]>
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) { | |
337 | ret = -EINVAL; | |
338 | goto out; | |
339 | } | |
340 | ||
341 | if (!asma->file) { | |
342 | ret = -EBADF; | |
343 | goto out; | |
344 | } | |
345 | ||
91360b02 | 346 | ret = vfs_llseek(asma->file, offset, origin); |
1efb3439 | 347 | if (ret < 0) |
5154b93b | 348 | goto out; |
5154b93b BB |
349 | |
350 | /** Copy f_pos from backing file, since f_ops->llseek() sets it */ | |
351 | file->f_pos = asma->file->f_pos; | |
853ca7ae BB |
352 | |
353 | out: | |
354 | mutex_unlock(&ashmem_mutex); | |
355 | return ret; | |
356 | } | |
357 | ||
7cfce77d | 358 | static inline vm_flags_t calc_vm_may_flags(unsigned long prot) |
56f76fc6 | 359 | { |
1efb3439 | 360 | return _calc_vm_trans(prot, PROT_READ, VM_MAYREAD) | |
56f76fc6 AH |
361 | _calc_vm_trans(prot, PROT_WRITE, VM_MAYWRITE) | |
362 | _calc_vm_trans(prot, PROT_EXEC, VM_MAYEXEC); | |
363 | } | |
853ca7ae | 364 | |
11980c2a RL |
365 | static int ashmem_mmap(struct file *file, struct vm_area_struct *vma) |
366 | { | |
367 | struct ashmem_area *asma = file->private_data; | |
368 | int ret = 0; | |
369 | ||
370 | mutex_lock(&ashmem_mutex); | |
371 | ||
372 | /* user needs to SET_SIZE before mapping */ | |
373 | if (unlikely(!asma->size)) { | |
374 | ret = -EINVAL; | |
375 | goto out; | |
376 | } | |
377 | ||
378 | /* requested protection bits must match our allowed protection mask */ | |
e6bfb709 DH |
379 | if (unlikely((vma->vm_flags & ~calc_vm_prot_bits(asma->prot_mask, 0)) & |
380 | calc_vm_prot_bits(PROT_MASK, 0))) { | |
11980c2a RL |
381 | ret = -EPERM; |
382 | goto out; | |
383 | } | |
56f76fc6 | 384 | vma->vm_flags &= ~calc_vm_may_flags(~asma->prot_mask); |
11980c2a RL |
385 | |
386 | if (!asma->file) { | |
387 | char *name = ASHMEM_NAME_DEF; | |
388 | struct file *vmfile; | |
389 | ||
390 | if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0') | |
391 | name = asma->name; | |
392 | ||
393 | /* ... and allocate the backing shmem file */ | |
394 | vmfile = shmem_file_setup(name, asma->size, vma->vm_flags); | |
7f44cb0b | 395 | if (IS_ERR(vmfile)) { |
11980c2a RL |
396 | ret = PTR_ERR(vmfile); |
397 | goto out; | |
398 | } | |
97fbfef6 | 399 | vmfile->f_mode |= FMODE_LSEEK; |
11980c2a RL |
400 | asma->file = vmfile; |
401 | } | |
402 | get_file(asma->file); | |
403 | ||
404 | /* | |
aa5af974 | 405 | * XXX - Reworked to use shmem_zero_setup() instead of |
11980c2a RL |
406 | * shmem_set_file while we're in staging. -jstultz |
407 | */ | |
408 | if (vma->vm_flags & VM_SHARED) { | |
409 | ret = shmem_zero_setup(vma); | |
410 | if (ret) { | |
411 | fput(asma->file); | |
412 | goto out; | |
413 | } | |
414 | } | |
415 | ||
416 | if (vma->vm_file) | |
417 | fput(vma->vm_file); | |
418 | vma->vm_file = asma->file; | |
11980c2a RL |
419 | |
420 | out: | |
421 | mutex_unlock(&ashmem_mutex); | |
422 | return ret; | |
423 | } | |
424 | ||
425 | /* | |
6b4f7799 | 426 | * ashmem_shrink - our cache shrinker, called from mm/vmscan.c |
11980c2a | 427 | * |
7dc19d5a | 428 | * 'nr_to_scan' is the number of objects to scan for freeing. |
11980c2a RL |
429 | * |
430 | * 'gfp_mask' is the mask of the allocation that got us into this mess. | |
431 | * | |
7dc19d5a | 432 | * Return value is the number of objects freed or -1 if we cannot |
11980c2a RL |
433 | * proceed without risk of deadlock (due to gfp_mask). |
434 | * | |
435 | * We approximate LRU via least-recently-unpinned, jettisoning unpinned partial | |
436 | * chunks of ashmem regions LRU-wise one-at-a-time until we hit 'nr_to_scan' | |
437 | * pages freed. | |
438 | */ | |
7dc19d5a DC |
439 | static unsigned long |
440 | ashmem_shrink_scan(struct shrinker *shrink, struct shrink_control *sc) | |
11980c2a RL |
441 | { |
442 | struct ashmem_range *range, *next; | |
7dc19d5a | 443 | unsigned long freed = 0; |
11980c2a RL |
444 | |
445 | /* We might recurse into filesystem code, so bail out if necessary */ | |
7dc19d5a DC |
446 | if (!(sc->gfp_mask & __GFP_FS)) |
447 | return SHRINK_STOP; | |
11980c2a | 448 | |
18e77054 LA |
449 | if (!mutex_trylock(&ashmem_mutex)) |
450 | return -1; | |
451 | ||
11980c2a | 452 | list_for_each_entry_safe(range, next, &ashmem_lru_list, lru) { |
11980c2a | 453 | loff_t start = range->pgstart * PAGE_SIZE; |
3f31d075 | 454 | loff_t end = (range->pgend + 1) * PAGE_SIZE; |
11980c2a | 455 | |
72c72bdf | 456 | vfs_fallocate(range->asma->file, |
002397c4 FH |
457 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
458 | start, end - start); | |
11980c2a RL |
459 | range->purged = ASHMEM_WAS_PURGED; |
460 | lru_del(range); | |
461 | ||
7dc19d5a DC |
462 | freed += range_size(range); |
463 | if (--sc->nr_to_scan <= 0) | |
11980c2a RL |
464 | break; |
465 | } | |
466 | mutex_unlock(&ashmem_mutex); | |
7dc19d5a DC |
467 | return freed; |
468 | } | |
11980c2a | 469 | |
7dc19d5a DC |
470 | static unsigned long |
471 | ashmem_shrink_count(struct shrinker *shrink, struct shrink_control *sc) | |
472 | { | |
473 | /* | |
474 | * note that lru_count is count of pages on the lru, not a count of | |
475 | * objects on the list. This means the scan function needs to return the | |
476 | * number of pages freed, not the number of objects scanned. | |
477 | */ | |
11980c2a RL |
478 | return lru_count; |
479 | } | |
480 | ||
481 | static struct shrinker ashmem_shrinker = { | |
7dc19d5a DC |
482 | .count_objects = ashmem_shrink_count, |
483 | .scan_objects = ashmem_shrink_scan, | |
484 | /* | |
485 | * XXX (dchinner): I wish people would comment on why they need on | |
486 | * significant changes to the default value here | |
487 | */ | |
11980c2a RL |
488 | .seeks = DEFAULT_SEEKS * 4, |
489 | }; | |
490 | ||
491 | static int set_prot_mask(struct ashmem_area *asma, unsigned long prot) | |
492 | { | |
493 | int ret = 0; | |
494 | ||
495 | mutex_lock(&ashmem_mutex); | |
496 | ||
497 | /* the user can only remove, not add, protection bits */ | |
498 | if (unlikely((asma->prot_mask & prot) != prot)) { | |
499 | ret = -EINVAL; | |
500 | goto out; | |
501 | } | |
502 | ||
503 | /* does the application expect PROT_READ to imply PROT_EXEC? */ | |
504 | if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) | |
505 | prot |= PROT_EXEC; | |
506 | ||
507 | asma->prot_mask = prot; | |
508 | ||
509 | out: | |
510 | mutex_unlock(&ashmem_mutex); | |
511 | return ret; | |
512 | } | |
513 | ||
514 | static int set_name(struct ashmem_area *asma, void __user *name) | |
515 | { | |
077f6db9 | 516 | int len; |
11980c2a | 517 | int ret = 0; |
e5834d62 | 518 | char local_name[ASHMEM_NAME_LEN]; |
11980c2a | 519 | |
e5834d62 SB |
520 | /* |
521 | * Holding the ashmem_mutex while doing a copy_from_user might cause | |
522 | * an data abort which would try to access mmap_sem. If another | |
523 | * thread has invoked ashmem_mmap then it will be holding the | |
524 | * semaphore and will be waiting for ashmem_mutex, there by leading to | |
525 | * deadlock. We'll release the mutex and take the name to a local | |
526 | * variable that does not need protection and later copy the local | |
527 | * variable to the structure member with lock held. | |
528 | */ | |
077f6db9 TP |
529 | len = strncpy_from_user(local_name, name, ASHMEM_NAME_LEN); |
530 | if (len < 0) | |
531 | return len; | |
532 | if (len == ASHMEM_NAME_LEN) | |
533 | local_name[ASHMEM_NAME_LEN - 1] = '\0'; | |
e5834d62 | 534 | mutex_lock(&ashmem_mutex); |
11980c2a | 535 | /* cannot change an existing mapping's name */ |
077f6db9 | 536 | if (unlikely(asma->file)) |
11980c2a | 537 | ret = -EINVAL; |
077f6db9 TP |
538 | else |
539 | strcpy(asma->name + ASHMEM_NAME_PREFIX_LEN, local_name); | |
11980c2a | 540 | |
077f6db9 | 541 | mutex_unlock(&ashmem_mutex); |
11980c2a RL |
542 | return ret; |
543 | } | |
544 | ||
545 | static int get_name(struct ashmem_area *asma, void __user *name) | |
546 | { | |
547 | int ret = 0; | |
e5834d62 SB |
548 | size_t len; |
549 | /* | |
550 | * Have a local variable to which we'll copy the content | |
551 | * from asma with the lock held. Later we can copy this to the user | |
552 | * space safely without holding any locks. So even if we proceed to | |
553 | * wait for mmap_sem, it won't lead to deadlock. | |
554 | */ | |
555 | char local_name[ASHMEM_NAME_LEN]; | |
11980c2a RL |
556 | |
557 | mutex_lock(&ashmem_mutex); | |
558 | if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0') { | |
11980c2a RL |
559 | /* |
560 | * Copying only `len', instead of ASHMEM_NAME_LEN, bytes | |
561 | * prevents us from revealing one user's stack to another. | |
562 | */ | |
563 | len = strlen(asma->name + ASHMEM_NAME_PREFIX_LEN) + 1; | |
e5834d62 | 564 | memcpy(local_name, asma->name + ASHMEM_NAME_PREFIX_LEN, len); |
11980c2a | 565 | } else { |
e5834d62 SB |
566 | len = sizeof(ASHMEM_NAME_DEF); |
567 | memcpy(local_name, ASHMEM_NAME_DEF, len); | |
11980c2a RL |
568 | } |
569 | mutex_unlock(&ashmem_mutex); | |
570 | ||
e5834d62 SB |
571 | /* |
572 | * Now we are just copying from the stack variable to userland | |
573 | * No lock held | |
574 | */ | |
575 | if (unlikely(copy_to_user(name, local_name, len))) | |
576 | ret = -EFAULT; | |
11980c2a RL |
577 | return ret; |
578 | } | |
579 | ||
580 | /* | |
581 | * ashmem_pin - pin the given ashmem region, returning whether it was | |
582 | * previously purged (ASHMEM_WAS_PURGED) or not (ASHMEM_NOT_PURGED). | |
583 | * | |
584 | * Caller must hold ashmem_mutex. | |
585 | */ | |
586 | static int ashmem_pin(struct ashmem_area *asma, size_t pgstart, size_t pgend) | |
587 | { | |
588 | struct ashmem_range *range, *next; | |
589 | int ret = ASHMEM_NOT_PURGED; | |
590 | ||
591 | list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) { | |
592 | /* moved past last applicable page; we can short circuit */ | |
593 | if (range_before_page(range, pgstart)) | |
594 | break; | |
595 | ||
596 | /* | |
597 | * The user can ask us to pin pages that span multiple ranges, | |
598 | * or to pin pages that aren't even unpinned, so this is messy. | |
599 | * | |
600 | * Four cases: | |
601 | * 1. The requested range subsumes an existing range, so we | |
602 | * just remove the entire matching range. | |
603 | * 2. The requested range overlaps the start of an existing | |
604 | * range, so we just update that range. | |
605 | * 3. The requested range overlaps the end of an existing | |
606 | * range, so we just update that range. | |
607 | * 4. The requested range punches a hole in an existing range, | |
608 | * so we have to update one side of the range and then | |
609 | * create a new range for the other side. | |
610 | */ | |
611 | if (page_range_in_range(range, pgstart, pgend)) { | |
612 | ret |= range->purged; | |
613 | ||
614 | /* Case #1: Easy. Just nuke the whole thing. */ | |
615 | if (page_range_subsumes_range(range, pgstart, pgend)) { | |
616 | range_del(range); | |
617 | continue; | |
618 | } | |
619 | ||
620 | /* Case #2: We overlap from the start, so adjust it */ | |
621 | if (range->pgstart >= pgstart) { | |
622 | range_shrink(range, pgend + 1, range->pgend); | |
623 | continue; | |
624 | } | |
625 | ||
626 | /* Case #3: We overlap from the rear, so adjust it */ | |
627 | if (range->pgend <= pgend) { | |
d2e4f687 PS |
628 | range_shrink(range, range->pgstart, |
629 | pgstart - 1); | |
11980c2a RL |
630 | continue; |
631 | } | |
632 | ||
633 | /* | |
634 | * Case #4: We eat a chunk out of the middle. A bit | |
635 | * more complicated, we allocate a new range for the | |
636 | * second half and adjust the first chunk's endpoint. | |
637 | */ | |
638 | range_alloc(asma, range, range->purged, | |
639 | pgend + 1, range->pgend); | |
640 | range_shrink(range, range->pgstart, pgstart - 1); | |
641 | break; | |
642 | } | |
643 | } | |
644 | ||
645 | return ret; | |
646 | } | |
647 | ||
648 | /* | |
649 | * ashmem_unpin - unpin the given range of pages. Returns zero on success. | |
650 | * | |
651 | * Caller must hold ashmem_mutex. | |
652 | */ | |
653 | static int ashmem_unpin(struct ashmem_area *asma, size_t pgstart, size_t pgend) | |
654 | { | |
655 | struct ashmem_range *range, *next; | |
656 | unsigned int purged = ASHMEM_NOT_PURGED; | |
657 | ||
658 | restart: | |
659 | list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) { | |
660 | /* short circuit: this is our insertion point */ | |
661 | if (range_before_page(range, pgstart)) | |
662 | break; | |
663 | ||
664 | /* | |
665 | * The user can ask us to unpin pages that are already entirely | |
666 | * or partially pinned. We handle those two cases here. | |
667 | */ | |
668 | if (page_range_subsumed_by_range(range, pgstart, pgend)) | |
669 | return 0; | |
670 | if (page_range_in_range(range, pgstart, pgend)) { | |
9f1c427d AKC |
671 | pgstart = min(range->pgstart, pgstart); |
672 | pgend = max(range->pgend, pgend); | |
11980c2a RL |
673 | purged |= range->purged; |
674 | range_del(range); | |
675 | goto restart; | |
676 | } | |
677 | } | |
678 | ||
679 | return range_alloc(asma, range, purged, pgstart, pgend); | |
680 | } | |
681 | ||
682 | /* | |
683 | * ashmem_get_pin_status - Returns ASHMEM_IS_UNPINNED if _any_ pages in the | |
684 | * given interval are unpinned and ASHMEM_IS_PINNED otherwise. | |
685 | * | |
686 | * Caller must hold ashmem_mutex. | |
687 | */ | |
688 | static int ashmem_get_pin_status(struct ashmem_area *asma, size_t pgstart, | |
689 | size_t pgend) | |
690 | { | |
691 | struct ashmem_range *range; | |
692 | int ret = ASHMEM_IS_PINNED; | |
693 | ||
694 | list_for_each_entry(range, &asma->unpinned_list, unpinned) { | |
695 | if (range_before_page(range, pgstart)) | |
696 | break; | |
697 | if (page_range_in_range(range, pgstart, pgend)) { | |
698 | ret = ASHMEM_IS_UNPINNED; | |
699 | break; | |
700 | } | |
701 | } | |
702 | ||
703 | return ret; | |
704 | } | |
705 | ||
706 | static int ashmem_pin_unpin(struct ashmem_area *asma, unsigned long cmd, | |
707 | void __user *p) | |
708 | { | |
709 | struct ashmem_pin pin; | |
710 | size_t pgstart, pgend; | |
711 | int ret = -EINVAL; | |
712 | ||
713 | if (unlikely(!asma->file)) | |
714 | return -EINVAL; | |
715 | ||
716 | if (unlikely(copy_from_user(&pin, p, sizeof(pin)))) | |
717 | return -EFAULT; | |
718 | ||
719 | /* per custom, you can pass zero for len to mean "everything onward" */ | |
720 | if (!pin.len) | |
721 | pin.len = PAGE_ALIGN(asma->size) - pin.offset; | |
722 | ||
723 | if (unlikely((pin.offset | pin.len) & ~PAGE_MASK)) | |
724 | return -EINVAL; | |
725 | ||
b8d3bfa7 | 726 | if (unlikely(((__u32)-1) - pin.offset < pin.len)) |
11980c2a RL |
727 | return -EINVAL; |
728 | ||
729 | if (unlikely(PAGE_ALIGN(asma->size) < pin.offset + pin.len)) | |
730 | return -EINVAL; | |
731 | ||
732 | pgstart = pin.offset / PAGE_SIZE; | |
733 | pgend = pgstart + (pin.len / PAGE_SIZE) - 1; | |
734 | ||
735 | mutex_lock(&ashmem_mutex); | |
736 | ||
737 | switch (cmd) { | |
738 | case ASHMEM_PIN: | |
739 | ret = ashmem_pin(asma, pgstart, pgend); | |
740 | break; | |
741 | case ASHMEM_UNPIN: | |
742 | ret = ashmem_unpin(asma, pgstart, pgend); | |
743 | break; | |
744 | case ASHMEM_GET_PIN_STATUS: | |
745 | ret = ashmem_get_pin_status(asma, pgstart, pgend); | |
746 | break; | |
747 | } | |
748 | ||
749 | mutex_unlock(&ashmem_mutex); | |
750 | ||
751 | return ret; | |
752 | } | |
753 | ||
754 | static long ashmem_ioctl(struct file *file, unsigned int cmd, unsigned long arg) | |
755 | { | |
756 | struct ashmem_area *asma = file->private_data; | |
757 | long ret = -ENOTTY; | |
758 | ||
759 | switch (cmd) { | |
760 | case ASHMEM_SET_NAME: | |
1703ca90 | 761 | ret = set_name(asma, (void __user *)arg); |
11980c2a RL |
762 | break; |
763 | case ASHMEM_GET_NAME: | |
1703ca90 | 764 | ret = get_name(asma, (void __user *)arg); |
11980c2a RL |
765 | break; |
766 | case ASHMEM_SET_SIZE: | |
767 | ret = -EINVAL; | |
443064cb | 768 | mutex_lock(&ashmem_mutex); |
11980c2a RL |
769 | if (!asma->file) { |
770 | ret = 0; | |
b8d3bfa7 | 771 | asma->size = (size_t)arg; |
11980c2a | 772 | } |
443064cb | 773 | mutex_unlock(&ashmem_mutex); |
11980c2a RL |
774 | break; |
775 | case ASHMEM_GET_SIZE: | |
776 | ret = asma->size; | |
777 | break; | |
778 | case ASHMEM_SET_PROT_MASK: | |
779 | ret = set_prot_mask(asma, arg); | |
780 | break; | |
781 | case ASHMEM_GET_PROT_MASK: | |
782 | ret = asma->prot_mask; | |
783 | break; | |
784 | case ASHMEM_PIN: | |
785 | case ASHMEM_UNPIN: | |
786 | case ASHMEM_GET_PIN_STATUS: | |
1703ca90 | 787 | ret = ashmem_pin_unpin(asma, cmd, (void __user *)arg); |
11980c2a RL |
788 | break; |
789 | case ASHMEM_PURGE_ALL_CACHES: | |
790 | ret = -EPERM; | |
791 | if (capable(CAP_SYS_ADMIN)) { | |
33e8fc46 CC |
792 | struct shrink_control sc = { |
793 | .gfp_mask = GFP_KERNEL, | |
7dc19d5a | 794 | .nr_to_scan = LONG_MAX, |
33e8fc46 | 795 | }; |
59573240 | 796 | ret = ashmem_shrink_count(&ashmem_shrinker, &sc); |
7dc19d5a | 797 | ashmem_shrink_scan(&ashmem_shrinker, &sc); |
11980c2a RL |
798 | } |
799 | break; | |
800 | } | |
801 | ||
802 | return ret; | |
803 | } | |
804 | ||
e9f5b814 SC |
805 | /* support of 32bit userspace on 64bit platforms */ |
806 | #ifdef CONFIG_COMPAT | |
017ca3bf MR |
807 | static long compat_ashmem_ioctl(struct file *file, unsigned int cmd, |
808 | unsigned long arg) | |
e9f5b814 | 809 | { |
e9f5b814 SC |
810 | switch (cmd) { |
811 | case COMPAT_ASHMEM_SET_SIZE: | |
812 | cmd = ASHMEM_SET_SIZE; | |
813 | break; | |
814 | case COMPAT_ASHMEM_SET_PROT_MASK: | |
815 | cmd = ASHMEM_SET_PROT_MASK; | |
816 | break; | |
817 | } | |
818 | return ashmem_ioctl(file, cmd, arg); | |
819 | } | |
820 | #endif | |
821 | ||
aa5af974 | 822 | static const struct file_operations ashmem_fops = { |
11980c2a RL |
823 | .owner = THIS_MODULE, |
824 | .open = ashmem_open, | |
825 | .release = ashmem_release, | |
8a2af064 | 826 | .read_iter = ashmem_read_iter, |
1efb3439 | 827 | .llseek = ashmem_llseek, |
11980c2a RL |
828 | .mmap = ashmem_mmap, |
829 | .unlocked_ioctl = ashmem_ioctl, | |
e9f5b814 SC |
830 | #ifdef CONFIG_COMPAT |
831 | .compat_ioctl = compat_ashmem_ioctl, | |
832 | #endif | |
11980c2a RL |
833 | }; |
834 | ||
835 | static struct miscdevice ashmem_misc = { | |
836 | .minor = MISC_DYNAMIC_MINOR, | |
837 | .name = "ashmem", | |
838 | .fops = &ashmem_fops, | |
839 | }; | |
840 | ||
841 | static int __init ashmem_init(void) | |
842 | { | |
a2df4e33 | 843 | int ret = -ENOMEM; |
11980c2a RL |
844 | |
845 | ashmem_area_cachep = kmem_cache_create("ashmem_area_cache", | |
7048c1fc PS |
846 | sizeof(struct ashmem_area), |
847 | 0, 0, NULL); | |
11980c2a | 848 | if (unlikely(!ashmem_area_cachep)) { |
c810a399 | 849 | pr_err("failed to create slab cache\n"); |
a2df4e33 | 850 | goto out; |
11980c2a RL |
851 | } |
852 | ||
853 | ashmem_range_cachep = kmem_cache_create("ashmem_range_cache", | |
7048c1fc PS |
854 | sizeof(struct ashmem_range), |
855 | 0, 0, NULL); | |
11980c2a | 856 | if (unlikely(!ashmem_range_cachep)) { |
c810a399 | 857 | pr_err("failed to create slab cache\n"); |
a2df4e33 | 858 | goto out_free1; |
11980c2a RL |
859 | } |
860 | ||
861 | ret = misc_register(&ashmem_misc); | |
862 | if (unlikely(ret)) { | |
c810a399 | 863 | pr_err("failed to register misc device!\n"); |
a2df4e33 | 864 | goto out_free2; |
11980c2a RL |
865 | } |
866 | ||
867 | register_shrinker(&ashmem_shrinker); | |
868 | ||
c810a399 | 869 | pr_info("initialized\n"); |
11980c2a RL |
870 | |
871 | return 0; | |
a2df4e33 WT |
872 | |
873 | out_free2: | |
874 | kmem_cache_destroy(ashmem_range_cachep); | |
875 | out_free1: | |
876 | kmem_cache_destroy(ashmem_area_cachep); | |
877 | out: | |
878 | return ret; | |
11980c2a | 879 | } |
12950595 | 880 | device_initcall(ashmem_init); |