Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * f2fs extent cache support
4 *
5 * Copyright (c) 2015 Motorola Mobility
6 * Copyright (c) 2015 Samsung Electronics
7 * Authors: Jaegeuk Kim <jaegeuk@kernel.org>
8 * Chao Yu <chao2.yu@samsung.com>
9 */
10
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13
14#include "f2fs.h"
15#include "node.h"
16#include <trace/events/f2fs.h>
17
18static struct rb_entry *__lookup_rb_tree_fast(struct rb_entry *cached_re,
19 unsigned int ofs)
20{
21 if (cached_re) {
22 if (cached_re->ofs <= ofs &&
23 cached_re->ofs + cached_re->len > ofs) {
24 return cached_re;
25 }
26 }
27 return NULL;
28}
29
30static struct rb_entry *__lookup_rb_tree_slow(struct rb_root_cached *root,
31 unsigned int ofs)
32{
33 struct rb_node *node = root->rb_root.rb_node;
34 struct rb_entry *re;
35
36 while (node) {
37 re = rb_entry(node, struct rb_entry, rb_node);
38
39 if (ofs < re->ofs)
40 node = node->rb_left;
41 else if (ofs >= re->ofs + re->len)
42 node = node->rb_right;
43 else
44 return re;
45 }
46 return NULL;
47}
48
49struct rb_entry *f2fs_lookup_rb_tree(struct rb_root_cached *root,
50 struct rb_entry *cached_re, unsigned int ofs)
51{
52 struct rb_entry *re;
53
54 re = __lookup_rb_tree_fast(cached_re, ofs);
55 if (!re)
56 return __lookup_rb_tree_slow(root, ofs);
57
58 return re;
59}
60
61struct rb_node **f2fs_lookup_rb_tree_ext(struct f2fs_sb_info *sbi,
62 struct rb_root_cached *root,
63 struct rb_node **parent,
64 unsigned long long key, bool *leftmost)
65{
66 struct rb_node **p = &root->rb_root.rb_node;
67 struct rb_entry *re;
68
69 while (*p) {
70 *parent = *p;
71 re = rb_entry(*parent, struct rb_entry, rb_node);
72
73 if (key < re->key) {
74 p = &(*p)->rb_left;
75 } else {
76 p = &(*p)->rb_right;
77 *leftmost = false;
78 }
79 }
80
81 return p;
82}
83
84struct rb_node **f2fs_lookup_rb_tree_for_insert(struct f2fs_sb_info *sbi,
85 struct rb_root_cached *root,
86 struct rb_node **parent,
87 unsigned int ofs, bool *leftmost)
88{
89 struct rb_node **p = &root->rb_root.rb_node;
90 struct rb_entry *re;
91
92 while (*p) {
93 *parent = *p;
94 re = rb_entry(*parent, struct rb_entry, rb_node);
95
96 if (ofs < re->ofs) {
97 p = &(*p)->rb_left;
98 } else if (ofs >= re->ofs + re->len) {
99 p = &(*p)->rb_right;
100 *leftmost = false;
101 } else {
102 f2fs_bug_on(sbi, 1);
103 }
104 }
105
106 return p;
107}
108
109/*
110 * lookup rb entry in position of @ofs in rb-tree,
111 * if hit, return the entry, otherwise, return NULL
112 * @prev_ex: extent before ofs
113 * @next_ex: extent after ofs
114 * @insert_p: insert point for new extent at ofs
115 * in order to simpfy the insertion after.
116 * tree must stay unchanged between lookup and insertion.
117 */
118struct rb_entry *f2fs_lookup_rb_tree_ret(struct rb_root_cached *root,
119 struct rb_entry *cached_re,
120 unsigned int ofs,
121 struct rb_entry **prev_entry,
122 struct rb_entry **next_entry,
123 struct rb_node ***insert_p,
124 struct rb_node **insert_parent,
125 bool force, bool *leftmost)
126{
127 struct rb_node **pnode = &root->rb_root.rb_node;
128 struct rb_node *parent = NULL, *tmp_node;
129 struct rb_entry *re = cached_re;
130
131 *insert_p = NULL;
132 *insert_parent = NULL;
133 *prev_entry = NULL;
134 *next_entry = NULL;
135
136 if (RB_EMPTY_ROOT(&root->rb_root))
137 return NULL;
138
139 if (re) {
140 if (re->ofs <= ofs && re->ofs + re->len > ofs)
141 goto lookup_neighbors;
142 }
143
144 if (leftmost)
145 *leftmost = true;
146
147 while (*pnode) {
148 parent = *pnode;
149 re = rb_entry(*pnode, struct rb_entry, rb_node);
150
151 if (ofs < re->ofs) {
152 pnode = &(*pnode)->rb_left;
153 } else if (ofs >= re->ofs + re->len) {
154 pnode = &(*pnode)->rb_right;
155 if (leftmost)
156 *leftmost = false;
157 } else {
158 goto lookup_neighbors;
159 }
160 }
161
162 *insert_p = pnode;
163 *insert_parent = parent;
164
165 re = rb_entry(parent, struct rb_entry, rb_node);
166 tmp_node = parent;
167 if (parent && ofs > re->ofs)
168 tmp_node = rb_next(parent);
169 *next_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
170
171 tmp_node = parent;
172 if (parent && ofs < re->ofs)
173 tmp_node = rb_prev(parent);
174 *prev_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
175 return NULL;
176
177lookup_neighbors:
178 if (ofs == re->ofs || force) {
179 /* lookup prev node for merging backward later */
180 tmp_node = rb_prev(&re->rb_node);
181 *prev_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
182 }
183 if (ofs == re->ofs + re->len - 1 || force) {
184 /* lookup next node for merging frontward later */
185 tmp_node = rb_next(&re->rb_node);
186 *next_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
187 }
188 return re;
189}
190
191bool f2fs_check_rb_tree_consistence(struct f2fs_sb_info *sbi,
192 struct rb_root_cached *root, bool check_key)
193{
194#ifdef CONFIG_F2FS_CHECK_FS
195 struct rb_node *cur = rb_first_cached(root), *next;
196 struct rb_entry *cur_re, *next_re;
197
198 if (!cur)
199 return true;
200
201 while (cur) {
202 next = rb_next(cur);
203 if (!next)
204 return true;
205
206 cur_re = rb_entry(cur, struct rb_entry, rb_node);
207 next_re = rb_entry(next, struct rb_entry, rb_node);
208
209 if (check_key) {
210 if (cur_re->key > next_re->key) {
211 f2fs_info(sbi, "inconsistent rbtree, "
212 "cur(%llu) next(%llu)",
213 cur_re->key, next_re->key);
214 return false;
215 }
216 goto next;
217 }
218
219 if (cur_re->ofs + cur_re->len > next_re->ofs) {
220 f2fs_info(sbi, "inconsistent rbtree, cur(%u, %u) next(%u, %u)",
221 cur_re->ofs, cur_re->len,
222 next_re->ofs, next_re->len);
223 return false;
224 }
225next:
226 cur = next;
227 }
228#endif
229 return true;
230}
231
232static struct kmem_cache *extent_tree_slab;
233static struct kmem_cache *extent_node_slab;
234
235static struct extent_node *__attach_extent_node(struct f2fs_sb_info *sbi,
236 struct extent_tree *et, struct extent_info *ei,
237 struct rb_node *parent, struct rb_node **p,
238 bool leftmost)
239{
240 struct extent_node *en;
241
242 en = kmem_cache_alloc(extent_node_slab, GFP_ATOMIC);
243 if (!en)
244 return NULL;
245
246 en->ei = *ei;
247 INIT_LIST_HEAD(&en->list);
248 en->et = et;
249
250 rb_link_node(&en->rb_node, parent, p);
251 rb_insert_color_cached(&en->rb_node, &et->root, leftmost);
252 atomic_inc(&et->node_cnt);
253 atomic_inc(&sbi->total_ext_node);
254 return en;
255}
256
257static void __detach_extent_node(struct f2fs_sb_info *sbi,
258 struct extent_tree *et, struct extent_node *en)
259{
260 rb_erase_cached(&en->rb_node, &et->root);
261 atomic_dec(&et->node_cnt);
262 atomic_dec(&sbi->total_ext_node);
263
264 if (et->cached_en == en)
265 et->cached_en = NULL;
266 kmem_cache_free(extent_node_slab, en);
267}
268
269/*
270 * Flow to release an extent_node:
271 * 1. list_del_init
272 * 2. __detach_extent_node
273 * 3. kmem_cache_free.
274 */
275static void __release_extent_node(struct f2fs_sb_info *sbi,
276 struct extent_tree *et, struct extent_node *en)
277{
278 spin_lock(&sbi->extent_lock);
279 f2fs_bug_on(sbi, list_empty(&en->list));
280 list_del_init(&en->list);
281 spin_unlock(&sbi->extent_lock);
282
283 __detach_extent_node(sbi, et, en);
284}
285
286static struct extent_tree *__grab_extent_tree(struct inode *inode)
287{
288 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
289 struct extent_tree *et;
290 nid_t ino = inode->i_ino;
291
292 mutex_lock(&sbi->extent_tree_lock);
293 et = radix_tree_lookup(&sbi->extent_tree_root, ino);
294 if (!et) {
295 et = f2fs_kmem_cache_alloc(extent_tree_slab, GFP_NOFS);
296 f2fs_radix_tree_insert(&sbi->extent_tree_root, ino, et);
297 memset(et, 0, sizeof(struct extent_tree));
298 et->ino = ino;
299 et->root = RB_ROOT_CACHED;
300 et->cached_en = NULL;
301 rwlock_init(&et->lock);
302 INIT_LIST_HEAD(&et->list);
303 atomic_set(&et->node_cnt, 0);
304 atomic_inc(&sbi->total_ext_tree);
305 } else {
306 atomic_dec(&sbi->total_zombie_tree);
307 list_del_init(&et->list);
308 }
309 mutex_unlock(&sbi->extent_tree_lock);
310
311 /* never died until evict_inode */
312 F2FS_I(inode)->extent_tree = et;
313
314 return et;
315}
316
317static struct extent_node *__init_extent_tree(struct f2fs_sb_info *sbi,
318 struct extent_tree *et, struct extent_info *ei)
319{
320 struct rb_node **p = &et->root.rb_root.rb_node;
321 struct extent_node *en;
322
323 en = __attach_extent_node(sbi, et, ei, NULL, p, true);
324 if (!en)
325 return NULL;
326
327 et->largest = en->ei;
328 et->cached_en = en;
329 return en;
330}
331
332static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi,
333 struct extent_tree *et)
334{
335 struct rb_node *node, *next;
336 struct extent_node *en;
337 unsigned int count = atomic_read(&et->node_cnt);
338
339 node = rb_first_cached(&et->root);
340 while (node) {
341 next = rb_next(node);
342 en = rb_entry(node, struct extent_node, rb_node);
343 __release_extent_node(sbi, et, en);
344 node = next;
345 }
346
347 return count - atomic_read(&et->node_cnt);
348}
349
350static void __drop_largest_extent(struct extent_tree *et,
351 pgoff_t fofs, unsigned int len)
352{
353 if (fofs < et->largest.fofs + et->largest.len &&
354 fofs + len > et->largest.fofs) {
355 et->largest.len = 0;
356 et->largest_updated = true;
357 }
358}
359
360/* return true, if inode page is changed */
361static void __f2fs_init_extent_tree(struct inode *inode, struct page *ipage)
362{
363 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
364 struct f2fs_extent *i_ext = ipage ? &F2FS_INODE(ipage)->i_ext : NULL;
365 struct extent_tree *et;
366 struct extent_node *en;
367 struct extent_info ei;
368
369 if (!f2fs_may_extent_tree(inode)) {
370 /* drop largest extent */
371 if (i_ext && i_ext->len) {
372 f2fs_wait_on_page_writeback(ipage, NODE, true, true);
373 i_ext->len = 0;
374 set_page_dirty(ipage);
375 return;
376 }
377 return;
378 }
379
380 et = __grab_extent_tree(inode);
381
382 if (!i_ext || !i_ext->len)
383 return;
384
385 get_extent_info(&ei, i_ext);
386
387 write_lock(&et->lock);
388 if (atomic_read(&et->node_cnt))
389 goto out;
390
391 en = __init_extent_tree(sbi, et, &ei);
392 if (en) {
393 spin_lock(&sbi->extent_lock);
394 list_add_tail(&en->list, &sbi->extent_list);
395 spin_unlock(&sbi->extent_lock);
396 }
397out:
398 write_unlock(&et->lock);
399}
400
401void f2fs_init_extent_tree(struct inode *inode, struct page *ipage)
402{
403 __f2fs_init_extent_tree(inode, ipage);
404
405 if (!F2FS_I(inode)->extent_tree)
406 set_inode_flag(inode, FI_NO_EXTENT);
407}
408
409static bool f2fs_lookup_extent_tree(struct inode *inode, pgoff_t pgofs,
410 struct extent_info *ei)
411{
412 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
413 struct extent_tree *et = F2FS_I(inode)->extent_tree;
414 struct extent_node *en;
415 bool ret = false;
416
417 f2fs_bug_on(sbi, !et);
418
419 trace_f2fs_lookup_extent_tree_start(inode, pgofs);
420
421 read_lock(&et->lock);
422
423 if (et->largest.fofs <= pgofs &&
424 et->largest.fofs + et->largest.len > pgofs) {
425 *ei = et->largest;
426 ret = true;
427 stat_inc_largest_node_hit(sbi);
428 goto out;
429 }
430
431 en = (struct extent_node *)f2fs_lookup_rb_tree(&et->root,
432 (struct rb_entry *)et->cached_en, pgofs);
433 if (!en)
434 goto out;
435
436 if (en == et->cached_en)
437 stat_inc_cached_node_hit(sbi);
438 else
439 stat_inc_rbtree_node_hit(sbi);
440
441 *ei = en->ei;
442 spin_lock(&sbi->extent_lock);
443 if (!list_empty(&en->list)) {
444 list_move_tail(&en->list, &sbi->extent_list);
445 et->cached_en = en;
446 }
447 spin_unlock(&sbi->extent_lock);
448 ret = true;
449out:
450 stat_inc_total_hit(sbi);
451 read_unlock(&et->lock);
452
453 trace_f2fs_lookup_extent_tree_end(inode, pgofs, ei);
454 return ret;
455}
456
457static struct extent_node *__try_merge_extent_node(struct f2fs_sb_info *sbi,
458 struct extent_tree *et, struct extent_info *ei,
459 struct extent_node *prev_ex,
460 struct extent_node *next_ex)
461{
462 struct extent_node *en = NULL;
463
464 if (prev_ex && __is_back_mergeable(ei, &prev_ex->ei)) {
465 prev_ex->ei.len += ei->len;
466 ei = &prev_ex->ei;
467 en = prev_ex;
468 }
469
470 if (next_ex && __is_front_mergeable(ei, &next_ex->ei)) {
471 next_ex->ei.fofs = ei->fofs;
472 next_ex->ei.blk = ei->blk;
473 next_ex->ei.len += ei->len;
474 if (en)
475 __release_extent_node(sbi, et, prev_ex);
476
477 en = next_ex;
478 }
479
480 if (!en)
481 return NULL;
482
483 __try_update_largest_extent(et, en);
484
485 spin_lock(&sbi->extent_lock);
486 if (!list_empty(&en->list)) {
487 list_move_tail(&en->list, &sbi->extent_list);
488 et->cached_en = en;
489 }
490 spin_unlock(&sbi->extent_lock);
491 return en;
492}
493
494static struct extent_node *__insert_extent_tree(struct f2fs_sb_info *sbi,
495 struct extent_tree *et, struct extent_info *ei,
496 struct rb_node **insert_p,
497 struct rb_node *insert_parent,
498 bool leftmost)
499{
500 struct rb_node **p;
501 struct rb_node *parent = NULL;
502 struct extent_node *en = NULL;
503
504 if (insert_p && insert_parent) {
505 parent = insert_parent;
506 p = insert_p;
507 goto do_insert;
508 }
509
510 leftmost = true;
511
512 p = f2fs_lookup_rb_tree_for_insert(sbi, &et->root, &parent,
513 ei->fofs, &leftmost);
514do_insert:
515 en = __attach_extent_node(sbi, et, ei, parent, p, leftmost);
516 if (!en)
517 return NULL;
518
519 __try_update_largest_extent(et, en);
520
521 /* update in global extent list */
522 spin_lock(&sbi->extent_lock);
523 list_add_tail(&en->list, &sbi->extent_list);
524 et->cached_en = en;
525 spin_unlock(&sbi->extent_lock);
526 return en;
527}
528
529static void f2fs_update_extent_tree_range(struct inode *inode,
530 pgoff_t fofs, block_t blkaddr, unsigned int len)
531{
532 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
533 struct extent_tree *et = F2FS_I(inode)->extent_tree;
534 struct extent_node *en = NULL, *en1 = NULL;
535 struct extent_node *prev_en = NULL, *next_en = NULL;
536 struct extent_info ei, dei, prev;
537 struct rb_node **insert_p = NULL, *insert_parent = NULL;
538 unsigned int end = fofs + len;
539 unsigned int pos = (unsigned int)fofs;
540 bool updated = false;
541 bool leftmost = false;
542
543 if (!et)
544 return;
545
546 trace_f2fs_update_extent_tree_range(inode, fofs, blkaddr, len);
547
548 write_lock(&et->lock);
549
550 if (is_inode_flag_set(inode, FI_NO_EXTENT)) {
551 write_unlock(&et->lock);
552 return;
553 }
554
555 prev = et->largest;
556 dei.len = 0;
557
558 /*
559 * drop largest extent before lookup, in case it's already
560 * been shrunk from extent tree
561 */
562 __drop_largest_extent(et, fofs, len);
563
564 /* 1. lookup first extent node in range [fofs, fofs + len - 1] */
565 en = (struct extent_node *)f2fs_lookup_rb_tree_ret(&et->root,
566 (struct rb_entry *)et->cached_en, fofs,
567 (struct rb_entry **)&prev_en,
568 (struct rb_entry **)&next_en,
569 &insert_p, &insert_parent, false,
570 &leftmost);
571 if (!en)
572 en = next_en;
573
574 /* 2. invlidate all extent nodes in range [fofs, fofs + len - 1] */
575 while (en && en->ei.fofs < end) {
576 unsigned int org_end;
577 int parts = 0; /* # of parts current extent split into */
578
579 next_en = en1 = NULL;
580
581 dei = en->ei;
582 org_end = dei.fofs + dei.len;
583 f2fs_bug_on(sbi, pos >= org_end);
584
585 if (pos > dei.fofs && pos - dei.fofs >= F2FS_MIN_EXTENT_LEN) {
586 en->ei.len = pos - en->ei.fofs;
587 prev_en = en;
588 parts = 1;
589 }
590
591 if (end < org_end && org_end - end >= F2FS_MIN_EXTENT_LEN) {
592 if (parts) {
593 set_extent_info(&ei, end,
594 end - dei.fofs + dei.blk,
595 org_end - end);
596 en1 = __insert_extent_tree(sbi, et, &ei,
597 NULL, NULL, true);
598 next_en = en1;
599 } else {
600 en->ei.fofs = end;
601 en->ei.blk += end - dei.fofs;
602 en->ei.len -= end - dei.fofs;
603 next_en = en;
604 }
605 parts++;
606 }
607
608 if (!next_en) {
609 struct rb_node *node = rb_next(&en->rb_node);
610
611 next_en = rb_entry_safe(node, struct extent_node,
612 rb_node);
613 }
614
615 if (parts)
616 __try_update_largest_extent(et, en);
617 else
618 __release_extent_node(sbi, et, en);
619
620 /*
621 * if original extent is split into zero or two parts, extent
622 * tree has been altered by deletion or insertion, therefore
623 * invalidate pointers regard to tree.
624 */
625 if (parts != 1) {
626 insert_p = NULL;
627 insert_parent = NULL;
628 }
629 en = next_en;
630 }
631
632 /* 3. update extent in extent cache */
633 if (blkaddr) {
634
635 set_extent_info(&ei, fofs, blkaddr, len);
636 if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
637 __insert_extent_tree(sbi, et, &ei,
638 insert_p, insert_parent, leftmost);
639
640 /* give up extent_cache, if split and small updates happen */
641 if (dei.len >= 1 &&
642 prev.len < F2FS_MIN_EXTENT_LEN &&
643 et->largest.len < F2FS_MIN_EXTENT_LEN) {
644 et->largest.len = 0;
645 et->largest_updated = true;
646 set_inode_flag(inode, FI_NO_EXTENT);
647 }
648 }
649
650 if (is_inode_flag_set(inode, FI_NO_EXTENT))
651 __free_extent_tree(sbi, et);
652
653 if (et->largest_updated) {
654 et->largest_updated = false;
655 updated = true;
656 }
657
658 write_unlock(&et->lock);
659
660 if (updated)
661 f2fs_mark_inode_dirty_sync(inode, true);
662}
663
664unsigned int f2fs_shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
665{
666 struct extent_tree *et, *next;
667 struct extent_node *en;
668 unsigned int node_cnt = 0, tree_cnt = 0;
669 int remained;
670
671 if (!test_opt(sbi, EXTENT_CACHE))
672 return 0;
673
674 if (!atomic_read(&sbi->total_zombie_tree))
675 goto free_node;
676
677 if (!mutex_trylock(&sbi->extent_tree_lock))
678 goto out;
679
680 /* 1. remove unreferenced extent tree */
681 list_for_each_entry_safe(et, next, &sbi->zombie_list, list) {
682 if (atomic_read(&et->node_cnt)) {
683 write_lock(&et->lock);
684 node_cnt += __free_extent_tree(sbi, et);
685 write_unlock(&et->lock);
686 }
687 f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
688 list_del_init(&et->list);
689 radix_tree_delete(&sbi->extent_tree_root, et->ino);
690 kmem_cache_free(extent_tree_slab, et);
691 atomic_dec(&sbi->total_ext_tree);
692 atomic_dec(&sbi->total_zombie_tree);
693 tree_cnt++;
694
695 if (node_cnt + tree_cnt >= nr_shrink)
696 goto unlock_out;
697 cond_resched();
698 }
699 mutex_unlock(&sbi->extent_tree_lock);
700
701free_node:
702 /* 2. remove LRU extent entries */
703 if (!mutex_trylock(&sbi->extent_tree_lock))
704 goto out;
705
706 remained = nr_shrink - (node_cnt + tree_cnt);
707
708 spin_lock(&sbi->extent_lock);
709 for (; remained > 0; remained--) {
710 if (list_empty(&sbi->extent_list))
711 break;
712 en = list_first_entry(&sbi->extent_list,
713 struct extent_node, list);
714 et = en->et;
715 if (!write_trylock(&et->lock)) {
716 /* refresh this extent node's position in extent list */
717 list_move_tail(&en->list, &sbi->extent_list);
718 continue;
719 }
720
721 list_del_init(&en->list);
722 spin_unlock(&sbi->extent_lock);
723
724 __detach_extent_node(sbi, et, en);
725
726 write_unlock(&et->lock);
727 node_cnt++;
728 spin_lock(&sbi->extent_lock);
729 }
730 spin_unlock(&sbi->extent_lock);
731
732unlock_out:
733 mutex_unlock(&sbi->extent_tree_lock);
734out:
735 trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt);
736
737 return node_cnt + tree_cnt;
738}
739
740unsigned int f2fs_destroy_extent_node(struct inode *inode)
741{
742 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
743 struct extent_tree *et = F2FS_I(inode)->extent_tree;
744 unsigned int node_cnt = 0;
745
746 if (!et || !atomic_read(&et->node_cnt))
747 return 0;
748
749 write_lock(&et->lock);
750 node_cnt = __free_extent_tree(sbi, et);
751 write_unlock(&et->lock);
752
753 return node_cnt;
754}
755
756void f2fs_drop_extent_tree(struct inode *inode)
757{
758 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
759 struct extent_tree *et = F2FS_I(inode)->extent_tree;
760 bool updated = false;
761
762 if (!f2fs_may_extent_tree(inode))
763 return;
764
765 set_inode_flag(inode, FI_NO_EXTENT);
766
767 write_lock(&et->lock);
768 __free_extent_tree(sbi, et);
769 if (et->largest.len) {
770 et->largest.len = 0;
771 updated = true;
772 }
773 write_unlock(&et->lock);
774 if (updated)
775 f2fs_mark_inode_dirty_sync(inode, true);
776}
777
778void f2fs_destroy_extent_tree(struct inode *inode)
779{
780 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
781 struct extent_tree *et = F2FS_I(inode)->extent_tree;
782 unsigned int node_cnt = 0;
783
784 if (!et)
785 return;
786
787 if (inode->i_nlink && !is_bad_inode(inode) &&
788 atomic_read(&et->node_cnt)) {
789 mutex_lock(&sbi->extent_tree_lock);
790 list_add_tail(&et->list, &sbi->zombie_list);
791 atomic_inc(&sbi->total_zombie_tree);
792 mutex_unlock(&sbi->extent_tree_lock);
793 return;
794 }
795
796 /* free all extent info belong to this extent tree */
797 node_cnt = f2fs_destroy_extent_node(inode);
798
799 /* delete extent tree entry in radix tree */
800 mutex_lock(&sbi->extent_tree_lock);
801 f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
802 radix_tree_delete(&sbi->extent_tree_root, inode->i_ino);
803 kmem_cache_free(extent_tree_slab, et);
804 atomic_dec(&sbi->total_ext_tree);
805 mutex_unlock(&sbi->extent_tree_lock);
806
807 F2FS_I(inode)->extent_tree = NULL;
808
809 trace_f2fs_destroy_extent_tree(inode, node_cnt);
810}
811
812bool f2fs_lookup_extent_cache(struct inode *inode, pgoff_t pgofs,
813 struct extent_info *ei)
814{
815 if (!f2fs_may_extent_tree(inode))
816 return false;
817
818 return f2fs_lookup_extent_tree(inode, pgofs, ei);
819}
820
821void f2fs_update_extent_cache(struct dnode_of_data *dn)
822{
823 pgoff_t fofs;
824 block_t blkaddr;
825
826 if (!f2fs_may_extent_tree(dn->inode))
827 return;
828
829 if (dn->data_blkaddr == NEW_ADDR)
830 blkaddr = NULL_ADDR;
831 else
832 blkaddr = dn->data_blkaddr;
833
834 fofs = f2fs_start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
835 dn->ofs_in_node;
836 f2fs_update_extent_tree_range(dn->inode, fofs, blkaddr, 1);
837}
838
839void f2fs_update_extent_cache_range(struct dnode_of_data *dn,
840 pgoff_t fofs, block_t blkaddr, unsigned int len)
841
842{
843 if (!f2fs_may_extent_tree(dn->inode))
844 return;
845
846 f2fs_update_extent_tree_range(dn->inode, fofs, blkaddr, len);
847}
848
849void f2fs_init_extent_cache_info(struct f2fs_sb_info *sbi)
850{
851 INIT_RADIX_TREE(&sbi->extent_tree_root, GFP_NOIO);
852 mutex_init(&sbi->extent_tree_lock);
853 INIT_LIST_HEAD(&sbi->extent_list);
854 spin_lock_init(&sbi->extent_lock);
855 atomic_set(&sbi->total_ext_tree, 0);
856 INIT_LIST_HEAD(&sbi->zombie_list);
857 atomic_set(&sbi->total_zombie_tree, 0);
858 atomic_set(&sbi->total_ext_node, 0);
859}
860
861int __init f2fs_create_extent_cache(void)
862{
863 extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree",
864 sizeof(struct extent_tree));
865 if (!extent_tree_slab)
866 return -ENOMEM;
867 extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node",
868 sizeof(struct extent_node));
869 if (!extent_node_slab) {
870 kmem_cache_destroy(extent_tree_slab);
871 return -ENOMEM;
872 }
873 return 0;
874}
875
876void f2fs_destroy_extent_cache(void)
877{
878 kmem_cache_destroy(extent_node_slab);
879 kmem_cache_destroy(extent_tree_slab);
880}
1/*
2 * f2fs extent cache support
3 *
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>
8 *
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.
12 */
13
14#include <linux/fs.h>
15#include <linux/f2fs_fs.h>
16
17#include "f2fs.h"
18#include "node.h"
19#include <trace/events/f2fs.h>
20
21static struct rb_entry *__lookup_rb_tree_fast(struct rb_entry *cached_re,
22 unsigned int ofs)
23{
24 if (cached_re) {
25 if (cached_re->ofs <= ofs &&
26 cached_re->ofs + cached_re->len > ofs) {
27 return cached_re;
28 }
29 }
30 return NULL;
31}
32
33static struct rb_entry *__lookup_rb_tree_slow(struct rb_root *root,
34 unsigned int ofs)
35{
36 struct rb_node *node = root->rb_node;
37 struct rb_entry *re;
38
39 while (node) {
40 re = rb_entry(node, struct rb_entry, rb_node);
41
42 if (ofs < re->ofs)
43 node = node->rb_left;
44 else if (ofs >= re->ofs + re->len)
45 node = node->rb_right;
46 else
47 return re;
48 }
49 return NULL;
50}
51
52struct rb_entry *__lookup_rb_tree(struct rb_root *root,
53 struct rb_entry *cached_re, unsigned int ofs)
54{
55 struct rb_entry *re;
56
57 re = __lookup_rb_tree_fast(cached_re, ofs);
58 if (!re)
59 return __lookup_rb_tree_slow(root, ofs);
60
61 return re;
62}
63
64struct rb_node **__lookup_rb_tree_for_insert(struct f2fs_sb_info *sbi,
65 struct rb_root *root, struct rb_node **parent,
66 unsigned int ofs)
67{
68 struct rb_node **p = &root->rb_node;
69 struct rb_entry *re;
70
71 while (*p) {
72 *parent = *p;
73 re = rb_entry(*parent, struct rb_entry, rb_node);
74
75 if (ofs < re->ofs)
76 p = &(*p)->rb_left;
77 else if (ofs >= re->ofs + re->len)
78 p = &(*p)->rb_right;
79 else
80 f2fs_bug_on(sbi, 1);
81 }
82
83 return p;
84}
85
86/*
87 * lookup rb entry in position of @ofs in rb-tree,
88 * if hit, return the entry, otherwise, return NULL
89 * @prev_ex: extent before ofs
90 * @next_ex: extent after ofs
91 * @insert_p: insert point for new extent at ofs
92 * in order to simpfy the insertion after.
93 * tree must stay unchanged between lookup and insertion.
94 */
95struct rb_entry *__lookup_rb_tree_ret(struct rb_root *root,
96 struct rb_entry *cached_re,
97 unsigned int ofs,
98 struct rb_entry **prev_entry,
99 struct rb_entry **next_entry,
100 struct rb_node ***insert_p,
101 struct rb_node **insert_parent,
102 bool force)
103{
104 struct rb_node **pnode = &root->rb_node;
105 struct rb_node *parent = NULL, *tmp_node;
106 struct rb_entry *re = cached_re;
107
108 *insert_p = NULL;
109 *insert_parent = NULL;
110 *prev_entry = NULL;
111 *next_entry = NULL;
112
113 if (RB_EMPTY_ROOT(root))
114 return NULL;
115
116 if (re) {
117 if (re->ofs <= ofs && re->ofs + re->len > ofs)
118 goto lookup_neighbors;
119 }
120
121 while (*pnode) {
122 parent = *pnode;
123 re = rb_entry(*pnode, struct rb_entry, rb_node);
124
125 if (ofs < re->ofs)
126 pnode = &(*pnode)->rb_left;
127 else if (ofs >= re->ofs + re->len)
128 pnode = &(*pnode)->rb_right;
129 else
130 goto lookup_neighbors;
131 }
132
133 *insert_p = pnode;
134 *insert_parent = parent;
135
136 re = rb_entry(parent, struct rb_entry, rb_node);
137 tmp_node = parent;
138 if (parent && ofs > re->ofs)
139 tmp_node = rb_next(parent);
140 *next_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
141
142 tmp_node = parent;
143 if (parent && ofs < re->ofs)
144 tmp_node = rb_prev(parent);
145 *prev_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
146 return NULL;
147
148lookup_neighbors:
149 if (ofs == re->ofs || force) {
150 /* lookup prev node for merging backward later */
151 tmp_node = rb_prev(&re->rb_node);
152 *prev_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
153 }
154 if (ofs == re->ofs + re->len - 1 || force) {
155 /* lookup next node for merging frontward later */
156 tmp_node = rb_next(&re->rb_node);
157 *next_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
158 }
159 return re;
160}
161
162bool __check_rb_tree_consistence(struct f2fs_sb_info *sbi,
163 struct rb_root *root)
164{
165#ifdef CONFIG_F2FS_CHECK_FS
166 struct rb_node *cur = rb_first(root), *next;
167 struct rb_entry *cur_re, *next_re;
168
169 if (!cur)
170 return true;
171
172 while (cur) {
173 next = rb_next(cur);
174 if (!next)
175 return true;
176
177 cur_re = rb_entry(cur, struct rb_entry, rb_node);
178 next_re = rb_entry(next, struct rb_entry, rb_node);
179
180 if (cur_re->ofs + cur_re->len > next_re->ofs) {
181 f2fs_msg(sbi->sb, KERN_INFO, "inconsistent rbtree, "
182 "cur(%u, %u) next(%u, %u)",
183 cur_re->ofs, cur_re->len,
184 next_re->ofs, next_re->len);
185 return false;
186 }
187
188 cur = next;
189 }
190#endif
191 return true;
192}
193
194static struct kmem_cache *extent_tree_slab;
195static struct kmem_cache *extent_node_slab;
196
197static struct extent_node *__attach_extent_node(struct f2fs_sb_info *sbi,
198 struct extent_tree *et, struct extent_info *ei,
199 struct rb_node *parent, struct rb_node **p)
200{
201 struct extent_node *en;
202
203 en = kmem_cache_alloc(extent_node_slab, GFP_ATOMIC);
204 if (!en)
205 return NULL;
206
207 en->ei = *ei;
208 INIT_LIST_HEAD(&en->list);
209 en->et = et;
210
211 rb_link_node(&en->rb_node, parent, p);
212 rb_insert_color(&en->rb_node, &et->root);
213 atomic_inc(&et->node_cnt);
214 atomic_inc(&sbi->total_ext_node);
215 return en;
216}
217
218static void __detach_extent_node(struct f2fs_sb_info *sbi,
219 struct extent_tree *et, struct extent_node *en)
220{
221 rb_erase(&en->rb_node, &et->root);
222 atomic_dec(&et->node_cnt);
223 atomic_dec(&sbi->total_ext_node);
224
225 if (et->cached_en == en)
226 et->cached_en = NULL;
227 kmem_cache_free(extent_node_slab, en);
228}
229
230/*
231 * Flow to release an extent_node:
232 * 1. list_del_init
233 * 2. __detach_extent_node
234 * 3. kmem_cache_free.
235 */
236static void __release_extent_node(struct f2fs_sb_info *sbi,
237 struct extent_tree *et, struct extent_node *en)
238{
239 spin_lock(&sbi->extent_lock);
240 f2fs_bug_on(sbi, list_empty(&en->list));
241 list_del_init(&en->list);
242 spin_unlock(&sbi->extent_lock);
243
244 __detach_extent_node(sbi, et, en);
245}
246
247static struct extent_tree *__grab_extent_tree(struct inode *inode)
248{
249 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
250 struct extent_tree *et;
251 nid_t ino = inode->i_ino;
252
253 mutex_lock(&sbi->extent_tree_lock);
254 et = radix_tree_lookup(&sbi->extent_tree_root, ino);
255 if (!et) {
256 et = f2fs_kmem_cache_alloc(extent_tree_slab, GFP_NOFS);
257 f2fs_radix_tree_insert(&sbi->extent_tree_root, ino, et);
258 memset(et, 0, sizeof(struct extent_tree));
259 et->ino = ino;
260 et->root = RB_ROOT;
261 et->cached_en = NULL;
262 rwlock_init(&et->lock);
263 INIT_LIST_HEAD(&et->list);
264 atomic_set(&et->node_cnt, 0);
265 atomic_inc(&sbi->total_ext_tree);
266 } else {
267 atomic_dec(&sbi->total_zombie_tree);
268 list_del_init(&et->list);
269 }
270 mutex_unlock(&sbi->extent_tree_lock);
271
272 /* never died until evict_inode */
273 F2FS_I(inode)->extent_tree = et;
274
275 return et;
276}
277
278static struct extent_node *__init_extent_tree(struct f2fs_sb_info *sbi,
279 struct extent_tree *et, struct extent_info *ei)
280{
281 struct rb_node **p = &et->root.rb_node;
282 struct extent_node *en;
283
284 en = __attach_extent_node(sbi, et, ei, NULL, p);
285 if (!en)
286 return NULL;
287
288 et->largest = en->ei;
289 et->cached_en = en;
290 return en;
291}
292
293static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi,
294 struct extent_tree *et)
295{
296 struct rb_node *node, *next;
297 struct extent_node *en;
298 unsigned int count = atomic_read(&et->node_cnt);
299
300 node = rb_first(&et->root);
301 while (node) {
302 next = rb_next(node);
303 en = rb_entry(node, struct extent_node, rb_node);
304 __release_extent_node(sbi, et, en);
305 node = next;
306 }
307
308 return count - atomic_read(&et->node_cnt);
309}
310
311static void __drop_largest_extent(struct inode *inode,
312 pgoff_t fofs, unsigned int len)
313{
314 struct extent_info *largest = &F2FS_I(inode)->extent_tree->largest;
315
316 if (fofs < largest->fofs + largest->len && fofs + len > largest->fofs) {
317 largest->len = 0;
318 f2fs_mark_inode_dirty_sync(inode, true);
319 }
320}
321
322/* return true, if inode page is changed */
323static bool __f2fs_init_extent_tree(struct inode *inode, struct f2fs_extent *i_ext)
324{
325 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
326 struct extent_tree *et;
327 struct extent_node *en;
328 struct extent_info ei;
329
330 if (!f2fs_may_extent_tree(inode)) {
331 /* drop largest extent */
332 if (i_ext && i_ext->len) {
333 i_ext->len = 0;
334 return true;
335 }
336 return false;
337 }
338
339 et = __grab_extent_tree(inode);
340
341 if (!i_ext || !i_ext->len)
342 return false;
343
344 get_extent_info(&ei, i_ext);
345
346 write_lock(&et->lock);
347 if (atomic_read(&et->node_cnt))
348 goto out;
349
350 en = __init_extent_tree(sbi, et, &ei);
351 if (en) {
352 spin_lock(&sbi->extent_lock);
353 list_add_tail(&en->list, &sbi->extent_list);
354 spin_unlock(&sbi->extent_lock);
355 }
356out:
357 write_unlock(&et->lock);
358 return false;
359}
360
361bool f2fs_init_extent_tree(struct inode *inode, struct f2fs_extent *i_ext)
362{
363 bool ret = __f2fs_init_extent_tree(inode, i_ext);
364
365 if (!F2FS_I(inode)->extent_tree)
366 set_inode_flag(inode, FI_NO_EXTENT);
367
368 return ret;
369}
370
371static bool f2fs_lookup_extent_tree(struct inode *inode, pgoff_t pgofs,
372 struct extent_info *ei)
373{
374 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
375 struct extent_tree *et = F2FS_I(inode)->extent_tree;
376 struct extent_node *en;
377 bool ret = false;
378
379 f2fs_bug_on(sbi, !et);
380
381 trace_f2fs_lookup_extent_tree_start(inode, pgofs);
382
383 read_lock(&et->lock);
384
385 if (et->largest.fofs <= pgofs &&
386 et->largest.fofs + et->largest.len > pgofs) {
387 *ei = et->largest;
388 ret = true;
389 stat_inc_largest_node_hit(sbi);
390 goto out;
391 }
392
393 en = (struct extent_node *)__lookup_rb_tree(&et->root,
394 (struct rb_entry *)et->cached_en, pgofs);
395 if (!en)
396 goto out;
397
398 if (en == et->cached_en)
399 stat_inc_cached_node_hit(sbi);
400 else
401 stat_inc_rbtree_node_hit(sbi);
402
403 *ei = en->ei;
404 spin_lock(&sbi->extent_lock);
405 if (!list_empty(&en->list)) {
406 list_move_tail(&en->list, &sbi->extent_list);
407 et->cached_en = en;
408 }
409 spin_unlock(&sbi->extent_lock);
410 ret = true;
411out:
412 stat_inc_total_hit(sbi);
413 read_unlock(&et->lock);
414
415 trace_f2fs_lookup_extent_tree_end(inode, pgofs, ei);
416 return ret;
417}
418
419static struct extent_node *__try_merge_extent_node(struct inode *inode,
420 struct extent_tree *et, struct extent_info *ei,
421 struct extent_node *prev_ex,
422 struct extent_node *next_ex)
423{
424 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
425 struct extent_node *en = NULL;
426
427 if (prev_ex && __is_back_mergeable(ei, &prev_ex->ei)) {
428 prev_ex->ei.len += ei->len;
429 ei = &prev_ex->ei;
430 en = prev_ex;
431 }
432
433 if (next_ex && __is_front_mergeable(ei, &next_ex->ei)) {
434 next_ex->ei.fofs = ei->fofs;
435 next_ex->ei.blk = ei->blk;
436 next_ex->ei.len += ei->len;
437 if (en)
438 __release_extent_node(sbi, et, prev_ex);
439
440 en = next_ex;
441 }
442
443 if (!en)
444 return NULL;
445
446 __try_update_largest_extent(inode, et, en);
447
448 spin_lock(&sbi->extent_lock);
449 if (!list_empty(&en->list)) {
450 list_move_tail(&en->list, &sbi->extent_list);
451 et->cached_en = en;
452 }
453 spin_unlock(&sbi->extent_lock);
454 return en;
455}
456
457static struct extent_node *__insert_extent_tree(struct inode *inode,
458 struct extent_tree *et, struct extent_info *ei,
459 struct rb_node **insert_p,
460 struct rb_node *insert_parent)
461{
462 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
463 struct rb_node **p;
464 struct rb_node *parent = NULL;
465 struct extent_node *en = NULL;
466
467 if (insert_p && insert_parent) {
468 parent = insert_parent;
469 p = insert_p;
470 goto do_insert;
471 }
472
473 p = __lookup_rb_tree_for_insert(sbi, &et->root, &parent, ei->fofs);
474do_insert:
475 en = __attach_extent_node(sbi, et, ei, parent, p);
476 if (!en)
477 return NULL;
478
479 __try_update_largest_extent(inode, et, en);
480
481 /* update in global extent list */
482 spin_lock(&sbi->extent_lock);
483 list_add_tail(&en->list, &sbi->extent_list);
484 et->cached_en = en;
485 spin_unlock(&sbi->extent_lock);
486 return en;
487}
488
489static void f2fs_update_extent_tree_range(struct inode *inode,
490 pgoff_t fofs, block_t blkaddr, unsigned int len)
491{
492 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
493 struct extent_tree *et = F2FS_I(inode)->extent_tree;
494 struct extent_node *en = NULL, *en1 = NULL;
495 struct extent_node *prev_en = NULL, *next_en = NULL;
496 struct extent_info ei, dei, prev;
497 struct rb_node **insert_p = NULL, *insert_parent = NULL;
498 unsigned int end = fofs + len;
499 unsigned int pos = (unsigned int)fofs;
500
501 if (!et)
502 return;
503
504 trace_f2fs_update_extent_tree_range(inode, fofs, blkaddr, len);
505
506 write_lock(&et->lock);
507
508 if (is_inode_flag_set(inode, FI_NO_EXTENT)) {
509 write_unlock(&et->lock);
510 return;
511 }
512
513 prev = et->largest;
514 dei.len = 0;
515
516 /*
517 * drop largest extent before lookup, in case it's already
518 * been shrunk from extent tree
519 */
520 __drop_largest_extent(inode, fofs, len);
521
522 /* 1. lookup first extent node in range [fofs, fofs + len - 1] */
523 en = (struct extent_node *)__lookup_rb_tree_ret(&et->root,
524 (struct rb_entry *)et->cached_en, fofs,
525 (struct rb_entry **)&prev_en,
526 (struct rb_entry **)&next_en,
527 &insert_p, &insert_parent, false);
528 if (!en)
529 en = next_en;
530
531 /* 2. invlidate all extent nodes in range [fofs, fofs + len - 1] */
532 while (en && en->ei.fofs < end) {
533 unsigned int org_end;
534 int parts = 0; /* # of parts current extent split into */
535
536 next_en = en1 = NULL;
537
538 dei = en->ei;
539 org_end = dei.fofs + dei.len;
540 f2fs_bug_on(sbi, pos >= org_end);
541
542 if (pos > dei.fofs && pos - dei.fofs >= F2FS_MIN_EXTENT_LEN) {
543 en->ei.len = pos - en->ei.fofs;
544 prev_en = en;
545 parts = 1;
546 }
547
548 if (end < org_end && org_end - end >= F2FS_MIN_EXTENT_LEN) {
549 if (parts) {
550 set_extent_info(&ei, end,
551 end - dei.fofs + dei.blk,
552 org_end - end);
553 en1 = __insert_extent_tree(inode, et, &ei,
554 NULL, NULL);
555 next_en = en1;
556 } else {
557 en->ei.fofs = end;
558 en->ei.blk += end - dei.fofs;
559 en->ei.len -= end - dei.fofs;
560 next_en = en;
561 }
562 parts++;
563 }
564
565 if (!next_en) {
566 struct rb_node *node = rb_next(&en->rb_node);
567
568 next_en = rb_entry_safe(node, struct extent_node,
569 rb_node);
570 }
571
572 if (parts)
573 __try_update_largest_extent(inode, et, en);
574 else
575 __release_extent_node(sbi, et, en);
576
577 /*
578 * if original extent is split into zero or two parts, extent
579 * tree has been altered by deletion or insertion, therefore
580 * invalidate pointers regard to tree.
581 */
582 if (parts != 1) {
583 insert_p = NULL;
584 insert_parent = NULL;
585 }
586 en = next_en;
587 }
588
589 /* 3. update extent in extent cache */
590 if (blkaddr) {
591
592 set_extent_info(&ei, fofs, blkaddr, len);
593 if (!__try_merge_extent_node(inode, et, &ei, prev_en, next_en))
594 __insert_extent_tree(inode, et, &ei,
595 insert_p, insert_parent);
596
597 /* give up extent_cache, if split and small updates happen */
598 if (dei.len >= 1 &&
599 prev.len < F2FS_MIN_EXTENT_LEN &&
600 et->largest.len < F2FS_MIN_EXTENT_LEN) {
601 __drop_largest_extent(inode, 0, UINT_MAX);
602 set_inode_flag(inode, FI_NO_EXTENT);
603 }
604 }
605
606 if (is_inode_flag_set(inode, FI_NO_EXTENT))
607 __free_extent_tree(sbi, et);
608
609 write_unlock(&et->lock);
610}
611
612unsigned int f2fs_shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
613{
614 struct extent_tree *et, *next;
615 struct extent_node *en;
616 unsigned int node_cnt = 0, tree_cnt = 0;
617 int remained;
618
619 if (!test_opt(sbi, EXTENT_CACHE))
620 return 0;
621
622 if (!atomic_read(&sbi->total_zombie_tree))
623 goto free_node;
624
625 if (!mutex_trylock(&sbi->extent_tree_lock))
626 goto out;
627
628 /* 1. remove unreferenced extent tree */
629 list_for_each_entry_safe(et, next, &sbi->zombie_list, list) {
630 if (atomic_read(&et->node_cnt)) {
631 write_lock(&et->lock);
632 node_cnt += __free_extent_tree(sbi, et);
633 write_unlock(&et->lock);
634 }
635 f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
636 list_del_init(&et->list);
637 radix_tree_delete(&sbi->extent_tree_root, et->ino);
638 kmem_cache_free(extent_tree_slab, et);
639 atomic_dec(&sbi->total_ext_tree);
640 atomic_dec(&sbi->total_zombie_tree);
641 tree_cnt++;
642
643 if (node_cnt + tree_cnt >= nr_shrink)
644 goto unlock_out;
645 cond_resched();
646 }
647 mutex_unlock(&sbi->extent_tree_lock);
648
649free_node:
650 /* 2. remove LRU extent entries */
651 if (!mutex_trylock(&sbi->extent_tree_lock))
652 goto out;
653
654 remained = nr_shrink - (node_cnt + tree_cnt);
655
656 spin_lock(&sbi->extent_lock);
657 for (; remained > 0; remained--) {
658 if (list_empty(&sbi->extent_list))
659 break;
660 en = list_first_entry(&sbi->extent_list,
661 struct extent_node, list);
662 et = en->et;
663 if (!write_trylock(&et->lock)) {
664 /* refresh this extent node's position in extent list */
665 list_move_tail(&en->list, &sbi->extent_list);
666 continue;
667 }
668
669 list_del_init(&en->list);
670 spin_unlock(&sbi->extent_lock);
671
672 __detach_extent_node(sbi, et, en);
673
674 write_unlock(&et->lock);
675 node_cnt++;
676 spin_lock(&sbi->extent_lock);
677 }
678 spin_unlock(&sbi->extent_lock);
679
680unlock_out:
681 mutex_unlock(&sbi->extent_tree_lock);
682out:
683 trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt);
684
685 return node_cnt + tree_cnt;
686}
687
688unsigned int f2fs_destroy_extent_node(struct inode *inode)
689{
690 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
691 struct extent_tree *et = F2FS_I(inode)->extent_tree;
692 unsigned int node_cnt = 0;
693
694 if (!et || !atomic_read(&et->node_cnt))
695 return 0;
696
697 write_lock(&et->lock);
698 node_cnt = __free_extent_tree(sbi, et);
699 write_unlock(&et->lock);
700
701 return node_cnt;
702}
703
704void f2fs_drop_extent_tree(struct inode *inode)
705{
706 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
707 struct extent_tree *et = F2FS_I(inode)->extent_tree;
708
709 if (!f2fs_may_extent_tree(inode))
710 return;
711
712 set_inode_flag(inode, FI_NO_EXTENT);
713
714 write_lock(&et->lock);
715 __free_extent_tree(sbi, et);
716 __drop_largest_extent(inode, 0, UINT_MAX);
717 write_unlock(&et->lock);
718}
719
720void f2fs_destroy_extent_tree(struct inode *inode)
721{
722 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
723 struct extent_tree *et = F2FS_I(inode)->extent_tree;
724 unsigned int node_cnt = 0;
725
726 if (!et)
727 return;
728
729 if (inode->i_nlink && !is_bad_inode(inode) &&
730 atomic_read(&et->node_cnt)) {
731 mutex_lock(&sbi->extent_tree_lock);
732 list_add_tail(&et->list, &sbi->zombie_list);
733 atomic_inc(&sbi->total_zombie_tree);
734 mutex_unlock(&sbi->extent_tree_lock);
735 return;
736 }
737
738 /* free all extent info belong to this extent tree */
739 node_cnt = f2fs_destroy_extent_node(inode);
740
741 /* delete extent tree entry in radix tree */
742 mutex_lock(&sbi->extent_tree_lock);
743 f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
744 radix_tree_delete(&sbi->extent_tree_root, inode->i_ino);
745 kmem_cache_free(extent_tree_slab, et);
746 atomic_dec(&sbi->total_ext_tree);
747 mutex_unlock(&sbi->extent_tree_lock);
748
749 F2FS_I(inode)->extent_tree = NULL;
750
751 trace_f2fs_destroy_extent_tree(inode, node_cnt);
752}
753
754bool f2fs_lookup_extent_cache(struct inode *inode, pgoff_t pgofs,
755 struct extent_info *ei)
756{
757 if (!f2fs_may_extent_tree(inode))
758 return false;
759
760 return f2fs_lookup_extent_tree(inode, pgofs, ei);
761}
762
763void f2fs_update_extent_cache(struct dnode_of_data *dn)
764{
765 pgoff_t fofs;
766 block_t blkaddr;
767
768 if (!f2fs_may_extent_tree(dn->inode))
769 return;
770
771 if (dn->data_blkaddr == NEW_ADDR)
772 blkaddr = NULL_ADDR;
773 else
774 blkaddr = dn->data_blkaddr;
775
776 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
777 dn->ofs_in_node;
778 f2fs_update_extent_tree_range(dn->inode, fofs, blkaddr, 1);
779}
780
781void f2fs_update_extent_cache_range(struct dnode_of_data *dn,
782 pgoff_t fofs, block_t blkaddr, unsigned int len)
783
784{
785 if (!f2fs_may_extent_tree(dn->inode))
786 return;
787
788 f2fs_update_extent_tree_range(dn->inode, fofs, blkaddr, len);
789}
790
791void init_extent_cache_info(struct f2fs_sb_info *sbi)
792{
793 INIT_RADIX_TREE(&sbi->extent_tree_root, GFP_NOIO);
794 mutex_init(&sbi->extent_tree_lock);
795 INIT_LIST_HEAD(&sbi->extent_list);
796 spin_lock_init(&sbi->extent_lock);
797 atomic_set(&sbi->total_ext_tree, 0);
798 INIT_LIST_HEAD(&sbi->zombie_list);
799 atomic_set(&sbi->total_zombie_tree, 0);
800 atomic_set(&sbi->total_ext_node, 0);
801}
802
803int __init create_extent_cache(void)
804{
805 extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree",
806 sizeof(struct extent_tree));
807 if (!extent_tree_slab)
808 return -ENOMEM;
809 extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node",
810 sizeof(struct extent_node));
811 if (!extent_node_slab) {
812 kmem_cache_destroy(extent_tree_slab);
813 return -ENOMEM;
814 }
815 return 0;
816}
817
818void destroy_extent_cache(void)
819{
820 kmem_cache_destroy(extent_node_slab);
821 kmem_cache_destroy(extent_tree_slab);
822}