xref: /freebsd/sys/kern/uipc_shm.c (revision ec0e626bafb335b30c499d06066997f54b10c092)
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/ktrace.h>
61 #include <sys/lock.h>
62 #include <sys/malloc.h>
63 #include <sys/mman.h>
64 #include <sys/mutex.h>
65 #include <sys/priv.h>
66 #include <sys/proc.h>
67 #include <sys/refcount.h>
68 #include <sys/resourcevar.h>
69 #include <sys/rwlock.h>
70 #include <sys/stat.h>
71 #include <sys/sysctl.h>
72 #include <sys/sysproto.h>
73 #include <sys/systm.h>
74 #include <sys/sx.h>
75 #include <sys/time.h>
76 #include <sys/vnode.h>
77 #include <sys/unistd.h>
78 #include <sys/user.h>
79 
80 #include <security/mac/mac_framework.h>
81 
82 #include <vm/vm.h>
83 #include <vm/vm_param.h>
84 #include <vm/pmap.h>
85 #include <vm/vm_extern.h>
86 #include <vm/vm_map.h>
87 #include <vm/vm_kern.h>
88 #include <vm/vm_object.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_pageout.h>
91 #include <vm/vm_pager.h>
92 #include <vm/swap_pager.h>
93 
94 struct shm_mapping {
95 	char		*sm_path;
96 	Fnv32_t		sm_fnv;
97 	struct shmfd	*sm_shmfd;
98 	LIST_ENTRY(shm_mapping) sm_link;
99 };
100 
101 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
102 static LIST_HEAD(, shm_mapping) *shm_dictionary;
103 static struct sx shm_dict_lock;
104 static struct mtx shm_timestamp_lock;
105 static u_long shm_hash;
106 static struct unrhdr *shm_ino_unr;
107 static dev_t shm_dev_ino;
108 
109 #define	SHM_HASH(fnv)	(&shm_dictionary[(fnv) & shm_hash])
110 
111 static int	shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags);
112 static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode);
113 static void	shm_init(void *arg);
114 static void	shm_drop(struct shmfd *shmfd);
115 static struct shmfd *shm_hold(struct shmfd *shmfd);
116 static void	shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
117 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
118 static int	shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
119 static int	shm_dotruncate(struct shmfd *shmfd, off_t length);
120 
121 static fo_rdwr_t	shm_read;
122 static fo_rdwr_t	shm_write;
123 static fo_truncate_t	shm_truncate;
124 static fo_stat_t	shm_stat;
125 static fo_close_t	shm_close;
126 static fo_chmod_t	shm_chmod;
127 static fo_chown_t	shm_chown;
128 static fo_seek_t	shm_seek;
129 static fo_fill_kinfo_t	shm_fill_kinfo;
130 
131 /* File descriptor operations. */
132 static struct fileops shm_ops = {
133 	.fo_read = shm_read,
134 	.fo_write = shm_write,
135 	.fo_truncate = shm_truncate,
136 	.fo_ioctl = invfo_ioctl,
137 	.fo_poll = invfo_poll,
138 	.fo_kqfilter = invfo_kqfilter,
139 	.fo_stat = shm_stat,
140 	.fo_close = shm_close,
141 	.fo_chmod = shm_chmod,
142 	.fo_chown = shm_chown,
143 	.fo_sendfile = vn_sendfile,
144 	.fo_seek = shm_seek,
145 	.fo_fill_kinfo = shm_fill_kinfo,
146 	.fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
147 };
148 
149 FEATURE(posix_shm, "POSIX shared memory");
150 
151 static int
152 uiomove_object_page(vm_object_t obj, size_t len, struct uio *uio)
153 {
154 	vm_page_t m;
155 	vm_pindex_t idx;
156 	size_t tlen;
157 	int error, offset, rv;
158 
159 	idx = OFF_TO_IDX(uio->uio_offset);
160 	offset = uio->uio_offset & PAGE_MASK;
161 	tlen = MIN(PAGE_SIZE - offset, len);
162 
163 	VM_OBJECT_WLOCK(obj);
164 
165 	/*
166 	 * Parallel reads of the page content from disk are prevented
167 	 * by exclusive busy.
168 	 *
169 	 * Although the tmpfs vnode lock is held here, it is
170 	 * nonetheless safe to sleep waiting for a free page.  The
171 	 * pageout daemon does not need to acquire the tmpfs vnode
172 	 * lock to page out tobj's pages because tobj is a OBJT_SWAP
173 	 * type object.
174 	 */
175 	m = vm_page_grab(obj, idx, VM_ALLOC_NORMAL);
176 	if (m->valid != VM_PAGE_BITS_ALL) {
177 		if (vm_pager_has_page(obj, idx, NULL, NULL)) {
178 			rv = vm_pager_get_pages(obj, &m, 1, 0);
179 			m = vm_page_lookup(obj, idx);
180 			if (m == NULL) {
181 				printf(
182 		    "uiomove_object: vm_obj %p idx %jd null lookup rv %d\n",
183 				    obj, idx, rv);
184 				VM_OBJECT_WUNLOCK(obj);
185 				return (EIO);
186 			}
187 			if (rv != VM_PAGER_OK) {
188 				printf(
189 	    "uiomove_object: vm_obj %p idx %jd valid %x pager error %d\n",
190 				    obj, idx, m->valid, rv);
191 				vm_page_lock(m);
192 				vm_page_free(m);
193 				vm_page_unlock(m);
194 				VM_OBJECT_WUNLOCK(obj);
195 				return (EIO);
196 			}
197 		} else
198 			vm_page_zero_invalid(m, TRUE);
199 	}
200 	vm_page_xunbusy(m);
201 	vm_page_lock(m);
202 	vm_page_hold(m);
203 	if (m->queue == PQ_NONE) {
204 		vm_page_deactivate(m);
205 	} else {
206 		/* Requeue to maintain LRU ordering. */
207 		vm_page_requeue(m);
208 	}
209 	vm_page_unlock(m);
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_lock(m);
219 	vm_page_unhold(m);
220 	vm_page_unlock(m);
221 
222 	return (error);
223 }
224 
225 int
226 uiomove_object(vm_object_t obj, off_t obj_size, struct uio *uio)
227 {
228 	ssize_t resid;
229 	size_t len;
230 	int error;
231 
232 	error = 0;
233 	while ((resid = uio->uio_resid) > 0) {
234 		if (obj_size <= uio->uio_offset)
235 			break;
236 		len = MIN(obj_size - uio->uio_offset, resid);
237 		if (len == 0)
238 			break;
239 		error = uiomove_object_page(obj, len, uio);
240 		if (error != 0 || resid == uio->uio_resid)
241 			break;
242 	}
243 	return (error);
244 }
245 
246 static int
247 shm_seek(struct file *fp, off_t offset, int whence, struct thread *td)
248 {
249 	struct shmfd *shmfd;
250 	off_t foffset;
251 	int error;
252 
253 	shmfd = fp->f_data;
254 	foffset = foffset_lock(fp, 0);
255 	error = 0;
256 	switch (whence) {
257 	case L_INCR:
258 		if (foffset < 0 ||
259 		    (offset > 0 && foffset > OFF_MAX - offset)) {
260 			error = EOVERFLOW;
261 			break;
262 		}
263 		offset += foffset;
264 		break;
265 	case L_XTND:
266 		if (offset > 0 && shmfd->shm_size > OFF_MAX - offset) {
267 			error = EOVERFLOW;
268 			break;
269 		}
270 		offset += shmfd->shm_size;
271 		break;
272 	case L_SET:
273 		break;
274 	default:
275 		error = EINVAL;
276 	}
277 	if (error == 0) {
278 		if (offset < 0 || offset > shmfd->shm_size)
279 			error = EINVAL;
280 		else
281 			td->td_uretoff.tdu_off = offset;
282 	}
283 	foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
284 	return (error);
285 }
286 
287 static int
288 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
289     int flags, struct thread *td)
290 {
291 	struct shmfd *shmfd;
292 	void *rl_cookie;
293 	int error;
294 
295 	shmfd = fp->f_data;
296 	foffset_lock_uio(fp, uio, flags);
297 	rl_cookie = rangelock_rlock(&shmfd->shm_rl, uio->uio_offset,
298 	    uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx);
299 #ifdef MAC
300 	error = mac_posixshm_check_read(active_cred, fp->f_cred, shmfd);
301 	if (error)
302 		return (error);
303 #endif
304 	error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio);
305 	rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx);
306 	foffset_unlock_uio(fp, uio, flags);
307 	return (error);
308 }
309 
310 static int
311 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
312     int flags, struct thread *td)
313 {
314 	struct shmfd *shmfd;
315 	void *rl_cookie;
316 	int error;
317 
318 	shmfd = fp->f_data;
319 #ifdef MAC
320 	error = mac_posixshm_check_write(active_cred, fp->f_cred, shmfd);
321 	if (error)
322 		return (error);
323 #endif
324 	foffset_lock_uio(fp, uio, flags);
325 	if ((flags & FOF_OFFSET) == 0) {
326 		rl_cookie = rangelock_wlock(&shmfd->shm_rl, 0, OFF_MAX,
327 		    &shmfd->shm_mtx);
328 	} else {
329 		rl_cookie = rangelock_wlock(&shmfd->shm_rl, uio->uio_offset,
330 		    uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx);
331 	}
332 
333 	error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio);
334 	rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx);
335 	foffset_unlock_uio(fp, uio, flags);
336 	return (error);
337 }
338 
339 static int
340 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
341     struct thread *td)
342 {
343 	struct shmfd *shmfd;
344 #ifdef MAC
345 	int error;
346 #endif
347 
348 	shmfd = fp->f_data;
349 #ifdef MAC
350 	error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
351 	if (error)
352 		return (error);
353 #endif
354 	return (shm_dotruncate(shmfd, length));
355 }
356 
357 static int
358 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
359     struct thread *td)
360 {
361 	struct shmfd *shmfd;
362 #ifdef MAC
363 	int error;
364 #endif
365 
366 	shmfd = fp->f_data;
367 
368 #ifdef MAC
369 	error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
370 	if (error)
371 		return (error);
372 #endif
373 
374 	/*
375 	 * Attempt to return sanish values for fstat() on a memory file
376 	 * descriptor.
377 	 */
378 	bzero(sb, sizeof(*sb));
379 	sb->st_blksize = PAGE_SIZE;
380 	sb->st_size = shmfd->shm_size;
381 	sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize;
382 	mtx_lock(&shm_timestamp_lock);
383 	sb->st_atim = shmfd->shm_atime;
384 	sb->st_ctim = shmfd->shm_ctime;
385 	sb->st_mtim = shmfd->shm_mtime;
386 	sb->st_birthtim = shmfd->shm_birthtime;
387 	sb->st_mode = S_IFREG | shmfd->shm_mode;		/* XXX */
388 	sb->st_uid = shmfd->shm_uid;
389 	sb->st_gid = shmfd->shm_gid;
390 	mtx_unlock(&shm_timestamp_lock);
391 	sb->st_dev = shm_dev_ino;
392 	sb->st_ino = shmfd->shm_ino;
393 
394 	return (0);
395 }
396 
397 static int
398 shm_close(struct file *fp, struct thread *td)
399 {
400 	struct shmfd *shmfd;
401 
402 	shmfd = fp->f_data;
403 	fp->f_data = NULL;
404 	shm_drop(shmfd);
405 
406 	return (0);
407 }
408 
409 static int
410 shm_dotruncate(struct shmfd *shmfd, off_t length)
411 {
412 	vm_object_t object;
413 	vm_page_t m, ma[1];
414 	vm_pindex_t idx, nobjsize;
415 	vm_ooffset_t delta;
416 	int base, rv;
417 
418 	object = shmfd->shm_object;
419 	VM_OBJECT_WLOCK(object);
420 	if (length == shmfd->shm_size) {
421 		VM_OBJECT_WUNLOCK(object);
422 		return (0);
423 	}
424 	nobjsize = OFF_TO_IDX(length + PAGE_MASK);
425 
426 	/* Are we shrinking?  If so, trim the end. */
427 	if (length < shmfd->shm_size) {
428 		/*
429 		 * Disallow any requests to shrink the size if this
430 		 * object is mapped into the kernel.
431 		 */
432 		if (shmfd->shm_kmappings > 0) {
433 			VM_OBJECT_WUNLOCK(object);
434 			return (EBUSY);
435 		}
436 
437 		/*
438 		 * Zero the truncated part of the last page.
439 		 */
440 		base = length & PAGE_MASK;
441 		if (base != 0) {
442 			idx = OFF_TO_IDX(length);
443 retry:
444 			m = vm_page_lookup(object, idx);
445 			if (m != NULL) {
446 				if (vm_page_sleep_if_busy(m, "shmtrc"))
447 					goto retry;
448 			} else if (vm_pager_has_page(object, idx, NULL, NULL)) {
449 				m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL);
450 				if (m == NULL) {
451 					VM_OBJECT_WUNLOCK(object);
452 					VM_WAIT;
453 					VM_OBJECT_WLOCK(object);
454 					goto retry;
455 				} else if (m->valid != VM_PAGE_BITS_ALL) {
456 					ma[0] = m;
457 					rv = vm_pager_get_pages(object, ma, 1,
458 					    0);
459 					m = vm_page_lookup(object, idx);
460 				} else
461 					/* A cached page was reactivated. */
462 					rv = VM_PAGER_OK;
463 				vm_page_lock(m);
464 				if (rv == VM_PAGER_OK) {
465 					vm_page_deactivate(m);
466 					vm_page_unlock(m);
467 					vm_page_xunbusy(m);
468 				} else {
469 					vm_page_free(m);
470 					vm_page_unlock(m);
471 					VM_OBJECT_WUNLOCK(object);
472 					return (EIO);
473 				}
474 			}
475 			if (m != NULL) {
476 				pmap_zero_page_area(m, base, PAGE_SIZE - base);
477 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
478 				    ("shm_dotruncate: page %p is invalid", m));
479 				vm_page_dirty(m);
480 				vm_pager_page_unswapped(m);
481 			}
482 		}
483 		delta = ptoa(object->size - nobjsize);
484 
485 		/* Toss in memory pages. */
486 		if (nobjsize < object->size)
487 			vm_object_page_remove(object, nobjsize, object->size,
488 			    0);
489 
490 		/* Toss pages from swap. */
491 		if (object->type == OBJT_SWAP)
492 			swap_pager_freespace(object, nobjsize, delta);
493 
494 		/* Free the swap accounted for shm */
495 		swap_release_by_cred(delta, object->cred);
496 		object->charge -= delta;
497 	} else {
498 		/* Attempt to reserve the swap */
499 		delta = ptoa(nobjsize - object->size);
500 		if (!swap_reserve_by_cred(delta, object->cred)) {
501 			VM_OBJECT_WUNLOCK(object);
502 			return (ENOMEM);
503 		}
504 		object->charge += delta;
505 	}
506 	shmfd->shm_size = length;
507 	mtx_lock(&shm_timestamp_lock);
508 	vfs_timestamp(&shmfd->shm_ctime);
509 	shmfd->shm_mtime = shmfd->shm_ctime;
510 	mtx_unlock(&shm_timestamp_lock);
511 	object->size = nobjsize;
512 	VM_OBJECT_WUNLOCK(object);
513 	return (0);
514 }
515 
516 /*
517  * shmfd object management including creation and reference counting
518  * routines.
519  */
520 static struct shmfd *
521 shm_alloc(struct ucred *ucred, mode_t mode)
522 {
523 	struct shmfd *shmfd;
524 	int ino;
525 
526 	shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
527 	shmfd->shm_size = 0;
528 	shmfd->shm_uid = ucred->cr_uid;
529 	shmfd->shm_gid = ucred->cr_gid;
530 	shmfd->shm_mode = mode;
531 	shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
532 	    shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
533 	KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
534 	shmfd->shm_object->pg_color = 0;
535 	VM_OBJECT_WLOCK(shmfd->shm_object);
536 	vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
537 	vm_object_set_flag(shmfd->shm_object, OBJ_COLORED | OBJ_NOSPLIT);
538 	VM_OBJECT_WUNLOCK(shmfd->shm_object);
539 	vfs_timestamp(&shmfd->shm_birthtime);
540 	shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
541 	    shmfd->shm_birthtime;
542 	ino = alloc_unr(shm_ino_unr);
543 	if (ino == -1)
544 		shmfd->shm_ino = 0;
545 	else
546 		shmfd->shm_ino = ino;
547 	refcount_init(&shmfd->shm_refs, 1);
548 	mtx_init(&shmfd->shm_mtx, "shmrl", NULL, MTX_DEF);
549 	rangelock_init(&shmfd->shm_rl);
550 #ifdef MAC
551 	mac_posixshm_init(shmfd);
552 	mac_posixshm_create(ucred, shmfd);
553 #endif
554 
555 	return (shmfd);
556 }
557 
558 static struct shmfd *
559 shm_hold(struct shmfd *shmfd)
560 {
561 
562 	refcount_acquire(&shmfd->shm_refs);
563 	return (shmfd);
564 }
565 
566 static void
567 shm_drop(struct shmfd *shmfd)
568 {
569 
570 	if (refcount_release(&shmfd->shm_refs)) {
571 #ifdef MAC
572 		mac_posixshm_destroy(shmfd);
573 #endif
574 		rangelock_destroy(&shmfd->shm_rl);
575 		mtx_destroy(&shmfd->shm_mtx);
576 		vm_object_deallocate(shmfd->shm_object);
577 		if (shmfd->shm_ino != 0)
578 			free_unr(shm_ino_unr, shmfd->shm_ino);
579 		free(shmfd, M_SHMFD);
580 	}
581 }
582 
583 /*
584  * Determine if the credentials have sufficient permissions for a
585  * specified combination of FREAD and FWRITE.
586  */
587 static int
588 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
589 {
590 	accmode_t accmode;
591 	int error;
592 
593 	accmode = 0;
594 	if (flags & FREAD)
595 		accmode |= VREAD;
596 	if (flags & FWRITE)
597 		accmode |= VWRITE;
598 	mtx_lock(&shm_timestamp_lock);
599 	error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
600 	    accmode, ucred, NULL);
601 	mtx_unlock(&shm_timestamp_lock);
602 	return (error);
603 }
604 
605 /*
606  * Dictionary management.  We maintain an in-kernel dictionary to map
607  * paths to shmfd objects.  We use the FNV hash on the path to store
608  * the mappings in a hash table.
609  */
610 static void
611 shm_init(void *arg)
612 {
613 
614 	mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
615 	sx_init(&shm_dict_lock, "shm dictionary");
616 	shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
617 	shm_ino_unr = new_unrhdr(1, INT32_MAX, NULL);
618 	KASSERT(shm_ino_unr != NULL, ("shm fake inodes not initialized"));
619 	shm_dev_ino = devfs_alloc_cdp_inode();
620 	KASSERT(shm_dev_ino > 0, ("shm dev inode not initialized"));
621 }
622 SYSINIT(shm_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_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 #ifdef KTRACE
730 		if (error == 0 && KTRPOINT(curthread, KTR_NAMEI))
731 			ktrnamei(path);
732 #endif
733 		/* Require paths to start with a '/' character. */
734 		if (error == 0 && path[0] != '/')
735 			error = EINVAL;
736 		if (error) {
737 			fdclose(fdp, fp, fd, td);
738 			fdrop(fp, td);
739 			free(path, M_SHMFD);
740 			return (error);
741 		}
742 
743 		fnv = fnv_32_str(path, FNV1_32_INIT);
744 		sx_xlock(&shm_dict_lock);
745 		shmfd = shm_lookup(path, fnv);
746 		if (shmfd == NULL) {
747 			/* Object does not yet exist, create it if requested. */
748 			if (uap->flags & O_CREAT) {
749 #ifdef MAC
750 				error = mac_posixshm_check_create(td->td_ucred,
751 				    path);
752 				if (error == 0) {
753 #endif
754 					shmfd = shm_alloc(td->td_ucred, cmode);
755 					shm_insert(path, fnv, shmfd);
756 #ifdef MAC
757 				}
758 #endif
759 			} else {
760 				free(path, M_SHMFD);
761 				error = ENOENT;
762 			}
763 		} else {
764 			/*
765 			 * Object already exists, obtain a new
766 			 * reference if requested and permitted.
767 			 */
768 			free(path, M_SHMFD);
769 			if ((uap->flags & (O_CREAT | O_EXCL)) ==
770 			    (O_CREAT | O_EXCL))
771 				error = EEXIST;
772 			else {
773 #ifdef MAC
774 				error = mac_posixshm_check_open(td->td_ucred,
775 				    shmfd, FFLAGS(uap->flags & O_ACCMODE));
776 				if (error == 0)
777 #endif
778 				error = shm_access(shmfd, td->td_ucred,
779 				    FFLAGS(uap->flags & O_ACCMODE));
780 			}
781 
782 			/*
783 			 * Truncate the file back to zero length if
784 			 * O_TRUNC was specified and the object was
785 			 * opened with read/write.
786 			 */
787 			if (error == 0 &&
788 			    (uap->flags & (O_ACCMODE | O_TRUNC)) ==
789 			    (O_RDWR | O_TRUNC)) {
790 #ifdef MAC
791 				error = mac_posixshm_check_truncate(
792 					td->td_ucred, fp->f_cred, shmfd);
793 				if (error == 0)
794 #endif
795 					shm_dotruncate(shmfd, 0);
796 			}
797 			if (error == 0)
798 				shm_hold(shmfd);
799 		}
800 		sx_xunlock(&shm_dict_lock);
801 
802 		if (error) {
803 			fdclose(fdp, fp, fd, td);
804 			fdrop(fp, td);
805 			return (error);
806 		}
807 	}
808 
809 	finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
810 
811 	td->td_retval[0] = fd;
812 	fdrop(fp, td);
813 
814 	return (0);
815 }
816 
817 int
818 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
819 {
820 	char *path;
821 	Fnv32_t fnv;
822 	int error;
823 
824 	path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
825 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
826 	if (error) {
827 		free(path, M_TEMP);
828 		return (error);
829 	}
830 #ifdef KTRACE
831 	if (KTRPOINT(curthread, KTR_NAMEI))
832 		ktrnamei(path);
833 #endif
834 	fnv = fnv_32_str(path, FNV1_32_INIT);
835 	sx_xlock(&shm_dict_lock);
836 	error = shm_remove(path, fnv, td->td_ucred);
837 	sx_xunlock(&shm_dict_lock);
838 	free(path, M_TEMP);
839 
840 	return (error);
841 }
842 
843 /*
844  * mmap() helper to validate mmap() requests against shm object state
845  * and give mmap() the vm_object to use for the mapping.
846  */
847 int
848 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
849     vm_object_t *obj)
850 {
851 
852 	/*
853 	 * XXXRW: This validation is probably insufficient, and subject to
854 	 * sign errors.  It should be fixed.
855 	 */
856 	if (foff >= shmfd->shm_size ||
857 	    foff + objsize > round_page(shmfd->shm_size))
858 		return (EINVAL);
859 
860 	mtx_lock(&shm_timestamp_lock);
861 	vfs_timestamp(&shmfd->shm_atime);
862 	mtx_unlock(&shm_timestamp_lock);
863 	vm_object_reference(shmfd->shm_object);
864 	*obj = shmfd->shm_object;
865 	return (0);
866 }
867 
868 static int
869 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
870     struct thread *td)
871 {
872 	struct shmfd *shmfd;
873 	int error;
874 
875 	error = 0;
876 	shmfd = fp->f_data;
877 	mtx_lock(&shm_timestamp_lock);
878 	/*
879 	 * SUSv4 says that x bits of permission need not be affected.
880 	 * Be consistent with our shm_open there.
881 	 */
882 #ifdef MAC
883 	error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
884 	if (error != 0)
885 		goto out;
886 #endif
887 	error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
888 	    shmfd->shm_gid, VADMIN, active_cred, NULL);
889 	if (error != 0)
890 		goto out;
891 	shmfd->shm_mode = mode & ACCESSPERMS;
892 out:
893 	mtx_unlock(&shm_timestamp_lock);
894 	return (error);
895 }
896 
897 static int
898 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
899     struct thread *td)
900 {
901 	struct shmfd *shmfd;
902 	int error;
903 
904 	error = 0;
905 	shmfd = fp->f_data;
906 	mtx_lock(&shm_timestamp_lock);
907 #ifdef MAC
908 	error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
909 	if (error != 0)
910 		goto out;
911 #endif
912 	if (uid == (uid_t)-1)
913 		uid = shmfd->shm_uid;
914 	if (gid == (gid_t)-1)
915                  gid = shmfd->shm_gid;
916 	if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
917 	    (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
918 	    (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
919 		goto out;
920 	shmfd->shm_uid = uid;
921 	shmfd->shm_gid = gid;
922 out:
923 	mtx_unlock(&shm_timestamp_lock);
924 	return (error);
925 }
926 
927 /*
928  * Helper routines to allow the backing object of a shared memory file
929  * descriptor to be mapped in the kernel.
930  */
931 int
932 shm_map(struct file *fp, size_t size, off_t offset, void **memp)
933 {
934 	struct shmfd *shmfd;
935 	vm_offset_t kva, ofs;
936 	vm_object_t obj;
937 	int rv;
938 
939 	if (fp->f_type != DTYPE_SHM)
940 		return (EINVAL);
941 	shmfd = fp->f_data;
942 	obj = shmfd->shm_object;
943 	VM_OBJECT_WLOCK(obj);
944 	/*
945 	 * XXXRW: This validation is probably insufficient, and subject to
946 	 * sign errors.  It should be fixed.
947 	 */
948 	if (offset >= shmfd->shm_size ||
949 	    offset + size > round_page(shmfd->shm_size)) {
950 		VM_OBJECT_WUNLOCK(obj);
951 		return (EINVAL);
952 	}
953 
954 	shmfd->shm_kmappings++;
955 	vm_object_reference_locked(obj);
956 	VM_OBJECT_WUNLOCK(obj);
957 
958 	/* Map the object into the kernel_map and wire it. */
959 	kva = vm_map_min(kernel_map);
960 	ofs = offset & PAGE_MASK;
961 	offset = trunc_page(offset);
962 	size = round_page(size + ofs);
963 	rv = vm_map_find(kernel_map, obj, offset, &kva, size, 0,
964 	    VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE,
965 	    VM_PROT_READ | VM_PROT_WRITE, 0);
966 	if (rv == KERN_SUCCESS) {
967 		rv = vm_map_wire(kernel_map, kva, kva + size,
968 		    VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
969 		if (rv == KERN_SUCCESS) {
970 			*memp = (void *)(kva + ofs);
971 			return (0);
972 		}
973 		vm_map_remove(kernel_map, kva, kva + size);
974 	} else
975 		vm_object_deallocate(obj);
976 
977 	/* On failure, drop our mapping reference. */
978 	VM_OBJECT_WLOCK(obj);
979 	shmfd->shm_kmappings--;
980 	VM_OBJECT_WUNLOCK(obj);
981 
982 	return (vm_mmap_to_errno(rv));
983 }
984 
985 /*
986  * We require the caller to unmap the entire entry.  This allows us to
987  * safely decrement shm_kmappings when a mapping is removed.
988  */
989 int
990 shm_unmap(struct file *fp, void *mem, size_t size)
991 {
992 	struct shmfd *shmfd;
993 	vm_map_entry_t entry;
994 	vm_offset_t kva, ofs;
995 	vm_object_t obj;
996 	vm_pindex_t pindex;
997 	vm_prot_t prot;
998 	boolean_t wired;
999 	vm_map_t map;
1000 	int rv;
1001 
1002 	if (fp->f_type != DTYPE_SHM)
1003 		return (EINVAL);
1004 	shmfd = fp->f_data;
1005 	kva = (vm_offset_t)mem;
1006 	ofs = kva & PAGE_MASK;
1007 	kva = trunc_page(kva);
1008 	size = round_page(size + ofs);
1009 	map = kernel_map;
1010 	rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
1011 	    &obj, &pindex, &prot, &wired);
1012 	if (rv != KERN_SUCCESS)
1013 		return (EINVAL);
1014 	if (entry->start != kva || entry->end != kva + size) {
1015 		vm_map_lookup_done(map, entry);
1016 		return (EINVAL);
1017 	}
1018 	vm_map_lookup_done(map, entry);
1019 	if (obj != shmfd->shm_object)
1020 		return (EINVAL);
1021 	vm_map_remove(map, kva, kva + size);
1022 	VM_OBJECT_WLOCK(obj);
1023 	KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
1024 	shmfd->shm_kmappings--;
1025 	VM_OBJECT_WUNLOCK(obj);
1026 	return (0);
1027 }
1028 
1029 static int
1030 shm_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
1031 {
1032 	struct shmfd *shmfd;
1033 
1034 	kif->kf_type = KF_TYPE_SHM;
1035 	shmfd = fp->f_data;
1036 
1037 	mtx_lock(&shm_timestamp_lock);
1038 	kif->kf_un.kf_file.kf_file_mode = S_IFREG | shmfd->shm_mode;	/* XXX */
1039 	mtx_unlock(&shm_timestamp_lock);
1040 	kif->kf_un.kf_file.kf_file_size = shmfd->shm_size;
1041 	if (shmfd->shm_path != NULL) {
1042 		sx_slock(&shm_dict_lock);
1043 		if (shmfd->shm_path != NULL)
1044 			strlcpy(kif->kf_path, shmfd->shm_path,
1045 			    sizeof(kif->kf_path));
1046 		sx_sunlock(&shm_dict_lock);
1047 	}
1048 	return (0);
1049 }
1050