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>
52 #include <sys/fcntl.h>
54 #include <sys/filedesc.h>
55 #include <sys/fnv_hash.h>
56 #include <sys/kernel.h>
58 #include <sys/malloc.h>
60 #include <sys/mutex.h>
63 #include <sys/refcount.h>
64 #include <sys/resourcevar.h>
66 #include <sys/sysctl.h>
67 #include <sys/sysproto.h>
68 #include <sys/systm.h>
71 #include <sys/vnode.h>
73 #include <security/mac/mac_framework.h>
76 #include <vm/vm_param.h>
78 #include <vm/vm_extern.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_kern.h>
81 #include <vm/vm_object.h>
82 #include <vm/vm_page.h>
83 #include <vm/vm_pageout.h>
84 #include <vm/vm_pager.h>
85 #include <vm/swap_pager.h>
90 struct shmfd *sm_shmfd;
91 LIST_ENTRY(shm_mapping) sm_link;
94 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
95 static LIST_HEAD(, shm_mapping) *shm_dictionary;
96 static struct sx shm_dict_lock;
97 static struct mtx shm_timestamp_lock;
98 static u_long shm_hash;
99 static struct unrhdr *shm_ino_unr;
100 static dev_t shm_dev_ino;
102 #define SHM_HASH(fnv) (&shm_dictionary[(fnv) & shm_hash])
104 static int shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags);
105 static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode);
106 static void shm_init(void *arg);
107 static void shm_drop(struct shmfd *shmfd);
108 static struct shmfd *shm_hold(struct shmfd *shmfd);
109 static void shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
110 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
111 static int shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
112 static int shm_dotruncate(struct shmfd *shmfd, off_t length);
114 static fo_rdwr_t shm_read;
115 static fo_rdwr_t shm_write;
116 static fo_truncate_t shm_truncate;
117 static fo_ioctl_t shm_ioctl;
118 static fo_poll_t shm_poll;
119 static fo_kqfilter_t shm_kqfilter;
120 static fo_stat_t shm_stat;
121 static fo_close_t shm_close;
122 static fo_chmod_t shm_chmod;
123 static fo_chown_t shm_chown;
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_flags = DFLAG_PASSABLE
140 FEATURE(posix_shm, "POSIX shared memory");
143 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
144 int flags, struct thread *td)
151 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
152 int flags, struct thread *td)
159 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
169 error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
173 return (shm_dotruncate(shmfd, length));
177 shm_ioctl(struct file *fp, u_long com, void *data,
178 struct ucred *active_cred, struct thread *td)
185 shm_poll(struct file *fp, int events, struct ucred *active_cred,
193 shm_kqfilter(struct file *fp, struct knote *kn)
200 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
211 error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
217 * Attempt to return sanish values for fstat() on a memory file
220 bzero(sb, sizeof(*sb));
221 sb->st_blksize = PAGE_SIZE;
222 sb->st_size = shmfd->shm_size;
223 sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize;
224 mtx_lock(&shm_timestamp_lock);
225 sb->st_atim = shmfd->shm_atime;
226 sb->st_ctim = shmfd->shm_ctime;
227 sb->st_mtim = shmfd->shm_mtime;
228 sb->st_birthtim = shmfd->shm_birthtime;
229 sb->st_mode = S_IFREG | shmfd->shm_mode; /* XXX */
230 sb->st_uid = shmfd->shm_uid;
231 sb->st_gid = shmfd->shm_gid;
232 mtx_unlock(&shm_timestamp_lock);
233 sb->st_dev = shm_dev_ino;
234 sb->st_ino = shmfd->shm_ino;
240 shm_close(struct file *fp, struct thread *td)
252 shm_dotruncate(struct shmfd *shmfd, off_t length)
256 vm_pindex_t idx, nobjsize;
260 object = shmfd->shm_object;
261 VM_OBJECT_LOCK(object);
262 if (length == shmfd->shm_size) {
263 VM_OBJECT_UNLOCK(object);
266 nobjsize = OFF_TO_IDX(length + PAGE_MASK);
268 /* Are we shrinking? If so, trim the end. */
269 if (length < shmfd->shm_size) {
271 * Disallow any requests to shrink the size if this
272 * object is mapped into the kernel.
274 if (shmfd->shm_kmappings > 0) {
275 VM_OBJECT_UNLOCK(object);
280 * Zero the truncated part of the last page.
282 base = length & PAGE_MASK;
284 idx = OFF_TO_IDX(length);
286 m = vm_page_lookup(object, idx);
288 if ((m->oflags & VPO_BUSY) != 0 ||
290 vm_page_sleep(m, "shmtrc");
293 } else if (vm_pager_has_page(object, idx, NULL, NULL)) {
294 m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL);
296 VM_OBJECT_UNLOCK(object);
298 VM_OBJECT_LOCK(object);
300 } else if (m->valid != VM_PAGE_BITS_ALL) {
302 rv = vm_pager_get_pages(object, ma, 1,
304 m = vm_page_lookup(object, idx);
306 /* A cached page was reactivated. */
309 if (rv == VM_PAGER_OK) {
310 vm_page_deactivate(m);
316 VM_OBJECT_UNLOCK(object);
321 pmap_zero_page_area(m, base, PAGE_SIZE - base);
322 KASSERT(m->valid == VM_PAGE_BITS_ALL,
323 ("shm_dotruncate: page %p is invalid", m));
325 vm_pager_page_unswapped(m);
328 delta = ptoa(object->size - nobjsize);
330 /* Toss in memory pages. */
331 if (nobjsize < object->size)
332 vm_object_page_remove(object, nobjsize, object->size,
335 /* Toss pages from swap. */
336 if (object->type == OBJT_SWAP)
337 swap_pager_freespace(object, nobjsize, delta);
339 /* Free the swap accounted for shm */
340 swap_release_by_cred(delta, object->cred);
341 object->charge -= delta;
343 /* Attempt to reserve the swap */
344 delta = ptoa(nobjsize - object->size);
345 if (!swap_reserve_by_cred(delta, object->cred)) {
346 VM_OBJECT_UNLOCK(object);
349 object->charge += delta;
351 shmfd->shm_size = length;
352 mtx_lock(&shm_timestamp_lock);
353 vfs_timestamp(&shmfd->shm_ctime);
354 shmfd->shm_mtime = shmfd->shm_ctime;
355 mtx_unlock(&shm_timestamp_lock);
356 object->size = nobjsize;
357 VM_OBJECT_UNLOCK(object);
362 * shmfd object management including creation and reference counting
365 static struct shmfd *
366 shm_alloc(struct ucred *ucred, mode_t mode)
371 shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
373 shmfd->shm_uid = ucred->cr_uid;
374 shmfd->shm_gid = ucred->cr_gid;
375 shmfd->shm_mode = mode;
376 shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
377 shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
378 KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
379 VM_OBJECT_LOCK(shmfd->shm_object);
380 vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
381 vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
382 VM_OBJECT_UNLOCK(shmfd->shm_object);
383 vfs_timestamp(&shmfd->shm_birthtime);
384 shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
385 shmfd->shm_birthtime;
386 ino = alloc_unr(shm_ino_unr);
390 shmfd->shm_ino = ino;
391 refcount_init(&shmfd->shm_refs, 1);
393 mac_posixshm_init(shmfd);
394 mac_posixshm_create(ucred, shmfd);
400 static struct shmfd *
401 shm_hold(struct shmfd *shmfd)
404 refcount_acquire(&shmfd->shm_refs);
409 shm_drop(struct shmfd *shmfd)
412 if (refcount_release(&shmfd->shm_refs)) {
414 mac_posixshm_destroy(shmfd);
416 vm_object_deallocate(shmfd->shm_object);
417 if (shmfd->shm_ino != 0)
418 free_unr(shm_ino_unr, shmfd->shm_ino);
419 free(shmfd, M_SHMFD);
424 * Determine if the credentials have sufficient permissions for a
425 * specified combination of FREAD and FWRITE.
428 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
438 mtx_lock(&shm_timestamp_lock);
439 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
440 accmode, ucred, NULL);
441 mtx_unlock(&shm_timestamp_lock);
446 * Dictionary management. We maintain an in-kernel dictionary to map
447 * paths to shmfd objects. We use the FNV hash on the path to store
448 * the mappings in a hash table.
454 mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
455 sx_init(&shm_dict_lock, "shm dictionary");
456 shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
457 shm_ino_unr = new_unrhdr(1, INT32_MAX, NULL);
458 KASSERT(shm_ino_unr != NULL, ("shm fake inodes not initialized"));
459 shm_dev_ino = devfs_alloc_cdp_inode();
460 KASSERT(shm_dev_ino > 0, ("shm dev inode not initialized"));
462 SYSINIT(shm_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_init, NULL);
464 static struct shmfd *
465 shm_lookup(char *path, Fnv32_t fnv)
467 struct shm_mapping *map;
469 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
470 if (map->sm_fnv != fnv)
472 if (strcmp(map->sm_path, path) == 0)
473 return (map->sm_shmfd);
480 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
482 struct shm_mapping *map;
484 map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
487 map->sm_shmfd = shm_hold(shmfd);
488 shmfd->shm_path = path;
489 LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
493 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
495 struct shm_mapping *map;
498 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
499 if (map->sm_fnv != fnv)
501 if (strcmp(map->sm_path, path) == 0) {
503 error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
507 error = shm_access(map->sm_shmfd, ucred,
511 map->sm_shmfd->shm_path = NULL;
512 LIST_REMOVE(map, sm_link);
513 shm_drop(map->sm_shmfd);
514 free(map->sm_path, M_SHMFD);
525 sys_shm_open(struct thread *td, struct shm_open_args *uap)
527 struct filedesc *fdp;
535 #ifdef CAPABILITY_MODE
537 * shm_open(2) is only allowed for anonymous objects.
539 if (IN_CAPABILITY_MODE(td) && (uap->path != SHM_ANON))
543 if ((uap->flags & O_ACCMODE) != O_RDONLY &&
544 (uap->flags & O_ACCMODE) != O_RDWR)
547 if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0)
550 fdp = td->td_proc->p_fd;
551 cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
553 error = falloc(td, &fp, &fd, O_CLOEXEC);
557 /* A SHM_ANON path pointer creates an anonymous object. */
558 if (uap->path == SHM_ANON) {
559 /* A read-only anonymous object is pointless. */
560 if ((uap->flags & O_ACCMODE) == O_RDONLY) {
561 fdclose(fdp, fp, fd, td);
565 shmfd = shm_alloc(td->td_ucred, cmode);
567 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
568 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
570 /* Require paths to start with a '/' character. */
571 if (error == 0 && path[0] != '/')
574 fdclose(fdp, fp, fd, td);
580 fnv = fnv_32_str(path, FNV1_32_INIT);
581 sx_xlock(&shm_dict_lock);
582 shmfd = shm_lookup(path, fnv);
584 /* Object does not yet exist, create it if requested. */
585 if (uap->flags & O_CREAT) {
587 error = mac_posixshm_check_create(td->td_ucred,
591 shmfd = shm_alloc(td->td_ucred, cmode);
592 shm_insert(path, fnv, shmfd);
602 * Object already exists, obtain a new
603 * reference if requested and permitted.
606 if ((uap->flags & (O_CREAT | O_EXCL)) ==
611 error = mac_posixshm_check_open(td->td_ucred,
612 shmfd, FFLAGS(uap->flags & O_ACCMODE));
615 error = shm_access(shmfd, td->td_ucred,
616 FFLAGS(uap->flags & O_ACCMODE));
620 * Truncate the file back to zero length if
621 * O_TRUNC was specified and the object was
622 * opened with read/write.
625 (uap->flags & (O_ACCMODE | O_TRUNC)) ==
626 (O_RDWR | O_TRUNC)) {
628 error = mac_posixshm_check_truncate(
629 td->td_ucred, fp->f_cred, shmfd);
632 shm_dotruncate(shmfd, 0);
637 sx_xunlock(&shm_dict_lock);
640 fdclose(fdp, fp, fd, td);
646 finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
648 td->td_retval[0] = fd;
655 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
661 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
662 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
668 fnv = fnv_32_str(path, FNV1_32_INIT);
669 sx_xlock(&shm_dict_lock);
670 error = shm_remove(path, fnv, td->td_ucred);
671 sx_xunlock(&shm_dict_lock);
678 * mmap() helper to validate mmap() requests against shm object state
679 * and give mmap() the vm_object to use for the mapping.
682 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
687 * XXXRW: This validation is probably insufficient, and subject to
688 * sign errors. It should be fixed.
690 if (foff >= shmfd->shm_size ||
691 foff + objsize > round_page(shmfd->shm_size))
694 mtx_lock(&shm_timestamp_lock);
695 vfs_timestamp(&shmfd->shm_atime);
696 mtx_unlock(&shm_timestamp_lock);
697 vm_object_reference(shmfd->shm_object);
698 *obj = shmfd->shm_object;
703 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
711 mtx_lock(&shm_timestamp_lock);
713 * SUSv4 says that x bits of permission need not be affected.
714 * Be consistent with our shm_open there.
717 error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
721 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
722 shmfd->shm_gid, VADMIN, active_cred, NULL);
725 shmfd->shm_mode = mode & ACCESSPERMS;
727 mtx_unlock(&shm_timestamp_lock);
732 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
740 mtx_lock(&shm_timestamp_lock);
742 error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
746 if (uid == (uid_t)-1)
747 uid = shmfd->shm_uid;
748 if (gid == (gid_t)-1)
749 gid = shmfd->shm_gid;
750 if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
751 (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
752 (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
754 shmfd->shm_uid = uid;
755 shmfd->shm_gid = gid;
757 mtx_unlock(&shm_timestamp_lock);
762 * Helper routines to allow the backing object of a shared memory file
763 * descriptor to be mapped in the kernel.
766 shm_map(struct file *fp, size_t size, off_t offset, void **memp)
769 vm_offset_t kva, ofs;
773 if (fp->f_type != DTYPE_SHM)
776 obj = shmfd->shm_object;
779 * XXXRW: This validation is probably insufficient, and subject to
780 * sign errors. It should be fixed.
782 if (offset >= shmfd->shm_size ||
783 offset + size > round_page(shmfd->shm_size)) {
784 VM_OBJECT_UNLOCK(obj);
788 shmfd->shm_kmappings++;
789 vm_object_reference_locked(obj);
790 VM_OBJECT_UNLOCK(obj);
792 /* Map the object into the kernel_map and wire it. */
793 kva = vm_map_min(kernel_map);
794 ofs = offset & PAGE_MASK;
795 offset = trunc_page(offset);
796 size = round_page(size + ofs);
797 rv = vm_map_find(kernel_map, obj, offset, &kva, size,
798 VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE,
799 VM_PROT_READ | VM_PROT_WRITE, 0);
800 if (rv == KERN_SUCCESS) {
801 rv = vm_map_wire(kernel_map, kva, kva + size,
802 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
803 if (rv == KERN_SUCCESS) {
804 *memp = (void *)(kva + ofs);
807 vm_map_remove(kernel_map, kva, kva + size);
809 vm_object_deallocate(obj);
811 /* On failure, drop our mapping reference. */
813 shmfd->shm_kmappings--;
814 VM_OBJECT_UNLOCK(obj);
816 return (vm_mmap_to_errno(rv));
820 * We require the caller to unmap the entire entry. This allows us to
821 * safely decrement shm_kmappings when a mapping is removed.
824 shm_unmap(struct file *fp, void *mem, size_t size)
827 vm_map_entry_t entry;
828 vm_offset_t kva, ofs;
836 if (fp->f_type != DTYPE_SHM)
839 kva = (vm_offset_t)mem;
840 ofs = kva & PAGE_MASK;
841 kva = trunc_page(kva);
842 size = round_page(size + ofs);
844 rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
845 &obj, &pindex, &prot, &wired);
846 if (rv != KERN_SUCCESS)
848 if (entry->start != kva || entry->end != kva + size) {
849 vm_map_lookup_done(map, entry);
852 vm_map_lookup_done(map, entry);
853 if (obj != shmfd->shm_object)
855 vm_map_remove(map, kva, kva + size);
857 KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
858 shmfd->shm_kmappings--;
859 VM_OBJECT_UNLOCK(obj);
864 shm_path(struct shmfd *shmfd, char *path, size_t size)
867 if (shmfd->shm_path == NULL)
869 sx_slock(&shm_dict_lock);
870 if (shmfd->shm_path != NULL)
871 strlcpy(path, shmfd->shm_path, size);
872 sx_sunlock(&shm_dict_lock);