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 | ||
11980c2a RL |
21 | #include <linux/module.h> |
22 | #include <linux/file.h> | |
23 | #include <linux/fs.h> | |
3f31d075 | 24 | #include <linux/falloc.h> |
11980c2a RL |
25 | #include <linux/miscdevice.h> |
26 | #include <linux/security.h> | |
27 | #include <linux/mm.h> | |
28 | #include <linux/mman.h> | |
29 | #include <linux/uaccess.h> | |
30 | #include <linux/personality.h> | |
31 | #include <linux/bitops.h> | |
32 | #include <linux/mutex.h> | |
33 | #include <linux/shmem_fs.h> | |
34 | #include "ashmem.h" | |
35 | ||
36 | #define ASHMEM_NAME_PREFIX "dev/ashmem/" | |
37 | #define ASHMEM_NAME_PREFIX_LEN (sizeof(ASHMEM_NAME_PREFIX) - 1) | |
38 | #define ASHMEM_FULL_NAME_LEN (ASHMEM_NAME_LEN + ASHMEM_NAME_PREFIX_LEN) | |
39 | ||
4d2c9d5d CJB |
40 | /** |
41 | * struct ashmem_area - The anonymous shared memory area | |
42 | * @name: The optional name in /proc/pid/maps | |
43 | * @unpinned_list: The list of all ashmem areas | |
44 | * @file: The shmem-based backing file | |
45 | * @size: The size of the mapping, in bytes | |
46 | * @prot_masks: The allowed protection bits, as vm_flags | |
47 | * | |
48 | * The lifecycle of this structure is from our parent file's open() until | |
49 | * its release(). It is also protected by 'ashmem_mutex' | |
50 | * | |
51 | * Warning: Mappings do NOT pin this structure; It dies on close() | |
11980c2a RL |
52 | */ |
53 | struct ashmem_area { | |
4d2c9d5d CJB |
54 | char name[ASHMEM_FULL_NAME_LEN]; |
55 | struct list_head unpinned_list; | |
56 | struct file *file; | |
57 | size_t size; | |
58 | unsigned long prot_mask; | |
11980c2a RL |
59 | }; |
60 | ||
4d2c9d5d CJB |
61 | /** |
62 | * struct ashmem_range - A range of unpinned/evictable pages | |
63 | * @lru: The entry in the LRU list | |
64 | * @unpinned: The entry in its area's unpinned list | |
65 | * @asma: The associated anonymous shared memory area. | |
66 | * @pgstart: The starting page (inclusive) | |
67 | * @pgend: The ending page (inclusive) | |
68 | * @purged: The purge status (ASHMEM_NOT or ASHMEM_WAS_PURGED) | |
69 | * | |
70 | * The lifecycle of this structure is from unpin to pin. | |
71 | * It is protected by 'ashmem_mutex' | |
11980c2a RL |
72 | */ |
73 | struct ashmem_range { | |
4d2c9d5d CJB |
74 | struct list_head lru; |
75 | struct list_head unpinned; | |
76 | struct ashmem_area *asma; | |
77 | size_t pgstart; | |
78 | size_t pgend; | |
79 | unsigned int purged; | |
11980c2a RL |
80 | }; |
81 | ||
82 | /* LRU list of unpinned pages, protected by ashmem_mutex */ | |
83 | static LIST_HEAD(ashmem_lru_list); | |
84 | ||
4d2c9d5d CJB |
85 | /** |
86 | * long lru_count - The count of pages on our LRU list. | |
87 | * | |
88 | * This is protected by ashmem_mutex. | |
89 | */ | |
11980c2a RL |
90 | static unsigned long lru_count; |
91 | ||
4d2c9d5d | 92 | /** |
11980c2a RL |
93 | * ashmem_mutex - protects the list of and each individual ashmem_area |
94 | * | |
95 | * Lock Ordering: ashmex_mutex -> i_mutex -> i_alloc_sem | |
96 | */ | |
97 | static DEFINE_MUTEX(ashmem_mutex); | |
98 | ||
99 | static struct kmem_cache *ashmem_area_cachep __read_mostly; | |
100 | static struct kmem_cache *ashmem_range_cachep __read_mostly; | |
101 | ||
102 | #define range_size(range) \ | |
1efb3439 | 103 | ((range)->pgend - (range)->pgstart + 1) |
11980c2a RL |
104 | |
105 | #define range_on_lru(range) \ | |
1efb3439 | 106 | ((range)->purged == ASHMEM_NOT_PURGED) |
11980c2a RL |
107 | |
108 | #define page_range_subsumes_range(range, start, end) \ | |
1efb3439 | 109 | (((range)->pgstart >= (start)) && ((range)->pgend <= (end))) |
11980c2a RL |
110 | |
111 | #define page_range_subsumed_by_range(range, start, end) \ | |
1efb3439 | 112 | (((range)->pgstart <= (start)) && ((range)->pgend >= (end))) |
11980c2a RL |
113 | |
114 | #define page_in_range(range, page) \ | |
1efb3439 | 115 | (((range)->pgstart <= (page)) && ((range)->pgend >= (page))) |
11980c2a RL |
116 | |
117 | #define page_range_in_range(range, start, end) \ | |
1efb3439 JS |
118 | (page_in_range(range, start) || page_in_range(range, end) || \ |
119 | page_range_subsumes_range(range, start, end)) | |
11980c2a RL |
120 | |
121 | #define range_before_page(range, page) \ | |
1efb3439 | 122 | ((range)->pgend < (page)) |
11980c2a RL |
123 | |
124 | #define PROT_MASK (PROT_EXEC | PROT_READ | PROT_WRITE) | |
125 | ||
4d2c9d5d CJB |
126 | /** |
127 | * lru_add() - Adds a range of memory to the LRU list | |
128 | * @range: The memory range being added. | |
129 | * | |
130 | * The range is first added to the end (tail) of the LRU list. | |
131 | * After this, the size of the range is added to @lru_count | |
132 | */ | |
11980c2a RL |
133 | static inline void lru_add(struct ashmem_range *range) |
134 | { | |
135 | list_add_tail(&range->lru, &ashmem_lru_list); | |
136 | lru_count += range_size(range); | |
137 | } | |
138 | ||
4d2c9d5d CJB |
139 | /** |
140 | * lru_del() - Removes a range of memory from the LRU list | |
141 | * @range: The memory range being removed | |
142 | * | |
143 | * The range is first deleted from the LRU list. | |
144 | * After this, the size of the range is removed from @lru_count | |
145 | */ | |
11980c2a RL |
146 | static inline void lru_del(struct ashmem_range *range) |
147 | { | |
148 | list_del(&range->lru); | |
149 | lru_count -= range_size(range); | |
150 | } | |
151 | ||
4d2c9d5d CJB |
152 | /** |
153 | * range_alloc() - Allocates and initializes a new ashmem_range structure | |
154 | * @asma: The associated ashmem_area | |
155 | * @prev_range: The previous ashmem_range in the sorted asma->unpinned list | |
156 | * @purged: Initial purge status (ASMEM_NOT_PURGED or ASHMEM_WAS_PURGED) | |
157 | * @start: The starting page (inclusive) | |
158 | * @end: The ending page (inclusive) | |
11980c2a | 159 | * |
4d2c9d5d | 160 | * This function is protected by ashmem_mutex. |
11980c2a | 161 | * |
4d2c9d5d | 162 | * Return: 0 if successful, or -ENOMEM if there is an error |
11980c2a RL |
163 | */ |
164 | static int range_alloc(struct ashmem_area *asma, | |
165 | struct ashmem_range *prev_range, unsigned int purged, | |
166 | size_t start, size_t end) | |
167 | { | |
168 | struct ashmem_range *range; | |
169 | ||
170 | range = kmem_cache_zalloc(ashmem_range_cachep, GFP_KERNEL); | |
171 | if (unlikely(!range)) | |
172 | return -ENOMEM; | |
173 | ||
174 | range->asma = asma; | |
175 | range->pgstart = start; | |
176 | range->pgend = end; | |
177 | range->purged = purged; | |
178 | ||
179 | list_add_tail(&range->unpinned, &prev_range->unpinned); | |
180 | ||
181 | if (range_on_lru(range)) | |
182 | lru_add(range); | |
183 | ||
184 | return 0; | |
185 | } | |
186 | ||
4d2c9d5d CJB |
187 | /** |
188 | * range_del() - Deletes and dealloctes an ashmem_range structure | |
189 | * @range: The associated ashmem_range that has previously been allocated | |
190 | */ | |
11980c2a RL |
191 | static void range_del(struct ashmem_range *range) |
192 | { | |
193 | list_del(&range->unpinned); | |
194 | if (range_on_lru(range)) | |
195 | lru_del(range); | |
196 | kmem_cache_free(ashmem_range_cachep, range); | |
197 | } | |
198 | ||
781114ce CJB |
199 | /** |
200 | * range_shrink() - Shrinks an ashmem_range | |
201 | * @range: The associated ashmem_range being shrunk | |
202 | * @start: The starting byte of the new range | |
203 | * @end: The ending byte of the new range | |
204 | * | |
205 | * This does not modify the data inside the existing range in any way - It | |
206 | * simply shrinks the boundaries of the range. | |
11980c2a | 207 | * |
781114ce CJB |
208 | * Theoretically, with a little tweaking, this could eventually be changed |
209 | * to range_resize, and expand the lru_count if the new range is larger. | |
11980c2a RL |
210 | */ |
211 | static inline void range_shrink(struct ashmem_range *range, | |
212 | size_t start, size_t end) | |
213 | { | |
214 | size_t pre = range_size(range); | |
215 | ||
216 | range->pgstart = start; | |
217 | range->pgend = end; | |
218 | ||
219 | if (range_on_lru(range)) | |
220 | lru_count -= pre - range_size(range); | |
221 | } | |
222 | ||
781114ce CJB |
223 | /** |
224 | * ashmem_open() - Opens an Anonymous Shared Memory structure | |
225 | * @inode: The backing file's index node(?) | |
226 | * @file: The backing file | |
227 | * | |
228 | * Please note that the ashmem_area is not returned by this function - It is | |
229 | * instead written to "file->private_data". | |
230 | * | |
231 | * Return: 0 if successful, or another code if unsuccessful. | |
232 | */ | |
11980c2a RL |
233 | static int ashmem_open(struct inode *inode, struct file *file) |
234 | { | |
235 | struct ashmem_area *asma; | |
236 | int ret; | |
237 | ||
5154b93b | 238 | ret = generic_file_open(inode, file); |
11980c2a RL |
239 | if (unlikely(ret)) |
240 | return ret; | |
241 | ||
242 | asma = kmem_cache_zalloc(ashmem_area_cachep, GFP_KERNEL); | |
243 | if (unlikely(!asma)) | |
244 | return -ENOMEM; | |
245 | ||
246 | INIT_LIST_HEAD(&asma->unpinned_list); | |
247 | memcpy(asma->name, ASHMEM_NAME_PREFIX, ASHMEM_NAME_PREFIX_LEN); | |
248 | asma->prot_mask = PROT_MASK; | |
249 | file->private_data = asma; | |
250 | ||
251 | return 0; | |
252 | } | |
253 | ||
781114ce CJB |
254 | /** |
255 | * ashmem_release() - Releases an Anonymous Shared Memory structure | |
256 | * @ignored: The backing file's Index Node(?) - It is ignored here. | |
257 | * @file: The backing file | |
258 | * | |
259 | * Return: 0 if successful. If it is anything else, go have a coffee and | |
260 | * try again. | |
261 | */ | |
11980c2a RL |
262 | static int ashmem_release(struct inode *ignored, struct file *file) |
263 | { | |
264 | struct ashmem_area *asma = file->private_data; | |
265 | struct ashmem_range *range, *next; | |
266 | ||
267 | mutex_lock(&ashmem_mutex); | |
268 | list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) | |
269 | range_del(range); | |
270 | mutex_unlock(&ashmem_mutex); | |
271 | ||
272 | if (asma->file) | |
273 | fput(asma->file); | |
274 | kmem_cache_free(ashmem_area_cachep, asma); | |
275 | ||
276 | return 0; | |
277 | } | |
278 | ||
781114ce CJB |
279 | /** |
280 | * ashmem_read() - Reads a set of bytes from an Ashmem-enabled file | |
281 | * @file: The associated backing file. | |
282 | * @buf: The buffer of data being written to | |
283 | * @len: The number of bytes being read | |
284 | * @pos: The position of the first byte to read. | |
285 | * | |
286 | * Return: 0 if successful, or another return code if not. | |
287 | */ | |
853ca7ae BB |
288 | static ssize_t ashmem_read(struct file *file, char __user *buf, |
289 | size_t len, loff_t *pos) | |
290 | { | |
291 | struct ashmem_area *asma = file->private_data; | |
292 | int ret = 0; | |
293 | ||
294 | mutex_lock(&ashmem_mutex); | |
295 | ||
296 | /* If size is not set, or set to 0, always return EOF. */ | |
1efb3439 | 297 | if (asma->size == 0) |
077f6db9 | 298 | goto out_unlock; |
853ca7ae BB |
299 | |
300 | if (!asma->file) { | |
301 | ret = -EBADF; | |
077f6db9 | 302 | goto out_unlock; |
853ca7ae BB |
303 | } |
304 | ||
077f6db9 | 305 | mutex_unlock(&ashmem_mutex); |
5154b93b | 306 | |
077f6db9 TP |
307 | /* |
308 | * asma and asma->file are used outside the lock here. We assume | |
309 | * once asma->file is set it will never be changed, and will not | |
310 | * be destroyed until all references to the file are dropped and | |
311 | * ashmem_release is called. | |
312 | */ | |
313 | ret = asma->file->f_op->read(asma->file, buf, len, pos); | |
314 | if (ret >= 0) { | |
315 | /** Update backing file pos, since f_ops->read() doesn't */ | |
316 | asma->file->f_pos = *pos; | |
317 | } | |
318 | return ret; | |
5154b93b | 319 | |
077f6db9 | 320 | out_unlock: |
5154b93b BB |
321 | mutex_unlock(&ashmem_mutex); |
322 | return ret; | |
323 | } | |
324 | ||
325 | static loff_t ashmem_llseek(struct file *file, loff_t offset, int origin) | |
326 | { | |
327 | struct ashmem_area *asma = file->private_data; | |
328 | int ret; | |
329 | ||
330 | mutex_lock(&ashmem_mutex); | |
331 | ||
332 | if (asma->size == 0) { | |
333 | ret = -EINVAL; | |
334 | goto out; | |
335 | } | |
336 | ||
337 | if (!asma->file) { | |
338 | ret = -EBADF; | |
339 | goto out; | |
340 | } | |
341 | ||
91360b02 | 342 | ret = vfs_llseek(asma->file, offset, origin); |
1efb3439 | 343 | if (ret < 0) |
5154b93b | 344 | goto out; |
5154b93b BB |
345 | |
346 | /** Copy f_pos from backing file, since f_ops->llseek() sets it */ | |
347 | file->f_pos = asma->file->f_pos; | |
853ca7ae BB |
348 | |
349 | out: | |
350 | mutex_unlock(&ashmem_mutex); | |
351 | return ret; | |
352 | } | |
353 | ||
7cfce77d | 354 | static inline vm_flags_t calc_vm_may_flags(unsigned long prot) |
56f76fc6 | 355 | { |
1efb3439 | 356 | return _calc_vm_trans(prot, PROT_READ, VM_MAYREAD) | |
56f76fc6 AH |
357 | _calc_vm_trans(prot, PROT_WRITE, VM_MAYWRITE) | |
358 | _calc_vm_trans(prot, PROT_EXEC, VM_MAYEXEC); | |
359 | } | |
853ca7ae | 360 | |
11980c2a RL |
361 | static int ashmem_mmap(struct file *file, struct vm_area_struct *vma) |
362 | { | |
363 | struct ashmem_area *asma = file->private_data; | |
364 | int ret = 0; | |
365 | ||
366 | mutex_lock(&ashmem_mutex); | |
367 | ||
368 | /* user needs to SET_SIZE before mapping */ | |
369 | if (unlikely(!asma->size)) { | |
370 | ret = -EINVAL; | |
371 | goto out; | |
372 | } | |
373 | ||
374 | /* requested protection bits must match our allowed protection mask */ | |
56f76fc6 | 375 | if (unlikely((vma->vm_flags & ~calc_vm_prot_bits(asma->prot_mask)) & |
1efb3439 | 376 | calc_vm_prot_bits(PROT_MASK))) { |
11980c2a RL |
377 | ret = -EPERM; |
378 | goto out; | |
379 | } | |
56f76fc6 | 380 | vma->vm_flags &= ~calc_vm_may_flags(~asma->prot_mask); |
11980c2a RL |
381 | |
382 | if (!asma->file) { | |
383 | char *name = ASHMEM_NAME_DEF; | |
384 | struct file *vmfile; | |
385 | ||
386 | if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0') | |
387 | name = asma->name; | |
388 | ||
389 | /* ... and allocate the backing shmem file */ | |
390 | vmfile = shmem_file_setup(name, asma->size, vma->vm_flags); | |
391 | if (unlikely(IS_ERR(vmfile))) { | |
392 | ret = PTR_ERR(vmfile); | |
393 | goto out; | |
394 | } | |
395 | asma->file = vmfile; | |
396 | } | |
397 | get_file(asma->file); | |
398 | ||
399 | /* | |
aa5af974 | 400 | * XXX - Reworked to use shmem_zero_setup() instead of |
11980c2a RL |
401 | * shmem_set_file while we're in staging. -jstultz |
402 | */ | |
403 | if (vma->vm_flags & VM_SHARED) { | |
404 | ret = shmem_zero_setup(vma); | |
405 | if (ret) { | |
406 | fput(asma->file); | |
407 | goto out; | |
408 | } | |
409 | } | |
410 | ||
411 | if (vma->vm_file) | |
412 | fput(vma->vm_file); | |
413 | vma->vm_file = asma->file; | |
11980c2a RL |
414 | |
415 | out: | |
416 | mutex_unlock(&ashmem_mutex); | |
417 | return ret; | |
418 | } | |
419 | ||
420 | /* | |
6b4f7799 | 421 | * ashmem_shrink - our cache shrinker, called from mm/vmscan.c |
11980c2a | 422 | * |
7dc19d5a | 423 | * 'nr_to_scan' is the number of objects to scan for freeing. |
11980c2a RL |
424 | * |
425 | * 'gfp_mask' is the mask of the allocation that got us into this mess. | |
426 | * | |
7dc19d5a | 427 | * Return value is the number of objects freed or -1 if we cannot |
11980c2a RL |
428 | * proceed without risk of deadlock (due to gfp_mask). |
429 | * | |
430 | * We approximate LRU via least-recently-unpinned, jettisoning unpinned partial | |
431 | * chunks of ashmem regions LRU-wise one-at-a-time until we hit 'nr_to_scan' | |
432 | * pages freed. | |
433 | */ | |
7dc19d5a DC |
434 | static unsigned long |
435 | ashmem_shrink_scan(struct shrinker *shrink, struct shrink_control *sc) | |
11980c2a RL |
436 | { |
437 | struct ashmem_range *range, *next; | |
7dc19d5a | 438 | unsigned long freed = 0; |
11980c2a RL |
439 | |
440 | /* We might recurse into filesystem code, so bail out if necessary */ | |
7dc19d5a DC |
441 | if (!(sc->gfp_mask & __GFP_FS)) |
442 | return SHRINK_STOP; | |
11980c2a RL |
443 | |
444 | mutex_lock(&ashmem_mutex); | |
445 | list_for_each_entry_safe(range, next, &ashmem_lru_list, lru) { | |
11980c2a | 446 | loff_t start = range->pgstart * PAGE_SIZE; |
3f31d075 | 447 | loff_t end = (range->pgend + 1) * PAGE_SIZE; |
11980c2a | 448 | |
72c72bdf | 449 | vfs_fallocate(range->asma->file, |
3f31d075 HD |
450 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
451 | start, end - start); | |
11980c2a RL |
452 | range->purged = ASHMEM_WAS_PURGED; |
453 | lru_del(range); | |
454 | ||
7dc19d5a DC |
455 | freed += range_size(range); |
456 | if (--sc->nr_to_scan <= 0) | |
11980c2a RL |
457 | break; |
458 | } | |
459 | mutex_unlock(&ashmem_mutex); | |
7dc19d5a DC |
460 | return freed; |
461 | } | |
11980c2a | 462 | |
7dc19d5a DC |
463 | static unsigned long |
464 | ashmem_shrink_count(struct shrinker *shrink, struct shrink_control *sc) | |
465 | { | |
466 | /* | |
467 | * note that lru_count is count of pages on the lru, not a count of | |
468 | * objects on the list. This means the scan function needs to return the | |
469 | * number of pages freed, not the number of objects scanned. | |
470 | */ | |
11980c2a RL |
471 | return lru_count; |
472 | } | |
473 | ||
474 | static struct shrinker ashmem_shrinker = { | |
7dc19d5a DC |
475 | .count_objects = ashmem_shrink_count, |
476 | .scan_objects = ashmem_shrink_scan, | |
477 | /* | |
478 | * XXX (dchinner): I wish people would comment on why they need on | |
479 | * significant changes to the default value here | |
480 | */ | |
11980c2a RL |
481 | .seeks = DEFAULT_SEEKS * 4, |
482 | }; | |
483 | ||
484 | static int set_prot_mask(struct ashmem_area *asma, unsigned long prot) | |
485 | { | |
486 | int ret = 0; | |
487 | ||
488 | mutex_lock(&ashmem_mutex); | |
489 | ||
490 | /* the user can only remove, not add, protection bits */ | |
491 | if (unlikely((asma->prot_mask & prot) != prot)) { | |
492 | ret = -EINVAL; | |
493 | goto out; | |
494 | } | |
495 | ||
496 | /* does the application expect PROT_READ to imply PROT_EXEC? */ | |
497 | if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) | |
498 | prot |= PROT_EXEC; | |
499 | ||
500 | asma->prot_mask = prot; | |
501 | ||
502 | out: | |
503 | mutex_unlock(&ashmem_mutex); | |
504 | return ret; | |
505 | } | |
506 | ||
507 | static int set_name(struct ashmem_area *asma, void __user *name) | |
508 | { | |
077f6db9 | 509 | int len; |
11980c2a | 510 | int ret = 0; |
e5834d62 | 511 | char local_name[ASHMEM_NAME_LEN]; |
11980c2a | 512 | |
e5834d62 SB |
513 | /* |
514 | * Holding the ashmem_mutex while doing a copy_from_user might cause | |
515 | * an data abort which would try to access mmap_sem. If another | |
516 | * thread has invoked ashmem_mmap then it will be holding the | |
517 | * semaphore and will be waiting for ashmem_mutex, there by leading to | |
518 | * deadlock. We'll release the mutex and take the name to a local | |
519 | * variable that does not need protection and later copy the local | |
520 | * variable to the structure member with lock held. | |
521 | */ | |
077f6db9 TP |
522 | len = strncpy_from_user(local_name, name, ASHMEM_NAME_LEN); |
523 | if (len < 0) | |
524 | return len; | |
525 | if (len == ASHMEM_NAME_LEN) | |
526 | local_name[ASHMEM_NAME_LEN - 1] = '\0'; | |
e5834d62 | 527 | mutex_lock(&ashmem_mutex); |
11980c2a | 528 | /* cannot change an existing mapping's name */ |
077f6db9 | 529 | if (unlikely(asma->file)) |
11980c2a | 530 | ret = -EINVAL; |
077f6db9 TP |
531 | else |
532 | strcpy(asma->name + ASHMEM_NAME_PREFIX_LEN, local_name); | |
11980c2a | 533 | |
077f6db9 | 534 | mutex_unlock(&ashmem_mutex); |
11980c2a RL |
535 | return ret; |
536 | } | |
537 | ||
538 | static int get_name(struct ashmem_area *asma, void __user *name) | |
539 | { | |
540 | int ret = 0; | |
e5834d62 SB |
541 | size_t len; |
542 | /* | |
543 | * Have a local variable to which we'll copy the content | |
544 | * from asma with the lock held. Later we can copy this to the user | |
545 | * space safely without holding any locks. So even if we proceed to | |
546 | * wait for mmap_sem, it won't lead to deadlock. | |
547 | */ | |
548 | char local_name[ASHMEM_NAME_LEN]; | |
11980c2a RL |
549 | |
550 | mutex_lock(&ashmem_mutex); | |
551 | if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0') { | |
11980c2a RL |
552 | |
553 | /* | |
554 | * Copying only `len', instead of ASHMEM_NAME_LEN, bytes | |
555 | * prevents us from revealing one user's stack to another. | |
556 | */ | |
557 | len = strlen(asma->name + ASHMEM_NAME_PREFIX_LEN) + 1; | |
e5834d62 | 558 | memcpy(local_name, asma->name + ASHMEM_NAME_PREFIX_LEN, len); |
11980c2a | 559 | } else { |
e5834d62 SB |
560 | len = sizeof(ASHMEM_NAME_DEF); |
561 | memcpy(local_name, ASHMEM_NAME_DEF, len); | |
11980c2a RL |
562 | } |
563 | mutex_unlock(&ashmem_mutex); | |
564 | ||
e5834d62 SB |
565 | /* |
566 | * Now we are just copying from the stack variable to userland | |
567 | * No lock held | |
568 | */ | |
569 | if (unlikely(copy_to_user(name, local_name, len))) | |
570 | ret = -EFAULT; | |
11980c2a RL |
571 | return ret; |
572 | } | |
573 | ||
574 | /* | |
575 | * ashmem_pin - pin the given ashmem region, returning whether it was | |
576 | * previously purged (ASHMEM_WAS_PURGED) or not (ASHMEM_NOT_PURGED). | |
577 | * | |
578 | * Caller must hold ashmem_mutex. | |
579 | */ | |
580 | static int ashmem_pin(struct ashmem_area *asma, size_t pgstart, size_t pgend) | |
581 | { | |
582 | struct ashmem_range *range, *next; | |
583 | int ret = ASHMEM_NOT_PURGED; | |
584 | ||
585 | list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) { | |
586 | /* moved past last applicable page; we can short circuit */ | |
587 | if (range_before_page(range, pgstart)) | |
588 | break; | |
589 | ||
590 | /* | |
591 | * The user can ask us to pin pages that span multiple ranges, | |
592 | * or to pin pages that aren't even unpinned, so this is messy. | |
593 | * | |
594 | * Four cases: | |
595 | * 1. The requested range subsumes an existing range, so we | |
596 | * just remove the entire matching range. | |
597 | * 2. The requested range overlaps the start of an existing | |
598 | * range, so we just update that range. | |
599 | * 3. The requested range overlaps the end of an existing | |
600 | * range, so we just update that range. | |
601 | * 4. The requested range punches a hole in an existing range, | |
602 | * so we have to update one side of the range and then | |
603 | * create a new range for the other side. | |
604 | */ | |
605 | if (page_range_in_range(range, pgstart, pgend)) { | |
606 | ret |= range->purged; | |
607 | ||
608 | /* Case #1: Easy. Just nuke the whole thing. */ | |
609 | if (page_range_subsumes_range(range, pgstart, pgend)) { | |
610 | range_del(range); | |
611 | continue; | |
612 | } | |
613 | ||
614 | /* Case #2: We overlap from the start, so adjust it */ | |
615 | if (range->pgstart >= pgstart) { | |
616 | range_shrink(range, pgend + 1, range->pgend); | |
617 | continue; | |
618 | } | |
619 | ||
620 | /* Case #3: We overlap from the rear, so adjust it */ | |
621 | if (range->pgend <= pgend) { | |
622 | range_shrink(range, range->pgstart, pgstart-1); | |
623 | continue; | |
624 | } | |
625 | ||
626 | /* | |
627 | * Case #4: We eat a chunk out of the middle. A bit | |
628 | * more complicated, we allocate a new range for the | |
629 | * second half and adjust the first chunk's endpoint. | |
630 | */ | |
631 | range_alloc(asma, range, range->purged, | |
632 | pgend + 1, range->pgend); | |
633 | range_shrink(range, range->pgstart, pgstart - 1); | |
634 | break; | |
635 | } | |
636 | } | |
637 | ||
638 | return ret; | |
639 | } | |
640 | ||
641 | /* | |
642 | * ashmem_unpin - unpin the given range of pages. Returns zero on success. | |
643 | * | |
644 | * Caller must hold ashmem_mutex. | |
645 | */ | |
646 | static int ashmem_unpin(struct ashmem_area *asma, size_t pgstart, size_t pgend) | |
647 | { | |
648 | struct ashmem_range *range, *next; | |
649 | unsigned int purged = ASHMEM_NOT_PURGED; | |
650 | ||
651 | restart: | |
652 | list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) { | |
653 | /* short circuit: this is our insertion point */ | |
654 | if (range_before_page(range, pgstart)) | |
655 | break; | |
656 | ||
657 | /* | |
658 | * The user can ask us to unpin pages that are already entirely | |
659 | * or partially pinned. We handle those two cases here. | |
660 | */ | |
661 | if (page_range_subsumed_by_range(range, pgstart, pgend)) | |
662 | return 0; | |
663 | if (page_range_in_range(range, pgstart, pgend)) { | |
664 | pgstart = min_t(size_t, range->pgstart, pgstart), | |
665 | pgend = max_t(size_t, range->pgend, pgend); | |
666 | purged |= range->purged; | |
667 | range_del(range); | |
668 | goto restart; | |
669 | } | |
670 | } | |
671 | ||
672 | return range_alloc(asma, range, purged, pgstart, pgend); | |
673 | } | |
674 | ||
675 | /* | |
676 | * ashmem_get_pin_status - Returns ASHMEM_IS_UNPINNED if _any_ pages in the | |
677 | * given interval are unpinned and ASHMEM_IS_PINNED otherwise. | |
678 | * | |
679 | * Caller must hold ashmem_mutex. | |
680 | */ | |
681 | static int ashmem_get_pin_status(struct ashmem_area *asma, size_t pgstart, | |
682 | size_t pgend) | |
683 | { | |
684 | struct ashmem_range *range; | |
685 | int ret = ASHMEM_IS_PINNED; | |
686 | ||
687 | list_for_each_entry(range, &asma->unpinned_list, unpinned) { | |
688 | if (range_before_page(range, pgstart)) | |
689 | break; | |
690 | if (page_range_in_range(range, pgstart, pgend)) { | |
691 | ret = ASHMEM_IS_UNPINNED; | |
692 | break; | |
693 | } | |
694 | } | |
695 | ||
696 | return ret; | |
697 | } | |
698 | ||
699 | static int ashmem_pin_unpin(struct ashmem_area *asma, unsigned long cmd, | |
700 | void __user *p) | |
701 | { | |
702 | struct ashmem_pin pin; | |
703 | size_t pgstart, pgend; | |
704 | int ret = -EINVAL; | |
705 | ||
706 | if (unlikely(!asma->file)) | |
707 | return -EINVAL; | |
708 | ||
709 | if (unlikely(copy_from_user(&pin, p, sizeof(pin)))) | |
710 | return -EFAULT; | |
711 | ||
712 | /* per custom, you can pass zero for len to mean "everything onward" */ | |
713 | if (!pin.len) | |
714 | pin.len = PAGE_ALIGN(asma->size) - pin.offset; | |
715 | ||
716 | if (unlikely((pin.offset | pin.len) & ~PAGE_MASK)) | |
717 | return -EINVAL; | |
718 | ||
719 | if (unlikely(((__u32) -1) - pin.offset < pin.len)) | |
720 | return -EINVAL; | |
721 | ||
722 | if (unlikely(PAGE_ALIGN(asma->size) < pin.offset + pin.len)) | |
723 | return -EINVAL; | |
724 | ||
725 | pgstart = pin.offset / PAGE_SIZE; | |
726 | pgend = pgstart + (pin.len / PAGE_SIZE) - 1; | |
727 | ||
728 | mutex_lock(&ashmem_mutex); | |
729 | ||
730 | switch (cmd) { | |
731 | case ASHMEM_PIN: | |
732 | ret = ashmem_pin(asma, pgstart, pgend); | |
733 | break; | |
734 | case ASHMEM_UNPIN: | |
735 | ret = ashmem_unpin(asma, pgstart, pgend); | |
736 | break; | |
737 | case ASHMEM_GET_PIN_STATUS: | |
738 | ret = ashmem_get_pin_status(asma, pgstart, pgend); | |
739 | break; | |
740 | } | |
741 | ||
742 | mutex_unlock(&ashmem_mutex); | |
743 | ||
744 | return ret; | |
745 | } | |
746 | ||
747 | static long ashmem_ioctl(struct file *file, unsigned int cmd, unsigned long arg) | |
748 | { | |
749 | struct ashmem_area *asma = file->private_data; | |
750 | long ret = -ENOTTY; | |
751 | ||
752 | switch (cmd) { | |
753 | case ASHMEM_SET_NAME: | |
754 | ret = set_name(asma, (void __user *) arg); | |
755 | break; | |
756 | case ASHMEM_GET_NAME: | |
757 | ret = get_name(asma, (void __user *) arg); | |
758 | break; | |
759 | case ASHMEM_SET_SIZE: | |
760 | ret = -EINVAL; | |
761 | if (!asma->file) { | |
762 | ret = 0; | |
763 | asma->size = (size_t) arg; | |
764 | } | |
765 | break; | |
766 | case ASHMEM_GET_SIZE: | |
767 | ret = asma->size; | |
768 | break; | |
769 | case ASHMEM_SET_PROT_MASK: | |
770 | ret = set_prot_mask(asma, arg); | |
771 | break; | |
772 | case ASHMEM_GET_PROT_MASK: | |
773 | ret = asma->prot_mask; | |
774 | break; | |
775 | case ASHMEM_PIN: | |
776 | case ASHMEM_UNPIN: | |
777 | case ASHMEM_GET_PIN_STATUS: | |
778 | ret = ashmem_pin_unpin(asma, cmd, (void __user *) arg); | |
779 | break; | |
780 | case ASHMEM_PURGE_ALL_CACHES: | |
781 | ret = -EPERM; | |
782 | if (capable(CAP_SYS_ADMIN)) { | |
33e8fc46 CC |
783 | struct shrink_control sc = { |
784 | .gfp_mask = GFP_KERNEL, | |
7dc19d5a | 785 | .nr_to_scan = LONG_MAX, |
33e8fc46 | 786 | }; |
59573240 | 787 | ret = ashmem_shrink_count(&ashmem_shrinker, &sc); |
7dc19d5a | 788 | ashmem_shrink_scan(&ashmem_shrinker, &sc); |
11980c2a RL |
789 | } |
790 | break; | |
791 | } | |
792 | ||
793 | return ret; | |
794 | } | |
795 | ||
e9f5b814 SC |
796 | /* support of 32bit userspace on 64bit platforms */ |
797 | #ifdef CONFIG_COMPAT | |
017ca3bf MR |
798 | static long compat_ashmem_ioctl(struct file *file, unsigned int cmd, |
799 | unsigned long arg) | |
e9f5b814 SC |
800 | { |
801 | ||
802 | switch (cmd) { | |
803 | case COMPAT_ASHMEM_SET_SIZE: | |
804 | cmd = ASHMEM_SET_SIZE; | |
805 | break; | |
806 | case COMPAT_ASHMEM_SET_PROT_MASK: | |
807 | cmd = ASHMEM_SET_PROT_MASK; | |
808 | break; | |
809 | } | |
810 | return ashmem_ioctl(file, cmd, arg); | |
811 | } | |
812 | #endif | |
813 | ||
aa5af974 | 814 | static const struct file_operations ashmem_fops = { |
11980c2a RL |
815 | .owner = THIS_MODULE, |
816 | .open = ashmem_open, | |
817 | .release = ashmem_release, | |
1efb3439 JS |
818 | .read = ashmem_read, |
819 | .llseek = ashmem_llseek, | |
11980c2a RL |
820 | .mmap = ashmem_mmap, |
821 | .unlocked_ioctl = ashmem_ioctl, | |
e9f5b814 SC |
822 | #ifdef CONFIG_COMPAT |
823 | .compat_ioctl = compat_ashmem_ioctl, | |
824 | #endif | |
11980c2a RL |
825 | }; |
826 | ||
827 | static struct miscdevice ashmem_misc = { | |
828 | .minor = MISC_DYNAMIC_MINOR, | |
829 | .name = "ashmem", | |
830 | .fops = &ashmem_fops, | |
831 | }; | |
832 | ||
833 | static int __init ashmem_init(void) | |
834 | { | |
835 | int ret; | |
836 | ||
837 | ashmem_area_cachep = kmem_cache_create("ashmem_area_cache", | |
838 | sizeof(struct ashmem_area), | |
839 | 0, 0, NULL); | |
840 | if (unlikely(!ashmem_area_cachep)) { | |
c810a399 | 841 | pr_err("failed to create slab cache\n"); |
11980c2a RL |
842 | return -ENOMEM; |
843 | } | |
844 | ||
845 | ashmem_range_cachep = kmem_cache_create("ashmem_range_cache", | |
846 | sizeof(struct ashmem_range), | |
847 | 0, 0, NULL); | |
848 | if (unlikely(!ashmem_range_cachep)) { | |
c810a399 | 849 | pr_err("failed to create slab cache\n"); |
11980c2a RL |
850 | return -ENOMEM; |
851 | } | |
852 | ||
853 | ret = misc_register(&ashmem_misc); | |
854 | if (unlikely(ret)) { | |
c810a399 | 855 | pr_err("failed to register misc device!\n"); |
11980c2a RL |
856 | return ret; |
857 | } | |
858 | ||
859 | register_shrinker(&ashmem_shrinker); | |
860 | ||
c810a399 | 861 | pr_info("initialized\n"); |
11980c2a RL |
862 | |
863 | return 0; | |
864 | } | |
865 | ||
866 | static void __exit ashmem_exit(void) | |
867 | { | |
868 | int ret; | |
869 | ||
870 | unregister_shrinker(&ashmem_shrinker); | |
871 | ||
872 | ret = misc_deregister(&ashmem_misc); | |
873 | if (unlikely(ret)) | |
c810a399 | 874 | pr_err("failed to unregister misc device!\n"); |
11980c2a RL |
875 | |
876 | kmem_cache_destroy(ashmem_range_cachep); | |
877 | kmem_cache_destroy(ashmem_area_cachep); | |
878 | ||
c810a399 | 879 | pr_info("unloaded\n"); |
11980c2a RL |
880 | } |
881 | ||
882 | module_init(ashmem_init); | |
883 | module_exit(ashmem_exit); | |
884 | ||
885 | MODULE_LICENSE("GPL"); |