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