3 * Anonymous Shared Memory Subsystem, ashmem
5 * Copyright (C) 2008 Google, Inc.
7 * Robert Love <rlove@google.com>
9 * This software is licensed under the terms of the GNU General Public
10 * License version 2, as published by the Free Software Foundation, and
11 * may be copied, distributed, and modified under those terms.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
19 #define pr_fmt(fmt) "ashmem: " fmt
21 #include <linux/module.h>
22 #include <linux/file.h>
24 #include <linux/falloc.h>
25 #include <linux/miscdevice.h>
26 #include <linux/security.h>
28 #include <linux/mman.h>
29 #include <linux/uaccess.h>
30 #include <linux/personality.h>
31 #include <linux/bitops.h>
32 #include <linux/mutex.h>
33 #include <linux/shmem_fs.h>
36 #define ASHMEM_NAME_PREFIX "dev/ashmem/"
37 #define ASHMEM_NAME_PREFIX_LEN (sizeof(ASHMEM_NAME_PREFIX) - 1)
38 #define ASHMEM_FULL_NAME_LEN (ASHMEM_NAME_LEN + ASHMEM_NAME_PREFIX_LEN)
41 * ashmem_area - anonymous shared memory area
42 * Lifecycle: From our parent file's open() until its release()
43 * Locking: Protected by `ashmem_mutex'
44 * Big Note: Mappings do NOT pin this structure; it dies on close()
47 char name
[ASHMEM_FULL_NAME_LEN
]; /* optional name in /proc/pid/maps */
48 struct list_head unpinned_list
; /* list of all ashmem areas */
49 struct file
*file
; /* the shmem-based backing file */
50 size_t size
; /* size of the mapping, in bytes */
51 unsigned long prot_mask
; /* allowed prot bits, as vm_flags */
55 * ashmem_range - represents an interval of unpinned (evictable) pages
56 * Lifecycle: From unpin to pin
57 * Locking: Protected by `ashmem_mutex'
60 struct list_head lru
; /* entry in LRU list */
61 struct list_head unpinned
; /* entry in its area's unpinned list */
62 struct ashmem_area
*asma
; /* associated area */
63 size_t pgstart
; /* starting page, inclusive */
64 size_t pgend
; /* ending page, inclusive */
65 unsigned int purged
; /* ASHMEM_NOT or ASHMEM_WAS_PURGED */
68 /* LRU list of unpinned pages, protected by ashmem_mutex */
69 static LIST_HEAD(ashmem_lru_list
);
71 /* Count of pages on our LRU list, protected by ashmem_mutex */
72 static unsigned long lru_count
;
75 * ashmem_mutex - protects the list of and each individual ashmem_area
77 * Lock Ordering: ashmex_mutex -> i_mutex -> i_alloc_sem
79 static DEFINE_MUTEX(ashmem_mutex
);
81 static struct kmem_cache
*ashmem_area_cachep __read_mostly
;
82 static struct kmem_cache
*ashmem_range_cachep __read_mostly
;
84 #define range_size(range) \
85 ((range)->pgend - (range)->pgstart + 1)
87 #define range_on_lru(range) \
88 ((range)->purged == ASHMEM_NOT_PURGED)
90 #define page_range_subsumes_range(range, start, end) \
91 (((range)->pgstart >= (start)) && ((range)->pgend <= (end)))
93 #define page_range_subsumed_by_range(range, start, end) \
94 (((range)->pgstart <= (start)) && ((range)->pgend >= (end)))
96 #define page_in_range(range, page) \
97 (((range)->pgstart <= (page)) && ((range)->pgend >= (page)))
99 #define page_range_in_range(range, start, end) \
100 (page_in_range(range, start) || page_in_range(range, end) || \
101 page_range_subsumes_range(range, start, end))
103 #define range_before_page(range, page) \
104 ((range)->pgend < (page))
106 #define PROT_MASK (PROT_EXEC | PROT_READ | PROT_WRITE)
108 static inline void lru_add(struct ashmem_range
*range
)
110 list_add_tail(&range
->lru
, &ashmem_lru_list
);
111 lru_count
+= range_size(range
);
114 static inline void lru_del(struct ashmem_range
*range
)
116 list_del(&range
->lru
);
117 lru_count
-= range_size(range
);
121 * range_alloc - allocate and initialize a new ashmem_range structure
123 * 'asma' - associated ashmem_area
124 * 'prev_range' - the previous ashmem_range in the sorted asma->unpinned list
125 * 'purged' - initial purge value (ASMEM_NOT_PURGED or ASHMEM_WAS_PURGED)
126 * 'start' - starting page, inclusive
127 * 'end' - ending page, inclusive
129 * Caller must hold ashmem_mutex.
131 static int range_alloc(struct ashmem_area
*asma
,
132 struct ashmem_range
*prev_range
, unsigned int purged
,
133 size_t start
, size_t end
)
135 struct ashmem_range
*range
;
137 range
= kmem_cache_zalloc(ashmem_range_cachep
, GFP_KERNEL
);
138 if (unlikely(!range
))
142 range
->pgstart
= start
;
144 range
->purged
= purged
;
146 list_add_tail(&range
->unpinned
, &prev_range
->unpinned
);
148 if (range_on_lru(range
))
154 static void range_del(struct ashmem_range
*range
)
156 list_del(&range
->unpinned
);
157 if (range_on_lru(range
))
159 kmem_cache_free(ashmem_range_cachep
, range
);
163 * range_shrink - shrinks a range
165 * Caller must hold ashmem_mutex.
167 static inline void range_shrink(struct ashmem_range
*range
,
168 size_t start
, size_t end
)
170 size_t pre
= range_size(range
);
172 range
->pgstart
= start
;
175 if (range_on_lru(range
))
176 lru_count
-= pre
- range_size(range
);
179 static int ashmem_open(struct inode
*inode
, struct file
*file
)
181 struct ashmem_area
*asma
;
184 ret
= generic_file_open(inode
, file
);
188 asma
= kmem_cache_zalloc(ashmem_area_cachep
, GFP_KERNEL
);
192 INIT_LIST_HEAD(&asma
->unpinned_list
);
193 memcpy(asma
->name
, ASHMEM_NAME_PREFIX
, ASHMEM_NAME_PREFIX_LEN
);
194 asma
->prot_mask
= PROT_MASK
;
195 file
->private_data
= asma
;
200 static int ashmem_release(struct inode
*ignored
, struct file
*file
)
202 struct ashmem_area
*asma
= file
->private_data
;
203 struct ashmem_range
*range
, *next
;
205 mutex_lock(&ashmem_mutex
);
206 list_for_each_entry_safe(range
, next
, &asma
->unpinned_list
, unpinned
)
208 mutex_unlock(&ashmem_mutex
);
212 kmem_cache_free(ashmem_area_cachep
, asma
);
217 static ssize_t
ashmem_read(struct file
*file
, char __user
*buf
,
218 size_t len
, loff_t
*pos
)
220 struct ashmem_area
*asma
= file
->private_data
;
223 mutex_lock(&ashmem_mutex
);
225 /* If size is not set, or set to 0, always return EOF. */
234 ret
= asma
->file
->f_op
->read(asma
->file
, buf
, len
, pos
);
238 /** Update backing file pos, since f_ops->read() doesn't */
239 asma
->file
->f_pos
= *pos
;
242 mutex_unlock(&ashmem_mutex
);
246 static loff_t
ashmem_llseek(struct file
*file
, loff_t offset
, int origin
)
248 struct ashmem_area
*asma
= file
->private_data
;
251 mutex_lock(&ashmem_mutex
);
253 if (asma
->size
== 0) {
263 ret
= asma
->file
->f_op
->llseek(asma
->file
, offset
, origin
);
267 /** Copy f_pos from backing file, since f_ops->llseek() sets it */
268 file
->f_pos
= asma
->file
->f_pos
;
271 mutex_unlock(&ashmem_mutex
);
275 static inline vm_flags_t
calc_vm_may_flags(unsigned long prot
)
277 return _calc_vm_trans(prot
, PROT_READ
, VM_MAYREAD
) |
278 _calc_vm_trans(prot
, PROT_WRITE
, VM_MAYWRITE
) |
279 _calc_vm_trans(prot
, PROT_EXEC
, VM_MAYEXEC
);
282 static int ashmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
284 struct ashmem_area
*asma
= file
->private_data
;
287 mutex_lock(&ashmem_mutex
);
289 /* user needs to SET_SIZE before mapping */
290 if (unlikely(!asma
->size
)) {
295 /* requested protection bits must match our allowed protection mask */
296 if (unlikely((vma
->vm_flags
& ~calc_vm_prot_bits(asma
->prot_mask
)) &
297 calc_vm_prot_bits(PROT_MASK
))) {
301 vma
->vm_flags
&= ~calc_vm_may_flags(~asma
->prot_mask
);
304 char *name
= ASHMEM_NAME_DEF
;
307 if (asma
->name
[ASHMEM_NAME_PREFIX_LEN
] != '\0')
310 /* ... and allocate the backing shmem file */
311 vmfile
= shmem_file_setup(name
, asma
->size
, vma
->vm_flags
);
312 if (unlikely(IS_ERR(vmfile
))) {
313 ret
= PTR_ERR(vmfile
);
318 get_file(asma
->file
);
321 * XXX - Reworked to use shmem_zero_setup() instead of
322 * shmem_set_file while we're in staging. -jstultz
324 if (vma
->vm_flags
& VM_SHARED
) {
325 ret
= shmem_zero_setup(vma
);
334 vma
->vm_file
= asma
->file
;
337 mutex_unlock(&ashmem_mutex
);
342 * ashmem_shrink - our cache shrinker, called from mm/vmscan.c :: shrink_slab
344 * 'nr_to_scan' is the number of objects (pages) to prune, or 0 to query how
345 * many objects (pages) we have in total.
347 * 'gfp_mask' is the mask of the allocation that got us into this mess.
349 * Return value is the number of objects (pages) remaining, or -1 if we cannot
350 * proceed without risk of deadlock (due to gfp_mask).
352 * We approximate LRU via least-recently-unpinned, jettisoning unpinned partial
353 * chunks of ashmem regions LRU-wise one-at-a-time until we hit 'nr_to_scan'
356 static int ashmem_shrink(struct shrinker
*s
, struct shrink_control
*sc
)
358 struct ashmem_range
*range
, *next
;
360 /* We might recurse into filesystem code, so bail out if necessary */
361 if (sc
->nr_to_scan
&& !(sc
->gfp_mask
& __GFP_FS
))
366 mutex_lock(&ashmem_mutex
);
367 list_for_each_entry_safe(range
, next
, &ashmem_lru_list
, lru
) {
368 loff_t start
= range
->pgstart
* PAGE_SIZE
;
369 loff_t end
= (range
->pgend
+ 1) * PAGE_SIZE
;
371 do_fallocate(range
->asma
->file
,
372 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
374 range
->purged
= ASHMEM_WAS_PURGED
;
377 sc
->nr_to_scan
-= range_size(range
);
378 if (sc
->nr_to_scan
<= 0)
381 mutex_unlock(&ashmem_mutex
);
386 static struct shrinker ashmem_shrinker
= {
387 .shrink
= ashmem_shrink
,
388 .seeks
= DEFAULT_SEEKS
* 4,
391 static int set_prot_mask(struct ashmem_area
*asma
, unsigned long prot
)
395 mutex_lock(&ashmem_mutex
);
397 /* the user can only remove, not add, protection bits */
398 if (unlikely((asma
->prot_mask
& prot
) != prot
)) {
403 /* does the application expect PROT_READ to imply PROT_EXEC? */
404 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
407 asma
->prot_mask
= prot
;
410 mutex_unlock(&ashmem_mutex
);
414 static int set_name(struct ashmem_area
*asma
, void __user
*name
)
418 mutex_lock(&ashmem_mutex
);
420 /* cannot change an existing mapping's name */
421 if (unlikely(asma
->file
)) {
426 if (unlikely(copy_from_user(asma
->name
+ ASHMEM_NAME_PREFIX_LEN
,
427 name
, ASHMEM_NAME_LEN
)))
429 asma
->name
[ASHMEM_FULL_NAME_LEN
-1] = '\0';
432 mutex_unlock(&ashmem_mutex
);
437 static int get_name(struct ashmem_area
*asma
, void __user
*name
)
441 mutex_lock(&ashmem_mutex
);
442 if (asma
->name
[ASHMEM_NAME_PREFIX_LEN
] != '\0') {
446 * Copying only `len', instead of ASHMEM_NAME_LEN, bytes
447 * prevents us from revealing one user's stack to another.
449 len
= strlen(asma
->name
+ ASHMEM_NAME_PREFIX_LEN
) + 1;
450 if (unlikely(copy_to_user(name
,
451 asma
->name
+ ASHMEM_NAME_PREFIX_LEN
, len
)))
454 if (unlikely(copy_to_user(name
, ASHMEM_NAME_DEF
,
455 sizeof(ASHMEM_NAME_DEF
))))
458 mutex_unlock(&ashmem_mutex
);
464 * ashmem_pin - pin the given ashmem region, returning whether it was
465 * previously purged (ASHMEM_WAS_PURGED) or not (ASHMEM_NOT_PURGED).
467 * Caller must hold ashmem_mutex.
469 static int ashmem_pin(struct ashmem_area
*asma
, size_t pgstart
, size_t pgend
)
471 struct ashmem_range
*range
, *next
;
472 int ret
= ASHMEM_NOT_PURGED
;
474 list_for_each_entry_safe(range
, next
, &asma
->unpinned_list
, unpinned
) {
475 /* moved past last applicable page; we can short circuit */
476 if (range_before_page(range
, pgstart
))
480 * The user can ask us to pin pages that span multiple ranges,
481 * or to pin pages that aren't even unpinned, so this is messy.
484 * 1. The requested range subsumes an existing range, so we
485 * just remove the entire matching range.
486 * 2. The requested range overlaps the start of an existing
487 * range, so we just update that range.
488 * 3. The requested range overlaps the end of an existing
489 * range, so we just update that range.
490 * 4. The requested range punches a hole in an existing range,
491 * so we have to update one side of the range and then
492 * create a new range for the other side.
494 if (page_range_in_range(range
, pgstart
, pgend
)) {
495 ret
|= range
->purged
;
497 /* Case #1: Easy. Just nuke the whole thing. */
498 if (page_range_subsumes_range(range
, pgstart
, pgend
)) {
503 /* Case #2: We overlap from the start, so adjust it */
504 if (range
->pgstart
>= pgstart
) {
505 range_shrink(range
, pgend
+ 1, range
->pgend
);
509 /* Case #3: We overlap from the rear, so adjust it */
510 if (range
->pgend
<= pgend
) {
511 range_shrink(range
, range
->pgstart
, pgstart
-1);
516 * Case #4: We eat a chunk out of the middle. A bit
517 * more complicated, we allocate a new range for the
518 * second half and adjust the first chunk's endpoint.
520 range_alloc(asma
, range
, range
->purged
,
521 pgend
+ 1, range
->pgend
);
522 range_shrink(range
, range
->pgstart
, pgstart
- 1);
531 * ashmem_unpin - unpin the given range of pages. Returns zero on success.
533 * Caller must hold ashmem_mutex.
535 static int ashmem_unpin(struct ashmem_area
*asma
, size_t pgstart
, size_t pgend
)
537 struct ashmem_range
*range
, *next
;
538 unsigned int purged
= ASHMEM_NOT_PURGED
;
541 list_for_each_entry_safe(range
, next
, &asma
->unpinned_list
, unpinned
) {
542 /* short circuit: this is our insertion point */
543 if (range_before_page(range
, pgstart
))
547 * The user can ask us to unpin pages that are already entirely
548 * or partially pinned. We handle those two cases here.
550 if (page_range_subsumed_by_range(range
, pgstart
, pgend
))
552 if (page_range_in_range(range
, pgstart
, pgend
)) {
553 pgstart
= min_t(size_t, range
->pgstart
, pgstart
),
554 pgend
= max_t(size_t, range
->pgend
, pgend
);
555 purged
|= range
->purged
;
561 return range_alloc(asma
, range
, purged
, pgstart
, pgend
);
565 * ashmem_get_pin_status - Returns ASHMEM_IS_UNPINNED if _any_ pages in the
566 * given interval are unpinned and ASHMEM_IS_PINNED otherwise.
568 * Caller must hold ashmem_mutex.
570 static int ashmem_get_pin_status(struct ashmem_area
*asma
, size_t pgstart
,
573 struct ashmem_range
*range
;
574 int ret
= ASHMEM_IS_PINNED
;
576 list_for_each_entry(range
, &asma
->unpinned_list
, unpinned
) {
577 if (range_before_page(range
, pgstart
))
579 if (page_range_in_range(range
, pgstart
, pgend
)) {
580 ret
= ASHMEM_IS_UNPINNED
;
588 static int ashmem_pin_unpin(struct ashmem_area
*asma
, unsigned long cmd
,
591 struct ashmem_pin pin
;
592 size_t pgstart
, pgend
;
595 if (unlikely(!asma
->file
))
598 if (unlikely(copy_from_user(&pin
, p
, sizeof(pin
))))
601 /* per custom, you can pass zero for len to mean "everything onward" */
603 pin
.len
= PAGE_ALIGN(asma
->size
) - pin
.offset
;
605 if (unlikely((pin
.offset
| pin
.len
) & ~PAGE_MASK
))
608 if (unlikely(((__u32
) -1) - pin
.offset
< pin
.len
))
611 if (unlikely(PAGE_ALIGN(asma
->size
) < pin
.offset
+ pin
.len
))
614 pgstart
= pin
.offset
/ PAGE_SIZE
;
615 pgend
= pgstart
+ (pin
.len
/ PAGE_SIZE
) - 1;
617 mutex_lock(&ashmem_mutex
);
621 ret
= ashmem_pin(asma
, pgstart
, pgend
);
624 ret
= ashmem_unpin(asma
, pgstart
, pgend
);
626 case ASHMEM_GET_PIN_STATUS
:
627 ret
= ashmem_get_pin_status(asma
, pgstart
, pgend
);
631 mutex_unlock(&ashmem_mutex
);
636 static long ashmem_ioctl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
638 struct ashmem_area
*asma
= file
->private_data
;
642 case ASHMEM_SET_NAME
:
643 ret
= set_name(asma
, (void __user
*) arg
);
645 case ASHMEM_GET_NAME
:
646 ret
= get_name(asma
, (void __user
*) arg
);
648 case ASHMEM_SET_SIZE
:
652 asma
->size
= (size_t) arg
;
655 case ASHMEM_GET_SIZE
:
658 case ASHMEM_SET_PROT_MASK
:
659 ret
= set_prot_mask(asma
, arg
);
661 case ASHMEM_GET_PROT_MASK
:
662 ret
= asma
->prot_mask
;
666 case ASHMEM_GET_PIN_STATUS
:
667 ret
= ashmem_pin_unpin(asma
, cmd
, (void __user
*) arg
);
669 case ASHMEM_PURGE_ALL_CACHES
:
671 if (capable(CAP_SYS_ADMIN
)) {
672 struct shrink_control sc
= {
673 .gfp_mask
= GFP_KERNEL
,
676 ret
= ashmem_shrink(&ashmem_shrinker
, &sc
);
678 ashmem_shrink(&ashmem_shrinker
, &sc
);
686 static const struct file_operations ashmem_fops
= {
687 .owner
= THIS_MODULE
,
689 .release
= ashmem_release
,
691 .llseek
= ashmem_llseek
,
693 .unlocked_ioctl
= ashmem_ioctl
,
694 .compat_ioctl
= ashmem_ioctl
,
697 static struct miscdevice ashmem_misc
= {
698 .minor
= MISC_DYNAMIC_MINOR
,
700 .fops
= &ashmem_fops
,
703 static int __init
ashmem_init(void)
707 ashmem_area_cachep
= kmem_cache_create("ashmem_area_cache",
708 sizeof(struct ashmem_area
),
710 if (unlikely(!ashmem_area_cachep
)) {
711 pr_err("failed to create slab cache\n");
715 ashmem_range_cachep
= kmem_cache_create("ashmem_range_cache",
716 sizeof(struct ashmem_range
),
718 if (unlikely(!ashmem_range_cachep
)) {
719 pr_err("failed to create slab cache\n");
723 ret
= misc_register(&ashmem_misc
);
725 pr_err("failed to register misc device!\n");
729 register_shrinker(&ashmem_shrinker
);
731 pr_info("initialized\n");
736 static void __exit
ashmem_exit(void)
740 unregister_shrinker(&ashmem_shrinker
);
742 ret
= misc_deregister(&ashmem_misc
);
744 pr_err("failed to unregister misc device!\n");
746 kmem_cache_destroy(ashmem_range_cachep
);
747 kmem_cache_destroy(ashmem_area_cachep
);
749 pr_info("unloaded\n");
752 module_init(ashmem_init
);
753 module_exit(ashmem_exit
);
755 MODULE_LICENSE("GPL");