3 * Anonymous Shared Memory Subsystem, ashmem
5 * Copyright (C) 2008 Google, Inc.
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)
417 char local_name[ASHMEM_NAME_LEN];
420 * Holding the ashmem_mutex while doing a copy_from_user might cause
421 * an data abort which would try to access mmap_sem. If another
422 * thread has invoked ashmem_mmap then it will be holding the
423 * semaphore and will be waiting for ashmem_mutex, there by leading to
424 * deadlock. We'll release the mutex and take the name to a local
425 * variable that does not need protection and later copy the local
426 * variable to the structure member with lock held.
428 if (copy_from_user(local_name, name, ASHMEM_NAME_LEN))
431 mutex_lock(&ashmem_mutex);
432 /* cannot change an existing mapping's name */
433 if (unlikely(asma->file)) {
437 memcpy(asma->name + ASHMEM_NAME_PREFIX_LEN,
438 local_name, ASHMEM_NAME_LEN);
439 asma->name[ASHMEM_FULL_NAME_LEN-1] = '\0';
441 mutex_unlock(&ashmem_mutex);
446 static int get_name(struct ashmem_area *asma, void __user *name)
451 * Have a local variable to which we'll copy the content
452 * from asma with the lock held. Later we can copy this to the user
453 * space safely without holding any locks. So even if we proceed to
454 * wait for mmap_sem, it won't lead to deadlock.
456 char local_name[ASHMEM_NAME_LEN];
458 mutex_lock(&ashmem_mutex);
459 if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0') {
462 * Copying only `len', instead of ASHMEM_NAME_LEN, bytes
463 * prevents us from revealing one user's stack to another.
465 len = strlen(asma->name + ASHMEM_NAME_PREFIX_LEN) + 1;
466 memcpy(local_name, asma->name + ASHMEM_NAME_PREFIX_LEN, len);
468 len = sizeof(ASHMEM_NAME_DEF);
469 memcpy(local_name, ASHMEM_NAME_DEF, len);
471 mutex_unlock(&ashmem_mutex);
474 * Now we are just copying from the stack variable to userland
477 if (unlikely(copy_to_user(name, local_name, len)))
483 * ashmem_pin - pin the given ashmem region, returning whether it was
484 * previously purged (ASHMEM_WAS_PURGED) or not (ASHMEM_NOT_PURGED).
486 * Caller must hold ashmem_mutex.
488 static int ashmem_pin(struct ashmem_area *asma, size_t pgstart, size_t pgend)
490 struct ashmem_range *range, *next;
491 int ret = ASHMEM_NOT_PURGED;
493 list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) {
494 /* moved past last applicable page; we can short circuit */
495 if (range_before_page(range, pgstart))
499 * The user can ask us to pin pages that span multiple ranges,
500 * or to pin pages that aren't even unpinned, so this is messy.
503 * 1. The requested range subsumes an existing range, so we
504 * just remove the entire matching range.
505 * 2. The requested range overlaps the start of an existing
506 * range, so we just update that range.
507 * 3. The requested range overlaps the end of an existing
508 * range, so we just update that range.
509 * 4. The requested range punches a hole in an existing range,
510 * so we have to update one side of the range and then
511 * create a new range for the other side.
513 if (page_range_in_range(range, pgstart, pgend)) {
514 ret |= range->purged;
516 /* Case #1: Easy. Just nuke the whole thing. */
517 if (page_range_subsumes_range(range, pgstart, pgend)) {
522 /* Case #2: We overlap from the start, so adjust it */
523 if (range->pgstart >= pgstart) {
524 range_shrink(range, pgend + 1, range->pgend);
528 /* Case #3: We overlap from the rear, so adjust it */
529 if (range->pgend <= pgend) {
530 range_shrink(range, range->pgstart, pgstart-1);
535 * Case #4: We eat a chunk out of the middle. A bit
536 * more complicated, we allocate a new range for the
537 * second half and adjust the first chunk's endpoint.
539 range_alloc(asma, range, range->purged,
540 pgend + 1, range->pgend);
541 range_shrink(range, range->pgstart, pgstart - 1);
550 * ashmem_unpin - unpin the given range of pages. Returns zero on success.
552 * Caller must hold ashmem_mutex.
554 static int ashmem_unpin(struct ashmem_area *asma, size_t pgstart, size_t pgend)
556 struct ashmem_range *range, *next;
557 unsigned int purged = ASHMEM_NOT_PURGED;
560 list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) {
561 /* short circuit: this is our insertion point */
562 if (range_before_page(range, pgstart))
566 * The user can ask us to unpin pages that are already entirely
567 * or partially pinned. We handle those two cases here.
569 if (page_range_subsumed_by_range(range, pgstart, pgend))
571 if (page_range_in_range(range, pgstart, pgend)) {
572 pgstart = min_t(size_t, range->pgstart, pgstart),
573 pgend = max_t(size_t, range->pgend, pgend);
574 purged |= range->purged;
580 return range_alloc(asma, range, purged, pgstart, pgend);
584 * ashmem_get_pin_status - Returns ASHMEM_IS_UNPINNED if _any_ pages in the
585 * given interval are unpinned and ASHMEM_IS_PINNED otherwise.
587 * Caller must hold ashmem_mutex.
589 static int ashmem_get_pin_status(struct ashmem_area *asma, size_t pgstart,
592 struct ashmem_range *range;
593 int ret = ASHMEM_IS_PINNED;
595 list_for_each_entry(range, &asma->unpinned_list, unpinned) {
596 if (range_before_page(range, pgstart))
598 if (page_range_in_range(range, pgstart, pgend)) {
599 ret = ASHMEM_IS_UNPINNED;
607 static int ashmem_pin_unpin(struct ashmem_area *asma, unsigned long cmd,
610 struct ashmem_pin pin;
611 size_t pgstart, pgend;
614 if (unlikely(!asma->file))
617 if (unlikely(copy_from_user(&pin, p, sizeof(pin))))
620 /* per custom, you can pass zero for len to mean "everything onward" */
622 pin.len = PAGE_ALIGN(asma->size) - pin.offset;
624 if (unlikely((pin.offset | pin.len) & ~PAGE_MASK))
627 if (unlikely(((__u32) -1) - pin.offset < pin.len))
630 if (unlikely(PAGE_ALIGN(asma->size) < pin.offset + pin.len))
633 pgstart = pin.offset / PAGE_SIZE;
634 pgend = pgstart + (pin.len / PAGE_SIZE) - 1;
636 mutex_lock(&ashmem_mutex);
640 ret = ashmem_pin(asma, pgstart, pgend);
643 ret = ashmem_unpin(asma, pgstart, pgend);
645 case ASHMEM_GET_PIN_STATUS:
646 ret = ashmem_get_pin_status(asma, pgstart, pgend);
650 mutex_unlock(&ashmem_mutex);
655 static long ashmem_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
657 struct ashmem_area *asma = file->private_data;
661 case ASHMEM_SET_NAME:
662 ret = set_name(asma, (void __user *) arg);
664 case ASHMEM_GET_NAME:
665 ret = get_name(asma, (void __user *) arg);
667 case ASHMEM_SET_SIZE:
671 asma->size = (size_t) arg;
674 case ASHMEM_GET_SIZE:
677 case ASHMEM_SET_PROT_MASK:
678 ret = set_prot_mask(asma, arg);
680 case ASHMEM_GET_PROT_MASK:
681 ret = asma->prot_mask;
685 case ASHMEM_GET_PIN_STATUS:
686 ret = ashmem_pin_unpin(asma, cmd, (void __user *) arg);
688 case ASHMEM_PURGE_ALL_CACHES:
690 if (capable(CAP_SYS_ADMIN)) {
691 struct shrink_control sc = {
692 .gfp_mask = GFP_KERNEL,
695 ret = ashmem_shrink(&ashmem_shrinker, &sc);
697 ashmem_shrink(&ashmem_shrinker, &sc);
705 /* support of 32bit userspace on 64bit platforms */
707 static long compat_ashmem_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
711 case COMPAT_ASHMEM_SET_SIZE:
712 cmd = ASHMEM_SET_SIZE;
714 case COMPAT_ASHMEM_SET_PROT_MASK:
715 cmd = ASHMEM_SET_PROT_MASK;
718 return ashmem_ioctl(file, cmd, arg);
722 static const struct file_operations ashmem_fops = {
723 .owner = THIS_MODULE,
725 .release = ashmem_release,
727 .llseek = ashmem_llseek,
729 .unlocked_ioctl = ashmem_ioctl,
731 .compat_ioctl = compat_ashmem_ioctl,
735 static struct miscdevice ashmem_misc = {
736 .minor = MISC_DYNAMIC_MINOR,
738 .fops = &ashmem_fops,
741 static int __init ashmem_init(void)
745 ashmem_area_cachep = kmem_cache_create("ashmem_area_cache",
746 sizeof(struct ashmem_area),
748 if (unlikely(!ashmem_area_cachep)) {
749 pr_err("failed to create slab cache\n");
753 ashmem_range_cachep = kmem_cache_create("ashmem_range_cache",
754 sizeof(struct ashmem_range),
756 if (unlikely(!ashmem_range_cachep)) {
757 pr_err("failed to create slab cache\n");
761 ret = misc_register(&ashmem_misc);
763 pr_err("failed to register misc device!\n");
767 register_shrinker(&ashmem_shrinker);
769 pr_info("initialized\n");
774 static void __exit ashmem_exit(void)
778 unregister_shrinker(&ashmem_shrinker);
780 ret = misc_deregister(&ashmem_misc);
782 pr_err("failed to unregister misc device!\n");
784 kmem_cache_destroy(ashmem_range_cachep);
785 kmem_cache_destroy(ashmem_area_cachep);
787 pr_info("unloaded\n");
790 module_init(ashmem_init);
791 module_exit(ashmem_exit);
793 MODULE_LICENSE("GPL");