]>
Commit | Line | Data |
---|---|---|
11980c2a | 1 | /* mm/ashmem.c |
2258937b CJB |
2 | * |
3 | * Anonymous Shared Memory Subsystem, ashmem | |
4 | * | |
5 | * Copyright (C) 2008 Google, Inc. | |
6 | * | |
7 | * Robert Love <rlove@google.com> | |
8 | * | |
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. | |
12 | * | |
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. | |
17 | */ | |
11980c2a | 18 | |
c810a399 SK |
19 | #define pr_fmt(fmt) "ashmem: " fmt |
20 | ||
12950595 PG |
21 | #include <linux/init.h> |
22 | #include <linux/export.h> | |
11980c2a RL |
23 | #include <linux/file.h> |
24 | #include <linux/fs.h> | |
3f31d075 | 25 | #include <linux/falloc.h> |
11980c2a RL |
26 | #include <linux/miscdevice.h> |
27 | #include <linux/security.h> | |
28 | #include <linux/mm.h> | |
29 | #include <linux/mman.h> | |
30 | #include <linux/uaccess.h> | |
31 | #include <linux/personality.h> | |
32 | #include <linux/bitops.h> | |
33 | #include <linux/mutex.h> | |
34 | #include <linux/shmem_fs.h> | |
35 | #include "ashmem.h" | |
36 | ||
37 | #define ASHMEM_NAME_PREFIX "dev/ashmem/" | |
38 | #define ASHMEM_NAME_PREFIX_LEN (sizeof(ASHMEM_NAME_PREFIX) - 1) | |
39 | #define ASHMEM_FULL_NAME_LEN (ASHMEM_NAME_LEN + ASHMEM_NAME_PREFIX_LEN) | |
40 | ||
4d2c9d5d CJB |
41 | /** |
42 | * struct ashmem_area - The anonymous shared memory area | |
43 | * @name: The optional name in /proc/pid/maps | |
44 | * @unpinned_list: The list of all ashmem areas | |
45 | * @file: The shmem-based backing file | |
46 | * @size: The size of the mapping, in bytes | |
7d92ea5d | 47 | * @prot_mask: The allowed protection bits, as vm_flags |
4d2c9d5d CJB |
48 | * |
49 | * The lifecycle of this structure is from our parent file's open() until | |
50 | * its release(). It is also protected by 'ashmem_mutex' | |
51 | * | |
52 | * Warning: Mappings do NOT pin this structure; It dies on close() | |
11980c2a RL |
53 | */ |
54 | struct ashmem_area { | |
4d2c9d5d CJB |
55 | char name[ASHMEM_FULL_NAME_LEN]; |
56 | struct list_head unpinned_list; | |
57 | struct file *file; | |
58 | size_t size; | |
59 | unsigned long prot_mask; | |
11980c2a RL |
60 | }; |
61 | ||
4d2c9d5d CJB |
62 | /** |
63 | * struct ashmem_range - A range of unpinned/evictable pages | |
64 | * @lru: The entry in the LRU list | |
65 | * @unpinned: The entry in its area's unpinned list | |
66 | * @asma: The associated anonymous shared memory area. | |
67 | * @pgstart: The starting page (inclusive) | |
68 | * @pgend: The ending page (inclusive) | |
69 | * @purged: The purge status (ASHMEM_NOT or ASHMEM_WAS_PURGED) | |
70 | * | |
71 | * The lifecycle of this structure is from unpin to pin. | |
72 | * It is protected by 'ashmem_mutex' | |
11980c2a RL |
73 | */ |
74 | struct ashmem_range { | |
4d2c9d5d CJB |
75 | struct list_head lru; |
76 | struct list_head unpinned; | |
77 | struct ashmem_area *asma; | |
78 | size_t pgstart; | |
79 | size_t pgend; | |
80 | unsigned int purged; | |
11980c2a RL |
81 | }; |
82 | ||
83 | /* LRU list of unpinned pages, protected by ashmem_mutex */ | |
84 | static LIST_HEAD(ashmem_lru_list); | |
85 | ||
801b798c | 86 | /* |
4d2c9d5d CJB |
87 | * long lru_count - The count of pages on our LRU list. |
88 | * | |
89 | * This is protected by ashmem_mutex. | |
90 | */ | |
11980c2a RL |
91 | static unsigned long lru_count; |
92 | ||
801b798c | 93 | /* |
11980c2a RL |
94 | * ashmem_mutex - protects the list of and each individual ashmem_area |
95 | * | |
96 | * Lock Ordering: ashmex_mutex -> i_mutex -> i_alloc_sem | |
97 | */ | |
98 | static DEFINE_MUTEX(ashmem_mutex); | |
99 | ||
100 | static struct kmem_cache *ashmem_area_cachep __read_mostly; | |
101 | static struct kmem_cache *ashmem_range_cachep __read_mostly; | |
102 | ||
40270ca0 GT |
103 | static inline unsigned long range_size(struct ashmem_range *range) |
104 | { | |
105 | return range->pgend - range->pgstart + 1; | |
106 | } | |
11980c2a | 107 | |
40270ca0 GT |
108 | static inline bool range_on_lru(struct ashmem_range *range) |
109 | { | |
110 | return range->purged == ASHMEM_NOT_PURGED; | |
111 | } | |
11980c2a | 112 | |
8d60b476 GT |
113 | static inline bool page_range_subsumes_range(struct ashmem_range *range, |
114 | size_t start, size_t end) | |
c0ece6c3 | 115 | { |
8d60b476 | 116 | return (range->pgstart >= start) && (range->pgend <= end); |
c0ece6c3 | 117 | } |
11980c2a | 118 | |
8d60b476 GT |
119 | static inline bool page_range_subsumed_by_range(struct ashmem_range *range, |
120 | size_t start, size_t end) | |
c0ece6c3 | 121 | { |
8d60b476 | 122 | return (range->pgstart <= start) && (range->pgend >= end); |
c0ece6c3 | 123 | } |
11980c2a | 124 | |
8d60b476 | 125 | static inline bool page_in_range(struct ashmem_range *range, size_t page) |
e2a83f32 | 126 | { |
8d60b476 | 127 | return (range->pgstart <= page) && (range->pgend >= page); |
e2a83f32 | 128 | } |
11980c2a | 129 | |
8d60b476 GT |
130 | static inline bool page_range_in_range(struct ashmem_range *range, |
131 | size_t start, size_t end) | |
c0ece6c3 | 132 | { |
8d60b476 GT |
133 | return page_in_range(range, start) || page_in_range(range, end) || |
134 | page_range_subsumes_range(range, start, end); | |
c0ece6c3 | 135 | } |
11980c2a | 136 | |
8d60b476 | 137 | static inline bool range_before_page(struct ashmem_range *range, size_t page) |
e2a83f32 | 138 | { |
8d60b476 | 139 | return range->pgend < page; |
e2a83f32 | 140 | } |
11980c2a RL |
141 | |
142 | #define PROT_MASK (PROT_EXEC | PROT_READ | PROT_WRITE) | |
143 | ||
4d2c9d5d CJB |
144 | /** |
145 | * lru_add() - Adds a range of memory to the LRU list | |
146 | * @range: The memory range being added. | |
147 | * | |
148 | * The range is first added to the end (tail) of the LRU list. | |
149 | * After this, the size of the range is added to @lru_count | |
150 | */ | |
11980c2a RL |
151 | static inline void lru_add(struct ashmem_range *range) |
152 | { | |
153 | list_add_tail(&range->lru, &ashmem_lru_list); | |
154 | lru_count += range_size(range); | |
155 | } | |
156 | ||
4d2c9d5d CJB |
157 | /** |
158 | * lru_del() - Removes a range of memory from the LRU list | |
159 | * @range: The memory range being removed | |
160 | * | |
161 | * The range is first deleted from the LRU list. | |
162 | * After this, the size of the range is removed from @lru_count | |
163 | */ | |
11980c2a RL |
164 | static inline void lru_del(struct ashmem_range *range) |
165 | { | |
166 | list_del(&range->lru); | |
167 | lru_count -= range_size(range); | |
168 | } | |
169 | ||
4d2c9d5d CJB |
170 | /** |
171 | * range_alloc() - Allocates and initializes a new ashmem_range structure | |
172 | * @asma: The associated ashmem_area | |
173 | * @prev_range: The previous ashmem_range in the sorted asma->unpinned list | |
174 | * @purged: Initial purge status (ASMEM_NOT_PURGED or ASHMEM_WAS_PURGED) | |
175 | * @start: The starting page (inclusive) | |
176 | * @end: The ending page (inclusive) | |
11980c2a | 177 | * |
4d2c9d5d | 178 | * This function is protected by ashmem_mutex. |
11980c2a | 179 | * |
4d2c9d5d | 180 | * Return: 0 if successful, or -ENOMEM if there is an error |
11980c2a RL |
181 | */ |
182 | static int range_alloc(struct ashmem_area *asma, | |
183 | struct ashmem_range *prev_range, unsigned int purged, | |
184 | size_t start, size_t end) | |
185 | { | |
186 | struct ashmem_range *range; | |
187 | ||
188 | range = kmem_cache_zalloc(ashmem_range_cachep, GFP_KERNEL); | |
189 | if (unlikely(!range)) | |
190 | return -ENOMEM; | |
191 | ||
192 | range->asma = asma; | |
193 | range->pgstart = start; | |
194 | range->pgend = end; | |
195 | range->purged = purged; | |
196 | ||
197 | list_add_tail(&range->unpinned, &prev_range->unpinned); | |
198 | ||
199 | if (range_on_lru(range)) | |
200 | lru_add(range); | |
201 | ||
202 | return 0; | |
203 | } | |
204 | ||
4d2c9d5d CJB |
205 | /** |
206 | * range_del() - Deletes and dealloctes an ashmem_range structure | |
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); |
11980c2a RL |
257 | if (unlikely(ret)) |
258 | return ret; | |
259 | ||
260 | asma = kmem_cache_zalloc(ashmem_area_cachep, GFP_KERNEL); | |
261 | if (unlikely(!asma)) | |
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; | |
332 | int ret; | |
333 | ||
334 | mutex_lock(&ashmem_mutex); | |
335 | ||
336 | if (asma->size == 0) { | |
078d8f48 JF |
337 | mutex_unlock(&ashmem_mutex); |
338 | return -EINVAL; | |
5154b93b BB |
339 | } |
340 | ||
341 | if (!asma->file) { | |
078d8f48 JF |
342 | mutex_unlock(&ashmem_mutex); |
343 | return -EBADF; | |
5154b93b BB |
344 | } |
345 | ||
078d8f48 JF |
346 | mutex_unlock(&ashmem_mutex); |
347 | ||
91360b02 | 348 | ret = vfs_llseek(asma->file, offset, origin); |
1efb3439 | 349 | if (ret < 0) |
078d8f48 | 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 | |
11980c2a RL |
364 | static int ashmem_mmap(struct file *file, struct vm_area_struct *vma) |
365 | { | |
366 | struct ashmem_area *asma = file->private_data; | |
367 | int ret = 0; | |
368 | ||
369 | mutex_lock(&ashmem_mutex); | |
370 | ||
371 | /* user needs to SET_SIZE before mapping */ | |
372 | if (unlikely(!asma->size)) { | |
373 | ret = -EINVAL; | |
374 | goto out; | |
375 | } | |
376 | ||
7d60aac2 AS |
377 | /* requested mapping size larger than object size */ |
378 | if (vma->vm_end - vma->vm_start > PAGE_ALIGN(asma->size)) { | |
379 | ret = -EINVAL; | |
380 | goto out; | |
381 | } | |
382 | ||
11980c2a | 383 | /* requested protection bits must match our allowed protection mask */ |
e6bfb709 DH |
384 | if (unlikely((vma->vm_flags & ~calc_vm_prot_bits(asma->prot_mask, 0)) & |
385 | calc_vm_prot_bits(PROT_MASK, 0))) { | |
11980c2a RL |
386 | ret = -EPERM; |
387 | goto out; | |
388 | } | |
56f76fc6 | 389 | vma->vm_flags &= ~calc_vm_may_flags(~asma->prot_mask); |
11980c2a RL |
390 | |
391 | if (!asma->file) { | |
392 | char *name = ASHMEM_NAME_DEF; | |
393 | struct file *vmfile; | |
394 | ||
395 | if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0') | |
396 | name = asma->name; | |
397 | ||
398 | /* ... and allocate the backing shmem file */ | |
399 | vmfile = shmem_file_setup(name, asma->size, vma->vm_flags); | |
7f44cb0b | 400 | if (IS_ERR(vmfile)) { |
11980c2a RL |
401 | ret = PTR_ERR(vmfile); |
402 | goto out; | |
403 | } | |
97fbfef6 | 404 | vmfile->f_mode |= FMODE_LSEEK; |
11980c2a RL |
405 | asma->file = vmfile; |
406 | } | |
407 | get_file(asma->file); | |
408 | ||
409 | /* | |
aa5af974 | 410 | * XXX - Reworked to use shmem_zero_setup() instead of |
11980c2a RL |
411 | * shmem_set_file while we're in staging. -jstultz |
412 | */ | |
413 | if (vma->vm_flags & VM_SHARED) { | |
414 | ret = shmem_zero_setup(vma); | |
415 | if (ret) { | |
416 | fput(asma->file); | |
417 | goto out; | |
418 | } | |
419 | } | |
420 | ||
421 | if (vma->vm_file) | |
422 | fput(vma->vm_file); | |
423 | vma->vm_file = asma->file; | |
11980c2a RL |
424 | |
425 | out: | |
426 | mutex_unlock(&ashmem_mutex); | |
427 | return ret; | |
428 | } | |
429 | ||
430 | /* | |
6b4f7799 | 431 | * ashmem_shrink - our cache shrinker, called from mm/vmscan.c |
11980c2a | 432 | * |
7dc19d5a | 433 | * 'nr_to_scan' is the number of objects to scan for freeing. |
11980c2a RL |
434 | * |
435 | * 'gfp_mask' is the mask of the allocation that got us into this mess. | |
436 | * | |
7dc19d5a | 437 | * Return value is the number of objects freed or -1 if we cannot |
11980c2a RL |
438 | * proceed without risk of deadlock (due to gfp_mask). |
439 | * | |
440 | * We approximate LRU via least-recently-unpinned, jettisoning unpinned partial | |
441 | * chunks of ashmem regions LRU-wise one-at-a-time until we hit 'nr_to_scan' | |
442 | * pages freed. | |
443 | */ | |
7dc19d5a DC |
444 | static unsigned long |
445 | ashmem_shrink_scan(struct shrinker *shrink, struct shrink_control *sc) | |
11980c2a RL |
446 | { |
447 | struct ashmem_range *range, *next; | |
7dc19d5a | 448 | unsigned long freed = 0; |
11980c2a RL |
449 | |
450 | /* We might recurse into filesystem code, so bail out if necessary */ | |
7dc19d5a DC |
451 | if (!(sc->gfp_mask & __GFP_FS)) |
452 | return SHRINK_STOP; | |
11980c2a | 453 | |
18e77054 LA |
454 | if (!mutex_trylock(&ashmem_mutex)) |
455 | return -1; | |
456 | ||
11980c2a | 457 | list_for_each_entry_safe(range, next, &ashmem_lru_list, lru) { |
11980c2a | 458 | loff_t start = range->pgstart * PAGE_SIZE; |
3f31d075 | 459 | loff_t end = (range->pgend + 1) * PAGE_SIZE; |
11980c2a | 460 | |
72c72bdf | 461 | vfs_fallocate(range->asma->file, |
002397c4 FH |
462 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
463 | start, end - start); | |
11980c2a RL |
464 | range->purged = ASHMEM_WAS_PURGED; |
465 | lru_del(range); | |
466 | ||
7dc19d5a DC |
467 | freed += range_size(range); |
468 | if (--sc->nr_to_scan <= 0) | |
11980c2a RL |
469 | break; |
470 | } | |
471 | mutex_unlock(&ashmem_mutex); | |
7dc19d5a DC |
472 | return freed; |
473 | } | |
11980c2a | 474 | |
7dc19d5a DC |
475 | static unsigned long |
476 | ashmem_shrink_count(struct shrinker *shrink, struct shrink_control *sc) | |
477 | { | |
478 | /* | |
479 | * note that lru_count is count of pages on the lru, not a count of | |
480 | * objects on the list. This means the scan function needs to return the | |
481 | * number of pages freed, not the number of objects scanned. | |
482 | */ | |
11980c2a RL |
483 | return lru_count; |
484 | } | |
485 | ||
486 | static struct shrinker ashmem_shrinker = { | |
7dc19d5a DC |
487 | .count_objects = ashmem_shrink_count, |
488 | .scan_objects = ashmem_shrink_scan, | |
489 | /* | |
490 | * XXX (dchinner): I wish people would comment on why they need on | |
491 | * significant changes to the default value here | |
492 | */ | |
11980c2a RL |
493 | .seeks = DEFAULT_SEEKS * 4, |
494 | }; | |
495 | ||
496 | static int set_prot_mask(struct ashmem_area *asma, unsigned long prot) | |
497 | { | |
498 | int ret = 0; | |
499 | ||
500 | mutex_lock(&ashmem_mutex); | |
501 | ||
502 | /* the user can only remove, not add, protection bits */ | |
503 | if (unlikely((asma->prot_mask & prot) != prot)) { | |
504 | ret = -EINVAL; | |
505 | goto out; | |
506 | } | |
507 | ||
508 | /* does the application expect PROT_READ to imply PROT_EXEC? */ | |
509 | if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) | |
510 | prot |= PROT_EXEC; | |
511 | ||
512 | asma->prot_mask = prot; | |
513 | ||
514 | out: | |
515 | mutex_unlock(&ashmem_mutex); | |
516 | return ret; | |
517 | } | |
518 | ||
519 | static int set_name(struct ashmem_area *asma, void __user *name) | |
520 | { | |
077f6db9 | 521 | int len; |
11980c2a | 522 | int ret = 0; |
e5834d62 | 523 | char local_name[ASHMEM_NAME_LEN]; |
11980c2a | 524 | |
e5834d62 SB |
525 | /* |
526 | * Holding the ashmem_mutex while doing a copy_from_user might cause | |
527 | * an data abort which would try to access mmap_sem. If another | |
528 | * thread has invoked ashmem_mmap then it will be holding the | |
529 | * semaphore and will be waiting for ashmem_mutex, there by leading to | |
530 | * deadlock. We'll release the mutex and take the name to a local | |
531 | * variable that does not need protection and later copy the local | |
532 | * variable to the structure member with lock held. | |
533 | */ | |
077f6db9 TP |
534 | len = strncpy_from_user(local_name, name, ASHMEM_NAME_LEN); |
535 | if (len < 0) | |
536 | return len; | |
537 | if (len == ASHMEM_NAME_LEN) | |
538 | local_name[ASHMEM_NAME_LEN - 1] = '\0'; | |
e5834d62 | 539 | mutex_lock(&ashmem_mutex); |
11980c2a | 540 | /* cannot change an existing mapping's name */ |
077f6db9 | 541 | if (unlikely(asma->file)) |
11980c2a | 542 | ret = -EINVAL; |
077f6db9 TP |
543 | else |
544 | strcpy(asma->name + ASHMEM_NAME_PREFIX_LEN, local_name); | |
11980c2a | 545 | |
077f6db9 | 546 | mutex_unlock(&ashmem_mutex); |
11980c2a RL |
547 | return ret; |
548 | } | |
549 | ||
550 | static int get_name(struct ashmem_area *asma, void __user *name) | |
551 | { | |
552 | int ret = 0; | |
e5834d62 SB |
553 | size_t len; |
554 | /* | |
555 | * Have a local variable to which we'll copy the content | |
556 | * from asma with the lock held. Later we can copy this to the user | |
557 | * space safely without holding any locks. So even if we proceed to | |
558 | * wait for mmap_sem, it won't lead to deadlock. | |
559 | */ | |
560 | char local_name[ASHMEM_NAME_LEN]; | |
11980c2a RL |
561 | |
562 | mutex_lock(&ashmem_mutex); | |
563 | if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0') { | |
11980c2a RL |
564 | /* |
565 | * Copying only `len', instead of ASHMEM_NAME_LEN, bytes | |
566 | * prevents us from revealing one user's stack to another. | |
567 | */ | |
568 | len = strlen(asma->name + ASHMEM_NAME_PREFIX_LEN) + 1; | |
e5834d62 | 569 | memcpy(local_name, asma->name + ASHMEM_NAME_PREFIX_LEN, len); |
11980c2a | 570 | } else { |
e5834d62 SB |
571 | len = sizeof(ASHMEM_NAME_DEF); |
572 | memcpy(local_name, ASHMEM_NAME_DEF, len); | |
11980c2a RL |
573 | } |
574 | mutex_unlock(&ashmem_mutex); | |
575 | ||
e5834d62 SB |
576 | /* |
577 | * Now we are just copying from the stack variable to userland | |
578 | * No lock held | |
579 | */ | |
580 | if (unlikely(copy_to_user(name, local_name, len))) | |
581 | ret = -EFAULT; | |
11980c2a RL |
582 | return ret; |
583 | } | |
584 | ||
585 | /* | |
586 | * ashmem_pin - pin the given ashmem region, returning whether it was | |
587 | * previously purged (ASHMEM_WAS_PURGED) or not (ASHMEM_NOT_PURGED). | |
588 | * | |
589 | * Caller must hold ashmem_mutex. | |
590 | */ | |
591 | static int ashmem_pin(struct ashmem_area *asma, size_t pgstart, size_t pgend) | |
592 | { | |
593 | struct ashmem_range *range, *next; | |
594 | int ret = ASHMEM_NOT_PURGED; | |
595 | ||
596 | list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) { | |
597 | /* moved past last applicable page; we can short circuit */ | |
598 | if (range_before_page(range, pgstart)) | |
599 | break; | |
600 | ||
601 | /* | |
602 | * The user can ask us to pin pages that span multiple ranges, | |
603 | * or to pin pages that aren't even unpinned, so this is messy. | |
604 | * | |
605 | * Four cases: | |
606 | * 1. The requested range subsumes an existing range, so we | |
607 | * just remove the entire matching range. | |
608 | * 2. The requested range overlaps the start of an existing | |
609 | * range, so we just update that range. | |
610 | * 3. The requested range overlaps the end of an existing | |
611 | * range, so we just update that range. | |
612 | * 4. The requested range punches a hole in an existing range, | |
613 | * so we have to update one side of the range and then | |
614 | * create a new range for the other side. | |
615 | */ | |
616 | if (page_range_in_range(range, pgstart, pgend)) { | |
617 | ret |= range->purged; | |
618 | ||
619 | /* Case #1: Easy. Just nuke the whole thing. */ | |
620 | if (page_range_subsumes_range(range, pgstart, pgend)) { | |
621 | range_del(range); | |
622 | continue; | |
623 | } | |
624 | ||
625 | /* Case #2: We overlap from the start, so adjust it */ | |
626 | if (range->pgstart >= pgstart) { | |
627 | range_shrink(range, pgend + 1, range->pgend); | |
628 | continue; | |
629 | } | |
630 | ||
631 | /* Case #3: We overlap from the rear, so adjust it */ | |
632 | if (range->pgend <= pgend) { | |
d2e4f687 PS |
633 | range_shrink(range, range->pgstart, |
634 | pgstart - 1); | |
11980c2a RL |
635 | continue; |
636 | } | |
637 | ||
638 | /* | |
639 | * Case #4: We eat a chunk out of the middle. A bit | |
640 | * more complicated, we allocate a new range for the | |
641 | * second half and adjust the first chunk's endpoint. | |
642 | */ | |
643 | range_alloc(asma, range, range->purged, | |
644 | pgend + 1, range->pgend); | |
645 | range_shrink(range, range->pgstart, pgstart - 1); | |
646 | break; | |
647 | } | |
648 | } | |
649 | ||
650 | return ret; | |
651 | } | |
652 | ||
653 | /* | |
654 | * ashmem_unpin - unpin the given range of pages. Returns zero on success. | |
655 | * | |
656 | * Caller must hold ashmem_mutex. | |
657 | */ | |
658 | static int ashmem_unpin(struct ashmem_area *asma, size_t pgstart, size_t pgend) | |
659 | { | |
660 | struct ashmem_range *range, *next; | |
661 | unsigned int purged = ASHMEM_NOT_PURGED; | |
662 | ||
663 | restart: | |
664 | list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) { | |
665 | /* short circuit: this is our insertion point */ | |
666 | if (range_before_page(range, pgstart)) | |
667 | break; | |
668 | ||
669 | /* | |
670 | * The user can ask us to unpin pages that are already entirely | |
671 | * or partially pinned. We handle those two cases here. | |
672 | */ | |
673 | if (page_range_subsumed_by_range(range, pgstart, pgend)) | |
674 | return 0; | |
675 | if (page_range_in_range(range, pgstart, pgend)) { | |
9f1c427d AKC |
676 | pgstart = min(range->pgstart, pgstart); |
677 | pgend = max(range->pgend, pgend); | |
11980c2a RL |
678 | purged |= range->purged; |
679 | range_del(range); | |
680 | goto restart; | |
681 | } | |
682 | } | |
683 | ||
684 | return range_alloc(asma, range, purged, pgstart, pgend); | |
685 | } | |
686 | ||
687 | /* | |
688 | * ashmem_get_pin_status - Returns ASHMEM_IS_UNPINNED if _any_ pages in the | |
689 | * given interval are unpinned and ASHMEM_IS_PINNED otherwise. | |
690 | * | |
691 | * Caller must hold ashmem_mutex. | |
692 | */ | |
693 | static int ashmem_get_pin_status(struct ashmem_area *asma, size_t pgstart, | |
694 | size_t pgend) | |
695 | { | |
696 | struct ashmem_range *range; | |
697 | int ret = ASHMEM_IS_PINNED; | |
698 | ||
699 | list_for_each_entry(range, &asma->unpinned_list, unpinned) { | |
700 | if (range_before_page(range, pgstart)) | |
701 | break; | |
702 | if (page_range_in_range(range, pgstart, pgend)) { | |
703 | ret = ASHMEM_IS_UNPINNED; | |
704 | break; | |
705 | } | |
706 | } | |
707 | ||
708 | return ret; | |
709 | } | |
710 | ||
711 | static int ashmem_pin_unpin(struct ashmem_area *asma, unsigned long cmd, | |
712 | void __user *p) | |
713 | { | |
714 | struct ashmem_pin pin; | |
715 | size_t pgstart, pgend; | |
716 | int ret = -EINVAL; | |
717 | ||
27fe2fd1 YX |
718 | if (unlikely(copy_from_user(&pin, p, sizeof(pin)))) |
719 | return -EFAULT; | |
720 | ||
421f4e7e BH |
721 | mutex_lock(&ashmem_mutex); |
722 | ||
11980c2a | 723 | if (unlikely(!asma->file)) |
421f4e7e | 724 | goto out_unlock; |
11980c2a | 725 | |
11980c2a RL |
726 | /* per custom, you can pass zero for len to mean "everything onward" */ |
727 | if (!pin.len) | |
728 | pin.len = PAGE_ALIGN(asma->size) - pin.offset; | |
729 | ||
730 | if (unlikely((pin.offset | pin.len) & ~PAGE_MASK)) | |
421f4e7e | 731 | goto out_unlock; |
11980c2a | 732 | |
b8d3bfa7 | 733 | if (unlikely(((__u32)-1) - pin.offset < pin.len)) |
421f4e7e | 734 | goto out_unlock; |
11980c2a RL |
735 | |
736 | if (unlikely(PAGE_ALIGN(asma->size) < pin.offset + pin.len)) | |
421f4e7e | 737 | goto out_unlock; |
11980c2a RL |
738 | |
739 | pgstart = pin.offset / PAGE_SIZE; | |
740 | pgend = pgstart + (pin.len / PAGE_SIZE) - 1; | |
741 | ||
11980c2a RL |
742 | switch (cmd) { |
743 | case ASHMEM_PIN: | |
744 | ret = ashmem_pin(asma, pgstart, pgend); | |
745 | break; | |
746 | case ASHMEM_UNPIN: | |
747 | ret = ashmem_unpin(asma, pgstart, pgend); | |
748 | break; | |
749 | case ASHMEM_GET_PIN_STATUS: | |
750 | ret = ashmem_get_pin_status(asma, pgstart, pgend); | |
751 | break; | |
752 | } | |
753 | ||
421f4e7e | 754 | out_unlock: |
11980c2a RL |
755 | mutex_unlock(&ashmem_mutex); |
756 | ||
757 | return ret; | |
758 | } | |
759 | ||
760 | static long ashmem_ioctl(struct file *file, unsigned int cmd, unsigned long arg) | |
761 | { | |
762 | struct ashmem_area *asma = file->private_data; | |
763 | long ret = -ENOTTY; | |
764 | ||
765 | switch (cmd) { | |
766 | case ASHMEM_SET_NAME: | |
1703ca90 | 767 | ret = set_name(asma, (void __user *)arg); |
11980c2a RL |
768 | break; |
769 | case ASHMEM_GET_NAME: | |
1703ca90 | 770 | ret = get_name(asma, (void __user *)arg); |
11980c2a RL |
771 | break; |
772 | case ASHMEM_SET_SIZE: | |
773 | ret = -EINVAL; | |
443064cb | 774 | mutex_lock(&ashmem_mutex); |
11980c2a RL |
775 | if (!asma->file) { |
776 | ret = 0; | |
b8d3bfa7 | 777 | asma->size = (size_t)arg; |
11980c2a | 778 | } |
443064cb | 779 | mutex_unlock(&ashmem_mutex); |
11980c2a RL |
780 | break; |
781 | case ASHMEM_GET_SIZE: | |
782 | ret = asma->size; | |
783 | break; | |
784 | case ASHMEM_SET_PROT_MASK: | |
785 | ret = set_prot_mask(asma, arg); | |
786 | break; | |
787 | case ASHMEM_GET_PROT_MASK: | |
788 | ret = asma->prot_mask; | |
789 | break; | |
790 | case ASHMEM_PIN: | |
791 | case ASHMEM_UNPIN: | |
792 | case ASHMEM_GET_PIN_STATUS: | |
1703ca90 | 793 | ret = ashmem_pin_unpin(asma, cmd, (void __user *)arg); |
11980c2a RL |
794 | break; |
795 | case ASHMEM_PURGE_ALL_CACHES: | |
796 | ret = -EPERM; | |
797 | if (capable(CAP_SYS_ADMIN)) { | |
33e8fc46 CC |
798 | struct shrink_control sc = { |
799 | .gfp_mask = GFP_KERNEL, | |
7dc19d5a | 800 | .nr_to_scan = LONG_MAX, |
33e8fc46 | 801 | }; |
59573240 | 802 | ret = ashmem_shrink_count(&ashmem_shrinker, &sc); |
7dc19d5a | 803 | ashmem_shrink_scan(&ashmem_shrinker, &sc); |
11980c2a RL |
804 | } |
805 | break; | |
806 | } | |
807 | ||
808 | return ret; | |
809 | } | |
810 | ||
e9f5b814 SC |
811 | /* support of 32bit userspace on 64bit platforms */ |
812 | #ifdef CONFIG_COMPAT | |
017ca3bf MR |
813 | static long compat_ashmem_ioctl(struct file *file, unsigned int cmd, |
814 | unsigned long arg) | |
e9f5b814 | 815 | { |
e9f5b814 SC |
816 | switch (cmd) { |
817 | case COMPAT_ASHMEM_SET_SIZE: | |
818 | cmd = ASHMEM_SET_SIZE; | |
819 | break; | |
820 | case COMPAT_ASHMEM_SET_PROT_MASK: | |
821 | cmd = ASHMEM_SET_PROT_MASK; | |
822 | break; | |
823 | } | |
824 | return ashmem_ioctl(file, cmd, arg); | |
825 | } | |
826 | #endif | |
827 | ||
aa5af974 | 828 | static const struct file_operations ashmem_fops = { |
11980c2a RL |
829 | .owner = THIS_MODULE, |
830 | .open = ashmem_open, | |
831 | .release = ashmem_release, | |
8a2af064 | 832 | .read_iter = ashmem_read_iter, |
1efb3439 | 833 | .llseek = ashmem_llseek, |
11980c2a RL |
834 | .mmap = ashmem_mmap, |
835 | .unlocked_ioctl = ashmem_ioctl, | |
e9f5b814 SC |
836 | #ifdef CONFIG_COMPAT |
837 | .compat_ioctl = compat_ashmem_ioctl, | |
838 | #endif | |
11980c2a RL |
839 | }; |
840 | ||
841 | static struct miscdevice ashmem_misc = { | |
842 | .minor = MISC_DYNAMIC_MINOR, | |
843 | .name = "ashmem", | |
844 | .fops = &ashmem_fops, | |
845 | }; | |
846 | ||
847 | static int __init ashmem_init(void) | |
848 | { | |
a2df4e33 | 849 | int ret = -ENOMEM; |
11980c2a RL |
850 | |
851 | ashmem_area_cachep = kmem_cache_create("ashmem_area_cache", | |
7048c1fc PS |
852 | sizeof(struct ashmem_area), |
853 | 0, 0, NULL); | |
11980c2a | 854 | if (unlikely(!ashmem_area_cachep)) { |
c810a399 | 855 | pr_err("failed to create slab cache\n"); |
a2df4e33 | 856 | goto out; |
11980c2a RL |
857 | } |
858 | ||
859 | ashmem_range_cachep = kmem_cache_create("ashmem_range_cache", | |
7048c1fc PS |
860 | sizeof(struct ashmem_range), |
861 | 0, 0, NULL); | |
11980c2a | 862 | if (unlikely(!ashmem_range_cachep)) { |
c810a399 | 863 | pr_err("failed to create slab cache\n"); |
a2df4e33 | 864 | goto out_free1; |
11980c2a RL |
865 | } |
866 | ||
867 | ret = misc_register(&ashmem_misc); | |
868 | if (unlikely(ret)) { | |
c810a399 | 869 | pr_err("failed to register misc device!\n"); |
a2df4e33 | 870 | goto out_free2; |
11980c2a RL |
871 | } |
872 | ||
873 | register_shrinker(&ashmem_shrinker); | |
874 | ||
c810a399 | 875 | pr_info("initialized\n"); |
11980c2a RL |
876 | |
877 | return 0; | |
a2df4e33 WT |
878 | |
879 | out_free2: | |
880 | kmem_cache_destroy(ashmem_range_cachep); | |
881 | out_free1: | |
882 | kmem_cache_destroy(ashmem_area_cachep); | |
883 | out: | |
884 | return ret; | |
11980c2a | 885 | } |
12950595 | 886 | device_initcall(ashmem_init); |