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_extern.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_kern.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_page.h>
82 #include <vm/vm_pageout.h>
83 #include <vm/vm_pager.h>
84 #include <vm/swap_pager.h>
89 struct shmfd *sm_shmfd;
90 LIST_ENTRY(shm_mapping) sm_link;
93 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
94 static LIST_HEAD(, shm_mapping) *shm_dictionary;
95 static struct sx shm_dict_lock;
96 static struct mtx shm_timestamp_lock;
97 static u_long shm_hash;
99 #define SHM_HASH(fnv) (&shm_dictionary[(fnv) & shm_hash])
101 static int shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags);
102 static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode);
103 static void shm_dict_init(void *arg);
104 static void shm_drop(struct shmfd *shmfd);
105 static struct shmfd *shm_hold(struct shmfd *shmfd);
106 static void shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
107 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
108 static int shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
109 static int shm_dotruncate(struct shmfd *shmfd, off_t length);
111 static fo_rdwr_t shm_read;
112 static fo_rdwr_t shm_write;
113 static fo_truncate_t shm_truncate;
114 static fo_ioctl_t shm_ioctl;
115 static fo_poll_t shm_poll;
116 static fo_kqfilter_t shm_kqfilter;
117 static fo_stat_t shm_stat;
118 static fo_close_t shm_close;
119 static fo_chmod_t shm_chmod;
120 static fo_chown_t shm_chown;
122 /* File descriptor operations. */
123 static struct fileops shm_ops = {
125 .fo_write = shm_write,
126 .fo_truncate = shm_truncate,
127 .fo_ioctl = shm_ioctl,
129 .fo_kqfilter = shm_kqfilter,
131 .fo_close = shm_close,
132 .fo_chmod = shm_chmod,
133 .fo_chown = shm_chown,
134 .fo_flags = DFLAG_PASSABLE
137 FEATURE(posix_shm, "POSIX shared memory");
140 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
141 int flags, struct thread *td)
148 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
149 int flags, struct thread *td)
156 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
166 error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
170 return (shm_dotruncate(shmfd, length));
174 shm_ioctl(struct file *fp, u_long com, void *data,
175 struct ucred *active_cred, struct thread *td)
182 shm_poll(struct file *fp, int events, struct ucred *active_cred,
190 shm_kqfilter(struct file *fp, struct knote *kn)
197 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
208 error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
214 * Attempt to return sanish values for fstat() on a memory file
217 bzero(sb, sizeof(*sb));
218 sb->st_blksize = PAGE_SIZE;
219 sb->st_size = shmfd->shm_size;
220 sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize;
221 mtx_lock(&shm_timestamp_lock);
222 sb->st_atim = shmfd->shm_atime;
223 sb->st_ctim = shmfd->shm_ctime;
224 sb->st_mtim = shmfd->shm_mtime;
225 sb->st_birthtim = shmfd->shm_birthtime;
226 sb->st_mode = S_IFREG | shmfd->shm_mode; /* XXX */
227 sb->st_uid = shmfd->shm_uid;
228 sb->st_gid = shmfd->shm_gid;
229 mtx_unlock(&shm_timestamp_lock);
235 shm_close(struct file *fp, struct thread *td)
247 shm_dotruncate(struct shmfd *shmfd, off_t length)
251 vm_pindex_t idx, nobjsize;
255 object = shmfd->shm_object;
256 VM_OBJECT_LOCK(object);
257 if (length == shmfd->shm_size) {
258 VM_OBJECT_UNLOCK(object);
261 nobjsize = OFF_TO_IDX(length + PAGE_MASK);
263 /* Are we shrinking? If so, trim the end. */
264 if (length < shmfd->shm_size) {
266 * Disallow any requests to shrink the size if this
267 * object is mapped into the kernel.
269 if (shmfd->shm_kmappings > 0) {
270 VM_OBJECT_UNLOCK(object);
275 * Zero the truncated part of the last page.
277 base = length & PAGE_MASK;
279 idx = OFF_TO_IDX(length);
281 m = vm_page_lookup(object, idx);
283 if ((m->oflags & VPO_BUSY) != 0 ||
285 vm_page_sleep(m, "shmtrc");
288 } else if (vm_pager_has_page(object, idx, NULL, NULL)) {
289 m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL);
291 VM_OBJECT_UNLOCK(object);
293 VM_OBJECT_LOCK(object);
295 } else if (m->valid != VM_PAGE_BITS_ALL) {
297 rv = vm_pager_get_pages(object, ma, 1,
299 m = vm_page_lookup(object, idx);
301 /* A cached page was reactivated. */
304 if (rv == VM_PAGER_OK) {
305 vm_page_deactivate(m);
311 VM_OBJECT_UNLOCK(object);
316 pmap_zero_page_area(m, base, PAGE_SIZE - base);
317 KASSERT(m->valid == VM_PAGE_BITS_ALL,
318 ("shm_dotruncate: page %p is invalid", m));
320 vm_pager_page_unswapped(m);
323 delta = ptoa(object->size - nobjsize);
325 /* Toss in memory pages. */
326 if (nobjsize < object->size)
327 vm_object_page_remove(object, nobjsize, object->size,
330 /* Toss pages from swap. */
331 if (object->type == OBJT_SWAP)
332 swap_pager_freespace(object, nobjsize, delta);
334 /* Free the swap accounted for shm */
335 swap_release_by_cred(delta, object->cred);
336 object->charge -= delta;
338 /* Attempt to reserve the swap */
339 delta = ptoa(nobjsize - object->size);
340 if (!swap_reserve_by_cred(delta, object->cred)) {
341 VM_OBJECT_UNLOCK(object);
344 object->charge += delta;
346 shmfd->shm_size = length;
347 mtx_lock(&shm_timestamp_lock);
348 vfs_timestamp(&shmfd->shm_ctime);
349 shmfd->shm_mtime = shmfd->shm_ctime;
350 mtx_unlock(&shm_timestamp_lock);
351 object->size = nobjsize;
352 VM_OBJECT_UNLOCK(object);
357 * shmfd object management including creation and reference counting
360 static struct shmfd *
361 shm_alloc(struct ucred *ucred, mode_t mode)
365 shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
367 shmfd->shm_uid = ucred->cr_uid;
368 shmfd->shm_gid = ucred->cr_gid;
369 shmfd->shm_mode = mode;
370 shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
371 shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
372 KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
373 VM_OBJECT_LOCK(shmfd->shm_object);
374 vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
375 vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
376 VM_OBJECT_UNLOCK(shmfd->shm_object);
377 vfs_timestamp(&shmfd->shm_birthtime);
378 shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
379 shmfd->shm_birthtime;
380 refcount_init(&shmfd->shm_refs, 1);
382 mac_posixshm_init(shmfd);
383 mac_posixshm_create(ucred, shmfd);
389 static struct shmfd *
390 shm_hold(struct shmfd *shmfd)
393 refcount_acquire(&shmfd->shm_refs);
398 shm_drop(struct shmfd *shmfd)
401 if (refcount_release(&shmfd->shm_refs)) {
403 mac_posixshm_destroy(shmfd);
405 vm_object_deallocate(shmfd->shm_object);
406 free(shmfd, M_SHMFD);
411 * Determine if the credentials have sufficient permissions for a
412 * specified combination of FREAD and FWRITE.
415 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
425 mtx_lock(&shm_timestamp_lock);
426 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
427 accmode, ucred, NULL);
428 mtx_unlock(&shm_timestamp_lock);
433 * Dictionary management. We maintain an in-kernel dictionary to map
434 * paths to shmfd objects. We use the FNV hash on the path to store
435 * the mappings in a hash table.
438 shm_dict_init(void *arg)
441 mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
442 sx_init(&shm_dict_lock, "shm dictionary");
443 shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
445 SYSINIT(shm_dict_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_dict_init, NULL);
447 static struct shmfd *
448 shm_lookup(char *path, Fnv32_t fnv)
450 struct shm_mapping *map;
452 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
453 if (map->sm_fnv != fnv)
455 if (strcmp(map->sm_path, path) == 0)
456 return (map->sm_shmfd);
463 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
465 struct shm_mapping *map;
467 map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
470 map->sm_shmfd = shm_hold(shmfd);
471 shmfd->shm_path = path;
472 LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
476 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
478 struct shm_mapping *map;
481 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
482 if (map->sm_fnv != fnv)
484 if (strcmp(map->sm_path, path) == 0) {
486 error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
490 error = shm_access(map->sm_shmfd, ucred,
494 map->sm_shmfd->shm_path = NULL;
495 LIST_REMOVE(map, sm_link);
496 shm_drop(map->sm_shmfd);
497 free(map->sm_path, M_SHMFD);
508 sys_shm_open(struct thread *td, struct shm_open_args *uap)
510 struct filedesc *fdp;
518 #ifdef CAPABILITY_MODE
520 * shm_open(2) is only allowed for anonymous objects.
522 if (IN_CAPABILITY_MODE(td) && (uap->path != SHM_ANON))
526 if ((uap->flags & O_ACCMODE) != O_RDONLY &&
527 (uap->flags & O_ACCMODE) != O_RDWR)
530 if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0)
533 fdp = td->td_proc->p_fd;
534 cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
536 error = falloc(td, &fp, &fd, O_CLOEXEC);
540 /* A SHM_ANON path pointer creates an anonymous object. */
541 if (uap->path == SHM_ANON) {
542 /* A read-only anonymous object is pointless. */
543 if ((uap->flags & O_ACCMODE) == O_RDONLY) {
544 fdclose(fdp, fp, fd, td);
548 shmfd = shm_alloc(td->td_ucred, cmode);
550 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
551 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
553 /* Require paths to start with a '/' character. */
554 if (error == 0 && path[0] != '/')
557 fdclose(fdp, fp, fd, td);
563 fnv = fnv_32_str(path, FNV1_32_INIT);
564 sx_xlock(&shm_dict_lock);
565 shmfd = shm_lookup(path, fnv);
567 /* Object does not yet exist, create it if requested. */
568 if (uap->flags & O_CREAT) {
570 error = mac_posixshm_check_create(td->td_ucred,
574 shmfd = shm_alloc(td->td_ucred, cmode);
575 shm_insert(path, fnv, shmfd);
585 * Object already exists, obtain a new
586 * reference if requested and permitted.
589 if ((uap->flags & (O_CREAT | O_EXCL)) ==
594 error = mac_posixshm_check_open(td->td_ucred,
595 shmfd, FFLAGS(uap->flags & O_ACCMODE));
598 error = shm_access(shmfd, td->td_ucred,
599 FFLAGS(uap->flags & O_ACCMODE));
603 * Truncate the file back to zero length if
604 * O_TRUNC was specified and the object was
605 * opened with read/write.
608 (uap->flags & (O_ACCMODE | O_TRUNC)) ==
609 (O_RDWR | O_TRUNC)) {
611 error = mac_posixshm_check_truncate(
612 td->td_ucred, fp->f_cred, shmfd);
615 shm_dotruncate(shmfd, 0);
620 sx_xunlock(&shm_dict_lock);
623 fdclose(fdp, fp, fd, td);
629 finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
631 td->td_retval[0] = fd;
638 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
644 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
645 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
651 fnv = fnv_32_str(path, FNV1_32_INIT);
652 sx_xlock(&shm_dict_lock);
653 error = shm_remove(path, fnv, td->td_ucred);
654 sx_xunlock(&shm_dict_lock);
661 * mmap() helper to validate mmap() requests against shm object state
662 * and give mmap() the vm_object to use for the mapping.
665 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
670 * XXXRW: This validation is probably insufficient, and subject to
671 * sign errors. It should be fixed.
673 if (foff >= shmfd->shm_size ||
674 foff + objsize > round_page(shmfd->shm_size))
677 mtx_lock(&shm_timestamp_lock);
678 vfs_timestamp(&shmfd->shm_atime);
679 mtx_unlock(&shm_timestamp_lock);
680 vm_object_reference(shmfd->shm_object);
681 *obj = shmfd->shm_object;
686 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
694 mtx_lock(&shm_timestamp_lock);
696 * SUSv4 says that x bits of permission need not be affected.
697 * Be consistent with our shm_open there.
700 error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
704 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
705 shmfd->shm_gid, VADMIN, active_cred, NULL);
708 shmfd->shm_mode = mode & ACCESSPERMS;
710 mtx_unlock(&shm_timestamp_lock);
715 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
723 mtx_lock(&shm_timestamp_lock);
725 error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
729 if (uid == (uid_t)-1)
730 uid = shmfd->shm_uid;
731 if (gid == (gid_t)-1)
732 gid = shmfd->shm_gid;
733 if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
734 (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
735 (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
737 shmfd->shm_uid = uid;
738 shmfd->shm_gid = gid;
740 mtx_unlock(&shm_timestamp_lock);
745 * Helper routines to allow the backing object of a shared memory file
746 * descriptor to be mapped in the kernel.
749 shm_map(struct file *fp, size_t size, off_t offset, void **memp)
752 vm_offset_t kva, ofs;
756 if (fp->f_type != DTYPE_SHM)
759 obj = shmfd->shm_object;
762 * XXXRW: This validation is probably insufficient, and subject to
763 * sign errors. It should be fixed.
765 if (offset >= shmfd->shm_size ||
766 offset + size > round_page(shmfd->shm_size)) {
767 VM_OBJECT_UNLOCK(obj);
771 shmfd->shm_kmappings++;
772 vm_object_reference_locked(obj);
773 VM_OBJECT_UNLOCK(obj);
775 /* Map the object into the kernel_map and wire it. */
776 kva = vm_map_min(kernel_map);
777 ofs = offset & PAGE_MASK;
778 offset = trunc_page(offset);
779 size = round_page(size + ofs);
780 rv = vm_map_find(kernel_map, obj, offset, &kva, size,
781 VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE,
782 VM_PROT_READ | VM_PROT_WRITE, 0);
783 if (rv == KERN_SUCCESS) {
784 rv = vm_map_wire(kernel_map, kva, kva + size,
785 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
786 if (rv == KERN_SUCCESS) {
787 *memp = (void *)(kva + ofs);
790 vm_map_remove(kernel_map, kva, kva + size);
792 vm_object_deallocate(obj);
794 /* On failure, drop our mapping reference. */
796 shmfd->shm_kmappings--;
797 VM_OBJECT_UNLOCK(obj);
799 return (vm_mmap_to_errno(rv));
803 * We require the caller to unmap the entire entry. This allows us to
804 * safely decrement shm_kmappings when a mapping is removed.
807 shm_unmap(struct file *fp, void *mem, size_t size)
810 vm_map_entry_t entry;
811 vm_offset_t kva, ofs;
819 if (fp->f_type != DTYPE_SHM)
822 kva = (vm_offset_t)mem;
823 ofs = kva & PAGE_MASK;
824 kva = trunc_page(kva);
825 size = round_page(size + ofs);
827 rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
828 &obj, &pindex, &prot, &wired);
829 if (rv != KERN_SUCCESS)
831 if (entry->start != kva || entry->end != kva + size) {
832 vm_map_lookup_done(map, entry);
835 vm_map_lookup_done(map, entry);
836 if (obj != shmfd->shm_object)
838 vm_map_remove(map, kva, kva + size);
840 KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
841 shmfd->shm_kmappings--;
842 VM_OBJECT_UNLOCK(obj);
847 shm_path(struct shmfd *shmfd, char *path, size_t size)
850 if (shmfd->shm_path == NULL)
852 sx_slock(&shm_dict_lock);
853 if (shmfd->shm_path != NULL)
854 strlcpy(path, shmfd->shm_path, size);
855 sx_sunlock(&shm_dict_lock);