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("$FreeBSD$");
60 #include <sys/param.h>
61 #include <sys/fcntl.h>
63 #include <sys/filedesc.h>
64 #include <sys/fnv_hash.h>
65 #include <sys/kernel.h>
67 #include <sys/malloc.h>
69 #include <sys/mutex.h>
71 #include <sys/refcount.h>
72 #include <sys/resourcevar.h>
74 #include <sys/sysctl.h>
75 #include <sys/sysproto.h>
76 #include <sys/systm.h>
79 #include <sys/vnode.h>
81 #include <security/mac/mac_framework.h>
84 #include <vm/vm_param.h>
86 #include <vm/vm_map.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_pager.h>
90 #include <vm/swap_pager.h>
95 struct shmfd *sm_shmfd;
96 LIST_ENTRY(shm_mapping) sm_link;
99 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
100 static LIST_HEAD(, shm_mapping) *shm_dictionary;
101 static struct sx shm_dict_lock;
102 static struct mtx shm_timestamp_lock;
103 static u_long shm_hash;
105 #define SHM_HASH(fnv) (&shm_dictionary[(fnv) & shm_hash])
107 static int shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags);
108 static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode);
109 static void shm_dict_init(void *arg);
110 static void shm_drop(struct shmfd *shmfd);
111 static struct shmfd *shm_hold(struct shmfd *shmfd);
112 static void shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
113 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
114 static int shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
115 static void shm_dotruncate(struct shmfd *shmfd, off_t length);
117 static fo_rdwr_t shm_read;
118 static fo_rdwr_t shm_write;
119 static fo_truncate_t shm_truncate;
120 static fo_ioctl_t shm_ioctl;
121 static fo_poll_t shm_poll;
122 static fo_kqfilter_t shm_kqfilter;
123 static fo_stat_t shm_stat;
124 static fo_close_t shm_close;
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_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 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_mode = S_IFREG | shmfd->shm_mode; /* XXX */
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 sb->st_atimespec = shmfd->shm_atime;
226 sb->st_ctimespec = shmfd->shm_ctime;
227 sb->st_mtimespec = shmfd->shm_mtime;
228 sb->st_birthtimespec = shmfd->shm_birthtime;
229 sb->st_uid = shmfd->shm_uid;
230 sb->st_gid = shmfd->shm_gid;
236 shm_close(struct file *fp, struct thread *td)
248 shm_dotruncate(struct shmfd *shmfd, off_t length)
252 vm_pindex_t nobjsize;
254 object = shmfd->shm_object;
255 VM_OBJECT_LOCK(object);
256 if (length == shmfd->shm_size) {
257 VM_OBJECT_UNLOCK(object);
260 nobjsize = OFF_TO_IDX(length + PAGE_MASK);
262 /* Are we shrinking? If so, trim the end. */
263 if (length < shmfd->shm_size) {
264 /* Toss in memory pages. */
265 if (nobjsize < object->size)
266 vm_object_page_remove(object, nobjsize, object->size,
269 /* Toss pages from swap. */
270 if (object->type == OBJT_SWAP)
271 swap_pager_freespace(object, nobjsize,
272 object->size - nobjsize);
275 * If the last page is partially mapped, then zero out
276 * the garbage at the end of the page. See comments
277 * in vnode_page_setsize() for more details.
279 * XXXJHB: This handles in memory pages, but what about
280 * a page swapped out to disk?
282 if ((length & PAGE_MASK) &&
283 (m = vm_page_lookup(object, OFF_TO_IDX(length))) != NULL &&
285 int base = (int)length & PAGE_MASK;
286 int size = PAGE_SIZE - base;
288 pmap_zero_page_area(m, base, size);
289 vm_page_lock_queues();
290 vm_page_set_validclean(m, base, size);
292 m->dirty = VM_PAGE_BITS_ALL;
293 vm_page_unlock_queues();
294 } else if ((length & PAGE_MASK) &&
295 __predict_false(object->cache != NULL)) {
296 vm_page_cache_free(object, OFF_TO_IDX(length),
300 shmfd->shm_size = length;
301 mtx_lock(&shm_timestamp_lock);
302 vfs_timestamp(&shmfd->shm_ctime);
303 shmfd->shm_mtime = shmfd->shm_ctime;
304 mtx_unlock(&shm_timestamp_lock);
305 object->size = nobjsize;
306 VM_OBJECT_UNLOCK(object);
310 * shmfd object management including creation and reference counting
313 static struct shmfd *
314 shm_alloc(struct ucred *ucred, mode_t mode)
318 shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
320 shmfd->shm_uid = ucred->cr_uid;
321 shmfd->shm_gid = ucred->cr_gid;
322 shmfd->shm_mode = mode;
323 shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
324 shmfd->shm_size, VM_PROT_DEFAULT, 0);
325 KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
326 VM_OBJECT_LOCK(shmfd->shm_object);
327 vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
328 vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
329 VM_OBJECT_UNLOCK(shmfd->shm_object);
330 vfs_timestamp(&shmfd->shm_birthtime);
331 shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
332 shmfd->shm_birthtime;
333 refcount_init(&shmfd->shm_refs, 1);
335 mac_posixshm_init(shmfd);
336 mac_posixshm_create(ucred, shmfd);
342 static struct shmfd *
343 shm_hold(struct shmfd *shmfd)
346 refcount_acquire(&shmfd->shm_refs);
351 shm_drop(struct shmfd *shmfd)
354 if (refcount_release(&shmfd->shm_refs)) {
356 mac_posixshm_destroy(shmfd);
358 vm_object_deallocate(shmfd->shm_object);
359 free(shmfd, M_SHMFD);
364 * Determine if the credentials have sufficient permissions for a
365 * specified combination of FREAD and FWRITE.
368 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
377 return (vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
378 acc_mode, ucred, NULL));
382 * Dictionary management. We maintain an in-kernel dictionary to map
383 * paths to shmfd objects. We use the FNV hash on the path to store
384 * the mappings in a hash table.
387 shm_dict_init(void *arg)
390 mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
391 sx_init(&shm_dict_lock, "shm dictionary");
392 shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
394 SYSINIT(shm_dict_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_dict_init, NULL);
396 static struct shmfd *
397 shm_lookup(char *path, Fnv32_t fnv)
399 struct shm_mapping *map;
401 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
402 if (map->sm_fnv != fnv)
404 if (strcmp(map->sm_path, path) == 0)
405 return (map->sm_shmfd);
412 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
414 struct shm_mapping *map;
416 map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
419 map->sm_shmfd = shm_hold(shmfd);
420 LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
424 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
426 struct shm_mapping *map;
429 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
430 if (map->sm_fnv != fnv)
432 if (strcmp(map->sm_path, path) == 0) {
434 error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
438 error = shm_access(map->sm_shmfd, ucred,
442 LIST_REMOVE(map, sm_link);
443 shm_drop(map->sm_shmfd);
444 free(map->sm_path, M_SHMFD);
455 shm_open(struct thread *td, struct shm_open_args *uap)
457 struct filedesc *fdp;
465 if ((uap->flags & O_ACCMODE) != O_RDONLY &&
466 (uap->flags & O_ACCMODE) != O_RDWR)
469 if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0)
472 fdp = td->td_proc->p_fd;
473 cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
475 error = falloc(td, &fp, &fd);
479 /* A SHM_ANON path pointer creates an anonymous object. */
480 if (uap->path == SHM_ANON) {
481 /* A read-only anonymous object is pointless. */
482 if ((uap->flags & O_ACCMODE) == O_RDONLY) {
483 fdclose(fdp, fp, fd, td);
487 shmfd = shm_alloc(td->td_ucred, cmode);
489 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
490 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
492 /* Require paths to start with a '/' character. */
493 if (error == 0 && path[0] != '/')
496 fdclose(fdp, fp, fd, td);
502 fnv = fnv_32_str(path, FNV1_32_INIT);
503 sx_xlock(&shm_dict_lock);
504 shmfd = shm_lookup(path, fnv);
506 /* Object does not yet exist, create it if requested. */
507 if (uap->flags & O_CREAT) {
508 shmfd = shm_alloc(td->td_ucred, cmode);
509 shm_insert(path, fnv, shmfd);
516 * Object already exists, obtain a new
517 * reference if requested and permitted.
520 if ((uap->flags & (O_CREAT | O_EXCL)) ==
525 error = mac_posixshm_check_open(td->td_ucred,
529 error = shm_access(shmfd, td->td_ucred,
530 FFLAGS(uap->flags & O_ACCMODE));
534 * Truncate the file back to zero length if
535 * O_TRUNC was specified and the object was
536 * opened with read/write.
539 (uap->flags & (O_ACCMODE | O_TRUNC)) ==
540 (O_RDWR | O_TRUNC)) {
542 error = mac_posixshm_check_truncate(
543 td->td_ucred, fp->f_cred, shmfd);
546 shm_dotruncate(shmfd, 0);
551 sx_xunlock(&shm_dict_lock);
554 fdclose(fdp, fp, fd, td);
560 finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
563 if (fdp->fd_ofiles[fd] == fp)
564 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
565 FILEDESC_XUNLOCK(fdp);
566 td->td_retval[0] = fd;
573 shm_unlink(struct thread *td, struct shm_unlink_args *uap)
579 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
580 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
586 fnv = fnv_32_str(path, FNV1_32_INIT);
587 sx_xlock(&shm_dict_lock);
588 error = shm_remove(path, fnv, td->td_ucred);
589 sx_xunlock(&shm_dict_lock);
596 * mmap() helper to validate mmap() requests against shm object state
597 * and give mmap() the vm_object to use for the mapping.
600 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
605 * XXXRW: This validation is probably insufficient, and subject to
606 * sign errors. It should be fixed.
608 if (foff >= shmfd->shm_size || foff + objsize > shmfd->shm_size)
611 mtx_lock(&shm_timestamp_lock);
612 vfs_timestamp(&shmfd->shm_atime);
613 mtx_unlock(&shm_timestamp_lock);
614 vm_object_reference(shmfd->shm_object);
615 *obj = shmfd->shm_object;