2 * Copyright (c) 2006 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 * (2) 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 * (3) Add support for this file type to fstat(1).
40 * (4) Resource limits? Does this need its own resource limits or are the
41 * existing limits in mmap(2) sufficient?
43 * (5) Partial page truncation. vnode_pager_setsize() will zero any parts
44 * of a partially mapped page as a result of ftruncate(2)/truncate(2).
45 * We can do the same (with the same pmap evil), but do we need to
46 * worry about the bits on disk if the page is swapped out or will the
47 * swapper zero the parts of a page that are invalid if the page is
48 * swapped back in for us?
50 * (6) Add MAC support in mac_biba(4) and mac_mls(4).
52 * (7) Add a MAC check_create() hook for creating new named objects.
55 #include <sys/cdefs.h>
56 __FBSDID("$FreeBSD$");
58 #include "opt_capsicum.h"
60 #include <sys/param.h>
61 #include <sys/capability.h>
62 #include <sys/fcntl.h>
64 #include <sys/filedesc.h>
65 #include <sys/fnv_hash.h>
66 #include <sys/kernel.h>
68 #include <sys/malloc.h>
70 #include <sys/mutex.h>
73 #include <sys/refcount.h>
74 #include <sys/resourcevar.h>
76 #include <sys/sysctl.h>
77 #include <sys/sysproto.h>
78 #include <sys/systm.h>
81 #include <sys/vnode.h>
83 #include <security/mac/mac_framework.h>
86 #include <vm/vm_param.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_object.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_pager.h>
92 #include <vm/swap_pager.h>
97 struct shmfd *sm_shmfd;
98 LIST_ENTRY(shm_mapping) sm_link;
101 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
102 static LIST_HEAD(, shm_mapping) *shm_dictionary;
103 static struct sx shm_dict_lock;
104 static struct mtx shm_timestamp_lock;
105 static u_long shm_hash;
107 #define SHM_HASH(fnv) (&shm_dictionary[(fnv) & shm_hash])
109 static int shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags);
110 static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode);
111 static void shm_dict_init(void *arg);
112 static void shm_drop(struct shmfd *shmfd);
113 static struct shmfd *shm_hold(struct shmfd *shmfd);
114 static void shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
115 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
116 static int shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
117 static int shm_dotruncate(struct shmfd *shmfd, off_t length);
119 static fo_rdwr_t shm_read;
120 static fo_rdwr_t shm_write;
121 static fo_truncate_t shm_truncate;
122 static fo_ioctl_t shm_ioctl;
123 static fo_poll_t shm_poll;
124 static fo_kqfilter_t shm_kqfilter;
125 static fo_stat_t shm_stat;
126 static fo_close_t shm_close;
127 static fo_chmod_t shm_chmod;
128 static fo_chown_t shm_chown;
130 /* File descriptor operations. */
131 static struct fileops shm_ops = {
133 .fo_write = shm_write,
134 .fo_truncate = shm_truncate,
135 .fo_ioctl = shm_ioctl,
137 .fo_kqfilter = shm_kqfilter,
139 .fo_close = shm_close,
140 .fo_chmod = shm_chmod,
141 .fo_chown = shm_chown,
142 .fo_flags = DFLAG_PASSABLE
145 FEATURE(posix_shm, "POSIX shared memory");
148 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
149 int flags, struct thread *td)
156 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
157 int flags, struct thread *td)
164 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
174 error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
178 return (shm_dotruncate(shmfd, length));
182 shm_ioctl(struct file *fp, u_long com, void *data,
183 struct ucred *active_cred, struct thread *td)
190 shm_poll(struct file *fp, int events, struct ucred *active_cred,
198 shm_kqfilter(struct file *fp, struct knote *kn)
205 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
216 error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
222 * Attempt to return sanish values for fstat() on a memory file
225 bzero(sb, sizeof(*sb));
226 sb->st_blksize = PAGE_SIZE;
227 sb->st_size = shmfd->shm_size;
228 sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize;
229 mtx_lock(&shm_timestamp_lock);
230 sb->st_atim = shmfd->shm_atime;
231 sb->st_ctim = shmfd->shm_ctime;
232 sb->st_mtim = shmfd->shm_mtime;
233 sb->st_birthtim = shmfd->shm_birthtime;
234 sb->st_mode = S_IFREG | shmfd->shm_mode; /* XXX */
235 sb->st_uid = shmfd->shm_uid;
236 sb->st_gid = shmfd->shm_gid;
237 mtx_unlock(&shm_timestamp_lock);
243 shm_close(struct file *fp, struct thread *td)
255 shm_dotruncate(struct shmfd *shmfd, off_t length)
259 vm_pindex_t nobjsize;
262 object = shmfd->shm_object;
263 VM_OBJECT_LOCK(object);
264 if (length == shmfd->shm_size) {
265 VM_OBJECT_UNLOCK(object);
268 nobjsize = OFF_TO_IDX(length + PAGE_MASK);
270 /* Are we shrinking? If so, trim the end. */
271 if (length < shmfd->shm_size) {
272 delta = ptoa(object->size - nobjsize);
274 /* Toss in memory pages. */
275 if (nobjsize < object->size)
276 vm_object_page_remove(object, nobjsize, object->size,
279 /* Toss pages from swap. */
280 if (object->type == OBJT_SWAP)
281 swap_pager_freespace(object, nobjsize, delta);
283 /* Free the swap accounted for shm */
284 swap_release_by_cred(delta, object->cred);
285 object->charge -= delta;
288 * If the last page is partially mapped, then zero out
289 * the garbage at the end of the page. See comments
290 * in vnode_pager_setsize() for more details.
292 * XXXJHB: This handles in memory pages, but what about
293 * a page swapped out to disk?
295 if ((length & PAGE_MASK) &&
296 (m = vm_page_lookup(object, OFF_TO_IDX(length))) != NULL &&
298 int base = (int)length & PAGE_MASK;
299 int size = PAGE_SIZE - base;
301 pmap_zero_page_area(m, base, size);
304 * Update the valid bits to reflect the blocks that
305 * have been zeroed. Some of these valid bits may
306 * have already been set.
308 vm_page_set_valid(m, base, size);
311 * Round "base" to the next block boundary so that the
312 * dirty bit for a partially zeroed block is not
315 base = roundup2(base, DEV_BSIZE);
317 vm_page_clear_dirty(m, base, PAGE_SIZE - base);
318 } else if ((length & PAGE_MASK) &&
319 __predict_false(object->cache != NULL)) {
320 vm_page_cache_free(object, OFF_TO_IDX(length),
325 /* Attempt to reserve the swap */
326 delta = ptoa(nobjsize - object->size);
327 if (!swap_reserve_by_cred(delta, object->cred)) {
328 VM_OBJECT_UNLOCK(object);
331 object->charge += delta;
333 shmfd->shm_size = length;
334 mtx_lock(&shm_timestamp_lock);
335 vfs_timestamp(&shmfd->shm_ctime);
336 shmfd->shm_mtime = shmfd->shm_ctime;
337 mtx_unlock(&shm_timestamp_lock);
338 object->size = nobjsize;
339 VM_OBJECT_UNLOCK(object);
344 * shmfd object management including creation and reference counting
347 static struct shmfd *
348 shm_alloc(struct ucred *ucred, mode_t mode)
352 shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
354 shmfd->shm_uid = ucred->cr_uid;
355 shmfd->shm_gid = ucred->cr_gid;
356 shmfd->shm_mode = mode;
357 shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
358 shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
359 KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
360 VM_OBJECT_LOCK(shmfd->shm_object);
361 vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
362 vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
363 VM_OBJECT_UNLOCK(shmfd->shm_object);
364 vfs_timestamp(&shmfd->shm_birthtime);
365 shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
366 shmfd->shm_birthtime;
367 refcount_init(&shmfd->shm_refs, 1);
369 mac_posixshm_init(shmfd);
370 mac_posixshm_create(ucred, shmfd);
376 static struct shmfd *
377 shm_hold(struct shmfd *shmfd)
380 refcount_acquire(&shmfd->shm_refs);
385 shm_drop(struct shmfd *shmfd)
388 if (refcount_release(&shmfd->shm_refs)) {
390 mac_posixshm_destroy(shmfd);
392 vm_object_deallocate(shmfd->shm_object);
393 free(shmfd, M_SHMFD);
398 * Determine if the credentials have sufficient permissions for a
399 * specified combination of FREAD and FWRITE.
402 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
412 mtx_lock(&shm_timestamp_lock);
413 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
414 accmode, ucred, NULL);
415 mtx_unlock(&shm_timestamp_lock);
420 * Dictionary management. We maintain an in-kernel dictionary to map
421 * paths to shmfd objects. We use the FNV hash on the path to store
422 * the mappings in a hash table.
425 shm_dict_init(void *arg)
428 mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
429 sx_init(&shm_dict_lock, "shm dictionary");
430 shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
432 SYSINIT(shm_dict_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_dict_init, NULL);
434 static struct shmfd *
435 shm_lookup(char *path, Fnv32_t fnv)
437 struct shm_mapping *map;
439 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
440 if (map->sm_fnv != fnv)
442 if (strcmp(map->sm_path, path) == 0)
443 return (map->sm_shmfd);
450 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
452 struct shm_mapping *map;
454 map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
457 map->sm_shmfd = shm_hold(shmfd);
458 LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
462 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
464 struct shm_mapping *map;
467 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
468 if (map->sm_fnv != fnv)
470 if (strcmp(map->sm_path, path) == 0) {
472 error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
476 error = shm_access(map->sm_shmfd, ucred,
480 LIST_REMOVE(map, sm_link);
481 shm_drop(map->sm_shmfd);
482 free(map->sm_path, M_SHMFD);
493 shm_open(struct thread *td, struct shm_open_args *uap)
495 struct filedesc *fdp;
503 #ifdef CAPABILITY_MODE
505 * shm_open(2) is only allowed for anonymous objects.
507 if (IN_CAPABILITY_MODE(td) && (uap->path != SHM_ANON))
511 if ((uap->flags & O_ACCMODE) != O_RDONLY &&
512 (uap->flags & O_ACCMODE) != O_RDWR)
515 if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0)
518 fdp = td->td_proc->p_fd;
519 cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
521 error = falloc(td, &fp, &fd, 0);
525 /* A SHM_ANON path pointer creates an anonymous object. */
526 if (uap->path == SHM_ANON) {
527 /* A read-only anonymous object is pointless. */
528 if ((uap->flags & O_ACCMODE) == O_RDONLY) {
529 fdclose(fdp, fp, fd, td);
533 shmfd = shm_alloc(td->td_ucred, cmode);
535 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
536 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
538 /* Require paths to start with a '/' character. */
539 if (error == 0 && path[0] != '/')
542 fdclose(fdp, fp, fd, td);
548 fnv = fnv_32_str(path, FNV1_32_INIT);
549 sx_xlock(&shm_dict_lock);
550 shmfd = shm_lookup(path, fnv);
552 /* Object does not yet exist, create it if requested. */
553 if (uap->flags & O_CREAT) {
554 shmfd = shm_alloc(td->td_ucred, cmode);
555 shm_insert(path, fnv, shmfd);
562 * Object already exists, obtain a new
563 * reference if requested and permitted.
566 if ((uap->flags & (O_CREAT | O_EXCL)) ==
571 error = mac_posixshm_check_open(td->td_ucred,
575 error = shm_access(shmfd, td->td_ucred,
576 FFLAGS(uap->flags & O_ACCMODE));
580 * Truncate the file back to zero length if
581 * O_TRUNC was specified and the object was
582 * opened with read/write.
585 (uap->flags & (O_ACCMODE | O_TRUNC)) ==
586 (O_RDWR | O_TRUNC)) {
588 error = mac_posixshm_check_truncate(
589 td->td_ucred, fp->f_cred, shmfd);
592 shm_dotruncate(shmfd, 0);
597 sx_xunlock(&shm_dict_lock);
600 fdclose(fdp, fp, fd, td);
606 finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
609 if (fdp->fd_ofiles[fd] == fp)
610 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
611 FILEDESC_XUNLOCK(fdp);
612 td->td_retval[0] = fd;
619 shm_unlink(struct thread *td, struct shm_unlink_args *uap)
625 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
626 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
632 fnv = fnv_32_str(path, FNV1_32_INIT);
633 sx_xlock(&shm_dict_lock);
634 error = shm_remove(path, fnv, td->td_ucred);
635 sx_xunlock(&shm_dict_lock);
642 * mmap() helper to validate mmap() requests against shm object state
643 * and give mmap() the vm_object to use for the mapping.
646 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
651 * XXXRW: This validation is probably insufficient, and subject to
652 * sign errors. It should be fixed.
654 if (foff >= shmfd->shm_size ||
655 foff + objsize > round_page(shmfd->shm_size))
658 mtx_lock(&shm_timestamp_lock);
659 vfs_timestamp(&shmfd->shm_atime);
660 mtx_unlock(&shm_timestamp_lock);
661 vm_object_reference(shmfd->shm_object);
662 *obj = shmfd->shm_object;
667 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
675 mtx_lock(&shm_timestamp_lock);
677 * SUSv4 says that x bits of permission need not be affected.
678 * Be consistent with our shm_open there.
681 error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
685 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
686 shmfd->shm_gid, VADMIN, active_cred, NULL);
689 shmfd->shm_mode = mode & ACCESSPERMS;
691 mtx_unlock(&shm_timestamp_lock);
696 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
704 mtx_lock(&shm_timestamp_lock);
706 error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
710 if (uid == (uid_t)-1)
711 uid = shmfd->shm_uid;
712 if (gid == (gid_t)-1)
713 gid = shmfd->shm_gid;
714 if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
715 (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
716 (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
718 shmfd->shm_uid = uid;
719 shmfd->shm_gid = gid;
721 mtx_unlock(&shm_timestamp_lock);