2 * f2fs extent cache support
4 * Copyright (c) 2015 Motorola Mobility
5 * Copyright (c) 2015 Samsung Electronics
6 * Authors: Jaegeuk Kim <jaegeuk@kernel.org>
7 * Chao Yu <chao2.yu@samsung.com>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
15 #include <linux/f2fs_fs.h>
19 #include <trace/events/f2fs.h>
21 static struct kmem_cache
*extent_tree_slab
;
22 static struct kmem_cache
*extent_node_slab
;
24 static struct extent_node
*__attach_extent_node(struct f2fs_sb_info
*sbi
,
25 struct extent_tree
*et
, struct extent_info
*ei
,
26 struct rb_node
*parent
, struct rb_node
**p
)
28 struct extent_node
*en
;
30 en
= kmem_cache_alloc(extent_node_slab
, GFP_ATOMIC
);
35 INIT_LIST_HEAD(&en
->list
);
38 rb_link_node(&en
->rb_node
, parent
, p
);
39 rb_insert_color(&en
->rb_node
, &et
->root
);
40 atomic_inc(&et
->node_cnt
);
41 atomic_inc(&sbi
->total_ext_node
);
45 static void __detach_extent_node(struct f2fs_sb_info
*sbi
,
46 struct extent_tree
*et
, struct extent_node
*en
)
48 rb_erase(&en
->rb_node
, &et
->root
);
49 atomic_dec(&et
->node_cnt
);
50 atomic_dec(&sbi
->total_ext_node
);
52 if (et
->cached_en
== en
)
54 kmem_cache_free(extent_node_slab
, en
);
58 * Flow to release an extent_node:
60 * 2. __detach_extent_node
63 static void __release_extent_node(struct f2fs_sb_info
*sbi
,
64 struct extent_tree
*et
, struct extent_node
*en
)
66 spin_lock(&sbi
->extent_lock
);
67 f2fs_bug_on(sbi
, list_empty(&en
->list
));
68 list_del_init(&en
->list
);
69 spin_unlock(&sbi
->extent_lock
);
71 __detach_extent_node(sbi
, et
, en
);
74 static struct extent_tree
*__grab_extent_tree(struct inode
*inode
)
76 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
77 struct extent_tree
*et
;
78 nid_t ino
= inode
->i_ino
;
80 down_write(&sbi
->extent_tree_lock
);
81 et
= radix_tree_lookup(&sbi
->extent_tree_root
, ino
);
83 et
= f2fs_kmem_cache_alloc(extent_tree_slab
, GFP_NOFS
);
84 f2fs_radix_tree_insert(&sbi
->extent_tree_root
, ino
, et
);
85 memset(et
, 0, sizeof(struct extent_tree
));
89 rwlock_init(&et
->lock
);
90 INIT_LIST_HEAD(&et
->list
);
91 atomic_set(&et
->node_cnt
, 0);
92 atomic_inc(&sbi
->total_ext_tree
);
94 atomic_dec(&sbi
->total_zombie_tree
);
95 list_del_init(&et
->list
);
97 up_write(&sbi
->extent_tree_lock
);
99 /* never died until evict_inode */
100 F2FS_I(inode
)->extent_tree
= et
;
105 static struct extent_node
*__lookup_extent_tree(struct f2fs_sb_info
*sbi
,
106 struct extent_tree
*et
, unsigned int fofs
)
108 struct rb_node
*node
= et
->root
.rb_node
;
109 struct extent_node
*en
= et
->cached_en
;
112 struct extent_info
*cei
= &en
->ei
;
114 if (cei
->fofs
<= fofs
&& cei
->fofs
+ cei
->len
> fofs
) {
115 stat_inc_cached_node_hit(sbi
);
121 en
= rb_entry(node
, struct extent_node
, rb_node
);
123 if (fofs
< en
->ei
.fofs
) {
124 node
= node
->rb_left
;
125 } else if (fofs
>= en
->ei
.fofs
+ en
->ei
.len
) {
126 node
= node
->rb_right
;
128 stat_inc_rbtree_node_hit(sbi
);
135 static struct extent_node
*__init_extent_tree(struct f2fs_sb_info
*sbi
,
136 struct extent_tree
*et
, struct extent_info
*ei
)
138 struct rb_node
**p
= &et
->root
.rb_node
;
139 struct extent_node
*en
;
141 en
= __attach_extent_node(sbi
, et
, ei
, NULL
, p
);
145 et
->largest
= en
->ei
;
150 static unsigned int __free_extent_tree(struct f2fs_sb_info
*sbi
,
151 struct extent_tree
*et
)
153 struct rb_node
*node
, *next
;
154 struct extent_node
*en
;
155 unsigned int count
= atomic_read(&et
->node_cnt
);
157 node
= rb_first(&et
->root
);
159 next
= rb_next(node
);
160 en
= rb_entry(node
, struct extent_node
, rb_node
);
161 __release_extent_node(sbi
, et
, en
);
165 return count
- atomic_read(&et
->node_cnt
);
168 static void __drop_largest_extent(struct inode
*inode
,
169 pgoff_t fofs
, unsigned int len
)
171 struct extent_info
*largest
= &F2FS_I(inode
)->extent_tree
->largest
;
173 if (fofs
< largest
->fofs
+ largest
->len
&& fofs
+ len
> largest
->fofs
) {
175 f2fs_mark_inode_dirty_sync(inode
);
179 /* return true, if inode page is changed */
180 bool f2fs_init_extent_tree(struct inode
*inode
, struct f2fs_extent
*i_ext
)
182 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
183 struct extent_tree
*et
;
184 struct extent_node
*en
;
185 struct extent_info ei
;
187 if (!f2fs_may_extent_tree(inode
)) {
188 /* drop largest extent */
189 if (i_ext
&& i_ext
->len
) {
196 et
= __grab_extent_tree(inode
);
198 if (!i_ext
|| !i_ext
->len
)
201 get_extent_info(&ei
, i_ext
);
203 write_lock(&et
->lock
);
204 if (atomic_read(&et
->node_cnt
))
207 en
= __init_extent_tree(sbi
, et
, &ei
);
209 spin_lock(&sbi
->extent_lock
);
210 list_add_tail(&en
->list
, &sbi
->extent_list
);
211 spin_unlock(&sbi
->extent_lock
);
214 write_unlock(&et
->lock
);
218 static bool f2fs_lookup_extent_tree(struct inode
*inode
, pgoff_t pgofs
,
219 struct extent_info
*ei
)
221 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
222 struct extent_tree
*et
= F2FS_I(inode
)->extent_tree
;
223 struct extent_node
*en
;
226 f2fs_bug_on(sbi
, !et
);
228 trace_f2fs_lookup_extent_tree_start(inode
, pgofs
);
230 read_lock(&et
->lock
);
232 if (et
->largest
.fofs
<= pgofs
&&
233 et
->largest
.fofs
+ et
->largest
.len
> pgofs
) {
236 stat_inc_largest_node_hit(sbi
);
240 en
= __lookup_extent_tree(sbi
, et
, pgofs
);
243 spin_lock(&sbi
->extent_lock
);
244 if (!list_empty(&en
->list
)) {
245 list_move_tail(&en
->list
, &sbi
->extent_list
);
248 spin_unlock(&sbi
->extent_lock
);
252 stat_inc_total_hit(sbi
);
253 read_unlock(&et
->lock
);
255 trace_f2fs_lookup_extent_tree_end(inode
, pgofs
, ei
);
261 * lookup extent at @fofs, if hit, return the extent
262 * if not, return NULL and
263 * @prev_ex: extent before fofs
264 * @next_ex: extent after fofs
265 * @insert_p: insert point for new extent at fofs
266 * in order to simpfy the insertion after.
267 * tree must stay unchanged between lookup and insertion.
269 static struct extent_node
*__lookup_extent_tree_ret(struct extent_tree
*et
,
271 struct extent_node
**prev_ex
,
272 struct extent_node
**next_ex
,
273 struct rb_node
***insert_p
,
274 struct rb_node
**insert_parent
)
276 struct rb_node
**pnode
= &et
->root
.rb_node
;
277 struct rb_node
*parent
= NULL
, *tmp_node
;
278 struct extent_node
*en
= et
->cached_en
;
281 *insert_parent
= NULL
;
285 if (RB_EMPTY_ROOT(&et
->root
))
289 struct extent_info
*cei
= &en
->ei
;
291 if (cei
->fofs
<= fofs
&& cei
->fofs
+ cei
->len
> fofs
)
292 goto lookup_neighbors
;
297 en
= rb_entry(*pnode
, struct extent_node
, rb_node
);
299 if (fofs
< en
->ei
.fofs
)
300 pnode
= &(*pnode
)->rb_left
;
301 else if (fofs
>= en
->ei
.fofs
+ en
->ei
.len
)
302 pnode
= &(*pnode
)->rb_right
;
304 goto lookup_neighbors
;
308 *insert_parent
= parent
;
310 en
= rb_entry(parent
, struct extent_node
, rb_node
);
312 if (parent
&& fofs
> en
->ei
.fofs
)
313 tmp_node
= rb_next(parent
);
314 *next_ex
= tmp_node
?
315 rb_entry(tmp_node
, struct extent_node
, rb_node
) : NULL
;
318 if (parent
&& fofs
< en
->ei
.fofs
)
319 tmp_node
= rb_prev(parent
);
320 *prev_ex
= tmp_node
?
321 rb_entry(tmp_node
, struct extent_node
, rb_node
) : NULL
;
325 if (fofs
== en
->ei
.fofs
) {
326 /* lookup prev node for merging backward later */
327 tmp_node
= rb_prev(&en
->rb_node
);
328 *prev_ex
= tmp_node
?
329 rb_entry(tmp_node
, struct extent_node
, rb_node
) : NULL
;
331 if (fofs
== en
->ei
.fofs
+ en
->ei
.len
- 1) {
332 /* lookup next node for merging frontward later */
333 tmp_node
= rb_next(&en
->rb_node
);
334 *next_ex
= tmp_node
?
335 rb_entry(tmp_node
, struct extent_node
, rb_node
) : NULL
;
340 static struct extent_node
*__try_merge_extent_node(struct inode
*inode
,
341 struct extent_tree
*et
, struct extent_info
*ei
,
342 struct extent_node
*prev_ex
,
343 struct extent_node
*next_ex
)
345 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
346 struct extent_node
*en
= NULL
;
348 if (prev_ex
&& __is_back_mergeable(ei
, &prev_ex
->ei
)) {
349 prev_ex
->ei
.len
+= ei
->len
;
354 if (next_ex
&& __is_front_mergeable(ei
, &next_ex
->ei
)) {
355 next_ex
->ei
.fofs
= ei
->fofs
;
356 next_ex
->ei
.blk
= ei
->blk
;
357 next_ex
->ei
.len
+= ei
->len
;
359 __release_extent_node(sbi
, et
, prev_ex
);
367 __try_update_largest_extent(inode
, et
, en
);
369 spin_lock(&sbi
->extent_lock
);
370 if (!list_empty(&en
->list
)) {
371 list_move_tail(&en
->list
, &sbi
->extent_list
);
374 spin_unlock(&sbi
->extent_lock
);
378 static struct extent_node
*__insert_extent_tree(struct inode
*inode
,
379 struct extent_tree
*et
, struct extent_info
*ei
,
380 struct rb_node
**insert_p
,
381 struct rb_node
*insert_parent
)
383 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
384 struct rb_node
**p
= &et
->root
.rb_node
;
385 struct rb_node
*parent
= NULL
;
386 struct extent_node
*en
= NULL
;
388 if (insert_p
&& insert_parent
) {
389 parent
= insert_parent
;
396 en
= rb_entry(parent
, struct extent_node
, rb_node
);
398 if (ei
->fofs
< en
->ei
.fofs
)
400 else if (ei
->fofs
>= en
->ei
.fofs
+ en
->ei
.len
)
406 en
= __attach_extent_node(sbi
, et
, ei
, parent
, p
);
410 __try_update_largest_extent(inode
, et
, en
);
412 /* update in global extent list */
413 spin_lock(&sbi
->extent_lock
);
414 list_add_tail(&en
->list
, &sbi
->extent_list
);
416 spin_unlock(&sbi
->extent_lock
);
420 static unsigned int f2fs_update_extent_tree_range(struct inode
*inode
,
421 pgoff_t fofs
, block_t blkaddr
, unsigned int len
)
423 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
424 struct extent_tree
*et
= F2FS_I(inode
)->extent_tree
;
425 struct extent_node
*en
= NULL
, *en1
= NULL
;
426 struct extent_node
*prev_en
= NULL
, *next_en
= NULL
;
427 struct extent_info ei
, dei
, prev
;
428 struct rb_node
**insert_p
= NULL
, *insert_parent
= NULL
;
429 unsigned int end
= fofs
+ len
;
430 unsigned int pos
= (unsigned int)fofs
;
435 trace_f2fs_update_extent_tree_range(inode
, fofs
, blkaddr
, len
);
437 write_lock(&et
->lock
);
439 if (is_inode_flag_set(inode
, FI_NO_EXTENT
)) {
440 write_unlock(&et
->lock
);
448 * drop largest extent before lookup, in case it's already
449 * been shrunk from extent tree
451 __drop_largest_extent(inode
, fofs
, len
);
453 /* 1. lookup first extent node in range [fofs, fofs + len - 1] */
454 en
= __lookup_extent_tree_ret(et
, fofs
, &prev_en
, &next_en
,
455 &insert_p
, &insert_parent
);
459 /* 2. invlidate all extent nodes in range [fofs, fofs + len - 1] */
460 while (en
&& en
->ei
.fofs
< end
) {
461 unsigned int org_end
;
462 int parts
= 0; /* # of parts current extent split into */
464 next_en
= en1
= NULL
;
467 org_end
= dei
.fofs
+ dei
.len
;
468 f2fs_bug_on(sbi
, pos
>= org_end
);
470 if (pos
> dei
.fofs
&& pos
- dei
.fofs
>= F2FS_MIN_EXTENT_LEN
) {
471 en
->ei
.len
= pos
- en
->ei
.fofs
;
476 if (end
< org_end
&& org_end
- end
>= F2FS_MIN_EXTENT_LEN
) {
478 set_extent_info(&ei
, end
,
479 end
- dei
.fofs
+ dei
.blk
,
481 en1
= __insert_extent_tree(inode
, et
, &ei
,
486 en
->ei
.blk
+= end
- dei
.fofs
;
487 en
->ei
.len
-= end
- dei
.fofs
;
494 struct rb_node
*node
= rb_next(&en
->rb_node
);
497 rb_entry(node
, struct extent_node
, rb_node
)
502 __try_update_largest_extent(inode
, et
, en
);
504 __release_extent_node(sbi
, et
, en
);
507 * if original extent is split into zero or two parts, extent
508 * tree has been altered by deletion or insertion, therefore
509 * invalidate pointers regard to tree.
513 insert_parent
= NULL
;
518 /* 3. update extent in extent cache */
521 set_extent_info(&ei
, fofs
, blkaddr
, len
);
522 if (!__try_merge_extent_node(inode
, et
, &ei
, prev_en
, next_en
))
523 __insert_extent_tree(inode
, et
, &ei
,
524 insert_p
, insert_parent
);
526 /* give up extent_cache, if split and small updates happen */
528 prev
.len
< F2FS_MIN_EXTENT_LEN
&&
529 et
->largest
.len
< F2FS_MIN_EXTENT_LEN
) {
530 __drop_largest_extent(inode
, 0, UINT_MAX
);
531 set_inode_flag(inode
, FI_NO_EXTENT
);
535 if (is_inode_flag_set(inode
, FI_NO_EXTENT
))
536 __free_extent_tree(sbi
, et
);
538 write_unlock(&et
->lock
);
540 return !__is_extent_same(&prev
, &et
->largest
);
543 unsigned int f2fs_shrink_extent_tree(struct f2fs_sb_info
*sbi
, int nr_shrink
)
545 struct extent_tree
*et
, *next
;
546 struct extent_node
*en
;
547 unsigned int node_cnt
= 0, tree_cnt
= 0;
550 if (!test_opt(sbi
, EXTENT_CACHE
))
553 if (!atomic_read(&sbi
->total_zombie_tree
))
556 if (!down_write_trylock(&sbi
->extent_tree_lock
))
559 /* 1. remove unreferenced extent tree */
560 list_for_each_entry_safe(et
, next
, &sbi
->zombie_list
, list
) {
561 if (atomic_read(&et
->node_cnt
)) {
562 write_lock(&et
->lock
);
563 node_cnt
+= __free_extent_tree(sbi
, et
);
564 write_unlock(&et
->lock
);
566 f2fs_bug_on(sbi
, atomic_read(&et
->node_cnt
));
567 list_del_init(&et
->list
);
568 radix_tree_delete(&sbi
->extent_tree_root
, et
->ino
);
569 kmem_cache_free(extent_tree_slab
, et
);
570 atomic_dec(&sbi
->total_ext_tree
);
571 atomic_dec(&sbi
->total_zombie_tree
);
574 if (node_cnt
+ tree_cnt
>= nr_shrink
)
578 up_write(&sbi
->extent_tree_lock
);
581 /* 2. remove LRU extent entries */
582 if (!down_write_trylock(&sbi
->extent_tree_lock
))
585 remained
= nr_shrink
- (node_cnt
+ tree_cnt
);
587 spin_lock(&sbi
->extent_lock
);
588 for (; remained
> 0; remained
--) {
589 if (list_empty(&sbi
->extent_list
))
591 en
= list_first_entry(&sbi
->extent_list
,
592 struct extent_node
, list
);
594 if (!write_trylock(&et
->lock
)) {
595 /* refresh this extent node's position in extent list */
596 list_move_tail(&en
->list
, &sbi
->extent_list
);
600 list_del_init(&en
->list
);
601 spin_unlock(&sbi
->extent_lock
);
603 __detach_extent_node(sbi
, et
, en
);
605 write_unlock(&et
->lock
);
607 spin_lock(&sbi
->extent_lock
);
609 spin_unlock(&sbi
->extent_lock
);
612 up_write(&sbi
->extent_tree_lock
);
614 trace_f2fs_shrink_extent_tree(sbi
, node_cnt
, tree_cnt
);
616 return node_cnt
+ tree_cnt
;
619 unsigned int f2fs_destroy_extent_node(struct inode
*inode
)
621 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
622 struct extent_tree
*et
= F2FS_I(inode
)->extent_tree
;
623 unsigned int node_cnt
= 0;
625 if (!et
|| !atomic_read(&et
->node_cnt
))
628 write_lock(&et
->lock
);
629 node_cnt
= __free_extent_tree(sbi
, et
);
630 write_unlock(&et
->lock
);
635 void f2fs_drop_extent_tree(struct inode
*inode
)
637 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
638 struct extent_tree
*et
= F2FS_I(inode
)->extent_tree
;
640 set_inode_flag(inode
, FI_NO_EXTENT
);
642 write_lock(&et
->lock
);
643 __free_extent_tree(sbi
, et
);
644 __drop_largest_extent(inode
, 0, UINT_MAX
);
645 write_unlock(&et
->lock
);
648 void f2fs_destroy_extent_tree(struct inode
*inode
)
650 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
651 struct extent_tree
*et
= F2FS_I(inode
)->extent_tree
;
652 unsigned int node_cnt
= 0;
657 if (inode
->i_nlink
&& !is_bad_inode(inode
) &&
658 atomic_read(&et
->node_cnt
)) {
659 down_write(&sbi
->extent_tree_lock
);
660 list_add_tail(&et
->list
, &sbi
->zombie_list
);
661 atomic_inc(&sbi
->total_zombie_tree
);
662 up_write(&sbi
->extent_tree_lock
);
666 /* free all extent info belong to this extent tree */
667 node_cnt
= f2fs_destroy_extent_node(inode
);
669 /* delete extent tree entry in radix tree */
670 down_write(&sbi
->extent_tree_lock
);
671 f2fs_bug_on(sbi
, atomic_read(&et
->node_cnt
));
672 radix_tree_delete(&sbi
->extent_tree_root
, inode
->i_ino
);
673 kmem_cache_free(extent_tree_slab
, et
);
674 atomic_dec(&sbi
->total_ext_tree
);
675 up_write(&sbi
->extent_tree_lock
);
677 F2FS_I(inode
)->extent_tree
= NULL
;
679 trace_f2fs_destroy_extent_tree(inode
, node_cnt
);
682 bool f2fs_lookup_extent_cache(struct inode
*inode
, pgoff_t pgofs
,
683 struct extent_info
*ei
)
685 if (!f2fs_may_extent_tree(inode
))
688 return f2fs_lookup_extent_tree(inode
, pgofs
, ei
);
691 void f2fs_update_extent_cache(struct dnode_of_data
*dn
)
696 if (!f2fs_may_extent_tree(dn
->inode
))
699 if (dn
->data_blkaddr
== NEW_ADDR
)
702 blkaddr
= dn
->data_blkaddr
;
704 fofs
= start_bidx_of_node(ofs_of_node(dn
->node_page
), dn
->inode
) +
706 f2fs_update_extent_tree_range(dn
->inode
, fofs
, blkaddr
, 1);
709 void f2fs_update_extent_cache_range(struct dnode_of_data
*dn
,
710 pgoff_t fofs
, block_t blkaddr
, unsigned int len
)
713 if (!f2fs_may_extent_tree(dn
->inode
))
716 f2fs_update_extent_tree_range(dn
->inode
, fofs
, blkaddr
, len
);
719 void init_extent_cache_info(struct f2fs_sb_info
*sbi
)
721 INIT_RADIX_TREE(&sbi
->extent_tree_root
, GFP_NOIO
);
722 init_rwsem(&sbi
->extent_tree_lock
);
723 INIT_LIST_HEAD(&sbi
->extent_list
);
724 spin_lock_init(&sbi
->extent_lock
);
725 atomic_set(&sbi
->total_ext_tree
, 0);
726 INIT_LIST_HEAD(&sbi
->zombie_list
);
727 atomic_set(&sbi
->total_zombie_tree
, 0);
728 atomic_set(&sbi
->total_ext_node
, 0);
731 int __init
create_extent_cache(void)
733 extent_tree_slab
= f2fs_kmem_cache_create("f2fs_extent_tree",
734 sizeof(struct extent_tree
));
735 if (!extent_tree_slab
)
737 extent_node_slab
= f2fs_kmem_cache_create("f2fs_extent_node",
738 sizeof(struct extent_node
));
739 if (!extent_node_slab
) {
740 kmem_cache_destroy(extent_tree_slab
);
746 void destroy_extent_cache(void)
748 kmem_cache_destroy(extent_node_slab
);
749 kmem_cache_destroy(extent_tree_slab
);