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>
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;
100 #define SHM_HASH(fnv) (&shm_dictionary[(fnv) & shm_hash])
102 static int shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags);
103 static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode);
104 static void shm_dict_init(void *arg);
105 static void shm_drop(struct shmfd *shmfd);
106 static struct shmfd *shm_hold(struct shmfd *shmfd);
107 static void shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
108 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
109 static int shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
110 static int shm_dotruncate(struct shmfd *shmfd, off_t length);
112 static fo_rdwr_t shm_read;
113 static fo_rdwr_t shm_write;
114 static fo_truncate_t shm_truncate;
115 static fo_ioctl_t shm_ioctl;
116 static fo_poll_t shm_poll;
117 static fo_kqfilter_t shm_kqfilter;
118 static fo_stat_t shm_stat;
119 static fo_close_t shm_close;
120 static fo_chmod_t shm_chmod;
121 static fo_chown_t shm_chown;
123 /* File descriptor operations. */
124 static struct fileops shm_ops = {
126 .fo_write = shm_write,
127 .fo_truncate = shm_truncate,
128 .fo_ioctl = shm_ioctl,
130 .fo_kqfilter = shm_kqfilter,
132 .fo_close = shm_close,
133 .fo_chmod = shm_chmod,
134 .fo_chown = shm_chown,
135 .fo_sendfile = invfo_sendfile,
136 .fo_flags = DFLAG_PASSABLE
139 FEATURE(posix_shm, "POSIX shared memory");
142 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
143 int flags, struct thread *td)
150 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
151 int flags, struct thread *td)
158 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
168 error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
172 return (shm_dotruncate(shmfd, length));
176 shm_ioctl(struct file *fp, u_long com, void *data,
177 struct ucred *active_cred, struct thread *td)
184 shm_poll(struct file *fp, int events, struct ucred *active_cred,
192 shm_kqfilter(struct file *fp, struct knote *kn)
199 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
210 error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
216 * Attempt to return sanish values for fstat() on a memory file
219 bzero(sb, sizeof(*sb));
220 sb->st_blksize = PAGE_SIZE;
221 sb->st_size = shmfd->shm_size;
222 sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize;
223 mtx_lock(&shm_timestamp_lock);
224 sb->st_atim = shmfd->shm_atime;
225 sb->st_ctim = shmfd->shm_ctime;
226 sb->st_mtim = shmfd->shm_mtime;
227 sb->st_birthtim = shmfd->shm_birthtime;
228 sb->st_mode = S_IFREG | shmfd->shm_mode; /* XXX */
229 sb->st_uid = shmfd->shm_uid;
230 sb->st_gid = shmfd->shm_gid;
231 mtx_unlock(&shm_timestamp_lock);
237 shm_close(struct file *fp, struct thread *td)
249 shm_dotruncate(struct shmfd *shmfd, off_t length)
253 vm_pindex_t idx, nobjsize;
257 object = shmfd->shm_object;
258 VM_OBJECT_WLOCK(object);
259 if (length == shmfd->shm_size) {
260 VM_OBJECT_WUNLOCK(object);
263 nobjsize = OFF_TO_IDX(length + PAGE_MASK);
265 /* Are we shrinking? If so, trim the end. */
266 if (length < shmfd->shm_size) {
268 * Disallow any requests to shrink the size if this
269 * object is mapped into the kernel.
271 if (shmfd->shm_kmappings > 0) {
272 VM_OBJECT_WUNLOCK(object);
277 * Zero the truncated part of the last page.
279 base = length & PAGE_MASK;
281 idx = OFF_TO_IDX(length);
283 m = vm_page_lookup(object, idx);
285 if (vm_page_sleep_if_busy(m, "shmtrc"))
287 } else if (vm_pager_has_page(object, idx, NULL, NULL)) {
288 m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL);
290 VM_OBJECT_WUNLOCK(object);
292 VM_OBJECT_WLOCK(object);
294 } else if (m->valid != VM_PAGE_BITS_ALL) {
296 rv = vm_pager_get_pages(object, ma, 1,
298 m = vm_page_lookup(object, idx);
300 /* A cached page was reactivated. */
303 if (rv == VM_PAGER_OK) {
304 vm_page_deactivate(m);
310 VM_OBJECT_WUNLOCK(object);
315 pmap_zero_page_area(m, base, PAGE_SIZE - base);
316 KASSERT(m->valid == VM_PAGE_BITS_ALL,
317 ("shm_dotruncate: page %p is invalid", m));
319 vm_pager_page_unswapped(m);
322 delta = ptoa(object->size - nobjsize);
324 /* Toss in memory pages. */
325 if (nobjsize < object->size)
326 vm_object_page_remove(object, nobjsize, object->size,
329 /* Toss pages from swap. */
330 if (object->type == OBJT_SWAP)
331 swap_pager_freespace(object, nobjsize, delta);
333 /* Free the swap accounted for shm */
334 swap_release_by_cred(delta, object->cred);
335 object->charge -= delta;
337 /* Attempt to reserve the swap */
338 delta = ptoa(nobjsize - object->size);
339 if (!swap_reserve_by_cred(delta, object->cred)) {
340 VM_OBJECT_WUNLOCK(object);
343 object->charge += delta;
345 shmfd->shm_size = length;
346 mtx_lock(&shm_timestamp_lock);
347 vfs_timestamp(&shmfd->shm_ctime);
348 shmfd->shm_mtime = shmfd->shm_ctime;
349 mtx_unlock(&shm_timestamp_lock);
350 object->size = nobjsize;
351 VM_OBJECT_WUNLOCK(object);
356 * shmfd object management including creation and reference counting
359 static struct shmfd *
360 shm_alloc(struct ucred *ucred, mode_t mode)
364 shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
366 shmfd->shm_uid = ucred->cr_uid;
367 shmfd->shm_gid = ucred->cr_gid;
368 shmfd->shm_mode = mode;
369 shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
370 shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
371 KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
372 VM_OBJECT_WLOCK(shmfd->shm_object);
373 vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
374 vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
375 VM_OBJECT_WUNLOCK(shmfd->shm_object);
376 vfs_timestamp(&shmfd->shm_birthtime);
377 shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
378 shmfd->shm_birthtime;
379 refcount_init(&shmfd->shm_refs, 1);
381 mac_posixshm_init(shmfd);
382 mac_posixshm_create(ucred, shmfd);
388 static struct shmfd *
389 shm_hold(struct shmfd *shmfd)
392 refcount_acquire(&shmfd->shm_refs);
397 shm_drop(struct shmfd *shmfd)
400 if (refcount_release(&shmfd->shm_refs)) {
402 mac_posixshm_destroy(shmfd);
404 vm_object_deallocate(shmfd->shm_object);
405 free(shmfd, M_SHMFD);
410 * Determine if the credentials have sufficient permissions for a
411 * specified combination of FREAD and FWRITE.
414 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
424 mtx_lock(&shm_timestamp_lock);
425 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
426 accmode, ucred, NULL);
427 mtx_unlock(&shm_timestamp_lock);
432 * Dictionary management. We maintain an in-kernel dictionary to map
433 * paths to shmfd objects. We use the FNV hash on the path to store
434 * the mappings in a hash table.
437 shm_dict_init(void *arg)
440 mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
441 sx_init(&shm_dict_lock, "shm dictionary");
442 shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
444 SYSINIT(shm_dict_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_dict_init, NULL);
446 static struct shmfd *
447 shm_lookup(char *path, Fnv32_t fnv)
449 struct shm_mapping *map;
451 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
452 if (map->sm_fnv != fnv)
454 if (strcmp(map->sm_path, path) == 0)
455 return (map->sm_shmfd);
462 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
464 struct shm_mapping *map;
466 map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
469 map->sm_shmfd = shm_hold(shmfd);
470 shmfd->shm_path = path;
471 LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
475 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
477 struct shm_mapping *map;
480 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
481 if (map->sm_fnv != fnv)
483 if (strcmp(map->sm_path, path) == 0) {
485 error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
489 error = shm_access(map->sm_shmfd, ucred,
493 map->sm_shmfd->shm_path = NULL;
494 LIST_REMOVE(map, sm_link);
495 shm_drop(map->sm_shmfd);
496 free(map->sm_path, M_SHMFD);
507 sys_shm_open(struct thread *td, struct shm_open_args *uap)
509 struct filedesc *fdp;
517 #ifdef CAPABILITY_MODE
519 * shm_open(2) is only allowed for anonymous objects.
521 if (IN_CAPABILITY_MODE(td) && (uap->path != SHM_ANON))
525 if ((uap->flags & O_ACCMODE) != O_RDONLY &&
526 (uap->flags & O_ACCMODE) != O_RDWR)
529 if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0)
532 fdp = td->td_proc->p_fd;
533 cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
535 error = falloc(td, &fp, &fd, O_CLOEXEC);
539 /* A SHM_ANON path pointer creates an anonymous object. */
540 if (uap->path == SHM_ANON) {
541 /* A read-only anonymous object is pointless. */
542 if ((uap->flags & O_ACCMODE) == O_RDONLY) {
543 fdclose(fdp, fp, fd, td);
547 shmfd = shm_alloc(td->td_ucred, cmode);
549 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
550 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
552 /* Require paths to start with a '/' character. */
553 if (error == 0 && path[0] != '/')
556 fdclose(fdp, fp, fd, td);
562 fnv = fnv_32_str(path, FNV1_32_INIT);
563 sx_xlock(&shm_dict_lock);
564 shmfd = shm_lookup(path, fnv);
566 /* Object does not yet exist, create it if requested. */
567 if (uap->flags & O_CREAT) {
569 error = mac_posixshm_check_create(td->td_ucred,
573 shmfd = shm_alloc(td->td_ucred, cmode);
574 shm_insert(path, fnv, shmfd);
584 * Object already exists, obtain a new
585 * reference if requested and permitted.
588 if ((uap->flags & (O_CREAT | O_EXCL)) ==
593 error = mac_posixshm_check_open(td->td_ucred,
594 shmfd, FFLAGS(uap->flags & O_ACCMODE));
597 error = shm_access(shmfd, td->td_ucred,
598 FFLAGS(uap->flags & O_ACCMODE));
602 * Truncate the file back to zero length if
603 * O_TRUNC was specified and the object was
604 * opened with read/write.
607 (uap->flags & (O_ACCMODE | O_TRUNC)) ==
608 (O_RDWR | O_TRUNC)) {
610 error = mac_posixshm_check_truncate(
611 td->td_ucred, fp->f_cred, shmfd);
614 shm_dotruncate(shmfd, 0);
619 sx_xunlock(&shm_dict_lock);
622 fdclose(fdp, fp, fd, td);
628 finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
630 td->td_retval[0] = fd;
637 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
643 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
644 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
650 fnv = fnv_32_str(path, FNV1_32_INIT);
651 sx_xlock(&shm_dict_lock);
652 error = shm_remove(path, fnv, td->td_ucred);
653 sx_xunlock(&shm_dict_lock);
660 * mmap() helper to validate mmap() requests against shm object state
661 * and give mmap() the vm_object to use for the mapping.
664 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
669 * XXXRW: This validation is probably insufficient, and subject to
670 * sign errors. It should be fixed.
672 if (foff >= shmfd->shm_size ||
673 foff + objsize > round_page(shmfd->shm_size))
676 mtx_lock(&shm_timestamp_lock);
677 vfs_timestamp(&shmfd->shm_atime);
678 mtx_unlock(&shm_timestamp_lock);
679 vm_object_reference(shmfd->shm_object);
680 *obj = shmfd->shm_object;
685 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
693 mtx_lock(&shm_timestamp_lock);
695 * SUSv4 says that x bits of permission need not be affected.
696 * Be consistent with our shm_open there.
699 error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
703 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
704 shmfd->shm_gid, VADMIN, active_cred, NULL);
707 shmfd->shm_mode = mode & ACCESSPERMS;
709 mtx_unlock(&shm_timestamp_lock);
714 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
722 mtx_lock(&shm_timestamp_lock);
724 error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
728 if (uid == (uid_t)-1)
729 uid = shmfd->shm_uid;
730 if (gid == (gid_t)-1)
731 gid = shmfd->shm_gid;
732 if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
733 (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
734 (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
736 shmfd->shm_uid = uid;
737 shmfd->shm_gid = gid;
739 mtx_unlock(&shm_timestamp_lock);
744 * Helper routines to allow the backing object of a shared memory file
745 * descriptor to be mapped in the kernel.
748 shm_map(struct file *fp, size_t size, off_t offset, void **memp)
751 vm_offset_t kva, ofs;
755 if (fp->f_type != DTYPE_SHM)
758 obj = shmfd->shm_object;
759 VM_OBJECT_WLOCK(obj);
761 * XXXRW: This validation is probably insufficient, and subject to
762 * sign errors. It should be fixed.
764 if (offset >= shmfd->shm_size ||
765 offset + size > round_page(shmfd->shm_size)) {
766 VM_OBJECT_WUNLOCK(obj);
770 shmfd->shm_kmappings++;
771 vm_object_reference_locked(obj);
772 VM_OBJECT_WUNLOCK(obj);
774 /* Map the object into the kernel_map and wire it. */
775 kva = vm_map_min(kernel_map);
776 ofs = offset & PAGE_MASK;
777 offset = trunc_page(offset);
778 size = round_page(size + ofs);
779 rv = vm_map_find(kernel_map, obj, offset, &kva, size,
780 VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE,
781 VM_PROT_READ | VM_PROT_WRITE, 0);
782 if (rv == KERN_SUCCESS) {
783 rv = vm_map_wire(kernel_map, kva, kva + size,
784 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
785 if (rv == KERN_SUCCESS) {
786 *memp = (void *)(kva + ofs);
789 vm_map_remove(kernel_map, kva, kva + size);
791 vm_object_deallocate(obj);
793 /* On failure, drop our mapping reference. */
794 VM_OBJECT_WLOCK(obj);
795 shmfd->shm_kmappings--;
796 VM_OBJECT_WUNLOCK(obj);
798 return (vm_mmap_to_errno(rv));
802 * We require the caller to unmap the entire entry. This allows us to
803 * safely decrement shm_kmappings when a mapping is removed.
806 shm_unmap(struct file *fp, void *mem, size_t size)
809 vm_map_entry_t entry;
810 vm_offset_t kva, ofs;
818 if (fp->f_type != DTYPE_SHM)
821 kva = (vm_offset_t)mem;
822 ofs = kva & PAGE_MASK;
823 kva = trunc_page(kva);
824 size = round_page(size + ofs);
826 rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
827 &obj, &pindex, &prot, &wired);
828 if (rv != KERN_SUCCESS)
830 if (entry->start != kva || entry->end != kva + size) {
831 vm_map_lookup_done(map, entry);
834 vm_map_lookup_done(map, entry);
835 if (obj != shmfd->shm_object)
837 vm_map_remove(map, kva, kva + size);
838 VM_OBJECT_WLOCK(obj);
839 KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
840 shmfd->shm_kmappings--;
841 VM_OBJECT_WUNLOCK(obj);
846 shm_path(struct shmfd *shmfd, char *path, size_t size)
849 if (shmfd->shm_path == NULL)
851 sx_slock(&shm_dict_lock);
852 if (shmfd->shm_path != NULL)
853 strlcpy(path, shmfd->shm_path, size);
854 sx_sunlock(&shm_dict_lock);