xref: /freebsd/sys/kern/uipc_shm.c (revision 0cdfe2ae89834a1a4b3468bfac870941ee17f2d5)
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 (vm_page_sleep_if_busy(m, "shmtrc"))
285 					goto retry;
286 			} else if (vm_pager_has_page(object, idx, NULL, NULL)) {
287 				m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL);
288 				if (m == NULL) {
289 					VM_OBJECT_WUNLOCK(object);
290 					VM_WAIT;
291 					VM_OBJECT_WLOCK(object);
292 					goto retry;
293 				} else if (m->valid != VM_PAGE_BITS_ALL) {
294 					ma[0] = m;
295 					rv = vm_pager_get_pages(object, ma, 1,
296 					    0);
297 					m = vm_page_lookup(object, idx);
298 				} else
299 					/* A cached page was reactivated. */
300 					rv = VM_PAGER_OK;
301 				vm_page_lock(m);
302 				if (rv == VM_PAGER_OK) {
303 					vm_page_deactivate(m);
304 					vm_page_unlock(m);
305 					vm_page_xunbusy(m);
306 				} else {
307 					vm_page_free(m);
308 					vm_page_unlock(m);
309 					VM_OBJECT_WUNLOCK(object);
310 					return (EIO);
311 				}
312 			}
313 			if (m != NULL) {
314 				pmap_zero_page_area(m, base, PAGE_SIZE - base);
315 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
316 				    ("shm_dotruncate: page %p is invalid", m));
317 				vm_page_dirty(m);
318 				vm_pager_page_unswapped(m);
319 			}
320 		}
321 		delta = ptoa(object->size - nobjsize);
322 
323 		/* Toss in memory pages. */
324 		if (nobjsize < object->size)
325 			vm_object_page_remove(object, nobjsize, object->size,
326 			    0);
327 
328 		/* Toss pages from swap. */
329 		if (object->type == OBJT_SWAP)
330 			swap_pager_freespace(object, nobjsize, delta);
331 
332 		/* Free the swap accounted for shm */
333 		swap_release_by_cred(delta, object->cred);
334 		object->charge -= delta;
335 	} else {
336 		/* Attempt to reserve the swap */
337 		delta = ptoa(nobjsize - object->size);
338 		if (!swap_reserve_by_cred(delta, object->cred)) {
339 			VM_OBJECT_WUNLOCK(object);
340 			return (ENOMEM);
341 		}
342 		object->charge += delta;
343 	}
344 	shmfd->shm_size = length;
345 	mtx_lock(&shm_timestamp_lock);
346 	vfs_timestamp(&shmfd->shm_ctime);
347 	shmfd->shm_mtime = shmfd->shm_ctime;
348 	mtx_unlock(&shm_timestamp_lock);
349 	object->size = nobjsize;
350 	VM_OBJECT_WUNLOCK(object);
351 	return (0);
352 }
353 
354 /*
355  * shmfd object management including creation and reference counting
356  * routines.
357  */
358 static struct shmfd *
359 shm_alloc(struct ucred *ucred, mode_t mode)
360 {
361 	struct shmfd *shmfd;
362 
363 	shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
364 	shmfd->shm_size = 0;
365 	shmfd->shm_uid = ucred->cr_uid;
366 	shmfd->shm_gid = ucred->cr_gid;
367 	shmfd->shm_mode = mode;
368 	shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
369 	    shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
370 	KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
371 	VM_OBJECT_WLOCK(shmfd->shm_object);
372 	vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
373 	vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
374 	VM_OBJECT_WUNLOCK(shmfd->shm_object);
375 	vfs_timestamp(&shmfd->shm_birthtime);
376 	shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
377 	    shmfd->shm_birthtime;
378 	refcount_init(&shmfd->shm_refs, 1);
379 #ifdef MAC
380 	mac_posixshm_init(shmfd);
381 	mac_posixshm_create(ucred, shmfd);
382 #endif
383 
384 	return (shmfd);
385 }
386 
387 static struct shmfd *
388 shm_hold(struct shmfd *shmfd)
389 {
390 
391 	refcount_acquire(&shmfd->shm_refs);
392 	return (shmfd);
393 }
394 
395 static void
396 shm_drop(struct shmfd *shmfd)
397 {
398 
399 	if (refcount_release(&shmfd->shm_refs)) {
400 #ifdef MAC
401 		mac_posixshm_destroy(shmfd);
402 #endif
403 		vm_object_deallocate(shmfd->shm_object);
404 		free(shmfd, M_SHMFD);
405 	}
406 }
407 
408 /*
409  * Determine if the credentials have sufficient permissions for a
410  * specified combination of FREAD and FWRITE.
411  */
412 static int
413 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
414 {
415 	accmode_t accmode;
416 	int error;
417 
418 	accmode = 0;
419 	if (flags & FREAD)
420 		accmode |= VREAD;
421 	if (flags & FWRITE)
422 		accmode |= VWRITE;
423 	mtx_lock(&shm_timestamp_lock);
424 	error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
425 	    accmode, ucred, NULL);
426 	mtx_unlock(&shm_timestamp_lock);
427 	return (error);
428 }
429 
430 /*
431  * Dictionary management.  We maintain an in-kernel dictionary to map
432  * paths to shmfd objects.  We use the FNV hash on the path to store
433  * the mappings in a hash table.
434  */
435 static void
436 shm_dict_init(void *arg)
437 {
438 
439 	mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
440 	sx_init(&shm_dict_lock, "shm dictionary");
441 	shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
442 }
443 SYSINIT(shm_dict_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_dict_init, NULL);
444 
445 static struct shmfd *
446 shm_lookup(char *path, Fnv32_t fnv)
447 {
448 	struct shm_mapping *map;
449 
450 	LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
451 		if (map->sm_fnv != fnv)
452 			continue;
453 		if (strcmp(map->sm_path, path) == 0)
454 			return (map->sm_shmfd);
455 	}
456 
457 	return (NULL);
458 }
459 
460 static void
461 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
462 {
463 	struct shm_mapping *map;
464 
465 	map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
466 	map->sm_path = path;
467 	map->sm_fnv = fnv;
468 	map->sm_shmfd = shm_hold(shmfd);
469 	shmfd->shm_path = path;
470 	LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
471 }
472 
473 static int
474 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
475 {
476 	struct shm_mapping *map;
477 	int error;
478 
479 	LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
480 		if (map->sm_fnv != fnv)
481 			continue;
482 		if (strcmp(map->sm_path, path) == 0) {
483 #ifdef MAC
484 			error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
485 			if (error)
486 				return (error);
487 #endif
488 			error = shm_access(map->sm_shmfd, ucred,
489 			    FREAD | FWRITE);
490 			if (error)
491 				return (error);
492 			map->sm_shmfd->shm_path = NULL;
493 			LIST_REMOVE(map, sm_link);
494 			shm_drop(map->sm_shmfd);
495 			free(map->sm_path, M_SHMFD);
496 			free(map, M_SHMFD);
497 			return (0);
498 		}
499 	}
500 
501 	return (ENOENT);
502 }
503 
504 /* System calls. */
505 int
506 sys_shm_open(struct thread *td, struct shm_open_args *uap)
507 {
508 	struct filedesc *fdp;
509 	struct shmfd *shmfd;
510 	struct file *fp;
511 	char *path;
512 	Fnv32_t fnv;
513 	mode_t cmode;
514 	int fd, error;
515 
516 #ifdef CAPABILITY_MODE
517 	/*
518 	 * shm_open(2) is only allowed for anonymous objects.
519 	 */
520 	if (IN_CAPABILITY_MODE(td) && (uap->path != SHM_ANON))
521 		return (ECAPMODE);
522 #endif
523 
524 	if ((uap->flags & O_ACCMODE) != O_RDONLY &&
525 	    (uap->flags & O_ACCMODE) != O_RDWR)
526 		return (EINVAL);
527 
528 	if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0)
529 		return (EINVAL);
530 
531 	fdp = td->td_proc->p_fd;
532 	cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
533 
534 	error = falloc(td, &fp, &fd, O_CLOEXEC);
535 	if (error)
536 		return (error);
537 
538 	/* A SHM_ANON path pointer creates an anonymous object. */
539 	if (uap->path == SHM_ANON) {
540 		/* A read-only anonymous object is pointless. */
541 		if ((uap->flags & O_ACCMODE) == O_RDONLY) {
542 			fdclose(fdp, fp, fd, td);
543 			fdrop(fp, td);
544 			return (EINVAL);
545 		}
546 		shmfd = shm_alloc(td->td_ucred, cmode);
547 	} else {
548 		path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
549 		error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
550 
551 		/* Require paths to start with a '/' character. */
552 		if (error == 0 && path[0] != '/')
553 			error = EINVAL;
554 		if (error) {
555 			fdclose(fdp, fp, fd, td);
556 			fdrop(fp, td);
557 			free(path, M_SHMFD);
558 			return (error);
559 		}
560 
561 		fnv = fnv_32_str(path, FNV1_32_INIT);
562 		sx_xlock(&shm_dict_lock);
563 		shmfd = shm_lookup(path, fnv);
564 		if (shmfd == NULL) {
565 			/* Object does not yet exist, create it if requested. */
566 			if (uap->flags & O_CREAT) {
567 #ifdef MAC
568 				error = mac_posixshm_check_create(td->td_ucred,
569 				    path);
570 				if (error == 0) {
571 #endif
572 					shmfd = shm_alloc(td->td_ucred, cmode);
573 					shm_insert(path, fnv, shmfd);
574 #ifdef MAC
575 				}
576 #endif
577 			} else {
578 				free(path, M_SHMFD);
579 				error = ENOENT;
580 			}
581 		} else {
582 			/*
583 			 * Object already exists, obtain a new
584 			 * reference if requested and permitted.
585 			 */
586 			free(path, M_SHMFD);
587 			if ((uap->flags & (O_CREAT | O_EXCL)) ==
588 			    (O_CREAT | O_EXCL))
589 				error = EEXIST;
590 			else {
591 #ifdef MAC
592 				error = mac_posixshm_check_open(td->td_ucred,
593 				    shmfd, FFLAGS(uap->flags & O_ACCMODE));
594 				if (error == 0)
595 #endif
596 				error = shm_access(shmfd, td->td_ucred,
597 				    FFLAGS(uap->flags & O_ACCMODE));
598 			}
599 
600 			/*
601 			 * Truncate the file back to zero length if
602 			 * O_TRUNC was specified and the object was
603 			 * opened with read/write.
604 			 */
605 			if (error == 0 &&
606 			    (uap->flags & (O_ACCMODE | O_TRUNC)) ==
607 			    (O_RDWR | O_TRUNC)) {
608 #ifdef MAC
609 				error = mac_posixshm_check_truncate(
610 					td->td_ucred, fp->f_cred, shmfd);
611 				if (error == 0)
612 #endif
613 					shm_dotruncate(shmfd, 0);
614 			}
615 			if (error == 0)
616 				shm_hold(shmfd);
617 		}
618 		sx_xunlock(&shm_dict_lock);
619 
620 		if (error) {
621 			fdclose(fdp, fp, fd, td);
622 			fdrop(fp, td);
623 			return (error);
624 		}
625 	}
626 
627 	finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
628 
629 	td->td_retval[0] = fd;
630 	fdrop(fp, td);
631 
632 	return (0);
633 }
634 
635 int
636 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
637 {
638 	char *path;
639 	Fnv32_t fnv;
640 	int error;
641 
642 	path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
643 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
644 	if (error) {
645 		free(path, M_TEMP);
646 		return (error);
647 	}
648 
649 	fnv = fnv_32_str(path, FNV1_32_INIT);
650 	sx_xlock(&shm_dict_lock);
651 	error = shm_remove(path, fnv, td->td_ucred);
652 	sx_xunlock(&shm_dict_lock);
653 	free(path, M_TEMP);
654 
655 	return (error);
656 }
657 
658 /*
659  * mmap() helper to validate mmap() requests against shm object state
660  * and give mmap() the vm_object to use for the mapping.
661  */
662 int
663 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
664     vm_object_t *obj)
665 {
666 
667 	/*
668 	 * XXXRW: This validation is probably insufficient, and subject to
669 	 * sign errors.  It should be fixed.
670 	 */
671 	if (foff >= shmfd->shm_size ||
672 	    foff + objsize > round_page(shmfd->shm_size))
673 		return (EINVAL);
674 
675 	mtx_lock(&shm_timestamp_lock);
676 	vfs_timestamp(&shmfd->shm_atime);
677 	mtx_unlock(&shm_timestamp_lock);
678 	vm_object_reference(shmfd->shm_object);
679 	*obj = shmfd->shm_object;
680 	return (0);
681 }
682 
683 static int
684 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
685     struct thread *td)
686 {
687 	struct shmfd *shmfd;
688 	int error;
689 
690 	error = 0;
691 	shmfd = fp->f_data;
692 	mtx_lock(&shm_timestamp_lock);
693 	/*
694 	 * SUSv4 says that x bits of permission need not be affected.
695 	 * Be consistent with our shm_open there.
696 	 */
697 #ifdef MAC
698 	error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
699 	if (error != 0)
700 		goto out;
701 #endif
702 	error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
703 	    shmfd->shm_gid, VADMIN, active_cred, NULL);
704 	if (error != 0)
705 		goto out;
706 	shmfd->shm_mode = mode & ACCESSPERMS;
707 out:
708 	mtx_unlock(&shm_timestamp_lock);
709 	return (error);
710 }
711 
712 static int
713 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
714     struct thread *td)
715 {
716 	struct shmfd *shmfd;
717 	int error;
718 
719 	error = 0;
720 	shmfd = fp->f_data;
721 	mtx_lock(&shm_timestamp_lock);
722 #ifdef MAC
723 	error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
724 	if (error != 0)
725 		goto out;
726 #endif
727 	if (uid == (uid_t)-1)
728 		uid = shmfd->shm_uid;
729 	if (gid == (gid_t)-1)
730                  gid = shmfd->shm_gid;
731 	if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
732 	    (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
733 	    (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
734 		goto out;
735 	shmfd->shm_uid = uid;
736 	shmfd->shm_gid = gid;
737 out:
738 	mtx_unlock(&shm_timestamp_lock);
739 	return (error);
740 }
741 
742 /*
743  * Helper routines to allow the backing object of a shared memory file
744  * descriptor to be mapped in the kernel.
745  */
746 int
747 shm_map(struct file *fp, size_t size, off_t offset, void **memp)
748 {
749 	struct shmfd *shmfd;
750 	vm_offset_t kva, ofs;
751 	vm_object_t obj;
752 	int rv;
753 
754 	if (fp->f_type != DTYPE_SHM)
755 		return (EINVAL);
756 	shmfd = fp->f_data;
757 	obj = shmfd->shm_object;
758 	VM_OBJECT_WLOCK(obj);
759 	/*
760 	 * XXXRW: This validation is probably insufficient, and subject to
761 	 * sign errors.  It should be fixed.
762 	 */
763 	if (offset >= shmfd->shm_size ||
764 	    offset + size > round_page(shmfd->shm_size)) {
765 		VM_OBJECT_WUNLOCK(obj);
766 		return (EINVAL);
767 	}
768 
769 	shmfd->shm_kmappings++;
770 	vm_object_reference_locked(obj);
771 	VM_OBJECT_WUNLOCK(obj);
772 
773 	/* Map the object into the kernel_map and wire it. */
774 	kva = vm_map_min(kernel_map);
775 	ofs = offset & PAGE_MASK;
776 	offset = trunc_page(offset);
777 	size = round_page(size + ofs);
778 	rv = vm_map_find(kernel_map, obj, offset, &kva, size,
779 	    VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE,
780 	    VM_PROT_READ | VM_PROT_WRITE, 0);
781 	if (rv == KERN_SUCCESS) {
782 		rv = vm_map_wire(kernel_map, kva, kva + size,
783 		    VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
784 		if (rv == KERN_SUCCESS) {
785 			*memp = (void *)(kva + ofs);
786 			return (0);
787 		}
788 		vm_map_remove(kernel_map, kva, kva + size);
789 	} else
790 		vm_object_deallocate(obj);
791 
792 	/* On failure, drop our mapping reference. */
793 	VM_OBJECT_WLOCK(obj);
794 	shmfd->shm_kmappings--;
795 	VM_OBJECT_WUNLOCK(obj);
796 
797 	return (vm_mmap_to_errno(rv));
798 }
799 
800 /*
801  * We require the caller to unmap the entire entry.  This allows us to
802  * safely decrement shm_kmappings when a mapping is removed.
803  */
804 int
805 shm_unmap(struct file *fp, void *mem, size_t size)
806 {
807 	struct shmfd *shmfd;
808 	vm_map_entry_t entry;
809 	vm_offset_t kva, ofs;
810 	vm_object_t obj;
811 	vm_pindex_t pindex;
812 	vm_prot_t prot;
813 	boolean_t wired;
814 	vm_map_t map;
815 	int rv;
816 
817 	if (fp->f_type != DTYPE_SHM)
818 		return (EINVAL);
819 	shmfd = fp->f_data;
820 	kva = (vm_offset_t)mem;
821 	ofs = kva & PAGE_MASK;
822 	kva = trunc_page(kva);
823 	size = round_page(size + ofs);
824 	map = kernel_map;
825 	rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
826 	    &obj, &pindex, &prot, &wired);
827 	if (rv != KERN_SUCCESS)
828 		return (EINVAL);
829 	if (entry->start != kva || entry->end != kva + size) {
830 		vm_map_lookup_done(map, entry);
831 		return (EINVAL);
832 	}
833 	vm_map_lookup_done(map, entry);
834 	if (obj != shmfd->shm_object)
835 		return (EINVAL);
836 	vm_map_remove(map, kva, kva + size);
837 	VM_OBJECT_WLOCK(obj);
838 	KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
839 	shmfd->shm_kmappings--;
840 	VM_OBJECT_WUNLOCK(obj);
841 	return (0);
842 }
843 
844 void
845 shm_path(struct shmfd *shmfd, char *path, size_t size)
846 {
847 
848 	if (shmfd->shm_path == NULL)
849 		return;
850 	sx_slock(&shm_dict_lock);
851 	if (shmfd->shm_path != NULL)
852 		strlcpy(path, shmfd->shm_path, size);
853 	sx_sunlock(&shm_dict_lock);
854 }
855