2 * f2fs extent cache support
4 * Copyright (c) 2015 Motorola Mobility
5 * Copyright (c) 2015 Samsung Electronics
6 * Authors: Jaegeuk Kim <jaegeuk@kernel.org>
7 * Chao Yu <chao2.yu@samsung.com>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
15 #include <linux/f2fs_fs.h>
19 #include <trace/events/f2fs.h>
21 static struct kmem_cache *extent_tree_slab;
22 static struct kmem_cache *extent_node_slab;
24 static struct extent_node *__attach_extent_node(struct f2fs_sb_info *sbi,
25 struct extent_tree *et, struct extent_info *ei,
26 struct rb_node *parent, struct rb_node **p)
28 struct extent_node *en;
30 en = kmem_cache_alloc(extent_node_slab, GFP_ATOMIC);
35 INIT_LIST_HEAD(&en->list);
38 rb_link_node(&en->rb_node, parent, p);
39 rb_insert_color(&en->rb_node, &et->root);
40 atomic_inc(&et->node_cnt);
41 atomic_inc(&sbi->total_ext_node);
45 static void __detach_extent_node(struct f2fs_sb_info *sbi,
46 struct extent_tree *et, struct extent_node *en)
48 rb_erase(&en->rb_node, &et->root);
49 atomic_dec(&et->node_cnt);
50 atomic_dec(&sbi->total_ext_node);
52 if (et->cached_en == en)
54 kmem_cache_free(extent_node_slab, en);
58 * Flow to release an extent_node:
60 * 2. __detach_extent_node
63 static void __release_extent_node(struct f2fs_sb_info *sbi,
64 struct extent_tree *et, struct extent_node *en)
66 spin_lock(&sbi->extent_lock);
67 f2fs_bug_on(sbi, list_empty(&en->list));
68 list_del_init(&en->list);
69 spin_unlock(&sbi->extent_lock);
71 __detach_extent_node(sbi, et, en);
74 static struct extent_tree *__grab_extent_tree(struct inode *inode)
76 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
77 struct extent_tree *et;
78 nid_t ino = inode->i_ino;
80 down_write(&sbi->extent_tree_lock);
81 et = radix_tree_lookup(&sbi->extent_tree_root, ino);
83 et = f2fs_kmem_cache_alloc(extent_tree_slab, GFP_NOFS);
84 f2fs_radix_tree_insert(&sbi->extent_tree_root, ino, et);
85 memset(et, 0, sizeof(struct extent_tree));
89 rwlock_init(&et->lock);
90 INIT_LIST_HEAD(&et->list);
91 atomic_set(&et->node_cnt, 0);
92 atomic_inc(&sbi->total_ext_tree);
94 atomic_dec(&sbi->total_zombie_tree);
95 list_del_init(&et->list);
97 up_write(&sbi->extent_tree_lock);
99 /* never died until evict_inode */
100 F2FS_I(inode)->extent_tree = et;
105 static struct extent_node *__lookup_extent_tree(struct f2fs_sb_info *sbi,
106 struct extent_tree *et, unsigned int fofs)
108 struct rb_node *node = et->root.rb_node;
109 struct extent_node *en = et->cached_en;
112 struct extent_info *cei = &en->ei;
114 if (cei->fofs <= fofs && cei->fofs + cei->len > fofs) {
115 stat_inc_cached_node_hit(sbi);
121 en = rb_entry(node, struct extent_node, rb_node);
123 if (fofs < en->ei.fofs) {
124 node = node->rb_left;
125 } else if (fofs >= en->ei.fofs + en->ei.len) {
126 node = node->rb_right;
128 stat_inc_rbtree_node_hit(sbi);
135 static struct extent_node *__init_extent_tree(struct f2fs_sb_info *sbi,
136 struct extent_tree *et, struct extent_info *ei)
138 struct rb_node **p = &et->root.rb_node;
139 struct extent_node *en;
141 en = __attach_extent_node(sbi, et, ei, NULL, p);
145 et->largest = en->ei;
150 static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi,
151 struct extent_tree *et)
153 struct rb_node *node, *next;
154 struct extent_node *en;
155 unsigned int count = atomic_read(&et->node_cnt);
157 node = rb_first(&et->root);
159 next = rb_next(node);
160 en = rb_entry(node, struct extent_node, rb_node);
161 __release_extent_node(sbi, et, en);
165 return count - atomic_read(&et->node_cnt);
168 static void __drop_largest_extent(struct inode *inode,
169 pgoff_t fofs, unsigned int len)
171 struct extent_info *largest = &F2FS_I(inode)->extent_tree->largest;
173 if (fofs < largest->fofs + largest->len && fofs + len > largest->fofs)
177 /* return true, if inode page is changed */
178 bool f2fs_init_extent_tree(struct inode *inode, struct f2fs_extent *i_ext)
180 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
181 struct extent_tree *et;
182 struct extent_node *en;
183 struct extent_info ei;
185 if (!f2fs_may_extent_tree(inode)) {
186 /* drop largest extent */
187 if (i_ext && i_ext->len) {
194 et = __grab_extent_tree(inode);
196 if (!i_ext || !i_ext->len)
199 get_extent_info(&ei, i_ext);
201 write_lock(&et->lock);
202 if (atomic_read(&et->node_cnt))
205 en = __init_extent_tree(sbi, et, &ei);
207 spin_lock(&sbi->extent_lock);
208 list_add_tail(&en->list, &sbi->extent_list);
209 spin_unlock(&sbi->extent_lock);
212 write_unlock(&et->lock);
216 static bool f2fs_lookup_extent_tree(struct inode *inode, pgoff_t pgofs,
217 struct extent_info *ei)
219 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
220 struct extent_tree *et = F2FS_I(inode)->extent_tree;
221 struct extent_node *en;
224 f2fs_bug_on(sbi, !et);
226 trace_f2fs_lookup_extent_tree_start(inode, pgofs);
228 read_lock(&et->lock);
230 if (et->largest.fofs <= pgofs &&
231 et->largest.fofs + et->largest.len > pgofs) {
234 stat_inc_largest_node_hit(sbi);
238 en = __lookup_extent_tree(sbi, et, pgofs);
241 spin_lock(&sbi->extent_lock);
242 if (!list_empty(&en->list)) {
243 list_move_tail(&en->list, &sbi->extent_list);
246 spin_unlock(&sbi->extent_lock);
250 stat_inc_total_hit(sbi);
251 read_unlock(&et->lock);
253 trace_f2fs_lookup_extent_tree_end(inode, pgofs, ei);
259 * lookup extent at @fofs, if hit, return the extent
260 * if not, return NULL and
261 * @prev_ex: extent before fofs
262 * @next_ex: extent after fofs
263 * @insert_p: insert point for new extent at fofs
264 * in order to simpfy the insertion after.
265 * tree must stay unchanged between lookup and insertion.
267 static struct extent_node *__lookup_extent_tree_ret(struct extent_tree *et,
269 struct extent_node **prev_ex,
270 struct extent_node **next_ex,
271 struct rb_node ***insert_p,
272 struct rb_node **insert_parent)
274 struct rb_node **pnode = &et->root.rb_node;
275 struct rb_node *parent = NULL, *tmp_node;
276 struct extent_node *en = et->cached_en;
279 *insert_parent = NULL;
283 if (RB_EMPTY_ROOT(&et->root))
287 struct extent_info *cei = &en->ei;
289 if (cei->fofs <= fofs && cei->fofs + cei->len > fofs)
290 goto lookup_neighbors;
295 en = rb_entry(*pnode, struct extent_node, rb_node);
297 if (fofs < en->ei.fofs)
298 pnode = &(*pnode)->rb_left;
299 else if (fofs >= en->ei.fofs + en->ei.len)
300 pnode = &(*pnode)->rb_right;
302 goto lookup_neighbors;
306 *insert_parent = parent;
308 en = rb_entry(parent, struct extent_node, rb_node);
310 if (parent && fofs > en->ei.fofs)
311 tmp_node = rb_next(parent);
312 *next_ex = tmp_node ?
313 rb_entry(tmp_node, struct extent_node, rb_node) : NULL;
316 if (parent && fofs < en->ei.fofs)
317 tmp_node = rb_prev(parent);
318 *prev_ex = tmp_node ?
319 rb_entry(tmp_node, struct extent_node, rb_node) : NULL;
323 if (fofs == en->ei.fofs) {
324 /* lookup prev node for merging backward later */
325 tmp_node = rb_prev(&en->rb_node);
326 *prev_ex = tmp_node ?
327 rb_entry(tmp_node, struct extent_node, rb_node) : NULL;
329 if (fofs == en->ei.fofs + en->ei.len - 1) {
330 /* lookup next node for merging frontward later */
331 tmp_node = rb_next(&en->rb_node);
332 *next_ex = tmp_node ?
333 rb_entry(tmp_node, struct extent_node, rb_node) : NULL;
338 static struct extent_node *__try_merge_extent_node(struct f2fs_sb_info *sbi,
339 struct extent_tree *et, struct extent_info *ei,
340 struct extent_node *prev_ex,
341 struct extent_node *next_ex)
343 struct extent_node *en = NULL;
345 if (prev_ex && __is_back_mergeable(ei, &prev_ex->ei)) {
346 prev_ex->ei.len += ei->len;
351 if (next_ex && __is_front_mergeable(ei, &next_ex->ei)) {
353 __release_extent_node(sbi, et, prev_ex);
354 next_ex->ei.fofs = ei->fofs;
355 next_ex->ei.blk = ei->blk;
356 next_ex->ei.len += ei->len;
363 __try_update_largest_extent(et, en);
365 spin_lock(&sbi->extent_lock);
366 if (!list_empty(&en->list)) {
367 list_move_tail(&en->list, &sbi->extent_list);
370 spin_unlock(&sbi->extent_lock);
374 static struct extent_node *__insert_extent_tree(struct f2fs_sb_info *sbi,
375 struct extent_tree *et, struct extent_info *ei,
376 struct rb_node **insert_p,
377 struct rb_node *insert_parent)
379 struct rb_node **p = &et->root.rb_node;
380 struct rb_node *parent = NULL;
381 struct extent_node *en = NULL;
383 if (insert_p && insert_parent) {
384 parent = insert_parent;
391 en = rb_entry(parent, struct extent_node, rb_node);
393 if (ei->fofs < en->ei.fofs)
395 else if (ei->fofs >= en->ei.fofs + en->ei.len)
401 en = __attach_extent_node(sbi, et, ei, parent, p);
405 __try_update_largest_extent(et, en);
407 /* update in global extent list */
408 spin_lock(&sbi->extent_lock);
409 list_add_tail(&en->list, &sbi->extent_list);
411 spin_unlock(&sbi->extent_lock);
415 static unsigned int f2fs_update_extent_tree_range(struct inode *inode,
416 pgoff_t fofs, block_t blkaddr, unsigned int len)
418 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
419 struct extent_tree *et = F2FS_I(inode)->extent_tree;
420 struct extent_node *en = NULL, *en1 = NULL;
421 struct extent_node *prev_en = NULL, *next_en = NULL;
422 struct extent_info ei, dei, prev;
423 struct rb_node **insert_p = NULL, *insert_parent = NULL;
424 unsigned int end = fofs + len;
425 unsigned int pos = (unsigned int)fofs;
430 trace_f2fs_update_extent_tree_range(inode, fofs, blkaddr, len);
432 write_lock(&et->lock);
434 if (is_inode_flag_set(F2FS_I(inode), FI_NO_EXTENT)) {
435 write_unlock(&et->lock);
443 * drop largest extent before lookup, in case it's already
444 * been shrunk from extent tree
446 __drop_largest_extent(inode, fofs, len);
448 /* 1. lookup first extent node in range [fofs, fofs + len - 1] */
449 en = __lookup_extent_tree_ret(et, fofs, &prev_en, &next_en,
450 &insert_p, &insert_parent);
454 /* 2. invlidate all extent nodes in range [fofs, fofs + len - 1] */
455 while (en && en->ei.fofs < end) {
456 unsigned int org_end;
457 int parts = 0; /* # of parts current extent split into */
459 next_en = en1 = NULL;
462 org_end = dei.fofs + dei.len;
463 f2fs_bug_on(sbi, pos >= org_end);
465 if (pos > dei.fofs && pos - dei.fofs >= F2FS_MIN_EXTENT_LEN) {
466 en->ei.len = pos - en->ei.fofs;
471 if (end < org_end && org_end - end >= F2FS_MIN_EXTENT_LEN) {
473 set_extent_info(&ei, end,
474 end - dei.fofs + dei.blk,
476 en1 = __insert_extent_tree(sbi, et, &ei,
481 en->ei.blk += end - dei.fofs;
482 en->ei.len -= end - dei.fofs;
489 struct rb_node *node = rb_next(&en->rb_node);
492 rb_entry(node, struct extent_node, rb_node)
497 __try_update_largest_extent(et, en);
499 __release_extent_node(sbi, et, en);
502 * if original extent is split into zero or two parts, extent
503 * tree has been altered by deletion or insertion, therefore
504 * invalidate pointers regard to tree.
508 insert_parent = NULL;
513 /* 3. update extent in extent cache */
516 set_extent_info(&ei, fofs, blkaddr, len);
517 if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
518 __insert_extent_tree(sbi, et, &ei,
519 insert_p, insert_parent);
521 /* give up extent_cache, if split and small updates happen */
523 prev.len < F2FS_MIN_EXTENT_LEN &&
524 et->largest.len < F2FS_MIN_EXTENT_LEN) {
526 set_inode_flag(F2FS_I(inode), FI_NO_EXTENT);
530 if (is_inode_flag_set(F2FS_I(inode), FI_NO_EXTENT))
531 __free_extent_tree(sbi, et);
533 write_unlock(&et->lock);
535 return !__is_extent_same(&prev, &et->largest);
538 unsigned int f2fs_shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
540 struct extent_tree *et, *next;
541 struct extent_node *en;
542 unsigned int node_cnt = 0, tree_cnt = 0;
545 if (!test_opt(sbi, EXTENT_CACHE))
548 if (!atomic_read(&sbi->total_zombie_tree))
551 if (!down_write_trylock(&sbi->extent_tree_lock))
554 /* 1. remove unreferenced extent tree */
555 list_for_each_entry_safe(et, next, &sbi->zombie_list, list) {
556 if (atomic_read(&et->node_cnt)) {
557 write_lock(&et->lock);
558 node_cnt += __free_extent_tree(sbi, et);
559 write_unlock(&et->lock);
561 f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
562 list_del_init(&et->list);
563 radix_tree_delete(&sbi->extent_tree_root, et->ino);
564 kmem_cache_free(extent_tree_slab, et);
565 atomic_dec(&sbi->total_ext_tree);
566 atomic_dec(&sbi->total_zombie_tree);
569 if (node_cnt + tree_cnt >= nr_shrink)
573 up_write(&sbi->extent_tree_lock);
576 /* 2. remove LRU extent entries */
577 if (!down_write_trylock(&sbi->extent_tree_lock))
580 remained = nr_shrink - (node_cnt + tree_cnt);
582 spin_lock(&sbi->extent_lock);
583 for (; remained > 0; remained--) {
584 if (list_empty(&sbi->extent_list))
586 en = list_first_entry(&sbi->extent_list,
587 struct extent_node, list);
589 if (!write_trylock(&et->lock)) {
590 /* refresh this extent node's position in extent list */
591 list_move_tail(&en->list, &sbi->extent_list);
595 list_del_init(&en->list);
596 spin_unlock(&sbi->extent_lock);
598 __detach_extent_node(sbi, et, en);
600 write_unlock(&et->lock);
602 spin_lock(&sbi->extent_lock);
604 spin_unlock(&sbi->extent_lock);
607 up_write(&sbi->extent_tree_lock);
609 trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt);
611 return node_cnt + tree_cnt;
614 unsigned int f2fs_destroy_extent_node(struct inode *inode)
616 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
617 struct extent_tree *et = F2FS_I(inode)->extent_tree;
618 unsigned int node_cnt = 0;
620 if (!et || !atomic_read(&et->node_cnt))
623 write_lock(&et->lock);
624 node_cnt = __free_extent_tree(sbi, et);
625 write_unlock(&et->lock);
630 void f2fs_destroy_extent_tree(struct inode *inode)
632 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
633 struct extent_tree *et = F2FS_I(inode)->extent_tree;
634 unsigned int node_cnt = 0;
639 if (inode->i_nlink && !is_bad_inode(inode) &&
640 atomic_read(&et->node_cnt)) {
641 down_write(&sbi->extent_tree_lock);
642 list_add_tail(&et->list, &sbi->zombie_list);
643 atomic_inc(&sbi->total_zombie_tree);
644 up_write(&sbi->extent_tree_lock);
648 /* free all extent info belong to this extent tree */
649 node_cnt = f2fs_destroy_extent_node(inode);
651 /* delete extent tree entry in radix tree */
652 down_write(&sbi->extent_tree_lock);
653 f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
654 radix_tree_delete(&sbi->extent_tree_root, inode->i_ino);
655 kmem_cache_free(extent_tree_slab, et);
656 atomic_dec(&sbi->total_ext_tree);
657 up_write(&sbi->extent_tree_lock);
659 F2FS_I(inode)->extent_tree = NULL;
661 trace_f2fs_destroy_extent_tree(inode, node_cnt);
664 bool f2fs_lookup_extent_cache(struct inode *inode, pgoff_t pgofs,
665 struct extent_info *ei)
667 if (!f2fs_may_extent_tree(inode))
670 return f2fs_lookup_extent_tree(inode, pgofs, ei);
673 void f2fs_update_extent_cache(struct dnode_of_data *dn)
678 if (!f2fs_may_extent_tree(dn->inode))
681 if (dn->data_blkaddr == NEW_ADDR)
684 blkaddr = dn->data_blkaddr;
686 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
689 if (f2fs_update_extent_tree_range(dn->inode, fofs, blkaddr, 1))
693 void f2fs_update_extent_cache_range(struct dnode_of_data *dn,
694 pgoff_t fofs, block_t blkaddr, unsigned int len)
697 if (!f2fs_may_extent_tree(dn->inode))
700 if (f2fs_update_extent_tree_range(dn->inode, fofs, blkaddr, len))
704 void init_extent_cache_info(struct f2fs_sb_info *sbi)
706 INIT_RADIX_TREE(&sbi->extent_tree_root, GFP_NOIO);
707 init_rwsem(&sbi->extent_tree_lock);
708 INIT_LIST_HEAD(&sbi->extent_list);
709 spin_lock_init(&sbi->extent_lock);
710 atomic_set(&sbi->total_ext_tree, 0);
711 INIT_LIST_HEAD(&sbi->zombie_list);
712 atomic_set(&sbi->total_zombie_tree, 0);
713 atomic_set(&sbi->total_ext_node, 0);
716 int __init create_extent_cache(void)
718 extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree",
719 sizeof(struct extent_tree));
720 if (!extent_tree_slab)
722 extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node",
723 sizeof(struct extent_node));
724 if (!extent_node_slab) {
725 kmem_cache_destroy(extent_tree_slab);
731 void destroy_extent_cache(void)
733 kmem_cache_destroy(extent_node_slab);
734 kmem_cache_destroy(extent_tree_slab);