FreeBSD/Linux Kernel Cross Reference
sys/kern/uipc_shm.c
1 /*-
2 * Copyright (c) 2006 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 * (2) 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 * (3) Add support for this file type to fstat(1).
39 *
40 * (4) Resource limits? Does this need its own resource limits or are the
41 * existing limits in mmap(2) sufficient?
42 *
43 * (5) Partial page truncation. vnode_pager_setsize() will zero any parts
44 * of a partially mapped page as a result of ftruncate(2)/truncate(2).
45 * We can do the same (with the same pmap evil), but do we need to
46 * worry about the bits on disk if the page is swapped out or will the
47 * swapper zero the parts of a page that are invalid if the page is
48 * swapped back in for us?
49 *
50 * (6) Add MAC support in mac_biba(4) and mac_mls(4).
51 *
52 * (7) Add a MAC check_create() hook for creating new named objects.
53 */
54
55 #include <sys/cdefs.h>
56 __FBSDID("$FreeBSD$");
57
58 #include <sys/param.h>
59 #include <sys/conf.h>
60 #include <sys/fcntl.h>
61 #include <sys/file.h>
62 #include <sys/filedesc.h>
63 #include <sys/fnv_hash.h>
64 #include <sys/kernel.h>
65 #include <sys/lock.h>
66 #include <sys/malloc.h>
67 #include <sys/mman.h>
68 #include <sys/mutex.h>
69 #include <sys/proc.h>
70 #include <sys/refcount.h>
71 #include <sys/resourcevar.h>
72 #include <sys/stat.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
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_pager.h>
91 #include <vm/swap_pager.h>
92
93 struct shm_mapping {
94 char *sm_path;
95 Fnv32_t sm_fnv;
96 struct shmfd *sm_shmfd;
97 LIST_ENTRY(shm_mapping) sm_link;
98 };
99
100 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
101 static LIST_HEAD(, shm_mapping) *shm_dictionary;
102 static struct sx shm_dict_lock;
103 static struct mtx shm_timestamp_lock;
104 static u_long shm_hash;
105 static struct unrhdr *shm_ino_unr;
106 static dev_t shm_dev_ino;
107
108 #define SHM_HASH(fnv) (&shm_dictionary[(fnv) & shm_hash])
109
110 static int shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags);
111 static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode);
112 static void shm_init(void *arg);
113 static void shm_drop(struct shmfd *shmfd);
114 static struct shmfd *shm_hold(struct shmfd *shmfd);
115 static void shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
116 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
117 static int shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
118 static int shm_dotruncate(struct shmfd *shmfd, off_t length);
119
120 static fo_rdwr_t shm_read;
121 static fo_rdwr_t shm_write;
122 static fo_truncate_t shm_truncate;
123 static fo_ioctl_t shm_ioctl;
124 static fo_poll_t shm_poll;
125 static fo_kqfilter_t shm_kqfilter;
126 static fo_stat_t shm_stat;
127 static fo_close_t shm_close;
128
129 /* File descriptor operations. */
130 static struct fileops shm_ops = {
131 .fo_read = shm_read,
132 .fo_write = shm_write,
133 .fo_truncate = shm_truncate,
134 .fo_ioctl = shm_ioctl,
135 .fo_poll = shm_poll,
136 .fo_kqfilter = shm_kqfilter,
137 .fo_stat = shm_stat,
138 .fo_close = shm_close,
139 .fo_flags = DFLAG_PASSABLE
140 };
141
142 FEATURE(posix_shm, "POSIX shared memory");
143
144 static int
145 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
146 int flags, struct thread *td)
147 {
148
149 return (EOPNOTSUPP);
150 }
151
152 static int
153 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
154 int flags, struct thread *td)
155 {
156
157 return (EOPNOTSUPP);
158 }
159
160 static int
161 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
162 struct thread *td)
163 {
164 struct shmfd *shmfd;
165 #ifdef MAC
166 int error;
167 #endif
168
169 shmfd = fp->f_data;
170 #ifdef MAC
171 error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
172 if (error)
173 return (error);
174 #endif
175 return (shm_dotruncate(shmfd, length));
176 }
177
178 static int
179 shm_ioctl(struct file *fp, u_long com, void *data,
180 struct ucred *active_cred, struct thread *td)
181 {
182
183 return (EOPNOTSUPP);
184 }
185
186 static int
187 shm_poll(struct file *fp, int events, struct ucred *active_cred,
188 struct thread *td)
189 {
190
191 return (EOPNOTSUPP);
192 }
193
194 static int
195 shm_kqfilter(struct file *fp, struct knote *kn)
196 {
197
198 return (EOPNOTSUPP);
199 }
200
201 static int
202 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
203 struct thread *td)
204 {
205 struct shmfd *shmfd;
206 #ifdef MAC
207 int error;
208 #endif
209
210 shmfd = fp->f_data;
211
212 #ifdef MAC
213 error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
214 if (error)
215 return (error);
216 #endif
217
218 /*
219 * Attempt to return sanish values for fstat() on a memory file
220 * descriptor.
221 */
222 bzero(sb, sizeof(*sb));
223 sb->st_mode = S_IFREG | shmfd->shm_mode; /* XXX */
224 sb->st_blksize = PAGE_SIZE;
225 sb->st_size = shmfd->shm_size;
226 sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize;
227 sb->st_atimespec = shmfd->shm_atime;
228 sb->st_ctimespec = shmfd->shm_ctime;
229 sb->st_mtimespec = shmfd->shm_mtime;
230 sb->st_birthtimespec = shmfd->shm_birthtime;
231 sb->st_uid = shmfd->shm_uid;
232 sb->st_gid = shmfd->shm_gid;
233 sb->st_dev = shm_dev_ino;
234 sb->st_ino = shmfd->shm_ino;
235
236 return (0);
237 }
238
239 static int
240 shm_close(struct file *fp, struct thread *td)
241 {
242 struct shmfd *shmfd;
243
244 shmfd = fp->f_data;
245 fp->f_data = NULL;
246 shm_drop(shmfd);
247
248 return (0);
249 }
250
251 static int
252 shm_dotruncate(struct shmfd *shmfd, off_t length)
253 {
254 vm_object_t object;
255 vm_page_t m;
256 vm_pindex_t nobjsize;
257 vm_ooffset_t delta;
258
259 object = shmfd->shm_object;
260 VM_OBJECT_LOCK(object);
261 if (length == shmfd->shm_size) {
262 VM_OBJECT_UNLOCK(object);
263 return (0);
264 }
265 nobjsize = OFF_TO_IDX(length + PAGE_MASK);
266
267 /* Are we shrinking? If so, trim the end. */
268 if (length < shmfd->shm_size) {
269 /*
270 * Disallow any requests to shrink the size if this
271 * object is mapped into the kernel.
272 */
273 if (shmfd->shm_kmappings > 0) {
274 VM_OBJECT_UNLOCK(object);
275 return (EBUSY);
276 }
277 delta = ptoa(object->size - nobjsize);
278
279 /* Toss in memory pages. */
280 if (nobjsize < object->size)
281 vm_object_page_remove(object, nobjsize, object->size,
282 FALSE);
283
284 /* Toss pages from swap. */
285 if (object->type == OBJT_SWAP)
286 swap_pager_freespace(object, nobjsize, delta);
287
288 /* Free the swap accounted for shm */
289 swap_release_by_uid(delta, object->uip);
290 object->charge -= delta;
291
292 /*
293 * If the last page is partially mapped, then zero out
294 * the garbage at the end of the page. See comments
295 * in vnode_pager_setsize() for more details.
296 *
297 * XXXJHB: This handles in memory pages, but what about
298 * a page swapped out to disk?
299 */
300 if ((length & PAGE_MASK) &&
301 (m = vm_page_lookup(object, OFF_TO_IDX(length))) != NULL &&
302 m->valid != 0) {
303 int base = (int)length & PAGE_MASK;
304 int size = PAGE_SIZE - base;
305
306 pmap_zero_page_area(m, base, size);
307
308 /*
309 * Update the valid bits to reflect the blocks that
310 * have been zeroed. Some of these valid bits may
311 * have already been set.
312 */
313 vm_page_set_valid(m, base, size);
314
315 /*
316 * Round "base" to the next block boundary so that the
317 * dirty bit for a partially zeroed block is not
318 * cleared.
319 */
320 base = roundup2(base, DEV_BSIZE);
321
322 vm_page_lock_queues();
323 vm_page_clear_dirty(m, base, PAGE_SIZE - base);
324 vm_page_unlock_queues();
325 } else if ((length & PAGE_MASK) &&
326 __predict_false(object->cache != NULL)) {
327 vm_page_cache_free(object, OFF_TO_IDX(length),
328 nobjsize);
329 }
330 } else {
331
332 /* Attempt to reserve the swap */
333 delta = ptoa(nobjsize - object->size);
334 if (!swap_reserve_by_uid(delta, object->uip)) {
335 VM_OBJECT_UNLOCK(object);
336 return (ENOMEM);
337 }
338 object->charge += delta;
339 }
340 shmfd->shm_size = length;
341 mtx_lock(&shm_timestamp_lock);
342 vfs_timestamp(&shmfd->shm_ctime);
343 shmfd->shm_mtime = shmfd->shm_ctime;
344 mtx_unlock(&shm_timestamp_lock);
345 object->size = nobjsize;
346 VM_OBJECT_UNLOCK(object);
347 return (0);
348 }
349
350 /*
351 * shmfd object management including creation and reference counting
352 * routines.
353 */
354 static struct shmfd *
355 shm_alloc(struct ucred *ucred, mode_t mode)
356 {
357 struct shmfd *shmfd;
358 int ino;
359
360 shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
361 shmfd->shm_size = 0;
362 shmfd->shm_uid = ucred->cr_uid;
363 shmfd->shm_gid = ucred->cr_gid;
364 shmfd->shm_mode = mode;
365 shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
366 shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
367 KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
368 VM_OBJECT_LOCK(shmfd->shm_object);
369 vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
370 vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
371 VM_OBJECT_UNLOCK(shmfd->shm_object);
372 vfs_timestamp(&shmfd->shm_birthtime);
373 shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
374 shmfd->shm_birthtime;
375 ino = alloc_unr(shm_ino_unr);
376 if (ino == -1)
377 shmfd->shm_ino = 0;
378 else
379 shmfd->shm_ino = ino;
380 refcount_init(&shmfd->shm_refs, 1);
381 #ifdef MAC
382 mac_posixshm_init(shmfd);
383 mac_posixshm_create(ucred, shmfd);
384 #endif
385
386 return (shmfd);
387 }
388
389 static struct shmfd *
390 shm_hold(struct shmfd *shmfd)
391 {
392
393 refcount_acquire(&shmfd->shm_refs);
394 return (shmfd);
395 }
396
397 static void
398 shm_drop(struct shmfd *shmfd)
399 {
400
401 if (refcount_release(&shmfd->shm_refs)) {
402 #ifdef MAC
403 mac_posixshm_destroy(shmfd);
404 #endif
405 vm_object_deallocate(shmfd->shm_object);
406 if (shmfd->shm_ino != 0)
407 free_unr(shm_ino_unr, shmfd->shm_ino);
408 free(shmfd, M_SHMFD);
409 }
410 }
411
412 /*
413 * Determine if the credentials have sufficient permissions for a
414 * specified combination of FREAD and FWRITE.
415 */
416 static int
417 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
418 {
419 accmode_t accmode;
420
421 accmode = 0;
422 if (flags & FREAD)
423 accmode |= VREAD;
424 if (flags & FWRITE)
425 accmode |= VWRITE;
426 return (vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
427 accmode, ucred, NULL));
428 }
429
430 /*
431 * Dictionary management. We maintain an in-kernel dictionary to map
432 * paths to shmfd objects. We use the FNV hash on the path to store
433 * the mappings in a hash table.
434 */
435 static void
436 shm_init(void *arg)
437 {
438
439 mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
440 sx_init(&shm_dict_lock, "shm dictionary");
441 shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
442 shm_ino_unr = new_unrhdr(1, INT32_MAX, NULL);
443 KASSERT(shm_ino_unr != NULL, ("shm fake inodes not initialized"));
444 shm_dev_ino = devfs_alloc_cdp_inode();
445 KASSERT(shm_dev_ino > 0, ("shm dev inode not initialized"));
446 }
447 SYSINIT(shm_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_init, NULL);
448
449 static struct shmfd *
450 shm_lookup(char *path, Fnv32_t fnv)
451 {
452 struct shm_mapping *map;
453
454 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
455 if (map->sm_fnv != fnv)
456 continue;
457 if (strcmp(map->sm_path, path) == 0)
458 return (map->sm_shmfd);
459 }
460
461 return (NULL);
462 }
463
464 static void
465 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
466 {
467 struct shm_mapping *map;
468
469 map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
470 map->sm_path = path;
471 map->sm_fnv = fnv;
472 map->sm_shmfd = shm_hold(shmfd);
473 shmfd->shm_path = path;
474 LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
475 }
476
477 static int
478 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
479 {
480 struct shm_mapping *map;
481 int error;
482
483 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
484 if (map->sm_fnv != fnv)
485 continue;
486 if (strcmp(map->sm_path, path) == 0) {
487 #ifdef MAC
488 error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
489 if (error)
490 return (error);
491 #endif
492 error = shm_access(map->sm_shmfd, ucred,
493 FREAD | FWRITE);
494 if (error)
495 return (error);
496 map->sm_shmfd->shm_path = NULL;
497 LIST_REMOVE(map, sm_link);
498 shm_drop(map->sm_shmfd);
499 free(map->sm_path, M_SHMFD);
500 free(map, M_SHMFD);
501 return (0);
502 }
503 }
504
505 return (ENOENT);
506 }
507
508 /* System calls. */
509 int
510 shm_open(struct thread *td, struct shm_open_args *uap)
511 {
512 struct filedesc *fdp;
513 struct shmfd *shmfd;
514 struct file *fp;
515 char *path;
516 Fnv32_t fnv;
517 mode_t cmode;
518 int fd, error;
519
520 if ((uap->flags & O_ACCMODE) != O_RDONLY &&
521 (uap->flags & O_ACCMODE) != O_RDWR)
522 return (EINVAL);
523
524 if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0)
525 return (EINVAL);
526
527 fdp = td->td_proc->p_fd;
528 cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
529
530 error = falloc(td, &fp, &fd);
531 if (error)
532 return (error);
533
534 /* A SHM_ANON path pointer creates an anonymous object. */
535 if (uap->path == SHM_ANON) {
536 /* A read-only anonymous object is pointless. */
537 if ((uap->flags & O_ACCMODE) == O_RDONLY) {
538 fdclose(fdp, fp, fd, td);
539 fdrop(fp, td);
540 return (EINVAL);
541 }
542 shmfd = shm_alloc(td->td_ucred, cmode);
543 } else {
544 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
545 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
546
547 /* Require paths to start with a '/' character. */
548 if (error == 0 && path[0] != '/')
549 error = EINVAL;
550 if (error) {
551 fdclose(fdp, fp, fd, td);
552 fdrop(fp, td);
553 free(path, M_SHMFD);
554 return (error);
555 }
556
557 fnv = fnv_32_str(path, FNV1_32_INIT);
558 sx_xlock(&shm_dict_lock);
559 shmfd = shm_lookup(path, fnv);
560 if (shmfd == NULL) {
561 /* Object does not yet exist, create it if requested. */
562 if (uap->flags & O_CREAT) {
563 shmfd = shm_alloc(td->td_ucred, cmode);
564 shm_insert(path, fnv, shmfd);
565 } else {
566 free(path, M_SHMFD);
567 error = ENOENT;
568 }
569 } else {
570 /*
571 * Object already exists, obtain a new
572 * reference if requested and permitted.
573 */
574 free(path, M_SHMFD);
575 if ((uap->flags & (O_CREAT | O_EXCL)) ==
576 (O_CREAT | O_EXCL))
577 error = EEXIST;
578 else {
579 #ifdef MAC
580 error = mac_posixshm_check_open(td->td_ucred,
581 shmfd);
582 if (error == 0)
583 #endif
584 error = shm_access(shmfd, td->td_ucred,
585 FFLAGS(uap->flags & O_ACCMODE));
586 }
587
588 /*
589 * Truncate the file back to zero length if
590 * O_TRUNC was specified and the object was
591 * opened with read/write.
592 */
593 if (error == 0 &&
594 (uap->flags & (O_ACCMODE | O_TRUNC)) ==
595 (O_RDWR | O_TRUNC)) {
596 #ifdef MAC
597 error = mac_posixshm_check_truncate(
598 td->td_ucred, fp->f_cred, shmfd);
599 if (error == 0)
600 #endif
601 shm_dotruncate(shmfd, 0);
602 }
603 if (error == 0)
604 shm_hold(shmfd);
605 }
606 sx_xunlock(&shm_dict_lock);
607
608 if (error) {
609 fdclose(fdp, fp, fd, td);
610 fdrop(fp, td);
611 return (error);
612 }
613 }
614
615 finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
616
617 FILEDESC_XLOCK(fdp);
618 if (fdp->fd_ofiles[fd] == fp)
619 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
620 FILEDESC_XUNLOCK(fdp);
621 td->td_retval[0] = fd;
622 fdrop(fp, td);
623
624 return (0);
625 }
626
627 int
628 shm_unlink(struct thread *td, struct shm_unlink_args *uap)
629 {
630 char *path;
631 Fnv32_t fnv;
632 int error;
633
634 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
635 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
636 if (error) {
637 free(path, M_TEMP);
638 return (error);
639 }
640
641 fnv = fnv_32_str(path, FNV1_32_INIT);
642 sx_xlock(&shm_dict_lock);
643 error = shm_remove(path, fnv, td->td_ucred);
644 sx_xunlock(&shm_dict_lock);
645 free(path, M_TEMP);
646
647 return (error);
648 }
649
650 /*
651 * mmap() helper to validate mmap() requests against shm object state
652 * and give mmap() the vm_object to use for the mapping.
653 */
654 int
655 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
656 vm_object_t *obj)
657 {
658
659 /*
660 * XXXRW: This validation is probably insufficient, and subject to
661 * sign errors. It should be fixed.
662 */
663 if (foff >= shmfd->shm_size ||
664 foff + objsize > round_page(shmfd->shm_size))
665 return (EINVAL);
666
667 mtx_lock(&shm_timestamp_lock);
668 vfs_timestamp(&shmfd->shm_atime);
669 mtx_unlock(&shm_timestamp_lock);
670 vm_object_reference(shmfd->shm_object);
671 *obj = shmfd->shm_object;
672 return (0);
673 }
674
675 /*
676 * Helper routines to allow the backing object of a shared memory file
677 * descriptor to be mapped in the kernel.
678 */
679 int
680 shm_map(struct file *fp, size_t size, off_t offset, void **memp)
681 {
682 struct shmfd *shmfd;
683 vm_offset_t kva, ofs;
684 vm_object_t obj;
685 int rv;
686
687 if (fp->f_type != DTYPE_SHM)
688 return (EINVAL);
689 shmfd = fp->f_data;
690 obj = shmfd->shm_object;
691 VM_OBJECT_LOCK(obj);
692 /*
693 * XXXRW: This validation is probably insufficient, and subject to
694 * sign errors. It should be fixed.
695 */
696 if (offset >= shmfd->shm_size ||
697 offset + size > round_page(shmfd->shm_size)) {
698 VM_OBJECT_UNLOCK(obj);
699 return (EINVAL);
700 }
701
702 shmfd->shm_kmappings++;
703 vm_object_reference_locked(obj);
704 VM_OBJECT_UNLOCK(obj);
705
706 /* Map the object into the kernel_map and wire it. */
707 kva = vm_map_min(kernel_map);
708 ofs = offset & PAGE_MASK;
709 offset = trunc_page(offset);
710 size = round_page(size + ofs);
711 rv = vm_map_find(kernel_map, obj, offset, &kva, size,
712 VMFS_ALIGNED_SPACE, VM_PROT_READ | VM_PROT_WRITE,
713 VM_PROT_READ | VM_PROT_WRITE, 0);
714 if (rv == KERN_SUCCESS) {
715 rv = vm_map_wire(kernel_map, kva, kva + size,
716 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
717 if (rv == KERN_SUCCESS) {
718 *memp = (void *)(kva + ofs);
719 return (0);
720 }
721 vm_map_remove(kernel_map, kva, kva + size);
722 } else
723 vm_object_deallocate(obj);
724
725 /* On failure, drop our mapping reference. */
726 VM_OBJECT_LOCK(obj);
727 shmfd->shm_kmappings--;
728 VM_OBJECT_UNLOCK(obj);
729
730 return (vm_mmap_to_errno(rv));
731 }
732
733 /*
734 * We require the caller to unmap the entire entry. This allows us to
735 * safely decrement shm_kmappings when a mapping is removed.
736 */
737 int
738 shm_unmap(struct file *fp, void *mem, size_t size)
739 {
740 struct shmfd *shmfd;
741 vm_map_entry_t entry;
742 vm_offset_t kva, ofs;
743 vm_object_t obj;
744 vm_pindex_t pindex;
745 vm_prot_t prot;
746 boolean_t wired;
747 vm_map_t map;
748 int rv;
749
750 if (fp->f_type != DTYPE_SHM)
751 return (EINVAL);
752 shmfd = fp->f_data;
753 kva = (vm_offset_t)mem;
754 ofs = kva & PAGE_MASK;
755 kva = trunc_page(kva);
756 size = round_page(size + ofs);
757 map = kernel_map;
758 rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
759 &obj, &pindex, &prot, &wired);
760 if (rv != KERN_SUCCESS)
761 return (EINVAL);
762 if (entry->start != kva || entry->end != kva + size) {
763 vm_map_lookup_done(map, entry);
764 return (EINVAL);
765 }
766 vm_map_lookup_done(map, entry);
767 if (obj != shmfd->shm_object)
768 return (EINVAL);
769 vm_map_remove(map, kva, kva + size);
770 VM_OBJECT_LOCK(obj);
771 KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
772 shmfd->shm_kmappings--;
773 VM_OBJECT_UNLOCK(obj);
774 return (0);
775 }
776
777 void
778 shm_path(struct shmfd *shmfd, char *path, size_t size)
779 {
780
781 if (shmfd->shm_path == NULL)
782 return;
783 sx_slock(&shm_dict_lock);
784 if (shmfd->shm_path != NULL)
785 strlcpy(path, shmfd->shm_path, size);
786 sx_sunlock(&shm_dict_lock);
787 }
Cache object: a699cd1d03c13da53b4e4940189a7f99
|