xref: /openbmc/linux/fs/f2fs/extent_cache.c (revision 4cd82a5b)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * f2fs extent cache support
4  *
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>
9  */
10 
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 
14 #include "f2fs.h"
15 #include "node.h"
16 #include <trace/events/f2fs.h>
17 
18 static struct rb_entry *__lookup_rb_tree_fast(struct rb_entry *cached_re,
19 							unsigned int ofs)
20 {
21 	if (cached_re) {
22 		if (cached_re->ofs <= ofs &&
23 				cached_re->ofs + cached_re->len > ofs) {
24 			return cached_re;
25 		}
26 	}
27 	return NULL;
28 }
29 
30 static struct rb_entry *__lookup_rb_tree_slow(struct rb_root_cached *root,
31 							unsigned int ofs)
32 {
33 	struct rb_node *node = root->rb_root.rb_node;
34 	struct rb_entry *re;
35 
36 	while (node) {
37 		re = rb_entry(node, struct rb_entry, rb_node);
38 
39 		if (ofs < re->ofs)
40 			node = node->rb_left;
41 		else if (ofs >= re->ofs + re->len)
42 			node = node->rb_right;
43 		else
44 			return re;
45 	}
46 	return NULL;
47 }
48 
49 struct rb_entry *f2fs_lookup_rb_tree(struct rb_root_cached *root,
50 				struct rb_entry *cached_re, unsigned int ofs)
51 {
52 	struct rb_entry *re;
53 
54 	re = __lookup_rb_tree_fast(cached_re, ofs);
55 	if (!re)
56 		return __lookup_rb_tree_slow(root, ofs);
57 
58 	return re;
59 }
60 
61 struct rb_node **f2fs_lookup_rb_tree_ext(struct f2fs_sb_info *sbi,
62 					struct rb_root_cached *root,
63 					struct rb_node **parent,
64 					unsigned long long key, bool *leftmost)
65 {
66 	struct rb_node **p = &root->rb_root.rb_node;
67 	struct rb_entry *re;
68 
69 	while (*p) {
70 		*parent = *p;
71 		re = rb_entry(*parent, struct rb_entry, rb_node);
72 
73 		if (key < re->key) {
74 			p = &(*p)->rb_left;
75 		} else {
76 			p = &(*p)->rb_right;
77 			*leftmost = false;
78 		}
79 	}
80 
81 	return p;
82 }
83 
84 struct rb_node **f2fs_lookup_rb_tree_for_insert(struct f2fs_sb_info *sbi,
85 				struct rb_root_cached *root,
86 				struct rb_node **parent,
87 				unsigned int ofs, bool *leftmost)
88 {
89 	struct rb_node **p = &root->rb_root.rb_node;
90 	struct rb_entry *re;
91 
92 	while (*p) {
93 		*parent = *p;
94 		re = rb_entry(*parent, struct rb_entry, rb_node);
95 
96 		if (ofs < re->ofs) {
97 			p = &(*p)->rb_left;
98 		} else if (ofs >= re->ofs + re->len) {
99 			p = &(*p)->rb_right;
100 			*leftmost = false;
101 		} else {
102 			f2fs_bug_on(sbi, 1);
103 		}
104 	}
105 
106 	return p;
107 }
108 
109 /*
110  * lookup rb entry in position of @ofs in rb-tree,
111  * if hit, return the entry, otherwise, return NULL
112  * @prev_ex: extent before ofs
113  * @next_ex: extent after ofs
114  * @insert_p: insert point for new extent at ofs
115  * in order to simpfy the insertion after.
116  * tree must stay unchanged between lookup and insertion.
117  */
118 struct rb_entry *f2fs_lookup_rb_tree_ret(struct rb_root_cached *root,
119 				struct rb_entry *cached_re,
120 				unsigned int ofs,
121 				struct rb_entry **prev_entry,
122 				struct rb_entry **next_entry,
123 				struct rb_node ***insert_p,
124 				struct rb_node **insert_parent,
125 				bool force, bool *leftmost)
126 {
127 	struct rb_node **pnode = &root->rb_root.rb_node;
128 	struct rb_node *parent = NULL, *tmp_node;
129 	struct rb_entry *re = cached_re;
130 
131 	*insert_p = NULL;
132 	*insert_parent = NULL;
133 	*prev_entry = NULL;
134 	*next_entry = NULL;
135 
136 	if (RB_EMPTY_ROOT(&root->rb_root))
137 		return NULL;
138 
139 	if (re) {
140 		if (re->ofs <= ofs && re->ofs + re->len > ofs)
141 			goto lookup_neighbors;
142 	}
143 
144 	if (leftmost)
145 		*leftmost = true;
146 
147 	while (*pnode) {
148 		parent = *pnode;
149 		re = rb_entry(*pnode, struct rb_entry, rb_node);
150 
151 		if (ofs < re->ofs) {
152 			pnode = &(*pnode)->rb_left;
153 		} else if (ofs >= re->ofs + re->len) {
154 			pnode = &(*pnode)->rb_right;
155 			if (leftmost)
156 				*leftmost = false;
157 		} else {
158 			goto lookup_neighbors;
159 		}
160 	}
161 
162 	*insert_p = pnode;
163 	*insert_parent = parent;
164 
165 	re = rb_entry(parent, struct rb_entry, rb_node);
166 	tmp_node = parent;
167 	if (parent && ofs > re->ofs)
168 		tmp_node = rb_next(parent);
169 	*next_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
170 
171 	tmp_node = parent;
172 	if (parent && ofs < re->ofs)
173 		tmp_node = rb_prev(parent);
174 	*prev_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
175 	return NULL;
176 
177 lookup_neighbors:
178 	if (ofs == re->ofs || force) {
179 		/* lookup prev node for merging backward later */
180 		tmp_node = rb_prev(&re->rb_node);
181 		*prev_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
182 	}
183 	if (ofs == re->ofs + re->len - 1 || force) {
184 		/* lookup next node for merging frontward later */
185 		tmp_node = rb_next(&re->rb_node);
186 		*next_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
187 	}
188 	return re;
189 }
190 
191 bool f2fs_check_rb_tree_consistence(struct f2fs_sb_info *sbi,
192 				struct rb_root_cached *root, bool check_key)
193 {
194 #ifdef CONFIG_F2FS_CHECK_FS
195 	struct rb_node *cur = rb_first_cached(root), *next;
196 	struct rb_entry *cur_re, *next_re;
197 
198 	if (!cur)
199 		return true;
200 
201 	while (cur) {
202 		next = rb_next(cur);
203 		if (!next)
204 			return true;
205 
206 		cur_re = rb_entry(cur, struct rb_entry, rb_node);
207 		next_re = rb_entry(next, struct rb_entry, rb_node);
208 
209 		if (check_key) {
210 			if (cur_re->key > next_re->key) {
211 				f2fs_info(sbi, "inconsistent rbtree, "
212 					"cur(%llu) next(%llu)",
213 					cur_re->key, next_re->key);
214 				return false;
215 			}
216 			goto next;
217 		}
218 
219 		if (cur_re->ofs + cur_re->len > next_re->ofs) {
220 			f2fs_info(sbi, "inconsistent rbtree, cur(%u, %u) next(%u, %u)",
221 				  cur_re->ofs, cur_re->len,
222 				  next_re->ofs, next_re->len);
223 			return false;
224 		}
225 next:
226 		cur = next;
227 	}
228 #endif
229 	return true;
230 }
231 
232 static struct kmem_cache *extent_tree_slab;
233 static struct kmem_cache *extent_node_slab;
234 
235 static struct extent_node *__attach_extent_node(struct f2fs_sb_info *sbi,
236 				struct extent_tree *et, struct extent_info *ei,
237 				struct rb_node *parent, struct rb_node **p,
238 				bool leftmost)
239 {
240 	struct extent_node *en;
241 
242 	en = f2fs_kmem_cache_alloc(extent_node_slab, GFP_ATOMIC, false, sbi);
243 	if (!en)
244 		return NULL;
245 
246 	en->ei = *ei;
247 	INIT_LIST_HEAD(&en->list);
248 	en->et = et;
249 
250 	rb_link_node(&en->rb_node, parent, p);
251 	rb_insert_color_cached(&en->rb_node, &et->root, leftmost);
252 	atomic_inc(&et->node_cnt);
253 	atomic_inc(&sbi->total_ext_node);
254 	return en;
255 }
256 
257 static void __detach_extent_node(struct f2fs_sb_info *sbi,
258 				struct extent_tree *et, struct extent_node *en)
259 {
260 	rb_erase_cached(&en->rb_node, &et->root);
261 	atomic_dec(&et->node_cnt);
262 	atomic_dec(&sbi->total_ext_node);
263 
264 	if (et->cached_en == en)
265 		et->cached_en = NULL;
266 	kmem_cache_free(extent_node_slab, en);
267 }
268 
269 /*
270  * Flow to release an extent_node:
271  * 1. list_del_init
272  * 2. __detach_extent_node
273  * 3. kmem_cache_free.
274  */
275 static void __release_extent_node(struct f2fs_sb_info *sbi,
276 			struct extent_tree *et, struct extent_node *en)
277 {
278 	spin_lock(&sbi->extent_lock);
279 	f2fs_bug_on(sbi, list_empty(&en->list));
280 	list_del_init(&en->list);
281 	spin_unlock(&sbi->extent_lock);
282 
283 	__detach_extent_node(sbi, et, en);
284 }
285 
286 static struct extent_tree *__grab_extent_tree(struct inode *inode)
287 {
288 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
289 	struct extent_tree *et;
290 	nid_t ino = inode->i_ino;
291 
292 	mutex_lock(&sbi->extent_tree_lock);
293 	et = radix_tree_lookup(&sbi->extent_tree_root, ino);
294 	if (!et) {
295 		et = f2fs_kmem_cache_alloc(extent_tree_slab,
296 					GFP_NOFS, true, NULL);
297 		f2fs_radix_tree_insert(&sbi->extent_tree_root, ino, et);
298 		memset(et, 0, sizeof(struct extent_tree));
299 		et->ino = ino;
300 		et->root = RB_ROOT_CACHED;
301 		et->cached_en = NULL;
302 		rwlock_init(&et->lock);
303 		INIT_LIST_HEAD(&et->list);
304 		atomic_set(&et->node_cnt, 0);
305 		atomic_inc(&sbi->total_ext_tree);
306 	} else {
307 		atomic_dec(&sbi->total_zombie_tree);
308 		list_del_init(&et->list);
309 	}
310 	mutex_unlock(&sbi->extent_tree_lock);
311 
312 	/* never died until evict_inode */
313 	F2FS_I(inode)->extent_tree = et;
314 
315 	return et;
316 }
317 
318 static struct extent_node *__init_extent_tree(struct f2fs_sb_info *sbi,
319 				struct extent_tree *et, struct extent_info *ei)
320 {
321 	struct rb_node **p = &et->root.rb_root.rb_node;
322 	struct extent_node *en;
323 
324 	en = __attach_extent_node(sbi, et, ei, NULL, p, true);
325 	if (!en)
326 		return NULL;
327 
328 	et->largest = en->ei;
329 	et->cached_en = en;
330 	return en;
331 }
332 
333 static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi,
334 					struct extent_tree *et)
335 {
336 	struct rb_node *node, *next;
337 	struct extent_node *en;
338 	unsigned int count = atomic_read(&et->node_cnt);
339 
340 	node = rb_first_cached(&et->root);
341 	while (node) {
342 		next = rb_next(node);
343 		en = rb_entry(node, struct extent_node, rb_node);
344 		__release_extent_node(sbi, et, en);
345 		node = next;
346 	}
347 
348 	return count - atomic_read(&et->node_cnt);
349 }
350 
351 static void __drop_largest_extent(struct extent_tree *et,
352 					pgoff_t fofs, unsigned int len)
353 {
354 	if (fofs < et->largest.fofs + et->largest.len &&
355 			fofs + len > et->largest.fofs) {
356 		et->largest.len = 0;
357 		et->largest_updated = true;
358 	}
359 }
360 
361 /* return true, if inode page is changed */
362 static void __f2fs_init_extent_tree(struct inode *inode, struct page *ipage)
363 {
364 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
365 	struct f2fs_extent *i_ext = ipage ? &F2FS_INODE(ipage)->i_ext : NULL;
366 	struct extent_tree *et;
367 	struct extent_node *en;
368 	struct extent_info ei;
369 
370 	if (!f2fs_may_extent_tree(inode)) {
371 		/* drop largest extent */
372 		if (i_ext && i_ext->len) {
373 			f2fs_wait_on_page_writeback(ipage, NODE, true, true);
374 			i_ext->len = 0;
375 			set_page_dirty(ipage);
376 			return;
377 		}
378 		return;
379 	}
380 
381 	et = __grab_extent_tree(inode);
382 
383 	if (!i_ext || !i_ext->len)
384 		return;
385 
386 	get_extent_info(&ei, i_ext);
387 
388 	write_lock(&et->lock);
389 	if (atomic_read(&et->node_cnt))
390 		goto out;
391 
392 	en = __init_extent_tree(sbi, et, &ei);
393 	if (en) {
394 		spin_lock(&sbi->extent_lock);
395 		list_add_tail(&en->list, &sbi->extent_list);
396 		spin_unlock(&sbi->extent_lock);
397 	}
398 out:
399 	write_unlock(&et->lock);
400 }
401 
402 void f2fs_init_extent_tree(struct inode *inode, struct page *ipage)
403 {
404 	__f2fs_init_extent_tree(inode, ipage);
405 
406 	if (!F2FS_I(inode)->extent_tree)
407 		set_inode_flag(inode, FI_NO_EXTENT);
408 }
409 
410 static bool f2fs_lookup_extent_tree(struct inode *inode, pgoff_t pgofs,
411 							struct extent_info *ei)
412 {
413 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
414 	struct extent_tree *et = F2FS_I(inode)->extent_tree;
415 	struct extent_node *en;
416 	bool ret = false;
417 
418 	f2fs_bug_on(sbi, !et);
419 
420 	trace_f2fs_lookup_extent_tree_start(inode, pgofs);
421 
422 	read_lock(&et->lock);
423 
424 	if (et->largest.fofs <= pgofs &&
425 			et->largest.fofs + et->largest.len > pgofs) {
426 		*ei = et->largest;
427 		ret = true;
428 		stat_inc_largest_node_hit(sbi);
429 		goto out;
430 	}
431 
432 	en = (struct extent_node *)f2fs_lookup_rb_tree(&et->root,
433 				(struct rb_entry *)et->cached_en, pgofs);
434 	if (!en)
435 		goto out;
436 
437 	if (en == et->cached_en)
438 		stat_inc_cached_node_hit(sbi);
439 	else
440 		stat_inc_rbtree_node_hit(sbi);
441 
442 	*ei = en->ei;
443 	spin_lock(&sbi->extent_lock);
444 	if (!list_empty(&en->list)) {
445 		list_move_tail(&en->list, &sbi->extent_list);
446 		et->cached_en = en;
447 	}
448 	spin_unlock(&sbi->extent_lock);
449 	ret = true;
450 out:
451 	stat_inc_total_hit(sbi);
452 	read_unlock(&et->lock);
453 
454 	trace_f2fs_lookup_extent_tree_end(inode, pgofs, ei);
455 	return ret;
456 }
457 
458 static struct extent_node *__try_merge_extent_node(struct f2fs_sb_info *sbi,
459 				struct extent_tree *et, struct extent_info *ei,
460 				struct extent_node *prev_ex,
461 				struct extent_node *next_ex)
462 {
463 	struct extent_node *en = NULL;
464 
465 	if (prev_ex && __is_back_mergeable(ei, &prev_ex->ei)) {
466 		prev_ex->ei.len += ei->len;
467 		ei = &prev_ex->ei;
468 		en = prev_ex;
469 	}
470 
471 	if (next_ex && __is_front_mergeable(ei, &next_ex->ei)) {
472 		next_ex->ei.fofs = ei->fofs;
473 		next_ex->ei.blk = ei->blk;
474 		next_ex->ei.len += ei->len;
475 		if (en)
476 			__release_extent_node(sbi, et, prev_ex);
477 
478 		en = next_ex;
479 	}
480 
481 	if (!en)
482 		return NULL;
483 
484 	__try_update_largest_extent(et, en);
485 
486 	spin_lock(&sbi->extent_lock);
487 	if (!list_empty(&en->list)) {
488 		list_move_tail(&en->list, &sbi->extent_list);
489 		et->cached_en = en;
490 	}
491 	spin_unlock(&sbi->extent_lock);
492 	return en;
493 }
494 
495 static struct extent_node *__insert_extent_tree(struct f2fs_sb_info *sbi,
496 				struct extent_tree *et, struct extent_info *ei,
497 				struct rb_node **insert_p,
498 				struct rb_node *insert_parent,
499 				bool leftmost)
500 {
501 	struct rb_node **p;
502 	struct rb_node *parent = NULL;
503 	struct extent_node *en = NULL;
504 
505 	if (insert_p && insert_parent) {
506 		parent = insert_parent;
507 		p = insert_p;
508 		goto do_insert;
509 	}
510 
511 	leftmost = true;
512 
513 	p = f2fs_lookup_rb_tree_for_insert(sbi, &et->root, &parent,
514 						ei->fofs, &leftmost);
515 do_insert:
516 	en = __attach_extent_node(sbi, et, ei, parent, p, leftmost);
517 	if (!en)
518 		return NULL;
519 
520 	__try_update_largest_extent(et, en);
521 
522 	/* update in global extent list */
523 	spin_lock(&sbi->extent_lock);
524 	list_add_tail(&en->list, &sbi->extent_list);
525 	et->cached_en = en;
526 	spin_unlock(&sbi->extent_lock);
527 	return en;
528 }
529 
530 static void f2fs_update_extent_tree_range(struct inode *inode,
531 				pgoff_t fofs, block_t blkaddr, unsigned int len)
532 {
533 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
534 	struct extent_tree *et = F2FS_I(inode)->extent_tree;
535 	struct extent_node *en = NULL, *en1 = NULL;
536 	struct extent_node *prev_en = NULL, *next_en = NULL;
537 	struct extent_info ei, dei, prev;
538 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
539 	unsigned int end = fofs + len;
540 	unsigned int pos = (unsigned int)fofs;
541 	bool updated = false;
542 	bool leftmost = false;
543 
544 	if (!et)
545 		return;
546 
547 	trace_f2fs_update_extent_tree_range(inode, fofs, blkaddr, len);
548 
549 	write_lock(&et->lock);
550 
551 	if (is_inode_flag_set(inode, FI_NO_EXTENT)) {
552 		write_unlock(&et->lock);
553 		return;
554 	}
555 
556 	prev = et->largest;
557 	dei.len = 0;
558 
559 	/*
560 	 * drop largest extent before lookup, in case it's already
561 	 * been shrunk from extent tree
562 	 */
563 	__drop_largest_extent(et, fofs, len);
564 
565 	/* 1. lookup first extent node in range [fofs, fofs + len - 1] */
566 	en = (struct extent_node *)f2fs_lookup_rb_tree_ret(&et->root,
567 					(struct rb_entry *)et->cached_en, fofs,
568 					(struct rb_entry **)&prev_en,
569 					(struct rb_entry **)&next_en,
570 					&insert_p, &insert_parent, false,
571 					&leftmost);
572 	if (!en)
573 		en = next_en;
574 
575 	/* 2. invlidate all extent nodes in range [fofs, fofs + len - 1] */
576 	while (en && en->ei.fofs < end) {
577 		unsigned int org_end;
578 		int parts = 0;	/* # of parts current extent split into */
579 
580 		next_en = en1 = NULL;
581 
582 		dei = en->ei;
583 		org_end = dei.fofs + dei.len;
584 		f2fs_bug_on(sbi, pos >= org_end);
585 
586 		if (pos > dei.fofs &&	pos - dei.fofs >= F2FS_MIN_EXTENT_LEN) {
587 			en->ei.len = pos - en->ei.fofs;
588 			prev_en = en;
589 			parts = 1;
590 		}
591 
592 		if (end < org_end && org_end - end >= F2FS_MIN_EXTENT_LEN) {
593 			if (parts) {
594 				set_extent_info(&ei, end,
595 						end - dei.fofs + dei.blk,
596 						org_end - end);
597 				en1 = __insert_extent_tree(sbi, et, &ei,
598 							NULL, NULL, true);
599 				next_en = en1;
600 			} else {
601 				en->ei.fofs = end;
602 				en->ei.blk += end - dei.fofs;
603 				en->ei.len -= end - dei.fofs;
604 				next_en = en;
605 			}
606 			parts++;
607 		}
608 
609 		if (!next_en) {
610 			struct rb_node *node = rb_next(&en->rb_node);
611 
612 			next_en = rb_entry_safe(node, struct extent_node,
613 						rb_node);
614 		}
615 
616 		if (parts)
617 			__try_update_largest_extent(et, en);
618 		else
619 			__release_extent_node(sbi, et, en);
620 
621 		/*
622 		 * if original extent is split into zero or two parts, extent
623 		 * tree has been altered by deletion or insertion, therefore
624 		 * invalidate pointers regard to tree.
625 		 */
626 		if (parts != 1) {
627 			insert_p = NULL;
628 			insert_parent = NULL;
629 		}
630 		en = next_en;
631 	}
632 
633 	/* 3. update extent in extent cache */
634 	if (blkaddr) {
635 
636 		set_extent_info(&ei, fofs, blkaddr, len);
637 		if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
638 			__insert_extent_tree(sbi, et, &ei,
639 					insert_p, insert_parent, leftmost);
640 
641 		/* give up extent_cache, if split and small updates happen */
642 		if (dei.len >= 1 &&
643 				prev.len < F2FS_MIN_EXTENT_LEN &&
644 				et->largest.len < F2FS_MIN_EXTENT_LEN) {
645 			et->largest.len = 0;
646 			et->largest_updated = true;
647 			set_inode_flag(inode, FI_NO_EXTENT);
648 		}
649 	}
650 
651 	if (is_inode_flag_set(inode, FI_NO_EXTENT))
652 		__free_extent_tree(sbi, et);
653 
654 	if (et->largest_updated) {
655 		et->largest_updated = false;
656 		updated = true;
657 	}
658 
659 	write_unlock(&et->lock);
660 
661 	if (updated)
662 		f2fs_mark_inode_dirty_sync(inode, true);
663 }
664 
665 #ifdef CONFIG_F2FS_FS_COMPRESSION
666 void f2fs_update_extent_tree_range_compressed(struct inode *inode,
667 				pgoff_t fofs, block_t blkaddr, unsigned int llen,
668 				unsigned int c_len)
669 {
670 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
671 	struct extent_tree *et = F2FS_I(inode)->extent_tree;
672 	struct extent_node *en = NULL;
673 	struct extent_node *prev_en = NULL, *next_en = NULL;
674 	struct extent_info ei;
675 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
676 	bool leftmost = false;
677 
678 	trace_f2fs_update_extent_tree_range(inode, fofs, blkaddr, llen);
679 
680 	/* it is safe here to check FI_NO_EXTENT w/o et->lock in ro image */
681 	if (is_inode_flag_set(inode, FI_NO_EXTENT))
682 		return;
683 
684 	write_lock(&et->lock);
685 
686 	en = (struct extent_node *)f2fs_lookup_rb_tree_ret(&et->root,
687 				(struct rb_entry *)et->cached_en, fofs,
688 				(struct rb_entry **)&prev_en,
689 				(struct rb_entry **)&next_en,
690 				&insert_p, &insert_parent, false,
691 				&leftmost);
692 	if (en)
693 		goto unlock_out;
694 
695 	set_extent_info(&ei, fofs, blkaddr, llen);
696 	ei.c_len = c_len;
697 
698 	if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
699 		__insert_extent_tree(sbi, et, &ei,
700 				insert_p, insert_parent, leftmost);
701 unlock_out:
702 	write_unlock(&et->lock);
703 }
704 #endif
705 
706 unsigned int f2fs_shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
707 {
708 	struct extent_tree *et, *next;
709 	struct extent_node *en;
710 	unsigned int node_cnt = 0, tree_cnt = 0;
711 	int remained;
712 
713 	if (!test_opt(sbi, EXTENT_CACHE))
714 		return 0;
715 
716 	if (!atomic_read(&sbi->total_zombie_tree))
717 		goto free_node;
718 
719 	if (!mutex_trylock(&sbi->extent_tree_lock))
720 		goto out;
721 
722 	/* 1. remove unreferenced extent tree */
723 	list_for_each_entry_safe(et, next, &sbi->zombie_list, list) {
724 		if (atomic_read(&et->node_cnt)) {
725 			write_lock(&et->lock);
726 			node_cnt += __free_extent_tree(sbi, et);
727 			write_unlock(&et->lock);
728 		}
729 		f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
730 		list_del_init(&et->list);
731 		radix_tree_delete(&sbi->extent_tree_root, et->ino);
732 		kmem_cache_free(extent_tree_slab, et);
733 		atomic_dec(&sbi->total_ext_tree);
734 		atomic_dec(&sbi->total_zombie_tree);
735 		tree_cnt++;
736 
737 		if (node_cnt + tree_cnt >= nr_shrink)
738 			goto unlock_out;
739 		cond_resched();
740 	}
741 	mutex_unlock(&sbi->extent_tree_lock);
742 
743 free_node:
744 	/* 2. remove LRU extent entries */
745 	if (!mutex_trylock(&sbi->extent_tree_lock))
746 		goto out;
747 
748 	remained = nr_shrink - (node_cnt + tree_cnt);
749 
750 	spin_lock(&sbi->extent_lock);
751 	for (; remained > 0; remained--) {
752 		if (list_empty(&sbi->extent_list))
753 			break;
754 		en = list_first_entry(&sbi->extent_list,
755 					struct extent_node, list);
756 		et = en->et;
757 		if (!write_trylock(&et->lock)) {
758 			/* refresh this extent node's position in extent list */
759 			list_move_tail(&en->list, &sbi->extent_list);
760 			continue;
761 		}
762 
763 		list_del_init(&en->list);
764 		spin_unlock(&sbi->extent_lock);
765 
766 		__detach_extent_node(sbi, et, en);
767 
768 		write_unlock(&et->lock);
769 		node_cnt++;
770 		spin_lock(&sbi->extent_lock);
771 	}
772 	spin_unlock(&sbi->extent_lock);
773 
774 unlock_out:
775 	mutex_unlock(&sbi->extent_tree_lock);
776 out:
777 	trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt);
778 
779 	return node_cnt + tree_cnt;
780 }
781 
782 unsigned int f2fs_destroy_extent_node(struct inode *inode)
783 {
784 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
785 	struct extent_tree *et = F2FS_I(inode)->extent_tree;
786 	unsigned int node_cnt = 0;
787 
788 	if (!et || !atomic_read(&et->node_cnt))
789 		return 0;
790 
791 	write_lock(&et->lock);
792 	node_cnt = __free_extent_tree(sbi, et);
793 	write_unlock(&et->lock);
794 
795 	return node_cnt;
796 }
797 
798 void f2fs_drop_extent_tree(struct inode *inode)
799 {
800 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
801 	struct extent_tree *et = F2FS_I(inode)->extent_tree;
802 	bool updated = false;
803 
804 	if (!f2fs_may_extent_tree(inode))
805 		return;
806 
807 	set_inode_flag(inode, FI_NO_EXTENT);
808 
809 	write_lock(&et->lock);
810 	__free_extent_tree(sbi, et);
811 	if (et->largest.len) {
812 		et->largest.len = 0;
813 		updated = true;
814 	}
815 	write_unlock(&et->lock);
816 	if (updated)
817 		f2fs_mark_inode_dirty_sync(inode, true);
818 }
819 
820 void f2fs_destroy_extent_tree(struct inode *inode)
821 {
822 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
823 	struct extent_tree *et = F2FS_I(inode)->extent_tree;
824 	unsigned int node_cnt = 0;
825 
826 	if (!et)
827 		return;
828 
829 	if (inode->i_nlink && !is_bad_inode(inode) &&
830 					atomic_read(&et->node_cnt)) {
831 		mutex_lock(&sbi->extent_tree_lock);
832 		list_add_tail(&et->list, &sbi->zombie_list);
833 		atomic_inc(&sbi->total_zombie_tree);
834 		mutex_unlock(&sbi->extent_tree_lock);
835 		return;
836 	}
837 
838 	/* free all extent info belong to this extent tree */
839 	node_cnt = f2fs_destroy_extent_node(inode);
840 
841 	/* delete extent tree entry in radix tree */
842 	mutex_lock(&sbi->extent_tree_lock);
843 	f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
844 	radix_tree_delete(&sbi->extent_tree_root, inode->i_ino);
845 	kmem_cache_free(extent_tree_slab, et);
846 	atomic_dec(&sbi->total_ext_tree);
847 	mutex_unlock(&sbi->extent_tree_lock);
848 
849 	F2FS_I(inode)->extent_tree = NULL;
850 
851 	trace_f2fs_destroy_extent_tree(inode, node_cnt);
852 }
853 
854 bool f2fs_lookup_extent_cache(struct inode *inode, pgoff_t pgofs,
855 					struct extent_info *ei)
856 {
857 	if (!f2fs_may_extent_tree(inode))
858 		return false;
859 
860 	return f2fs_lookup_extent_tree(inode, pgofs, ei);
861 }
862 
863 void f2fs_update_extent_cache(struct dnode_of_data *dn)
864 {
865 	pgoff_t fofs;
866 	block_t blkaddr;
867 
868 	if (!f2fs_may_extent_tree(dn->inode))
869 		return;
870 
871 	if (dn->data_blkaddr == NEW_ADDR)
872 		blkaddr = NULL_ADDR;
873 	else
874 		blkaddr = dn->data_blkaddr;
875 
876 	fofs = f2fs_start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
877 								dn->ofs_in_node;
878 	f2fs_update_extent_tree_range(dn->inode, fofs, blkaddr, 1);
879 }
880 
881 void f2fs_update_extent_cache_range(struct dnode_of_data *dn,
882 				pgoff_t fofs, block_t blkaddr, unsigned int len)
883 
884 {
885 	if (!f2fs_may_extent_tree(dn->inode))
886 		return;
887 
888 	f2fs_update_extent_tree_range(dn->inode, fofs, blkaddr, len);
889 }
890 
891 void f2fs_init_extent_cache_info(struct f2fs_sb_info *sbi)
892 {
893 	INIT_RADIX_TREE(&sbi->extent_tree_root, GFP_NOIO);
894 	mutex_init(&sbi->extent_tree_lock);
895 	INIT_LIST_HEAD(&sbi->extent_list);
896 	spin_lock_init(&sbi->extent_lock);
897 	atomic_set(&sbi->total_ext_tree, 0);
898 	INIT_LIST_HEAD(&sbi->zombie_list);
899 	atomic_set(&sbi->total_zombie_tree, 0);
900 	atomic_set(&sbi->total_ext_node, 0);
901 }
902 
903 int __init f2fs_create_extent_cache(void)
904 {
905 	extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree",
906 			sizeof(struct extent_tree));
907 	if (!extent_tree_slab)
908 		return -ENOMEM;
909 	extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node",
910 			sizeof(struct extent_node));
911 	if (!extent_node_slab) {
912 		kmem_cache_destroy(extent_tree_slab);
913 		return -ENOMEM;
914 	}
915 	return 0;
916 }
917 
918 void f2fs_destroy_extent_cache(void)
919 {
920 	kmem_cache_destroy(extent_node_slab);
921 	kmem_cache_destroy(extent_tree_slab);
922 }
923