ashmem.c revision 853ca7ae292f9c4809e1e42914e81453eaa15367 
1/* mm/ashmem.c
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*/
18
19#include <linux/module.h>
20#include <linux/file.h>
21#include <linux/fs.h>
22#include <linux/miscdevice.h>
23#include <linux/security.h>
24#include <linux/mm.h>
25#include <linux/mman.h>
26#include <linux/uaccess.h>
27#include <linux/personality.h>
28#include <linux/bitops.h>
29#include <linux/mutex.h>
30#include <linux/shmem_fs.h>
31#include "ashmem.h"
32
33#define ASHMEM_NAME_PREFIX "dev/ashmem/"
34#define ASHMEM_NAME_PREFIX_LEN (sizeof(ASHMEM_NAME_PREFIX) - 1)
35#define ASHMEM_FULL_NAME_LEN (ASHMEM_NAME_LEN + ASHMEM_NAME_PREFIX_LEN)
36
37/*
38 * ashmem_area - anonymous shared memory area
39 * Lifecycle: From our parent file's open() until its release()
40 * Locking: Protected by `ashmem_mutex'
41 * Big Note: Mappings do NOT pin this structure; it dies on close()
42 */
43struct ashmem_area {
44	char name[ASHMEM_FULL_NAME_LEN];/* optional name for /proc/pid/maps */
45	struct list_head unpinned_list;	/* list of all ashmem areas */
46	struct file *file;		/* the shmem-based backing file */
47	size_t size;			/* size of the mapping, in bytes */
48	unsigned long prot_mask;	/* allowed prot bits, as vm_flags */
49};
50
51/*
52 * ashmem_range - represents an interval of unpinned (evictable) pages
53 * Lifecycle: From unpin to pin
54 * Locking: Protected by `ashmem_mutex'
55 */
56struct ashmem_range {
57	struct list_head lru;		/* entry in LRU list */
58	struct list_head unpinned;	/* entry in its area's unpinned list */
59	struct ashmem_area *asma;	/* associated area */
60	size_t pgstart;			/* starting page, inclusive */
61	size_t pgend;			/* ending page, inclusive */
62	unsigned int purged;		/* ASHMEM_NOT or ASHMEM_WAS_PURGED */
63};
64
65/* LRU list of unpinned pages, protected by ashmem_mutex */
66static LIST_HEAD(ashmem_lru_list);
67
68/* Count of pages on our LRU list, protected by ashmem_mutex */
69static unsigned long lru_count;
70
71/*
72 * ashmem_mutex - protects the list of and each individual ashmem_area
73 *
74 * Lock Ordering: ashmex_mutex -> i_mutex -> i_alloc_sem
75 */
76static DEFINE_MUTEX(ashmem_mutex);
77
78static struct kmem_cache *ashmem_area_cachep __read_mostly;
79static struct kmem_cache *ashmem_range_cachep __read_mostly;
80
81#define range_size(range) \
82  ((range)->pgend - (range)->pgstart + 1)
83
84#define range_on_lru(range) \
85  ((range)->purged == ASHMEM_NOT_PURGED)
86
87#define page_range_subsumes_range(range, start, end) \
88  (((range)->pgstart >= (start)) && ((range)->pgend <= (end)))
89
90#define page_range_subsumed_by_range(range, start, end) \
91  (((range)->pgstart <= (start)) && ((range)->pgend >= (end)))
92
93#define page_in_range(range, page) \
94 (((range)->pgstart <= (page)) && ((range)->pgend >= (page)))
95
96#define page_range_in_range(range, start, end) \
97  (page_in_range(range, start) || page_in_range(range, end) || \
98   page_range_subsumes_range(range, start, end))
99
100#define range_before_page(range, page) \
101  ((range)->pgend < (page))
102
103#define PROT_MASK		(PROT_EXEC | PROT_READ | PROT_WRITE)
104
105static inline void lru_add(struct ashmem_range *range)
106{
107	list_add_tail(&range->lru, &ashmem_lru_list);
108	lru_count += range_size(range);
109}
110
111static inline void lru_del(struct ashmem_range *range)
112{
113	list_del(&range->lru);
114	lru_count -= range_size(range);
115}
116
117/*
118 * range_alloc - allocate and initialize a new ashmem_range structure
119 *
120 * 'asma' - associated ashmem_area
121 * 'prev_range' - the previous ashmem_range in the sorted asma->unpinned list
122 * 'purged' - initial purge value (ASMEM_NOT_PURGED or ASHMEM_WAS_PURGED)
123 * 'start' - starting page, inclusive
124 * 'end' - ending page, inclusive
125 *
126 * Caller must hold ashmem_mutex.
127 */
128static int range_alloc(struct ashmem_area *asma,
129		       struct ashmem_range *prev_range, unsigned int purged,
130		       size_t start, size_t end)
131{
132	struct ashmem_range *range;
133
134	range = kmem_cache_zalloc(ashmem_range_cachep, GFP_KERNEL);
135	if (unlikely(!range))
136		return -ENOMEM;
137
138	range->asma = asma;
139	range->pgstart = start;
140	range->pgend = end;
141	range->purged = purged;
142
143	list_add_tail(&range->unpinned, &prev_range->unpinned);
144
145	if (range_on_lru(range))
146		lru_add(range);
147
148	return 0;
149}
150
151static void range_del(struct ashmem_range *range)
152{
153	list_del(&range->unpinned);
154	if (range_on_lru(range))
155		lru_del(range);
156	kmem_cache_free(ashmem_range_cachep, range);
157}
158
159/*
160 * range_shrink - shrinks a range
161 *
162 * Caller must hold ashmem_mutex.
163 */
164static inline void range_shrink(struct ashmem_range *range,
165				size_t start, size_t end)
166{
167	size_t pre = range_size(range);
168
169	range->pgstart = start;
170	range->pgend = end;
171
172	if (range_on_lru(range))
173		lru_count -= pre - range_size(range);
174}
175
176static int ashmem_open(struct inode *inode, struct file *file)
177{
178	struct ashmem_area *asma;
179	int ret;
180
181	ret = nonseekable_open(inode, file);
182	if (unlikely(ret))
183		return ret;
184
185	asma = kmem_cache_zalloc(ashmem_area_cachep, GFP_KERNEL);
186	if (unlikely(!asma))
187		return -ENOMEM;
188
189	INIT_LIST_HEAD(&asma->unpinned_list);
190	memcpy(asma->name, ASHMEM_NAME_PREFIX, ASHMEM_NAME_PREFIX_LEN);
191	asma->prot_mask = PROT_MASK;
192	file->private_data = asma;
193
194	return 0;
195}
196
197static int ashmem_release(struct inode *ignored, struct file *file)
198{
199	struct ashmem_area *asma = file->private_data;
200	struct ashmem_range *range, *next;
201
202	mutex_lock(&ashmem_mutex);
203	list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned)
204		range_del(range);
205	mutex_unlock(&ashmem_mutex);
206
207	if (asma->file)
208		fput(asma->file);
209	kmem_cache_free(ashmem_area_cachep, asma);
210
211	return 0;
212}
213
214static ssize_t ashmem_read(struct file *file, char __user *buf,
215			   size_t len, loff_t *pos)
216{
217	struct ashmem_area *asma = file->private_data;
218	int ret = 0;
219
220	mutex_lock(&ashmem_mutex);
221
222	/* If size is not set, or set to 0, always return EOF. */
223	if (asma->size == 0) {
224		goto out;
225        }
226
227	if (!asma->file) {
228		ret = -EBADF;
229		goto out;
230	}
231
232	ret = asma->file->f_op->read(asma->file, buf, len, pos);
233
234out:
235	mutex_unlock(&ashmem_mutex);
236	return ret;
237}
238
239
240static int ashmem_mmap(struct file *file, struct vm_area_struct *vma)
241{
242	struct ashmem_area *asma = file->private_data;
243	int ret = 0;
244
245	mutex_lock(&ashmem_mutex);
246
247	/* user needs to SET_SIZE before mapping */
248	if (unlikely(!asma->size)) {
249		ret = -EINVAL;
250		goto out;
251	}
252
253	/* requested protection bits must match our allowed protection mask */
254	if (unlikely((vma->vm_flags & ~asma->prot_mask) & PROT_MASK)) {
255		ret = -EPERM;
256		goto out;
257	}
258
259	if (!asma->file) {
260		char *name = ASHMEM_NAME_DEF;
261		struct file *vmfile;
262
263		if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0')
264			name = asma->name;
265
266		/* ... and allocate the backing shmem file */
267		vmfile = shmem_file_setup(name, asma->size, vma->vm_flags);
268		if (unlikely(IS_ERR(vmfile))) {
269			ret = PTR_ERR(vmfile);
270			goto out;
271		}
272		asma->file = vmfile;
273	}
274	get_file(asma->file);
275
276	/*
277	 * XXX - Reworked to use shmem_zero_setup() instead of
278	 * shmem_set_file while we're in staging. -jstultz
279	 */
280	if (vma->vm_flags & VM_SHARED) {
281		ret = shmem_zero_setup(vma);
282		if (ret) {
283			fput(asma->file);
284			goto out;
285		}
286	}
287
288	if (vma->vm_file)
289		fput(vma->vm_file);
290	vma->vm_file = asma->file;
291	vma->vm_flags |= VM_CAN_NONLINEAR;
292
293out:
294	mutex_unlock(&ashmem_mutex);
295	return ret;
296}
297
298/*
299 * ashmem_shrink - our cache shrinker, called from mm/vmscan.c :: shrink_slab
300 *
301 * 'nr_to_scan' is the number of objects (pages) to prune, or 0 to query how
302 * many objects (pages) we have in total.
303 *
304 * 'gfp_mask' is the mask of the allocation that got us into this mess.
305 *
306 * Return value is the number of objects (pages) remaining, or -1 if we cannot
307 * proceed without risk of deadlock (due to gfp_mask).
308 *
309 * We approximate LRU via least-recently-unpinned, jettisoning unpinned partial
310 * chunks of ashmem regions LRU-wise one-at-a-time until we hit 'nr_to_scan'
311 * pages freed.
312 */
313static int ashmem_shrink(int nr_to_scan, gfp_t gfp_mask)
314{
315	struct ashmem_range *range, *next;
316
317	/* We might recurse into filesystem code, so bail out if necessary */
318	if (nr_to_scan && !(gfp_mask & __GFP_FS))
319		return -1;
320	if (!nr_to_scan)
321		return lru_count;
322
323	mutex_lock(&ashmem_mutex);
324	list_for_each_entry_safe(range, next, &ashmem_lru_list, lru) {
325		struct inode *inode = range->asma->file->f_dentry->d_inode;
326		loff_t start = range->pgstart * PAGE_SIZE;
327		loff_t end = (range->pgend + 1) * PAGE_SIZE - 1;
328
329		vmtruncate_range(inode, start, end);
330		range->purged = ASHMEM_WAS_PURGED;
331		lru_del(range);
332
333		nr_to_scan -= range_size(range);
334		if (nr_to_scan <= 0)
335			break;
336	}
337	mutex_unlock(&ashmem_mutex);
338
339	return lru_count;
340}
341
342static struct shrinker ashmem_shrinker = {
343	.shrink = ashmem_shrink,
344	.seeks = DEFAULT_SEEKS * 4,
345};
346
347static int set_prot_mask(struct ashmem_area *asma, unsigned long prot)
348{
349	int ret = 0;
350
351	mutex_lock(&ashmem_mutex);
352
353	/* the user can only remove, not add, protection bits */
354	if (unlikely((asma->prot_mask & prot) != prot)) {
355		ret = -EINVAL;
356		goto out;
357	}
358
359	/* does the application expect PROT_READ to imply PROT_EXEC? */
360	if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
361		prot |= PROT_EXEC;
362
363	asma->prot_mask = prot;
364
365out:
366	mutex_unlock(&ashmem_mutex);
367	return ret;
368}
369
370static int set_name(struct ashmem_area *asma, void __user *name)
371{
372	int ret = 0;
373
374	mutex_lock(&ashmem_mutex);
375
376	/* cannot change an existing mapping's name */
377	if (unlikely(asma->file)) {
378		ret = -EINVAL;
379		goto out;
380	}
381
382	if (unlikely(copy_from_user(asma->name + ASHMEM_NAME_PREFIX_LEN,
383				    name, ASHMEM_NAME_LEN)))
384		ret = -EFAULT;
385	asma->name[ASHMEM_FULL_NAME_LEN-1] = '\0';
386
387out:
388	mutex_unlock(&ashmem_mutex);
389
390	return ret;
391}
392
393static int get_name(struct ashmem_area *asma, void __user *name)
394{
395	int ret = 0;
396
397	mutex_lock(&ashmem_mutex);
398	if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0') {
399		size_t len;
400
401		/*
402		 * Copying only `len', instead of ASHMEM_NAME_LEN, bytes
403		 * prevents us from revealing one user's stack to another.
404		 */
405		len = strlen(asma->name + ASHMEM_NAME_PREFIX_LEN) + 1;
406		if (unlikely(copy_to_user(name,
407				asma->name + ASHMEM_NAME_PREFIX_LEN, len)))
408			ret = -EFAULT;
409	} else {
410		if (unlikely(copy_to_user(name, ASHMEM_NAME_DEF,
411					  sizeof(ASHMEM_NAME_DEF))))
412			ret = -EFAULT;
413	}
414	mutex_unlock(&ashmem_mutex);
415
416	return ret;
417}
418
419/*
420 * ashmem_pin - pin the given ashmem region, returning whether it was
421 * previously purged (ASHMEM_WAS_PURGED) or not (ASHMEM_NOT_PURGED).
422 *
423 * Caller must hold ashmem_mutex.
424 */
425static int ashmem_pin(struct ashmem_area *asma, size_t pgstart, size_t pgend)
426{
427	struct ashmem_range *range, *next;
428	int ret = ASHMEM_NOT_PURGED;
429
430	list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) {
431		/* moved past last applicable page; we can short circuit */
432		if (range_before_page(range, pgstart))
433			break;
434
435		/*
436		 * The user can ask us to pin pages that span multiple ranges,
437		 * or to pin pages that aren't even unpinned, so this is messy.
438		 *
439		 * Four cases:
440		 * 1. The requested range subsumes an existing range, so we
441		 *    just remove the entire matching range.
442		 * 2. The requested range overlaps the start of an existing
443		 *    range, so we just update that range.
444		 * 3. The requested range overlaps the end of an existing
445		 *    range, so we just update that range.
446		 * 4. The requested range punches a hole in an existing range,
447		 *    so we have to update one side of the range and then
448		 *    create a new range for the other side.
449		 */
450		if (page_range_in_range(range, pgstart, pgend)) {
451			ret |= range->purged;
452
453			/* Case #1: Easy. Just nuke the whole thing. */
454			if (page_range_subsumes_range(range, pgstart, pgend)) {
455				range_del(range);
456				continue;
457			}
458
459			/* Case #2: We overlap from the start, so adjust it */
460			if (range->pgstart >= pgstart) {
461				range_shrink(range, pgend + 1, range->pgend);
462				continue;
463			}
464
465			/* Case #3: We overlap from the rear, so adjust it */
466			if (range->pgend <= pgend) {
467				range_shrink(range, range->pgstart, pgstart-1);
468				continue;
469			}
470
471			/*
472			 * Case #4: We eat a chunk out of the middle. A bit
473			 * more complicated, we allocate a new range for the
474			 * second half and adjust the first chunk's endpoint.
475			 */
476			range_alloc(asma, range, range->purged,
477				    pgend + 1, range->pgend);
478			range_shrink(range, range->pgstart, pgstart - 1);
479			break;
480		}
481	}
482
483	return ret;
484}
485
486/*
487 * ashmem_unpin - unpin the given range of pages. Returns zero on success.
488 *
489 * Caller must hold ashmem_mutex.
490 */
491static int ashmem_unpin(struct ashmem_area *asma, size_t pgstart, size_t pgend)
492{
493	struct ashmem_range *range, *next;
494	unsigned int purged = ASHMEM_NOT_PURGED;
495
496restart:
497	list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) {
498		/* short circuit: this is our insertion point */
499		if (range_before_page(range, pgstart))
500			break;
501
502		/*
503		 * The user can ask us to unpin pages that are already entirely
504		 * or partially pinned. We handle those two cases here.
505		 */
506		if (page_range_subsumed_by_range(range, pgstart, pgend))
507			return 0;
508		if (page_range_in_range(range, pgstart, pgend)) {
509			pgstart = min_t(size_t, range->pgstart, pgstart),
510			pgend = max_t(size_t, range->pgend, pgend);
511			purged |= range->purged;
512			range_del(range);
513			goto restart;
514		}
515	}
516
517	return range_alloc(asma, range, purged, pgstart, pgend);
518}
519
520/*
521 * ashmem_get_pin_status - Returns ASHMEM_IS_UNPINNED if _any_ pages in the
522 * given interval are unpinned and ASHMEM_IS_PINNED otherwise.
523 *
524 * Caller must hold ashmem_mutex.
525 */
526static int ashmem_get_pin_status(struct ashmem_area *asma, size_t pgstart,
527				 size_t pgend)
528{
529	struct ashmem_range *range;
530	int ret = ASHMEM_IS_PINNED;
531
532	list_for_each_entry(range, &asma->unpinned_list, unpinned) {
533		if (range_before_page(range, pgstart))
534			break;
535		if (page_range_in_range(range, pgstart, pgend)) {
536			ret = ASHMEM_IS_UNPINNED;
537			break;
538		}
539	}
540
541	return ret;
542}
543
544static int ashmem_pin_unpin(struct ashmem_area *asma, unsigned long cmd,
545			    void __user *p)
546{
547	struct ashmem_pin pin;
548	size_t pgstart, pgend;
549	int ret = -EINVAL;
550
551	if (unlikely(!asma->file))
552		return -EINVAL;
553
554	if (unlikely(copy_from_user(&pin, p, sizeof(pin))))
555		return -EFAULT;
556
557	/* per custom, you can pass zero for len to mean "everything onward" */
558	if (!pin.len)
559		pin.len = PAGE_ALIGN(asma->size) - pin.offset;
560
561	if (unlikely((pin.offset | pin.len) & ~PAGE_MASK))
562		return -EINVAL;
563
564	if (unlikely(((__u32) -1) - pin.offset < pin.len))
565		return -EINVAL;
566
567	if (unlikely(PAGE_ALIGN(asma->size) < pin.offset + pin.len))
568		return -EINVAL;
569
570	pgstart = pin.offset / PAGE_SIZE;
571	pgend = pgstart + (pin.len / PAGE_SIZE) - 1;
572
573	mutex_lock(&ashmem_mutex);
574
575	switch (cmd) {
576	case ASHMEM_PIN:
577		ret = ashmem_pin(asma, pgstart, pgend);
578		break;
579	case ASHMEM_UNPIN:
580		ret = ashmem_unpin(asma, pgstart, pgend);
581		break;
582	case ASHMEM_GET_PIN_STATUS:
583		ret = ashmem_get_pin_status(asma, pgstart, pgend);
584		break;
585	}
586
587	mutex_unlock(&ashmem_mutex);
588
589	return ret;
590}
591
592static long ashmem_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
593{
594	struct ashmem_area *asma = file->private_data;
595	long ret = -ENOTTY;
596
597	switch (cmd) {
598	case ASHMEM_SET_NAME:
599		ret = set_name(asma, (void __user *) arg);
600		break;
601	case ASHMEM_GET_NAME:
602		ret = get_name(asma, (void __user *) arg);
603		break;
604	case ASHMEM_SET_SIZE:
605		ret = -EINVAL;
606		if (!asma->file) {
607			ret = 0;
608			asma->size = (size_t) arg;
609		}
610		break;
611	case ASHMEM_GET_SIZE:
612		ret = asma->size;
613		break;
614	case ASHMEM_SET_PROT_MASK:
615		ret = set_prot_mask(asma, arg);
616		break;
617	case ASHMEM_GET_PROT_MASK:
618		ret = asma->prot_mask;
619		break;
620	case ASHMEM_PIN:
621	case ASHMEM_UNPIN:
622	case ASHMEM_GET_PIN_STATUS:
623		ret = ashmem_pin_unpin(asma, cmd, (void __user *) arg);
624		break;
625	case ASHMEM_PURGE_ALL_CACHES:
626		ret = -EPERM;
627		if (capable(CAP_SYS_ADMIN)) {
628			ret = ashmem_shrink(0, GFP_KERNEL);
629			ashmem_shrink(ret, GFP_KERNEL);
630		}
631		break;
632	}
633
634	return ret;
635}
636
637static struct file_operations ashmem_fops = {
638	.owner = THIS_MODULE,
639	.open = ashmem_open,
640	.release = ashmem_release,
641        .read = ashmem_read,
642	.mmap = ashmem_mmap,
643	.unlocked_ioctl = ashmem_ioctl,
644	.compat_ioctl = ashmem_ioctl,
645};
646
647static struct miscdevice ashmem_misc = {
648	.minor = MISC_DYNAMIC_MINOR,
649	.name = "ashmem",
650	.fops = &ashmem_fops,
651};
652
653static int __init ashmem_init(void)
654{
655	int ret;
656
657	ashmem_area_cachep = kmem_cache_create("ashmem_area_cache",
658					  sizeof(struct ashmem_area),
659					  0, 0, NULL);
660	if (unlikely(!ashmem_area_cachep)) {
661		printk(KERN_ERR "ashmem: failed to create slab cache\n");
662		return -ENOMEM;
663	}
664
665	ashmem_range_cachep = kmem_cache_create("ashmem_range_cache",
666					  sizeof(struct ashmem_range),
667					  0, 0, NULL);
668	if (unlikely(!ashmem_range_cachep)) {
669		printk(KERN_ERR "ashmem: failed to create slab cache\n");
670		return -ENOMEM;
671	}
672
673	ret = misc_register(&ashmem_misc);
674	if (unlikely(ret)) {
675		printk(KERN_ERR "ashmem: failed to register misc device!\n");
676		return ret;
677	}
678
679	register_shrinker(&ashmem_shrinker);
680
681	printk(KERN_INFO "ashmem: initialized\n");
682
683	return 0;
684}
685
686static void __exit ashmem_exit(void)
687{
688	int ret;
689
690	unregister_shrinker(&ashmem_shrinker);
691
692	ret = misc_deregister(&ashmem_misc);
693	if (unlikely(ret))
694		printk(KERN_ERR "ashmem: failed to unregister misc device!\n");
695
696	kmem_cache_destroy(ashmem_range_cachep);
697	kmem_cache_destroy(ashmem_area_cachep);
698
699	printk(KERN_INFO "ashmem: unloaded\n");
700}
701
702module_init(ashmem_init);
703module_exit(ashmem_exit);
704
705MODULE_LICENSE("GPL");
706