xref: /freebsd/sys/kern/uipc_shm.c (revision 595e514d0df2bac5b813d35f83e32875dbf16a83)
1 /*-
2  * Copyright (c) 2006, 2011 Robert N. M. Watson
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
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.
13  *
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
24  * SUCH DAMAGE.
25  */
26 
27 /*
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.
31  *
32  * TODO:
33  *
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?
37  *
38  * (2) Add support for this file type to fstat(1).
39  *
40  * (3) Resource limits?  Does this need its own resource limits or are the
41  *     existing limits in mmap(2) sufficient?
42  */
43 
44 #include <sys/cdefs.h>
45 __FBSDID("$FreeBSD$");
46 
47 #include "opt_capsicum.h"
48 
49 #include <sys/param.h>
50 #include <sys/capability.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/fnv_hash.h>
55 #include <sys/kernel.h>
56 #include <sys/lock.h>
57 #include <sys/malloc.h>
58 #include <sys/mman.h>
59 #include <sys/mutex.h>
60 #include <sys/priv.h>
61 #include <sys/proc.h>
62 #include <sys/refcount.h>
63 #include <sys/resourcevar.h>
64 #include <sys/rwlock.h>
65 #include <sys/stat.h>
66 #include <sys/sysctl.h>
67 #include <sys/sysproto.h>
68 #include <sys/systm.h>
69 #include <sys/sx.h>
70 #include <sys/time.h>
71 #include <sys/vnode.h>
72 
73 #include <security/mac/mac_framework.h>
74 
75 #include <vm/vm.h>
76 #include <vm/vm_param.h>
77 #include <vm/pmap.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>
86 
87 struct shm_mapping {
88 	char		*sm_path;
89 	Fnv32_t		sm_fnv;
90 	struct shmfd	*sm_shmfd;
91 	LIST_ENTRY(shm_mapping) sm_link;
92 };
93 
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 
100 #define	SHM_HASH(fnv)	(&shm_dictionary[(fnv) & shm_hash])
101 
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);
111 
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;
122 
123 /* File descriptor operations. */
124 static struct fileops shm_ops = {
125 	.fo_read = shm_read,
126 	.fo_write = shm_write,
127 	.fo_truncate = shm_truncate,
128 	.fo_ioctl = shm_ioctl,
129 	.fo_poll = shm_poll,
130 	.fo_kqfilter = shm_kqfilter,
131 	.fo_stat = shm_stat,
132 	.fo_close = shm_close,
133 	.fo_chmod = shm_chmod,
134 	.fo_chown = shm_chown,
135 	.fo_flags = DFLAG_PASSABLE
136 };
137 
138 FEATURE(posix_shm, "POSIX shared memory");
139 
140 static int
141 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
142     int flags, struct thread *td)
143 {
144 
145 	return (EOPNOTSUPP);
146 }
147 
148 static int
149 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
150     int flags, struct thread *td)
151 {
152 
153 	return (EOPNOTSUPP);
154 }
155 
156 static int
157 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
158     struct thread *td)
159 {
160 	struct shmfd *shmfd;
161 #ifdef MAC
162 	int error;
163 #endif
164 
165 	shmfd = fp->f_data;
166 #ifdef MAC
167 	error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
168 	if (error)
169 		return (error);
170 #endif
171 	return (shm_dotruncate(shmfd, length));
172 }
173 
174 static int
175 shm_ioctl(struct file *fp, u_long com, void *data,
176     struct ucred *active_cred, struct thread *td)
177 {
178 
179 	return (EOPNOTSUPP);
180 }
181 
182 static int
183 shm_poll(struct file *fp, int events, struct ucred *active_cred,
184     struct thread *td)
185 {
186 
187 	return (EOPNOTSUPP);
188 }
189 
190 static int
191 shm_kqfilter(struct file *fp, struct knote *kn)
192 {
193 
194 	return (EOPNOTSUPP);
195 }
196 
197 static int
198 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
199     struct thread *td)
200 {
201 	struct shmfd *shmfd;
202 #ifdef MAC
203 	int error;
204 #endif
205 
206 	shmfd = fp->f_data;
207 
208 #ifdef MAC
209 	error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
210 	if (error)
211 		return (error);
212 #endif
213 
214 	/*
215 	 * Attempt to return sanish values for fstat() on a memory file
216 	 * descriptor.
217 	 */
218 	bzero(sb, sizeof(*sb));
219 	sb->st_blksize = PAGE_SIZE;
220 	sb->st_size = shmfd->shm_size;
221 	sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize;
222 	mtx_lock(&shm_timestamp_lock);
223 	sb->st_atim = shmfd->shm_atime;
224 	sb->st_ctim = shmfd->shm_ctime;
225 	sb->st_mtim = shmfd->shm_mtime;
226 	sb->st_birthtim = shmfd->shm_birthtime;
227 	sb->st_mode = S_IFREG | shmfd->shm_mode;		/* XXX */
228 	sb->st_uid = shmfd->shm_uid;
229 	sb->st_gid = shmfd->shm_gid;
230 	mtx_unlock(&shm_timestamp_lock);
231 
232 	return (0);
233 }
234 
235 static int
236 shm_close(struct file *fp, struct thread *td)
237 {
238 	struct shmfd *shmfd;
239 
240 	shmfd = fp->f_data;
241 	fp->f_data = NULL;
242 	shm_drop(shmfd);
243 
244 	return (0);
245 }
246 
247 static int
248 shm_dotruncate(struct shmfd *shmfd, off_t length)
249 {
250 	vm_object_t object;
251 	vm_page_t m, ma[1];
252 	vm_pindex_t idx, nobjsize;
253 	vm_ooffset_t delta;
254 	int base, rv;
255 
256 	object = shmfd->shm_object;
257 	VM_OBJECT_WLOCK(object);
258 	if (length == shmfd->shm_size) {
259 		VM_OBJECT_WUNLOCK(object);
260 		return (0);
261 	}
262 	nobjsize = OFF_TO_IDX(length + PAGE_MASK);
263 
264 	/* Are we shrinking?  If so, trim the end. */
265 	if (length < shmfd->shm_size) {
266 		/*
267 		 * Disallow any requests to shrink the size if this
268 		 * object is mapped into the kernel.
269 		 */
270 		if (shmfd->shm_kmappings > 0) {
271 			VM_OBJECT_WUNLOCK(object);
272 			return (EBUSY);
273 		}
274 
275 		/*
276 		 * Zero the truncated part of the last page.
277 		 */
278 		base = length & PAGE_MASK;
279 		if (base != 0) {
280 			idx = OFF_TO_IDX(length);
281 retry:
282 			m = vm_page_lookup(object, idx);
283 			if (m != NULL) {
284 				if ((m->oflags & VPO_BUSY) != 0 ||
285 				    m->busy != 0) {
286 					vm_page_sleep(m, "shmtrc");
287 					goto retry;
288 				}
289 			} else if (vm_pager_has_page(object, idx, NULL, NULL)) {
290 				m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL);
291 				if (m == NULL) {
292 					VM_OBJECT_WUNLOCK(object);
293 					VM_WAIT;
294 					VM_OBJECT_WLOCK(object);
295 					goto retry;
296 				} else if (m->valid != VM_PAGE_BITS_ALL) {
297 					ma[0] = m;
298 					rv = vm_pager_get_pages(object, ma, 1,
299 					    0);
300 					m = vm_page_lookup(object, idx);
301 				} else
302 					/* A cached page was reactivated. */
303 					rv = VM_PAGER_OK;
304 				vm_page_lock(m);
305 				if (rv == VM_PAGER_OK) {
306 					vm_page_deactivate(m);
307 					vm_page_unlock(m);
308 					vm_page_wakeup(m);
309 				} else {
310 					vm_page_free(m);
311 					vm_page_unlock(m);
312 					VM_OBJECT_WUNLOCK(object);
313 					return (EIO);
314 				}
315 			}
316 			if (m != NULL) {
317 				pmap_zero_page_area(m, base, PAGE_SIZE - base);
318 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
319 				    ("shm_dotruncate: page %p is invalid", m));
320 				vm_page_dirty(m);
321 				vm_pager_page_unswapped(m);
322 			}
323 		}
324 		delta = ptoa(object->size - nobjsize);
325 
326 		/* Toss in memory pages. */
327 		if (nobjsize < object->size)
328 			vm_object_page_remove(object, nobjsize, object->size,
329 			    0);
330 
331 		/* Toss pages from swap. */
332 		if (object->type == OBJT_SWAP)
333 			swap_pager_freespace(object, nobjsize, delta);
334 
335 		/* Free the swap accounted for shm */
336 		swap_release_by_cred(delta, object->cred);
337 		object->charge -= delta;
338 	} else {
339 		/* Attempt to reserve the swap */
340 		delta = ptoa(nobjsize - object->size);
341 		if (!swap_reserve_by_cred(delta, object->cred)) {
342 			VM_OBJECT_WUNLOCK(object);
343 			return (ENOMEM);
344 		}
345 		object->charge += delta;
346 	}
347 	shmfd->shm_size = length;
348 	mtx_lock(&shm_timestamp_lock);
349 	vfs_timestamp(&shmfd->shm_ctime);
350 	shmfd->shm_mtime = shmfd->shm_ctime;
351 	mtx_unlock(&shm_timestamp_lock);
352 	object->size = nobjsize;
353 	VM_OBJECT_WUNLOCK(object);
354 	return (0);
355 }
356 
357 /*
358  * shmfd object management including creation and reference counting
359  * routines.
360  */
361 static struct shmfd *
362 shm_alloc(struct ucred *ucred, mode_t mode)
363 {
364 	struct shmfd *shmfd;
365 
366 	shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
367 	shmfd->shm_size = 0;
368 	shmfd->shm_uid = ucred->cr_uid;
369 	shmfd->shm_gid = ucred->cr_gid;
370 	shmfd->shm_mode = mode;
371 	shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
372 	    shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
373 	KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
374 	VM_OBJECT_WLOCK(shmfd->shm_object);
375 	vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
376 	vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
377 	VM_OBJECT_WUNLOCK(shmfd->shm_object);
378 	vfs_timestamp(&shmfd->shm_birthtime);
379 	shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
380 	    shmfd->shm_birthtime;
381 	refcount_init(&shmfd->shm_refs, 1);
382 #ifdef MAC
383 	mac_posixshm_init(shmfd);
384 	mac_posixshm_create(ucred, shmfd);
385 #endif
386 
387 	return (shmfd);
388 }
389 
390 static struct shmfd *
391 shm_hold(struct shmfd *shmfd)
392 {
393 
394 	refcount_acquire(&shmfd->shm_refs);
395 	return (shmfd);
396 }
397 
398 static void
399 shm_drop(struct shmfd *shmfd)
400 {
401 
402 	if (refcount_release(&shmfd->shm_refs)) {
403 #ifdef MAC
404 		mac_posixshm_destroy(shmfd);
405 #endif
406 		vm_object_deallocate(shmfd->shm_object);
407 		free(shmfd, M_SHMFD);
408 	}
409 }
410 
411 /*
412  * Determine if the credentials have sufficient permissions for a
413  * specified combination of FREAD and FWRITE.
414  */
415 static int
416 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
417 {
418 	accmode_t accmode;
419 	int error;
420 
421 	accmode = 0;
422 	if (flags & FREAD)
423 		accmode |= VREAD;
424 	if (flags & FWRITE)
425 		accmode |= VWRITE;
426 	mtx_lock(&shm_timestamp_lock);
427 	error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
428 	    accmode, ucred, NULL);
429 	mtx_unlock(&shm_timestamp_lock);
430 	return (error);
431 }
432 
433 /*
434  * Dictionary management.  We maintain an in-kernel dictionary to map
435  * paths to shmfd objects.  We use the FNV hash on the path to store
436  * the mappings in a hash table.
437  */
438 static void
439 shm_dict_init(void *arg)
440 {
441 
442 	mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
443 	sx_init(&shm_dict_lock, "shm dictionary");
444 	shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
445 }
446 SYSINIT(shm_dict_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_dict_init, NULL);
447 
448 static struct shmfd *
449 shm_lookup(char *path, Fnv32_t fnv)
450 {
451 	struct shm_mapping *map;
452 
453 	LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
454 		if (map->sm_fnv != fnv)
455 			continue;
456 		if (strcmp(map->sm_path, path) == 0)
457 			return (map->sm_shmfd);
458 	}
459 
460 	return (NULL);
461 }
462 
463 static void
464 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
465 {
466 	struct shm_mapping *map;
467 
468 	map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
469 	map->sm_path = path;
470 	map->sm_fnv = fnv;
471 	map->sm_shmfd = shm_hold(shmfd);
472 	shmfd->shm_path = path;
473 	LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
474 }
475 
476 static int
477 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
478 {
479 	struct shm_mapping *map;
480 	int error;
481 
482 	LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
483 		if (map->sm_fnv != fnv)
484 			continue;
485 		if (strcmp(map->sm_path, path) == 0) {
486 #ifdef MAC
487 			error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
488 			if (error)
489 				return (error);
490 #endif
491 			error = shm_access(map->sm_shmfd, ucred,
492 			    FREAD | FWRITE);
493 			if (error)
494 				return (error);
495 			map->sm_shmfd->shm_path = NULL;
496 			LIST_REMOVE(map, sm_link);
497 			shm_drop(map->sm_shmfd);
498 			free(map->sm_path, M_SHMFD);
499 			free(map, M_SHMFD);
500 			return (0);
501 		}
502 	}
503 
504 	return (ENOENT);
505 }
506 
507 /* System calls. */
508 int
509 sys_shm_open(struct thread *td, struct shm_open_args *uap)
510 {
511 	struct filedesc *fdp;
512 	struct shmfd *shmfd;
513 	struct file *fp;
514 	char *path;
515 	Fnv32_t fnv;
516 	mode_t cmode;
517 	int fd, error;
518 
519 #ifdef CAPABILITY_MODE
520 	/*
521 	 * shm_open(2) is only allowed for anonymous objects.
522 	 */
523 	if (IN_CAPABILITY_MODE(td) && (uap->path != SHM_ANON))
524 		return (ECAPMODE);
525 #endif
526 
527 	if ((uap->flags & O_ACCMODE) != O_RDONLY &&
528 	    (uap->flags & O_ACCMODE) != O_RDWR)
529 		return (EINVAL);
530 
531 	if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0)
532 		return (EINVAL);
533 
534 	fdp = td->td_proc->p_fd;
535 	cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
536 
537 	error = falloc(td, &fp, &fd, O_CLOEXEC);
538 	if (error)
539 		return (error);
540 
541 	/* A SHM_ANON path pointer creates an anonymous object. */
542 	if (uap->path == SHM_ANON) {
543 		/* A read-only anonymous object is pointless. */
544 		if ((uap->flags & O_ACCMODE) == O_RDONLY) {
545 			fdclose(fdp, fp, fd, td);
546 			fdrop(fp, td);
547 			return (EINVAL);
548 		}
549 		shmfd = shm_alloc(td->td_ucred, cmode);
550 	} else {
551 		path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
552 		error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
553 
554 		/* Require paths to start with a '/' character. */
555 		if (error == 0 && path[0] != '/')
556 			error = EINVAL;
557 		if (error) {
558 			fdclose(fdp, fp, fd, td);
559 			fdrop(fp, td);
560 			free(path, M_SHMFD);
561 			return (error);
562 		}
563 
564 		fnv = fnv_32_str(path, FNV1_32_INIT);
565 		sx_xlock(&shm_dict_lock);
566 		shmfd = shm_lookup(path, fnv);
567 		if (shmfd == NULL) {
568 			/* Object does not yet exist, create it if requested. */
569 			if (uap->flags & O_CREAT) {
570 #ifdef MAC
571 				error = mac_posixshm_check_create(td->td_ucred,
572 				    path);
573 				if (error == 0) {
574 #endif
575 					shmfd = shm_alloc(td->td_ucred, cmode);
576 					shm_insert(path, fnv, shmfd);
577 #ifdef MAC
578 				}
579 #endif
580 			} else {
581 				free(path, M_SHMFD);
582 				error = ENOENT;
583 			}
584 		} else {
585 			/*
586 			 * Object already exists, obtain a new
587 			 * reference if requested and permitted.
588 			 */
589 			free(path, M_SHMFD);
590 			if ((uap->flags & (O_CREAT | O_EXCL)) ==
591 			    (O_CREAT | O_EXCL))
592 				error = EEXIST;
593 			else {
594 #ifdef MAC
595 				error = mac_posixshm_check_open(td->td_ucred,
596 				    shmfd, FFLAGS(uap->flags & O_ACCMODE));
597 				if (error == 0)
598 #endif
599 				error = shm_access(shmfd, td->td_ucred,
600 				    FFLAGS(uap->flags & O_ACCMODE));
601 			}
602 
603 			/*
604 			 * Truncate the file back to zero length if
605 			 * O_TRUNC was specified and the object was
606 			 * opened with read/write.
607 			 */
608 			if (error == 0 &&
609 			    (uap->flags & (O_ACCMODE | O_TRUNC)) ==
610 			    (O_RDWR | O_TRUNC)) {
611 #ifdef MAC
612 				error = mac_posixshm_check_truncate(
613 					td->td_ucred, fp->f_cred, shmfd);
614 				if (error == 0)
615 #endif
616 					shm_dotruncate(shmfd, 0);
617 			}
618 			if (error == 0)
619 				shm_hold(shmfd);
620 		}
621 		sx_xunlock(&shm_dict_lock);
622 
623 		if (error) {
624 			fdclose(fdp, fp, fd, td);
625 			fdrop(fp, td);
626 			return (error);
627 		}
628 	}
629 
630 	finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
631 
632 	td->td_retval[0] = fd;
633 	fdrop(fp, td);
634 
635 	return (0);
636 }
637 
638 int
639 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
640 {
641 	char *path;
642 	Fnv32_t fnv;
643 	int error;
644 
645 	path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
646 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
647 	if (error) {
648 		free(path, M_TEMP);
649 		return (error);
650 	}
651 
652 	fnv = fnv_32_str(path, FNV1_32_INIT);
653 	sx_xlock(&shm_dict_lock);
654 	error = shm_remove(path, fnv, td->td_ucred);
655 	sx_xunlock(&shm_dict_lock);
656 	free(path, M_TEMP);
657 
658 	return (error);
659 }
660 
661 /*
662  * mmap() helper to validate mmap() requests against shm object state
663  * and give mmap() the vm_object to use for the mapping.
664  */
665 int
666 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
667     vm_object_t *obj)
668 {
669 
670 	/*
671 	 * XXXRW: This validation is probably insufficient, and subject to
672 	 * sign errors.  It should be fixed.
673 	 */
674 	if (foff >= shmfd->shm_size ||
675 	    foff + objsize > round_page(shmfd->shm_size))
676 		return (EINVAL);
677 
678 	mtx_lock(&shm_timestamp_lock);
679 	vfs_timestamp(&shmfd->shm_atime);
680 	mtx_unlock(&shm_timestamp_lock);
681 	vm_object_reference(shmfd->shm_object);
682 	*obj = shmfd->shm_object;
683 	return (0);
684 }
685 
686 static int
687 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
688     struct thread *td)
689 {
690 	struct shmfd *shmfd;
691 	int error;
692 
693 	error = 0;
694 	shmfd = fp->f_data;
695 	mtx_lock(&shm_timestamp_lock);
696 	/*
697 	 * SUSv4 says that x bits of permission need not be affected.
698 	 * Be consistent with our shm_open there.
699 	 */
700 #ifdef MAC
701 	error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
702 	if (error != 0)
703 		goto out;
704 #endif
705 	error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
706 	    shmfd->shm_gid, VADMIN, active_cred, NULL);
707 	if (error != 0)
708 		goto out;
709 	shmfd->shm_mode = mode & ACCESSPERMS;
710 out:
711 	mtx_unlock(&shm_timestamp_lock);
712 	return (error);
713 }
714 
715 static int
716 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
717     struct thread *td)
718 {
719 	struct shmfd *shmfd;
720 	int error;
721 
722 	error = 0;
723 	shmfd = fp->f_data;
724 	mtx_lock(&shm_timestamp_lock);
725 #ifdef MAC
726 	error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
727 	if (error != 0)
728 		goto out;
729 #endif
730 	if (uid == (uid_t)-1)
731 		uid = shmfd->shm_uid;
732 	if (gid == (gid_t)-1)
733                  gid = shmfd->shm_gid;
734 	if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
735 	    (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
736 	    (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
737 		goto out;
738 	shmfd->shm_uid = uid;
739 	shmfd->shm_gid = gid;
740 out:
741 	mtx_unlock(&shm_timestamp_lock);
742 	return (error);
743 }
744 
745 /*
746  * Helper routines to allow the backing object of a shared memory file
747  * descriptor to be mapped in the kernel.
748  */
749 int
750 shm_map(struct file *fp, size_t size, off_t offset, void **memp)
751 {
752 	struct shmfd *shmfd;
753 	vm_offset_t kva, ofs;
754 	vm_object_t obj;
755 	int rv;
756 
757 	if (fp->f_type != DTYPE_SHM)
758 		return (EINVAL);
759 	shmfd = fp->f_data;
760 	obj = shmfd->shm_object;
761 	VM_OBJECT_WLOCK(obj);
762 	/*
763 	 * XXXRW: This validation is probably insufficient, and subject to
764 	 * sign errors.  It should be fixed.
765 	 */
766 	if (offset >= shmfd->shm_size ||
767 	    offset + size > round_page(shmfd->shm_size)) {
768 		VM_OBJECT_WUNLOCK(obj);
769 		return (EINVAL);
770 	}
771 
772 	shmfd->shm_kmappings++;
773 	vm_object_reference_locked(obj);
774 	VM_OBJECT_WUNLOCK(obj);
775 
776 	/* Map the object into the kernel_map and wire it. */
777 	kva = vm_map_min(kernel_map);
778 	ofs = offset & PAGE_MASK;
779 	offset = trunc_page(offset);
780 	size = round_page(size + ofs);
781 	rv = vm_map_find(kernel_map, obj, offset, &kva, size,
782 	    VMFS_ALIGNED_SPACE, VM_PROT_READ | VM_PROT_WRITE,
783 	    VM_PROT_READ | VM_PROT_WRITE, 0);
784 	if (rv == KERN_SUCCESS) {
785 		rv = vm_map_wire(kernel_map, kva, kva + size,
786 		    VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
787 		if (rv == KERN_SUCCESS) {
788 			*memp = (void *)(kva + ofs);
789 			return (0);
790 		}
791 		vm_map_remove(kernel_map, kva, kva + size);
792 	} else
793 		vm_object_deallocate(obj);
794 
795 	/* On failure, drop our mapping reference. */
796 	VM_OBJECT_WLOCK(obj);
797 	shmfd->shm_kmappings--;
798 	VM_OBJECT_WUNLOCK(obj);
799 
800 	return (vm_mmap_to_errno(rv));
801 }
802 
803 /*
804  * We require the caller to unmap the entire entry.  This allows us to
805  * safely decrement shm_kmappings when a mapping is removed.
806  */
807 int
808 shm_unmap(struct file *fp, void *mem, size_t size)
809 {
810 	struct shmfd *shmfd;
811 	vm_map_entry_t entry;
812 	vm_offset_t kva, ofs;
813 	vm_object_t obj;
814 	vm_pindex_t pindex;
815 	vm_prot_t prot;
816 	boolean_t wired;
817 	vm_map_t map;
818 	int rv;
819 
820 	if (fp->f_type != DTYPE_SHM)
821 		return (EINVAL);
822 	shmfd = fp->f_data;
823 	kva = (vm_offset_t)mem;
824 	ofs = kva & PAGE_MASK;
825 	kva = trunc_page(kva);
826 	size = round_page(size + ofs);
827 	map = kernel_map;
828 	rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
829 	    &obj, &pindex, &prot, &wired);
830 	if (rv != KERN_SUCCESS)
831 		return (EINVAL);
832 	if (entry->start != kva || entry->end != kva + size) {
833 		vm_map_lookup_done(map, entry);
834 		return (EINVAL);
835 	}
836 	vm_map_lookup_done(map, entry);
837 	if (obj != shmfd->shm_object)
838 		return (EINVAL);
839 	vm_map_remove(map, kva, kva + size);
840 	VM_OBJECT_WLOCK(obj);
841 	KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
842 	shmfd->shm_kmappings--;
843 	VM_OBJECT_WUNLOCK(obj);
844 	return (0);
845 }
846 
847 void
848 shm_path(struct shmfd *shmfd, char *path, size_t size)
849 {
850 
851 	if (shmfd->shm_path == NULL)
852 		return;
853 	sx_slock(&shm_dict_lock);
854 	if (shmfd->shm_path != NULL)
855 		strlcpy(path, shmfd->shm_path, size);
856 	sx_sunlock(&shm_dict_lock);
857 }
858