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);
37 rb_link_node(&en->rb_node, parent, p);
38 rb_insert_color(&en->rb_node, &et->root);
40 atomic_inc(&sbi->total_ext_node);
44 static void __detach_extent_node(struct f2fs_sb_info *sbi,
45 struct extent_tree *et, struct extent_node *en)
47 rb_erase(&en->rb_node, &et->root);
49 atomic_dec(&sbi->total_ext_node);
51 if (et->cached_en == en)
55 static struct extent_tree *__grab_extent_tree(struct inode *inode)
57 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
58 struct extent_tree *et;
59 nid_t ino = inode->i_ino;
61 down_write(&sbi->extent_tree_lock);
62 et = radix_tree_lookup(&sbi->extent_tree_root, ino);
64 et = f2fs_kmem_cache_alloc(extent_tree_slab, GFP_NOFS);
65 f2fs_radix_tree_insert(&sbi->extent_tree_root, ino, et);
66 memset(et, 0, sizeof(struct extent_tree));
70 rwlock_init(&et->lock);
71 atomic_set(&et->refcount, 0);
73 atomic_inc(&sbi->total_ext_tree);
75 atomic_dec(&sbi->total_zombie_tree);
77 atomic_inc(&et->refcount);
78 up_write(&sbi->extent_tree_lock);
80 /* never died until evict_inode */
81 F2FS_I(inode)->extent_tree = et;
86 static struct extent_node *__lookup_extent_tree(struct f2fs_sb_info *sbi,
87 struct extent_tree *et, unsigned int fofs)
89 struct rb_node *node = et->root.rb_node;
90 struct extent_node *en = et->cached_en;
93 struct extent_info *cei = &en->ei;
95 if (cei->fofs <= fofs && cei->fofs + cei->len > fofs) {
96 stat_inc_cached_node_hit(sbi);
102 en = rb_entry(node, struct extent_node, rb_node);
104 if (fofs < en->ei.fofs) {
105 node = node->rb_left;
106 } else if (fofs >= en->ei.fofs + en->ei.len) {
107 node = node->rb_right;
109 stat_inc_rbtree_node_hit(sbi);
116 static struct extent_node *__init_extent_tree(struct f2fs_sb_info *sbi,
117 struct extent_tree *et, struct extent_info *ei)
119 struct rb_node **p = &et->root.rb_node;
120 struct extent_node *en;
122 en = __attach_extent_node(sbi, et, ei, NULL, p);
126 et->largest = en->ei;
131 static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi,
132 struct extent_tree *et, bool free_all)
134 struct rb_node *node, *next;
135 struct extent_node *en;
136 unsigned int count = et->count;
138 node = rb_first(&et->root);
140 next = rb_next(node);
141 en = rb_entry(node, struct extent_node, rb_node);
144 spin_lock(&sbi->extent_lock);
145 if (!list_empty(&en->list))
146 list_del_init(&en->list);
147 spin_unlock(&sbi->extent_lock);
150 if (free_all || list_empty(&en->list)) {
151 __detach_extent_node(sbi, et, en);
152 kmem_cache_free(extent_node_slab, en);
157 return count - et->count;
160 static void __drop_largest_extent(struct inode *inode,
161 pgoff_t fofs, unsigned int len)
163 struct extent_info *largest = &F2FS_I(inode)->extent_tree->largest;
165 if (fofs < largest->fofs + largest->len && fofs + len > largest->fofs)
169 void f2fs_init_extent_tree(struct inode *inode, struct f2fs_extent *i_ext)
171 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
172 struct extent_tree *et;
173 struct extent_node *en;
174 struct extent_info ei;
176 if (!f2fs_may_extent_tree(inode))
179 et = __grab_extent_tree(inode);
181 if (!i_ext || le32_to_cpu(i_ext->len) < F2FS_MIN_EXTENT_LEN)
184 set_extent_info(&ei, le32_to_cpu(i_ext->fofs),
185 le32_to_cpu(i_ext->blk), le32_to_cpu(i_ext->len));
187 write_lock(&et->lock);
191 en = __init_extent_tree(sbi, et, &ei);
193 spin_lock(&sbi->extent_lock);
194 list_add_tail(&en->list, &sbi->extent_list);
195 spin_unlock(&sbi->extent_lock);
198 write_unlock(&et->lock);
201 static bool f2fs_lookup_extent_tree(struct inode *inode, pgoff_t pgofs,
202 struct extent_info *ei)
204 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
205 struct extent_tree *et = F2FS_I(inode)->extent_tree;
206 struct extent_node *en;
209 f2fs_bug_on(sbi, !et);
211 trace_f2fs_lookup_extent_tree_start(inode, pgofs);
213 read_lock(&et->lock);
215 if (et->largest.fofs <= pgofs &&
216 et->largest.fofs + et->largest.len > pgofs) {
219 stat_inc_largest_node_hit(sbi);
223 en = __lookup_extent_tree(sbi, et, pgofs);
226 spin_lock(&sbi->extent_lock);
227 if (!list_empty(&en->list))
228 list_move_tail(&en->list, &sbi->extent_list);
230 spin_unlock(&sbi->extent_lock);
234 stat_inc_total_hit(sbi);
235 read_unlock(&et->lock);
237 trace_f2fs_lookup_extent_tree_end(inode, pgofs, ei);
243 * lookup extent at @fofs, if hit, return the extent
244 * if not, return NULL and
245 * @prev_ex: extent before fofs
246 * @next_ex: extent after fofs
247 * @insert_p: insert point for new extent at fofs
248 * in order to simpfy the insertion after.
249 * tree must stay unchanged between lookup and insertion.
251 static struct extent_node *__lookup_extent_tree_ret(struct extent_tree *et,
253 struct extent_node **prev_ex,
254 struct extent_node **next_ex,
255 struct rb_node ***insert_p,
256 struct rb_node **insert_parent)
258 struct rb_node **pnode = &et->root.rb_node;
259 struct rb_node *parent = NULL, *tmp_node;
260 struct extent_node *en = et->cached_en;
263 *insert_parent = NULL;
267 if (RB_EMPTY_ROOT(&et->root))
271 struct extent_info *cei = &en->ei;
273 if (cei->fofs <= fofs && cei->fofs + cei->len > fofs)
274 goto lookup_neighbors;
279 en = rb_entry(*pnode, struct extent_node, rb_node);
281 if (fofs < en->ei.fofs)
282 pnode = &(*pnode)->rb_left;
283 else if (fofs >= en->ei.fofs + en->ei.len)
284 pnode = &(*pnode)->rb_right;
286 goto lookup_neighbors;
290 *insert_parent = parent;
292 en = rb_entry(parent, struct extent_node, rb_node);
294 if (parent && fofs > en->ei.fofs)
295 tmp_node = rb_next(parent);
296 *next_ex = tmp_node ?
297 rb_entry(tmp_node, struct extent_node, rb_node) : NULL;
300 if (parent && fofs < en->ei.fofs)
301 tmp_node = rb_prev(parent);
302 *prev_ex = tmp_node ?
303 rb_entry(tmp_node, struct extent_node, rb_node) : NULL;
307 if (fofs == en->ei.fofs) {
308 /* lookup prev node for merging backward later */
309 tmp_node = rb_prev(&en->rb_node);
310 *prev_ex = tmp_node ?
311 rb_entry(tmp_node, struct extent_node, rb_node) : NULL;
313 if (fofs == en->ei.fofs + en->ei.len - 1) {
314 /* lookup next node for merging frontward later */
315 tmp_node = rb_next(&en->rb_node);
316 *next_ex = tmp_node ?
317 rb_entry(tmp_node, struct extent_node, rb_node) : NULL;
322 static struct extent_node *__try_merge_extent_node(struct f2fs_sb_info *sbi,
323 struct extent_tree *et, struct extent_info *ei,
324 struct extent_node **den,
325 struct extent_node *prev_ex,
326 struct extent_node *next_ex)
328 struct extent_node *en = NULL;
330 if (prev_ex && __is_back_mergeable(ei, &prev_ex->ei)) {
331 prev_ex->ei.len += ei->len;
336 if (next_ex && __is_front_mergeable(ei, &next_ex->ei)) {
338 __detach_extent_node(sbi, et, prev_ex);
341 next_ex->ei.fofs = ei->fofs;
342 next_ex->ei.blk = ei->blk;
343 next_ex->ei.len += ei->len;
348 __try_update_largest_extent(et, en);
354 static struct extent_node *__insert_extent_tree(struct f2fs_sb_info *sbi,
355 struct extent_tree *et, struct extent_info *ei,
356 struct rb_node **insert_p,
357 struct rb_node *insert_parent)
359 struct rb_node **p = &et->root.rb_node;
360 struct rb_node *parent = NULL;
361 struct extent_node *en = NULL;
363 if (insert_p && insert_parent) {
364 parent = insert_parent;
371 en = rb_entry(parent, struct extent_node, rb_node);
373 if (ei->fofs < en->ei.fofs)
375 else if (ei->fofs >= en->ei.fofs + en->ei.len)
381 en = __attach_extent_node(sbi, et, ei, parent, p);
385 __try_update_largest_extent(et, en);
390 static unsigned int f2fs_update_extent_tree_range(struct inode *inode,
391 pgoff_t fofs, block_t blkaddr, unsigned int len)
393 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
394 struct extent_tree *et = F2FS_I(inode)->extent_tree;
395 struct extent_node *en = NULL, *en1 = NULL;
396 struct extent_node *prev_en = NULL, *next_en = NULL;
397 struct extent_info ei, dei, prev;
398 struct rb_node **insert_p = NULL, *insert_parent = NULL;
399 unsigned int end = fofs + len;
400 unsigned int pos = (unsigned int)fofs;
405 trace_f2fs_update_extent_tree_range(inode, fofs, blkaddr, len);
407 write_lock(&et->lock);
409 if (is_inode_flag_set(F2FS_I(inode), FI_NO_EXTENT)) {
410 write_unlock(&et->lock);
418 * drop largest extent before lookup, in case it's already
419 * been shrunk from extent tree
421 __drop_largest_extent(inode, fofs, len);
423 /* 1. lookup first extent node in range [fofs, fofs + len - 1] */
424 en = __lookup_extent_tree_ret(et, fofs, &prev_en, &next_en,
425 &insert_p, &insert_parent);
429 /* 2. invlidate all extent nodes in range [fofs, fofs + len - 1] */
430 while (en && en->ei.fofs < end) {
431 unsigned int org_end;
432 int parts = 0; /* # of parts current extent split into */
434 next_en = en1 = NULL;
437 org_end = dei.fofs + dei.len;
438 f2fs_bug_on(sbi, pos >= org_end);
440 if (pos > dei.fofs && pos - dei.fofs >= F2FS_MIN_EXTENT_LEN) {
441 en->ei.len = pos - en->ei.fofs;
446 if (end < org_end && org_end - end >= F2FS_MIN_EXTENT_LEN) {
448 set_extent_info(&ei, end,
449 end - dei.fofs + dei.blk,
451 en1 = __insert_extent_tree(sbi, et, &ei,
456 en->ei.blk += end - dei.fofs;
457 en->ei.len -= end - dei.fofs;
464 struct rb_node *node = rb_next(&en->rb_node);
467 rb_entry(node, struct extent_node, rb_node)
472 __try_update_largest_extent(et, en);
474 __detach_extent_node(sbi, et, en);
477 * if original extent is split into zero or two parts, extent
478 * tree has been altered by deletion or insertion, therefore
479 * invalidate pointers regard to tree.
483 insert_parent = NULL;
486 /* update in global extent list */
487 spin_lock(&sbi->extent_lock);
488 if (!parts && !list_empty(&en->list))
491 list_add_tail(&en1->list, &sbi->extent_list);
492 spin_unlock(&sbi->extent_lock);
494 /* release extent node */
496 kmem_cache_free(extent_node_slab, en);
501 /* 3. update extent in extent cache */
503 struct extent_node *den = NULL;
505 set_extent_info(&ei, fofs, blkaddr, len);
506 en1 = __try_merge_extent_node(sbi, et, &ei, &den,
509 en1 = __insert_extent_tree(sbi, et, &ei,
510 insert_p, insert_parent);
512 /* give up extent_cache, if split and small updates happen */
514 prev.len < F2FS_MIN_EXTENT_LEN &&
515 et->largest.len < F2FS_MIN_EXTENT_LEN) {
517 set_inode_flag(F2FS_I(inode), FI_NO_EXTENT);
520 spin_lock(&sbi->extent_lock);
522 if (list_empty(&en1->list))
523 list_add_tail(&en1->list, &sbi->extent_list);
525 list_move_tail(&en1->list, &sbi->extent_list);
527 if (den && !list_empty(&den->list))
528 list_del(&den->list);
529 spin_unlock(&sbi->extent_lock);
532 kmem_cache_free(extent_node_slab, den);
535 if (is_inode_flag_set(F2FS_I(inode), FI_NO_EXTENT))
536 __free_extent_tree(sbi, et, true);
538 write_unlock(&et->lock);
540 return !__is_extent_same(&prev, &et->largest);
543 unsigned int f2fs_shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
545 struct extent_tree *treevec[EXT_TREE_VEC_SIZE];
546 struct extent_node *en, *tmp;
547 unsigned long ino = F2FS_ROOT_INO(sbi);
548 struct radix_tree_root *root = &sbi->extent_tree_root;
550 unsigned int node_cnt = 0, tree_cnt = 0;
552 bool do_free = false;
554 if (!test_opt(sbi, EXTENT_CACHE))
557 if (!atomic_read(&sbi->total_zombie_tree))
560 if (!down_write_trylock(&sbi->extent_tree_lock))
563 /* 1. remove unreferenced extent tree */
564 while ((found = radix_tree_gang_lookup(root,
565 (void **)treevec, ino, EXT_TREE_VEC_SIZE))) {
568 ino = treevec[found - 1]->ino + 1;
569 for (i = 0; i < found; i++) {
570 struct extent_tree *et = treevec[i];
572 if (!atomic_read(&et->refcount)) {
573 write_lock(&et->lock);
574 node_cnt += __free_extent_tree(sbi, et, true);
575 write_unlock(&et->lock);
577 radix_tree_delete(root, et->ino);
578 kmem_cache_free(extent_tree_slab, et);
579 atomic_dec(&sbi->total_ext_tree);
580 atomic_dec(&sbi->total_zombie_tree);
583 if (node_cnt + tree_cnt >= nr_shrink)
588 up_write(&sbi->extent_tree_lock);
591 /* 2. remove LRU extent entries */
592 if (!down_write_trylock(&sbi->extent_tree_lock))
595 remained = nr_shrink - (node_cnt + tree_cnt);
597 spin_lock(&sbi->extent_lock);
598 list_for_each_entry_safe(en, tmp, &sbi->extent_list, list) {
601 list_del_init(&en->list);
604 spin_unlock(&sbi->extent_lock);
606 if (do_free == false)
610 * reset ino for searching victims from beginning of global extent tree.
612 ino = F2FS_ROOT_INO(sbi);
614 while ((found = radix_tree_gang_lookup(root,
615 (void **)treevec, ino, EXT_TREE_VEC_SIZE))) {
618 ino = treevec[found - 1]->ino + 1;
619 for (i = 0; i < found; i++) {
620 struct extent_tree *et = treevec[i];
622 if (write_trylock(&et->lock)) {
623 node_cnt += __free_extent_tree(sbi, et, false);
624 write_unlock(&et->lock);
627 if (node_cnt + tree_cnt >= nr_shrink)
632 up_write(&sbi->extent_tree_lock);
634 trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt);
636 return node_cnt + tree_cnt;
639 unsigned int f2fs_destroy_extent_node(struct inode *inode)
641 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
642 struct extent_tree *et = F2FS_I(inode)->extent_tree;
643 unsigned int node_cnt = 0;
648 write_lock(&et->lock);
649 node_cnt = __free_extent_tree(sbi, et, true);
650 write_unlock(&et->lock);
655 void f2fs_destroy_extent_tree(struct inode *inode)
657 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
658 struct extent_tree *et = F2FS_I(inode)->extent_tree;
659 unsigned int node_cnt = 0;
664 if (inode->i_nlink && !is_bad_inode(inode) && et->count) {
665 atomic_dec(&et->refcount);
666 atomic_inc(&sbi->total_zombie_tree);
670 /* free all extent info belong to this extent tree */
671 node_cnt = f2fs_destroy_extent_node(inode);
673 /* delete extent tree entry in radix tree */
674 down_write(&sbi->extent_tree_lock);
675 atomic_dec(&et->refcount);
676 f2fs_bug_on(sbi, atomic_read(&et->refcount) || et->count);
677 radix_tree_delete(&sbi->extent_tree_root, inode->i_ino);
678 kmem_cache_free(extent_tree_slab, et);
679 atomic_dec(&sbi->total_ext_tree);
680 up_write(&sbi->extent_tree_lock);
682 F2FS_I(inode)->extent_tree = NULL;
684 trace_f2fs_destroy_extent_tree(inode, node_cnt);
687 bool f2fs_lookup_extent_cache(struct inode *inode, pgoff_t pgofs,
688 struct extent_info *ei)
690 if (!f2fs_may_extent_tree(inode))
693 return f2fs_lookup_extent_tree(inode, pgofs, ei);
696 void f2fs_update_extent_cache(struct dnode_of_data *dn)
698 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
701 if (!f2fs_may_extent_tree(dn->inode))
704 f2fs_bug_on(F2FS_I_SB(dn->inode), dn->data_blkaddr == NEW_ADDR);
707 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
710 if (f2fs_update_extent_tree_range(dn->inode, fofs, dn->data_blkaddr, 1))
714 void f2fs_update_extent_cache_range(struct dnode_of_data *dn,
715 pgoff_t fofs, block_t blkaddr, unsigned int len)
718 if (!f2fs_may_extent_tree(dn->inode))
721 if (f2fs_update_extent_tree_range(dn->inode, fofs, blkaddr, len))
725 void init_extent_cache_info(struct f2fs_sb_info *sbi)
727 INIT_RADIX_TREE(&sbi->extent_tree_root, GFP_NOIO);
728 init_rwsem(&sbi->extent_tree_lock);
729 INIT_LIST_HEAD(&sbi->extent_list);
730 spin_lock_init(&sbi->extent_lock);
731 atomic_set(&sbi->total_ext_tree, 0);
732 atomic_set(&sbi->total_zombie_tree, 0);
733 atomic_set(&sbi->total_ext_node, 0);
736 int __init create_extent_cache(void)
738 extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree",
739 sizeof(struct extent_tree));
740 if (!extent_tree_slab)
742 extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node",
743 sizeof(struct extent_node));
744 if (!extent_node_slab) {
745 kmem_cache_destroy(extent_tree_slab);
751 void destroy_extent_cache(void)
753 kmem_cache_destroy(extent_node_slab);
754 kmem_cache_destroy(extent_tree_slab);