xref: /freebsd/sys/kern/uipc_shm.c (revision d0ef721ed3dc99bddc1e48605a6921ec60322efc)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2006, 2011, 2016-2017 Robert N. M. Watson
5  * All rights reserved.
6  *
7  * Portions of this software were developed by BAE Systems, the University of
8  * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL
9  * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent
10  * Computing (TC) research program.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 /*
35  * Support for shared swap-backed anonymous memory objects via
36  * shm_open(2) and shm_unlink(2).  While most of the implementation is
37  * here, vm_mmap.c contains mapping logic changes.
38  *
39  * posixshmcontrol(1) allows users to inspect the state of the memory
40  * objects.  Per-uid swap resource limit controls total amount of
41  * memory that user can consume for anonymous objects, including
42  * shared.
43  */
44 
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 
48 #include "opt_capsicum.h"
49 #include "opt_ktrace.h"
50 
51 #include <sys/param.h>
52 #include <sys/capsicum.h>
53 #include <sys/conf.h>
54 #include <sys/fcntl.h>
55 #include <sys/file.h>
56 #include <sys/filedesc.h>
57 #include <sys/filio.h>
58 #include <sys/fnv_hash.h>
59 #include <sys/kernel.h>
60 #include <sys/limits.h>
61 #include <sys/uio.h>
62 #include <sys/signal.h>
63 #include <sys/jail.h>
64 #include <sys/ktrace.h>
65 #include <sys/lock.h>
66 #include <sys/malloc.h>
67 #include <sys/mman.h>
68 #include <sys/mutex.h>
69 #include <sys/priv.h>
70 #include <sys/proc.h>
71 #include <sys/refcount.h>
72 #include <sys/resourcevar.h>
73 #include <sys/rwlock.h>
74 #include <sys/sbuf.h>
75 #include <sys/stat.h>
76 #include <sys/syscallsubr.h>
77 #include <sys/sysctl.h>
78 #include <sys/sysproto.h>
79 #include <sys/systm.h>
80 #include <sys/sx.h>
81 #include <sys/time.h>
82 #include <sys/vnode.h>
83 #include <sys/unistd.h>
84 #include <sys/user.h>
85 
86 #include <security/audit/audit.h>
87 #include <security/mac/mac_framework.h>
88 
89 #include <vm/vm.h>
90 #include <vm/vm_param.h>
91 #include <vm/pmap.h>
92 #include <vm/vm_extern.h>
93 #include <vm/vm_map.h>
94 #include <vm/vm_kern.h>
95 #include <vm/vm_object.h>
96 #include <vm/vm_page.h>
97 #include <vm/vm_pageout.h>
98 #include <vm/vm_pager.h>
99 #include <vm/swap_pager.h>
100 
101 struct shm_mapping {
102 	char		*sm_path;
103 	Fnv32_t		sm_fnv;
104 	struct shmfd	*sm_shmfd;
105 	LIST_ENTRY(shm_mapping) sm_link;
106 };
107 
108 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
109 static LIST_HEAD(, shm_mapping) *shm_dictionary;
110 static struct sx shm_dict_lock;
111 static struct mtx shm_timestamp_lock;
112 static u_long shm_hash;
113 static struct unrhdr64 shm_ino_unr;
114 static dev_t shm_dev_ino;
115 
116 #define	SHM_HASH(fnv)	(&shm_dictionary[(fnv) & shm_hash])
117 
118 static void	shm_init(void *arg);
119 static void	shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
120 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
121 static int	shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
122 
123 static fo_rdwr_t	shm_read;
124 static fo_rdwr_t	shm_write;
125 static fo_truncate_t	shm_truncate;
126 static fo_ioctl_t	shm_ioctl;
127 static fo_stat_t	shm_stat;
128 static fo_close_t	shm_close;
129 static fo_chmod_t	shm_chmod;
130 static fo_chown_t	shm_chown;
131 static fo_seek_t	shm_seek;
132 static fo_fill_kinfo_t	shm_fill_kinfo;
133 static fo_mmap_t	shm_mmap;
134 
135 /* File descriptor operations. */
136 struct fileops shm_ops = {
137 	.fo_read = shm_read,
138 	.fo_write = shm_write,
139 	.fo_truncate = shm_truncate,
140 	.fo_ioctl = shm_ioctl,
141 	.fo_poll = invfo_poll,
142 	.fo_kqfilter = invfo_kqfilter,
143 	.fo_stat = shm_stat,
144 	.fo_close = shm_close,
145 	.fo_chmod = shm_chmod,
146 	.fo_chown = shm_chown,
147 	.fo_sendfile = vn_sendfile,
148 	.fo_seek = shm_seek,
149 	.fo_fill_kinfo = shm_fill_kinfo,
150 	.fo_mmap = shm_mmap,
151 	.fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
152 };
153 
154 FEATURE(posix_shm, "POSIX shared memory");
155 
156 static int
157 uiomove_object_page(vm_object_t obj, size_t len, struct uio *uio)
158 {
159 	vm_page_t m;
160 	vm_pindex_t idx;
161 	size_t tlen;
162 	int error, offset, rv;
163 
164 	idx = OFF_TO_IDX(uio->uio_offset);
165 	offset = uio->uio_offset & PAGE_MASK;
166 	tlen = MIN(PAGE_SIZE - offset, len);
167 
168 	VM_OBJECT_WLOCK(obj);
169 
170 	/*
171 	 * Read I/O without either a corresponding resident page or swap
172 	 * page: use zero_region.  This is intended to avoid instantiating
173 	 * pages on read from a sparse region.
174 	 */
175 	if (uio->uio_rw == UIO_READ && vm_page_lookup(obj, idx) == NULL &&
176 	    !vm_pager_has_page(obj, idx, NULL, NULL)) {
177 		VM_OBJECT_WUNLOCK(obj);
178 		return (uiomove(__DECONST(void *, zero_region), tlen, uio));
179 	}
180 
181 	/*
182 	 * Parallel reads of the page content from disk are prevented
183 	 * by exclusive busy.
184 	 *
185 	 * Although the tmpfs vnode lock is held here, it is
186 	 * nonetheless safe to sleep waiting for a free page.  The
187 	 * pageout daemon does not need to acquire the tmpfs vnode
188 	 * lock to page out tobj's pages because tobj is a OBJT_SWAP
189 	 * type object.
190 	 */
191 	m = vm_page_grab(obj, idx, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY |
192 	    VM_ALLOC_WIRED);
193 	if (m->valid != VM_PAGE_BITS_ALL) {
194 		vm_page_xbusy(m);
195 		if (vm_pager_has_page(obj, idx, NULL, NULL)) {
196 			rv = vm_pager_get_pages(obj, &m, 1, NULL, NULL);
197 			if (rv != VM_PAGER_OK) {
198 				printf(
199 	    "uiomove_object: vm_obj %p idx %jd valid %x pager error %d\n",
200 				    obj, idx, m->valid, rv);
201 				vm_page_unwire_noq(m);
202 				vm_page_free(m);
203 				VM_OBJECT_WUNLOCK(obj);
204 				return (EIO);
205 			}
206 		} else
207 			vm_page_zero_invalid(m, TRUE);
208 		vm_page_xunbusy(m);
209 	}
210 	VM_OBJECT_WUNLOCK(obj);
211 	error = uiomove_fromphys(&m, offset, tlen, uio);
212 	if (uio->uio_rw == UIO_WRITE && error == 0) {
213 		VM_OBJECT_WLOCK(obj);
214 		vm_page_dirty(m);
215 		vm_pager_page_unswapped(m);
216 		VM_OBJECT_WUNLOCK(obj);
217 	}
218 	vm_page_unwire(m, PQ_ACTIVE);
219 
220 	return (error);
221 }
222 
223 int
224 uiomove_object(vm_object_t obj, off_t obj_size, struct uio *uio)
225 {
226 	ssize_t resid;
227 	size_t len;
228 	int error;
229 
230 	error = 0;
231 	while ((resid = uio->uio_resid) > 0) {
232 		if (obj_size <= uio->uio_offset)
233 			break;
234 		len = MIN(obj_size - uio->uio_offset, resid);
235 		if (len == 0)
236 			break;
237 		error = uiomove_object_page(obj, len, uio);
238 		if (error != 0 || resid == uio->uio_resid)
239 			break;
240 	}
241 	return (error);
242 }
243 
244 static int
245 shm_seek(struct file *fp, off_t offset, int whence, struct thread *td)
246 {
247 	struct shmfd *shmfd;
248 	off_t foffset;
249 	int error;
250 
251 	shmfd = fp->f_data;
252 	foffset = foffset_lock(fp, 0);
253 	error = 0;
254 	switch (whence) {
255 	case L_INCR:
256 		if (foffset < 0 ||
257 		    (offset > 0 && foffset > OFF_MAX - offset)) {
258 			error = EOVERFLOW;
259 			break;
260 		}
261 		offset += foffset;
262 		break;
263 	case L_XTND:
264 		if (offset > 0 && shmfd->shm_size > OFF_MAX - offset) {
265 			error = EOVERFLOW;
266 			break;
267 		}
268 		offset += shmfd->shm_size;
269 		break;
270 	case L_SET:
271 		break;
272 	default:
273 		error = EINVAL;
274 	}
275 	if (error == 0) {
276 		if (offset < 0 || offset > shmfd->shm_size)
277 			error = EINVAL;
278 		else
279 			td->td_uretoff.tdu_off = offset;
280 	}
281 	foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
282 	return (error);
283 }
284 
285 static int
286 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
287     int flags, struct thread *td)
288 {
289 	struct shmfd *shmfd;
290 	void *rl_cookie;
291 	int error;
292 
293 	shmfd = fp->f_data;
294 #ifdef MAC
295 	error = mac_posixshm_check_read(active_cred, fp->f_cred, shmfd);
296 	if (error)
297 		return (error);
298 #endif
299 	foffset_lock_uio(fp, uio, flags);
300 	rl_cookie = rangelock_rlock(&shmfd->shm_rl, uio->uio_offset,
301 	    uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx);
302 	error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio);
303 	rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx);
304 	foffset_unlock_uio(fp, uio, flags);
305 	return (error);
306 }
307 
308 static int
309 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
310     int flags, struct thread *td)
311 {
312 	struct shmfd *shmfd;
313 	void *rl_cookie;
314 	int error;
315 
316 	shmfd = fp->f_data;
317 #ifdef MAC
318 	error = mac_posixshm_check_write(active_cred, fp->f_cred, shmfd);
319 	if (error)
320 		return (error);
321 #endif
322 	foffset_lock_uio(fp, uio, flags);
323 	if ((flags & FOF_OFFSET) == 0) {
324 		rl_cookie = rangelock_wlock(&shmfd->shm_rl, 0, OFF_MAX,
325 		    &shmfd->shm_mtx);
326 	} else {
327 		rl_cookie = rangelock_wlock(&shmfd->shm_rl, uio->uio_offset,
328 		    uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx);
329 	}
330 
331 	error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio);
332 	rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx);
333 	foffset_unlock_uio(fp, uio, flags);
334 	return (error);
335 }
336 
337 static int
338 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
339     struct thread *td)
340 {
341 	struct shmfd *shmfd;
342 #ifdef MAC
343 	int error;
344 #endif
345 
346 	shmfd = fp->f_data;
347 #ifdef MAC
348 	error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
349 	if (error)
350 		return (error);
351 #endif
352 	return (shm_dotruncate(shmfd, length));
353 }
354 
355 int
356 shm_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
357     struct thread *td)
358 {
359 
360 	switch (com) {
361 	case FIONBIO:
362 	case FIOASYNC:
363 		/*
364 		 * Allow fcntl(fd, F_SETFL, O_NONBLOCK) to work,
365 		 * just like it would on an unlinked regular file
366 		 */
367 		return (0);
368 	default:
369 		return (ENOTTY);
370 	}
371 }
372 
373 static int
374 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
375     struct thread *td)
376 {
377 	struct shmfd *shmfd;
378 #ifdef MAC
379 	int error;
380 #endif
381 
382 	shmfd = fp->f_data;
383 
384 #ifdef MAC
385 	error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
386 	if (error)
387 		return (error);
388 #endif
389 
390 	/*
391 	 * Attempt to return sanish values for fstat() on a memory file
392 	 * descriptor.
393 	 */
394 	bzero(sb, sizeof(*sb));
395 	sb->st_blksize = PAGE_SIZE;
396 	sb->st_size = shmfd->shm_size;
397 	sb->st_blocks = howmany(sb->st_size, sb->st_blksize);
398 	mtx_lock(&shm_timestamp_lock);
399 	sb->st_atim = shmfd->shm_atime;
400 	sb->st_ctim = shmfd->shm_ctime;
401 	sb->st_mtim = shmfd->shm_mtime;
402 	sb->st_birthtim = shmfd->shm_birthtime;
403 	sb->st_mode = S_IFREG | shmfd->shm_mode;		/* XXX */
404 	sb->st_uid = shmfd->shm_uid;
405 	sb->st_gid = shmfd->shm_gid;
406 	mtx_unlock(&shm_timestamp_lock);
407 	sb->st_dev = shm_dev_ino;
408 	sb->st_ino = shmfd->shm_ino;
409 	sb->st_nlink = shmfd->shm_object->ref_count;
410 
411 	return (0);
412 }
413 
414 static int
415 shm_close(struct file *fp, struct thread *td)
416 {
417 	struct shmfd *shmfd;
418 
419 	shmfd = fp->f_data;
420 	fp->f_data = NULL;
421 	shm_drop(shmfd);
422 
423 	return (0);
424 }
425 
426 int
427 shm_dotruncate(struct shmfd *shmfd, off_t length)
428 {
429 	vm_object_t object;
430 	vm_page_t m;
431 	vm_pindex_t idx, nobjsize;
432 	vm_ooffset_t delta;
433 	int base, rv;
434 
435 	KASSERT(length >= 0, ("shm_dotruncate: length < 0"));
436 	object = shmfd->shm_object;
437 	VM_OBJECT_WLOCK(object);
438 	if (length == shmfd->shm_size) {
439 		VM_OBJECT_WUNLOCK(object);
440 		return (0);
441 	}
442 	nobjsize = OFF_TO_IDX(length + PAGE_MASK);
443 
444 	/* Are we shrinking?  If so, trim the end. */
445 	if (length < shmfd->shm_size) {
446 		/*
447 		 * Disallow any requests to shrink the size if this
448 		 * object is mapped into the kernel.
449 		 */
450 		if (shmfd->shm_kmappings > 0) {
451 			VM_OBJECT_WUNLOCK(object);
452 			return (EBUSY);
453 		}
454 
455 		/*
456 		 * Zero the truncated part of the last page.
457 		 */
458 		base = length & PAGE_MASK;
459 		if (base != 0) {
460 			idx = OFF_TO_IDX(length);
461 retry:
462 			m = vm_page_lookup(object, idx);
463 			if (m != NULL) {
464 				if (vm_page_sleep_if_busy(m, "shmtrc"))
465 					goto retry;
466 			} else if (vm_pager_has_page(object, idx, NULL, NULL)) {
467 				m = vm_page_alloc(object, idx,
468 				    VM_ALLOC_NORMAL | VM_ALLOC_WAITFAIL);
469 				if (m == NULL)
470 					goto retry;
471 				rv = vm_pager_get_pages(object, &m, 1, NULL,
472 				    NULL);
473 				if (rv == VM_PAGER_OK) {
474 					/*
475 					 * Since the page was not resident,
476 					 * and therefore not recently
477 					 * accessed, immediately enqueue it
478 					 * for asynchronous laundering.  The
479 					 * current operation is not regarded
480 					 * as an access.
481 					 */
482 					vm_page_launder(m);
483 					vm_page_xunbusy(m);
484 				} else {
485 					vm_page_free(m);
486 					VM_OBJECT_WUNLOCK(object);
487 					return (EIO);
488 				}
489 			}
490 			if (m != NULL) {
491 				pmap_zero_page_area(m, base, PAGE_SIZE - base);
492 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
493 				    ("shm_dotruncate: page %p is invalid", m));
494 				vm_page_dirty(m);
495 				vm_pager_page_unswapped(m);
496 			}
497 		}
498 		delta = IDX_TO_OFF(object->size - nobjsize);
499 
500 		/* Toss in memory pages. */
501 		if (nobjsize < object->size)
502 			vm_object_page_remove(object, nobjsize, object->size,
503 			    0);
504 
505 		/* Toss pages from swap. */
506 		if (object->type == OBJT_SWAP)
507 			swap_pager_freespace(object, nobjsize, delta);
508 
509 		/* Free the swap accounted for shm */
510 		swap_release_by_cred(delta, object->cred);
511 		object->charge -= delta;
512 	} else {
513 		/* Try to reserve additional swap space. */
514 		delta = IDX_TO_OFF(nobjsize - object->size);
515 		if (!swap_reserve_by_cred(delta, object->cred)) {
516 			VM_OBJECT_WUNLOCK(object);
517 			return (ENOMEM);
518 		}
519 		object->charge += delta;
520 	}
521 	shmfd->shm_size = length;
522 	mtx_lock(&shm_timestamp_lock);
523 	vfs_timestamp(&shmfd->shm_ctime);
524 	shmfd->shm_mtime = shmfd->shm_ctime;
525 	mtx_unlock(&shm_timestamp_lock);
526 	object->size = nobjsize;
527 	VM_OBJECT_WUNLOCK(object);
528 	return (0);
529 }
530 
531 /*
532  * shmfd object management including creation and reference counting
533  * routines.
534  */
535 struct shmfd *
536 shm_alloc(struct ucred *ucred, mode_t mode)
537 {
538 	struct shmfd *shmfd;
539 
540 	shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
541 	shmfd->shm_size = 0;
542 	shmfd->shm_uid = ucred->cr_uid;
543 	shmfd->shm_gid = ucred->cr_gid;
544 	shmfd->shm_mode = mode;
545 	shmfd->shm_object = vm_pager_allocate(OBJT_SWAP, NULL,
546 	    shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
547 	KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
548 	shmfd->shm_object->pg_color = 0;
549 	VM_OBJECT_WLOCK(shmfd->shm_object);
550 	vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
551 	vm_object_set_flag(shmfd->shm_object, OBJ_COLORED | OBJ_NOSPLIT);
552 	VM_OBJECT_WUNLOCK(shmfd->shm_object);
553 	vfs_timestamp(&shmfd->shm_birthtime);
554 	shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
555 	    shmfd->shm_birthtime;
556 	shmfd->shm_ino = alloc_unr64(&shm_ino_unr);
557 	refcount_init(&shmfd->shm_refs, 1);
558 	mtx_init(&shmfd->shm_mtx, "shmrl", NULL, MTX_DEF);
559 	rangelock_init(&shmfd->shm_rl);
560 #ifdef MAC
561 	mac_posixshm_init(shmfd);
562 	mac_posixshm_create(ucred, shmfd);
563 #endif
564 
565 	return (shmfd);
566 }
567 
568 struct shmfd *
569 shm_hold(struct shmfd *shmfd)
570 {
571 
572 	refcount_acquire(&shmfd->shm_refs);
573 	return (shmfd);
574 }
575 
576 void
577 shm_drop(struct shmfd *shmfd)
578 {
579 
580 	if (refcount_release(&shmfd->shm_refs)) {
581 #ifdef MAC
582 		mac_posixshm_destroy(shmfd);
583 #endif
584 		rangelock_destroy(&shmfd->shm_rl);
585 		mtx_destroy(&shmfd->shm_mtx);
586 		vm_object_deallocate(shmfd->shm_object);
587 		free(shmfd, M_SHMFD);
588 	}
589 }
590 
591 /*
592  * Determine if the credentials have sufficient permissions for a
593  * specified combination of FREAD and FWRITE.
594  */
595 int
596 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
597 {
598 	accmode_t accmode;
599 	int error;
600 
601 	accmode = 0;
602 	if (flags & FREAD)
603 		accmode |= VREAD;
604 	if (flags & FWRITE)
605 		accmode |= VWRITE;
606 	mtx_lock(&shm_timestamp_lock);
607 	error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
608 	    accmode, ucred, NULL);
609 	mtx_unlock(&shm_timestamp_lock);
610 	return (error);
611 }
612 
613 /*
614  * Dictionary management.  We maintain an in-kernel dictionary to map
615  * paths to shmfd objects.  We use the FNV hash on the path to store
616  * the mappings in a hash table.
617  */
618 static void
619 shm_init(void *arg)
620 {
621 
622 	mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
623 	sx_init(&shm_dict_lock, "shm dictionary");
624 	shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
625 	new_unrhdr64(&shm_ino_unr, 1);
626 	shm_dev_ino = devfs_alloc_cdp_inode();
627 	KASSERT(shm_dev_ino > 0, ("shm dev inode not initialized"));
628 }
629 SYSINIT(shm_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_init, NULL);
630 
631 static struct shmfd *
632 shm_lookup(char *path, Fnv32_t fnv)
633 {
634 	struct shm_mapping *map;
635 
636 	LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
637 		if (map->sm_fnv != fnv)
638 			continue;
639 		if (strcmp(map->sm_path, path) == 0)
640 			return (map->sm_shmfd);
641 	}
642 
643 	return (NULL);
644 }
645 
646 static void
647 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
648 {
649 	struct shm_mapping *map;
650 
651 	map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
652 	map->sm_path = path;
653 	map->sm_fnv = fnv;
654 	map->sm_shmfd = shm_hold(shmfd);
655 	shmfd->shm_path = path;
656 	LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
657 }
658 
659 static int
660 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
661 {
662 	struct shm_mapping *map;
663 	int error;
664 
665 	LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
666 		if (map->sm_fnv != fnv)
667 			continue;
668 		if (strcmp(map->sm_path, path) == 0) {
669 #ifdef MAC
670 			error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
671 			if (error)
672 				return (error);
673 #endif
674 			error = shm_access(map->sm_shmfd, ucred,
675 			    FREAD | FWRITE);
676 			if (error)
677 				return (error);
678 			map->sm_shmfd->shm_path = NULL;
679 			LIST_REMOVE(map, sm_link);
680 			shm_drop(map->sm_shmfd);
681 			free(map->sm_path, M_SHMFD);
682 			free(map, M_SHMFD);
683 			return (0);
684 		}
685 	}
686 
687 	return (ENOENT);
688 }
689 
690 int
691 kern_shm_open(struct thread *td, const char *userpath, int flags, mode_t mode,
692     struct filecaps *fcaps)
693 {
694 	struct filedesc *fdp;
695 	struct shmfd *shmfd;
696 	struct file *fp;
697 	char *path;
698 	const char *pr_path;
699 	size_t pr_pathlen;
700 	Fnv32_t fnv;
701 	mode_t cmode;
702 	int fd, error;
703 
704 #ifdef CAPABILITY_MODE
705 	/*
706 	 * shm_open(2) is only allowed for anonymous objects.
707 	 */
708 	if (IN_CAPABILITY_MODE(td) && (userpath != SHM_ANON))
709 		return (ECAPMODE);
710 #endif
711 
712 	AUDIT_ARG_FFLAGS(flags);
713 	AUDIT_ARG_MODE(mode);
714 
715 	if ((flags & O_ACCMODE) != O_RDONLY && (flags & O_ACCMODE) != O_RDWR)
716 		return (EINVAL);
717 
718 	if ((flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC | O_CLOEXEC)) != 0)
719 		return (EINVAL);
720 
721 	fdp = td->td_proc->p_fd;
722 	cmode = (mode & ~fdp->fd_cmask) & ACCESSPERMS;
723 
724 	/*
725 	 * shm_open(2) created shm should always have O_CLOEXEC set, as mandated
726 	 * by POSIX.  We allow it to be unset here so that an in-kernel
727 	 * interface may be written as a thin layer around shm, optionally not
728 	 * setting CLOEXEC.  For shm_open(2), O_CLOEXEC is set unconditionally
729 	 * in sys_shm_open() to keep this implementation compliant.
730 	 */
731 	error = falloc_caps(td, &fp, &fd, flags & O_CLOEXEC, fcaps);
732 	if (error)
733 		return (error);
734 
735 	/* A SHM_ANON path pointer creates an anonymous object. */
736 	if (userpath == SHM_ANON) {
737 		/* A read-only anonymous object is pointless. */
738 		if ((flags & O_ACCMODE) == O_RDONLY) {
739 			fdclose(td, fp, fd);
740 			fdrop(fp, td);
741 			return (EINVAL);
742 		}
743 		shmfd = shm_alloc(td->td_ucred, cmode);
744 	} else {
745 		path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
746 		pr_path = td->td_ucred->cr_prison->pr_path;
747 
748 		/* Construct a full pathname for jailed callers. */
749 		pr_pathlen = strcmp(pr_path, "/") == 0 ? 0
750 		    : strlcpy(path, pr_path, MAXPATHLEN);
751 		error = copyinstr(userpath, path + pr_pathlen,
752 		    MAXPATHLEN - pr_pathlen, NULL);
753 #ifdef KTRACE
754 		if (error == 0 && KTRPOINT(curthread, KTR_NAMEI))
755 			ktrnamei(path);
756 #endif
757 		/* Require paths to start with a '/' character. */
758 		if (error == 0 && path[pr_pathlen] != '/')
759 			error = EINVAL;
760 		if (error) {
761 			fdclose(td, fp, fd);
762 			fdrop(fp, td);
763 			free(path, M_SHMFD);
764 			return (error);
765 		}
766 
767 		AUDIT_ARG_UPATH1_CANON(path);
768 		fnv = fnv_32_str(path, FNV1_32_INIT);
769 		sx_xlock(&shm_dict_lock);
770 		shmfd = shm_lookup(path, fnv);
771 		if (shmfd == NULL) {
772 			/* Object does not yet exist, create it if requested. */
773 			if (flags & O_CREAT) {
774 #ifdef MAC
775 				error = mac_posixshm_check_create(td->td_ucred,
776 				    path);
777 				if (error == 0) {
778 #endif
779 					shmfd = shm_alloc(td->td_ucred, cmode);
780 					shm_insert(path, fnv, shmfd);
781 #ifdef MAC
782 				}
783 #endif
784 			} else {
785 				free(path, M_SHMFD);
786 				error = ENOENT;
787 			}
788 		} else {
789 			/*
790 			 * Object already exists, obtain a new
791 			 * reference if requested and permitted.
792 			 */
793 			free(path, M_SHMFD);
794 			if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
795 				error = EEXIST;
796 			else {
797 #ifdef MAC
798 				error = mac_posixshm_check_open(td->td_ucred,
799 				    shmfd, FFLAGS(flags & O_ACCMODE));
800 				if (error == 0)
801 #endif
802 				error = shm_access(shmfd, td->td_ucred,
803 				    FFLAGS(flags & O_ACCMODE));
804 			}
805 
806 			/*
807 			 * Truncate the file back to zero length if
808 			 * O_TRUNC was specified and the object was
809 			 * opened with read/write.
810 			 */
811 			if (error == 0 &&
812 			    (flags & (O_ACCMODE | O_TRUNC)) ==
813 			    (O_RDWR | O_TRUNC)) {
814 #ifdef MAC
815 				error = mac_posixshm_check_truncate(
816 					td->td_ucred, fp->f_cred, shmfd);
817 				if (error == 0)
818 #endif
819 					shm_dotruncate(shmfd, 0);
820 			}
821 			if (error == 0)
822 				shm_hold(shmfd);
823 		}
824 		sx_xunlock(&shm_dict_lock);
825 
826 		if (error) {
827 			fdclose(td, fp, fd);
828 			fdrop(fp, td);
829 			return (error);
830 		}
831 	}
832 
833 	finit(fp, FFLAGS(flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
834 
835 	td->td_retval[0] = fd;
836 	fdrop(fp, td);
837 
838 	return (0);
839 }
840 
841 /* System calls. */
842 int
843 sys_shm_open(struct thread *td, struct shm_open_args *uap)
844 {
845 
846 	return (kern_shm_open(td, uap->path, uap->flags | O_CLOEXEC, uap->mode,
847 	    NULL));
848 }
849 
850 int
851 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
852 {
853 	char *path;
854 	const char *pr_path;
855 	size_t pr_pathlen;
856 	Fnv32_t fnv;
857 	int error;
858 
859 	path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
860 	pr_path = td->td_ucred->cr_prison->pr_path;
861 	pr_pathlen = strcmp(pr_path, "/") == 0 ? 0
862 	    : strlcpy(path, pr_path, MAXPATHLEN);
863 	error = copyinstr(uap->path, path + pr_pathlen, MAXPATHLEN - pr_pathlen,
864 	    NULL);
865 	if (error) {
866 		free(path, M_TEMP);
867 		return (error);
868 	}
869 #ifdef KTRACE
870 	if (KTRPOINT(curthread, KTR_NAMEI))
871 		ktrnamei(path);
872 #endif
873 	AUDIT_ARG_UPATH1_CANON(path);
874 	fnv = fnv_32_str(path, FNV1_32_INIT);
875 	sx_xlock(&shm_dict_lock);
876 	error = shm_remove(path, fnv, td->td_ucred);
877 	sx_xunlock(&shm_dict_lock);
878 	free(path, M_TEMP);
879 
880 	return (error);
881 }
882 
883 int
884 shm_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t objsize,
885     vm_prot_t prot, vm_prot_t cap_maxprot, int flags,
886     vm_ooffset_t foff, struct thread *td)
887 {
888 	struct shmfd *shmfd;
889 	vm_prot_t maxprot;
890 	int error;
891 	bool writecnt;
892 
893 	shmfd = fp->f_data;
894 	maxprot = VM_PROT_NONE;
895 
896 	/* FREAD should always be set. */
897 	if ((fp->f_flag & FREAD) != 0)
898 		maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
899 	if ((fp->f_flag & FWRITE) != 0)
900 		maxprot |= VM_PROT_WRITE;
901 
902 	writecnt = (flags & MAP_SHARED) != 0 && (prot & VM_PROT_WRITE) != 0;
903 
904 	/* Don't permit shared writable mappings on read-only descriptors. */
905 	if (writecnt && (maxprot & VM_PROT_WRITE) == 0)
906 		return (EACCES);
907 	maxprot &= cap_maxprot;
908 
909 	/* See comment in vn_mmap(). */
910 	if (
911 #ifdef _LP64
912 	    objsize > OFF_MAX ||
913 #endif
914 	    foff < 0 || foff > OFF_MAX - objsize)
915 		return (EINVAL);
916 
917 #ifdef MAC
918 	error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, flags);
919 	if (error != 0)
920 		return (error);
921 #endif
922 
923 	mtx_lock(&shm_timestamp_lock);
924 	vfs_timestamp(&shmfd->shm_atime);
925 	mtx_unlock(&shm_timestamp_lock);
926 	vm_object_reference(shmfd->shm_object);
927 
928 	if (writecnt)
929 		vm_pager_update_writecount(shmfd->shm_object, 0, objsize);
930 	error = vm_mmap_object(map, addr, objsize, prot, maxprot, flags,
931 	    shmfd->shm_object, foff, writecnt, td);
932 	if (error != 0) {
933 		if (writecnt)
934 			vm_pager_release_writecount(shmfd->shm_object, 0,
935 			    objsize);
936 		vm_object_deallocate(shmfd->shm_object);
937 	}
938 	return (error);
939 }
940 
941 static int
942 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
943     struct thread *td)
944 {
945 	struct shmfd *shmfd;
946 	int error;
947 
948 	error = 0;
949 	shmfd = fp->f_data;
950 	mtx_lock(&shm_timestamp_lock);
951 	/*
952 	 * SUSv4 says that x bits of permission need not be affected.
953 	 * Be consistent with our shm_open there.
954 	 */
955 #ifdef MAC
956 	error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
957 	if (error != 0)
958 		goto out;
959 #endif
960 	error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
961 	    shmfd->shm_gid, VADMIN, active_cred, NULL);
962 	if (error != 0)
963 		goto out;
964 	shmfd->shm_mode = mode & ACCESSPERMS;
965 out:
966 	mtx_unlock(&shm_timestamp_lock);
967 	return (error);
968 }
969 
970 static int
971 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
972     struct thread *td)
973 {
974 	struct shmfd *shmfd;
975 	int error;
976 
977 	error = 0;
978 	shmfd = fp->f_data;
979 	mtx_lock(&shm_timestamp_lock);
980 #ifdef MAC
981 	error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
982 	if (error != 0)
983 		goto out;
984 #endif
985 	if (uid == (uid_t)-1)
986 		uid = shmfd->shm_uid;
987 	if (gid == (gid_t)-1)
988                  gid = shmfd->shm_gid;
989 	if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
990 	    (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
991 	    (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN)))
992 		goto out;
993 	shmfd->shm_uid = uid;
994 	shmfd->shm_gid = gid;
995 out:
996 	mtx_unlock(&shm_timestamp_lock);
997 	return (error);
998 }
999 
1000 /*
1001  * Helper routines to allow the backing object of a shared memory file
1002  * descriptor to be mapped in the kernel.
1003  */
1004 int
1005 shm_map(struct file *fp, size_t size, off_t offset, void **memp)
1006 {
1007 	struct shmfd *shmfd;
1008 	vm_offset_t kva, ofs;
1009 	vm_object_t obj;
1010 	int rv;
1011 
1012 	if (fp->f_type != DTYPE_SHM)
1013 		return (EINVAL);
1014 	shmfd = fp->f_data;
1015 	obj = shmfd->shm_object;
1016 	VM_OBJECT_WLOCK(obj);
1017 	/*
1018 	 * XXXRW: This validation is probably insufficient, and subject to
1019 	 * sign errors.  It should be fixed.
1020 	 */
1021 	if (offset >= shmfd->shm_size ||
1022 	    offset + size > round_page(shmfd->shm_size)) {
1023 		VM_OBJECT_WUNLOCK(obj);
1024 		return (EINVAL);
1025 	}
1026 
1027 	shmfd->shm_kmappings++;
1028 	vm_object_reference_locked(obj);
1029 	VM_OBJECT_WUNLOCK(obj);
1030 
1031 	/* Map the object into the kernel_map and wire it. */
1032 	kva = vm_map_min(kernel_map);
1033 	ofs = offset & PAGE_MASK;
1034 	offset = trunc_page(offset);
1035 	size = round_page(size + ofs);
1036 	rv = vm_map_find(kernel_map, obj, offset, &kva, size, 0,
1037 	    VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE,
1038 	    VM_PROT_READ | VM_PROT_WRITE, 0);
1039 	if (rv == KERN_SUCCESS) {
1040 		rv = vm_map_wire(kernel_map, kva, kva + size,
1041 		    VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
1042 		if (rv == KERN_SUCCESS) {
1043 			*memp = (void *)(kva + ofs);
1044 			return (0);
1045 		}
1046 		vm_map_remove(kernel_map, kva, kva + size);
1047 	} else
1048 		vm_object_deallocate(obj);
1049 
1050 	/* On failure, drop our mapping reference. */
1051 	VM_OBJECT_WLOCK(obj);
1052 	shmfd->shm_kmappings--;
1053 	VM_OBJECT_WUNLOCK(obj);
1054 
1055 	return (vm_mmap_to_errno(rv));
1056 }
1057 
1058 /*
1059  * We require the caller to unmap the entire entry.  This allows us to
1060  * safely decrement shm_kmappings when a mapping is removed.
1061  */
1062 int
1063 shm_unmap(struct file *fp, void *mem, size_t size)
1064 {
1065 	struct shmfd *shmfd;
1066 	vm_map_entry_t entry;
1067 	vm_offset_t kva, ofs;
1068 	vm_object_t obj;
1069 	vm_pindex_t pindex;
1070 	vm_prot_t prot;
1071 	boolean_t wired;
1072 	vm_map_t map;
1073 	int rv;
1074 
1075 	if (fp->f_type != DTYPE_SHM)
1076 		return (EINVAL);
1077 	shmfd = fp->f_data;
1078 	kva = (vm_offset_t)mem;
1079 	ofs = kva & PAGE_MASK;
1080 	kva = trunc_page(kva);
1081 	size = round_page(size + ofs);
1082 	map = kernel_map;
1083 	rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
1084 	    &obj, &pindex, &prot, &wired);
1085 	if (rv != KERN_SUCCESS)
1086 		return (EINVAL);
1087 	if (entry->start != kva || entry->end != kva + size) {
1088 		vm_map_lookup_done(map, entry);
1089 		return (EINVAL);
1090 	}
1091 	vm_map_lookup_done(map, entry);
1092 	if (obj != shmfd->shm_object)
1093 		return (EINVAL);
1094 	vm_map_remove(map, kva, kva + size);
1095 	VM_OBJECT_WLOCK(obj);
1096 	KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
1097 	shmfd->shm_kmappings--;
1098 	VM_OBJECT_WUNLOCK(obj);
1099 	return (0);
1100 }
1101 
1102 static int
1103 shm_fill_kinfo_locked(struct shmfd *shmfd, struct kinfo_file *kif, bool list)
1104 {
1105 	const char *path, *pr_path;
1106 	size_t pr_pathlen;
1107 	bool visible;
1108 
1109 	sx_assert(&shm_dict_lock, SA_LOCKED);
1110 	kif->kf_type = KF_TYPE_SHM;
1111 	kif->kf_un.kf_file.kf_file_mode = S_IFREG | shmfd->shm_mode;
1112 	kif->kf_un.kf_file.kf_file_size = shmfd->shm_size;
1113 	if (shmfd->shm_path != NULL) {
1114 		if (shmfd->shm_path != NULL) {
1115 			path = shmfd->shm_path;
1116 			pr_path = curthread->td_ucred->cr_prison->pr_path;
1117 			if (strcmp(pr_path, "/") != 0) {
1118 				/* Return the jail-rooted pathname. */
1119 				pr_pathlen = strlen(pr_path);
1120 				visible = strncmp(path, pr_path, pr_pathlen)
1121 				    == 0 && path[pr_pathlen] == '/';
1122 				if (list && !visible)
1123 					return (EPERM);
1124 				if (visible)
1125 					path += pr_pathlen;
1126 			}
1127 			strlcpy(kif->kf_path, path, sizeof(kif->kf_path));
1128 		}
1129 	}
1130 	return (0);
1131 }
1132 
1133 static int
1134 shm_fill_kinfo(struct file *fp, struct kinfo_file *kif,
1135     struct filedesc *fdp __unused)
1136 {
1137 	int res;
1138 
1139 	sx_slock(&shm_dict_lock);
1140 	res = shm_fill_kinfo_locked(fp->f_data, kif, false);
1141 	sx_sunlock(&shm_dict_lock);
1142 	return (res);
1143 }
1144 
1145 static int
1146 sysctl_posix_shm_list(SYSCTL_HANDLER_ARGS)
1147 {
1148 	struct shm_mapping *shmm;
1149 	struct sbuf sb;
1150 	struct kinfo_file kif;
1151 	u_long i;
1152 	ssize_t curlen;
1153 	int error, error2;
1154 
1155 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file) * 5, req);
1156 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
1157 	curlen = 0;
1158 	error = 0;
1159 	sx_slock(&shm_dict_lock);
1160 	for (i = 0; i < shm_hash + 1; i++) {
1161 		LIST_FOREACH(shmm, &shm_dictionary[i], sm_link) {
1162 			error = shm_fill_kinfo_locked(shmm->sm_shmfd,
1163 			    &kif, true);
1164 			if (error == EPERM)
1165 				continue;
1166 			if (error != 0)
1167 				break;
1168 			pack_kinfo(&kif);
1169 			if (req->oldptr != NULL &&
1170 			    kif.kf_structsize + curlen > req->oldlen)
1171 				break;
1172 			error = sbuf_bcat(&sb, &kif, kif.kf_structsize) == 0 ?
1173 			    0 : ENOMEM;
1174 			if (error != 0)
1175 				break;
1176 			curlen += kif.kf_structsize;
1177 		}
1178 	}
1179 	sx_sunlock(&shm_dict_lock);
1180 	error2 = sbuf_finish(&sb);
1181 	sbuf_delete(&sb);
1182 	return (error != 0 ? error : error2);
1183 }
1184 
1185 SYSCTL_PROC(_kern_ipc, OID_AUTO, posix_shm_list,
1186     CTLFLAG_RD | CTLFLAG_MPSAFE | CTLTYPE_OPAQUE,
1187     NULL, 0, sysctl_posix_shm_list, "",
1188     "POSIX SHM list");
1189