfd->shm_size > OFF_MAX - offset) {
262 offset += shmfd->shm_size;
270 if (offset < 0 || offset > shmfd->shm_size)
273 *(off_t *)(td->td_retval) = offset;
275 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
280 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
281 int flags, struct thread *td)
288 foffset_lock_uio(fp, uio, flags);
289 rl_cookie = rangelock_rlock(&shmfd->shm_rl, uio->uio_offset,
290 uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx);
292 error = mac_posixshm_check_read(active_cred, fp->f_cred, shmfd);
296 error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio);
297 rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx);
298 foffset_unlock_uio(fp, uio, flags);
303 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
304 int flags, struct thread *td)
312 error = mac_posixshm_check_write(active_cred, fp->f_cred, shmfd);
316 foffset_lock_uio(fp, uio, flags);
317 if ((flags & FOF_OFFSET) == 0) {
318 rl_cookie = rangelock_wlock(&shmfd->shm_rl, 0, OFF_MAX,
321 rl_cookie = rangelock_wlock(&shmfd->shm_rl, uio->uio_offset,
322 uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx);
325 error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio);
326 rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx);
327 foffset_unlock_uio(fp, uio, flags);
332 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
342 error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
346 return (shm_dotruncate(shmfd, length));
350 shm_ioctl(struct file *fp, u_long com, void *data,
351 struct ucred *active_cred, struct thread *td)
358 shm_poll(struct file *fp, int events, struct ucred *active_cred,
366 shm_kqfilter(struct file *fp, struct knote *kn)
373 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
384 error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
390 * Attempt to return sanish values for fstat() on a memory file
393 bzero(sb, sizeof(*sb));
394 sb->st_blksize = PAGE_SIZE;
395 sb->st_size = shmfd->shm_size;
396 sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize;
397 mtx_lock(&shm_timestamp_lock);
398 sb->st_atim = shmfd->shm_atime;
399 sb->st_ctim = shmfd->shm_ctime;
400 sb->st_mtim = shmfd->shm_mtime;
401 sb->st_birthtim = shmfd->shm_birthtime;
402 sb->st_mode = S_IFREG | shmfd->shm_mode; /* XXX */
403 sb->st_uid = shmfd->shm_uid;
404 sb->st_gid = shmfd->shm_gid;
405 mtx_unlock(&shm_timestamp_lock);
411 shm_close(struct file *fp, struct thread *td)
423 shm_dotruncate(struct shmfd *shmfd, off_t length)
427 vm_pindex_t idx, nobjsize;
431 object = shmfd->shm_object;
432 VM_OBJECT_WLOCK(object);
433 if (length == shmfd->shm_size) {
434 VM_OBJECT_WUNLOCK(object);
437 nobjsize = OFF_TO_IDX(length + PAGE_MASK);
439 /* Are we shrinking? If so, trim the end. */
440 if (length < shmfd->shm_size) {
442 * Disallow any requests to shrink the size if this
443 * object is mapped into the kernel.
445 if (shmfd->shm_kmappings > 0) {
446 VM_OBJECT_WUNLOCK(object);
451 * Zero the truncated part of the last page.
453 base = length & PAGE_MASK;
455 idx = OFF_TO_IDX(length);
457 m = vm_page_lookup(object, idx);
459 if (vm_page_sleep_if_busy(m, "shmtrc"))
461 } else if (vm_pager_has_page(object, idx, NULL, NULL)) {
462 m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL);
464 VM_OBJECT_WUNLOCK(object);
466 VM_OBJECT_WLOCK(object);
468 } else if (m->valid != VM_PAGE_BITS_ALL) {
470 rv = vm_pager_get_pages(object, ma, 1,
472 m = vm_page_lookup(object, idx);
474 /* A cached page was reactivated. */
477 if (rv == VM_PAGER_OK) {
478 vm_page_deactivate(m);
484 VM_OBJECT_WUNLOCK(object);
489 pmap_zero_page_area(m, base, PAGE_SIZE - base);
490 KASSERT(m->valid == VM_PAGE_BITS_ALL,
491 ("shm_dotruncate: page %p is invalid", m));
493 vm_pager_page_unswapped(m);
496 delta = ptoa(object->size - nobjsize);
498 /* Toss in memory pages. */
499 if (nobjsize < object->size)
500 vm_object_page_remove(object, nobjsize, object->size,
503 /* Toss pages from swap. */
504 if (object->type == OBJT_SWAP)
505 swap_pager_freespace(object, nobjsize, delta);
507 /* Free the swap accounted for shm */
508 swap_release_by_cred(delta, object->cred);
509 object->charge -= delta;
511 /* Attempt to reserve the swap */
512 delta = ptoa(nobjsize - object->size);
513 if (!swap_reserve_by_cred(delta, object->cred)) {
514 VM_OBJECT_WUNLOCK(object);
517 object->charge += delta;
519 shmfd->shm_size = length;
520 mtx_lock(&shm_timestamp_lock);
521 vfs_timestamp(&shmfd->shm_ctime);
522 shmfd->shm_mtime = shmfd->shm_ctime;
523 mtx_unlock(&shm_timestamp_lock);
524 object->size = nobjsize;
525 VM_OBJECT_WUNLOCK(object);
530 * shmfd object management including creation and reference counting
533 static struct shmfd *
534 shm_alloc(struct ucred *ucred, mode_t mode)
538 shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
540 shmfd->shm_uid = ucred->cr_uid;
541 shmfd->shm_gid = ucred->cr_gid;
542 shmfd->shm_mode = mode;
543 shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
544 shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
545 KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
546 VM_OBJECT_WLOCK(shmfd->shm_object);
547 vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
548 vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
549 VM_OBJECT_WUNLOCK(shmfd->shm_object);
550 vfs_timestamp(&shmfd->shm_birthtime);
551 shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
552 shmfd->shm_birthtime;
553 refcount_init(&shmfd->shm_refs, 1);
554 mtx_init(&shmfd->shm_mtx, "shmrl", NULL, MTX_DEF);
555 rangelock_init(&shmfd->shm_rl);
557 mac_posixshm_init(shmfd);
558 mac_posixshm_create(ucred, shmfd);
564 static struct shmfd *
565 shm_hold(struct shmfd *shmfd)
568 refcount_acquire(&shmfd->shm_refs);
573 shm_drop(struct shmfd *shmfd)
576 if (refcount_release(&shmfd->shm_refs)) {
578 mac_posixshm_destroy(shmfd);
580 rangelock_destroy(&shmfd->shm_rl);
581 mtx_destroy(&shmfd->shm_mtx);
582 vm_object_deallocate(shmfd->shm_object);
583 free(shmfd, M_SHMFD);
588 * Determine if the credentials have sufficient permissions for a
589 * specified combination of FREAD and FWRITE.
592 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
602 mtx_lock(&shm_timestamp_lock);
603 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
604 accmode, ucred, NULL);
605 mtx_unlock(&shm_timestamp_lock);
610 * Dictionary management. We maintain an in-kernel dictionary to map
611 * paths to shmfd objects. We use the FNV hash on the path to store
612 * the mappings in a hash table.
615 shm_dict_init(void *arg)
618 mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
619 sx_init(&shm_dict_lock, "shm dictionary");
620 shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
622 SYSINIT(shm_dict_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_dict_init, NULL);
624 static struct shmfd *
625 shm_lookup(char *path, Fnv32_t fnv)
627 struct shm_mapping *map;
629 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
630 if (map->sm_fnv != fnv)
632 if (strcmp(map->sm_path, path) == 0)
633 return (map->sm_shmfd);
640 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
642 struct shm_mapping *map;
644 map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
647 map->sm_shmfd = shm_hold(shmfd);
648 shmfd->shm_path = path;
649 LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
653 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
655 struct shm_mapping *map;
658 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
659 if (map->sm_fnv != fnv)
661 if (strcmp(map->sm_path, path) == 0) {
663 error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
667 error = shm_access(map->sm_shmfd, ucred,
671 map->sm_shmfd->shm_path = NULL;
672 LIST_REMOVE(map, sm_link);
673 shm_drop(map->sm_shmfd);
674 free(map->sm_path, M_SHMFD);
685 sys_shm_open(struct thread *td, struct shm_open_args *uap)
687 struct filedesc *fdp;
695 #ifdef CAPABILITY_MODE
697 * shm_open(2) is only allowed for anonymous objects.
699 if (IN_CAPABILITY_MODE(td) && (uap->path != SHM_ANON))
703 if ((uap->flags & O_ACCMODE) != O_RDONLY &&
704 (uap->flags & O_ACCMODE) != O_RDWR)
707 if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0)
710 fdp = td->td_proc->p_fd;
711 cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
713 error = falloc(td, &fp, &fd, O_CLOEXEC);
717 /* A SHM_ANON path pointer creates an anonymous object. */
718 if (uap->path == SHM_ANON) {
719 /* A read-only anonymous object is pointless. */
720 if ((uap->flags & O_ACCMODE) == O_RDONLY) {
721 fdclose(fdp, fp, fd, td);
725 shmfd = shm_alloc(td->td_ucred, cmode);
727 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
728 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
730 /* Require paths to start with a '/' character. */
731 if (error == 0 && path[0] != '/')
734 fdclose(fdp, fp, fd, td);
740 fnv = fnv_32_str(path, FNV1_32_INIT);
741 sx_xlock(&shm_dict_lock);
742 shmfd = shm_lookup(path, fnv);
744 /* Object does not yet exist, create it if requested. */
745 if (uap->flags & O_CREAT) {
747 error = mac_posixshm_check_create(td->td_ucred,
751 shmfd = shm_alloc(td->td_ucred, cmode);
752 shm_insert(path, fnv, shmfd);
762 * Object already exists, obtain a new
763 * reference if requested and permitted.
766 if ((uap->flags & (O_CREAT | O_EXCL)) ==
771 error = mac_posixshm_check_open(td->td_ucred,
772 shmfd, FFLAGS(uap->flags & O_ACCMODE));
775 error = shm_access(shmfd, td->td_ucred,
776 FFLAGS(uap->flags & O_ACCMODE));
780 * Truncate the file back to zero length if
781 * O_TRUNC was specified and the object was
782 * opened with read/write.
785 (uap->flags & (O_ACCMODE | O_TRUNC)) ==
786 (O_RDWR | O_TRUNC)) {
788 error = mac_posixshm_check_truncate(
789 td->td_ucred, fp->f_cred, shmfd);
792 shm_dotruncate(shmfd, 0);
797 sx_xunlock(&shm_dict_lock);
800 fdclose(fdp, fp, fd, td);
806 finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
808 td->td_retval[0] = fd;
815 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
821 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
822 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
828 fnv = fnv_32_str(path, FNV1_32_INIT);
829 sx_xlock(&shm_dict_lock);
830 error = shm_remove(path, fnv, td->td_ucred);
831 sx_xunlock(&shm_dict_lock);
838 * mmap() helper to validate mmap() requests against shm object state
839 * and give mmap() the vm_object to use for the mapping.
842 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
847 * XXXRW: This validation is probably insufficient, and subject to
848 * sign errors. It should be fixed.
850 if (foff >= shmfd->shm_size ||
851 foff + objsize > round_page(shmfd->shm_size))
854 mtx_lock(&shm_timestamp_lock);
855 vfs_timestamp(&shmfd->shm_atime);
856 mtx_unlock(&shm_timestamp_lock);
857 vm_object_reference(shmfd->shm_object);
858 *obj = shmfd->shm_object;
863 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
871 mtx_lock(&shm_timestamp_lock);
873 * SUSv4 says that x bits of permission need not be affected.
874 * Be consistent with our shm_open there.
877 error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
881 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
882 shmfd->shm_gid, VADMIN, active_cred, NULL);
885 shmfd->shm_mode = mode & ACCESSPERMS;
887 mtx_unlock(&shm_timestamp_lock);
892 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
900 mtx_lock(&shm_timestamp_lock);
902 error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
906 if (uid == (uid_t)-1)
907 uid = shmfd->shm_uid;
908 if (gid == (gid_t)-1)
909 gid = shmfd->shm_gid;
910 if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
911 (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
912 (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
914 shmfd->shm_uid = uid;
915 shmfd->shm_gid = gid;
917 mtx_unlock(&shm_timestamp_lock);
922 * Helper routines to allow the backing object of a shared memory file
923 * descriptor to be mapped in the kernel.
926 shm_map(struct file *fp, size_t size, off_t offset, void **memp)
929 vm_offset_t kva, ofs;
933 if (fp->f_type != DTYPE_SHM)
936 obj = shmfd->shm_object;
937 VM_OBJECT_WLOCK(obj);
939 * XXXRW: This validation is probably insufficient, and subject to
940 * sign errors. It should be fixed.
942 if (offset >= shmfd->shm_size ||
943 offset + size > round_page(shmfd->shm_size)) {
944 VM_OBJECT_WUNLOCK(obj);
948 shmfd->shm_kmappings++;
949 vm_object_reference_locked(obj);
950 VM_OBJECT_WUNLOCK(obj);
952 /* Map the object into the kernel_map and wire it. */
953 kva = vm_map_min(kernel_map);
954 ofs = offset & PAGE_MASK;
955 offset = trunc_page(offset);
956 size = round_page(size + ofs);
957 rv = vm_map_find(kernel_map, obj, offset, &kva, size, 0,
958 VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE,
959 VM_PROT_READ | VM_PROT_WRITE, 0);
960 if (rv == KERN_SUCCESS) {
961 rv = vm_map_wire(kernel_map, kva, kva + size,
962 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
963 if (rv == KERN_SUCCESS) {
964 *memp = (void *)(kva + ofs);
967 vm_map_remove(kernel_map, kva, kva + size);
969 vm_object_deallocate(obj);
971 /* On failure, drop our mapping reference. */
972 VM_OBJECT_WLOCK(obj);
973 shmfd->shm_kmappings--;
974 VM_OBJECT_WUNLOCK(obj);
976 return (vm_mmap_to_errno(rv));
980 * We require the caller to unmap the entire entry. This allows us to
981 * safely decrement shm_kmappings when a mapping is removed.
984 shm_unmap(struct file *fp, void *mem, size_t size)
987 vm_map_entry_t entry;
988 vm_offset_t kva, ofs;
996 if (fp->f_type != DTYPE_SHM)
999 kva = (vm_offset_t)mem;
1000 ofs = kva & PAGE_MASK;
1001 kva = trunc_page(kva);
1002 size = round_page(size + ofs);
1004 rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
1005 &obj, &pindex, &prot, &wired);
1006 if (rv != KERN_SUCCESS)
1008 if (entry->start != kva || entry->end != kva + size) {
1009 vm_map_lookup_done(map, entry);
1012 vm_map_lookup_done(map, entry);
1013 if (obj != shmfd->shm_object)
1015 vm_map_remove(map, kva, kva + size);
1016 VM_OBJECT_WLOCK(obj);
1017 KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
1018 shmfd->shm_kmappings--;
1019 VM_OBJECT_WUNLOCK(obj);
1024 shm_path(struct shmfd *shmfd, char *path, size_t size)
1027 if (shmfd->shm_path == NULL)
1029 sx_slock(&shm_dict_lock);
1030 if (shmfd->shm_path != NULL)
1031 strlcpy(path, shmfd->shm_path, size);
1032 sx_sunlock(&shm_dict_lock);