xref: /freebsd/sys/kern/uipc_shm.c (revision 5dae51da3da0cc94d17bd67b308fad304ebec7e0)
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 #include "opt_ktrace.h"
49 
50 #include <sys/param.h>
51 #include <sys/capsicum.h>
52 #include <sys/conf.h>
53 #include <sys/fcntl.h>
54 #include <sys/file.h>
55 #include <sys/filedesc.h>
56 #include <sys/fnv_hash.h>
57 #include <sys/kernel.h>
58 #include <sys/uio.h>
59 #include <sys/signal.h>
60 #include <sys/jail.h>
61 #include <sys/ktrace.h>
62 #include <sys/lock.h>
63 #include <sys/malloc.h>
64 #include <sys/mman.h>
65 #include <sys/mutex.h>
66 #include <sys/priv.h>
67 #include <sys/proc.h>
68 #include <sys/refcount.h>
69 #include <sys/resourcevar.h>
70 #include <sys/rwlock.h>
71 #include <sys/stat.h>
72 #include <sys/syscallsubr.h>
73 #include <sys/sysctl.h>
74 #include <sys/sysproto.h>
75 #include <sys/systm.h>
76 #include <sys/sx.h>
77 #include <sys/time.h>
78 #include <sys/vnode.h>
79 #include <sys/unistd.h>
80 #include <sys/user.h>
81 
82 #include <security/mac/mac_framework.h>
83 
84 #include <vm/vm.h>
85 #include <vm/vm_param.h>
86 #include <vm/pmap.h>
87 #include <vm/vm_extern.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_kern.h>
90 #include <vm/vm_object.h>
91 #include <vm/vm_page.h>
92 #include <vm/vm_pageout.h>
93 #include <vm/vm_pager.h>
94 #include <vm/swap_pager.h>
95 
96 struct shm_mapping {
97 	char		*sm_path;
98 	Fnv32_t		sm_fnv;
99 	struct shmfd	*sm_shmfd;
100 	LIST_ENTRY(shm_mapping) sm_link;
101 };
102 
103 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
104 static LIST_HEAD(, shm_mapping) *shm_dictionary;
105 static struct sx shm_dict_lock;
106 static struct mtx shm_timestamp_lock;
107 static u_long shm_hash;
108 static struct unrhdr *shm_ino_unr;
109 static dev_t shm_dev_ino;
110 
111 #define	SHM_HASH(fnv)	(&shm_dictionary[(fnv) & shm_hash])
112 
113 static void	shm_init(void *arg);
114 static void	shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
115 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
116 static int	shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
117 
118 static fo_rdwr_t	shm_read;
119 static fo_rdwr_t	shm_write;
120 static fo_truncate_t	shm_truncate;
121 static fo_stat_t	shm_stat;
122 static fo_close_t	shm_close;
123 static fo_chmod_t	shm_chmod;
124 static fo_chown_t	shm_chown;
125 static fo_seek_t	shm_seek;
126 static fo_fill_kinfo_t	shm_fill_kinfo;
127 static fo_mmap_t	shm_mmap;
128 
129 /* File descriptor operations. */
130 struct fileops shm_ops = {
131 	.fo_read = shm_read,
132 	.fo_write = shm_write,
133 	.fo_truncate = shm_truncate,
134 	.fo_ioctl = invfo_ioctl,
135 	.fo_poll = invfo_poll,
136 	.fo_kqfilter = invfo_kqfilter,
137 	.fo_stat = shm_stat,
138 	.fo_close = shm_close,
139 	.fo_chmod = shm_chmod,
140 	.fo_chown = shm_chown,
141 	.fo_sendfile = vn_sendfile,
142 	.fo_seek = shm_seek,
143 	.fo_fill_kinfo = shm_fill_kinfo,
144 	.fo_mmap = shm_mmap,
145 	.fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
146 };
147 
148 FEATURE(posix_shm, "POSIX shared memory");
149 
150 static int
151 uiomove_object_page(vm_object_t obj, size_t len, struct uio *uio)
152 {
153 	vm_page_t m;
154 	vm_pindex_t idx;
155 	size_t tlen;
156 	int error, offset, rv;
157 
158 	idx = OFF_TO_IDX(uio->uio_offset);
159 	offset = uio->uio_offset & PAGE_MASK;
160 	tlen = MIN(PAGE_SIZE - offset, len);
161 
162 	VM_OBJECT_WLOCK(obj);
163 
164 	/*
165 	 * Read I/O without either a corresponding resident page or swap
166 	 * page: use zero_region.  This is intended to avoid instantiating
167 	 * pages on read from a sparse region.
168 	 */
169 	if (uio->uio_rw == UIO_READ && vm_page_lookup(obj, idx) == NULL &&
170 	    !vm_pager_has_page(obj, idx, NULL, NULL)) {
171 		VM_OBJECT_WUNLOCK(obj);
172 		return (uiomove(__DECONST(void *, zero_region), tlen, uio));
173 	}
174 
175 	/*
176 	 * Parallel reads of the page content from disk are prevented
177 	 * by exclusive busy.
178 	 *
179 	 * Although the tmpfs vnode lock is held here, it is
180 	 * nonetheless safe to sleep waiting for a free page.  The
181 	 * pageout daemon does not need to acquire the tmpfs vnode
182 	 * lock to page out tobj's pages because tobj is a OBJT_SWAP
183 	 * type object.
184 	 */
185 	m = vm_page_grab(obj, idx, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
186 	if (m->valid != VM_PAGE_BITS_ALL) {
187 		vm_page_xbusy(m);
188 		if (vm_pager_has_page(obj, idx, NULL, NULL)) {
189 			rv = vm_pager_get_pages(obj, &m, 1, NULL, NULL);
190 			if (rv != VM_PAGER_OK) {
191 				printf(
192 	    "uiomove_object: vm_obj %p idx %jd valid %x pager error %d\n",
193 				    obj, idx, m->valid, rv);
194 				vm_page_lock(m);
195 				vm_page_free(m);
196 				vm_page_unlock(m);
197 				VM_OBJECT_WUNLOCK(obj);
198 				return (EIO);
199 			}
200 		} else
201 			vm_page_zero_invalid(m, TRUE);
202 		vm_page_xunbusy(m);
203 	}
204 	vm_page_lock(m);
205 	vm_page_hold(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 	VM_OBJECT_WUNLOCK(obj);
214 	error = uiomove_fromphys(&m, offset, tlen, uio);
215 	if (uio->uio_rw == UIO_WRITE && error == 0) {
216 		VM_OBJECT_WLOCK(obj);
217 		vm_page_dirty(m);
218 		vm_pager_page_unswapped(m);
219 		VM_OBJECT_WUNLOCK(obj);
220 	}
221 	vm_page_lock(m);
222 	vm_page_unhold(m);
223 	vm_page_unlock(m);
224 
225 	return (error);
226 }
227 
228 int
229 uiomove_object(vm_object_t obj, off_t obj_size, struct uio *uio)
230 {
231 	ssize_t resid;
232 	size_t len;
233 	int error;
234 
235 	error = 0;
236 	while ((resid = uio->uio_resid) > 0) {
237 		if (obj_size <= uio->uio_offset)
238 			break;
239 		len = MIN(obj_size - uio->uio_offset, resid);
240 		if (len == 0)
241 			break;
242 		error = uiomove_object_page(obj, len, uio);
243 		if (error != 0 || resid == uio->uio_resid)
244 			break;
245 	}
246 	return (error);
247 }
248 
249 static int
250 shm_seek(struct file *fp, off_t offset, int whence, struct thread *td)
251 {
252 	struct shmfd *shmfd;
253 	off_t foffset;
254 	int error;
255 
256 	shmfd = fp->f_data;
257 	foffset = foffset_lock(fp, 0);
258 	error = 0;
259 	switch (whence) {
260 	case L_INCR:
261 		if (foffset < 0 ||
262 		    (offset > 0 && foffset > OFF_MAX - offset)) {
263 			error = EOVERFLOW;
264 			break;
265 		}
266 		offset += foffset;
267 		break;
268 	case L_XTND:
269 		if (offset > 0 && shmfd->shm_size > OFF_MAX - offset) {
270 			error = EOVERFLOW;
271 			break;
272 		}
273 		offset += shmfd->shm_size;
274 		break;
275 	case L_SET:
276 		break;
277 	default:
278 		error = EINVAL;
279 	}
280 	if (error == 0) {
281 		if (offset < 0 || offset > shmfd->shm_size)
282 			error = EINVAL;
283 		else
284 			td->td_uretoff.tdu_off = offset;
285 	}
286 	foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
287 	return (error);
288 }
289 
290 static int
291 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
292     int flags, struct thread *td)
293 {
294 	struct shmfd *shmfd;
295 	void *rl_cookie;
296 	int error;
297 
298 	shmfd = fp->f_data;
299 #ifdef MAC
300 	error = mac_posixshm_check_read(active_cred, fp->f_cred, shmfd);
301 	if (error)
302 		return (error);
303 #endif
304 	foffset_lock_uio(fp, uio, flags);
305 	rl_cookie = rangelock_rlock(&shmfd->shm_rl, uio->uio_offset,
306 	    uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx);
307 	error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio);
308 	rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx);
309 	foffset_unlock_uio(fp, uio, flags);
310 	return (error);
311 }
312 
313 static int
314 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
315     int flags, struct thread *td)
316 {
317 	struct shmfd *shmfd;
318 	void *rl_cookie;
319 	int error;
320 
321 	shmfd = fp->f_data;
322 #ifdef MAC
323 	error = mac_posixshm_check_write(active_cred, fp->f_cred, shmfd);
324 	if (error)
325 		return (error);
326 #endif
327 	foffset_lock_uio(fp, uio, flags);
328 	if ((flags & FOF_OFFSET) == 0) {
329 		rl_cookie = rangelock_wlock(&shmfd->shm_rl, 0, OFF_MAX,
330 		    &shmfd->shm_mtx);
331 	} else {
332 		rl_cookie = rangelock_wlock(&shmfd->shm_rl, uio->uio_offset,
333 		    uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx);
334 	}
335 
336 	error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio);
337 	rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx);
338 	foffset_unlock_uio(fp, uio, flags);
339 	return (error);
340 }
341 
342 static int
343 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
344     struct thread *td)
345 {
346 	struct shmfd *shmfd;
347 #ifdef MAC
348 	int error;
349 #endif
350 
351 	shmfd = fp->f_data;
352 #ifdef MAC
353 	error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
354 	if (error)
355 		return (error);
356 #endif
357 	return (shm_dotruncate(shmfd, length));
358 }
359 
360 static int
361 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
362     struct thread *td)
363 {
364 	struct shmfd *shmfd;
365 #ifdef MAC
366 	int error;
367 #endif
368 
369 	shmfd = fp->f_data;
370 
371 #ifdef MAC
372 	error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
373 	if (error)
374 		return (error);
375 #endif
376 
377 	/*
378 	 * Attempt to return sanish values for fstat() on a memory file
379 	 * descriptor.
380 	 */
381 	bzero(sb, sizeof(*sb));
382 	sb->st_blksize = PAGE_SIZE;
383 	sb->st_size = shmfd->shm_size;
384 	sb->st_blocks = howmany(sb->st_size, sb->st_blksize);
385 	mtx_lock(&shm_timestamp_lock);
386 	sb->st_atim = shmfd->shm_atime;
387 	sb->st_ctim = shmfd->shm_ctime;
388 	sb->st_mtim = shmfd->shm_mtime;
389 	sb->st_birthtim = shmfd->shm_birthtime;
390 	sb->st_mode = S_IFREG | shmfd->shm_mode;		/* XXX */
391 	sb->st_uid = shmfd->shm_uid;
392 	sb->st_gid = shmfd->shm_gid;
393 	mtx_unlock(&shm_timestamp_lock);
394 	sb->st_dev = shm_dev_ino;
395 	sb->st_ino = shmfd->shm_ino;
396 
397 	return (0);
398 }
399 
400 static int
401 shm_close(struct file *fp, struct thread *td)
402 {
403 	struct shmfd *shmfd;
404 
405 	shmfd = fp->f_data;
406 	fp->f_data = NULL;
407 	shm_drop(shmfd);
408 
409 	return (0);
410 }
411 
412 int
413 shm_dotruncate(struct shmfd *shmfd, off_t length)
414 {
415 	vm_object_t object;
416 	vm_page_t m;
417 	vm_pindex_t idx, nobjsize;
418 	vm_ooffset_t delta;
419 	int base, rv;
420 
421 	object = shmfd->shm_object;
422 	VM_OBJECT_WLOCK(object);
423 	if (length == shmfd->shm_size) {
424 		VM_OBJECT_WUNLOCK(object);
425 		return (0);
426 	}
427 	nobjsize = OFF_TO_IDX(length + PAGE_MASK);
428 
429 	/* Are we shrinking?  If so, trim the end. */
430 	if (length < shmfd->shm_size) {
431 		/*
432 		 * Disallow any requests to shrink the size if this
433 		 * object is mapped into the kernel.
434 		 */
435 		if (shmfd->shm_kmappings > 0) {
436 			VM_OBJECT_WUNLOCK(object);
437 			return (EBUSY);
438 		}
439 
440 		/*
441 		 * Zero the truncated part of the last page.
442 		 */
443 		base = length & PAGE_MASK;
444 		if (base != 0) {
445 			idx = OFF_TO_IDX(length);
446 retry:
447 			m = vm_page_lookup(object, idx);
448 			if (m != NULL) {
449 				if (vm_page_sleep_if_busy(m, "shmtrc"))
450 					goto retry;
451 			} else if (vm_pager_has_page(object, idx, NULL, NULL)) {
452 				m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL);
453 				if (m == NULL) {
454 					VM_OBJECT_WUNLOCK(object);
455 					VM_WAIT;
456 					VM_OBJECT_WLOCK(object);
457 					goto retry;
458 				} else if (m->valid != VM_PAGE_BITS_ALL)
459 					rv = vm_pager_get_pages(object, &m, 1,
460 					    NULL, NULL);
461 				else
462 					/* A cached page was reactivated. */
463 					rv = VM_PAGER_OK;
464 				vm_page_lock(m);
465 				if (rv == VM_PAGER_OK) {
466 					vm_page_deactivate(m);
467 					vm_page_unlock(m);
468 					vm_page_xunbusy(m);
469 				} else {
470 					vm_page_free(m);
471 					vm_page_unlock(m);
472 					VM_OBJECT_WUNLOCK(object);
473 					return (EIO);
474 				}
475 			}
476 			if (m != NULL) {
477 				pmap_zero_page_area(m, base, PAGE_SIZE - base);
478 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
479 				    ("shm_dotruncate: page %p is invalid", m));
480 				vm_page_dirty(m);
481 				vm_pager_page_unswapped(m);
482 			}
483 		}
484 		delta = ptoa(object->size - nobjsize);
485 
486 		/* Toss in memory pages. */
487 		if (nobjsize < object->size)
488 			vm_object_page_remove(object, nobjsize, object->size,
489 			    0);
490 
491 		/* Toss pages from swap. */
492 		if (object->type == OBJT_SWAP)
493 			swap_pager_freespace(object, nobjsize, delta);
494 
495 		/* Free the swap accounted for shm */
496 		swap_release_by_cred(delta, object->cred);
497 		object->charge -= delta;
498 	} else {
499 		/* Attempt to reserve the swap */
500 		delta = ptoa(nobjsize - object->size);
501 		if (!swap_reserve_by_cred(delta, object->cred)) {
502 			VM_OBJECT_WUNLOCK(object);
503 			return (ENOMEM);
504 		}
505 		object->charge += delta;
506 	}
507 	shmfd->shm_size = length;
508 	mtx_lock(&shm_timestamp_lock);
509 	vfs_timestamp(&shmfd->shm_ctime);
510 	shmfd->shm_mtime = shmfd->shm_ctime;
511 	mtx_unlock(&shm_timestamp_lock);
512 	object->size = nobjsize;
513 	VM_OBJECT_WUNLOCK(object);
514 	return (0);
515 }
516 
517 /*
518  * shmfd object management including creation and reference counting
519  * routines.
520  */
521 struct shmfd *
522 shm_alloc(struct ucred *ucred, mode_t mode)
523 {
524 	struct shmfd *shmfd;
525 	int ino;
526 
527 	shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
528 	shmfd->shm_size = 0;
529 	shmfd->shm_uid = ucred->cr_uid;
530 	shmfd->shm_gid = ucred->cr_gid;
531 	shmfd->shm_mode = mode;
532 	shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
533 	    shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
534 	KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
535 	shmfd->shm_object->pg_color = 0;
536 	VM_OBJECT_WLOCK(shmfd->shm_object);
537 	vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
538 	vm_object_set_flag(shmfd->shm_object, OBJ_COLORED | OBJ_NOSPLIT);
539 	VM_OBJECT_WUNLOCK(shmfd->shm_object);
540 	vfs_timestamp(&shmfd->shm_birthtime);
541 	shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
542 	    shmfd->shm_birthtime;
543 	ino = alloc_unr(shm_ino_unr);
544 	if (ino == -1)
545 		shmfd->shm_ino = 0;
546 	else
547 		shmfd->shm_ino = ino;
548 	refcount_init(&shmfd->shm_refs, 1);
549 	mtx_init(&shmfd->shm_mtx, "shmrl", NULL, MTX_DEF);
550 	rangelock_init(&shmfd->shm_rl);
551 #ifdef MAC
552 	mac_posixshm_init(shmfd);
553 	mac_posixshm_create(ucred, shmfd);
554 #endif
555 
556 	return (shmfd);
557 }
558 
559 struct shmfd *
560 shm_hold(struct shmfd *shmfd)
561 {
562 
563 	refcount_acquire(&shmfd->shm_refs);
564 	return (shmfd);
565 }
566 
567 void
568 shm_drop(struct shmfd *shmfd)
569 {
570 
571 	if (refcount_release(&shmfd->shm_refs)) {
572 #ifdef MAC
573 		mac_posixshm_destroy(shmfd);
574 #endif
575 		rangelock_destroy(&shmfd->shm_rl);
576 		mtx_destroy(&shmfd->shm_mtx);
577 		vm_object_deallocate(shmfd->shm_object);
578 		if (shmfd->shm_ino != 0)
579 			free_unr(shm_ino_unr, shmfd->shm_ino);
580 		free(shmfd, M_SHMFD);
581 	}
582 }
583 
584 /*
585  * Determine if the credentials have sufficient permissions for a
586  * specified combination of FREAD and FWRITE.
587  */
588 int
589 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
590 {
591 	accmode_t accmode;
592 	int error;
593 
594 	accmode = 0;
595 	if (flags & FREAD)
596 		accmode |= VREAD;
597 	if (flags & FWRITE)
598 		accmode |= VWRITE;
599 	mtx_lock(&shm_timestamp_lock);
600 	error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
601 	    accmode, ucred, NULL);
602 	mtx_unlock(&shm_timestamp_lock);
603 	return (error);
604 }
605 
606 /*
607  * Dictionary management.  We maintain an in-kernel dictionary to map
608  * paths to shmfd objects.  We use the FNV hash on the path to store
609  * the mappings in a hash table.
610  */
611 static void
612 shm_init(void *arg)
613 {
614 
615 	mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
616 	sx_init(&shm_dict_lock, "shm dictionary");
617 	shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
618 	shm_ino_unr = new_unrhdr(1, INT32_MAX, NULL);
619 	KASSERT(shm_ino_unr != NULL, ("shm fake inodes not initialized"));
620 	shm_dev_ino = devfs_alloc_cdp_inode();
621 	KASSERT(shm_dev_ino > 0, ("shm dev inode not initialized"));
622 }
623 SYSINIT(shm_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_init, NULL);
624 
625 static struct shmfd *
626 shm_lookup(char *path, Fnv32_t fnv)
627 {
628 	struct shm_mapping *map;
629 
630 	LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
631 		if (map->sm_fnv != fnv)
632 			continue;
633 		if (strcmp(map->sm_path, path) == 0)
634 			return (map->sm_shmfd);
635 	}
636 
637 	return (NULL);
638 }
639 
640 static void
641 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
642 {
643 	struct shm_mapping *map;
644 
645 	map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
646 	map->sm_path = path;
647 	map->sm_fnv = fnv;
648 	map->sm_shmfd = shm_hold(shmfd);
649 	shmfd->shm_path = path;
650 	LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
651 }
652 
653 static int
654 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
655 {
656 	struct shm_mapping *map;
657 	int error;
658 
659 	LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
660 		if (map->sm_fnv != fnv)
661 			continue;
662 		if (strcmp(map->sm_path, path) == 0) {
663 #ifdef MAC
664 			error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
665 			if (error)
666 				return (error);
667 #endif
668 			error = shm_access(map->sm_shmfd, ucred,
669 			    FREAD | FWRITE);
670 			if (error)
671 				return (error);
672 			map->sm_shmfd->shm_path = NULL;
673 			LIST_REMOVE(map, sm_link);
674 			shm_drop(map->sm_shmfd);
675 			free(map->sm_path, M_SHMFD);
676 			free(map, M_SHMFD);
677 			return (0);
678 		}
679 	}
680 
681 	return (ENOENT);
682 }
683 
684 int
685 kern_shm_open(struct thread *td, const char *userpath, int flags, mode_t mode,
686     struct filecaps *fcaps)
687 {
688 	struct filedesc *fdp;
689 	struct shmfd *shmfd;
690 	struct file *fp;
691 	char *path;
692 	const char *pr_path;
693 	size_t pr_pathlen;
694 	Fnv32_t fnv;
695 	mode_t cmode;
696 	int fd, error;
697 
698 #ifdef CAPABILITY_MODE
699 	/*
700 	 * shm_open(2) is only allowed for anonymous objects.
701 	 */
702 	if (IN_CAPABILITY_MODE(td) && (userpath != SHM_ANON))
703 		return (ECAPMODE);
704 #endif
705 
706 	if ((flags & O_ACCMODE) != O_RDONLY && (flags & O_ACCMODE) != O_RDWR)
707 		return (EINVAL);
708 
709 	if ((flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC | O_CLOEXEC)) != 0)
710 		return (EINVAL);
711 
712 	fdp = td->td_proc->p_fd;
713 	cmode = (mode & ~fdp->fd_cmask) & ACCESSPERMS;
714 
715 	error = falloc_caps(td, &fp, &fd, O_CLOEXEC, fcaps);
716 	if (error)
717 		return (error);
718 
719 	/* A SHM_ANON path pointer creates an anonymous object. */
720 	if (userpath == SHM_ANON) {
721 		/* A read-only anonymous object is pointless. */
722 		if ((flags & O_ACCMODE) == O_RDONLY) {
723 			fdclose(td, fp, fd);
724 			fdrop(fp, td);
725 			return (EINVAL);
726 		}
727 		shmfd = shm_alloc(td->td_ucred, cmode);
728 	} else {
729 		path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
730 		pr_path = td->td_ucred->cr_prison->pr_path;
731 
732 		/* Construct a full pathname for jailed callers. */
733 		pr_pathlen = strcmp(pr_path, "/") == 0 ? 0
734 		    : strlcpy(path, pr_path, MAXPATHLEN);
735 		error = copyinstr(userpath, path + pr_pathlen,
736 		    MAXPATHLEN - pr_pathlen, NULL);
737 #ifdef KTRACE
738 		if (error == 0 && KTRPOINT(curthread, KTR_NAMEI))
739 			ktrnamei(path);
740 #endif
741 		/* Require paths to start with a '/' character. */
742 		if (error == 0 && path[pr_pathlen] != '/')
743 			error = EINVAL;
744 		if (error) {
745 			fdclose(td, fp, fd);
746 			fdrop(fp, td);
747 			free(path, M_SHMFD);
748 			return (error);
749 		}
750 
751 		fnv = fnv_32_str(path, FNV1_32_INIT);
752 		sx_xlock(&shm_dict_lock);
753 		shmfd = shm_lookup(path, fnv);
754 		if (shmfd == NULL) {
755 			/* Object does not yet exist, create it if requested. */
756 			if (flags & O_CREAT) {
757 #ifdef MAC
758 				error = mac_posixshm_check_create(td->td_ucred,
759 				    path);
760 				if (error == 0) {
761 #endif
762 					shmfd = shm_alloc(td->td_ucred, cmode);
763 					shm_insert(path, fnv, shmfd);
764 #ifdef MAC
765 				}
766 #endif
767 			} else {
768 				free(path, M_SHMFD);
769 				error = ENOENT;
770 			}
771 		} else {
772 			/*
773 			 * Object already exists, obtain a new
774 			 * reference if requested and permitted.
775 			 */
776 			free(path, M_SHMFD);
777 			if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
778 				error = EEXIST;
779 			else {
780 #ifdef MAC
781 				error = mac_posixshm_check_open(td->td_ucred,
782 				    shmfd, FFLAGS(flags & O_ACCMODE));
783 				if (error == 0)
784 #endif
785 				error = shm_access(shmfd, td->td_ucred,
786 				    FFLAGS(flags & O_ACCMODE));
787 			}
788 
789 			/*
790 			 * Truncate the file back to zero length if
791 			 * O_TRUNC was specified and the object was
792 			 * opened with read/write.
793 			 */
794 			if (error == 0 &&
795 			    (flags & (O_ACCMODE | O_TRUNC)) ==
796 			    (O_RDWR | O_TRUNC)) {
797 #ifdef MAC
798 				error = mac_posixshm_check_truncate(
799 					td->td_ucred, fp->f_cred, shmfd);
800 				if (error == 0)
801 #endif
802 					shm_dotruncate(shmfd, 0);
803 			}
804 			if (error == 0)
805 				shm_hold(shmfd);
806 		}
807 		sx_xunlock(&shm_dict_lock);
808 
809 		if (error) {
810 			fdclose(td, fp, fd);
811 			fdrop(fp, td);
812 			return (error);
813 		}
814 	}
815 
816 	finit(fp, FFLAGS(flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
817 
818 	td->td_retval[0] = fd;
819 	fdrop(fp, td);
820 
821 	return (0);
822 }
823 
824 /* System calls. */
825 int
826 sys_shm_open(struct thread *td, struct shm_open_args *uap)
827 {
828 
829 	return (kern_shm_open(td, uap->path, uap->flags, uap->mode, NULL));
830 }
831 
832 int
833 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
834 {
835 	char *path;
836 	const char *pr_path;
837 	size_t pr_pathlen;
838 	Fnv32_t fnv;
839 	int error;
840 
841 	path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
842 	pr_path = td->td_ucred->cr_prison->pr_path;
843 	pr_pathlen = strcmp(pr_path, "/") == 0 ? 0
844 	    : strlcpy(path, pr_path, MAXPATHLEN);
845 	error = copyinstr(uap->path, path + pr_pathlen, MAXPATHLEN - pr_pathlen,
846 	    NULL);
847 	if (error) {
848 		free(path, M_TEMP);
849 		return (error);
850 	}
851 #ifdef KTRACE
852 	if (KTRPOINT(curthread, KTR_NAMEI))
853 		ktrnamei(path);
854 #endif
855 	fnv = fnv_32_str(path, FNV1_32_INIT);
856 	sx_xlock(&shm_dict_lock);
857 	error = shm_remove(path, fnv, td->td_ucred);
858 	sx_xunlock(&shm_dict_lock);
859 	free(path, M_TEMP);
860 
861 	return (error);
862 }
863 
864 int
865 shm_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t objsize,
866     vm_prot_t prot, vm_prot_t cap_maxprot, int flags,
867     vm_ooffset_t foff, struct thread *td)
868 {
869 	struct shmfd *shmfd;
870 	vm_prot_t maxprot;
871 	int error;
872 
873 	shmfd = fp->f_data;
874 	maxprot = VM_PROT_NONE;
875 
876 	/* FREAD should always be set. */
877 	if ((fp->f_flag & FREAD) != 0)
878 		maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
879 	if ((fp->f_flag & FWRITE) != 0)
880 		maxprot |= VM_PROT_WRITE;
881 
882 	/* Don't permit shared writable mappings on read-only descriptors. */
883 	if ((flags & MAP_SHARED) != 0 &&
884 	    (maxprot & VM_PROT_WRITE) == 0 &&
885 	    (prot & VM_PROT_WRITE) != 0)
886 		return (EACCES);
887 	maxprot &= cap_maxprot;
888 
889 #ifdef MAC
890 	error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, flags);
891 	if (error != 0)
892 		return (error);
893 #endif
894 
895 	/*
896 	 * XXXRW: This validation is probably insufficient, and subject to
897 	 * sign errors.  It should be fixed.
898 	 */
899 	if (foff >= shmfd->shm_size ||
900 	    foff + objsize > round_page(shmfd->shm_size))
901 		return (EINVAL);
902 
903 	mtx_lock(&shm_timestamp_lock);
904 	vfs_timestamp(&shmfd->shm_atime);
905 	mtx_unlock(&shm_timestamp_lock);
906 	vm_object_reference(shmfd->shm_object);
907 
908 	error = vm_mmap_object(map, addr, objsize, prot, maxprot, flags,
909 	    shmfd->shm_object, foff, FALSE, td);
910 	if (error != 0)
911 		vm_object_deallocate(shmfd->shm_object);
912 	return (0);
913 }
914 
915 static int
916 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
917     struct thread *td)
918 {
919 	struct shmfd *shmfd;
920 	int error;
921 
922 	error = 0;
923 	shmfd = fp->f_data;
924 	mtx_lock(&shm_timestamp_lock);
925 	/*
926 	 * SUSv4 says that x bits of permission need not be affected.
927 	 * Be consistent with our shm_open there.
928 	 */
929 #ifdef MAC
930 	error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
931 	if (error != 0)
932 		goto out;
933 #endif
934 	error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
935 	    shmfd->shm_gid, VADMIN, active_cred, NULL);
936 	if (error != 0)
937 		goto out;
938 	shmfd->shm_mode = mode & ACCESSPERMS;
939 out:
940 	mtx_unlock(&shm_timestamp_lock);
941 	return (error);
942 }
943 
944 static int
945 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
946     struct thread *td)
947 {
948 	struct shmfd *shmfd;
949 	int error;
950 
951 	error = 0;
952 	shmfd = fp->f_data;
953 	mtx_lock(&shm_timestamp_lock);
954 #ifdef MAC
955 	error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
956 	if (error != 0)
957 		goto out;
958 #endif
959 	if (uid == (uid_t)-1)
960 		uid = shmfd->shm_uid;
961 	if (gid == (gid_t)-1)
962                  gid = shmfd->shm_gid;
963 	if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
964 	    (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
965 	    (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
966 		goto out;
967 	shmfd->shm_uid = uid;
968 	shmfd->shm_gid = gid;
969 out:
970 	mtx_unlock(&shm_timestamp_lock);
971 	return (error);
972 }
973 
974 /*
975  * Helper routines to allow the backing object of a shared memory file
976  * descriptor to be mapped in the kernel.
977  */
978 int
979 shm_map(struct file *fp, size_t size, off_t offset, void **memp)
980 {
981 	struct shmfd *shmfd;
982 	vm_offset_t kva, ofs;
983 	vm_object_t obj;
984 	int rv;
985 
986 	if (fp->f_type != DTYPE_SHM)
987 		return (EINVAL);
988 	shmfd = fp->f_data;
989 	obj = shmfd->shm_object;
990 	VM_OBJECT_WLOCK(obj);
991 	/*
992 	 * XXXRW: This validation is probably insufficient, and subject to
993 	 * sign errors.  It should be fixed.
994 	 */
995 	if (offset >= shmfd->shm_size ||
996 	    offset + size > round_page(shmfd->shm_size)) {
997 		VM_OBJECT_WUNLOCK(obj);
998 		return (EINVAL);
999 	}
1000 
1001 	shmfd->shm_kmappings++;
1002 	vm_object_reference_locked(obj);
1003 	VM_OBJECT_WUNLOCK(obj);
1004 
1005 	/* Map the object into the kernel_map and wire it. */
1006 	kva = vm_map_min(kernel_map);
1007 	ofs = offset & PAGE_MASK;
1008 	offset = trunc_page(offset);
1009 	size = round_page(size + ofs);
1010 	rv = vm_map_find(kernel_map, obj, offset, &kva, size, 0,
1011 	    VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE,
1012 	    VM_PROT_READ | VM_PROT_WRITE, 0);
1013 	if (rv == KERN_SUCCESS) {
1014 		rv = vm_map_wire(kernel_map, kva, kva + size,
1015 		    VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
1016 		if (rv == KERN_SUCCESS) {
1017 			*memp = (void *)(kva + ofs);
1018 			return (0);
1019 		}
1020 		vm_map_remove(kernel_map, kva, kva + size);
1021 	} else
1022 		vm_object_deallocate(obj);
1023 
1024 	/* On failure, drop our mapping reference. */
1025 	VM_OBJECT_WLOCK(obj);
1026 	shmfd->shm_kmappings--;
1027 	VM_OBJECT_WUNLOCK(obj);
1028 
1029 	return (vm_mmap_to_errno(rv));
1030 }
1031 
1032 /*
1033  * We require the caller to unmap the entire entry.  This allows us to
1034  * safely decrement shm_kmappings when a mapping is removed.
1035  */
1036 int
1037 shm_unmap(struct file *fp, void *mem, size_t size)
1038 {
1039 	struct shmfd *shmfd;
1040 	vm_map_entry_t entry;
1041 	vm_offset_t kva, ofs;
1042 	vm_object_t obj;
1043 	vm_pindex_t pindex;
1044 	vm_prot_t prot;
1045 	boolean_t wired;
1046 	vm_map_t map;
1047 	int rv;
1048 
1049 	if (fp->f_type != DTYPE_SHM)
1050 		return (EINVAL);
1051 	shmfd = fp->f_data;
1052 	kva = (vm_offset_t)mem;
1053 	ofs = kva & PAGE_MASK;
1054 	kva = trunc_page(kva);
1055 	size = round_page(size + ofs);
1056 	map = kernel_map;
1057 	rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
1058 	    &obj, &pindex, &prot, &wired);
1059 	if (rv != KERN_SUCCESS)
1060 		return (EINVAL);
1061 	if (entry->start != kva || entry->end != kva + size) {
1062 		vm_map_lookup_done(map, entry);
1063 		return (EINVAL);
1064 	}
1065 	vm_map_lookup_done(map, entry);
1066 	if (obj != shmfd->shm_object)
1067 		return (EINVAL);
1068 	vm_map_remove(map, kva, kva + size);
1069 	VM_OBJECT_WLOCK(obj);
1070 	KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
1071 	shmfd->shm_kmappings--;
1072 	VM_OBJECT_WUNLOCK(obj);
1073 	return (0);
1074 }
1075 
1076 static int
1077 shm_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
1078 {
1079 	const char *path, *pr_path;
1080 	struct shmfd *shmfd;
1081 	size_t pr_pathlen;
1082 
1083 	kif->kf_type = KF_TYPE_SHM;
1084 	shmfd = fp->f_data;
1085 
1086 	mtx_lock(&shm_timestamp_lock);
1087 	kif->kf_un.kf_file.kf_file_mode = S_IFREG | shmfd->shm_mode;	/* XXX */
1088 	mtx_unlock(&shm_timestamp_lock);
1089 	kif->kf_un.kf_file.kf_file_size = shmfd->shm_size;
1090 	if (shmfd->shm_path != NULL) {
1091 		sx_slock(&shm_dict_lock);
1092 		if (shmfd->shm_path != NULL) {
1093 			path = shmfd->shm_path;
1094 			pr_path = curthread->td_ucred->cr_prison->pr_path;
1095 			if (strcmp(pr_path, "/") != 0) {
1096 				/* Return the jail-rooted pathname. */
1097 				pr_pathlen = strlen(pr_path);
1098 				if (strncmp(path, pr_path, pr_pathlen) == 0 &&
1099 				    path[pr_pathlen] == '/')
1100 					path += pr_pathlen;
1101 			}
1102 			strlcpy(kif->kf_path, path, sizeof(kif->kf_path));
1103 		}
1104 		sx_sunlock(&shm_dict_lock);
1105 	}
1106 	return (0);
1107 }
1108