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 *root,
33 struct rb_node *node = 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 *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 *root, struct rb_node **parent,
65 struct rb_node **p = &root->rb_node;
70 re = rb_entry(*parent, struct rb_entry, rb_node);
74 else if (ofs >= re->ofs + re->len)
84 * lookup rb entry in position of @ofs in rb-tree,
85 * if hit, return the entry, otherwise, return NULL
86 * @prev_ex: extent before ofs
87 * @next_ex: extent after ofs
88 * @insert_p: insert point for new extent at ofs
89 * in order to simpfy the insertion after.
90 * tree must stay unchanged between lookup and insertion.
92 struct rb_entry *f2fs_lookup_rb_tree_ret(struct rb_root *root,
93 struct rb_entry *cached_re,
95 struct rb_entry **prev_entry,
96 struct rb_entry **next_entry,
97 struct rb_node ***insert_p,
98 struct rb_node **insert_parent,
101 struct rb_node **pnode = &root->rb_node;
102 struct rb_node *parent = NULL, *tmp_node;
103 struct rb_entry *re = cached_re;
106 *insert_parent = NULL;
110 if (RB_EMPTY_ROOT(root))
114 if (re->ofs <= ofs && re->ofs + re->len > ofs)
115 goto lookup_neighbors;
120 re = rb_entry(*pnode, struct rb_entry, rb_node);
123 pnode = &(*pnode)->rb_left;
124 else if (ofs >= re->ofs + re->len)
125 pnode = &(*pnode)->rb_right;
127 goto lookup_neighbors;
131 *insert_parent = parent;
133 re = rb_entry(parent, struct rb_entry, rb_node);
135 if (parent && ofs > re->ofs)
136 tmp_node = rb_next(parent);
137 *next_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
140 if (parent && ofs < re->ofs)
141 tmp_node = rb_prev(parent);
142 *prev_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
146 if (ofs == re->ofs || force) {
147 /* lookup prev node for merging backward later */
148 tmp_node = rb_prev(&re->rb_node);
149 *prev_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
151 if (ofs == re->ofs + re->len - 1 || force) {
152 /* lookup next node for merging frontward later */
153 tmp_node = rb_next(&re->rb_node);
154 *next_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
159 bool f2fs_check_rb_tree_consistence(struct f2fs_sb_info *sbi,
160 struct rb_root *root)
162 #ifdef CONFIG_F2FS_CHECK_FS
163 struct rb_node *cur = rb_first(root), *next;
164 struct rb_entry *cur_re, *next_re;
174 cur_re = rb_entry(cur, struct rb_entry, rb_node);
175 next_re = rb_entry(next, struct rb_entry, rb_node);
177 if (cur_re->ofs + cur_re->len > next_re->ofs) {
178 f2fs_info(sbi, "inconsistent rbtree, cur(%u, %u) next(%u, %u)",
179 cur_re->ofs, cur_re->len,
180 next_re->ofs, next_re->len);
190 static struct kmem_cache *extent_tree_slab;
191 static struct kmem_cache *extent_node_slab;
193 static struct extent_node *__attach_extent_node(struct f2fs_sb_info *sbi,
194 struct extent_tree *et, struct extent_info *ei,
195 struct rb_node *parent, struct rb_node **p)
197 struct extent_node *en;
199 en = kmem_cache_alloc(extent_node_slab, GFP_ATOMIC);
204 INIT_LIST_HEAD(&en->list);
207 rb_link_node(&en->rb_node, parent, p);
208 rb_insert_color(&en->rb_node, &et->root);
209 atomic_inc(&et->node_cnt);
210 atomic_inc(&sbi->total_ext_node);
214 static void __detach_extent_node(struct f2fs_sb_info *sbi,
215 struct extent_tree *et, struct extent_node *en)
217 rb_erase(&en->rb_node, &et->root);
218 atomic_dec(&et->node_cnt);
219 atomic_dec(&sbi->total_ext_node);
221 if (et->cached_en == en)
222 et->cached_en = NULL;
223 kmem_cache_free(extent_node_slab, en);
227 * Flow to release an extent_node:
229 * 2. __detach_extent_node
230 * 3. kmem_cache_free.
232 static void __release_extent_node(struct f2fs_sb_info *sbi,
233 struct extent_tree *et, struct extent_node *en)
235 spin_lock(&sbi->extent_lock);
236 f2fs_bug_on(sbi, list_empty(&en->list));
237 list_del_init(&en->list);
238 spin_unlock(&sbi->extent_lock);
240 __detach_extent_node(sbi, et, en);
243 static struct extent_tree *__grab_extent_tree(struct inode *inode)
245 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
246 struct extent_tree *et;
247 nid_t ino = inode->i_ino;
249 mutex_lock(&sbi->extent_tree_lock);
250 et = radix_tree_lookup(&sbi->extent_tree_root, ino);
252 et = f2fs_kmem_cache_alloc(extent_tree_slab, GFP_NOFS);
253 f2fs_radix_tree_insert(&sbi->extent_tree_root, ino, et);
254 memset(et, 0, sizeof(struct extent_tree));
257 et->cached_en = NULL;
258 rwlock_init(&et->lock);
259 INIT_LIST_HEAD(&et->list);
260 atomic_set(&et->node_cnt, 0);
261 atomic_inc(&sbi->total_ext_tree);
263 atomic_dec(&sbi->total_zombie_tree);
264 list_del_init(&et->list);
266 mutex_unlock(&sbi->extent_tree_lock);
268 /* never died until evict_inode */
269 F2FS_I(inode)->extent_tree = et;
274 static struct extent_node *__init_extent_tree(struct f2fs_sb_info *sbi,
275 struct extent_tree *et, struct extent_info *ei)
277 struct rb_node **p = &et->root.rb_node;
278 struct extent_node *en;
280 en = __attach_extent_node(sbi, et, ei, NULL, p);
284 et->largest = en->ei;
289 static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi,
290 struct extent_tree *et)
292 struct rb_node *node, *next;
293 struct extent_node *en;
294 unsigned int count = atomic_read(&et->node_cnt);
296 node = rb_first(&et->root);
298 next = rb_next(node);
299 en = rb_entry(node, struct extent_node, rb_node);
300 __release_extent_node(sbi, et, en);
304 return count - atomic_read(&et->node_cnt);
307 static void __drop_largest_extent(struct extent_tree *et,
308 pgoff_t fofs, unsigned int len)
310 if (fofs < et->largest.fofs + et->largest.len &&
311 fofs + len > et->largest.fofs) {
313 et->largest_updated = true;
317 /* return true, if inode page is changed */
318 static bool __f2fs_init_extent_tree(struct inode *inode, struct f2fs_extent *i_ext)
320 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
321 struct extent_tree *et;
322 struct extent_node *en;
323 struct extent_info ei;
325 if (!f2fs_may_extent_tree(inode)) {
326 /* drop largest extent */
327 if (i_ext && i_ext->len) {
334 et = __grab_extent_tree(inode);
336 if (!i_ext || !i_ext->len)
339 get_extent_info(&ei, i_ext);
341 write_lock(&et->lock);
342 if (atomic_read(&et->node_cnt))
345 en = __init_extent_tree(sbi, et, &ei);
347 spin_lock(&sbi->extent_lock);
348 list_add_tail(&en->list, &sbi->extent_list);
349 spin_unlock(&sbi->extent_lock);
352 write_unlock(&et->lock);
356 bool f2fs_init_extent_tree(struct inode *inode, struct f2fs_extent *i_ext)
358 bool ret = __f2fs_init_extent_tree(inode, i_ext);
360 if (!F2FS_I(inode)->extent_tree)
361 set_inode_flag(inode, FI_NO_EXTENT);
366 static bool f2fs_lookup_extent_tree(struct inode *inode, pgoff_t pgofs,
367 struct extent_info *ei)
369 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
370 struct extent_tree *et = F2FS_I(inode)->extent_tree;
371 struct extent_node *en;
374 f2fs_bug_on(sbi, !et);
376 trace_f2fs_lookup_extent_tree_start(inode, pgofs);
378 read_lock(&et->lock);
380 if (et->largest.fofs <= pgofs &&
381 et->largest.fofs + et->largest.len > pgofs) {
384 stat_inc_largest_node_hit(sbi);
388 en = (struct extent_node *)f2fs_lookup_rb_tree(&et->root,
389 (struct rb_entry *)et->cached_en, pgofs);
393 if (en == et->cached_en)
394 stat_inc_cached_node_hit(sbi);
396 stat_inc_rbtree_node_hit(sbi);
399 spin_lock(&sbi->extent_lock);
400 if (!list_empty(&en->list)) {
401 list_move_tail(&en->list, &sbi->extent_list);
404 spin_unlock(&sbi->extent_lock);
407 stat_inc_total_hit(sbi);
408 read_unlock(&et->lock);
410 trace_f2fs_lookup_extent_tree_end(inode, pgofs, ei);
414 static struct extent_node *__try_merge_extent_node(struct f2fs_sb_info *sbi,
415 struct extent_tree *et, struct extent_info *ei,
416 struct extent_node *prev_ex,
417 struct extent_node *next_ex)
419 struct extent_node *en = NULL;
421 if (prev_ex && __is_back_mergeable(ei, &prev_ex->ei)) {
422 prev_ex->ei.len += ei->len;
427 if (next_ex && __is_front_mergeable(ei, &next_ex->ei)) {
428 next_ex->ei.fofs = ei->fofs;
429 next_ex->ei.blk = ei->blk;
430 next_ex->ei.len += ei->len;
432 __release_extent_node(sbi, et, prev_ex);
440 __try_update_largest_extent(et, en);
442 spin_lock(&sbi->extent_lock);
443 if (!list_empty(&en->list)) {
444 list_move_tail(&en->list, &sbi->extent_list);
447 spin_unlock(&sbi->extent_lock);
451 static struct extent_node *__insert_extent_tree(struct f2fs_sb_info *sbi,
452 struct extent_tree *et, struct extent_info *ei,
453 struct rb_node **insert_p,
454 struct rb_node *insert_parent)
457 struct rb_node *parent = NULL;
458 struct extent_node *en = NULL;
460 if (insert_p && insert_parent) {
461 parent = insert_parent;
466 p = f2fs_lookup_rb_tree_for_insert(sbi, &et->root, &parent, ei->fofs);
468 en = __attach_extent_node(sbi, et, ei, parent, p);
472 __try_update_largest_extent(et, en);
474 /* update in global extent list */
475 spin_lock(&sbi->extent_lock);
476 list_add_tail(&en->list, &sbi->extent_list);
478 spin_unlock(&sbi->extent_lock);
482 static void f2fs_update_extent_tree_range(struct inode *inode,
483 pgoff_t fofs, block_t blkaddr, unsigned int len)
485 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
486 struct extent_tree *et = F2FS_I(inode)->extent_tree;
487 struct extent_node *en = NULL, *en1 = NULL;
488 struct extent_node *prev_en = NULL, *next_en = NULL;
489 struct extent_info ei, dei, prev;
490 struct rb_node **insert_p = NULL, *insert_parent = NULL;
491 unsigned int end = fofs + len;
492 unsigned int pos = (unsigned int)fofs;
493 bool updated = false;
498 trace_f2fs_update_extent_tree_range(inode, fofs, blkaddr, len);
500 write_lock(&et->lock);
502 if (is_inode_flag_set(inode, FI_NO_EXTENT)) {
503 write_unlock(&et->lock);
511 * drop largest extent before lookup, in case it's already
512 * been shrunk from extent tree
514 __drop_largest_extent(et, fofs, len);
516 /* 1. lookup first extent node in range [fofs, fofs + len - 1] */
517 en = (struct extent_node *)f2fs_lookup_rb_tree_ret(&et->root,
518 (struct rb_entry *)et->cached_en, fofs,
519 (struct rb_entry **)&prev_en,
520 (struct rb_entry **)&next_en,
521 &insert_p, &insert_parent, false);
525 /* 2. invlidate all extent nodes in range [fofs, fofs + len - 1] */
526 while (en && en->ei.fofs < end) {
527 unsigned int org_end;
528 int parts = 0; /* # of parts current extent split into */
530 next_en = en1 = NULL;
533 org_end = dei.fofs + dei.len;
534 f2fs_bug_on(sbi, pos >= org_end);
536 if (pos > dei.fofs && pos - dei.fofs >= F2FS_MIN_EXTENT_LEN) {
537 en->ei.len = pos - en->ei.fofs;
542 if (end < org_end && org_end - end >= F2FS_MIN_EXTENT_LEN) {
544 set_extent_info(&ei, end,
545 end - dei.fofs + dei.blk,
547 en1 = __insert_extent_tree(sbi, et, &ei,
552 en->ei.blk += end - dei.fofs;
553 en->ei.len -= end - dei.fofs;
560 struct rb_node *node = rb_next(&en->rb_node);
562 next_en = rb_entry_safe(node, struct extent_node,
567 __try_update_largest_extent(et, en);
569 __release_extent_node(sbi, et, en);
572 * if original extent is split into zero or two parts, extent
573 * tree has been altered by deletion or insertion, therefore
574 * invalidate pointers regard to tree.
578 insert_parent = NULL;
583 /* 3. update extent in extent cache */
586 set_extent_info(&ei, fofs, blkaddr, len);
587 if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
588 __insert_extent_tree(sbi, et, &ei,
589 insert_p, insert_parent);
591 /* give up extent_cache, if split and small updates happen */
593 prev.len < F2FS_MIN_EXTENT_LEN &&
594 et->largest.len < F2FS_MIN_EXTENT_LEN) {
596 et->largest_updated = true;
597 set_inode_flag(inode, FI_NO_EXTENT);
601 if (is_inode_flag_set(inode, FI_NO_EXTENT))
602 __free_extent_tree(sbi, et);
604 if (et->largest_updated) {
605 et->largest_updated = false;
609 write_unlock(&et->lock);
612 f2fs_mark_inode_dirty_sync(inode, true);
615 unsigned int f2fs_shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
617 struct extent_tree *et, *next;
618 struct extent_node *en;
619 unsigned int node_cnt = 0, tree_cnt = 0;
622 if (!test_opt(sbi, EXTENT_CACHE))
625 if (!atomic_read(&sbi->total_zombie_tree))
628 if (!mutex_trylock(&sbi->extent_tree_lock))
631 /* 1. remove unreferenced extent tree */
632 list_for_each_entry_safe(et, next, &sbi->zombie_list, list) {
633 if (atomic_read(&et->node_cnt)) {
634 write_lock(&et->lock);
635 node_cnt += __free_extent_tree(sbi, et);
636 write_unlock(&et->lock);
638 f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
639 list_del_init(&et->list);
640 radix_tree_delete(&sbi->extent_tree_root, et->ino);
641 kmem_cache_free(extent_tree_slab, et);
642 atomic_dec(&sbi->total_ext_tree);
643 atomic_dec(&sbi->total_zombie_tree);
646 if (node_cnt + tree_cnt >= nr_shrink)
650 mutex_unlock(&sbi->extent_tree_lock);
653 /* 2. remove LRU extent entries */
654 if (!mutex_trylock(&sbi->extent_tree_lock))
657 remained = nr_shrink - (node_cnt + tree_cnt);
659 spin_lock(&sbi->extent_lock);
660 for (; remained > 0; remained--) {
661 if (list_empty(&sbi->extent_list))
663 en = list_first_entry(&sbi->extent_list,
664 struct extent_node, list);
666 if (!write_trylock(&et->lock)) {
667 /* refresh this extent node's position in extent list */
668 list_move_tail(&en->list, &sbi->extent_list);
672 list_del_init(&en->list);
673 spin_unlock(&sbi->extent_lock);
675 __detach_extent_node(sbi, et, en);
677 write_unlock(&et->lock);
679 spin_lock(&sbi->extent_lock);
681 spin_unlock(&sbi->extent_lock);
684 mutex_unlock(&sbi->extent_tree_lock);
686 trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt);
688 return node_cnt + tree_cnt;
691 unsigned int f2fs_destroy_extent_node(struct inode *inode)
693 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
694 struct extent_tree *et = F2FS_I(inode)->extent_tree;
695 unsigned int node_cnt = 0;
697 if (!et || !atomic_read(&et->node_cnt))
700 write_lock(&et->lock);
701 node_cnt = __free_extent_tree(sbi, et);
702 write_unlock(&et->lock);
707 void f2fs_drop_extent_tree(struct inode *inode)
709 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
710 struct extent_tree *et = F2FS_I(inode)->extent_tree;
711 bool updated = false;
713 if (!f2fs_may_extent_tree(inode))
716 set_inode_flag(inode, FI_NO_EXTENT);
718 write_lock(&et->lock);
719 __free_extent_tree(sbi, et);
720 if (et->largest.len) {
724 write_unlock(&et->lock);
726 f2fs_mark_inode_dirty_sync(inode, true);
729 void f2fs_destroy_extent_tree(struct inode *inode)
731 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
732 struct extent_tree *et = F2FS_I(inode)->extent_tree;
733 unsigned int node_cnt = 0;
738 if (inode->i_nlink && !is_bad_inode(inode) &&
739 atomic_read(&et->node_cnt)) {
740 mutex_lock(&sbi->extent_tree_lock);
741 list_add_tail(&et->list, &sbi->zombie_list);
742 atomic_inc(&sbi->total_zombie_tree);
743 mutex_unlock(&sbi->extent_tree_lock);
747 /* free all extent info belong to this extent tree */
748 node_cnt = f2fs_destroy_extent_node(inode);
750 /* delete extent tree entry in radix tree */
751 mutex_lock(&sbi->extent_tree_lock);
752 f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
753 radix_tree_delete(&sbi->extent_tree_root, inode->i_ino);
754 kmem_cache_free(extent_tree_slab, et);
755 atomic_dec(&sbi->total_ext_tree);
756 mutex_unlock(&sbi->extent_tree_lock);
758 F2FS_I(inode)->extent_tree = NULL;
760 trace_f2fs_destroy_extent_tree(inode, node_cnt);
763 bool f2fs_lookup_extent_cache(struct inode *inode, pgoff_t pgofs,
764 struct extent_info *ei)
766 if (!f2fs_may_extent_tree(inode))
769 return f2fs_lookup_extent_tree(inode, pgofs, ei);
772 void f2fs_update_extent_cache(struct dnode_of_data *dn)
777 if (!f2fs_may_extent_tree(dn->inode))
780 if (dn->data_blkaddr == NEW_ADDR)
783 blkaddr = dn->data_blkaddr;
785 fofs = f2fs_start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
787 f2fs_update_extent_tree_range(dn->inode, fofs, blkaddr, 1);
790 void f2fs_update_extent_cache_range(struct dnode_of_data *dn,
791 pgoff_t fofs, block_t blkaddr, unsigned int len)
794 if (!f2fs_may_extent_tree(dn->inode))
797 f2fs_update_extent_tree_range(dn->inode, fofs, blkaddr, len);
800 void f2fs_init_extent_cache_info(struct f2fs_sb_info *sbi)
802 INIT_RADIX_TREE(&sbi->extent_tree_root, GFP_NOIO);
803 mutex_init(&sbi->extent_tree_lock);
804 INIT_LIST_HEAD(&sbi->extent_list);
805 spin_lock_init(&sbi->extent_lock);
806 atomic_set(&sbi->total_ext_tree, 0);
807 INIT_LIST_HEAD(&sbi->zombie_list);
808 atomic_set(&sbi->total_zombie_tree, 0);
809 atomic_set(&sbi->total_ext_node, 0);
812 int __init f2fs_create_extent_cache(void)
814 extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree",
815 sizeof(struct extent_tree));
816 if (!extent_tree_slab)
818 extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node",
819 sizeof(struct extent_node));
820 if (!extent_node_slab) {
821 kmem_cache_destroy(extent_tree_slab);
827 void f2fs_destroy_extent_cache(void)
829 kmem_cache_destroy(extent_node_slab);
830 kmem_cache_destroy(extent_tree_slab);