1
2
3
4
5
6
7
8
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
35
36
37
38
39
40
41
42
43
44
45
46struct 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
56
57
58
59
60
61
62
63
64
65
66struct 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
76static LIST_HEAD(ashmem_lru_list);
77
78static atomic_t ashmem_shrink_inflight = ATOMIC_INIT(0);
79static DECLARE_WAIT_QUEUE_HEAD(ashmem_shrink_wait);
80
81
82
83
84
85
86static unsigned long lru_count;
87
88
89
90
91
92
93static DEFINE_MUTEX(ashmem_mutex);
94
95static struct kmem_cache *ashmem_area_cachep __read_mostly;
96static struct kmem_cache *ashmem_range_cachep __read_mostly;
97
98
99
100
101
102
103
104
105static struct lock_class_key backing_shmem_inode_class;
106
107static inline unsigned long range_size(struct ashmem_range *range)
108{
109 return range->pgend - range->pgstart + 1;
110}
111
112static inline bool range_on_lru(struct ashmem_range *range)
113{
114 return range->purged == ASHMEM_NOT_PURGED;
115}
116
117static 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
123static 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
129static inline bool page_in_range(struct ashmem_range *range, size_t page)
130{
131 return (range->pgstart <= page) && (range->pgend >= page);
132}
133
134static 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
141static 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
151
152
153
154
155
156static 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
164
165
166
167
168
169static inline void lru_del(struct ashmem_range *range)
170{
171 list_del(&range->lru);
172 lru_count -= range_size(range);
173}
174
175
176
177
178
179
180
181
182
183
184
185
186static void range_alloc(struct ashmem_area *asma,
187 struct ashmem_range *prev_range, unsigned int purged,
188 size_t start, size_t end,
189 struct ashmem_range **new_range)
190{
191 struct ashmem_range *range = *new_range;
192
193 *new_range = NULL;
194 range->asma = asma;
195 range->pgstart = start;
196 range->pgend = end;
197 range->purged = purged;
198
199 list_add_tail(&range->unpinned, &prev_range->unpinned);
200
201 if (range_on_lru(range))
202 lru_add(range);
203}
204
205
206
207
208
209static void range_del(struct ashmem_range *range)
210{
211 list_del(&range->unpinned);
212 if (range_on_lru(range))
213 lru_del(range);
214 kmem_cache_free(ashmem_range_cachep, range);
215}
216
217
218
219
220
221
222
223
224
225
226
227
228
229static inline void range_shrink(struct ashmem_range *range,
230 size_t start, size_t end)
231{
232 size_t pre = range_size(range);
233
234 range->pgstart = start;
235 range->pgend = end;
236
237 if (range_on_lru(range))
238 lru_count -= pre - range_size(range);
239}
240
241
242
243
244
245
246
247
248
249
250
251static int ashmem_open(struct inode *inode, struct file *file)
252{
253 struct ashmem_area *asma;
254 int ret;
255
256 ret = generic_file_open(inode, file);
257 if (ret)
258 return ret;
259
260 asma = kmem_cache_zalloc(ashmem_area_cachep, GFP_KERNEL);
261 if (!asma)
262 return -ENOMEM;
263
264 INIT_LIST_HEAD(&asma->unpinned_list);
265 memcpy(asma->name, ASHMEM_NAME_PREFIX, ASHMEM_NAME_PREFIX_LEN);
266 asma->prot_mask = PROT_MASK;
267 file->private_data = asma;
268
269 return 0;
270}
271
272
273
274
275
276
277
278
279
280static int ashmem_release(struct inode *ignored, struct file *file)
281{
282 struct ashmem_area *asma = file->private_data;
283 struct ashmem_range *range, *next;
284
285 mutex_lock(&ashmem_mutex);
286 list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned)
287 range_del(range);
288 mutex_unlock(&ashmem_mutex);
289
290 if (asma->file)
291 fput(asma->file);
292 kmem_cache_free(ashmem_area_cachep, asma);
293
294 return 0;
295}
296
297static ssize_t ashmem_read_iter(struct kiocb *iocb, struct iov_iter *iter)
298{
299 struct ashmem_area *asma = iocb->ki_filp->private_data;
300 int ret = 0;
301
302 mutex_lock(&ashmem_mutex);
303
304
305 if (asma->size == 0)
306 goto out_unlock;
307
308 if (!asma->file) {
309 ret = -EBADF;
310 goto out_unlock;
311 }
312
313
314
315
316
317
318
319 mutex_unlock(&ashmem_mutex);
320 ret = vfs_iter_read(asma->file, iter, &iocb->ki_pos, 0);
321 mutex_lock(&ashmem_mutex);
322 if (ret > 0)
323 asma->file->f_pos = iocb->ki_pos;
324out_unlock:
325 mutex_unlock(&ashmem_mutex);
326 return ret;
327}
328
329static loff_t ashmem_llseek(struct file *file, loff_t offset, int origin)
330{
331 struct ashmem_area *asma = file->private_data;
332 loff_t ret;
333
334 mutex_lock(&ashmem_mutex);
335
336 if (asma->size == 0) {
337 mutex_unlock(&ashmem_mutex);
338 return -EINVAL;
339 }
340
341 if (!asma->file) {
342 mutex_unlock(&ashmem_mutex);
343 return -EBADF;
344 }
345
346 mutex_unlock(&ashmem_mutex);
347
348 ret = vfs_llseek(asma->file, offset, origin);
349 if (ret < 0)
350 return ret;
351
352
353 file->f_pos = asma->file->f_pos;
354 return ret;
355}
356
357static inline vm_flags_t calc_vm_may_flags(unsigned long prot)
358{
359 return _calc_vm_trans(prot, PROT_READ, VM_MAYREAD) |
360 _calc_vm_trans(prot, PROT_WRITE, VM_MAYWRITE) |
361 _calc_vm_trans(prot, PROT_EXEC, VM_MAYEXEC);
362}
363
364static int ashmem_vmfile_mmap(struct file *file, struct vm_area_struct *vma)
365{
366
367 return -EPERM;
368}
369
370static unsigned long
371ashmem_vmfile_get_unmapped_area(struct file *file, unsigned long addr,
372 unsigned long len, unsigned long pgoff,
373 unsigned long flags)
374{
375 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
376}
377
378static int ashmem_mmap(struct file *file, struct vm_area_struct *vma)
379{
380 static struct file_operations vmfile_fops;
381 struct ashmem_area *asma = file->private_data;
382 int ret = 0;
383
384 mutex_lock(&ashmem_mutex);
385
386
387 if (!asma->size) {
388 ret = -EINVAL;
389 goto out;
390 }
391
392
393 if (vma->vm_end - vma->vm_start > PAGE_ALIGN(asma->size)) {
394 ret = -EINVAL;
395 goto out;
396 }
397
398
399 if ((vma->vm_flags & ~calc_vm_prot_bits(asma->prot_mask, 0)) &
400 calc_vm_prot_bits(PROT_MASK, 0)) {
401 ret = -EPERM;
402 goto out;
403 }
404 vma->vm_flags &= ~calc_vm_may_flags(~asma->prot_mask);
405
406 if (!asma->file) {
407 char *name = ASHMEM_NAME_DEF;
408 struct file *vmfile;
409 struct inode *inode;
410
411 if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0')
412 name = asma->name;
413
414
415 vmfile = shmem_file_setup(name, asma->size, vma->vm_flags);
416 if (IS_ERR(vmfile)) {
417 ret = PTR_ERR(vmfile);
418 goto out;
419 }
420 vmfile->f_mode |= FMODE_LSEEK;
421 inode = file_inode(vmfile);
422 lockdep_set_class(&inode->i_rwsem, &backing_shmem_inode_class);
423 asma->file = vmfile;
424
425
426
427
428
429
430 if (!vmfile_fops.mmap) {
431 vmfile_fops = *vmfile->f_op;
432 vmfile_fops.mmap = ashmem_vmfile_mmap;
433 vmfile_fops.get_unmapped_area =
434 ashmem_vmfile_get_unmapped_area;
435 }
436 vmfile->f_op = &vmfile_fops;
437 }
438 get_file(asma->file);
439
440
441
442
443
444 if (vma->vm_flags & VM_SHARED) {
445 ret = shmem_zero_setup(vma);
446 if (ret) {
447 fput(asma->file);
448 goto out;
449 }
450 } else {
451 vma_set_anonymous(vma);
452 }
453
454 vma_set_file(vma, asma->file);
455
456 fput(asma->file);
457
458out:
459 mutex_unlock(&ashmem_mutex);
460 return ret;
461}
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477static unsigned long
478ashmem_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
479{
480 unsigned long freed = 0;
481
482
483 if (!(sc->gfp_mask & __GFP_FS))
484 return SHRINK_STOP;
485
486 if (!mutex_trylock(&ashmem_mutex))
487 return -1;
488
489 while (!list_empty(&ashmem_lru_list)) {
490 struct ashmem_range *range =
491 list_first_entry(&ashmem_lru_list, typeof(*range), lru);
492 loff_t start = range->pgstart * PAGE_SIZE;
493 loff_t end = (range->pgend + 1) * PAGE_SIZE;
494 struct file *f = range->asma->file;
495
496 get_file(f);
497 atomic_inc(&ashmem_shrink_inflight);
498 range->purged = ASHMEM_WAS_PURGED;
499 lru_del(range);
500
501 freed += range_size(range);
502 mutex_unlock(&ashmem_mutex);
503 f->f_op->fallocate(f,
504 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
505 start, end - start);
506 fput(f);
507 if (atomic_dec_and_test(&ashmem_shrink_inflight))
508 wake_up_all(&ashmem_shrink_wait);
509 if (!mutex_trylock(&ashmem_mutex))
510 goto out;
511 if (--sc->nr_to_scan <= 0)
512 break;
513 }
514 mutex_unlock(&ashmem_mutex);
515out:
516 return freed;
517}
518
519static unsigned long
520ashmem_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
521{
522
523
524
525
526
527 return lru_count;
528}
529
530static struct shrinker ashmem_shrinker = {
531 .count_objects = ashmem_shrink_count,
532 .scan_objects = ashmem_shrink_scan,
533
534
535
536
537 .seeks = DEFAULT_SEEKS * 4,
538};
539
540static int set_prot_mask(struct ashmem_area *asma, unsigned long prot)
541{
542 int ret = 0;
543
544 mutex_lock(&ashmem_mutex);
545
546
547 if ((asma->prot_mask & prot) != prot) {
548 ret = -EINVAL;
549 goto out;
550 }
551
552
553 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
554 prot |= PROT_EXEC;
555
556 asma->prot_mask = prot;
557
558out:
559 mutex_unlock(&ashmem_mutex);
560 return ret;
561}
562
563static int set_name(struct ashmem_area *asma, void __user *name)
564{
565 int len;
566 int ret = 0;
567 char local_name[ASHMEM_NAME_LEN];
568
569
570
571
572
573
574
575
576
577
578 len = strncpy_from_user(local_name, name, ASHMEM_NAME_LEN);
579 if (len < 0)
580 return len;
581
582 mutex_lock(&ashmem_mutex);
583
584 if (asma->file)
585 ret = -EINVAL;
586 else
587 strscpy(asma->name + ASHMEM_NAME_PREFIX_LEN, local_name,
588 ASHMEM_NAME_LEN);
589
590 mutex_unlock(&ashmem_mutex);
591 return ret;
592}
593
594static int get_name(struct ashmem_area *asma, void __user *name)
595{
596 int ret = 0;
597 size_t len;
598
599
600
601
602
603
604 char local_name[ASHMEM_NAME_LEN];
605
606 mutex_lock(&ashmem_mutex);
607 if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0') {
608
609
610
611
612 len = strlen(asma->name + ASHMEM_NAME_PREFIX_LEN) + 1;
613 memcpy(local_name, asma->name + ASHMEM_NAME_PREFIX_LEN, len);
614 } else {
615 len = sizeof(ASHMEM_NAME_DEF);
616 memcpy(local_name, ASHMEM_NAME_DEF, len);
617 }
618 mutex_unlock(&ashmem_mutex);
619
620
621
622
623
624 if (copy_to_user(name, local_name, len))
625 ret = -EFAULT;
626 return ret;
627}
628
629
630
631
632
633
634
635static int ashmem_pin(struct ashmem_area *asma, size_t pgstart, size_t pgend,
636 struct ashmem_range **new_range)
637{
638 struct ashmem_range *range, *next;
639 int ret = ASHMEM_NOT_PURGED;
640
641 list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) {
642
643 if (range_before_page(range, pgstart))
644 break;
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661 if (page_range_in_range(range, pgstart, pgend)) {
662 ret |= range->purged;
663
664
665 if (page_range_subsumes_range(range, pgstart, pgend)) {
666 range_del(range);
667 continue;
668 }
669
670
671 if (range->pgstart >= pgstart) {
672 range_shrink(range, pgend + 1, range->pgend);
673 continue;
674 }
675
676
677 if (range->pgend <= pgend) {
678 range_shrink(range, range->pgstart,
679 pgstart - 1);
680 continue;
681 }
682
683
684
685
686
687
688 range_alloc(asma, range, range->purged,
689 pgend + 1, range->pgend, new_range);
690 range_shrink(range, range->pgstart, pgstart - 1);
691 break;
692 }
693 }
694
695 return ret;
696}
697
698
699
700
701
702
703static int ashmem_unpin(struct ashmem_area *asma, size_t pgstart, size_t pgend,
704 struct ashmem_range **new_range)
705{
706 struct ashmem_range *range, *next;
707 unsigned int purged = ASHMEM_NOT_PURGED;
708
709restart:
710 list_for_each_entry_safe(range, next, &asma->unpinned_list, unpinned) {
711
712 if (range_before_page(range, pgstart))
713 break;
714
715
716
717
718
719 if (page_range_subsumed_by_range(range, pgstart, pgend))
720 return 0;
721 if (page_range_in_range(range, pgstart, pgend)) {
722 pgstart = min(range->pgstart, pgstart);
723 pgend = max(range->pgend, pgend);
724 purged |= range->purged;
725 range_del(range);
726 goto restart;
727 }
728 }
729
730 range_alloc(asma, range, purged, pgstart, pgend, new_range);
731 return 0;
732}
733
734
735
736
737
738
739
740static int ashmem_get_pin_status(struct ashmem_area *asma, size_t pgstart,
741 size_t pgend)
742{
743 struct ashmem_range *range;
744 int ret = ASHMEM_IS_PINNED;
745
746 list_for_each_entry(range, &asma->unpinned_list, unpinned) {
747 if (range_before_page(range, pgstart))
748 break;
749 if (page_range_in_range(range, pgstart, pgend)) {
750 ret = ASHMEM_IS_UNPINNED;
751 break;
752 }
753 }
754
755 return ret;
756}
757
758static int ashmem_pin_unpin(struct ashmem_area *asma, unsigned long cmd,
759 void __user *p)
760{
761 struct ashmem_pin pin;
762 size_t pgstart, pgend;
763 int ret = -EINVAL;
764 struct ashmem_range *range = NULL;
765
766 if (copy_from_user(&pin, p, sizeof(pin)))
767 return -EFAULT;
768
769 if (cmd == ASHMEM_PIN || cmd == ASHMEM_UNPIN) {
770 range = kmem_cache_zalloc(ashmem_range_cachep, GFP_KERNEL);
771 if (!range)
772 return -ENOMEM;
773 }
774
775 mutex_lock(&ashmem_mutex);
776 wait_event(ashmem_shrink_wait, !atomic_read(&ashmem_shrink_inflight));
777
778 if (!asma->file)
779 goto out_unlock;
780
781
782 if (!pin.len)
783 pin.len = PAGE_ALIGN(asma->size) - pin.offset;
784
785 if ((pin.offset | pin.len) & ~PAGE_MASK)
786 goto out_unlock;
787
788 if (((__u32)-1) - pin.offset < pin.len)
789 goto out_unlock;
790
791 if (PAGE_ALIGN(asma->size) < pin.offset + pin.len)
792 goto out_unlock;
793
794 pgstart = pin.offset / PAGE_SIZE;
795 pgend = pgstart + (pin.len / PAGE_SIZE) - 1;
796
797 switch (cmd) {
798 case ASHMEM_PIN:
799 ret = ashmem_pin(asma, pgstart, pgend, &range);
800 break;
801 case ASHMEM_UNPIN:
802 ret = ashmem_unpin(asma, pgstart, pgend, &range);
803 break;
804 case ASHMEM_GET_PIN_STATUS:
805 ret = ashmem_get_pin_status(asma, pgstart, pgend);
806 break;
807 }
808
809out_unlock:
810 mutex_unlock(&ashmem_mutex);
811 if (range)
812 kmem_cache_free(ashmem_range_cachep, range);
813
814 return ret;
815}
816
817static long ashmem_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
818{
819 struct ashmem_area *asma = file->private_data;
820 long ret = -ENOTTY;
821
822 switch (cmd) {
823 case ASHMEM_SET_NAME:
824 ret = set_name(asma, (void __user *)arg);
825 break;
826 case ASHMEM_GET_NAME:
827 ret = get_name(asma, (void __user *)arg);
828 break;
829 case ASHMEM_SET_SIZE:
830 ret = -EINVAL;
831 mutex_lock(&ashmem_mutex);
832 if (!asma->file) {
833 ret = 0;
834 asma->size = (size_t)arg;
835 }
836 mutex_unlock(&ashmem_mutex);
837 break;
838 case ASHMEM_GET_SIZE:
839 ret = asma->size;
840 break;
841 case ASHMEM_SET_PROT_MASK:
842 ret = set_prot_mask(asma, arg);
843 break;
844 case ASHMEM_GET_PROT_MASK:
845 ret = asma->prot_mask;
846 break;
847 case ASHMEM_PIN:
848 case ASHMEM_UNPIN:
849 case ASHMEM_GET_PIN_STATUS:
850 ret = ashmem_pin_unpin(asma, cmd, (void __user *)arg);
851 break;
852 case ASHMEM_PURGE_ALL_CACHES:
853 ret = -EPERM;
854 if (capable(CAP_SYS_ADMIN)) {
855 struct shrink_control sc = {
856 .gfp_mask = GFP_KERNEL,
857 .nr_to_scan = LONG_MAX,
858 };
859 ret = ashmem_shrink_count(&ashmem_shrinker, &sc);
860 ashmem_shrink_scan(&ashmem_shrinker, &sc);
861 }
862 break;
863 }
864
865 return ret;
866}
867
868
869#ifdef CONFIG_COMPAT
870static long compat_ashmem_ioctl(struct file *file, unsigned int cmd,
871 unsigned long arg)
872{
873 switch (cmd) {
874 case COMPAT_ASHMEM_SET_SIZE:
875 cmd = ASHMEM_SET_SIZE;
876 break;
877 case COMPAT_ASHMEM_SET_PROT_MASK:
878 cmd = ASHMEM_SET_PROT_MASK;
879 break;
880 }
881 return ashmem_ioctl(file, cmd, arg);
882}
883#endif
884#ifdef CONFIG_PROC_FS
885static void ashmem_show_fdinfo(struct seq_file *m, struct file *file)
886{
887 struct ashmem_area *asma = file->private_data;
888
889 mutex_lock(&ashmem_mutex);
890
891 if (asma->file)
892 seq_printf(m, "inode:\t%ld\n", file_inode(asma->file)->i_ino);
893
894 if (asma->name[ASHMEM_NAME_PREFIX_LEN] != '\0')
895 seq_printf(m, "name:\t%s\n",
896 asma->name + ASHMEM_NAME_PREFIX_LEN);
897
898 seq_printf(m, "size:\t%zu\n", asma->size);
899
900 mutex_unlock(&ashmem_mutex);
901}
902#endif
903static const struct file_operations ashmem_fops = {
904 .owner = THIS_MODULE,
905 .open = ashmem_open,
906 .release = ashmem_release,
907 .read_iter = ashmem_read_iter,
908 .llseek = ashmem_llseek,
909 .mmap = ashmem_mmap,
910 .unlocked_ioctl = ashmem_ioctl,
911#ifdef CONFIG_COMPAT
912 .compat_ioctl = compat_ashmem_ioctl,
913#endif
914#ifdef CONFIG_PROC_FS
915 .show_fdinfo = ashmem_show_fdinfo,
916#endif
917};
918
919static struct miscdevice ashmem_misc = {
920 .minor = MISC_DYNAMIC_MINOR,
921 .name = "ashmem",
922 .fops = &ashmem_fops,
923};
924
925static int __init ashmem_init(void)
926{
927 int ret = -ENOMEM;
928
929 ashmem_area_cachep = kmem_cache_create("ashmem_area_cache",
930 sizeof(struct ashmem_area),
931 0, 0, NULL);
932 if (!ashmem_area_cachep) {
933 pr_err("failed to create slab cache\n");
934 goto out;
935 }
936
937 ashmem_range_cachep = kmem_cache_create("ashmem_range_cache",
938 sizeof(struct ashmem_range),
939 0, SLAB_RECLAIM_ACCOUNT, NULL);
940 if (!ashmem_range_cachep) {
941 pr_err("failed to create slab cache\n");
942 goto out_free1;
943 }
944
945 ret = misc_register(&ashmem_misc);
946 if (ret) {
947 pr_err("failed to register misc device!\n");
948 goto out_free2;
949 }
950
951 ret = register_shrinker(&ashmem_shrinker);
952 if (ret) {
953 pr_err("failed to register shrinker!\n");
954 goto out_demisc;
955 }
956
957 pr_info("initialized\n");
958
959 return 0;
960
961out_demisc:
962 misc_deregister(&ashmem_misc);
963out_free2:
964 kmem_cache_destroy(ashmem_range_cachep);
965out_free1:
966 kmem_cache_destroy(ashmem_area_cachep);
967out:
968 return ret;
969}
970device_initcall(ashmem_init);
971