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 rb_entry *__lookup_rb_tree_fast(struct rb_entry *cached_re,
25 if (cached_re->ofs <= ofs &&
26 cached_re->ofs + cached_re->len > ofs) {
33 static struct rb_entry *__lookup_rb_tree_slow(struct rb_root *root,
36 struct rb_node *node = root->rb_node;
40 re = rb_entry(node, struct rb_entry, rb_node);
44 else if (ofs >= re->ofs + re->len)
45 node = node->rb_right;
52 struct rb_entry *__lookup_rb_tree(struct rb_root *root,
53 struct rb_entry *cached_re, unsigned int ofs)
57 re = __lookup_rb_tree_fast(cached_re, ofs);
59 return __lookup_rb_tree_slow(root, ofs);
64 struct rb_node **__lookup_rb_tree_for_insert(struct f2fs_sb_info *sbi,
65 struct rb_root *root, struct rb_node **parent,
68 struct rb_node **p = &root->rb_node;
73 re = rb_entry(*parent, struct rb_entry, rb_node);
77 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 *__lookup_rb_tree_ret(struct rb_root *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,
104 struct rb_node **pnode = &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))
117 if (re->ofs <= ofs && re->ofs + re->len > ofs)
118 goto lookup_neighbors;
123 re = rb_entry(*pnode, struct rb_entry, rb_node);
126 pnode = &(*pnode)->rb_left;
127 else if (ofs >= re->ofs + re->len)
128 pnode = &(*pnode)->rb_right;
130 goto lookup_neighbors;
134 *insert_parent = parent;
136 re = rb_entry(parent, struct rb_entry, rb_node);
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);
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);
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);
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);
162 bool __check_rb_tree_consistence(struct f2fs_sb_info *sbi,
163 struct rb_root *root)
165 #ifdef CONFIG_F2FS_CHECK_FS
166 struct rb_node *cur = rb_first(root), *next;
167 struct rb_entry *cur_re, *next_re;
177 cur_re = rb_entry(cur, struct rb_entry, rb_node);
178 next_re = rb_entry(next, struct rb_entry, rb_node);
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);
194 static struct kmem_cache *extent_tree_slab;
195 static struct kmem_cache *extent_node_slab;
197 static 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)
201 struct extent_node *en;
203 en = kmem_cache_alloc(extent_node_slab, GFP_ATOMIC);
208 INIT_LIST_HEAD(&en->list);
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);
218 static void __detach_extent_node(struct f2fs_sb_info *sbi,
219 struct extent_tree *et, struct extent_node *en)
221 rb_erase(&en->rb_node, &et->root);
222 atomic_dec(&et->node_cnt);
223 atomic_dec(&sbi->total_ext_node);
225 if (et->cached_en == en)
226 et->cached_en = NULL;
227 kmem_cache_free(extent_node_slab, en);
231 * Flow to release an extent_node:
233 * 2. __detach_extent_node
234 * 3. kmem_cache_free.
236 static void __release_extent_node(struct f2fs_sb_info *sbi,
237 struct extent_tree *et, struct extent_node *en)
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);
244 __detach_extent_node(sbi, et, en);
247 static struct extent_tree *__grab_extent_tree(struct inode *inode)
249 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
250 struct extent_tree *et;
251 nid_t ino = inode->i_ino;
253 mutex_lock(&sbi->extent_tree_lock);
254 et = radix_tree_lookup(&sbi->extent_tree_root, ino);
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));
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);
267 atomic_dec(&sbi->total_zombie_tree);
268 list_del_init(&et->list);
270 mutex_unlock(&sbi->extent_tree_lock);
272 /* never died until evict_inode */
273 F2FS_I(inode)->extent_tree = et;
278 static struct extent_node *__init_extent_tree(struct f2fs_sb_info *sbi,
279 struct extent_tree *et, struct extent_info *ei)
281 struct rb_node **p = &et->root.rb_node;
282 struct extent_node *en;
284 en = __attach_extent_node(sbi, et, ei, NULL, p);
288 et->largest = en->ei;
293 static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi,
294 struct extent_tree *et)
296 struct rb_node *node, *next;
297 struct extent_node *en;
298 unsigned int count = atomic_read(&et->node_cnt);
300 node = rb_first(&et->root);
302 next = rb_next(node);
303 en = rb_entry(node, struct extent_node, rb_node);
304 __release_extent_node(sbi, et, en);
308 return count - atomic_read(&et->node_cnt);
311 static void __drop_largest_extent(struct inode *inode,
312 pgoff_t fofs, unsigned int len)
314 struct extent_info *largest = &F2FS_I(inode)->extent_tree->largest;
316 if (fofs < largest->fofs + largest->len && fofs + len > largest->fofs) {
318 f2fs_mark_inode_dirty_sync(inode, true);
322 /* return true, if inode page is changed */
323 static bool __f2fs_init_extent_tree(struct inode *inode, struct f2fs_extent *i_ext)
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;
330 if (!f2fs_may_extent_tree(inode)) {
331 /* drop largest extent */
332 if (i_ext && i_ext->len) {
339 et = __grab_extent_tree(inode);
341 if (!i_ext || !i_ext->len)
344 get_extent_info(&ei, i_ext);
346 write_lock(&et->lock);
347 if (atomic_read(&et->node_cnt))
350 en = __init_extent_tree(sbi, et, &ei);
352 spin_lock(&sbi->extent_lock);
353 list_add_tail(&en->list, &sbi->extent_list);
354 spin_unlock(&sbi->extent_lock);
357 write_unlock(&et->lock);
361 bool f2fs_init_extent_tree(struct inode *inode, struct f2fs_extent *i_ext)
363 bool ret = __f2fs_init_extent_tree(inode, i_ext);
365 if (!F2FS_I(inode)->extent_tree)
366 set_inode_flag(inode, FI_NO_EXTENT);
371 static bool f2fs_lookup_extent_tree(struct inode *inode, pgoff_t pgofs,
372 struct extent_info *ei)
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;
379 f2fs_bug_on(sbi, !et);
381 trace_f2fs_lookup_extent_tree_start(inode, pgofs);
383 read_lock(&et->lock);
385 if (et->largest.fofs <= pgofs &&
386 et->largest.fofs + et->largest.len > pgofs) {
389 stat_inc_largest_node_hit(sbi);
393 en = (struct extent_node *)__lookup_rb_tree(&et->root,
394 (struct rb_entry *)et->cached_en, pgofs);
398 if (en == et->cached_en)
399 stat_inc_cached_node_hit(sbi);
401 stat_inc_rbtree_node_hit(sbi);
404 spin_lock(&sbi->extent_lock);
405 if (!list_empty(&en->list)) {
406 list_move_tail(&en->list, &sbi->extent_list);
409 spin_unlock(&sbi->extent_lock);
412 stat_inc_total_hit(sbi);
413 read_unlock(&et->lock);
415 trace_f2fs_lookup_extent_tree_end(inode, pgofs, ei);
419 static 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)
424 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
425 struct extent_node *en = NULL;
427 if (prev_ex && __is_back_mergeable(ei, &prev_ex->ei)) {
428 prev_ex->ei.len += ei->len;
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;
438 __release_extent_node(sbi, et, prev_ex);
446 __try_update_largest_extent(inode, et, en);
448 spin_lock(&sbi->extent_lock);
449 if (!list_empty(&en->list)) {
450 list_move_tail(&en->list, &sbi->extent_list);
453 spin_unlock(&sbi->extent_lock);
457 static 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)
462 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
463 struct rb_node **p = &et->root.rb_node;
464 struct rb_node *parent = NULL;
465 struct extent_node *en = NULL;
467 if (insert_p && insert_parent) {
468 parent = insert_parent;
473 p = __lookup_rb_tree_for_insert(sbi, &et->root, &parent, ei->fofs);
475 en = __attach_extent_node(sbi, et, ei, parent, p);
479 __try_update_largest_extent(inode, et, en);
481 /* update in global extent list */
482 spin_lock(&sbi->extent_lock);
483 list_add_tail(&en->list, &sbi->extent_list);
485 spin_unlock(&sbi->extent_lock);
489 static void f2fs_update_extent_tree_range(struct inode *inode,
490 pgoff_t fofs, block_t blkaddr, unsigned int len)
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;
504 trace_f2fs_update_extent_tree_range(inode, fofs, blkaddr, len);
506 write_lock(&et->lock);
508 if (is_inode_flag_set(inode, FI_NO_EXTENT)) {
509 write_unlock(&et->lock);
517 * drop largest extent before lookup, in case it's already
518 * been shrunk from extent tree
520 __drop_largest_extent(inode, fofs, len);
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);
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 */
536 next_en = en1 = NULL;
539 org_end = dei.fofs + dei.len;
540 f2fs_bug_on(sbi, pos >= org_end);
542 if (pos > dei.fofs && pos - dei.fofs >= F2FS_MIN_EXTENT_LEN) {
543 en->ei.len = pos - en->ei.fofs;
548 if (end < org_end && org_end - end >= F2FS_MIN_EXTENT_LEN) {
550 set_extent_info(&ei, end,
551 end - dei.fofs + dei.blk,
553 en1 = __insert_extent_tree(inode, et, &ei,
558 en->ei.blk += end - dei.fofs;
559 en->ei.len -= end - dei.fofs;
566 struct rb_node *node = rb_next(&en->rb_node);
568 next_en = rb_entry_safe(node, struct extent_node,
573 __try_update_largest_extent(inode, et, en);
575 __release_extent_node(sbi, et, en);
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.
584 insert_parent = NULL;
589 /* 3. update extent in extent cache */
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);
597 /* give up extent_cache, if split and small updates happen */
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);
606 if (is_inode_flag_set(inode, FI_NO_EXTENT))
607 __free_extent_tree(sbi, et);
609 write_unlock(&et->lock);
612 unsigned int f2fs_shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
614 struct extent_tree *et, *next;
615 struct extent_node *en;
616 unsigned int node_cnt = 0, tree_cnt = 0;
619 if (!test_opt(sbi, EXTENT_CACHE))
622 if (!atomic_read(&sbi->total_zombie_tree))
625 if (!mutex_trylock(&sbi->extent_tree_lock))
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);
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);
643 if (node_cnt + tree_cnt >= nr_shrink)
647 mutex_unlock(&sbi->extent_tree_lock);
650 /* 2. remove LRU extent entries */
651 if (!mutex_trylock(&sbi->extent_tree_lock))
654 remained = nr_shrink - (node_cnt + tree_cnt);
656 spin_lock(&sbi->extent_lock);
657 for (; remained > 0; remained--) {
658 if (list_empty(&sbi->extent_list))
660 en = list_first_entry(&sbi->extent_list,
661 struct extent_node, list);
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);
669 list_del_init(&en->list);
670 spin_unlock(&sbi->extent_lock);
672 __detach_extent_node(sbi, et, en);
674 write_unlock(&et->lock);
676 spin_lock(&sbi->extent_lock);
678 spin_unlock(&sbi->extent_lock);
681 mutex_unlock(&sbi->extent_tree_lock);
683 trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt);
685 return node_cnt + tree_cnt;
688 unsigned int f2fs_destroy_extent_node(struct inode *inode)
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;
694 if (!et || !atomic_read(&et->node_cnt))
697 write_lock(&et->lock);
698 node_cnt = __free_extent_tree(sbi, et);
699 write_unlock(&et->lock);
704 void f2fs_drop_extent_tree(struct inode *inode)
706 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
707 struct extent_tree *et = F2FS_I(inode)->extent_tree;
709 if (!f2fs_may_extent_tree(inode))
712 set_inode_flag(inode, FI_NO_EXTENT);
714 write_lock(&et->lock);
715 __free_extent_tree(sbi, et);
716 __drop_largest_extent(inode, 0, UINT_MAX);
717 write_unlock(&et->lock);
720 void f2fs_destroy_extent_tree(struct inode *inode)
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;
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);
738 /* free all extent info belong to this extent tree */
739 node_cnt = f2fs_destroy_extent_node(inode);
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);
749 F2FS_I(inode)->extent_tree = NULL;
751 trace_f2fs_destroy_extent_tree(inode, node_cnt);
754 bool f2fs_lookup_extent_cache(struct inode *inode, pgoff_t pgofs,
755 struct extent_info *ei)
757 if (!f2fs_may_extent_tree(inode))
760 return f2fs_lookup_extent_tree(inode, pgofs, ei);
763 void f2fs_update_extent_cache(struct dnode_of_data *dn)
768 if (!f2fs_may_extent_tree(dn->inode))
771 if (dn->data_blkaddr == NEW_ADDR)
774 blkaddr = dn->data_blkaddr;
776 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
778 f2fs_update_extent_tree_range(dn->inode, fofs, blkaddr, 1);
781 void f2fs_update_extent_cache_range(struct dnode_of_data *dn,
782 pgoff_t fofs, block_t blkaddr, unsigned int len)
785 if (!f2fs_may_extent_tree(dn->inode))
788 f2fs_update_extent_tree_range(dn->inode, fofs, blkaddr, len);
791 void init_extent_cache_info(struct f2fs_sb_info *sbi)
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);
803 int __init create_extent_cache(void)
805 extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree",
806 sizeof(struct extent_tree));
807 if (!extent_tree_slab)
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);
818 void destroy_extent_cache(void)
820 kmem_cache_destroy(extent_node_slab);
821 kmem_cache_destroy(extent_tree_slab);