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?
43 * (4) 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?
51 #include <sys/cdefs.h>
52 __FBSDID("$FreeBSD$");
54 #include "opt_capsicum.h"
56 #include <sys/param.h>
57 #include <sys/capability.h>
58 #include <sys/fcntl.h>
60 #include <sys/filedesc.h>
61 #include <sys/fnv_hash.h>
62 #include <sys/kernel.h>
64 #include <sys/malloc.h>
66 #include <sys/mutex.h>
69 #include <sys/refcount.h>
70 #include <sys/resourcevar.h>
72 #include <sys/sysctl.h>
73 #include <sys/sysproto.h>
74 #include <sys/systm.h>
77 #include <sys/vnode.h>
79 #include <security/mac/mac_framework.h>
82 #include <vm/vm_param.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_page.h>
87 #include <vm/vm_pager.h>
88 #include <vm/swap_pager.h>
93 struct shmfd *sm_shmfd;
94 LIST_ENTRY(shm_mapping) sm_link;
97 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
98 static LIST_HEAD(, shm_mapping) *shm_dictionary;
99 static struct sx shm_dict_lock;
100 static struct mtx shm_timestamp_lock;
101 static u_long shm_hash;
103 #define SHM_HASH(fnv) (&shm_dictionary[(fnv) & shm_hash])
105 static int shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags);
106 static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode);
107 static void shm_dict_init(void *arg);
108 static void shm_drop(struct shmfd *shmfd);
109 static struct shmfd *shm_hold(struct shmfd *shmfd);
110 static void shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
111 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
112 static int shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
113 static int shm_dotruncate(struct shmfd *shmfd, off_t length);
115 static fo_rdwr_t shm_read;
116 static fo_rdwr_t shm_write;
117 static fo_truncate_t shm_truncate;
118 static fo_ioctl_t shm_ioctl;
119 static fo_poll_t shm_poll;
120 static fo_kqfilter_t shm_kqfilter;
121 static fo_stat_t shm_stat;
122 static fo_close_t shm_close;
123 static fo_chmod_t shm_chmod;
124 static fo_chown_t shm_chown;
126 /* File descriptor operations. */
127 static struct fileops shm_ops = {
129 .fo_write = shm_write,
130 .fo_truncate = shm_truncate,
131 .fo_ioctl = shm_ioctl,
133 .fo_kqfilter = shm_kqfilter,
135 .fo_close = shm_close,
136 .fo_chmod = shm_chmod,
137 .fo_chown = shm_chown,
138 .fo_flags = DFLAG_PASSABLE
141 FEATURE(posix_shm, "POSIX shared memory");
144 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
145 int flags, struct thread *td)
152 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
153 int flags, struct thread *td)
160 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
170 error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
174 return (shm_dotruncate(shmfd, length));
178 shm_ioctl(struct file *fp, u_long com, void *data,
179 struct ucred *active_cred, struct thread *td)
186 shm_poll(struct file *fp, int events, struct ucred *active_cred,
194 shm_kqfilter(struct file *fp, struct knote *kn)
201 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
212 error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
218 * Attempt to return sanish values for fstat() on a memory file
221 bzero(sb, sizeof(*sb));
222 sb->st_blksize = PAGE_SIZE;
223 sb->st_size = shmfd->shm_size;
224 sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize;
225 mtx_lock(&shm_timestamp_lock);
226 sb->st_atim = shmfd->shm_atime;
227 sb->st_ctim = shmfd->shm_ctime;
228 sb->st_mtim = shmfd->shm_mtime;
229 sb->st_birthtim = shmfd->shm_birthtime;
230 sb->st_mode = S_IFREG | shmfd->shm_mode; /* XXX */
231 sb->st_uid = shmfd->shm_uid;
232 sb->st_gid = shmfd->shm_gid;
233 mtx_unlock(&shm_timestamp_lock);
239 shm_close(struct file *fp, struct thread *td)
251 shm_dotruncate(struct shmfd *shmfd, off_t length)
255 vm_pindex_t nobjsize;
258 object = shmfd->shm_object;
259 VM_OBJECT_LOCK(object);
260 if (length == shmfd->shm_size) {
261 VM_OBJECT_UNLOCK(object);
264 nobjsize = OFF_TO_IDX(length + PAGE_MASK);
266 /* Are we shrinking? If so, trim the end. */
267 if (length < shmfd->shm_size) {
268 delta = ptoa(object->size - nobjsize);
270 /* Toss in memory pages. */
271 if (nobjsize < object->size)
272 vm_object_page_remove(object, nobjsize, object->size,
275 /* Toss pages from swap. */
276 if (object->type == OBJT_SWAP)
277 swap_pager_freespace(object, nobjsize, delta);
279 /* Free the swap accounted for shm */
280 swap_release_by_cred(delta, object->cred);
281 object->charge -= delta;
284 * If the last page is partially mapped, then zero out
285 * the garbage at the end of the page. See comments
286 * in vnode_pager_setsize() for more details.
288 * XXXJHB: This handles in memory pages, but what about
289 * a page swapped out to disk?
291 if ((length & PAGE_MASK) &&
292 (m = vm_page_lookup(object, OFF_TO_IDX(length))) != NULL &&
294 int base = (int)length & PAGE_MASK;
295 int size = PAGE_SIZE - base;
297 pmap_zero_page_area(m, base, size);
300 * Update the valid bits to reflect the blocks that
301 * have been zeroed. Some of these valid bits may
302 * have already been set.
304 vm_page_set_valid(m, base, size);
307 * Round "base" to the next block boundary so that the
308 * dirty bit for a partially zeroed block is not
311 base = roundup2(base, DEV_BSIZE);
313 vm_page_clear_dirty(m, base, PAGE_SIZE - base);
314 } else if ((length & PAGE_MASK) &&
315 __predict_false(object->cache != NULL)) {
316 vm_page_cache_free(object, OFF_TO_IDX(length),
321 /* Attempt to reserve the swap */
322 delta = ptoa(nobjsize - object->size);
323 if (!swap_reserve_by_cred(delta, object->cred)) {
324 VM_OBJECT_UNLOCK(object);
327 object->charge += delta;
329 shmfd->shm_size = length;
330 mtx_lock(&shm_timestamp_lock);
331 vfs_timestamp(&shmfd->shm_ctime);
332 shmfd->shm_mtime = shmfd->shm_ctime;
333 mtx_unlock(&shm_timestamp_lock);
334 object->size = nobjsize;
335 VM_OBJECT_UNLOCK(object);
340 * shmfd object management including creation and reference counting
343 static struct shmfd *
344 shm_alloc(struct ucred *ucred, mode_t mode)
348 shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
350 shmfd->shm_uid = ucred->cr_uid;
351 shmfd->shm_gid = ucred->cr_gid;
352 shmfd->shm_mode = mode;
353 shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
354 shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
355 KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
356 VM_OBJECT_LOCK(shmfd->shm_object);
357 vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
358 vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
359 VM_OBJECT_UNLOCK(shmfd->shm_object);
360 vfs_timestamp(&shmfd->shm_birthtime);
361 shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
362 shmfd->shm_birthtime;
363 refcount_init(&shmfd->shm_refs, 1);
365 mac_posixshm_init(shmfd);
366 mac_posixshm_create(ucred, shmfd);
372 static struct shmfd *
373 shm_hold(struct shmfd *shmfd)
376 refcount_acquire(&shmfd->shm_refs);
381 shm_drop(struct shmfd *shmfd)
384 if (refcount_release(&shmfd->shm_refs)) {
386 mac_posixshm_destroy(shmfd);
388 vm_object_deallocate(shmfd->shm_object);
389 free(shmfd, M_SHMFD);
394 * Determine if the credentials have sufficient permissions for a
395 * specified combination of FREAD and FWRITE.
398 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
408 mtx_lock(&shm_timestamp_lock);
409 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
410 accmode, ucred, NULL);
411 mtx_unlock(&shm_timestamp_lock);
416 * Dictionary management. We maintain an in-kernel dictionary to map
417 * paths to shmfd objects. We use the FNV hash on the path to store
418 * the mappings in a hash table.
421 shm_dict_init(void *arg)
424 mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
425 sx_init(&shm_dict_lock, "shm dictionary");
426 shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
428 SYSINIT(shm_dict_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_dict_init, NULL);
430 static struct shmfd *
431 shm_lookup(char *path, Fnv32_t fnv)
433 struct shm_mapping *map;
435 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
436 if (map->sm_fnv != fnv)
438 if (strcmp(map->sm_path, path) == 0)
439 return (map->sm_shmfd);
446 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
448 struct shm_mapping *map;
450 map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
453 map->sm_shmfd = shm_hold(shmfd);
454 LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
458 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
460 struct shm_mapping *map;
463 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
464 if (map->sm_fnv != fnv)
466 if (strcmp(map->sm_path, path) == 0) {
468 error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
472 error = shm_access(map->sm_shmfd, ucred,
476 LIST_REMOVE(map, sm_link);
477 shm_drop(map->sm_shmfd);
478 free(map->sm_path, M_SHMFD);
489 sys_shm_open(struct thread *td, struct shm_open_args *uap)
491 struct filedesc *fdp;
499 #ifdef CAPABILITY_MODE
501 * shm_open(2) is only allowed for anonymous objects.
503 if (IN_CAPABILITY_MODE(td) && (uap->path != SHM_ANON))
507 if ((uap->flags & O_ACCMODE) != O_RDONLY &&
508 (uap->flags & O_ACCMODE) != O_RDWR)
511 if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0)
514 fdp = td->td_proc->p_fd;
515 cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
517 error = falloc(td, &fp, &fd, 0);
521 /* A SHM_ANON path pointer creates an anonymous object. */
522 if (uap->path == SHM_ANON) {
523 /* A read-only anonymous object is pointless. */
524 if ((uap->flags & O_ACCMODE) == O_RDONLY) {
525 fdclose(fdp, fp, fd, td);
529 shmfd = shm_alloc(td->td_ucred, cmode);
531 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
532 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
534 /* Require paths to start with a '/' character. */
535 if (error == 0 && path[0] != '/')
538 fdclose(fdp, fp, fd, td);
544 fnv = fnv_32_str(path, FNV1_32_INIT);
545 sx_xlock(&shm_dict_lock);
546 shmfd = shm_lookup(path, fnv);
548 /* Object does not yet exist, create it if requested. */
549 if (uap->flags & O_CREAT) {
551 error = mac_posixshm_check_create(td->td_ucred,
555 shmfd = shm_alloc(td->td_ucred, cmode);
556 shm_insert(path, fnv, shmfd);
566 * Object already exists, obtain a new
567 * reference if requested and permitted.
570 if ((uap->flags & (O_CREAT | O_EXCL)) ==
575 error = mac_posixshm_check_open(td->td_ucred,
576 shmfd, FFLAGS(uap->flags & O_ACCMODE));
579 error = shm_access(shmfd, td->td_ucred,
580 FFLAGS(uap->flags & O_ACCMODE));
584 * Truncate the file back to zero length if
585 * O_TRUNC was specified and the object was
586 * opened with read/write.
589 (uap->flags & (O_ACCMODE | O_TRUNC)) ==
590 (O_RDWR | O_TRUNC)) {
592 error = mac_posixshm_check_truncate(
593 td->td_ucred, fp->f_cred, shmfd);
596 shm_dotruncate(shmfd, 0);
601 sx_xunlock(&shm_dict_lock);
604 fdclose(fdp, fp, fd, td);
610 finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
613 if (fdp->fd_ofiles[fd] == fp)
614 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
615 FILEDESC_XUNLOCK(fdp);
616 td->td_retval[0] = fd;
623 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
629 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
630 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
636 fnv = fnv_32_str(path, FNV1_32_INIT);
637 sx_xlock(&shm_dict_lock);
638 error = shm_remove(path, fnv, td->td_ucred);
639 sx_xunlock(&shm_dict_lock);
646 * mmap() helper to validate mmap() requests against shm object state
647 * and give mmap() the vm_object to use for the mapping.
650 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
655 * XXXRW: This validation is probably insufficient, and subject to
656 * sign errors. It should be fixed.
658 if (foff >= shmfd->shm_size ||
659 foff + objsize > round_page(shmfd->shm_size))
662 mtx_lock(&shm_timestamp_lock);
663 vfs_timestamp(&shmfd->shm_atime);
664 mtx_unlock(&shm_timestamp_lock);
665 vm_object_reference(shmfd->shm_object);
666 *obj = shmfd->shm_object;
671 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
679 mtx_lock(&shm_timestamp_lock);
681 * SUSv4 says that x bits of permission need not be affected.
682 * Be consistent with our shm_open there.
685 error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
689 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
690 shmfd->shm_gid, VADMIN, active_cred, NULL);
693 shmfd->shm_mode = mode & ACCESSPERMS;
695 mtx_unlock(&shm_timestamp_lock);
700 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
708 mtx_lock(&shm_timestamp_lock);
710 error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
714 if (uid == (uid_t)-1)
715 uid = shmfd->shm_uid;
716 if (gid == (gid_t)-1)
717 gid = shmfd->shm_gid;
718 if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
719 (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
720 (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
722 shmfd->shm_uid = uid;
723 shmfd->shm_gid = gid;
725 mtx_unlock(&shm_timestamp_lock);