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: releng/8.4/sys/kern/uipc_shm.c 236699 2012-06-06 21:57:03Z jhb $");
57
58 #include <sys/param.h>
59 #include <sys/fcntl.h>
60 #include <sys/file.h>
61 #include <sys/filedesc.h>
62 #include <sys/fnv_hash.h>
63 #include <sys/kernel.h>
64 #include <sys/lock.h>
65 #include <sys/malloc.h>
66 #include <sys/mman.h>
67 #include <sys/mutex.h>
68 #include <sys/proc.h>
69 #include <sys/refcount.h>
70 #include <sys/resourcevar.h>
71 #include <sys/stat.h>
72 #include <sys/sysctl.h>
73 #include <sys/sysproto.h>
74 #include <sys/systm.h>
75 #include <sys/sx.h>
76 #include <sys/time.h>
77 #include <sys/vnode.h>
78
79 #include <security/mac/mac_framework.h>
80
81 #include <vm/vm.h>
82 #include <vm/vm_param.h>
83 #include <vm/pmap.h>
84 #include <vm/vm_extern.h>
85 #include <vm/vm_map.h>
86 #include <vm/vm_kern.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_pager.h>
90 #include <vm/swap_pager.h>
91
92 struct shm_mapping {
93 char *sm_path;
94 Fnv32_t sm_fnv;
95 struct shmfd *sm_shmfd;
96 LIST_ENTRY(shm_mapping) sm_link;
97 };
98
99 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
100 static LIST_HEAD(, shm_mapping) *shm_dictionary;
101 static struct sx shm_dict_lock;
102 static struct mtx shm_timestamp_lock;
103 static u_long shm_hash;
104
105 #define SHM_HASH(fnv) (&shm_dictionary[(fnv) & shm_hash])
106
107 static int shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags);
108 static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode);
109 static void shm_dict_init(void *arg);
110 static void shm_drop(struct shmfd *shmfd);
111 static struct shmfd *shm_hold(struct shmfd *shmfd);
112 static void shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
113 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
114 static int shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
115 static int shm_dotruncate(struct shmfd *shmfd, off_t length);
116
117 static fo_rdwr_t shm_read;
118 static fo_rdwr_t shm_write;
119 static fo_truncate_t shm_truncate;
120 static fo_ioctl_t shm_ioctl;
121 static fo_poll_t shm_poll;
122 static fo_kqfilter_t shm_kqfilter;
123 static fo_stat_t shm_stat;
124 static fo_close_t shm_close;
125
126 /* File descriptor operations. */
127 static struct fileops shm_ops = {
128 .fo_read = shm_read,
129 .fo_write = shm_write,
130 .fo_truncate = shm_truncate,
131 .fo_ioctl = shm_ioctl,
132 .fo_poll = shm_poll,
133 .fo_kqfilter = shm_kqfilter,
134 .fo_stat = shm_stat,
135 .fo_close = shm_close,
136 .fo_flags = DFLAG_PASSABLE
137 };
138
139 FEATURE(posix_shm, "POSIX shared memory");
140
141 static int
142 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
143 int flags, struct thread *td)
144 {
145
146 return (EOPNOTSUPP);
147 }
148
149 static int
150 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
151 int flags, struct thread *td)
152 {
153
154 return (EOPNOTSUPP);
155 }
156
157 static int
158 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
159 struct thread *td)
160 {
161 struct shmfd *shmfd;
162 #ifdef MAC
163 int error;
164 #endif
165
166 shmfd = fp->f_data;
167 #ifdef MAC
168 error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
169 if (error)
170 return (error);
171 #endif
172 return (shm_dotruncate(shmfd, length));
173 }
174
175 static int
176 shm_ioctl(struct file *fp, u_long com, void *data,
177 struct ucred *active_cred, struct thread *td)
178 {
179
180 return (EOPNOTSUPP);
181 }
182
183 static int
184 shm_poll(struct file *fp, int events, struct ucred *active_cred,
185 struct thread *td)
186 {
187
188 return (EOPNOTSUPP);
189 }
190
191 static int
192 shm_kqfilter(struct file *fp, struct knote *kn)
193 {
194
195 return (EOPNOTSUPP);
196 }
197
198 static int
199 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
200 struct thread *td)
201 {
202 struct shmfd *shmfd;
203 #ifdef MAC
204 int error;
205 #endif
206
207 shmfd = fp->f_data;
208
209 #ifdef MAC
210 error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
211 if (error)
212 return (error);
213 #endif
214
215 /*
216 * Attempt to return sanish values for fstat() on a memory file
217 * descriptor.
218 */
219 bzero(sb, sizeof(*sb));
220 sb->st_mode = S_IFREG | shmfd->shm_mode; /* XXX */
221 sb->st_blksize = PAGE_SIZE;
222 sb->st_size = shmfd->shm_size;
223 sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize;
224 sb->st_atimespec = shmfd->shm_atime;
225 sb->st_ctimespec = shmfd->shm_ctime;
226 sb->st_mtimespec = shmfd->shm_mtime;
227 sb->st_birthtimespec = shmfd->shm_birthtime;
228 sb->st_uid = shmfd->shm_uid;
229 sb->st_gid = shmfd->shm_gid;
230
231 return (0);
232 }
233
234 static int
235 shm_close(struct file *fp, struct thread *td)
236 {
237 struct shmfd *shmfd;
238
239 shmfd = fp->f_data;
240 fp->f_data = NULL;
241 shm_drop(shmfd);
242
243 return (0);
244 }
245
246 static int
247 shm_dotruncate(struct shmfd *shmfd, off_t length)
248 {
249 vm_object_t object;
250 vm_page_t m;
251 vm_pindex_t nobjsize;
252 vm_ooffset_t delta;
253
254 object = shmfd->shm_object;
255 VM_OBJECT_LOCK(object);
256 if (length == shmfd->shm_size) {
257 VM_OBJECT_UNLOCK(object);
258 return (0);
259 }
260 nobjsize = OFF_TO_IDX(length + PAGE_MASK);
261
262 /* Are we shrinking? If so, trim the end. */
263 if (length < shmfd->shm_size) {
264 /*
265 * Disallow any requests to shrink the size if this
266 * object is mapped into the kernel.
267 */
268 if (shmfd->shm_kmappings > 0) {
269 VM_OBJECT_UNLOCK(object);
270 return (EBUSY);
271 }
272 delta = ptoa(object->size - nobjsize);
273
274 /* Toss in memory pages. */
275 if (nobjsize < object->size)
276 vm_object_page_remove(object, nobjsize, object->size,
277 FALSE);
278
279 /* Toss pages from swap. */
280 if (object->type == OBJT_SWAP)
281 swap_pager_freespace(object, nobjsize, delta);
282
283 /* Free the swap accounted for shm */
284 swap_release_by_uid(delta, object->uip);
285 object->charge -= delta;
286
287 /*
288 * If the last page is partially mapped, then zero out
289 * the garbage at the end of the page. See comments
290 * in vnode_pager_setsize() for more details.
291 *
292 * XXXJHB: This handles in memory pages, but what about
293 * a page swapped out to disk?
294 */
295 if ((length & PAGE_MASK) &&
296 (m = vm_page_lookup(object, OFF_TO_IDX(length))) != NULL &&
297 m->valid != 0) {
298 int base = (int)length & PAGE_MASK;
299 int size = PAGE_SIZE - base;
300
301 pmap_zero_page_area(m, base, size);
302
303 /*
304 * Update the valid bits to reflect the blocks that
305 * have been zeroed. Some of these valid bits may
306 * have already been set.
307 */
308 vm_page_set_valid(m, base, size);
309
310 /*
311 * Round "base" to the next block boundary so that the
312 * dirty bit for a partially zeroed block is not
313 * cleared.
314 */
315 base = roundup2(base, DEV_BSIZE);
316
317 vm_page_lock_queues();
318 vm_page_clear_dirty(m, base, PAGE_SIZE - base);
319 vm_page_unlock_queues();
320 } else if ((length & PAGE_MASK) &&
321 __predict_false(object->cache != NULL)) {
322 vm_page_cache_free(object, OFF_TO_IDX(length),
323 nobjsize);
324 }
325 } else {
326
327 /* Attempt to reserve the swap */
328 delta = ptoa(nobjsize - object->size);
329 if (!swap_reserve_by_uid(delta, object->uip)) {
330 VM_OBJECT_UNLOCK(object);
331 return (ENOMEM);
332 }
333 object->charge += delta;
334 }
335 shmfd->shm_size = length;
336 mtx_lock(&shm_timestamp_lock);
337 vfs_timestamp(&shmfd->shm_ctime);
338 shmfd->shm_mtime = shmfd->shm_ctime;
339 mtx_unlock(&shm_timestamp_lock);
340 object->size = nobjsize;
341 VM_OBJECT_UNLOCK(object);
342 return (0);
343 }
344
345 /*
346 * shmfd object management including creation and reference counting
347 * routines.
348 */
349 static struct shmfd *
350 shm_alloc(struct ucred *ucred, mode_t mode)
351 {
352 struct shmfd *shmfd;
353
354 shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
355 shmfd->shm_size = 0;
356 shmfd->shm_uid = ucred->cr_uid;
357 shmfd->shm_gid = ucred->cr_gid;
358 shmfd->shm_mode = mode;
359 shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
360 shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
361 KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
362 VM_OBJECT_LOCK(shmfd->shm_object);
363 vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
364 vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
365 VM_OBJECT_UNLOCK(shmfd->shm_object);
366 vfs_timestamp(&shmfd->shm_birthtime);
367 shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
368 shmfd->shm_birthtime;
369 refcount_init(&shmfd->shm_refs, 1);
370 #ifdef MAC
371 mac_posixshm_init(shmfd);
372 mac_posixshm_create(ucred, shmfd);
373 #endif
374
375 return (shmfd);
376 }
377
378 static struct shmfd *
379 shm_hold(struct shmfd *shmfd)
380 {
381
382 refcount_acquire(&shmfd->shm_refs);
383 return (shmfd);
384 }
385
386 static void
387 shm_drop(struct shmfd *shmfd)
388 {
389
390 if (refcount_release(&shmfd->shm_refs)) {
391 #ifdef MAC
392 mac_posixshm_destroy(shmfd);
393 #endif
394 vm_object_deallocate(shmfd->shm_object);
395 free(shmfd, M_SHMFD);
396 }
397 }
398
399 /*
400 * Determine if the credentials have sufficient permissions for a
401 * specified combination of FREAD and FWRITE.
402 */
403 static int
404 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
405 {
406 accmode_t accmode;
407
408 accmode = 0;
409 if (flags & FREAD)
410 accmode |= VREAD;
411 if (flags & FWRITE)
412 accmode |= VWRITE;
413 return (vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
414 accmode, ucred, NULL));
415 }
416
417 /*
418 * Dictionary management. We maintain an in-kernel dictionary to map
419 * paths to shmfd objects. We use the FNV hash on the path to store
420 * the mappings in a hash table.
421 */
422 static void
423 shm_dict_init(void *arg)
424 {
425
426 mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
427 sx_init(&shm_dict_lock, "shm dictionary");
428 shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
429 }
430 SYSINIT(shm_dict_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_dict_init, NULL);
431
432 static struct shmfd *
433 shm_lookup(char *path, Fnv32_t fnv)
434 {
435 struct shm_mapping *map;
436
437 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
438 if (map->sm_fnv != fnv)
439 continue;
440 if (strcmp(map->sm_path, path) == 0)
441 return (map->sm_shmfd);
442 }
443
444 return (NULL);
445 }
446
447 static void
448 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
449 {
450 struct shm_mapping *map;
451
452 map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
453 map->sm_path = path;
454 map->sm_fnv = fnv;
455 map->sm_shmfd = shm_hold(shmfd);
456 shmfd->shm_path = path;
457 LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
458 }
459
460 static int
461 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
462 {
463 struct shm_mapping *map;
464 int error;
465
466 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
467 if (map->sm_fnv != fnv)
468 continue;
469 if (strcmp(map->sm_path, path) == 0) {
470 #ifdef MAC
471 error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
472 if (error)
473 return (error);
474 #endif
475 error = shm_access(map->sm_shmfd, ucred,
476 FREAD | FWRITE);
477 if (error)
478 return (error);
479 map->sm_shmfd->shm_path = NULL;
480 LIST_REMOVE(map, sm_link);
481 shm_drop(map->sm_shmfd);
482 free(map->sm_path, M_SHMFD);
483 free(map, M_SHMFD);
484 return (0);
485 }
486 }
487
488 return (ENOENT);
489 }
490
491 /* System calls. */
492 int
493 shm_open(struct thread *td, struct shm_open_args *uap)
494 {
495 struct filedesc *fdp;
496 struct shmfd *shmfd;
497 struct file *fp;
498 char *path;
499 Fnv32_t fnv;
500 mode_t cmode;
501 int fd, error;
502
503 if ((uap->flags & O_ACCMODE) != O_RDONLY &&
504 (uap->flags & O_ACCMODE) != O_RDWR)
505 return (EINVAL);
506
507 if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0)
508 return (EINVAL);
509
510 fdp = td->td_proc->p_fd;
511 cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
512
513 error = falloc(td, &fp, &fd);
514 if (error)
515 return (error);
516
517 /* A SHM_ANON path pointer creates an anonymous object. */
518 if (uap->path == SHM_ANON) {
519 /* A read-only anonymous object is pointless. */
520 if ((uap->flags & O_ACCMODE) == O_RDONLY) {
521 fdclose(fdp, fp, fd, td);
522 fdrop(fp, td);
523 return (EINVAL);
524 }
525 shmfd = shm_alloc(td->td_ucred, cmode);
526 } else {
527 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
528 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
529
530 /* Require paths to start with a '/' character. */
531 if (error == 0 && path[0] != '/')
532 error = EINVAL;
533 if (error) {
534 fdclose(fdp, fp, fd, td);
535 fdrop(fp, td);
536 free(path, M_SHMFD);
537 return (error);
538 }
539
540 fnv = fnv_32_str(path, FNV1_32_INIT);
541 sx_xlock(&shm_dict_lock);
542 shmfd = shm_lookup(path, fnv);
543 if (shmfd == NULL) {
544 /* Object does not yet exist, create it if requested. */
545 if (uap->flags & O_CREAT) {
546 shmfd = shm_alloc(td->td_ucred, cmode);
547 shm_insert(path, fnv, shmfd);
548 } else {
549 free(path, M_SHMFD);
550 error = ENOENT;
551 }
552 } else {
553 /*
554 * Object already exists, obtain a new
555 * reference if requested and permitted.
556 */
557 free(path, M_SHMFD);
558 if ((uap->flags & (O_CREAT | O_EXCL)) ==
559 (O_CREAT | O_EXCL))
560 error = EEXIST;
561 else {
562 #ifdef MAC
563 error = mac_posixshm_check_open(td->td_ucred,
564 shmfd);
565 if (error == 0)
566 #endif
567 error = shm_access(shmfd, td->td_ucred,
568 FFLAGS(uap->flags & O_ACCMODE));
569 }
570
571 /*
572 * Truncate the file back to zero length if
573 * O_TRUNC was specified and the object was
574 * opened with read/write.
575 */
576 if (error == 0 &&
577 (uap->flags & (O_ACCMODE | O_TRUNC)) ==
578 (O_RDWR | O_TRUNC)) {
579 #ifdef MAC
580 error = mac_posixshm_check_truncate(
581 td->td_ucred, fp->f_cred, shmfd);
582 if (error == 0)
583 #endif
584 shm_dotruncate(shmfd, 0);
585 }
586 if (error == 0)
587 shm_hold(shmfd);
588 }
589 sx_xunlock(&shm_dict_lock);
590
591 if (error) {
592 fdclose(fdp, fp, fd, td);
593 fdrop(fp, td);
594 return (error);
595 }
596 }
597
598 finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
599
600 FILEDESC_XLOCK(fdp);
601 if (fdp->fd_ofiles[fd] == fp)
602 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
603 FILEDESC_XUNLOCK(fdp);
604 td->td_retval[0] = fd;
605 fdrop(fp, td);
606
607 return (0);
608 }
609
610 int
611 shm_unlink(struct thread *td, struct shm_unlink_args *uap)
612 {
613 char *path;
614 Fnv32_t fnv;
615 int error;
616
617 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
618 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
619 if (error) {
620 free(path, M_TEMP);
621 return (error);
622 }
623
624 fnv = fnv_32_str(path, FNV1_32_INIT);
625 sx_xlock(&shm_dict_lock);
626 error = shm_remove(path, fnv, td->td_ucred);
627 sx_xunlock(&shm_dict_lock);
628 free(path, M_TEMP);
629
630 return (error);
631 }
632
633 /*
634 * mmap() helper to validate mmap() requests against shm object state
635 * and give mmap() the vm_object to use for the mapping.
636 */
637 int
638 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
639 vm_object_t *obj)
640 {
641
642 /*
643 * XXXRW: This validation is probably insufficient, and subject to
644 * sign errors. It should be fixed.
645 */
646 if (foff >= shmfd->shm_size ||
647 foff + objsize > round_page(shmfd->shm_size))
648 return (EINVAL);
649
650 mtx_lock(&shm_timestamp_lock);
651 vfs_timestamp(&shmfd->shm_atime);
652 mtx_unlock(&shm_timestamp_lock);
653 vm_object_reference(shmfd->shm_object);
654 *obj = shmfd->shm_object;
655 return (0);
656 }
657
658 /*
659 * Helper routines to allow the backing object of a shared memory file
660 * descriptor to be mapped in the kernel.
661 */
662 int
663 shm_map(struct file *fp, size_t size, off_t offset, void **memp)
664 {
665 struct shmfd *shmfd;
666 vm_offset_t kva, ofs;
667 vm_object_t obj;
668 int rv;
669
670 if (fp->f_type != DTYPE_SHM)
671 return (EINVAL);
672 shmfd = fp->f_data;
673 obj = shmfd->shm_object;
674 VM_OBJECT_LOCK(obj);
675 /*
676 * XXXRW: This validation is probably insufficient, and subject to
677 * sign errors. It should be fixed.
678 */
679 if (offset >= shmfd->shm_size ||
680 offset + size > round_page(shmfd->shm_size)) {
681 VM_OBJECT_UNLOCK(obj);
682 return (EINVAL);
683 }
684
685 shmfd->shm_kmappings++;
686 vm_object_reference_locked(obj);
687 VM_OBJECT_UNLOCK(obj);
688
689 /* Map the object into the kernel_map and wire it. */
690 kva = vm_map_min(kernel_map);
691 ofs = offset & PAGE_MASK;
692 offset = trunc_page(offset);
693 size = round_page(size + ofs);
694 rv = vm_map_find(kernel_map, obj, offset, &kva, size,
695 VMFS_ALIGNED_SPACE, VM_PROT_READ | VM_PROT_WRITE,
696 VM_PROT_READ | VM_PROT_WRITE, 0);
697 if (rv == KERN_SUCCESS) {
698 rv = vm_map_wire(kernel_map, kva, kva + size,
699 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
700 if (rv == KERN_SUCCESS) {
701 *memp = (void *)(kva + ofs);
702 return (0);
703 }
704 vm_map_remove(kernel_map, kva, kva + size);
705 } else
706 vm_object_deallocate(obj);
707
708 /* On failure, drop our mapping reference. */
709 VM_OBJECT_LOCK(obj);
710 shmfd->shm_kmappings--;
711 VM_OBJECT_UNLOCK(obj);
712
713 return (vm_mmap_to_errno(rv));
714 }
715
716 /*
717 * We require the caller to unmap the entire entry. This allows us to
718 * safely decrement shm_kmappings when a mapping is removed.
719 */
720 int
721 shm_unmap(struct file *fp, void *mem, size_t size)
722 {
723 struct shmfd *shmfd;
724 vm_map_entry_t entry;
725 vm_offset_t kva, ofs;
726 vm_object_t obj;
727 vm_pindex_t pindex;
728 vm_prot_t prot;
729 boolean_t wired;
730 vm_map_t map;
731 int rv;
732
733 if (fp->f_type != DTYPE_SHM)
734 return (EINVAL);
735 shmfd = fp->f_data;
736 kva = (vm_offset_t)mem;
737 ofs = kva & PAGE_MASK;
738 kva = trunc_page(kva);
739 size = round_page(size + ofs);
740 map = kernel_map;
741 rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
742 &obj, &pindex, &prot, &wired);
743 if (rv != KERN_SUCCESS)
744 return (EINVAL);
745 if (entry->start != kva || entry->end != kva + size) {
746 vm_map_lookup_done(map, entry);
747 return (EINVAL);
748 }
749 vm_map_lookup_done(map, entry);
750 if (obj != shmfd->shm_object)
751 return (EINVAL);
752 vm_map_remove(map, kva, kva + size);
753 VM_OBJECT_LOCK(obj);
754 KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
755 shmfd->shm_kmappings--;
756 VM_OBJECT_UNLOCK(obj);
757 return (0);
758 }
759
760 void
761 shm_path(struct shmfd *shmfd, char *path, size_t size)
762 {
763
764 if (shmfd->shm_path == NULL)
765 return;
766 sx_slock(&shm_dict_lock);
767 if (shmfd->shm_path != NULL)
768 strlcpy(path, shmfd->shm_path, size);
769 sx_sunlock(&shm_dict_lock);
770 }
Cache object: 53f036e88592c050fdf46970e887180d
|