1 // SPDX-License-Identifier: GPL-2.0
3 * f2fs extent cache support
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>
12 #include <linux/f2fs_fs.h>
16 #include <trace/events/f2fs.h>
18 static struct rb_entry *__lookup_rb_tree_fast(struct rb_entry *cached_re,
22 if (cached_re->ofs <= ofs &&
23 cached_re->ofs + cached_re->len > ofs) {
30 static struct rb_entry *__lookup_rb_tree_slow(struct rb_root_cached *root,
33 struct rb_node *node = root->rb_root.rb_node;
37 re = rb_entry(node, struct rb_entry, rb_node);
41 else if (ofs >= re->ofs + re->len)
42 node = node->rb_right;
49 struct rb_entry *f2fs_lookup_rb_tree(struct rb_root_cached *root,
50 struct rb_entry *cached_re, unsigned int ofs)
54 re = __lookup_rb_tree_fast(cached_re, ofs);
56 return __lookup_rb_tree_slow(root, ofs);
61 struct rb_node **f2fs_lookup_rb_tree_for_insert(struct f2fs_sb_info *sbi,
62 struct rb_root_cached *root,
63 struct rb_node **parent,
64 unsigned int ofs, bool *leftmost)
66 struct rb_node **p = &root->rb_root.rb_node;
71 re = rb_entry(*parent, struct rb_entry, rb_node);
75 } else if (ofs >= re->ofs + re->len) {
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.
95 struct rb_entry *f2fs_lookup_rb_tree_ret(struct rb_root_cached *root,
96 struct rb_entry *cached_re,
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, bool *leftmost)
104 struct rb_node **pnode = &root->rb_root.rb_node;
105 struct rb_node *parent = NULL, *tmp_node;
106 struct rb_entry *re = cached_re;
109 *insert_parent = NULL;
113 if (RB_EMPTY_ROOT(&root->rb_root))
117 if (re->ofs <= ofs && re->ofs + re->len > ofs)
118 goto lookup_neighbors;
126 re = rb_entry(*pnode, struct rb_entry, rb_node);
129 pnode = &(*pnode)->rb_left;
130 } else if (ofs >= re->ofs + re->len) {
131 pnode = &(*pnode)->rb_right;
135 goto lookup_neighbors;
140 *insert_parent = parent;
142 re = rb_entry(parent, struct rb_entry, rb_node);
144 if (parent && ofs > re->ofs)
145 tmp_node = rb_next(parent);
146 *next_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
149 if (parent && ofs < re->ofs)
150 tmp_node = rb_prev(parent);
151 *prev_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
155 if (ofs == re->ofs || force) {
156 /* lookup prev node for merging backward later */
157 tmp_node = rb_prev(&re->rb_node);
158 *prev_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
160 if (ofs == re->ofs + re->len - 1 || force) {
161 /* lookup next node for merging frontward later */
162 tmp_node = rb_next(&re->rb_node);
163 *next_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
168 bool f2fs_check_rb_tree_consistence(struct f2fs_sb_info *sbi,
169 struct rb_root_cached *root)
171 #ifdef CONFIG_F2FS_CHECK_FS
172 struct rb_node *cur = rb_first_cached(root), *next;
173 struct rb_entry *cur_re, *next_re;
183 cur_re = rb_entry(cur, struct rb_entry, rb_node);
184 next_re = rb_entry(next, struct rb_entry, rb_node);
186 if (cur_re->ofs + cur_re->len > next_re->ofs) {
187 f2fs_info(sbi, "inconsistent rbtree, cur(%u, %u) next(%u, %u)",
188 cur_re->ofs, cur_re->len,
189 next_re->ofs, next_re->len);
199 static struct kmem_cache *extent_tree_slab;
200 static struct kmem_cache *extent_node_slab;
202 static struct extent_node *__attach_extent_node(struct f2fs_sb_info *sbi,
203 struct extent_tree *et, struct extent_info *ei,
204 struct rb_node *parent, struct rb_node **p,
207 struct extent_node *en;
209 en = kmem_cache_alloc(extent_node_slab, GFP_ATOMIC);
214 INIT_LIST_HEAD(&en->list);
217 rb_link_node(&en->rb_node, parent, p);
218 rb_insert_color_cached(&en->rb_node, &et->root, leftmost);
219 atomic_inc(&et->node_cnt);
220 atomic_inc(&sbi->total_ext_node);
224 static void __detach_extent_node(struct f2fs_sb_info *sbi,
225 struct extent_tree *et, struct extent_node *en)
227 rb_erase_cached(&en->rb_node, &et->root);
228 atomic_dec(&et->node_cnt);
229 atomic_dec(&sbi->total_ext_node);
231 if (et->cached_en == en)
232 et->cached_en = NULL;
233 kmem_cache_free(extent_node_slab, en);
237 * Flow to release an extent_node:
239 * 2. __detach_extent_node
240 * 3. kmem_cache_free.
242 static void __release_extent_node(struct f2fs_sb_info *sbi,
243 struct extent_tree *et, struct extent_node *en)
245 spin_lock(&sbi->extent_lock);
246 f2fs_bug_on(sbi, list_empty(&en->list));
247 list_del_init(&en->list);
248 spin_unlock(&sbi->extent_lock);
250 __detach_extent_node(sbi, et, en);
253 static struct extent_tree *__grab_extent_tree(struct inode *inode)
255 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
256 struct extent_tree *et;
257 nid_t ino = inode->i_ino;
259 mutex_lock(&sbi->extent_tree_lock);
260 et = radix_tree_lookup(&sbi->extent_tree_root, ino);
262 et = f2fs_kmem_cache_alloc(extent_tree_slab, GFP_NOFS);
263 f2fs_radix_tree_insert(&sbi->extent_tree_root, ino, et);
264 memset(et, 0, sizeof(struct extent_tree));
266 et->root = RB_ROOT_CACHED;
267 et->cached_en = NULL;
268 rwlock_init(&et->lock);
269 INIT_LIST_HEAD(&et->list);
270 atomic_set(&et->node_cnt, 0);
271 atomic_inc(&sbi->total_ext_tree);
273 atomic_dec(&sbi->total_zombie_tree);
274 list_del_init(&et->list);
276 mutex_unlock(&sbi->extent_tree_lock);
278 /* never died until evict_inode */
279 F2FS_I(inode)->extent_tree = et;
284 static struct extent_node *__init_extent_tree(struct f2fs_sb_info *sbi,
285 struct extent_tree *et, struct extent_info *ei)
287 struct rb_node **p = &et->root.rb_root.rb_node;
288 struct extent_node *en;
290 en = __attach_extent_node(sbi, et, ei, NULL, p, true);
294 et->largest = en->ei;
299 static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi,
300 struct extent_tree *et)
302 struct rb_node *node, *next;
303 struct extent_node *en;
304 unsigned int count = atomic_read(&et->node_cnt);
306 node = rb_first_cached(&et->root);
308 next = rb_next(node);
309 en = rb_entry(node, struct extent_node, rb_node);
310 __release_extent_node(sbi, et, en);
314 return count - atomic_read(&et->node_cnt);
317 static void __drop_largest_extent(struct extent_tree *et,
318 pgoff_t fofs, unsigned int len)
320 if (fofs < et->largest.fofs + et->largest.len &&
321 fofs + len > et->largest.fofs) {
323 et->largest_updated = true;
327 /* return true, if inode page is changed */
328 static bool __f2fs_init_extent_tree(struct inode *inode, struct f2fs_extent *i_ext)
330 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
331 struct extent_tree *et;
332 struct extent_node *en;
333 struct extent_info ei;
335 if (!f2fs_may_extent_tree(inode)) {
336 /* drop largest extent */
337 if (i_ext && i_ext->len) {
344 et = __grab_extent_tree(inode);
346 if (!i_ext || !i_ext->len)
349 get_extent_info(&ei, i_ext);
351 write_lock(&et->lock);
352 if (atomic_read(&et->node_cnt))
355 en = __init_extent_tree(sbi, et, &ei);
357 spin_lock(&sbi->extent_lock);
358 list_add_tail(&en->list, &sbi->extent_list);
359 spin_unlock(&sbi->extent_lock);
362 write_unlock(&et->lock);
366 bool f2fs_init_extent_tree(struct inode *inode, struct f2fs_extent *i_ext)
368 bool ret = __f2fs_init_extent_tree(inode, i_ext);
370 if (!F2FS_I(inode)->extent_tree)
371 set_inode_flag(inode, FI_NO_EXTENT);
376 static bool f2fs_lookup_extent_tree(struct inode *inode, pgoff_t pgofs,
377 struct extent_info *ei)
379 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
380 struct extent_tree *et = F2FS_I(inode)->extent_tree;
381 struct extent_node *en;
384 f2fs_bug_on(sbi, !et);
386 trace_f2fs_lookup_extent_tree_start(inode, pgofs);
388 read_lock(&et->lock);
390 if (et->largest.fofs <= pgofs &&
391 et->largest.fofs + et->largest.len > pgofs) {
394 stat_inc_largest_node_hit(sbi);
398 en = (struct extent_node *)f2fs_lookup_rb_tree(&et->root,
399 (struct rb_entry *)et->cached_en, pgofs);
403 if (en == et->cached_en)
404 stat_inc_cached_node_hit(sbi);
406 stat_inc_rbtree_node_hit(sbi);
409 spin_lock(&sbi->extent_lock);
410 if (!list_empty(&en->list)) {
411 list_move_tail(&en->list, &sbi->extent_list);
414 spin_unlock(&sbi->extent_lock);
417 stat_inc_total_hit(sbi);
418 read_unlock(&et->lock);
420 trace_f2fs_lookup_extent_tree_end(inode, pgofs, ei);
424 static struct extent_node *__try_merge_extent_node(struct f2fs_sb_info *sbi,
425 struct extent_tree *et, struct extent_info *ei,
426 struct extent_node *prev_ex,
427 struct extent_node *next_ex)
429 struct extent_node *en = NULL;
431 if (prev_ex && __is_back_mergeable(ei, &prev_ex->ei)) {
432 prev_ex->ei.len += ei->len;
437 if (next_ex && __is_front_mergeable(ei, &next_ex->ei)) {
438 next_ex->ei.fofs = ei->fofs;
439 next_ex->ei.blk = ei->blk;
440 next_ex->ei.len += ei->len;
442 __release_extent_node(sbi, et, prev_ex);
450 __try_update_largest_extent(et, en);
452 spin_lock(&sbi->extent_lock);
453 if (!list_empty(&en->list)) {
454 list_move_tail(&en->list, &sbi->extent_list);
457 spin_unlock(&sbi->extent_lock);
461 static struct extent_node *__insert_extent_tree(struct f2fs_sb_info *sbi,
462 struct extent_tree *et, struct extent_info *ei,
463 struct rb_node **insert_p,
464 struct rb_node *insert_parent,
468 struct rb_node *parent = NULL;
469 struct extent_node *en = NULL;
471 if (insert_p && insert_parent) {
472 parent = insert_parent;
479 p = f2fs_lookup_rb_tree_for_insert(sbi, &et->root, &parent,
480 ei->fofs, &leftmost);
482 en = __attach_extent_node(sbi, et, ei, parent, p, leftmost);
486 __try_update_largest_extent(et, en);
488 /* update in global extent list */
489 spin_lock(&sbi->extent_lock);
490 list_add_tail(&en->list, &sbi->extent_list);
492 spin_unlock(&sbi->extent_lock);
496 static void f2fs_update_extent_tree_range(struct inode *inode,
497 pgoff_t fofs, block_t blkaddr, unsigned int len)
499 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
500 struct extent_tree *et = F2FS_I(inode)->extent_tree;
501 struct extent_node *en = NULL, *en1 = NULL;
502 struct extent_node *prev_en = NULL, *next_en = NULL;
503 struct extent_info ei, dei, prev;
504 struct rb_node **insert_p = NULL, *insert_parent = NULL;
505 unsigned int end = fofs + len;
506 unsigned int pos = (unsigned int)fofs;
507 bool updated = false;
508 bool leftmost = false;
513 trace_f2fs_update_extent_tree_range(inode, fofs, blkaddr, len);
515 write_lock(&et->lock);
517 if (is_inode_flag_set(inode, FI_NO_EXTENT)) {
518 write_unlock(&et->lock);
526 * drop largest extent before lookup, in case it's already
527 * been shrunk from extent tree
529 __drop_largest_extent(et, fofs, len);
531 /* 1. lookup first extent node in range [fofs, fofs + len - 1] */
532 en = (struct extent_node *)f2fs_lookup_rb_tree_ret(&et->root,
533 (struct rb_entry *)et->cached_en, fofs,
534 (struct rb_entry **)&prev_en,
535 (struct rb_entry **)&next_en,
536 &insert_p, &insert_parent, false,
541 /* 2. invlidate all extent nodes in range [fofs, fofs + len - 1] */
542 while (en && en->ei.fofs < end) {
543 unsigned int org_end;
544 int parts = 0; /* # of parts current extent split into */
546 next_en = en1 = NULL;
549 org_end = dei.fofs + dei.len;
550 f2fs_bug_on(sbi, pos >= org_end);
552 if (pos > dei.fofs && pos - dei.fofs >= F2FS_MIN_EXTENT_LEN) {
553 en->ei.len = pos - en->ei.fofs;
558 if (end < org_end && org_end - end >= F2FS_MIN_EXTENT_LEN) {
560 set_extent_info(&ei, end,
561 end - dei.fofs + dei.blk,
563 en1 = __insert_extent_tree(sbi, et, &ei,
568 en->ei.blk += end - dei.fofs;
569 en->ei.len -= end - dei.fofs;
576 struct rb_node *node = rb_next(&en->rb_node);
578 next_en = rb_entry_safe(node, struct extent_node,
583 __try_update_largest_extent(et, en);
585 __release_extent_node(sbi, et, en);
588 * if original extent is split into zero or two parts, extent
589 * tree has been altered by deletion or insertion, therefore
590 * invalidate pointers regard to tree.
594 insert_parent = NULL;
599 /* 3. update extent in extent cache */
602 set_extent_info(&ei, fofs, blkaddr, len);
603 if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
604 __insert_extent_tree(sbi, et, &ei,
605 insert_p, insert_parent, leftmost);
607 /* give up extent_cache, if split and small updates happen */
609 prev.len < F2FS_MIN_EXTENT_LEN &&
610 et->largest.len < F2FS_MIN_EXTENT_LEN) {
612 et->largest_updated = true;
613 set_inode_flag(inode, FI_NO_EXTENT);
617 if (is_inode_flag_set(inode, FI_NO_EXTENT))
618 __free_extent_tree(sbi, et);
620 if (et->largest_updated) {
621 et->largest_updated = false;
625 write_unlock(&et->lock);
628 f2fs_mark_inode_dirty_sync(inode, true);
631 unsigned int f2fs_shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
633 struct extent_tree *et, *next;
634 struct extent_node *en;
635 unsigned int node_cnt = 0, tree_cnt = 0;
638 if (!test_opt(sbi, EXTENT_CACHE))
641 if (!atomic_read(&sbi->total_zombie_tree))
644 if (!mutex_trylock(&sbi->extent_tree_lock))
647 /* 1. remove unreferenced extent tree */
648 list_for_each_entry_safe(et, next, &sbi->zombie_list, list) {
649 if (atomic_read(&et->node_cnt)) {
650 write_lock(&et->lock);
651 node_cnt += __free_extent_tree(sbi, et);
652 write_unlock(&et->lock);
654 f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
655 list_del_init(&et->list);
656 radix_tree_delete(&sbi->extent_tree_root, et->ino);
657 kmem_cache_free(extent_tree_slab, et);
658 atomic_dec(&sbi->total_ext_tree);
659 atomic_dec(&sbi->total_zombie_tree);
662 if (node_cnt + tree_cnt >= nr_shrink)
666 mutex_unlock(&sbi->extent_tree_lock);
669 /* 2. remove LRU extent entries */
670 if (!mutex_trylock(&sbi->extent_tree_lock))
673 remained = nr_shrink - (node_cnt + tree_cnt);
675 spin_lock(&sbi->extent_lock);
676 for (; remained > 0; remained--) {
677 if (list_empty(&sbi->extent_list))
679 en = list_first_entry(&sbi->extent_list,
680 struct extent_node, list);
682 if (!write_trylock(&et->lock)) {
683 /* refresh this extent node's position in extent list */
684 list_move_tail(&en->list, &sbi->extent_list);
688 list_del_init(&en->list);
689 spin_unlock(&sbi->extent_lock);
691 __detach_extent_node(sbi, et, en);
693 write_unlock(&et->lock);
695 spin_lock(&sbi->extent_lock);
697 spin_unlock(&sbi->extent_lock);
700 mutex_unlock(&sbi->extent_tree_lock);
702 trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt);
704 return node_cnt + tree_cnt;
707 unsigned int f2fs_destroy_extent_node(struct inode *inode)
709 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
710 struct extent_tree *et = F2FS_I(inode)->extent_tree;
711 unsigned int node_cnt = 0;
713 if (!et || !atomic_read(&et->node_cnt))
716 write_lock(&et->lock);
717 node_cnt = __free_extent_tree(sbi, et);
718 write_unlock(&et->lock);
723 void f2fs_drop_extent_tree(struct inode *inode)
725 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
726 struct extent_tree *et = F2FS_I(inode)->extent_tree;
727 bool updated = false;
729 if (!f2fs_may_extent_tree(inode))
732 set_inode_flag(inode, FI_NO_EXTENT);
734 write_lock(&et->lock);
735 __free_extent_tree(sbi, et);
736 if (et->largest.len) {
740 write_unlock(&et->lock);
742 f2fs_mark_inode_dirty_sync(inode, true);
745 void f2fs_destroy_extent_tree(struct inode *inode)
747 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
748 struct extent_tree *et = F2FS_I(inode)->extent_tree;
749 unsigned int node_cnt = 0;
754 if (inode->i_nlink && !is_bad_inode(inode) &&
755 atomic_read(&et->node_cnt)) {
756 mutex_lock(&sbi->extent_tree_lock);
757 list_add_tail(&et->list, &sbi->zombie_list);
758 atomic_inc(&sbi->total_zombie_tree);
759 mutex_unlock(&sbi->extent_tree_lock);
763 /* free all extent info belong to this extent tree */
764 node_cnt = f2fs_destroy_extent_node(inode);
766 /* delete extent tree entry in radix tree */
767 mutex_lock(&sbi->extent_tree_lock);
768 f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
769 radix_tree_delete(&sbi->extent_tree_root, inode->i_ino);
770 kmem_cache_free(extent_tree_slab, et);
771 atomic_dec(&sbi->total_ext_tree);
772 mutex_unlock(&sbi->extent_tree_lock);
774 F2FS_I(inode)->extent_tree = NULL;
776 trace_f2fs_destroy_extent_tree(inode, node_cnt);
779 bool f2fs_lookup_extent_cache(struct inode *inode, pgoff_t pgofs,
780 struct extent_info *ei)
782 if (!f2fs_may_extent_tree(inode))
785 return f2fs_lookup_extent_tree(inode, pgofs, ei);
788 void f2fs_update_extent_cache(struct dnode_of_data *dn)
793 if (!f2fs_may_extent_tree(dn->inode))
796 if (dn->data_blkaddr == NEW_ADDR)
799 blkaddr = dn->data_blkaddr;
801 fofs = f2fs_start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
803 f2fs_update_extent_tree_range(dn->inode, fofs, blkaddr, 1);
806 void f2fs_update_extent_cache_range(struct dnode_of_data *dn,
807 pgoff_t fofs, block_t blkaddr, unsigned int len)
810 if (!f2fs_may_extent_tree(dn->inode))
813 f2fs_update_extent_tree_range(dn->inode, fofs, blkaddr, len);
816 void f2fs_init_extent_cache_info(struct f2fs_sb_info *sbi)
818 INIT_RADIX_TREE(&sbi->extent_tree_root, GFP_NOIO);
819 mutex_init(&sbi->extent_tree_lock);
820 INIT_LIST_HEAD(&sbi->extent_list);
821 spin_lock_init(&sbi->extent_lock);
822 atomic_set(&sbi->total_ext_tree, 0);
823 INIT_LIST_HEAD(&sbi->zombie_list);
824 atomic_set(&sbi->total_zombie_tree, 0);
825 atomic_set(&sbi->total_ext_node, 0);
828 int __init f2fs_create_extent_cache(void)
830 extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree",
831 sizeof(struct extent_tree));
832 if (!extent_tree_slab)
834 extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node",
835 sizeof(struct extent_node));
836 if (!extent_node_slab) {
837 kmem_cache_destroy(extent_tree_slab);
843 void f2fs_destroy_extent_cache(void)
845 kmem_cache_destroy(extent_node_slab);
846 kmem_cache_destroy(extent_tree_slab);