xref: /freebsd/sys/kern/uipc_shm.c (revision 38f0b757fd84d17d0fc24739a7cda160c4516d81)
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/capsicum.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 #include <sys/unistd.h>
73 
74 #include <security/mac/mac_framework.h>
75 
76 #include <vm/vm.h>
77 #include <vm/vm_param.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_kern.h>
82 #include <vm/vm_object.h>
83 #include <vm/vm_page.h>
84 #include <vm/vm_pageout.h>
85 #include <vm/vm_pager.h>
86 #include <vm/swap_pager.h>
87 
88 struct shm_mapping {
89 	char		*sm_path;
90 	Fnv32_t		sm_fnv;
91 	struct shmfd	*sm_shmfd;
92 	LIST_ENTRY(shm_mapping) sm_link;
93 };
94 
95 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
96 static LIST_HEAD(, shm_mapping) *shm_dictionary;
97 static struct sx shm_dict_lock;
98 static struct mtx shm_timestamp_lock;
99 static u_long shm_hash;
100 
101 #define	SHM_HASH(fnv)	(&shm_dictionary[(fnv) & shm_hash])
102 
103 static int	shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags);
104 static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode);
105 static void	shm_dict_init(void *arg);
106 static void	shm_drop(struct shmfd *shmfd);
107 static struct shmfd *shm_hold(struct shmfd *shmfd);
108 static void	shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
109 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
110 static int	shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
111 static int	shm_dotruncate(struct shmfd *shmfd, off_t length);
112 
113 static fo_rdwr_t	shm_read;
114 static fo_rdwr_t	shm_write;
115 static fo_truncate_t	shm_truncate;
116 static fo_ioctl_t	shm_ioctl;
117 static fo_poll_t	shm_poll;
118 static fo_kqfilter_t	shm_kqfilter;
119 static fo_stat_t	shm_stat;
120 static fo_close_t	shm_close;
121 static fo_chmod_t	shm_chmod;
122 static fo_chown_t	shm_chown;
123 static fo_seek_t	shm_seek;
124 
125 /* File descriptor operations. */
126 static struct fileops shm_ops = {
127 	.fo_read = shm_read,
128 	.fo_write = shm_write,
129 	.fo_truncate = shm_truncate,
130 	.fo_ioctl = shm_ioctl,
131 	.fo_poll = shm_poll,
132 	.fo_kqfilter = shm_kqfilter,
133 	.fo_stat = shm_stat,
134 	.fo_close = shm_close,
135 	.fo_chmod = shm_chmod,
136 	.fo_chown = shm_chown,
137 	.fo_sendfile = vn_sendfile,
138 	.fo_seek = shm_seek,
139 	.fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
140 };
141 
142 FEATURE(posix_shm, "POSIX shared memory");
143 
144 static int
145 uiomove_object_page(vm_object_t obj, size_t len, struct uio *uio)
146 {
147 	vm_page_t m;
148 	vm_pindex_t idx;
149 	size_t tlen;
150 	int error, offset, rv;
151 
152 	idx = OFF_TO_IDX(uio->uio_offset);
153 	offset = uio->uio_offset & PAGE_MASK;
154 	tlen = MIN(PAGE_SIZE - offset, len);
155 
156 	VM_OBJECT_WLOCK(obj);
157 
158 	/*
159 	 * Parallel reads of the page content from disk are prevented
160 	 * by exclusive busy.
161 	 *
162 	 * Although the tmpfs vnode lock is held here, it is
163 	 * nonetheless safe to sleep waiting for a free page.  The
164 	 * pageout daemon does not need to acquire the tmpfs vnode
165 	 * lock to page out tobj's pages because tobj is a OBJT_SWAP
166 	 * type object.
167 	 */
168 	m = vm_page_grab(obj, idx, VM_ALLOC_NORMAL);
169 	if (m->valid != VM_PAGE_BITS_ALL) {
170 		if (vm_pager_has_page(obj, idx, NULL, NULL)) {
171 			rv = vm_pager_get_pages(obj, &m, 1, 0);
172 			m = vm_page_lookup(obj, idx);
173 			if (m == NULL) {
174 				printf(
175 		    "uiomove_object: vm_obj %p idx %jd null lookup rv %d\n",
176 				    obj, idx, rv);
177 				VM_OBJECT_WUNLOCK(obj);
178 				return (EIO);
179 			}
180 			if (rv != VM_PAGER_OK) {
181 				printf(
182 	    "uiomove_object: vm_obj %p idx %jd valid %x pager error %d\n",
183 				    obj, idx, m->valid, rv);
184 				vm_page_lock(m);
185 				vm_page_free(m);
186 				vm_page_unlock(m);
187 				VM_OBJECT_WUNLOCK(obj);
188 				return (EIO);
189 			}
190 		} else
191 			vm_page_zero_invalid(m, TRUE);
192 	}
193 	vm_page_xunbusy(m);
194 	vm_page_lock(m);
195 	vm_page_hold(m);
196 	vm_page_unlock(m);
197 	VM_OBJECT_WUNLOCK(obj);
198 	error = uiomove_fromphys(&m, offset, tlen, uio);
199 	if (uio->uio_rw == UIO_WRITE && error == 0) {
200 		VM_OBJECT_WLOCK(obj);
201 		vm_page_dirty(m);
202 		VM_OBJECT_WUNLOCK(obj);
203 	}
204 	vm_page_lock(m);
205 	vm_page_unhold(m);
206 	if (m->queue == PQ_NONE) {
207 		vm_page_deactivate(m);
208 	} else {
209 		/* Requeue to maintain LRU ordering. */
210 		vm_page_requeue(m);
211 	}
212 	vm_page_unlock(m);
213 
214 	return (error);
215 }
216 
217 int
218 uiomove_object(vm_object_t obj, off_t obj_size, struct uio *uio)
219 {
220 	ssize_t resid;
221 	size_t len;
222 	int error;
223 
224 	error = 0;
225 	while ((resid = uio->uio_resid) > 0) {
226 		if (obj_size <= uio->uio_offset)
227 			break;
228 		len = MIN(obj_size - uio->uio_offset, resid);
229 		if (len == 0)
230 			break;
231 		error = uiomove_object_page(obj, len, uio);
232 		if (error != 0 || resid == uio->uio_resid)
233 			break;
234 	}
235 	return (error);
236 }
237 
238 static int
239 shm_seek(struct file *fp, off_t offset, int whence, struct thread *td)
240 {
241 	struct shmfd *shmfd;
242 	off_t foffset;
243 	int error;
244 
245 	shmfd = fp->f_data;
246 	foffset = foffset_lock(fp, 0);
247 	error = 0;
248 	switch (whence) {
249 	case L_INCR:
250 		if (foffset < 0 ||
251 		    (offset > 0 && foffset > OFF_MAX - offset)) {
252 			error = EOVERFLOW;
253 			break;
254 		}
255 		offset += foffset;
256 		break;
257 	case L_XTND:
258 		if (offset > 0 && shmfd->shm_size > OFF_MAX - offset) {
259 			error = EOVERFLOW;
260 			break;
261 		}
262 		offset += shmfd->shm_size;
263 		break;
264 	case L_SET:
265 		break;
266 	default:
267 		error = EINVAL;
268 	}
269 	if (error == 0) {
270 		if (offset < 0 || offset > shmfd->shm_size)
271 			error = EINVAL;
272 		else
273 			td->td_uretoff.tdu_off = offset;
274 	}
275 	foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
276 	return (error);
277 }
278 
279 static int
280 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
281     int flags, struct thread *td)
282 {
283 	struct shmfd *shmfd;
284 	void *rl_cookie;
285 	int error;
286 
287 	shmfd = fp->f_data;
288 	foffset_lock_uio(fp, uio, flags);
289 	rl_cookie = rangelock_rlock(&shmfd->shm_rl, uio->uio_offset,
290 	    uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx);
291 #ifdef MAC
292 	error = mac_posixshm_check_read(active_cred, fp->f_cred, shmfd);
293 	if (error)
294 		return (error);
295 #endif
296 	error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio);
297 	rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx);
298 	foffset_unlock_uio(fp, uio, flags);
299 	return (error);
300 }
301 
302 static int
303 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
304     int flags, struct thread *td)
305 {
306 	struct shmfd *shmfd;
307 	void *rl_cookie;
308 	int error;
309 
310 	shmfd = fp->f_data;
311 #ifdef MAC
312 	error = mac_posixshm_check_write(active_cred, fp->f_cred, shmfd);
313 	if (error)
314 		return (error);
315 #endif
316 	foffset_lock_uio(fp, uio, flags);
317 	if ((flags & FOF_OFFSET) == 0) {
318 		rl_cookie = rangelock_wlock(&shmfd->shm_rl, 0, OFF_MAX,
319 		    &shmfd->shm_mtx);
320 	} else {
321 		rl_cookie = rangelock_wlock(&shmfd->shm_rl, uio->uio_offset,
322 		    uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx);
323 	}
324 
325 	error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio);
326 	rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx);
327 	foffset_unlock_uio(fp, uio, flags);
328 	return (error);
329 }
330 
331 static int
332 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
333     struct thread *td)
334 {
335 	struct shmfd *shmfd;
336 #ifdef MAC
337 	int error;
338 #endif
339 
340 	shmfd = fp->f_data;
341 #ifdef MAC
342 	error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
343 	if (error)
344 		return (error);
345 #endif
346 	return (shm_dotruncate(shmfd, length));
347 }
348 
349 static int
350 shm_ioctl(struct file *fp, u_long com, void *data,
351     struct ucred *active_cred, struct thread *td)
352 {
353 
354 	return (EOPNOTSUPP);
355 }
356 
357 static int
358 shm_poll(struct file *fp, int events, struct ucred *active_cred,
359     struct thread *td)
360 {
361 
362 	return (EOPNOTSUPP);
363 }
364 
365 static int
366 shm_kqfilter(struct file *fp, struct knote *kn)
367 {
368 
369 	return (EOPNOTSUPP);
370 }
371 
372 static int
373 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
374     struct thread *td)
375 {
376 	struct shmfd *shmfd;
377 #ifdef MAC
378 	int error;
379 #endif
380 
381 	shmfd = fp->f_data;
382 
383 #ifdef MAC
384 	error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
385 	if (error)
386 		return (error);
387 #endif
388 
389 	/*
390 	 * Attempt to return sanish values for fstat() on a memory file
391 	 * descriptor.
392 	 */
393 	bzero(sb, sizeof(*sb));
394 	sb->st_blksize = PAGE_SIZE;
395 	sb->st_size = shmfd->shm_size;
396 	sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize;
397 	mtx_lock(&shm_timestamp_lock);
398 	sb->st_atim = shmfd->shm_atime;
399 	sb->st_ctim = shmfd->shm_ctime;
400 	sb->st_mtim = shmfd->shm_mtime;
401 	sb->st_birthtim = shmfd->shm_birthtime;
402 	sb->st_mode = S_IFREG | shmfd->shm_mode;		/* XXX */
403 	sb->st_uid = shmfd->shm_uid;
404 	sb->st_gid = shmfd->shm_gid;
405 	mtx_unlock(&shm_timestamp_lock);
406 
407 	return (0);
408 }
409 
410 static int
411 shm_close(struct file *fp, struct thread *td)
412 {
413 	struct shmfd *shmfd;
414 
415 	shmfd = fp->f_data;
416 	fp->f_data = NULL;
417 	shm_drop(shmfd);
418 
419 	return (0);
420 }
421 
422 static int
423 shm_dotruncate(struct shmfd *shmfd, off_t length)
424 {
425 	vm_object_t object;
426 	vm_page_t m, ma[1];
427 	vm_pindex_t idx, nobjsize;
428 	vm_ooffset_t delta;
429 	int base, rv;
430 
431 	object = shmfd->shm_object;
432 	VM_OBJECT_WLOCK(object);
433 	if (length == shmfd->shm_size) {
434 		VM_OBJECT_WUNLOCK(object);
435 		return (0);
436 	}
437 	nobjsize = OFF_TO_IDX(length + PAGE_MASK);
438 
439 	/* Are we shrinking?  If so, trim the end. */
440 	if (length < shmfd->shm_size) {
441 		/*
442 		 * Disallow any requests to shrink the size if this
443 		 * object is mapped into the kernel.
444 		 */
445 		if (shmfd->shm_kmappings > 0) {
446 			VM_OBJECT_WUNLOCK(object);
447 			return (EBUSY);
448 		}
449 
450 		/*
451 		 * Zero the truncated part of the last page.
452 		 */
453 		base = length & PAGE_MASK;
454 		if (base != 0) {
455 			idx = OFF_TO_IDX(length);
456 retry:
457 			m = vm_page_lookup(object, idx);
458 			if (m != NULL) {
459 				if (vm_page_sleep_if_busy(m, "shmtrc"))
460 					goto retry;
461 			} else if (vm_pager_has_page(object, idx, NULL, NULL)) {
462 				m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL);
463 				if (m == NULL) {
464 					VM_OBJECT_WUNLOCK(object);
465 					VM_WAIT;
466 					VM_OBJECT_WLOCK(object);
467 					goto retry;
468 				} else if (m->valid != VM_PAGE_BITS_ALL) {
469 					ma[0] = m;
470 					rv = vm_pager_get_pages(object, ma, 1,
471 					    0);
472 					m = vm_page_lookup(object, idx);
473 				} else
474 					/* A cached page was reactivated. */
475 					rv = VM_PAGER_OK;
476 				vm_page_lock(m);
477 				if (rv == VM_PAGER_OK) {
478 					vm_page_deactivate(m);
479 					vm_page_unlock(m);
480 					vm_page_xunbusy(m);
481 				} else {
482 					vm_page_free(m);
483 					vm_page_unlock(m);
484 					VM_OBJECT_WUNLOCK(object);
485 					return (EIO);
486 				}
487 			}
488 			if (m != NULL) {
489 				pmap_zero_page_area(m, base, PAGE_SIZE - base);
490 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
491 				    ("shm_dotruncate: page %p is invalid", m));
492 				vm_page_dirty(m);
493 				vm_pager_page_unswapped(m);
494 			}
495 		}
496 		delta = ptoa(object->size - nobjsize);
497 
498 		/* Toss in memory pages. */
499 		if (nobjsize < object->size)
500 			vm_object_page_remove(object, nobjsize, object->size,
501 			    0);
502 
503 		/* Toss pages from swap. */
504 		if (object->type == OBJT_SWAP)
505 			swap_pager_freespace(object, nobjsize, delta);
506 
507 		/* Free the swap accounted for shm */
508 		swap_release_by_cred(delta, object->cred);
509 		object->charge -= delta;
510 	} else {
511 		/* Attempt to reserve the swap */
512 		delta = ptoa(nobjsize - object->size);
513 		if (!swap_reserve_by_cred(delta, object->cred)) {
514 			VM_OBJECT_WUNLOCK(object);
515 			return (ENOMEM);
516 		}
517 		object->charge += delta;
518 	}
519 	shmfd->shm_size = length;
520 	mtx_lock(&shm_timestamp_lock);
521 	vfs_timestamp(&shmfd->shm_ctime);
522 	shmfd->shm_mtime = shmfd->shm_ctime;
523 	mtx_unlock(&shm_timestamp_lock);
524 	object->size = nobjsize;
525 	VM_OBJECT_WUNLOCK(object);
526 	return (0);
527 }
528 
529 /*
530  * shmfd object management including creation and reference counting
531  * routines.
532  */
533 static struct shmfd *
534 shm_alloc(struct ucred *ucred, mode_t mode)
535 {
536 	struct shmfd *shmfd;
537 
538 	shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
539 	shmfd->shm_size = 0;
540 	shmfd->shm_uid = ucred->cr_uid;
541 	shmfd->shm_gid = ucred->cr_gid;
542 	shmfd->shm_mode = mode;
543 	shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
544 	    shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
545 	KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
546 	VM_OBJECT_WLOCK(shmfd->shm_object);
547 	vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
548 	vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
549 	VM_OBJECT_WUNLOCK(shmfd->shm_object);
550 	vfs_timestamp(&shmfd->shm_birthtime);
551 	shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
552 	    shmfd->shm_birthtime;
553 	refcount_init(&shmfd->shm_refs, 1);
554 	mtx_init(&shmfd->shm_mtx, "shmrl", NULL, MTX_DEF);
555 	rangelock_init(&shmfd->shm_rl);
556 #ifdef MAC
557 	mac_posixshm_init(shmfd);
558 	mac_posixshm_create(ucred, shmfd);
559 #endif
560 
561 	return (shmfd);
562 }
563 
564 static struct shmfd *
565 shm_hold(struct shmfd *shmfd)
566 {
567 
568 	refcount_acquire(&shmfd->shm_refs);
569 	return (shmfd);
570 }
571 
572 static void
573 shm_drop(struct shmfd *shmfd)
574 {
575 
576 	if (refcount_release(&shmfd->shm_refs)) {
577 #ifdef MAC
578 		mac_posixshm_destroy(shmfd);
579 #endif
580 		rangelock_destroy(&shmfd->shm_rl);
581 		mtx_destroy(&shmfd->shm_mtx);
582 		vm_object_deallocate(shmfd->shm_object);
583 		free(shmfd, M_SHMFD);
584 	}
585 }
586 
587 /*
588  * Determine if the credentials have sufficient permissions for a
589  * specified combination of FREAD and FWRITE.
590  */
591 static int
592 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
593 {
594 	accmode_t accmode;
595 	int error;
596 
597 	accmode = 0;
598 	if (flags & FREAD)
599 		accmode |= VREAD;
600 	if (flags & FWRITE)
601 		accmode |= VWRITE;
602 	mtx_lock(&shm_timestamp_lock);
603 	error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
604 	    accmode, ucred, NULL);
605 	mtx_unlock(&shm_timestamp_lock);
606 	return (error);
607 }
608 
609 /*
610  * Dictionary management.  We maintain an in-kernel dictionary to map
611  * paths to shmfd objects.  We use the FNV hash on the path to store
612  * the mappings in a hash table.
613  */
614 static void
615 shm_dict_init(void *arg)
616 {
617 
618 	mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
619 	sx_init(&shm_dict_lock, "shm dictionary");
620 	shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
621 }
622 SYSINIT(shm_dict_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_dict_init, NULL);
623 
624 static struct shmfd *
625 shm_lookup(char *path, Fnv32_t fnv)
626 {
627 	struct shm_mapping *map;
628 
629 	LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
630 		if (map->sm_fnv != fnv)
631 			continue;
632 		if (strcmp(map->sm_path, path) == 0)
633 			return (map->sm_shmfd);
634 	}
635 
636 	return (NULL);
637 }
638 
639 static void
640 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
641 {
642 	struct shm_mapping *map;
643 
644 	map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
645 	map->sm_path = path;
646 	map->sm_fnv = fnv;
647 	map->sm_shmfd = shm_hold(shmfd);
648 	shmfd->shm_path = path;
649 	LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
650 }
651 
652 static int
653 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
654 {
655 	struct shm_mapping *map;
656 	int error;
657 
658 	LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
659 		if (map->sm_fnv != fnv)
660 			continue;
661 		if (strcmp(map->sm_path, path) == 0) {
662 #ifdef MAC
663 			error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
664 			if (error)
665 				return (error);
666 #endif
667 			error = shm_access(map->sm_shmfd, ucred,
668 			    FREAD | FWRITE);
669 			if (error)
670 				return (error);
671 			map->sm_shmfd->shm_path = NULL;
672 			LIST_REMOVE(map, sm_link);
673 			shm_drop(map->sm_shmfd);
674 			free(map->sm_path, M_SHMFD);
675 			free(map, M_SHMFD);
676 			return (0);
677 		}
678 	}
679 
680 	return (ENOENT);
681 }
682 
683 /* System calls. */
684 int
685 sys_shm_open(struct thread *td, struct shm_open_args *uap)
686 {
687 	struct filedesc *fdp;
688 	struct shmfd *shmfd;
689 	struct file *fp;
690 	char *path;
691 	Fnv32_t fnv;
692 	mode_t cmode;
693 	int fd, error;
694 
695 #ifdef CAPABILITY_MODE
696 	/*
697 	 * shm_open(2) is only allowed for anonymous objects.
698 	 */
699 	if (IN_CAPABILITY_MODE(td) && (uap->path != SHM_ANON))
700 		return (ECAPMODE);
701 #endif
702 
703 	if ((uap->flags & O_ACCMODE) != O_RDONLY &&
704 	    (uap->flags & O_ACCMODE) != O_RDWR)
705 		return (EINVAL);
706 
707 	if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC | O_CLOEXEC)) != 0)
708 		return (EINVAL);
709 
710 	fdp = td->td_proc->p_fd;
711 	cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
712 
713 	error = falloc(td, &fp, &fd, O_CLOEXEC);
714 	if (error)
715 		return (error);
716 
717 	/* A SHM_ANON path pointer creates an anonymous object. */
718 	if (uap->path == SHM_ANON) {
719 		/* A read-only anonymous object is pointless. */
720 		if ((uap->flags & O_ACCMODE) == O_RDONLY) {
721 			fdclose(fdp, fp, fd, td);
722 			fdrop(fp, td);
723 			return (EINVAL);
724 		}
725 		shmfd = shm_alloc(td->td_ucred, cmode);
726 	} else {
727 		path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
728 		error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
729 
730 		/* Require paths to start with a '/' character. */
731 		if (error == 0 && path[0] != '/')
732 			error = EINVAL;
733 		if (error) {
734 			fdclose(fdp, fp, fd, td);
735 			fdrop(fp, td);
736 			free(path, M_SHMFD);
737 			return (error);
738 		}
739 
740 		fnv = fnv_32_str(path, FNV1_32_INIT);
741 		sx_xlock(&shm_dict_lock);
742 		shmfd = shm_lookup(path, fnv);
743 		if (shmfd == NULL) {
744 			/* Object does not yet exist, create it if requested. */
745 			if (uap->flags & O_CREAT) {
746 #ifdef MAC
747 				error = mac_posixshm_check_create(td->td_ucred,
748 				    path);
749 				if (error == 0) {
750 #endif
751 					shmfd = shm_alloc(td->td_ucred, cmode);
752 					shm_insert(path, fnv, shmfd);
753 #ifdef MAC
754 				}
755 #endif
756 			} else {
757 				free(path, M_SHMFD);
758 				error = ENOENT;
759 			}
760 		} else {
761 			/*
762 			 * Object already exists, obtain a new
763 			 * reference if requested and permitted.
764 			 */
765 			free(path, M_SHMFD);
766 			if ((uap->flags & (O_CREAT | O_EXCL)) ==
767 			    (O_CREAT | O_EXCL))
768 				error = EEXIST;
769 			else {
770 #ifdef MAC
771 				error = mac_posixshm_check_open(td->td_ucred,
772 				    shmfd, FFLAGS(uap->flags & O_ACCMODE));
773 				if (error == 0)
774 #endif
775 				error = shm_access(shmfd, td->td_ucred,
776 				    FFLAGS(uap->flags & O_ACCMODE));
777 			}
778 
779 			/*
780 			 * Truncate the file back to zero length if
781 			 * O_TRUNC was specified and the object was
782 			 * opened with read/write.
783 			 */
784 			if (error == 0 &&
785 			    (uap->flags & (O_ACCMODE | O_TRUNC)) ==
786 			    (O_RDWR | O_TRUNC)) {
787 #ifdef MAC
788 				error = mac_posixshm_check_truncate(
789 					td->td_ucred, fp->f_cred, shmfd);
790 				if (error == 0)
791 #endif
792 					shm_dotruncate(shmfd, 0);
793 			}
794 			if (error == 0)
795 				shm_hold(shmfd);
796 		}
797 		sx_xunlock(&shm_dict_lock);
798 
799 		if (error) {
800 			fdclose(fdp, fp, fd, td);
801 			fdrop(fp, td);
802 			return (error);
803 		}
804 	}
805 
806 	finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
807 
808 	td->td_retval[0] = fd;
809 	fdrop(fp, td);
810 
811 	return (0);
812 }
813 
814 int
815 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
816 {
817 	char *path;
818 	Fnv32_t fnv;
819 	int error;
820 
821 	path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
822 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
823 	if (error) {
824 		free(path, M_TEMP);
825 		return (error);
826 	}
827 
828 	fnv = fnv_32_str(path, FNV1_32_INIT);
829 	sx_xlock(&shm_dict_lock);
830 	error = shm_remove(path, fnv, td->td_ucred);
831 	sx_xunlock(&shm_dict_lock);
832 	free(path, M_TEMP);
833 
834 	return (error);
835 }
836 
837 /*
838  * mmap() helper to validate mmap() requests against shm object state
839  * and give mmap() the vm_object to use for the mapping.
840  */
841 int
842 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
843     vm_object_t *obj)
844 {
845 
846 	/*
847 	 * XXXRW: This validation is probably insufficient, and subject to
848 	 * sign errors.  It should be fixed.
849 	 */
850 	if (foff >= shmfd->shm_size ||
851 	    foff + objsize > round_page(shmfd->shm_size))
852 		return (EINVAL);
853 
854 	mtx_lock(&shm_timestamp_lock);
855 	vfs_timestamp(&shmfd->shm_atime);
856 	mtx_unlock(&shm_timestamp_lock);
857 	vm_object_reference(shmfd->shm_object);
858 	*obj = shmfd->shm_object;
859 	return (0);
860 }
861 
862 static int
863 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
864     struct thread *td)
865 {
866 	struct shmfd *shmfd;
867 	int error;
868 
869 	error = 0;
870 	shmfd = fp->f_data;
871 	mtx_lock(&shm_timestamp_lock);
872 	/*
873 	 * SUSv4 says that x bits of permission need not be affected.
874 	 * Be consistent with our shm_open there.
875 	 */
876 #ifdef MAC
877 	error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
878 	if (error != 0)
879 		goto out;
880 #endif
881 	error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
882 	    shmfd->shm_gid, VADMIN, active_cred, NULL);
883 	if (error != 0)
884 		goto out;
885 	shmfd->shm_mode = mode & ACCESSPERMS;
886 out:
887 	mtx_unlock(&shm_timestamp_lock);
888 	return (error);
889 }
890 
891 static int
892 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
893     struct thread *td)
894 {
895 	struct shmfd *shmfd;
896 	int error;
897 
898 	error = 0;
899 	shmfd = fp->f_data;
900 	mtx_lock(&shm_timestamp_lock);
901 #ifdef MAC
902 	error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
903 	if (error != 0)
904 		goto out;
905 #endif
906 	if (uid == (uid_t)-1)
907 		uid = shmfd->shm_uid;
908 	if (gid == (gid_t)-1)
909                  gid = shmfd->shm_gid;
910 	if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
911 	    (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
912 	    (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
913 		goto out;
914 	shmfd->shm_uid = uid;
915 	shmfd->shm_gid = gid;
916 out:
917 	mtx_unlock(&shm_timestamp_lock);
918 	return (error);
919 }
920 
921 /*
922  * Helper routines to allow the backing object of a shared memory file
923  * descriptor to be mapped in the kernel.
924  */
925 int
926 shm_map(struct file *fp, size_t size, off_t offset, void **memp)
927 {
928 	struct shmfd *shmfd;
929 	vm_offset_t kva, ofs;
930 	vm_object_t obj;
931 	int rv;
932 
933 	if (fp->f_type != DTYPE_SHM)
934 		return (EINVAL);
935 	shmfd = fp->f_data;
936 	obj = shmfd->shm_object;
937 	VM_OBJECT_WLOCK(obj);
938 	/*
939 	 * XXXRW: This validation is probably insufficient, and subject to
940 	 * sign errors.  It should be fixed.
941 	 */
942 	if (offset >= shmfd->shm_size ||
943 	    offset + size > round_page(shmfd->shm_size)) {
944 		VM_OBJECT_WUNLOCK(obj);
945 		return (EINVAL);
946 	}
947 
948 	shmfd->shm_kmappings++;
949 	vm_object_reference_locked(obj);
950 	VM_OBJECT_WUNLOCK(obj);
951 
952 	/* Map the object into the kernel_map and wire it. */
953 	kva = vm_map_min(kernel_map);
954 	ofs = offset & PAGE_MASK;
955 	offset = trunc_page(offset);
956 	size = round_page(size + ofs);
957 	rv = vm_map_find(kernel_map, obj, offset, &kva, size, 0,
958 	    VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE,
959 	    VM_PROT_READ | VM_PROT_WRITE, 0);
960 	if (rv == KERN_SUCCESS) {
961 		rv = vm_map_wire(kernel_map, kva, kva + size,
962 		    VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
963 		if (rv == KERN_SUCCESS) {
964 			*memp = (void *)(kva + ofs);
965 			return (0);
966 		}
967 		vm_map_remove(kernel_map, kva, kva + size);
968 	} else
969 		vm_object_deallocate(obj);
970 
971 	/* On failure, drop our mapping reference. */
972 	VM_OBJECT_WLOCK(obj);
973 	shmfd->shm_kmappings--;
974 	VM_OBJECT_WUNLOCK(obj);
975 
976 	return (vm_mmap_to_errno(rv));
977 }
978 
979 /*
980  * We require the caller to unmap the entire entry.  This allows us to
981  * safely decrement shm_kmappings when a mapping is removed.
982  */
983 int
984 shm_unmap(struct file *fp, void *mem, size_t size)
985 {
986 	struct shmfd *shmfd;
987 	vm_map_entry_t entry;
988 	vm_offset_t kva, ofs;
989 	vm_object_t obj;
990 	vm_pindex_t pindex;
991 	vm_prot_t prot;
992 	boolean_t wired;
993 	vm_map_t map;
994 	int rv;
995 
996 	if (fp->f_type != DTYPE_SHM)
997 		return (EINVAL);
998 	shmfd = fp->f_data;
999 	kva = (vm_offset_t)mem;
1000 	ofs = kva & PAGE_MASK;
1001 	kva = trunc_page(kva);
1002 	size = round_page(size + ofs);
1003 	map = kernel_map;
1004 	rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
1005 	    &obj, &pindex, &prot, &wired);
1006 	if (rv != KERN_SUCCESS)
1007 		return (EINVAL);
1008 	if (entry->start != kva || entry->end != kva + size) {
1009 		vm_map_lookup_done(map, entry);
1010 		return (EINVAL);
1011 	}
1012 	vm_map_lookup_done(map, entry);
1013 	if (obj != shmfd->shm_object)
1014 		return (EINVAL);
1015 	vm_map_remove(map, kva, kva + size);
1016 	VM_OBJECT_WLOCK(obj);
1017 	KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
1018 	shmfd->shm_kmappings--;
1019 	VM_OBJECT_WUNLOCK(obj);
1020 	return (0);
1021 }
1022 
1023 void
1024 shm_path(struct shmfd *shmfd, char *path, size_t size)
1025 {
1026 
1027 	if (shmfd->shm_path == NULL)
1028 		return;
1029 	sx_slock(&shm_dict_lock);
1030 	if (shmfd->shm_path != NULL)
1031 		strlcpy(path, shmfd->shm_path, size);
1032 	sx_sunlock(&shm_dict_lock);
1033 }
1034