]>
Commit | Line | Data |
---|---|---|
bdcb7be6 | 1 | // SPDX-License-Identifier: GPL-2.0 |
11980c2a | 2 | /* mm/ashmem.c |
2258937b CJB |
3 | * |
4 | * Anonymous Shared Memory Subsystem, ashmem | |
5 | * | |
6 | * Copyright (C) 2008 Google, Inc. | |
7 | * | |
8 | * Robert Love <rlove@google.com> | |
2258937b | 9 | */ |
11980c2a | 10 | |
c810a399 SK |
11 | #define pr_fmt(fmt) "ashmem: " fmt |
12 | ||
12950595 PG |
13 | #include <linux/init.h> |
14 | #include <linux/export.h> | |
11980c2a RL |
15 | #include <linux/file.h> |
16 | #include <linux/fs.h> | |
3f31d075 | 17 | #include <linux/falloc.h> |
11980c2a RL |
18 | #include <linux/miscdevice.h> |
19 | #include <linux/security.h> | |
20 | #include <linux/mm.h> | |
21 | #include <linux/mman.h> | |
7670e053 | 22 | #include <linux/module.h> |
11980c2a RL |
23 | #include <linux/uaccess.h> |
24 | #include <linux/personality.h> | |
25 | #include <linux/bitops.h> | |
26 | #include <linux/mutex.h> | |
27 | #include <linux/shmem_fs.h> | |
28 | #include "ashmem.h" | |
29 | ||
30 | #define ASHMEM_NAME_PREFIX "dev/ashmem/" | |
31 | #define ASHMEM_NAME_PREFIX_LEN (sizeof(ASHMEM_NAME_PREFIX) - 1) | |
32 | #define ASHMEM_FULL_NAME_LEN (ASHMEM_NAME_LEN + ASHMEM_NAME_PREFIX_LEN) | |
33 | ||
4d2c9d5d CJB |
34 | /** |
35 | * struct ashmem_area - The anonymous shared memory area | |
36 | * @name: The optional name in /proc/pid/maps | |
37 | * @unpinned_list: The list of all ashmem areas | |
38 | * @file: The shmem-based backing file | |
39 | * @size: The size of the mapping, in bytes | |
7d92ea5d | 40 | * @prot_mask: The allowed protection bits, as vm_flags |
4d2c9d5d CJB |
41 | * |
42 | * The lifecycle of this structure is from our parent file's open() until | |
43 | * its release(). It is also protected by 'ashmem_mutex' | |
44 | * | |
45 | * Warning: Mappings do NOT pin this structure; It dies on close() | |
11980c2a RL |
46 | */ |
47 | struct ashmem_area { | |
4d2c9d5d CJB |
48 | char name[ASHMEM_FULL_NAME_LEN]; |
49 | struct list_head unpinned_list; | |
50 | struct file *file; | |
51 | size_t size; | |
52 | unsigned long prot_mask; | |
11980c2a RL |
53 | }; |
54 | ||
4d2c9d5d CJB |
55 | /** |
56 | * struct ashmem_range - A range of unpinned/evictable pages | |
57 | * @lru: The entry in the LRU list | |
58 | * @unpinned: The entry in its area's unpinned list | |
59 | * @asma: The associated anonymous shared memory area. | |
60 | * @pgstart: The starting page (inclusive) | |
61 | * @pgend: The ending page (inclusive) | |
62 | * @purged: The purge status (ASHMEM_NOT or ASHMEM_WAS_PURGED) | |
63 | * | |
64 | * The lifecycle of this structure is from unpin to pin. | |
65 | * It is protected by 'ashmem_mutex' | |
11980c2a RL |
66 | */ |
67 | struct ashmem_range { | |
4d2c9d5d CJB |
68 | struct list_head lru; |
69 | struct list_head unpinned; | |
70 | struct ashmem_area *asma; | |
71 | size_t pgstart; | |
72 | size_t pgend; | |
73 | unsigned int purged; | |
11980c2a RL |
74 | }; |
75 | ||
76 | /* LRU list of unpinned pages, protected by ashmem_mutex */ | |
77 | static LIST_HEAD(ashmem_lru_list); | |
78 | ||
fb4415a1 TH |
79 | static atomic_t ashmem_shrink_inflight = ATOMIC_INIT(0); |
80 | static DECLARE_WAIT_QUEUE_HEAD(ashmem_shrink_wait); | |
81 | ||
801b798c | 82 | /* |
4d2c9d5d CJB |
83 | * long lru_count - The count of pages on our LRU list. |
84 | * | |
85 | * This is protected by ashmem_mutex. | |
86 | */ | |
11980c2a RL |
87 | static unsigned long lru_count; |
88 | ||
801b798c | 89 | /* |
11980c2a RL |
90 | * ashmem_mutex - protects the list of and each individual ashmem_area |
91 | * | |
92 | * Lock Ordering: ashmex_mutex -> i_mutex -> i_alloc_sem | |
93 | */ | |
94 | static DEFINE_MUTEX(ashmem_mutex); | |
95 | ||
96 | static struct kmem_cache *ashmem_area_cachep __read_mostly; | |
97 | static struct kmem_cache *ashmem_range_cachep __read_mostly; | |
98 | ||
1e1f7ae0 SB |
99 | /* |
100 | * A separate lockdep class for the backing shmem inodes to resolve the lockdep | |
101 | * warning about the race between kswapd taking fs_reclaim before inode_lock | |
102 | * and write syscall taking inode_lock and then fs_reclaim. | |
103 | * Note that such race is impossible because ashmem does not support write | |
104 | * syscalls operating on the backing shmem. | |
105 | */ | |
106 | static struct lock_class_key backing_shmem_inode_class; | |
107 | ||
40270ca0 GT |
108 | static inline unsigned long range_size(struct ashmem_range *range) |
109 | { | |
110 | return range->pgend - range->pgstart + 1; | |
111 | } | |
11980c2a | 112 | |
40270ca0 GT |
113 | static inline bool range_on_lru(struct ashmem_range *range) |
114 | { | |
115 | return range->purged == ASHMEM_NOT_PURGED; | |
116 | } | |
11980c2a | 117 | |
8d60b476 GT |
118 | static inline bool page_range_subsumes_range(struct ashmem_range *range, |
119 | size_t start, size_t end) | |
c0ece6c3 | 120 | { |
8d60b476 | 121 | return (range->pgstart >= start) && (range->pgend <= end); |
c0ece6c3 | 122 | } |
11980c2a | 123 | |
8d60b476 GT |
124 | static inline bool page_range_subsumed_by_range(struct ashmem_range *range, |
125 | size_t start, size_t end) | |
c0ece6c3 | 126 | { |
8d60b476 | 127 | return (range->pgstart <= start) && (range->pgend >= end); |
c0ece6c3 | 128 | } |
11980c2a | 129 | |
8d60b476 | 130 | static inline bool page_in_range(struct ashmem_range *range, size_t page) |
e2a83f32 | 131 | { |
8d60b476 | 132 | return (range->pgstart <= page) && (range->pgend >= page); |
e2a83f32 | 133 | } |
11980c2a | 134 | |
8d60b476 GT |
135 | static inline bool page_range_in_range(struct ashmem_range *range, |
136 | size_t start, size_t end) | |
c0ece6c3 | 137 | { |
8d60b476 GT |
138 | return page_in_range(range, start) || page_in_range(range, end) || |
139 | page_range_subsumes_range(range, start, end); | |
c0ece6c3 | 140 | } |
11980c2a | 141 | |
c4089864 H |
142 | static inline bool range_before_page(struct ashmem_range *range, |
143 | size_t page) | |
e2a83f32 | 144 | { |
8d60b476 | 145 | return range->pgend < page; |
e2a83f32 | 146 | } |
11980c2a RL |
147 | |
148 | #define PROT_MASK (PROT_EXEC | PROT_READ | PROT_WRITE) | |
149 | ||
4d2c9d5d CJB |
150 | /** |
151 | * lru_add() - Adds a range of memory to the LRU list | |
152 | * @range: The memory range being added. | |
153 | * | |
154 | * The range is first added to the end (tail) of the LRU list. | |
155 | * After this, the size of the range is added to @lru_count | |
156 | */ | |
11980c2a RL |
157 | static inline void lru_add(struct ashmem_range *range) |
158 | { | |
159 | list_add_tail(&range->lru, &ashmem_lru_list); | |
160 | lru_count += range_size(range); | |
161 | } | |
162 | ||
4d2c9d5d CJB |
163 | /** |
164 | * lru_del() - Removes a range of memory from the LRU list | |
165 | * @range: The memory range being removed | |
166 | * | |
167 | * The range is first deleted from the LRU list. | |
168 | * After this, the size of the range is removed from @lru_count | |
169 | */ | |
11980c2a RL |
170 | static inline void lru_del(struct ashmem_range *range) |
171 | { | |
172 | list_del(&range->lru); | |
173 | lru_count -= range_size(range); | |
174 | } | |
175 | ||
4d2c9d5d CJB |
176 | /** |
177 | * range_alloc() - Allocates and initializes a new ashmem_range structure | |
178 | * @asma: The associated ashmem_area | |
179 | * @prev_range: The previous ashmem_range in the sorted asma->unpinned list | |
180 | * @purged: Initial purge status (ASMEM_NOT_PURGED or ASHMEM_WAS_PURGED) | |
181 | * @start: The starting page (inclusive) | |
182 | * @end: The ending page (inclusive) | |
11980c2a | 183 | * |
4d2c9d5d | 184 | * This function is protected by ashmem_mutex. |
11980c2a | 185 | */ |
ecd182cb TH |
186 | static void range_alloc(struct ashmem_area *asma, |
187 | struct ashmem_range *prev_range, unsigned int purged, | |
188 | size_t start, size_t end, | |
189 | struct ashmem_range **new_range) | |
11980c2a | 190 | { |
ecd182cb | 191 | struct ashmem_range *range = *new_range; |
11980c2a | 192 | |
ecd182cb | 193 | *new_range = NULL; |
11980c2a RL |
194 | range->asma = asma; |
195 | range->pgstart = start; | |
196 | range->pgend = end; | |
197 | range->purged = purged; | |
198 | ||
199 | list_add_tail(&range->unpinned, &prev_range->unpinned); | |
200 | ||
201 | if (range_on_lru(range)) | |
202 | lru_add(range); | |
11980c2a RL |
203 | } |
204 | ||
4d2c9d5d | 205 | /** |
1de24f36 | 206 | * range_del() - Deletes and deallocates an ashmem_range structure |
4d2c9d5d CJB |
207 | * @range: The associated ashmem_range that has previously been allocated |
208 | */ | |
11980c2a RL |
209 | static void range_del(struct ashmem_range *range) |
210 | { | |
211 | list_del(&range->unpinned); | |
212 | if (range_on_lru(range)) | |
213 | lru_del(range); | |
214 | kmem_cache_free(ashmem_range_cachep, range); | |
215 | } | |
216 | ||
781114ce CJB |
217 | /** |
218 | * range_shrink() - Shrinks an ashmem_range | |
219 | * @range: The associated ashmem_range being shrunk | |
220 | * @start: The starting byte of the new range | |
221 | * @end: The ending byte of the new range | |
222 | * | |
223 | * This does not modify the data inside the existing range in any way - It | |
224 | * simply shrinks the boundaries of the range. | |
11980c2a | 225 | * |
781114ce CJB |
226 | * Theoretically, with a little tweaking, this could eventually be changed |
227 | * to range_resize, and expand the lru_count if the new range is larger. | |
11980c2a RL |
228 | */ |
229 | static inline void range_shrink(struct ashmem_range *range, | |
230 | size_t start, size_t end) | |
231 | { | |
232 | size_t pre = range_size(range); | |
233 | ||
234 | range->pgstart = start; | |
235 | range->pgend = end; | |
236 | ||
237 | if (range_on_lru(range)) | |
238 | lru_count -= pre - range_size(range); | |
239 | } | |
240 | ||
781114ce CJB |
241 | /** |
242 | * ashmem_open() - Opens an Anonymous Shared Memory structure | |
243 | * @inode: The backing file's index node(?) | |
244 | * @file: The backing file | |
245 | * | |
246 | * Please note that the ashmem_area is not returned by this function - It is | |
247 | * instead written to "file->private_data". | |
248 | * | |
249 | * Return: 0 if successful, or another code if unsuccessful. | |
250 | */ | |
11980c2a RL |
251 | static int ashmem_open(struct inode *inode, struct file *file) |
252 | { | |
253 | struct ashmem_area *asma; | |
254 | int ret; | |
255 | ||
5154b93b | 256 | ret = generic_file_open(inode, file); |
59848d6a | 257 | if (ret) |
11980c2a RL |
258 | return ret; |
259 | ||
260 | asma = kmem_cache_zalloc(ashmem_area_cachep, GFP_KERNEL); | |
59848d6a | 261 | if (!asma) |
11980c2a RL |
262 | return -ENOMEM; |
263 | ||
264 | INIT_LIST_HEAD(&asma->unpinned_list); | |
265 | memcpy(asma->name, ASHMEM_NAME_PREFIX, ASHMEM_NAME_PREFIX_LEN); | |
266 | asma->prot_mask = PROT_MASK; | |
267 | file->private_data = asma; | |
268 | ||
269 | return 0; | |
270 | } | |
271 | ||
781114ce CJB |
272 | /** |
273 | * ashmem_release() - Releases an Anonymous Shared Memory structure | |
274 | * @ignored: The backing file's Index Node(?) - It is ignored here. | |
275 | * @file: The backing file | |
276 | * | |
277 | * Return: 0 if successful. If it is anything else, go have a coffee and | |
278 | * try again. | |
279 | */ | |
11980c2a RL |
280 | static int ashmem_release(struct inode *ignored, struct file *file) |
281 | { | |
282 | struct ashmem_area *asma = file->private_data; | |
283 | struct ashmem_range *range, *next; | |
284 | ||
285 | mutex_lock(&ashmem_mutex); | |
286 | list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) | |
287 | range_del(range); | |
288 | mutex_unlock(&ashmem_mutex); | |
289 | ||
290 | if (asma->file) | |
291 | fput(asma->file); | |
292 | kmem_cache_free(ashmem_area_cachep, asma); | |
293 | ||
294 | return 0; | |
295 | } | |
296 | ||
8a2af064 | 297 | static ssize_t ashmem_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
853ca7ae | 298 | { |
8a2af064 | 299 | struct ashmem_area *asma = iocb->ki_filp->private_data; |
853ca7ae BB |
300 | int ret = 0; |
301 | ||
302 | mutex_lock(&ashmem_mutex); | |
303 | ||
304 | /* If size is not set, or set to 0, always return EOF. */ | |
1efb3439 | 305 | if (asma->size == 0) |
077f6db9 | 306 | goto out_unlock; |
853ca7ae BB |
307 | |
308 | if (!asma->file) { | |
309 | ret = -EBADF; | |
077f6db9 | 310 | goto out_unlock; |
853ca7ae BB |
311 | } |
312 | ||
077f6db9 TP |
313 | /* |
314 | * asma and asma->file are used outside the lock here. We assume | |
315 | * once asma->file is set it will never be changed, and will not | |
316 | * be destroyed until all references to the file are dropped and | |
317 | * ashmem_release is called. | |
318 | */ | |
8a2af064 CH |
319 | mutex_unlock(&ashmem_mutex); |
320 | ret = vfs_iter_read(asma->file, iter, &iocb->ki_pos, 0); | |
321 | mutex_lock(&ashmem_mutex); | |
322 | if (ret > 0) | |
323 | asma->file->f_pos = iocb->ki_pos; | |
077f6db9 | 324 | out_unlock: |
5154b93b BB |
325 | mutex_unlock(&ashmem_mutex); |
326 | return ret; | |
327 | } | |
328 | ||
329 | static loff_t ashmem_llseek(struct file *file, loff_t offset, int origin) | |
330 | { | |
331 | struct ashmem_area *asma = file->private_data; | |
44f68936 | 332 | loff_t ret; |
5154b93b BB |
333 | |
334 | mutex_lock(&ashmem_mutex); | |
335 | ||
336 | if (asma->size == 0) { | |
cb57469c JF |
337 | mutex_unlock(&ashmem_mutex); |
338 | return -EINVAL; | |
5154b93b BB |
339 | } |
340 | ||
341 | if (!asma->file) { | |
cb57469c JF |
342 | mutex_unlock(&ashmem_mutex); |
343 | return -EBADF; | |
5154b93b BB |
344 | } |
345 | ||
cb57469c JF |
346 | mutex_unlock(&ashmem_mutex); |
347 | ||
91360b02 | 348 | ret = vfs_llseek(asma->file, offset, origin); |
1efb3439 | 349 | if (ret < 0) |
cb57469c | 350 | return ret; |
5154b93b BB |
351 | |
352 | /** Copy f_pos from backing file, since f_ops->llseek() sets it */ | |
353 | file->f_pos = asma->file->f_pos; | |
853ca7ae BB |
354 | return ret; |
355 | } | |
356 | ||
7cfce77d | 357 | static inline vm_flags_t calc_vm_may_flags(unsigned long prot) |
56f76fc6 | 358 | { |
1efb3439 | 359 | return _calc_vm_trans(prot, PROT_READ, VM_MAYREAD) | |
56f76fc6 AH |
360 | _calc_vm_trans(prot, PROT_WRITE, VM_MAYWRITE) | |
361 | _calc_vm_trans(prot, PROT_EXEC, VM_MAYEXEC); | |
362 | } | |
853ca7ae | 363 | |
d024db74 SB |
364 | static int ashmem_vmfile_mmap(struct file *file, struct vm_area_struct *vma) |
365 | { | |
366 | /* do not allow to mmap ashmem backing shmem file directly */ | |
367 | return -EPERM; | |
368 | } | |
369 | ||
370 | static unsigned long | |
371 | ashmem_vmfile_get_unmapped_area(struct file *file, unsigned long addr, | |
372 | unsigned long len, unsigned long pgoff, | |
373 | unsigned long flags) | |
374 | { | |
375 | return current->mm->get_unmapped_area(file, addr, len, pgoff, flags); | |
376 | } | |
377 | ||
11980c2a RL |
378 | static int ashmem_mmap(struct file *file, struct vm_area_struct *vma) |
379 | { | |
d024db74 | 380 | static struct file_operations vmfile_fops; |
11980c2a RL |
381 | struct ashmem_area *asma = file->private_data; |
382 | int ret = 0; | |
383 | ||
384 | mutex_lock(&ashmem_mutex); | |
385 | ||
386 | /* user needs to SET_SIZE before mapping */ | |
59848d6a | 387 | if (!asma->size) { |
11980c2a RL |
388 | ret = -EINVAL; |
389 | goto out; | |
390 | } | |
391 | ||
8632c614 AS |
392 | /* requested mapping size larger than object size */ |
393 | if (vma->vm_end - vma->vm_start > PAGE_ALIGN(asma->size)) { | |
11980c2a RL |
394 | ret = -EINVAL; |
395 | goto out; | |
396 | } | |
397 | ||
398 | /* requested protection bits must match our allowed protection mask */ | |
59848d6a AS |
399 | if ((vma->vm_flags & ~calc_vm_prot_bits(asma->prot_mask, 0)) & |
400 | calc_vm_prot_bits(PROT_MASK, 0)) { | |
11980c2a RL |
401 | ret = -EPERM; |
402 | goto out; | |
403 | } | |
56f76fc6 | 404 | vma->vm_flags &= ~calc_vm_may_flags(~asma->prot_mask); |
11980c2a RL |
405 | |
406 | if (!asma->file) { | |
407 | char *name = ASHMEM_NAME_DEF; | |
408 | struct file *vmfile; | |
1e1f7ae0 | 409 | struct inode *inode; |
11980c2a RL |
410 | |
411 | if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0') | |
412 | name = asma->name; | |
413 | ||
414 | /* ... and allocate the backing shmem file */ | |
415 | vmfile = shmem_file_setup(name, asma->size, vma->vm_flags); | |
7f44cb0b | 416 | if (IS_ERR(vmfile)) { |
11980c2a RL |
417 | ret = PTR_ERR(vmfile); |
418 | goto out; | |
419 | } | |
97fbfef6 | 420 | vmfile->f_mode |= FMODE_LSEEK; |
1e1f7ae0 SB |
421 | inode = file_inode(vmfile); |
422 | lockdep_set_class(&inode->i_rwsem, &backing_shmem_inode_class); | |
11980c2a | 423 | asma->file = vmfile; |
d024db74 SB |
424 | /* |
425 | * override mmap operation of the vmfile so that it can't be | |
426 | * remapped which would lead to creation of a new vma with no | |
427 | * asma permission checks. Have to override get_unmapped_area | |
428 | * as well to prevent VM_BUG_ON check for f_ops modification. | |
429 | */ | |
430 | if (!vmfile_fops.mmap) { | |
431 | vmfile_fops = *vmfile->f_op; | |
432 | vmfile_fops.mmap = ashmem_vmfile_mmap; | |
433 | vmfile_fops.get_unmapped_area = | |
434 | ashmem_vmfile_get_unmapped_area; | |
435 | } | |
436 | vmfile->f_op = &vmfile_fops; | |
11980c2a RL |
437 | } |
438 | get_file(asma->file); | |
439 | ||
440 | /* | |
aa5af974 | 441 | * XXX - Reworked to use shmem_zero_setup() instead of |
11980c2a RL |
442 | * shmem_set_file while we're in staging. -jstultz |
443 | */ | |
444 | if (vma->vm_flags & VM_SHARED) { | |
445 | ret = shmem_zero_setup(vma); | |
446 | if (ret) { | |
447 | fput(asma->file); | |
448 | goto out; | |
449 | } | |
44960f2a JS |
450 | } else { |
451 | vma_set_anonymous(vma); | |
11980c2a RL |
452 | } |
453 | ||
454 | if (vma->vm_file) | |
455 | fput(vma->vm_file); | |
456 | vma->vm_file = asma->file; | |
11980c2a RL |
457 | |
458 | out: | |
459 | mutex_unlock(&ashmem_mutex); | |
460 | return ret; | |
461 | } | |
462 | ||
463 | /* | |
6b4f7799 | 464 | * ashmem_shrink - our cache shrinker, called from mm/vmscan.c |
11980c2a | 465 | * |
7dc19d5a | 466 | * 'nr_to_scan' is the number of objects to scan for freeing. |
11980c2a RL |
467 | * |
468 | * 'gfp_mask' is the mask of the allocation that got us into this mess. | |
469 | * | |
7dc19d5a | 470 | * Return value is the number of objects freed or -1 if we cannot |
11980c2a RL |
471 | * proceed without risk of deadlock (due to gfp_mask). |
472 | * | |
473 | * We approximate LRU via least-recently-unpinned, jettisoning unpinned partial | |
474 | * chunks of ashmem regions LRU-wise one-at-a-time until we hit 'nr_to_scan' | |
475 | * pages freed. | |
476 | */ | |
7dc19d5a DC |
477 | static unsigned long |
478 | ashmem_shrink_scan(struct shrinker *shrink, struct shrink_control *sc) | |
11980c2a | 479 | { |
7dc19d5a | 480 | unsigned long freed = 0; |
11980c2a RL |
481 | |
482 | /* We might recurse into filesystem code, so bail out if necessary */ | |
7dc19d5a DC |
483 | if (!(sc->gfp_mask & __GFP_FS)) |
484 | return SHRINK_STOP; | |
11980c2a | 485 | |
18e77054 LA |
486 | if (!mutex_trylock(&ashmem_mutex)) |
487 | return -1; | |
488 | ||
fb4415a1 TH |
489 | while (!list_empty(&ashmem_lru_list)) { |
490 | struct ashmem_range *range = | |
491 | list_first_entry(&ashmem_lru_list, typeof(*range), lru); | |
11980c2a | 492 | loff_t start = range->pgstart * PAGE_SIZE; |
3f31d075 | 493 | loff_t end = (range->pgend + 1) * PAGE_SIZE; |
fb4415a1 | 494 | struct file *f = range->asma->file; |
11980c2a | 495 | |
fb4415a1 TH |
496 | get_file(f); |
497 | atomic_inc(&ashmem_shrink_inflight); | |
11980c2a RL |
498 | range->purged = ASHMEM_WAS_PURGED; |
499 | lru_del(range); | |
500 | ||
7dc19d5a | 501 | freed += range_size(range); |
fb4415a1 TH |
502 | mutex_unlock(&ashmem_mutex); |
503 | f->f_op->fallocate(f, | |
504 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, | |
505 | start, end - start); | |
506 | fput(f); | |
507 | if (atomic_dec_and_test(&ashmem_shrink_inflight)) | |
508 | wake_up_all(&ashmem_shrink_wait); | |
509 | if (!mutex_trylock(&ashmem_mutex)) | |
510 | goto out; | |
7dc19d5a | 511 | if (--sc->nr_to_scan <= 0) |
11980c2a RL |
512 | break; |
513 | } | |
514 | mutex_unlock(&ashmem_mutex); | |
fb4415a1 | 515 | out: |
7dc19d5a DC |
516 | return freed; |
517 | } | |
11980c2a | 518 | |
7dc19d5a DC |
519 | static unsigned long |
520 | ashmem_shrink_count(struct shrinker *shrink, struct shrink_control *sc) | |
521 | { | |
522 | /* | |
523 | * note that lru_count is count of pages on the lru, not a count of | |
524 | * objects on the list. This means the scan function needs to return the | |
525 | * number of pages freed, not the number of objects scanned. | |
526 | */ | |
11980c2a RL |
527 | return lru_count; |
528 | } | |
529 | ||
530 | static struct shrinker ashmem_shrinker = { | |
7dc19d5a DC |
531 | .count_objects = ashmem_shrink_count, |
532 | .scan_objects = ashmem_shrink_scan, | |
533 | /* | |
534 | * XXX (dchinner): I wish people would comment on why they need on | |
535 | * significant changes to the default value here | |
536 | */ | |
11980c2a RL |
537 | .seeks = DEFAULT_SEEKS * 4, |
538 | }; | |
539 | ||
540 | static int set_prot_mask(struct ashmem_area *asma, unsigned long prot) | |
541 | { | |
542 | int ret = 0; | |
543 | ||
544 | mutex_lock(&ashmem_mutex); | |
545 | ||
546 | /* the user can only remove, not add, protection bits */ | |
59848d6a | 547 | if ((asma->prot_mask & prot) != prot) { |
11980c2a RL |
548 | ret = -EINVAL; |
549 | goto out; | |
550 | } | |
551 | ||
552 | /* does the application expect PROT_READ to imply PROT_EXEC? */ | |
553 | if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) | |
554 | prot |= PROT_EXEC; | |
555 | ||
556 | asma->prot_mask = prot; | |
557 | ||
558 | out: | |
559 | mutex_unlock(&ashmem_mutex); | |
560 | return ret; | |
561 | } | |
562 | ||
563 | static int set_name(struct ashmem_area *asma, void __user *name) | |
564 | { | |
077f6db9 | 565 | int len; |
11980c2a | 566 | int ret = 0; |
e5834d62 | 567 | char local_name[ASHMEM_NAME_LEN]; |
11980c2a | 568 | |
e5834d62 SB |
569 | /* |
570 | * Holding the ashmem_mutex while doing a copy_from_user might cause | |
571 | * an data abort which would try to access mmap_sem. If another | |
572 | * thread has invoked ashmem_mmap then it will be holding the | |
573 | * semaphore and will be waiting for ashmem_mutex, there by leading to | |
1de24f36 | 574 | * deadlock. We'll release the mutex and take the name to a local |
e5834d62 SB |
575 | * variable that does not need protection and later copy the local |
576 | * variable to the structure member with lock held. | |
577 | */ | |
077f6db9 TP |
578 | len = strncpy_from_user(local_name, name, ASHMEM_NAME_LEN); |
579 | if (len < 0) | |
580 | return len; | |
581 | if (len == ASHMEM_NAME_LEN) | |
582 | local_name[ASHMEM_NAME_LEN - 1] = '\0'; | |
e5834d62 | 583 | mutex_lock(&ashmem_mutex); |
11980c2a | 584 | /* cannot change an existing mapping's name */ |
59848d6a | 585 | if (asma->file) |
11980c2a | 586 | ret = -EINVAL; |
077f6db9 TP |
587 | else |
588 | strcpy(asma->name + ASHMEM_NAME_PREFIX_LEN, local_name); | |
11980c2a | 589 | |
077f6db9 | 590 | mutex_unlock(&ashmem_mutex); |
11980c2a RL |
591 | return ret; |
592 | } | |
593 | ||
594 | static int get_name(struct ashmem_area *asma, void __user *name) | |
595 | { | |
596 | int ret = 0; | |
e5834d62 SB |
597 | size_t len; |
598 | /* | |
599 | * Have a local variable to which we'll copy the content | |
600 | * from asma with the lock held. Later we can copy this to the user | |
601 | * space safely without holding any locks. So even if we proceed to | |
602 | * wait for mmap_sem, it won't lead to deadlock. | |
603 | */ | |
604 | char local_name[ASHMEM_NAME_LEN]; | |
11980c2a RL |
605 | |
606 | mutex_lock(&ashmem_mutex); | |
607 | if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0') { | |
11980c2a RL |
608 | /* |
609 | * Copying only `len', instead of ASHMEM_NAME_LEN, bytes | |
610 | * prevents us from revealing one user's stack to another. | |
611 | */ | |
612 | len = strlen(asma->name + ASHMEM_NAME_PREFIX_LEN) + 1; | |
e5834d62 | 613 | memcpy(local_name, asma->name + ASHMEM_NAME_PREFIX_LEN, len); |
11980c2a | 614 | } else { |
e5834d62 SB |
615 | len = sizeof(ASHMEM_NAME_DEF); |
616 | memcpy(local_name, ASHMEM_NAME_DEF, len); | |
11980c2a RL |
617 | } |
618 | mutex_unlock(&ashmem_mutex); | |
619 | ||
e5834d62 SB |
620 | /* |
621 | * Now we are just copying from the stack variable to userland | |
622 | * No lock held | |
623 | */ | |
59848d6a | 624 | if (copy_to_user(name, local_name, len)) |
e5834d62 | 625 | ret = -EFAULT; |
11980c2a RL |
626 | return ret; |
627 | } | |
628 | ||
629 | /* | |
630 | * ashmem_pin - pin the given ashmem region, returning whether it was | |
631 | * previously purged (ASHMEM_WAS_PURGED) or not (ASHMEM_NOT_PURGED). | |
632 | * | |
633 | * Caller must hold ashmem_mutex. | |
634 | */ | |
ecd182cb TH |
635 | static int ashmem_pin(struct ashmem_area *asma, size_t pgstart, size_t pgend, |
636 | struct ashmem_range **new_range) | |
11980c2a RL |
637 | { |
638 | struct ashmem_range *range, *next; | |
639 | int ret = ASHMEM_NOT_PURGED; | |
640 | ||
641 | list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) { | |
642 | /* moved past last applicable page; we can short circuit */ | |
643 | if (range_before_page(range, pgstart)) | |
644 | break; | |
645 | ||
646 | /* | |
647 | * The user can ask us to pin pages that span multiple ranges, | |
648 | * or to pin pages that aren't even unpinned, so this is messy. | |
649 | * | |
650 | * Four cases: | |
651 | * 1. The requested range subsumes an existing range, so we | |
652 | * just remove the entire matching range. | |
653 | * 2. The requested range overlaps the start of an existing | |
654 | * range, so we just update that range. | |
655 | * 3. The requested range overlaps the end of an existing | |
656 | * range, so we just update that range. | |
657 | * 4. The requested range punches a hole in an existing range, | |
658 | * so we have to update one side of the range and then | |
659 | * create a new range for the other side. | |
660 | */ | |
661 | if (page_range_in_range(range, pgstart, pgend)) { | |
662 | ret |= range->purged; | |
663 | ||
664 | /* Case #1: Easy. Just nuke the whole thing. */ | |
665 | if (page_range_subsumes_range(range, pgstart, pgend)) { | |
666 | range_del(range); | |
667 | continue; | |
668 | } | |
669 | ||
670 | /* Case #2: We overlap from the start, so adjust it */ | |
671 | if (range->pgstart >= pgstart) { | |
672 | range_shrink(range, pgend + 1, range->pgend); | |
673 | continue; | |
674 | } | |
675 | ||
676 | /* Case #3: We overlap from the rear, so adjust it */ | |
677 | if (range->pgend <= pgend) { | |
d2e4f687 PS |
678 | range_shrink(range, range->pgstart, |
679 | pgstart - 1); | |
11980c2a RL |
680 | continue; |
681 | } | |
682 | ||
683 | /* | |
684 | * Case #4: We eat a chunk out of the middle. A bit | |
685 | * more complicated, we allocate a new range for the | |
686 | * second half and adjust the first chunk's endpoint. | |
687 | */ | |
688 | range_alloc(asma, range, range->purged, | |
ecd182cb | 689 | pgend + 1, range->pgend, new_range); |
11980c2a RL |
690 | range_shrink(range, range->pgstart, pgstart - 1); |
691 | break; | |
692 | } | |
693 | } | |
694 | ||
695 | return ret; | |
696 | } | |
697 | ||
698 | /* | |
699 | * ashmem_unpin - unpin the given range of pages. Returns zero on success. | |
700 | * | |
701 | * Caller must hold ashmem_mutex. | |
702 | */ | |
ecd182cb TH |
703 | static int ashmem_unpin(struct ashmem_area *asma, size_t pgstart, size_t pgend, |
704 | struct ashmem_range **new_range) | |
11980c2a RL |
705 | { |
706 | struct ashmem_range *range, *next; | |
707 | unsigned int purged = ASHMEM_NOT_PURGED; | |
708 | ||
709 | restart: | |
710 | list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) { | |
711 | /* short circuit: this is our insertion point */ | |
712 | if (range_before_page(range, pgstart)) | |
713 | break; | |
714 | ||
715 | /* | |
716 | * The user can ask us to unpin pages that are already entirely | |
717 | * or partially pinned. We handle those two cases here. | |
718 | */ | |
719 | if (page_range_subsumed_by_range(range, pgstart, pgend)) | |
720 | return 0; | |
721 | if (page_range_in_range(range, pgstart, pgend)) { | |
9f1c427d AKC |
722 | pgstart = min(range->pgstart, pgstart); |
723 | pgend = max(range->pgend, pgend); | |
11980c2a RL |
724 | purged |= range->purged; |
725 | range_del(range); | |
726 | goto restart; | |
727 | } | |
728 | } | |
729 | ||
ecd182cb TH |
730 | range_alloc(asma, range, purged, pgstart, pgend, new_range); |
731 | return 0; | |
11980c2a RL |
732 | } |
733 | ||
734 | /* | |
735 | * ashmem_get_pin_status - Returns ASHMEM_IS_UNPINNED if _any_ pages in the | |
736 | * given interval are unpinned and ASHMEM_IS_PINNED otherwise. | |
737 | * | |
738 | * Caller must hold ashmem_mutex. | |
739 | */ | |
740 | static int ashmem_get_pin_status(struct ashmem_area *asma, size_t pgstart, | |
741 | size_t pgend) | |
742 | { | |
743 | struct ashmem_range *range; | |
744 | int ret = ASHMEM_IS_PINNED; | |
745 | ||
746 | list_for_each_entry(range, &asma->unpinned_list, unpinned) { | |
747 | if (range_before_page(range, pgstart)) | |
748 | break; | |
749 | if (page_range_in_range(range, pgstart, pgend)) { | |
750 | ret = ASHMEM_IS_UNPINNED; | |
751 | break; | |
752 | } | |
753 | } | |
754 | ||
755 | return ret; | |
756 | } | |
757 | ||
758 | static int ashmem_pin_unpin(struct ashmem_area *asma, unsigned long cmd, | |
759 | void __user *p) | |
760 | { | |
761 | struct ashmem_pin pin; | |
762 | size_t pgstart, pgend; | |
763 | int ret = -EINVAL; | |
ecd182cb | 764 | struct ashmem_range *range = NULL; |
11980c2a | 765 | |
59848d6a | 766 | if (copy_from_user(&pin, p, sizeof(pin))) |
740a5759 YX |
767 | return -EFAULT; |
768 | ||
ecd182cb TH |
769 | if (cmd == ASHMEM_PIN || cmd == ASHMEM_UNPIN) { |
770 | range = kmem_cache_zalloc(ashmem_range_cachep, GFP_KERNEL); | |
771 | if (!range) | |
772 | return -ENOMEM; | |
773 | } | |
774 | ||
ce8a3a9e | 775 | mutex_lock(&ashmem_mutex); |
fb4415a1 | 776 | wait_event(ashmem_shrink_wait, !atomic_read(&ashmem_shrink_inflight)); |
ce8a3a9e | 777 | |
59848d6a | 778 | if (!asma->file) |
ce8a3a9e | 779 | goto out_unlock; |
11980c2a | 780 | |
11980c2a RL |
781 | /* per custom, you can pass zero for len to mean "everything onward" */ |
782 | if (!pin.len) | |
783 | pin.len = PAGE_ALIGN(asma->size) - pin.offset; | |
784 | ||
59848d6a | 785 | if ((pin.offset | pin.len) & ~PAGE_MASK) |
ce8a3a9e | 786 | goto out_unlock; |
11980c2a | 787 | |
59848d6a | 788 | if (((__u32)-1) - pin.offset < pin.len) |
ce8a3a9e | 789 | goto out_unlock; |
11980c2a | 790 | |
59848d6a | 791 | if (PAGE_ALIGN(asma->size) < pin.offset + pin.len) |
ce8a3a9e | 792 | goto out_unlock; |
11980c2a RL |
793 | |
794 | pgstart = pin.offset / PAGE_SIZE; | |
795 | pgend = pgstart + (pin.len / PAGE_SIZE) - 1; | |
796 | ||
11980c2a RL |
797 | switch (cmd) { |
798 | case ASHMEM_PIN: | |
ecd182cb | 799 | ret = ashmem_pin(asma, pgstart, pgend, &range); |
11980c2a RL |
800 | break; |
801 | case ASHMEM_UNPIN: | |
ecd182cb | 802 | ret = ashmem_unpin(asma, pgstart, pgend, &range); |
11980c2a RL |
803 | break; |
804 | case ASHMEM_GET_PIN_STATUS: | |
805 | ret = ashmem_get_pin_status(asma, pgstart, pgend); | |
806 | break; | |
807 | } | |
808 | ||
ce8a3a9e | 809 | out_unlock: |
11980c2a | 810 | mutex_unlock(&ashmem_mutex); |
ecd182cb TH |
811 | if (range) |
812 | kmem_cache_free(ashmem_range_cachep, range); | |
11980c2a RL |
813 | |
814 | return ret; | |
815 | } | |
816 | ||
817 | static long ashmem_ioctl(struct file *file, unsigned int cmd, unsigned long arg) | |
818 | { | |
819 | struct ashmem_area *asma = file->private_data; | |
820 | long ret = -ENOTTY; | |
821 | ||
822 | switch (cmd) { | |
823 | case ASHMEM_SET_NAME: | |
1703ca90 | 824 | ret = set_name(asma, (void __user *)arg); |
11980c2a RL |
825 | break; |
826 | case ASHMEM_GET_NAME: | |
1703ca90 | 827 | ret = get_name(asma, (void __user *)arg); |
11980c2a RL |
828 | break; |
829 | case ASHMEM_SET_SIZE: | |
830 | ret = -EINVAL; | |
443064cb | 831 | mutex_lock(&ashmem_mutex); |
11980c2a RL |
832 | if (!asma->file) { |
833 | ret = 0; | |
b8d3bfa7 | 834 | asma->size = (size_t)arg; |
11980c2a | 835 | } |
443064cb | 836 | mutex_unlock(&ashmem_mutex); |
11980c2a RL |
837 | break; |
838 | case ASHMEM_GET_SIZE: | |
839 | ret = asma->size; | |
840 | break; | |
841 | case ASHMEM_SET_PROT_MASK: | |
842 | ret = set_prot_mask(asma, arg); | |
843 | break; | |
844 | case ASHMEM_GET_PROT_MASK: | |
845 | ret = asma->prot_mask; | |
846 | break; | |
847 | case ASHMEM_PIN: | |
848 | case ASHMEM_UNPIN: | |
849 | case ASHMEM_GET_PIN_STATUS: | |
1703ca90 | 850 | ret = ashmem_pin_unpin(asma, cmd, (void __user *)arg); |
11980c2a RL |
851 | break; |
852 | case ASHMEM_PURGE_ALL_CACHES: | |
853 | ret = -EPERM; | |
854 | if (capable(CAP_SYS_ADMIN)) { | |
33e8fc46 CC |
855 | struct shrink_control sc = { |
856 | .gfp_mask = GFP_KERNEL, | |
7dc19d5a | 857 | .nr_to_scan = LONG_MAX, |
33e8fc46 | 858 | }; |
59573240 | 859 | ret = ashmem_shrink_count(&ashmem_shrinker, &sc); |
7dc19d5a | 860 | ashmem_shrink_scan(&ashmem_shrinker, &sc); |
11980c2a RL |
861 | } |
862 | break; | |
863 | } | |
864 | ||
865 | return ret; | |
866 | } | |
867 | ||
e9f5b814 SC |
868 | /* support of 32bit userspace on 64bit platforms */ |
869 | #ifdef CONFIG_COMPAT | |
017ca3bf MR |
870 | static long compat_ashmem_ioctl(struct file *file, unsigned int cmd, |
871 | unsigned long arg) | |
e9f5b814 | 872 | { |
e9f5b814 SC |
873 | switch (cmd) { |
874 | case COMPAT_ASHMEM_SET_SIZE: | |
875 | cmd = ASHMEM_SET_SIZE; | |
876 | break; | |
877 | case COMPAT_ASHMEM_SET_PROT_MASK: | |
878 | cmd = ASHMEM_SET_PROT_MASK; | |
879 | break; | |
880 | } | |
881 | return ashmem_ioctl(file, cmd, arg); | |
882 | } | |
883 | #endif | |
a941f70e ZZ |
884 | #ifdef CONFIG_PROC_FS |
885 | static void ashmem_show_fdinfo(struct seq_file *m, struct file *file) | |
886 | { | |
887 | struct ashmem_area *asma = file->private_data; | |
888 | ||
889 | mutex_lock(&ashmem_mutex); | |
890 | ||
891 | if (asma->file) | |
892 | seq_printf(m, "inode:\t%ld\n", file_inode(asma->file)->i_ino); | |
e9f5b814 | 893 | |
a941f70e ZZ |
894 | if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0') |
895 | seq_printf(m, "name:\t%s\n", | |
896 | asma->name + ASHMEM_NAME_PREFIX_LEN); | |
e9f5b814 | 897 | |
a941f70e ZZ |
898 | mutex_unlock(&ashmem_mutex); |
899 | } | |
900 | #endif | |
aa5af974 | 901 | static const struct file_operations ashmem_fops = { |
11980c2a RL |
902 | .owner = THIS_MODULE, |
903 | .open = ashmem_open, | |
904 | .release = ashmem_release, | |
8a2af064 | 905 | .read_iter = ashmem_read_iter, |
1efb3439 | 906 | .llseek = ashmem_llseek, |
11980c2a RL |
907 | .mmap = ashmem_mmap, |
908 | .unlocked_ioctl = ashmem_ioctl, | |
e9f5b814 SC |
909 | #ifdef CONFIG_COMPAT |
910 | .compat_ioctl = compat_ashmem_ioctl, | |
911 | #endif | |
a941f70e ZZ |
912 | #ifdef CONFIG_PROC_FS |
913 | .show_fdinfo = ashmem_show_fdinfo, | |
914 | #endif | |
11980c2a RL |
915 | }; |
916 | ||
917 | static struct miscdevice ashmem_misc = { | |
918 | .minor = MISC_DYNAMIC_MINOR, | |
919 | .name = "ashmem", | |
920 | .fops = &ashmem_fops, | |
921 | }; | |
922 | ||
923 | static int __init ashmem_init(void) | |
924 | { | |
a2df4e33 | 925 | int ret = -ENOMEM; |
11980c2a RL |
926 | |
927 | ashmem_area_cachep = kmem_cache_create("ashmem_area_cache", | |
7048c1fc PS |
928 | sizeof(struct ashmem_area), |
929 | 0, 0, NULL); | |
59848d6a | 930 | if (!ashmem_area_cachep) { |
c810a399 | 931 | pr_err("failed to create slab cache\n"); |
a2df4e33 | 932 | goto out; |
11980c2a RL |
933 | } |
934 | ||
935 | ashmem_range_cachep = kmem_cache_create("ashmem_range_cache", | |
7048c1fc PS |
936 | sizeof(struct ashmem_range), |
937 | 0, 0, NULL); | |
59848d6a | 938 | if (!ashmem_range_cachep) { |
c810a399 | 939 | pr_err("failed to create slab cache\n"); |
a2df4e33 | 940 | goto out_free1; |
11980c2a RL |
941 | } |
942 | ||
943 | ret = misc_register(&ashmem_misc); | |
59848d6a | 944 | if (ret) { |
c810a399 | 945 | pr_err("failed to register misc device!\n"); |
a2df4e33 | 946 | goto out_free2; |
11980c2a RL |
947 | } |
948 | ||
fb02205a XS |
949 | ret = register_shrinker(&ashmem_shrinker); |
950 | if (ret) { | |
951 | pr_err("failed to register shrinker!\n"); | |
952 | goto out_demisc; | |
953 | } | |
11980c2a | 954 | |
c810a399 | 955 | pr_info("initialized\n"); |
11980c2a RL |
956 | |
957 | return 0; | |
a2df4e33 | 958 | |
fb02205a XS |
959 | out_demisc: |
960 | misc_deregister(&ashmem_misc); | |
a2df4e33 WT |
961 | out_free2: |
962 | kmem_cache_destroy(ashmem_range_cachep); | |
963 | out_free1: | |
964 | kmem_cache_destroy(ashmem_area_cachep); | |
965 | out: | |
966 | return ret; | |
11980c2a | 967 | } |
7670e053 CB |
968 | |
969 | static void __exit ashmem_exit(void) | |
970 | { | |
971 | misc_deregister(&ashmem_misc); | |
972 | unregister_shrinker(&ashmem_shrinker); | |
973 | kmem_cache_destroy(ashmem_range_cachep); | |
974 | kmem_cache_destroy(ashmem_area_cachep); | |
975 | } | |
976 | ||
977 | module_init(ashmem_init); | |
978 | module_exit(ashmem_exit); | |
979 | ||
980 | MODULE_AUTHOR("Google, Inc."); | |
981 | MODULE_DESCRIPTION("Driver for Android shared memory device"); | |
982 | MODULE_LICENSE("GPL v2"); |