1 // SPDX-License-Identifier: GPL-2.0
2 /* mm/ashmem.c
3 *
4 * Anonymous Shared Memory Subsystem, ashmem
5 *
6 * Copyright (C) 2008 Google, Inc.
7 *
8 * Robert Love <rlove@google.com>
9 */
10
11 #define pr_fmt(fmt) "ashmem: " fmt
12
13 #include <linux/init.h>
14 #include <linux/export.h>
15 #include <linux/file.h>
16 #include <linux/fs.h>
17 #include <linux/falloc.h>
18 #include <linux/miscdevice.h>
19 #include <linux/security.h>
20 #include <linux/mm.h>
21 #include <linux/mman.h>
22 #include <linux/uaccess.h>
23 #include <linux/personality.h>
24 #include <linux/bitops.h>
25 #include <linux/mutex.h>
26 #include <linux/shmem_fs.h>
27 #include <linux/memcheck.h>
28 #include "ashmem.h"
29
30 #define ASHMEM_NAME_PREFIX "dev/ashmem/"
31 #define ASHMEM_NAME_PREFIX_LEN (sizeof(ASHMEM_NAME_PREFIX) - 1)
32 #define ASHMEM_FULL_NAME_LEN (ASHMEM_NAME_LEN + ASHMEM_NAME_PREFIX_LEN)
33
34 /**
35 * struct ashmem_area - The anonymous shared memory area
36 * @name: The optional name in /proc/pid/maps
37 * @unpinned_list: The list of all ashmem areas
38 * @file: The shmem-based backing file
39 * @size: The size of the mapping, in bytes
40 * @prot_mask: The allowed protection bits, as vm_flags
41 *
42 * The lifecycle of this structure is from our parent file's open() until
43 * its release(). It is also protected by 'ashmem_mutex'
44 *
45 * Warning: Mappings do NOT pin this structure; It dies on close()
46 */
47 struct ashmem_area {
48 char name[ASHMEM_FULL_NAME_LEN];
49 struct list_head unpinned_list;
50 struct file *file;
51 size_t size;
52 unsigned long prot_mask;
53 };
54
55 /**
56 * struct ashmem_range - A range of unpinned/evictable pages
57 * @lru: The entry in the LRU list
58 * @unpinned: The entry in its area's unpinned list
59 * @asma: The associated anonymous shared memory area.
60 * @pgstart: The starting page (inclusive)
61 * @pgend: The ending page (inclusive)
62 * @purged: The purge status (ASHMEM_NOT or ASHMEM_WAS_PURGED)
63 *
64 * The lifecycle of this structure is from unpin to pin.
65 * It is protected by 'ashmem_mutex'
66 */
67 struct ashmem_range {
68 struct list_head lru;
69 struct list_head unpinned;
70 struct ashmem_area *asma;
71 size_t pgstart;
72 size_t pgend;
73 unsigned int purged;
74 };
75
76 /* LRU list of unpinned pages, protected by ashmem_mutex */
77 static LIST_HEAD(ashmem_lru_list);
78
79 static atomic_t ashmem_shrink_inflight = ATOMIC_INIT(0);
80 static DECLARE_WAIT_QUEUE_HEAD(ashmem_shrink_wait);
81
82 /*
83 * long lru_count - The count of pages on our LRU list.
84 *
85 * This is protected by ashmem_mutex.
86 */
87 static unsigned long lru_count;
88
89 /*
90 * ashmem_mutex - protects the list of and each individual ashmem_area
91 *
92 * Lock Ordering: ashmex_mutex -> i_mutex -> i_alloc_sem
93 */
94 static DEFINE_MUTEX(ashmem_mutex);
95
96 static struct kmem_cache *ashmem_area_cachep __read_mostly;
97 static struct kmem_cache *ashmem_range_cachep __read_mostly;
98
99 /*
100 * A separate lockdep class for the backing shmem inodes to resolve the lockdep
101 * warning about the race between kswapd taking fs_reclaim before inode_lock
102 * and write syscall taking inode_lock and then fs_reclaim.
103 * Note that such race is impossible because ashmem does not support write
104 * syscalls operating on the backing shmem.
105 */
106 static struct lock_class_key backing_shmem_inode_class;
107
ashmem_mutex_lock(void)108 void ashmem_mutex_lock(void)
109 {
110 mutex_lock(&ashmem_mutex);
111 }
112
ashmem_mutex_unlock(void)113 void ashmem_mutex_unlock(void)
114 {
115 mutex_unlock(&ashmem_mutex);
116 }
117
range_size(struct ashmem_range * range)118 static inline unsigned long range_size(struct ashmem_range *range)
119 {
120 return range->pgend - range->pgstart + 1;
121 }
122
range_on_lru(struct ashmem_range * range)123 static inline bool range_on_lru(struct ashmem_range *range)
124 {
125 return range->purged == ASHMEM_NOT_PURGED;
126 }
127
page_range_subsumes_range(struct ashmem_range * range,size_t start,size_t end)128 static inline bool page_range_subsumes_range(struct ashmem_range *range,
129 size_t start, size_t end)
130 {
131 return (range->pgstart >= start) && (range->pgend <= end);
132 }
133
page_range_subsumed_by_range(struct ashmem_range * range,size_t start,size_t end)134 static inline bool page_range_subsumed_by_range(struct ashmem_range *range,
135 size_t start, size_t end)
136 {
137 return (range->pgstart <= start) && (range->pgend >= end);
138 }
139
page_in_range(struct ashmem_range * range,size_t page)140 static inline bool page_in_range(struct ashmem_range *range, size_t page)
141 {
142 return (range->pgstart <= page) && (range->pgend >= page);
143 }
144
page_range_in_range(struct ashmem_range * range,size_t start,size_t end)145 static inline bool page_range_in_range(struct ashmem_range *range,
146 size_t start, size_t end)
147 {
148 return page_in_range(range, start) || page_in_range(range, end) ||
149 page_range_subsumes_range(range, start, end);
150 }
151
range_before_page(struct ashmem_range * range,size_t page)152 static inline bool range_before_page(struct ashmem_range *range,
153 size_t page)
154 {
155 return range->pgend < page;
156 }
157
158 #define PROT_MASK (PROT_EXEC | PROT_READ | PROT_WRITE)
159
160 /**
161 * lru_add() - Adds a range of memory to the LRU list
162 * @range: The memory range being added.
163 *
164 * The range is first added to the end (tail) of the LRU list.
165 * After this, the size of the range is added to @lru_count
166 */
lru_add(struct ashmem_range * range)167 static inline void lru_add(struct ashmem_range *range)
168 {
169 list_add_tail(&range->lru, &ashmem_lru_list);
170 lru_count += range_size(range);
171 }
172
173 /**
174 * lru_del() - Removes a range of memory from the LRU list
175 * @range: The memory range being removed
176 *
177 * The range is first deleted from the LRU list.
178 * After this, the size of the range is removed from @lru_count
179 */
lru_del(struct ashmem_range * range)180 static inline void lru_del(struct ashmem_range *range)
181 {
182 list_del(&range->lru);
183 lru_count -= range_size(range);
184 }
185
186 /**
187 * range_alloc() - Allocates and initializes a new ashmem_range structure
188 * @asma: The associated ashmem_area
189 * @prev_range: The previous ashmem_range in the sorted asma->unpinned list
190 * @purged: Initial purge status (ASMEM_NOT_PURGED or ASHMEM_WAS_PURGED)
191 * @start: The starting page (inclusive)
192 * @end: The ending page (inclusive)
193 *
194 * This function is protected by ashmem_mutex.
195 */
range_alloc(struct ashmem_area * asma,struct ashmem_range * prev_range,unsigned int purged,size_t start,size_t end,struct ashmem_range ** new_range)196 static void range_alloc(struct ashmem_area *asma,
197 struct ashmem_range *prev_range, unsigned int purged,
198 size_t start, size_t end,
199 struct ashmem_range **new_range)
200 {
201 struct ashmem_range *range = *new_range;
202
203 *new_range = NULL;
204 range->asma = asma;
205 range->pgstart = start;
206 range->pgend = end;
207 range->purged = purged;
208
209 list_add_tail(&range->unpinned, &prev_range->unpinned);
210
211 if (range_on_lru(range))
212 lru_add(range);
213 }
214
215 /**
216 * range_del() - Deletes and deallocates an ashmem_range structure
217 * @range: The associated ashmem_range that has previously been allocated
218 */
range_del(struct ashmem_range * range)219 static void range_del(struct ashmem_range *range)
220 {
221 list_del(&range->unpinned);
222 if (range_on_lru(range))
223 lru_del(range);
224 kmem_cache_free(ashmem_range_cachep, range);
225 }
226
227 /**
228 * range_shrink() - Shrinks an ashmem_range
229 * @range: The associated ashmem_range being shrunk
230 * @start: The starting byte of the new range
231 * @end: The ending byte of the new range
232 *
233 * This does not modify the data inside the existing range in any way - It
234 * simply shrinks the boundaries of the range.
235 *
236 * Theoretically, with a little tweaking, this could eventually be changed
237 * to range_resize, and expand the lru_count if the new range is larger.
238 */
range_shrink(struct ashmem_range * range,size_t start,size_t end)239 static inline void range_shrink(struct ashmem_range *range,
240 size_t start, size_t end)
241 {
242 size_t pre = range_size(range);
243
244 range->pgstart = start;
245 range->pgend = end;
246
247 if (range_on_lru(range))
248 lru_count -= pre - range_size(range);
249 }
250
251 /**
252 * ashmem_open() - Opens an Anonymous Shared Memory structure
253 * @inode: The backing file's index node(?)
254 * @file: The backing file
255 *
256 * Please note that the ashmem_area is not returned by this function - It is
257 * instead written to "file->private_data".
258 *
259 * Return: 0 if successful, or another code if unsuccessful.
260 */
ashmem_open(struct inode * inode,struct file * file)261 static int ashmem_open(struct inode *inode, struct file *file)
262 {
263 struct ashmem_area *asma;
264 int ret;
265
266 ret = generic_file_open(inode, file);
267 if (ret)
268 return ret;
269
270 asma = kmem_cache_zalloc(ashmem_area_cachep, GFP_KERNEL);
271 if (!asma)
272 return -ENOMEM;
273
274 INIT_LIST_HEAD(&asma->unpinned_list);
275 memcpy(asma->name, ASHMEM_NAME_PREFIX, ASHMEM_NAME_PREFIX_LEN);
276 asma->prot_mask = PROT_MASK;
277 file->private_data = asma;
278
279 return 0;
280 }
281
282 /**
283 * ashmem_release() - Releases an Anonymous Shared Memory structure
284 * @ignored: The backing file's Index Node(?) - It is ignored here.
285 * @file: The backing file
286 *
287 * Return: 0 if successful. If it is anything else, go have a coffee and
288 * try again.
289 */
ashmem_release(struct inode * ignored,struct file * file)290 static int ashmem_release(struct inode *ignored, struct file *file)
291 {
292 struct ashmem_area *asma = file->private_data;
293 struct ashmem_range *range, *next;
294
295 mutex_lock(&ashmem_mutex);
296 list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned)
297 range_del(range);
298 mutex_unlock(&ashmem_mutex);
299
300 if (asma->file)
301 fput(asma->file);
302 kmem_cache_free(ashmem_area_cachep, asma);
303
304 return 0;
305 }
306
ashmem_read_iter(struct kiocb * iocb,struct iov_iter * iter)307 static ssize_t ashmem_read_iter(struct kiocb *iocb, struct iov_iter *iter)
308 {
309 struct ashmem_area *asma = iocb->ki_filp->private_data;
310 int ret = 0;
311
312 mutex_lock(&ashmem_mutex);
313
314 /* If size is not set, or set to 0, always return EOF. */
315 if (asma->size == 0)
316 goto out_unlock;
317
318 if (!asma->file) {
319 ret = -EBADF;
320 goto out_unlock;
321 }
322
323 /*
324 * asma and asma->file are used outside the lock here. We assume
325 * once asma->file is set it will never be changed, and will not
326 * be destroyed until all references to the file are dropped and
327 * ashmem_release is called.
328 */
329 mutex_unlock(&ashmem_mutex);
330 ret = vfs_iter_read(asma->file, iter, &iocb->ki_pos, 0);
331 mutex_lock(&ashmem_mutex);
332 if (ret > 0)
333 asma->file->f_pos = iocb->ki_pos;
334 out_unlock:
335 mutex_unlock(&ashmem_mutex);
336 return ret;
337 }
338
ashmem_llseek(struct file * file,loff_t offset,int origin)339 static loff_t ashmem_llseek(struct file *file, loff_t offset, int origin)
340 {
341 struct ashmem_area *asma = file->private_data;
342 loff_t ret;
343
344 mutex_lock(&ashmem_mutex);
345
346 if (asma->size == 0) {
347 mutex_unlock(&ashmem_mutex);
348 return -EINVAL;
349 }
350
351 if (!asma->file) {
352 mutex_unlock(&ashmem_mutex);
353 return -EBADF;
354 }
355
356 mutex_unlock(&ashmem_mutex);
357
358 ret = vfs_llseek(asma->file, offset, origin);
359 if (ret < 0)
360 return ret;
361
362 /** Copy f_pos from backing file, since f_ops->llseek() sets it */
363 file->f_pos = asma->file->f_pos;
364 return ret;
365 }
366
calc_vm_may_flags(unsigned long prot)367 static inline vm_flags_t calc_vm_may_flags(unsigned long prot)
368 {
369 return _calc_vm_trans(prot, PROT_READ, VM_MAYREAD) |
370 _calc_vm_trans(prot, PROT_WRITE, VM_MAYWRITE) |
371 _calc_vm_trans(prot, PROT_EXEC, VM_MAYEXEC);
372 }
373
ashmem_vmfile_mmap(struct file * file,struct vm_area_struct * vma)374 static int ashmem_vmfile_mmap(struct file *file, struct vm_area_struct *vma)
375 {
376 /* do not allow to mmap ashmem backing shmem file directly */
377 return -EPERM;
378 }
379
380 static unsigned long
ashmem_vmfile_get_unmapped_area(struct file * file,unsigned long addr,unsigned long len,unsigned long pgoff,unsigned long flags)381 ashmem_vmfile_get_unmapped_area(struct file *file, unsigned long addr,
382 unsigned long len, unsigned long pgoff,
383 unsigned long flags)
384 {
385 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
386 }
387
ashmem_mmap(struct file * file,struct vm_area_struct * vma)388 static int ashmem_mmap(struct file *file, struct vm_area_struct *vma)
389 {
390 static struct file_operations vmfile_fops;
391 struct ashmem_area *asma = file->private_data;
392 int ret = 0;
393
394 mutex_lock(&ashmem_mutex);
395
396 /* user needs to SET_SIZE before mapping */
397 if (!asma->size) {
398 ret = -EINVAL;
399 goto out;
400 }
401
402 /* requested mapping size larger than object size */
403 if (vma->vm_end - vma->vm_start > PAGE_ALIGN(asma->size)) {
404 ret = -EINVAL;
405 goto out;
406 }
407
408 /* requested protection bits must match our allowed protection mask */
409 if ((vma->vm_flags & ~calc_vm_prot_bits(asma->prot_mask, 0)) &
410 calc_vm_prot_bits(PROT_MASK, 0)) {
411 ret = -EPERM;
412 goto out;
413 }
414 vma->vm_flags &= ~calc_vm_may_flags(~asma->prot_mask);
415
416 if (!asma->file) {
417 char *name = ASHMEM_NAME_DEF;
418 struct file *vmfile;
419 struct inode *inode;
420
421 if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0')
422 name = asma->name;
423
424 /* ... and allocate the backing shmem file */
425 vmfile = shmem_file_setup(name, asma->size, vma->vm_flags);
426 if (IS_ERR(vmfile)) {
427 ret = PTR_ERR(vmfile);
428 goto out;
429 }
430 vmfile->f_mode |= FMODE_LSEEK;
431 inode = file_inode(vmfile);
432 lockdep_set_class(&inode->i_rwsem, &backing_shmem_inode_class);
433 asma->file = vmfile;
434 /*
435 * override mmap operation of the vmfile so that it can't be
436 * remapped which would lead to creation of a new vma with no
437 * asma permission checks. Have to override get_unmapped_area
438 * as well to prevent VM_BUG_ON check for f_ops modification.
439 */
440 if (!vmfile_fops.mmap) {
441 vmfile_fops = *vmfile->f_op;
442 vmfile_fops.mmap = ashmem_vmfile_mmap;
443 vmfile_fops.get_unmapped_area =
444 ashmem_vmfile_get_unmapped_area;
445 }
446 vmfile->f_op = &vmfile_fops;
447 }
448 get_file(asma->file);
449
450 /*
451 * XXX - Reworked to use shmem_zero_setup() instead of
452 * shmem_set_file while we're in staging. -jstultz
453 */
454 if (vma->vm_flags & VM_SHARED) {
455 ret = shmem_zero_setup(vma);
456 if (ret) {
457 fput(asma->file);
458 goto out;
459 }
460 } else {
461 vma_set_anonymous(vma);
462 }
463
464 if (vma->vm_file)
465 fput(vma->vm_file);
466 vma->vm_file = asma->file;
467
468 out:
469 mutex_unlock(&ashmem_mutex);
470 return ret;
471 }
472
473 /*
474 * ashmem_shrink - our cache shrinker, called from mm/vmscan.c
475 *
476 * 'nr_to_scan' is the number of objects to scan for freeing.
477 *
478 * 'gfp_mask' is the mask of the allocation that got us into this mess.
479 *
480 * Return value is the number of objects freed or -1 if we cannot
481 * proceed without risk of deadlock (due to gfp_mask).
482 *
483 * We approximate LRU via least-recently-unpinned, jettisoning unpinned partial
484 * chunks of ashmem regions LRU-wise one-at-a-time until we hit 'nr_to_scan'
485 * pages freed.
486 */
487 static unsigned long
ashmem_shrink_scan(struct shrinker * shrink,struct shrink_control * sc)488 ashmem_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
489 {
490 unsigned long freed = 0;
491
492 /* We might recurse into filesystem code, so bail out if necessary */
493 if (!(sc->gfp_mask & __GFP_FS))
494 return SHRINK_STOP;
495
496 if (!mutex_trylock(&ashmem_mutex))
497 return -1;
498
499 while (!list_empty(&ashmem_lru_list)) {
500 struct ashmem_range *range =
501 list_first_entry(&ashmem_lru_list, typeof(*range), lru);
502 loff_t start = range->pgstart * PAGE_SIZE;
503 loff_t end = (range->pgend + 1) * PAGE_SIZE;
504 struct file *f = range->asma->file;
505
506 get_file(f);
507 atomic_inc(&ashmem_shrink_inflight);
508 range->purged = ASHMEM_WAS_PURGED;
509 lru_del(range);
510
511 freed += range_size(range);
512 mutex_unlock(&ashmem_mutex);
513 f->f_op->fallocate(f,
514 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
515 start, end - start);
516 fput(f);
517 if (atomic_dec_and_test(&ashmem_shrink_inflight))
518 wake_up_all(&ashmem_shrink_wait);
519 if (!mutex_trylock(&ashmem_mutex))
520 goto out;
521 if (--sc->nr_to_scan <= 0)
522 break;
523 }
524 mutex_unlock(&ashmem_mutex);
525 out:
526 return freed;
527 }
528
529 static unsigned long
ashmem_shrink_count(struct shrinker * shrink,struct shrink_control * sc)530 ashmem_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
531 {
532 /*
533 * note that lru_count is count of pages on the lru, not a count of
534 * objects on the list. This means the scan function needs to return the
535 * number of pages freed, not the number of objects scanned.
536 */
537 return lru_count;
538 }
539
540 static struct shrinker ashmem_shrinker = {
541 .count_objects = ashmem_shrink_count,
542 .scan_objects = ashmem_shrink_scan,
543 /*
544 * XXX (dchinner): I wish people would comment on why they need on
545 * significant changes to the default value here
546 */
547 .seeks = DEFAULT_SEEKS * 4,
548 };
549
set_prot_mask(struct ashmem_area * asma,unsigned long prot)550 static int set_prot_mask(struct ashmem_area *asma, unsigned long prot)
551 {
552 int ret = 0;
553
554 mutex_lock(&ashmem_mutex);
555
556 /* the user can only remove, not add, protection bits */
557 if ((asma->prot_mask & prot) != prot) {
558 ret = -EINVAL;
559 goto out;
560 }
561
562 /* does the application expect PROT_READ to imply PROT_EXEC? */
563 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
564 prot |= PROT_EXEC;
565
566 asma->prot_mask = prot;
567
568 out:
569 mutex_unlock(&ashmem_mutex);
570 return ret;
571 }
572
set_name(struct ashmem_area * asma,void __user * name)573 static int set_name(struct ashmem_area *asma, void __user *name)
574 {
575 int len;
576 int ret = 0;
577 char local_name[ASHMEM_NAME_LEN];
578
579 /*
580 * Holding the ashmem_mutex while doing a copy_from_user might cause
581 * an data abort which would try to access mmap_lock. If another
582 * thread has invoked ashmem_mmap then it will be holding the
583 * semaphore and will be waiting for ashmem_mutex, there by leading to
584 * deadlock. We'll release the mutex and take the name to a local
585 * variable that does not need protection and later copy the local
586 * variable to the structure member with lock held.
587 */
588 len = strncpy_from_user(local_name, name, ASHMEM_NAME_LEN);
589 if (len < 0)
590 return len;
591
592 mutex_lock(&ashmem_mutex);
593 /* cannot change an existing mapping's name */
594 if (asma->file)
595 ret = -EINVAL;
596 else
597 strscpy(asma->name + ASHMEM_NAME_PREFIX_LEN, local_name,
598 ASHMEM_NAME_LEN);
599
600 mutex_unlock(&ashmem_mutex);
601 return ret;
602 }
603
get_name(struct ashmem_area * asma,void __user * name)604 static int get_name(struct ashmem_area *asma, void __user *name)
605 {
606 int ret = 0;
607 size_t len;
608 /*
609 * Have a local variable to which we'll copy the content
610 * from asma with the lock held. Later we can copy this to the user
611 * space safely without holding any locks. So even if we proceed to
612 * wait for mmap_lock, it won't lead to deadlock.
613 */
614 char local_name[ASHMEM_NAME_LEN];
615
616 mutex_lock(&ashmem_mutex);
617 if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0') {
618 /*
619 * Copying only `len', instead of ASHMEM_NAME_LEN, bytes
620 * prevents us from revealing one user's stack to another.
621 */
622 len = strlen(asma->name + ASHMEM_NAME_PREFIX_LEN) + 1;
623 memcpy(local_name, asma->name + ASHMEM_NAME_PREFIX_LEN, len);
624 } else {
625 len = sizeof(ASHMEM_NAME_DEF);
626 memcpy(local_name, ASHMEM_NAME_DEF, len);
627 }
628 mutex_unlock(&ashmem_mutex);
629
630 /*
631 * Now we are just copying from the stack variable to userland
632 * No lock held
633 */
634 if (copy_to_user(name, local_name, len))
635 ret = -EFAULT;
636 return ret;
637 }
638
639 /*
640 * ashmem_pin - pin the given ashmem region, returning whether it was
641 * previously purged (ASHMEM_WAS_PURGED) or not (ASHMEM_NOT_PURGED).
642 *
643 * Caller must hold ashmem_mutex.
644 */
ashmem_pin(struct ashmem_area * asma,size_t pgstart,size_t pgend,struct ashmem_range ** new_range)645 static int ashmem_pin(struct ashmem_area *asma, size_t pgstart, size_t pgend,
646 struct ashmem_range **new_range)
647 {
648 struct ashmem_range *range, *next;
649 int ret = ASHMEM_NOT_PURGED;
650
651 list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) {
652 /* moved past last applicable page; we can short circuit */
653 if (range_before_page(range, pgstart))
654 break;
655
656 /*
657 * The user can ask us to pin pages that span multiple ranges,
658 * or to pin pages that aren't even unpinned, so this is messy.
659 *
660 * Four cases:
661 * 1. The requested range subsumes an existing range, so we
662 * just remove the entire matching range.
663 * 2. The requested range overlaps the start of an existing
664 * range, so we just update that range.
665 * 3. The requested range overlaps the end of an existing
666 * range, so we just update that range.
667 * 4. The requested range punches a hole in an existing range,
668 * so we have to update one side of the range and then
669 * create a new range for the other side.
670 */
671 if (page_range_in_range(range, pgstart, pgend)) {
672 ret |= range->purged;
673
674 /* Case #1: Easy. Just nuke the whole thing. */
675 if (page_range_subsumes_range(range, pgstart, pgend)) {
676 range_del(range);
677 continue;
678 }
679
680 /* Case #2: We overlap from the start, so adjust it */
681 if (range->pgstart >= pgstart) {
682 range_shrink(range, pgend + 1, range->pgend);
683 continue;
684 }
685
686 /* Case #3: We overlap from the rear, so adjust it */
687 if (range->pgend <= pgend) {
688 range_shrink(range, range->pgstart,
689 pgstart - 1);
690 continue;
691 }
692
693 /*
694 * Case #4: We eat a chunk out of the middle. A bit
695 * more complicated, we allocate a new range for the
696 * second half and adjust the first chunk's endpoint.
697 */
698 range_alloc(asma, range, range->purged,
699 pgend + 1, range->pgend, new_range);
700 range_shrink(range, range->pgstart, pgstart - 1);
701 break;
702 }
703 }
704
705 return ret;
706 }
707
708 /*
709 * ashmem_unpin - unpin the given range of pages. Returns zero on success.
710 *
711 * Caller must hold ashmem_mutex.
712 */
ashmem_unpin(struct ashmem_area * asma,size_t pgstart,size_t pgend,struct ashmem_range ** new_range)713 static int ashmem_unpin(struct ashmem_area *asma, size_t pgstart, size_t pgend,
714 struct ashmem_range **new_range)
715 {
716 struct ashmem_range *range, *next;
717 unsigned int purged = ASHMEM_NOT_PURGED;
718
719 restart:
720 list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) {
721 /* short circuit: this is our insertion point */
722 if (range_before_page(range, pgstart))
723 break;
724
725 /*
726 * The user can ask us to unpin pages that are already entirely
727 * or partially pinned. We handle those two cases here.
728 */
729 if (page_range_subsumed_by_range(range, pgstart, pgend))
730 return 0;
731 if (page_range_in_range(range, pgstart, pgend)) {
732 pgstart = min(range->pgstart, pgstart);
733 pgend = max(range->pgend, pgend);
734 purged |= range->purged;
735 range_del(range);
736 goto restart;
737 }
738 }
739
740 range_alloc(asma, range, purged, pgstart, pgend, new_range);
741 return 0;
742 }
743
744 /*
745 * ashmem_get_pin_status - Returns ASHMEM_IS_UNPINNED if _any_ pages in the
746 * given interval are unpinned and ASHMEM_IS_PINNED otherwise.
747 *
748 * Caller must hold ashmem_mutex.
749 */
ashmem_get_pin_status(struct ashmem_area * asma,size_t pgstart,size_t pgend)750 static int ashmem_get_pin_status(struct ashmem_area *asma, size_t pgstart,
751 size_t pgend)
752 {
753 struct ashmem_range *range;
754 int ret = ASHMEM_IS_PINNED;
755
756 list_for_each_entry(range, &asma->unpinned_list, unpinned) {
757 if (range_before_page(range, pgstart))
758 break;
759 if (page_range_in_range(range, pgstart, pgend)) {
760 ret = ASHMEM_IS_UNPINNED;
761 break;
762 }
763 }
764
765 return ret;
766 }
767
ashmem_pin_unpin(struct ashmem_area * asma,unsigned long cmd,void __user * p)768 static int ashmem_pin_unpin(struct ashmem_area *asma, unsigned long cmd,
769 void __user *p)
770 {
771 struct ashmem_pin pin;
772 size_t pgstart, pgend;
773 int ret = -EINVAL;
774 struct ashmem_range *range = NULL;
775
776 if (copy_from_user(&pin, p, sizeof(pin)))
777 return -EFAULT;
778
779 if (cmd == ASHMEM_PIN || cmd == ASHMEM_UNPIN) {
780 range = kmem_cache_zalloc(ashmem_range_cachep, GFP_KERNEL);
781 if (!range)
782 return -ENOMEM;
783 }
784
785 mutex_lock(&ashmem_mutex);
786 wait_event(ashmem_shrink_wait, !atomic_read(&ashmem_shrink_inflight));
787
788 if (!asma->file)
789 goto out_unlock;
790
791 /* per custom, you can pass zero for len to mean "everything onward" */
792 if (!pin.len)
793 pin.len = PAGE_ALIGN(asma->size) - pin.offset;
794
795 if ((pin.offset | pin.len) & ~PAGE_MASK)
796 goto out_unlock;
797
798 if (((__u32)-1) - pin.offset < pin.len)
799 goto out_unlock;
800
801 if (PAGE_ALIGN(asma->size) < pin.offset + pin.len)
802 goto out_unlock;
803
804 pgstart = pin.offset / PAGE_SIZE;
805 pgend = pgstart + (pin.len / PAGE_SIZE) - 1;
806
807 switch (cmd) {
808 case ASHMEM_PIN:
809 ret = ashmem_pin(asma, pgstart, pgend, &range);
810 break;
811 case ASHMEM_UNPIN:
812 ret = ashmem_unpin(asma, pgstart, pgend, &range);
813 break;
814 case ASHMEM_GET_PIN_STATUS:
815 ret = ashmem_get_pin_status(asma, pgstart, pgend);
816 break;
817 }
818
819 out_unlock:
820 mutex_unlock(&ashmem_mutex);
821 if (range)
822 kmem_cache_free(ashmem_range_cachep, range);
823
824 return ret;
825 }
826
ashmem_ioctl(struct file * file,unsigned int cmd,unsigned long arg)827 static long ashmem_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
828 {
829 struct ashmem_area *asma = file->private_data;
830 long ret = -ENOTTY;
831
832 switch (cmd) {
833 case ASHMEM_SET_NAME:
834 ret = set_name(asma, (void __user *)arg);
835 break;
836 case ASHMEM_GET_NAME:
837 ret = get_name(asma, (void __user *)arg);
838 break;
839 case ASHMEM_SET_SIZE:
840 ret = -EINVAL;
841 mutex_lock(&ashmem_mutex);
842 if (!asma->file) {
843 ret = 0;
844 asma->size = (size_t)arg;
845 }
846 mutex_unlock(&ashmem_mutex);
847 break;
848 case ASHMEM_GET_SIZE:
849 ret = asma->size;
850 break;
851 case ASHMEM_SET_PROT_MASK:
852 ret = set_prot_mask(asma, arg);
853 break;
854 case ASHMEM_GET_PROT_MASK:
855 ret = asma->prot_mask;
856 break;
857 case ASHMEM_PIN:
858 case ASHMEM_UNPIN:
859 case ASHMEM_GET_PIN_STATUS:
860 ret = ashmem_pin_unpin(asma, cmd, (void __user *)arg);
861 break;
862 case ASHMEM_PURGE_ALL_CACHES:
863 ret = -EPERM;
864 if (capable(CAP_SYS_ADMIN)) {
865 struct shrink_control sc = {
866 .gfp_mask = GFP_KERNEL,
867 .nr_to_scan = LONG_MAX,
868 };
869 ret = ashmem_shrink_count(&ashmem_shrinker, &sc);
870 ashmem_shrink_scan(&ashmem_shrinker, &sc);
871 }
872 break;
873 }
874
875 return ret;
876 }
877
878 /* support of 32bit userspace on 64bit platforms */
879 #ifdef CONFIG_COMPAT
compat_ashmem_ioctl(struct file * file,unsigned int cmd,unsigned long arg)880 static long compat_ashmem_ioctl(struct file *file, unsigned int cmd,
881 unsigned long arg)
882 {
883 switch (cmd) {
884 case COMPAT_ASHMEM_SET_SIZE:
885 cmd = ASHMEM_SET_SIZE;
886 break;
887 case COMPAT_ASHMEM_SET_PROT_MASK:
888 cmd = ASHMEM_SET_PROT_MASK;
889 break;
890 }
891 return ashmem_ioctl(file, cmd, arg);
892 }
893 #endif
894 #ifdef CONFIG_PROC_FS
ashmem_show_fdinfo(struct seq_file * m,struct file * file)895 static void ashmem_show_fdinfo(struct seq_file *m, struct file *file)
896 {
897 struct ashmem_area *asma = file->private_data;
898
899 mutex_lock(&ashmem_mutex);
900
901 if (asma->file)
902 seq_printf(m, "inode:\t%ld\n", file_inode(asma->file)->i_ino);
903
904 if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0')
905 seq_printf(m, "name:\t%s\n",
906 asma->name + ASHMEM_NAME_PREFIX_LEN);
907
908 mutex_unlock(&ashmem_mutex);
909 }
910 #endif
911 static const struct file_operations ashmem_fops = {
912 .owner = THIS_MODULE,
913 .open = ashmem_open,
914 .release = ashmem_release,
915 .read_iter = ashmem_read_iter,
916 .llseek = ashmem_llseek,
917 .mmap = ashmem_mmap,
918 .unlocked_ioctl = ashmem_ioctl,
919 #ifdef CONFIG_COMPAT
920 .compat_ioctl = compat_ashmem_ioctl,
921 #endif
922 #ifdef CONFIG_PROC_FS
923 .show_fdinfo = ashmem_show_fdinfo,
924 #endif
925 };
926
is_ashmem_file(struct file * file)927 int is_ashmem_file(struct file *file)
928 {
929 return file->f_op == &ashmem_fops;
930 }
931
932 static struct miscdevice ashmem_misc = {
933 .minor = MISC_DYNAMIC_MINOR,
934 .name = "ashmem",
935 .fops = &ashmem_fops,
936 };
937
get_ashmem_size_by_file(struct file * f)938 size_t get_ashmem_size_by_file(struct file *f)
939 {
940 struct ashmem_area *asma = f->private_data;
941
942 if (asma)
943 return asma->size;
944 return 0;
945 }
946
get_ashmem_name_by_file(struct file * f)947 char *get_ashmem_name_by_file(struct file *f)
948 {
949 struct ashmem_area *asma = f->private_data;
950
951 if (asma)
952 return asma->name;
953 return NULL;
954 }
955
ashmem_init(void)956 static int __init ashmem_init(void)
957 {
958 int ret = -ENOMEM;
959
960 ashmem_area_cachep = kmem_cache_create("ashmem_area_cache",
961 sizeof(struct ashmem_area),
962 0, 0, NULL);
963 if (!ashmem_area_cachep) {
964 pr_err("failed to create slab cache\n");
965 goto out;
966 }
967
968 ashmem_range_cachep = kmem_cache_create("ashmem_range_cache",
969 sizeof(struct ashmem_range),
970 0, 0, NULL);
971 if (!ashmem_range_cachep) {
972 pr_err("failed to create slab cache\n");
973 goto out_free1;
974 }
975
976 ret = misc_register(&ashmem_misc);
977 if (ret) {
978 pr_err("failed to register misc device!\n");
979 goto out_free2;
980 }
981
982 ret = register_shrinker(&ashmem_shrinker);
983 if (ret) {
984 pr_err("failed to register shrinker!\n");
985 goto out_demisc;
986 }
987 init_ashmem_process_info();
988 pr_info("initialized\n");
989
990 return 0;
991
992 out_demisc:
993 misc_deregister(&ashmem_misc);
994 out_free2:
995 kmem_cache_destroy(ashmem_range_cachep);
996 out_free1:
997 kmem_cache_destroy(ashmem_area_cachep);
998 out:
999 return ret;
1000 }
1001 device_initcall(ashmem_init);
1002