xref: /freebsd/sys/kern/uipc_shm.c (revision 7f9dff23d3092aa33ad45b2b63e52469b3c13a6e)
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 				}
459 				rv = vm_pager_get_pages(object, &m, 1, NULL,
460 				    NULL);
461 				vm_page_lock(m);
462 				if (rv == VM_PAGER_OK) {
463 					vm_page_deactivate(m);
464 					vm_page_unlock(m);
465 					vm_page_xunbusy(m);
466 				} else {
467 					vm_page_free(m);
468 					vm_page_unlock(m);
469 					VM_OBJECT_WUNLOCK(object);
470 					return (EIO);
471 				}
472 			}
473 			if (m != NULL) {
474 				pmap_zero_page_area(m, base, PAGE_SIZE - base);
475 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
476 				    ("shm_dotruncate: page %p is invalid", m));
477 				vm_page_dirty(m);
478 				vm_pager_page_unswapped(m);
479 			}
480 		}
481 		delta = ptoa(object->size - nobjsize);
482 
483 		/* Toss in memory pages. */
484 		if (nobjsize < object->size)
485 			vm_object_page_remove(object, nobjsize, object->size,
486 			    0);
487 
488 		/* Toss pages from swap. */
489 		if (object->type == OBJT_SWAP)
490 			swap_pager_freespace(object, nobjsize, delta);
491 
492 		/* Free the swap accounted for shm */
493 		swap_release_by_cred(delta, object->cred);
494 		object->charge -= delta;
495 	} else {
496 		/* Attempt to reserve the swap */
497 		delta = ptoa(nobjsize - object->size);
498 		if (!swap_reserve_by_cred(delta, object->cred)) {
499 			VM_OBJECT_WUNLOCK(object);
500 			return (ENOMEM);
501 		}
502 		object->charge += delta;
503 	}
504 	shmfd->shm_size = length;
505 	mtx_lock(&shm_timestamp_lock);
506 	vfs_timestamp(&shmfd->shm_ctime);
507 	shmfd->shm_mtime = shmfd->shm_ctime;
508 	mtx_unlock(&shm_timestamp_lock);
509 	object->size = nobjsize;
510 	VM_OBJECT_WUNLOCK(object);
511 	return (0);
512 }
513 
514 /*
515  * shmfd object management including creation and reference counting
516  * routines.
517  */
518 struct shmfd *
519 shm_alloc(struct ucred *ucred, mode_t mode)
520 {
521 	struct shmfd *shmfd;
522 	int ino;
523 
524 	shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
525 	shmfd->shm_size = 0;
526 	shmfd->shm_uid = ucred->cr_uid;
527 	shmfd->shm_gid = ucred->cr_gid;
528 	shmfd->shm_mode = mode;
529 	shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
530 	    shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
531 	KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
532 	shmfd->shm_object->pg_color = 0;
533 	VM_OBJECT_WLOCK(shmfd->shm_object);
534 	vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
535 	vm_object_set_flag(shmfd->shm_object, OBJ_COLORED | OBJ_NOSPLIT);
536 	VM_OBJECT_WUNLOCK(shmfd->shm_object);
537 	vfs_timestamp(&shmfd->shm_birthtime);
538 	shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
539 	    shmfd->shm_birthtime;
540 	ino = alloc_unr(shm_ino_unr);
541 	if (ino == -1)
542 		shmfd->shm_ino = 0;
543 	else
544 		shmfd->shm_ino = ino;
545 	refcount_init(&shmfd->shm_refs, 1);
546 	mtx_init(&shmfd->shm_mtx, "shmrl", NULL, MTX_DEF);
547 	rangelock_init(&shmfd->shm_rl);
548 #ifdef MAC
549 	mac_posixshm_init(shmfd);
550 	mac_posixshm_create(ucred, shmfd);
551 #endif
552 
553 	return (shmfd);
554 }
555 
556 struct shmfd *
557 shm_hold(struct shmfd *shmfd)
558 {
559 
560 	refcount_acquire(&shmfd->shm_refs);
561 	return (shmfd);
562 }
563 
564 void
565 shm_drop(struct shmfd *shmfd)
566 {
567 
568 	if (refcount_release(&shmfd->shm_refs)) {
569 #ifdef MAC
570 		mac_posixshm_destroy(shmfd);
571 #endif
572 		rangelock_destroy(&shmfd->shm_rl);
573 		mtx_destroy(&shmfd->shm_mtx);
574 		vm_object_deallocate(shmfd->shm_object);
575 		if (shmfd->shm_ino != 0)
576 			free_unr(shm_ino_unr, shmfd->shm_ino);
577 		free(shmfd, M_SHMFD);
578 	}
579 }
580 
581 /*
582  * Determine if the credentials have sufficient permissions for a
583  * specified combination of FREAD and FWRITE.
584  */
585 int
586 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
587 {
588 	accmode_t accmode;
589 	int error;
590 
591 	accmode = 0;
592 	if (flags & FREAD)
593 		accmode |= VREAD;
594 	if (flags & FWRITE)
595 		accmode |= VWRITE;
596 	mtx_lock(&shm_timestamp_lock);
597 	error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
598 	    accmode, ucred, NULL);
599 	mtx_unlock(&shm_timestamp_lock);
600 	return (error);
601 }
602 
603 /*
604  * Dictionary management.  We maintain an in-kernel dictionary to map
605  * paths to shmfd objects.  We use the FNV hash on the path to store
606  * the mappings in a hash table.
607  */
608 static void
609 shm_init(void *arg)
610 {
611 
612 	mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
613 	sx_init(&shm_dict_lock, "shm dictionary");
614 	shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
615 	shm_ino_unr = new_unrhdr(1, INT32_MAX, NULL);
616 	KASSERT(shm_ino_unr != NULL, ("shm fake inodes not initialized"));
617 	shm_dev_ino = devfs_alloc_cdp_inode();
618 	KASSERT(shm_dev_ino > 0, ("shm dev inode not initialized"));
619 }
620 SYSINIT(shm_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_init, NULL);
621 
622 static struct shmfd *
623 shm_lookup(char *path, Fnv32_t fnv)
624 {
625 	struct shm_mapping *map;
626 
627 	LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
628 		if (map->sm_fnv != fnv)
629 			continue;
630 		if (strcmp(map->sm_path, path) == 0)
631 			return (map->sm_shmfd);
632 	}
633 
634 	return (NULL);
635 }
636 
637 static void
638 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
639 {
640 	struct shm_mapping *map;
641 
642 	map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
643 	map->sm_path = path;
644 	map->sm_fnv = fnv;
645 	map->sm_shmfd = shm_hold(shmfd);
646 	shmfd->shm_path = path;
647 	LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
648 }
649 
650 static int
651 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
652 {
653 	struct shm_mapping *map;
654 	int error;
655 
656 	LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
657 		if (map->sm_fnv != fnv)
658 			continue;
659 		if (strcmp(map->sm_path, path) == 0) {
660 #ifdef MAC
661 			error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
662 			if (error)
663 				return (error);
664 #endif
665 			error = shm_access(map->sm_shmfd, ucred,
666 			    FREAD | FWRITE);
667 			if (error)
668 				return (error);
669 			map->sm_shmfd->shm_path = NULL;
670 			LIST_REMOVE(map, sm_link);
671 			shm_drop(map->sm_shmfd);
672 			free(map->sm_path, M_SHMFD);
673 			free(map, M_SHMFD);
674 			return (0);
675 		}
676 	}
677 
678 	return (ENOENT);
679 }
680 
681 int
682 kern_shm_open(struct thread *td, const char *userpath, int flags, mode_t mode,
683     struct filecaps *fcaps)
684 {
685 	struct filedesc *fdp;
686 	struct shmfd *shmfd;
687 	struct file *fp;
688 	char *path;
689 	const char *pr_path;
690 	size_t pr_pathlen;
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) && (userpath != SHM_ANON))
700 		return (ECAPMODE);
701 #endif
702 
703 	if ((flags & O_ACCMODE) != O_RDONLY && (flags & O_ACCMODE) != O_RDWR)
704 		return (EINVAL);
705 
706 	if ((flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC | O_CLOEXEC)) != 0)
707 		return (EINVAL);
708 
709 	fdp = td->td_proc->p_fd;
710 	cmode = (mode & ~fdp->fd_cmask) & ACCESSPERMS;
711 
712 	error = falloc_caps(td, &fp, &fd, O_CLOEXEC, fcaps);
713 	if (error)
714 		return (error);
715 
716 	/* A SHM_ANON path pointer creates an anonymous object. */
717 	if (userpath == SHM_ANON) {
718 		/* A read-only anonymous object is pointless. */
719 		if ((flags & O_ACCMODE) == O_RDONLY) {
720 			fdclose(td, fp, fd);
721 			fdrop(fp, td);
722 			return (EINVAL);
723 		}
724 		shmfd = shm_alloc(td->td_ucred, cmode);
725 	} else {
726 		path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
727 		pr_path = td->td_ucred->cr_prison->pr_path;
728 
729 		/* Construct a full pathname for jailed callers. */
730 		pr_pathlen = strcmp(pr_path, "/") == 0 ? 0
731 		    : strlcpy(path, pr_path, MAXPATHLEN);
732 		error = copyinstr(userpath, path + pr_pathlen,
733 		    MAXPATHLEN - pr_pathlen, NULL);
734 #ifdef KTRACE
735 		if (error == 0 && KTRPOINT(curthread, KTR_NAMEI))
736 			ktrnamei(path);
737 #endif
738 		/* Require paths to start with a '/' character. */
739 		if (error == 0 && path[pr_pathlen] != '/')
740 			error = EINVAL;
741 		if (error) {
742 			fdclose(td, fp, fd);
743 			fdrop(fp, td);
744 			free(path, M_SHMFD);
745 			return (error);
746 		}
747 
748 		fnv = fnv_32_str(path, FNV1_32_INIT);
749 		sx_xlock(&shm_dict_lock);
750 		shmfd = shm_lookup(path, fnv);
751 		if (shmfd == NULL) {
752 			/* Object does not yet exist, create it if requested. */
753 			if (flags & O_CREAT) {
754 #ifdef MAC
755 				error = mac_posixshm_check_create(td->td_ucred,
756 				    path);
757 				if (error == 0) {
758 #endif
759 					shmfd = shm_alloc(td->td_ucred, cmode);
760 					shm_insert(path, fnv, shmfd);
761 #ifdef MAC
762 				}
763 #endif
764 			} else {
765 				free(path, M_SHMFD);
766 				error = ENOENT;
767 			}
768 		} else {
769 			/*
770 			 * Object already exists, obtain a new
771 			 * reference if requested and permitted.
772 			 */
773 			free(path, M_SHMFD);
774 			if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
775 				error = EEXIST;
776 			else {
777 #ifdef MAC
778 				error = mac_posixshm_check_open(td->td_ucred,
779 				    shmfd, FFLAGS(flags & O_ACCMODE));
780 				if (error == 0)
781 #endif
782 				error = shm_access(shmfd, td->td_ucred,
783 				    FFLAGS(flags & O_ACCMODE));
784 			}
785 
786 			/*
787 			 * Truncate the file back to zero length if
788 			 * O_TRUNC was specified and the object was
789 			 * opened with read/write.
790 			 */
791 			if (error == 0 &&
792 			    (flags & (O_ACCMODE | O_TRUNC)) ==
793 			    (O_RDWR | O_TRUNC)) {
794 #ifdef MAC
795 				error = mac_posixshm_check_truncate(
796 					td->td_ucred, fp->f_cred, shmfd);
797 				if (error == 0)
798 #endif
799 					shm_dotruncate(shmfd, 0);
800 			}
801 			if (error == 0)
802 				shm_hold(shmfd);
803 		}
804 		sx_xunlock(&shm_dict_lock);
805 
806 		if (error) {
807 			fdclose(td, fp, fd);
808 			fdrop(fp, td);
809 			return (error);
810 		}
811 	}
812 
813 	finit(fp, FFLAGS(flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
814 
815 	td->td_retval[0] = fd;
816 	fdrop(fp, td);
817 
818 	return (0);
819 }
820 
821 /* System calls. */
822 int
823 sys_shm_open(struct thread *td, struct shm_open_args *uap)
824 {
825 
826 	return (kern_shm_open(td, uap->path, uap->flags, uap->mode, NULL));
827 }
828 
829 int
830 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
831 {
832 	char *path;
833 	const char *pr_path;
834 	size_t pr_pathlen;
835 	Fnv32_t fnv;
836 	int error;
837 
838 	path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
839 	pr_path = td->td_ucred->cr_prison->pr_path;
840 	pr_pathlen = strcmp(pr_path, "/") == 0 ? 0
841 	    : strlcpy(path, pr_path, MAXPATHLEN);
842 	error = copyinstr(uap->path, path + pr_pathlen, MAXPATHLEN - pr_pathlen,
843 	    NULL);
844 	if (error) {
845 		free(path, M_TEMP);
846 		return (error);
847 	}
848 #ifdef KTRACE
849 	if (KTRPOINT(curthread, KTR_NAMEI))
850 		ktrnamei(path);
851 #endif
852 	fnv = fnv_32_str(path, FNV1_32_INIT);
853 	sx_xlock(&shm_dict_lock);
854 	error = shm_remove(path, fnv, td->td_ucred);
855 	sx_xunlock(&shm_dict_lock);
856 	free(path, M_TEMP);
857 
858 	return (error);
859 }
860 
861 int
862 shm_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t objsize,
863     vm_prot_t prot, vm_prot_t cap_maxprot, int flags,
864     vm_ooffset_t foff, struct thread *td)
865 {
866 	struct shmfd *shmfd;
867 	vm_prot_t maxprot;
868 	int error;
869 
870 	shmfd = fp->f_data;
871 	maxprot = VM_PROT_NONE;
872 
873 	/* FREAD should always be set. */
874 	if ((fp->f_flag & FREAD) != 0)
875 		maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
876 	if ((fp->f_flag & FWRITE) != 0)
877 		maxprot |= VM_PROT_WRITE;
878 
879 	/* Don't permit shared writable mappings on read-only descriptors. */
880 	if ((flags & MAP_SHARED) != 0 &&
881 	    (maxprot & VM_PROT_WRITE) == 0 &&
882 	    (prot & VM_PROT_WRITE) != 0)
883 		return (EACCES);
884 	maxprot &= cap_maxprot;
885 
886 #ifdef MAC
887 	error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, flags);
888 	if (error != 0)
889 		return (error);
890 #endif
891 
892 	/*
893 	 * XXXRW: This validation is probably insufficient, and subject to
894 	 * sign errors.  It should be fixed.
895 	 */
896 	if (foff >= shmfd->shm_size ||
897 	    foff + objsize > round_page(shmfd->shm_size))
898 		return (EINVAL);
899 
900 	mtx_lock(&shm_timestamp_lock);
901 	vfs_timestamp(&shmfd->shm_atime);
902 	mtx_unlock(&shm_timestamp_lock);
903 	vm_object_reference(shmfd->shm_object);
904 
905 	error = vm_mmap_object(map, addr, objsize, prot, maxprot, flags,
906 	    shmfd->shm_object, foff, FALSE, td);
907 	if (error != 0)
908 		vm_object_deallocate(shmfd->shm_object);
909 	return (0);
910 }
911 
912 static int
913 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
914     struct thread *td)
915 {
916 	struct shmfd *shmfd;
917 	int error;
918 
919 	error = 0;
920 	shmfd = fp->f_data;
921 	mtx_lock(&shm_timestamp_lock);
922 	/*
923 	 * SUSv4 says that x bits of permission need not be affected.
924 	 * Be consistent with our shm_open there.
925 	 */
926 #ifdef MAC
927 	error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
928 	if (error != 0)
929 		goto out;
930 #endif
931 	error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
932 	    shmfd->shm_gid, VADMIN, active_cred, NULL);
933 	if (error != 0)
934 		goto out;
935 	shmfd->shm_mode = mode & ACCESSPERMS;
936 out:
937 	mtx_unlock(&shm_timestamp_lock);
938 	return (error);
939 }
940 
941 static int
942 shm_chown(struct file *fp, uid_t uid, gid_t gid, 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 #ifdef MAC
952 	error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
953 	if (error != 0)
954 		goto out;
955 #endif
956 	if (uid == (uid_t)-1)
957 		uid = shmfd->shm_uid;
958 	if (gid == (gid_t)-1)
959                  gid = shmfd->shm_gid;
960 	if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
961 	    (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
962 	    (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
963 		goto out;
964 	shmfd->shm_uid = uid;
965 	shmfd->shm_gid = gid;
966 out:
967 	mtx_unlock(&shm_timestamp_lock);
968 	return (error);
969 }
970 
971 /*
972  * Helper routines to allow the backing object of a shared memory file
973  * descriptor to be mapped in the kernel.
974  */
975 int
976 shm_map(struct file *fp, size_t size, off_t offset, void **memp)
977 {
978 	struct shmfd *shmfd;
979 	vm_offset_t kva, ofs;
980 	vm_object_t obj;
981 	int rv;
982 
983 	if (fp->f_type != DTYPE_SHM)
984 		return (EINVAL);
985 	shmfd = fp->f_data;
986 	obj = shmfd->shm_object;
987 	VM_OBJECT_WLOCK(obj);
988 	/*
989 	 * XXXRW: This validation is probably insufficient, and subject to
990 	 * sign errors.  It should be fixed.
991 	 */
992 	if (offset >= shmfd->shm_size ||
993 	    offset + size > round_page(shmfd->shm_size)) {
994 		VM_OBJECT_WUNLOCK(obj);
995 		return (EINVAL);
996 	}
997 
998 	shmfd->shm_kmappings++;
999 	vm_object_reference_locked(obj);
1000 	VM_OBJECT_WUNLOCK(obj);
1001 
1002 	/* Map the object into the kernel_map and wire it. */
1003 	kva = vm_map_min(kernel_map);
1004 	ofs = offset & PAGE_MASK;
1005 	offset = trunc_page(offset);
1006 	size = round_page(size + ofs);
1007 	rv = vm_map_find(kernel_map, obj, offset, &kva, size, 0,
1008 	    VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE,
1009 	    VM_PROT_READ | VM_PROT_WRITE, 0);
1010 	if (rv == KERN_SUCCESS) {
1011 		rv = vm_map_wire(kernel_map, kva, kva + size,
1012 		    VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
1013 		if (rv == KERN_SUCCESS) {
1014 			*memp = (void *)(kva + ofs);
1015 			return (0);
1016 		}
1017 		vm_map_remove(kernel_map, kva, kva + size);
1018 	} else
1019 		vm_object_deallocate(obj);
1020 
1021 	/* On failure, drop our mapping reference. */
1022 	VM_OBJECT_WLOCK(obj);
1023 	shmfd->shm_kmappings--;
1024 	VM_OBJECT_WUNLOCK(obj);
1025 
1026 	return (vm_mmap_to_errno(rv));
1027 }
1028 
1029 /*
1030  * We require the caller to unmap the entire entry.  This allows us to
1031  * safely decrement shm_kmappings when a mapping is removed.
1032  */
1033 int
1034 shm_unmap(struct file *fp, void *mem, size_t size)
1035 {
1036 	struct shmfd *shmfd;
1037 	vm_map_entry_t entry;
1038 	vm_offset_t kva, ofs;
1039 	vm_object_t obj;
1040 	vm_pindex_t pindex;
1041 	vm_prot_t prot;
1042 	boolean_t wired;
1043 	vm_map_t map;
1044 	int rv;
1045 
1046 	if (fp->f_type != DTYPE_SHM)
1047 		return (EINVAL);
1048 	shmfd = fp->f_data;
1049 	kva = (vm_offset_t)mem;
1050 	ofs = kva & PAGE_MASK;
1051 	kva = trunc_page(kva);
1052 	size = round_page(size + ofs);
1053 	map = kernel_map;
1054 	rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
1055 	    &obj, &pindex, &prot, &wired);
1056 	if (rv != KERN_SUCCESS)
1057 		return (EINVAL);
1058 	if (entry->start != kva || entry->end != kva + size) {
1059 		vm_map_lookup_done(map, entry);
1060 		return (EINVAL);
1061 	}
1062 	vm_map_lookup_done(map, entry);
1063 	if (obj != shmfd->shm_object)
1064 		return (EINVAL);
1065 	vm_map_remove(map, kva, kva + size);
1066 	VM_OBJECT_WLOCK(obj);
1067 	KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
1068 	shmfd->shm_kmappings--;
1069 	VM_OBJECT_WUNLOCK(obj);
1070 	return (0);
1071 }
1072 
1073 static int
1074 shm_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
1075 {
1076 	const char *path, *pr_path;
1077 	struct shmfd *shmfd;
1078 	size_t pr_pathlen;
1079 
1080 	kif->kf_type = KF_TYPE_SHM;
1081 	shmfd = fp->f_data;
1082 
1083 	mtx_lock(&shm_timestamp_lock);
1084 	kif->kf_un.kf_file.kf_file_mode = S_IFREG | shmfd->shm_mode;	/* XXX */
1085 	mtx_unlock(&shm_timestamp_lock);
1086 	kif->kf_un.kf_file.kf_file_size = shmfd->shm_size;
1087 	if (shmfd->shm_path != NULL) {
1088 		sx_slock(&shm_dict_lock);
1089 		if (shmfd->shm_path != NULL) {
1090 			path = shmfd->shm_path;
1091 			pr_path = curthread->td_ucred->cr_prison->pr_path;
1092 			if (strcmp(pr_path, "/") != 0) {
1093 				/* Return the jail-rooted pathname. */
1094 				pr_pathlen = strlen(pr_path);
1095 				if (strncmp(path, pr_path, pr_pathlen) == 0 &&
1096 				    path[pr_pathlen] == '/')
1097 					path += pr_pathlen;
1098 			}
1099 			strlcpy(kif->kf_path, path, sizeof(kif->kf_path));
1100 		}
1101 		sx_sunlock(&shm_dict_lock);
1102 	}
1103 	return (0);
1104 }
1105