xref: /openbmc/linux/fs/f2fs/extent_cache.c (revision c699ce1a)
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  * block_age-based extent cache added by:
11  * Copyright (c) 2022 xiaomi Co., Ltd.
12  *             http://www.xiaomi.com/
13  */
14 
15 #include <linux/fs.h>
16 #include <linux/f2fs_fs.h>
17 
18 #include "f2fs.h"
19 #include "node.h"
20 #include <trace/events/f2fs.h>
21 
22 static void __set_extent_info(struct extent_info *ei,
23 				unsigned int fofs, unsigned int len,
24 				block_t blk, bool keep_clen,
25 				unsigned long age, unsigned long last_blocks,
26 				enum extent_type type)
27 {
28 	ei->fofs = fofs;
29 	ei->len = len;
30 
31 	if (type == EX_READ) {
32 		ei->blk = blk;
33 		if (keep_clen)
34 			return;
35 #ifdef CONFIG_F2FS_FS_COMPRESSION
36 		ei->c_len = 0;
37 #endif
38 	} else if (type == EX_BLOCK_AGE) {
39 		ei->age = age;
40 		ei->last_blocks = last_blocks;
41 	}
42 }
43 
44 static bool __may_read_extent_tree(struct inode *inode)
45 {
46 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
47 
48 	if (!test_opt(sbi, READ_EXTENT_CACHE))
49 		return false;
50 	if (is_inode_flag_set(inode, FI_NO_EXTENT))
51 		return false;
52 	if (is_inode_flag_set(inode, FI_COMPRESSED_FILE) &&
53 			 !f2fs_sb_has_readonly(sbi))
54 		return false;
55 	return S_ISREG(inode->i_mode);
56 }
57 
58 static bool __may_age_extent_tree(struct inode *inode)
59 {
60 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
61 
62 	if (!test_opt(sbi, AGE_EXTENT_CACHE))
63 		return false;
64 	/* don't cache block age info for cold file */
65 	if (is_inode_flag_set(inode, FI_COMPRESSED_FILE))
66 		return false;
67 	if (file_is_cold(inode))
68 		return false;
69 
70 	return S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode);
71 }
72 
73 static bool __init_may_extent_tree(struct inode *inode, enum extent_type type)
74 {
75 	if (type == EX_READ)
76 		return __may_read_extent_tree(inode);
77 	else if (type == EX_BLOCK_AGE)
78 		return __may_age_extent_tree(inode);
79 	return false;
80 }
81 
82 static bool __may_extent_tree(struct inode *inode, enum extent_type type)
83 {
84 	/*
85 	 * for recovered files during mount do not create extents
86 	 * if shrinker is not registered.
87 	 */
88 	if (list_empty(&F2FS_I_SB(inode)->s_list))
89 		return false;
90 
91 	return __init_may_extent_tree(inode, type);
92 }
93 
94 static void __try_update_largest_extent(struct extent_tree *et,
95 						struct extent_node *en)
96 {
97 	if (et->type != EX_READ)
98 		return;
99 	if (en->ei.len <= et->largest.len)
100 		return;
101 
102 	et->largest = en->ei;
103 	et->largest_updated = true;
104 }
105 
106 static bool __is_extent_mergeable(struct extent_info *back,
107 		struct extent_info *front, enum extent_type type)
108 {
109 	if (type == EX_READ) {
110 #ifdef CONFIG_F2FS_FS_COMPRESSION
111 		if (back->c_len && back->len != back->c_len)
112 			return false;
113 		if (front->c_len && front->len != front->c_len)
114 			return false;
115 #endif
116 		return (back->fofs + back->len == front->fofs &&
117 				back->blk + back->len == front->blk);
118 	} else if (type == EX_BLOCK_AGE) {
119 		return (back->fofs + back->len == front->fofs &&
120 			abs(back->age - front->age) <= SAME_AGE_REGION &&
121 			abs(back->last_blocks - front->last_blocks) <=
122 							SAME_AGE_REGION);
123 	}
124 	return false;
125 }
126 
127 static bool __is_back_mergeable(struct extent_info *cur,
128 		struct extent_info *back, enum extent_type type)
129 {
130 	return __is_extent_mergeable(back, cur, type);
131 }
132 
133 static bool __is_front_mergeable(struct extent_info *cur,
134 		struct extent_info *front, enum extent_type type)
135 {
136 	return __is_extent_mergeable(cur, front, type);
137 }
138 
139 static struct rb_entry *__lookup_rb_tree_fast(struct rb_entry *cached_re,
140 							unsigned int ofs)
141 {
142 	if (cached_re) {
143 		if (cached_re->ofs <= ofs &&
144 				cached_re->ofs + cached_re->len > ofs) {
145 			return cached_re;
146 		}
147 	}
148 	return NULL;
149 }
150 
151 static struct rb_entry *__lookup_rb_tree_slow(struct rb_root_cached *root,
152 							unsigned int ofs)
153 {
154 	struct rb_node *node = root->rb_root.rb_node;
155 	struct rb_entry *re;
156 
157 	while (node) {
158 		re = rb_entry(node, struct rb_entry, rb_node);
159 
160 		if (ofs < re->ofs)
161 			node = node->rb_left;
162 		else if (ofs >= re->ofs + re->len)
163 			node = node->rb_right;
164 		else
165 			return re;
166 	}
167 	return NULL;
168 }
169 
170 struct rb_entry *f2fs_lookup_rb_tree(struct rb_root_cached *root,
171 				struct rb_entry *cached_re, unsigned int ofs)
172 {
173 	struct rb_entry *re;
174 
175 	re = __lookup_rb_tree_fast(cached_re, ofs);
176 	if (!re)
177 		return __lookup_rb_tree_slow(root, ofs);
178 
179 	return re;
180 }
181 
182 struct rb_node **f2fs_lookup_rb_tree_ext(struct f2fs_sb_info *sbi,
183 					struct rb_root_cached *root,
184 					struct rb_node **parent,
185 					unsigned long long key, bool *leftmost)
186 {
187 	struct rb_node **p = &root->rb_root.rb_node;
188 	struct rb_entry *re;
189 
190 	while (*p) {
191 		*parent = *p;
192 		re = rb_entry(*parent, struct rb_entry, rb_node);
193 
194 		if (key < re->key) {
195 			p = &(*p)->rb_left;
196 		} else {
197 			p = &(*p)->rb_right;
198 			*leftmost = false;
199 		}
200 	}
201 
202 	return p;
203 }
204 
205 struct rb_node **f2fs_lookup_rb_tree_for_insert(struct f2fs_sb_info *sbi,
206 				struct rb_root_cached *root,
207 				struct rb_node **parent,
208 				unsigned int ofs, bool *leftmost)
209 {
210 	struct rb_node **p = &root->rb_root.rb_node;
211 	struct rb_entry *re;
212 
213 	while (*p) {
214 		*parent = *p;
215 		re = rb_entry(*parent, struct rb_entry, rb_node);
216 
217 		if (ofs < re->ofs) {
218 			p = &(*p)->rb_left;
219 		} else if (ofs >= re->ofs + re->len) {
220 			p = &(*p)->rb_right;
221 			*leftmost = false;
222 		} else {
223 			f2fs_bug_on(sbi, 1);
224 		}
225 	}
226 
227 	return p;
228 }
229 
230 /*
231  * lookup rb entry in position of @ofs in rb-tree,
232  * if hit, return the entry, otherwise, return NULL
233  * @prev_ex: extent before ofs
234  * @next_ex: extent after ofs
235  * @insert_p: insert point for new extent at ofs
236  * in order to simpfy the insertion after.
237  * tree must stay unchanged between lookup and insertion.
238  */
239 struct rb_entry *f2fs_lookup_rb_tree_ret(struct rb_root_cached *root,
240 				struct rb_entry *cached_re,
241 				unsigned int ofs,
242 				struct rb_entry **prev_entry,
243 				struct rb_entry **next_entry,
244 				struct rb_node ***insert_p,
245 				struct rb_node **insert_parent,
246 				bool force, bool *leftmost)
247 {
248 	struct rb_node **pnode = &root->rb_root.rb_node;
249 	struct rb_node *parent = NULL, *tmp_node;
250 	struct rb_entry *re = cached_re;
251 
252 	*insert_p = NULL;
253 	*insert_parent = NULL;
254 	*prev_entry = NULL;
255 	*next_entry = NULL;
256 
257 	if (RB_EMPTY_ROOT(&root->rb_root))
258 		return NULL;
259 
260 	if (re) {
261 		if (re->ofs <= ofs && re->ofs + re->len > ofs)
262 			goto lookup_neighbors;
263 	}
264 
265 	if (leftmost)
266 		*leftmost = true;
267 
268 	while (*pnode) {
269 		parent = *pnode;
270 		re = rb_entry(*pnode, struct rb_entry, rb_node);
271 
272 		if (ofs < re->ofs) {
273 			pnode = &(*pnode)->rb_left;
274 		} else if (ofs >= re->ofs + re->len) {
275 			pnode = &(*pnode)->rb_right;
276 			if (leftmost)
277 				*leftmost = false;
278 		} else {
279 			goto lookup_neighbors;
280 		}
281 	}
282 
283 	*insert_p = pnode;
284 	*insert_parent = parent;
285 
286 	re = rb_entry(parent, struct rb_entry, rb_node);
287 	tmp_node = parent;
288 	if (parent && ofs > re->ofs)
289 		tmp_node = rb_next(parent);
290 	*next_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
291 
292 	tmp_node = parent;
293 	if (parent && ofs < re->ofs)
294 		tmp_node = rb_prev(parent);
295 	*prev_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
296 	return NULL;
297 
298 lookup_neighbors:
299 	if (ofs == re->ofs || force) {
300 		/* lookup prev node for merging backward later */
301 		tmp_node = rb_prev(&re->rb_node);
302 		*prev_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
303 	}
304 	if (ofs == re->ofs + re->len - 1 || force) {
305 		/* lookup next node for merging frontward later */
306 		tmp_node = rb_next(&re->rb_node);
307 		*next_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
308 	}
309 	return re;
310 }
311 
312 bool f2fs_check_rb_tree_consistence(struct f2fs_sb_info *sbi,
313 				struct rb_root_cached *root, bool check_key)
314 {
315 #ifdef CONFIG_F2FS_CHECK_FS
316 	struct rb_node *cur = rb_first_cached(root), *next;
317 	struct rb_entry *cur_re, *next_re;
318 
319 	if (!cur)
320 		return true;
321 
322 	while (cur) {
323 		next = rb_next(cur);
324 		if (!next)
325 			return true;
326 
327 		cur_re = rb_entry(cur, struct rb_entry, rb_node);
328 		next_re = rb_entry(next, struct rb_entry, rb_node);
329 
330 		if (check_key) {
331 			if (cur_re->key > next_re->key) {
332 				f2fs_info(sbi, "inconsistent rbtree, "
333 					"cur(%llu) next(%llu)",
334 					cur_re->key, next_re->key);
335 				return false;
336 			}
337 			goto next;
338 		}
339 
340 		if (cur_re->ofs + cur_re->len > next_re->ofs) {
341 			f2fs_info(sbi, "inconsistent rbtree, cur(%u, %u) next(%u, %u)",
342 				  cur_re->ofs, cur_re->len,
343 				  next_re->ofs, next_re->len);
344 			return false;
345 		}
346 next:
347 		cur = next;
348 	}
349 #endif
350 	return true;
351 }
352 
353 static struct kmem_cache *extent_tree_slab;
354 static struct kmem_cache *extent_node_slab;
355 
356 static struct extent_node *__attach_extent_node(struct f2fs_sb_info *sbi,
357 				struct extent_tree *et, struct extent_info *ei,
358 				struct rb_node *parent, struct rb_node **p,
359 				bool leftmost)
360 {
361 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
362 	struct extent_node *en;
363 
364 	en = f2fs_kmem_cache_alloc(extent_node_slab, GFP_ATOMIC, false, sbi);
365 	if (!en)
366 		return NULL;
367 
368 	en->ei = *ei;
369 	INIT_LIST_HEAD(&en->list);
370 	en->et = et;
371 
372 	rb_link_node(&en->rb_node, parent, p);
373 	rb_insert_color_cached(&en->rb_node, &et->root, leftmost);
374 	atomic_inc(&et->node_cnt);
375 	atomic_inc(&eti->total_ext_node);
376 	return en;
377 }
378 
379 static void __detach_extent_node(struct f2fs_sb_info *sbi,
380 				struct extent_tree *et, struct extent_node *en)
381 {
382 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
383 
384 	rb_erase_cached(&en->rb_node, &et->root);
385 	atomic_dec(&et->node_cnt);
386 	atomic_dec(&eti->total_ext_node);
387 
388 	if (et->cached_en == en)
389 		et->cached_en = NULL;
390 	kmem_cache_free(extent_node_slab, en);
391 }
392 
393 /*
394  * Flow to release an extent_node:
395  * 1. list_del_init
396  * 2. __detach_extent_node
397  * 3. kmem_cache_free.
398  */
399 static void __release_extent_node(struct f2fs_sb_info *sbi,
400 			struct extent_tree *et, struct extent_node *en)
401 {
402 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
403 
404 	spin_lock(&eti->extent_lock);
405 	f2fs_bug_on(sbi, list_empty(&en->list));
406 	list_del_init(&en->list);
407 	spin_unlock(&eti->extent_lock);
408 
409 	__detach_extent_node(sbi, et, en);
410 }
411 
412 static struct extent_tree *__grab_extent_tree(struct inode *inode,
413 						enum extent_type type)
414 {
415 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
416 	struct extent_tree_info *eti = &sbi->extent_tree[type];
417 	struct extent_tree *et;
418 	nid_t ino = inode->i_ino;
419 
420 	mutex_lock(&eti->extent_tree_lock);
421 	et = radix_tree_lookup(&eti->extent_tree_root, ino);
422 	if (!et) {
423 		et = f2fs_kmem_cache_alloc(extent_tree_slab,
424 					GFP_NOFS, true, NULL);
425 		f2fs_radix_tree_insert(&eti->extent_tree_root, ino, et);
426 		memset(et, 0, sizeof(struct extent_tree));
427 		et->ino = ino;
428 		et->type = type;
429 		et->root = RB_ROOT_CACHED;
430 		et->cached_en = NULL;
431 		rwlock_init(&et->lock);
432 		INIT_LIST_HEAD(&et->list);
433 		atomic_set(&et->node_cnt, 0);
434 		atomic_inc(&eti->total_ext_tree);
435 	} else {
436 		atomic_dec(&eti->total_zombie_tree);
437 		list_del_init(&et->list);
438 	}
439 	mutex_unlock(&eti->extent_tree_lock);
440 
441 	/* never died until evict_inode */
442 	F2FS_I(inode)->extent_tree[type] = et;
443 
444 	return et;
445 }
446 
447 static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi,
448 					struct extent_tree *et)
449 {
450 	struct rb_node *node, *next;
451 	struct extent_node *en;
452 	unsigned int count = atomic_read(&et->node_cnt);
453 
454 	node = rb_first_cached(&et->root);
455 	while (node) {
456 		next = rb_next(node);
457 		en = rb_entry(node, struct extent_node, rb_node);
458 		__release_extent_node(sbi, et, en);
459 		node = next;
460 	}
461 
462 	return count - atomic_read(&et->node_cnt);
463 }
464 
465 static void __drop_largest_extent(struct extent_tree *et,
466 					pgoff_t fofs, unsigned int len)
467 {
468 	if (fofs < et->largest.fofs + et->largest.len &&
469 			fofs + len > et->largest.fofs) {
470 		et->largest.len = 0;
471 		et->largest_updated = true;
472 	}
473 }
474 
475 void f2fs_init_read_extent_tree(struct inode *inode, struct page *ipage)
476 {
477 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
478 	struct extent_tree_info *eti = &sbi->extent_tree[EX_READ];
479 	struct f2fs_extent *i_ext = &F2FS_INODE(ipage)->i_ext;
480 	struct extent_tree *et;
481 	struct extent_node *en;
482 	struct extent_info ei;
483 
484 	if (!__may_extent_tree(inode, EX_READ)) {
485 		/* drop largest read extent */
486 		if (i_ext && i_ext->len) {
487 			f2fs_wait_on_page_writeback(ipage, NODE, true, true);
488 			i_ext->len = 0;
489 			set_page_dirty(ipage);
490 		}
491 		goto out;
492 	}
493 
494 	et = __grab_extent_tree(inode, EX_READ);
495 
496 	if (!i_ext || !i_ext->len)
497 		goto out;
498 
499 	get_read_extent_info(&ei, i_ext);
500 
501 	write_lock(&et->lock);
502 	if (atomic_read(&et->node_cnt))
503 		goto unlock_out;
504 
505 	en = __attach_extent_node(sbi, et, &ei, NULL,
506 				&et->root.rb_root.rb_node, true);
507 	if (en) {
508 		et->largest = en->ei;
509 		et->cached_en = en;
510 
511 		spin_lock(&eti->extent_lock);
512 		list_add_tail(&en->list, &eti->extent_list);
513 		spin_unlock(&eti->extent_lock);
514 	}
515 unlock_out:
516 	write_unlock(&et->lock);
517 out:
518 	if (!F2FS_I(inode)->extent_tree[EX_READ])
519 		set_inode_flag(inode, FI_NO_EXTENT);
520 }
521 
522 void f2fs_init_age_extent_tree(struct inode *inode)
523 {
524 	if (!__init_may_extent_tree(inode, EX_BLOCK_AGE))
525 		return;
526 	__grab_extent_tree(inode, EX_BLOCK_AGE);
527 }
528 
529 void f2fs_init_extent_tree(struct inode *inode)
530 {
531 	/* initialize read cache */
532 	if (__init_may_extent_tree(inode, EX_READ))
533 		__grab_extent_tree(inode, EX_READ);
534 
535 	/* initialize block age cache */
536 	if (__init_may_extent_tree(inode, EX_BLOCK_AGE))
537 		__grab_extent_tree(inode, EX_BLOCK_AGE);
538 }
539 
540 static bool __lookup_extent_tree(struct inode *inode, pgoff_t pgofs,
541 			struct extent_info *ei, enum extent_type type)
542 {
543 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
544 	struct extent_tree_info *eti = &sbi->extent_tree[type];
545 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
546 	struct extent_node *en;
547 	bool ret = false;
548 
549 	f2fs_bug_on(sbi, !et);
550 
551 	trace_f2fs_lookup_extent_tree_start(inode, pgofs, type);
552 
553 	read_lock(&et->lock);
554 
555 	if (type == EX_READ &&
556 			et->largest.fofs <= pgofs &&
557 			et->largest.fofs + et->largest.len > pgofs) {
558 		*ei = et->largest;
559 		ret = true;
560 		stat_inc_largest_node_hit(sbi);
561 		goto out;
562 	}
563 
564 	en = (struct extent_node *)f2fs_lookup_rb_tree(&et->root,
565 				(struct rb_entry *)et->cached_en, pgofs);
566 	if (!en)
567 		goto out;
568 
569 	if (en == et->cached_en)
570 		stat_inc_cached_node_hit(sbi, type);
571 	else
572 		stat_inc_rbtree_node_hit(sbi, type);
573 
574 	*ei = en->ei;
575 	spin_lock(&eti->extent_lock);
576 	if (!list_empty(&en->list)) {
577 		list_move_tail(&en->list, &eti->extent_list);
578 		et->cached_en = en;
579 	}
580 	spin_unlock(&eti->extent_lock);
581 	ret = true;
582 out:
583 	stat_inc_total_hit(sbi, type);
584 	read_unlock(&et->lock);
585 
586 	if (type == EX_READ)
587 		trace_f2fs_lookup_read_extent_tree_end(inode, pgofs, ei);
588 	else if (type == EX_BLOCK_AGE)
589 		trace_f2fs_lookup_age_extent_tree_end(inode, pgofs, ei);
590 	return ret;
591 }
592 
593 static struct extent_node *__try_merge_extent_node(struct f2fs_sb_info *sbi,
594 				struct extent_tree *et, struct extent_info *ei,
595 				struct extent_node *prev_ex,
596 				struct extent_node *next_ex)
597 {
598 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
599 	struct extent_node *en = NULL;
600 
601 	if (prev_ex && __is_back_mergeable(ei, &prev_ex->ei, et->type)) {
602 		prev_ex->ei.len += ei->len;
603 		ei = &prev_ex->ei;
604 		en = prev_ex;
605 	}
606 
607 	if (next_ex && __is_front_mergeable(ei, &next_ex->ei, et->type)) {
608 		next_ex->ei.fofs = ei->fofs;
609 		next_ex->ei.len += ei->len;
610 		if (et->type == EX_READ)
611 			next_ex->ei.blk = ei->blk;
612 		if (en)
613 			__release_extent_node(sbi, et, prev_ex);
614 
615 		en = next_ex;
616 	}
617 
618 	if (!en)
619 		return NULL;
620 
621 	__try_update_largest_extent(et, en);
622 
623 	spin_lock(&eti->extent_lock);
624 	if (!list_empty(&en->list)) {
625 		list_move_tail(&en->list, &eti->extent_list);
626 		et->cached_en = en;
627 	}
628 	spin_unlock(&eti->extent_lock);
629 	return en;
630 }
631 
632 static struct extent_node *__insert_extent_tree(struct f2fs_sb_info *sbi,
633 				struct extent_tree *et, struct extent_info *ei,
634 				struct rb_node **insert_p,
635 				struct rb_node *insert_parent,
636 				bool leftmost)
637 {
638 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
639 	struct rb_node **p;
640 	struct rb_node *parent = NULL;
641 	struct extent_node *en = NULL;
642 
643 	if (insert_p && insert_parent) {
644 		parent = insert_parent;
645 		p = insert_p;
646 		goto do_insert;
647 	}
648 
649 	leftmost = true;
650 
651 	p = f2fs_lookup_rb_tree_for_insert(sbi, &et->root, &parent,
652 						ei->fofs, &leftmost);
653 do_insert:
654 	en = __attach_extent_node(sbi, et, ei, parent, p, leftmost);
655 	if (!en)
656 		return NULL;
657 
658 	__try_update_largest_extent(et, en);
659 
660 	/* update in global extent list */
661 	spin_lock(&eti->extent_lock);
662 	list_add_tail(&en->list, &eti->extent_list);
663 	et->cached_en = en;
664 	spin_unlock(&eti->extent_lock);
665 	return en;
666 }
667 
668 static void __update_extent_tree_range(struct inode *inode,
669 			struct extent_info *tei, enum extent_type type)
670 {
671 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
672 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
673 	struct extent_node *en = NULL, *en1 = NULL;
674 	struct extent_node *prev_en = NULL, *next_en = NULL;
675 	struct extent_info ei, dei, prev;
676 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
677 	unsigned int fofs = tei->fofs, len = tei->len;
678 	unsigned int end = fofs + len;
679 	bool updated = false;
680 	bool leftmost = false;
681 
682 	if (!et)
683 		return;
684 
685 	if (type == EX_READ)
686 		trace_f2fs_update_read_extent_tree_range(inode, fofs, len,
687 						tei->blk, 0);
688 	else if (type == EX_BLOCK_AGE)
689 		trace_f2fs_update_age_extent_tree_range(inode, fofs, len,
690 						tei->age, tei->last_blocks);
691 
692 	write_lock(&et->lock);
693 
694 	if (type == EX_READ) {
695 		if (is_inode_flag_set(inode, FI_NO_EXTENT)) {
696 			write_unlock(&et->lock);
697 			return;
698 		}
699 
700 		prev = et->largest;
701 		dei.len = 0;
702 
703 		/*
704 		 * drop largest extent before lookup, in case it's already
705 		 * been shrunk from extent tree
706 		 */
707 		__drop_largest_extent(et, fofs, len);
708 	}
709 
710 	/* 1. lookup first extent node in range [fofs, fofs + len - 1] */
711 	en = (struct extent_node *)f2fs_lookup_rb_tree_ret(&et->root,
712 					(struct rb_entry *)et->cached_en, fofs,
713 					(struct rb_entry **)&prev_en,
714 					(struct rb_entry **)&next_en,
715 					&insert_p, &insert_parent, false,
716 					&leftmost);
717 	if (!en)
718 		en = next_en;
719 
720 	/* 2. invlidate all extent nodes in range [fofs, fofs + len - 1] */
721 	while (en && en->ei.fofs < end) {
722 		unsigned int org_end;
723 		int parts = 0;	/* # of parts current extent split into */
724 
725 		next_en = en1 = NULL;
726 
727 		dei = en->ei;
728 		org_end = dei.fofs + dei.len;
729 		f2fs_bug_on(sbi, fofs >= org_end);
730 
731 		if (fofs > dei.fofs && (type != EX_READ ||
732 				fofs - dei.fofs >= F2FS_MIN_EXTENT_LEN)) {
733 			en->ei.len = fofs - en->ei.fofs;
734 			prev_en = en;
735 			parts = 1;
736 		}
737 
738 		if (end < org_end && (type != EX_READ ||
739 				org_end - end >= F2FS_MIN_EXTENT_LEN)) {
740 			if (parts) {
741 				__set_extent_info(&ei,
742 					end, org_end - end,
743 					end - dei.fofs + dei.blk, false,
744 					dei.age, dei.last_blocks,
745 					type);
746 				en1 = __insert_extent_tree(sbi, et, &ei,
747 							NULL, NULL, true);
748 				next_en = en1;
749 			} else {
750 				__set_extent_info(&en->ei,
751 					end, en->ei.len - (end - dei.fofs),
752 					en->ei.blk + (end - dei.fofs), true,
753 					dei.age, dei.last_blocks,
754 					type);
755 				next_en = en;
756 			}
757 			parts++;
758 		}
759 
760 		if (!next_en) {
761 			struct rb_node *node = rb_next(&en->rb_node);
762 
763 			next_en = rb_entry_safe(node, struct extent_node,
764 						rb_node);
765 		}
766 
767 		if (parts)
768 			__try_update_largest_extent(et, en);
769 		else
770 			__release_extent_node(sbi, et, en);
771 
772 		/*
773 		 * if original extent is split into zero or two parts, extent
774 		 * tree has been altered by deletion or insertion, therefore
775 		 * invalidate pointers regard to tree.
776 		 */
777 		if (parts != 1) {
778 			insert_p = NULL;
779 			insert_parent = NULL;
780 		}
781 		en = next_en;
782 	}
783 
784 	if (type == EX_BLOCK_AGE)
785 		goto update_age_extent_cache;
786 
787 	/* 3. update extent in read extent cache */
788 	BUG_ON(type != EX_READ);
789 
790 	if (tei->blk) {
791 		__set_extent_info(&ei, fofs, len, tei->blk, false,
792 				  0, 0, EX_READ);
793 		if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
794 			__insert_extent_tree(sbi, et, &ei,
795 					insert_p, insert_parent, leftmost);
796 
797 		/* give up extent_cache, if split and small updates happen */
798 		if (dei.len >= 1 &&
799 				prev.len < F2FS_MIN_EXTENT_LEN &&
800 				et->largest.len < F2FS_MIN_EXTENT_LEN) {
801 			et->largest.len = 0;
802 			et->largest_updated = true;
803 			set_inode_flag(inode, FI_NO_EXTENT);
804 		}
805 	}
806 
807 	if (is_inode_flag_set(inode, FI_NO_EXTENT))
808 		__free_extent_tree(sbi, et);
809 
810 	if (et->largest_updated) {
811 		et->largest_updated = false;
812 		updated = true;
813 	}
814 	goto out_read_extent_cache;
815 update_age_extent_cache:
816 	if (!tei->last_blocks)
817 		goto out_read_extent_cache;
818 
819 	__set_extent_info(&ei, fofs, len, 0, false,
820 			tei->age, tei->last_blocks, EX_BLOCK_AGE);
821 	if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
822 		__insert_extent_tree(sbi, et, &ei,
823 					insert_p, insert_parent, leftmost);
824 out_read_extent_cache:
825 	write_unlock(&et->lock);
826 
827 	if (updated)
828 		f2fs_mark_inode_dirty_sync(inode, true);
829 }
830 
831 #ifdef CONFIG_F2FS_FS_COMPRESSION
832 void f2fs_update_read_extent_tree_range_compressed(struct inode *inode,
833 				pgoff_t fofs, block_t blkaddr, unsigned int llen,
834 				unsigned int c_len)
835 {
836 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
837 	struct extent_tree *et = F2FS_I(inode)->extent_tree[EX_READ];
838 	struct extent_node *en = NULL;
839 	struct extent_node *prev_en = NULL, *next_en = NULL;
840 	struct extent_info ei;
841 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
842 	bool leftmost = false;
843 
844 	trace_f2fs_update_read_extent_tree_range(inode, fofs, llen,
845 						blkaddr, c_len);
846 
847 	/* it is safe here to check FI_NO_EXTENT w/o et->lock in ro image */
848 	if (is_inode_flag_set(inode, FI_NO_EXTENT))
849 		return;
850 
851 	write_lock(&et->lock);
852 
853 	en = (struct extent_node *)f2fs_lookup_rb_tree_ret(&et->root,
854 				(struct rb_entry *)et->cached_en, fofs,
855 				(struct rb_entry **)&prev_en,
856 				(struct rb_entry **)&next_en,
857 				&insert_p, &insert_parent, false,
858 				&leftmost);
859 	if (en)
860 		goto unlock_out;
861 
862 	__set_extent_info(&ei, fofs, llen, blkaddr, true, 0, 0, EX_READ);
863 	ei.c_len = c_len;
864 
865 	if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
866 		__insert_extent_tree(sbi, et, &ei,
867 				insert_p, insert_parent, leftmost);
868 unlock_out:
869 	write_unlock(&et->lock);
870 }
871 #endif
872 
873 static unsigned long long __calculate_block_age(unsigned long long new,
874 						unsigned long long old)
875 {
876 	unsigned long long diff;
877 
878 	diff = (new >= old) ? new - (new - old) : new + (old - new);
879 
880 	return div_u64(diff * LAST_AGE_WEIGHT, 100);
881 }
882 
883 /* This returns a new age and allocated blocks in ei */
884 static int __get_new_block_age(struct inode *inode, struct extent_info *ei)
885 {
886 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
887 	loff_t f_size = i_size_read(inode);
888 	unsigned long long cur_blocks =
889 				atomic64_read(&sbi->allocated_data_blocks);
890 
891 	/*
892 	 * When I/O is not aligned to a PAGE_SIZE, update will happen to the last
893 	 * file block even in seq write. So don't record age for newly last file
894 	 * block here.
895 	 */
896 	if ((f_size >> PAGE_SHIFT) == ei->fofs && f_size & (PAGE_SIZE - 1) &&
897 			ei->blk == NEW_ADDR)
898 		return -EINVAL;
899 
900 	if (__lookup_extent_tree(inode, ei->fofs, ei, EX_BLOCK_AGE)) {
901 		unsigned long long cur_age;
902 
903 		if (cur_blocks >= ei->last_blocks)
904 			cur_age = cur_blocks - ei->last_blocks;
905 		else
906 			/* allocated_data_blocks overflow */
907 			cur_age = ULLONG_MAX - ei->last_blocks + cur_blocks;
908 
909 		if (ei->age)
910 			ei->age = __calculate_block_age(cur_age, ei->age);
911 		else
912 			ei->age = cur_age;
913 		ei->last_blocks = cur_blocks;
914 		WARN_ON(ei->age > cur_blocks);
915 		return 0;
916 	}
917 
918 	f2fs_bug_on(sbi, ei->blk == NULL_ADDR);
919 
920 	/* the data block was allocated for the first time */
921 	if (ei->blk == NEW_ADDR)
922 		goto out;
923 
924 	if (__is_valid_data_blkaddr(ei->blk) &&
925 			!f2fs_is_valid_blkaddr(sbi, ei->blk, DATA_GENERIC_ENHANCE)) {
926 		f2fs_bug_on(sbi, 1);
927 		return -EINVAL;
928 	}
929 out:
930 	/*
931 	 * init block age with zero, this can happen when the block age extent
932 	 * was reclaimed due to memory constraint or system reboot
933 	 */
934 	ei->age = 0;
935 	ei->last_blocks = cur_blocks;
936 	return 0;
937 }
938 
939 static void __update_extent_cache(struct dnode_of_data *dn, enum extent_type type)
940 {
941 	struct extent_info ei;
942 
943 	if (!__may_extent_tree(dn->inode, type))
944 		return;
945 
946 	ei.fofs = f2fs_start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
947 								dn->ofs_in_node;
948 	ei.len = 1;
949 
950 	if (type == EX_READ) {
951 		if (dn->data_blkaddr == NEW_ADDR)
952 			ei.blk = NULL_ADDR;
953 		else
954 			ei.blk = dn->data_blkaddr;
955 	} else if (type == EX_BLOCK_AGE) {
956 		ei.blk = dn->data_blkaddr;
957 		if (__get_new_block_age(dn->inode, &ei))
958 			return;
959 	}
960 	__update_extent_tree_range(dn->inode, &ei, type);
961 }
962 
963 static unsigned int __shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink,
964 					enum extent_type type)
965 {
966 	struct extent_tree_info *eti = &sbi->extent_tree[type];
967 	struct extent_tree *et, *next;
968 	struct extent_node *en;
969 	unsigned int node_cnt = 0, tree_cnt = 0;
970 	int remained;
971 
972 	if (!atomic_read(&eti->total_zombie_tree))
973 		goto free_node;
974 
975 	if (!mutex_trylock(&eti->extent_tree_lock))
976 		goto out;
977 
978 	/* 1. remove unreferenced extent tree */
979 	list_for_each_entry_safe(et, next, &eti->zombie_list, list) {
980 		if (atomic_read(&et->node_cnt)) {
981 			write_lock(&et->lock);
982 			node_cnt += __free_extent_tree(sbi, et);
983 			write_unlock(&et->lock);
984 		}
985 		f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
986 		list_del_init(&et->list);
987 		radix_tree_delete(&eti->extent_tree_root, et->ino);
988 		kmem_cache_free(extent_tree_slab, et);
989 		atomic_dec(&eti->total_ext_tree);
990 		atomic_dec(&eti->total_zombie_tree);
991 		tree_cnt++;
992 
993 		if (node_cnt + tree_cnt >= nr_shrink)
994 			goto unlock_out;
995 		cond_resched();
996 	}
997 	mutex_unlock(&eti->extent_tree_lock);
998 
999 free_node:
1000 	/* 2. remove LRU extent entries */
1001 	if (!mutex_trylock(&eti->extent_tree_lock))
1002 		goto out;
1003 
1004 	remained = nr_shrink - (node_cnt + tree_cnt);
1005 
1006 	spin_lock(&eti->extent_lock);
1007 	for (; remained > 0; remained--) {
1008 		if (list_empty(&eti->extent_list))
1009 			break;
1010 		en = list_first_entry(&eti->extent_list,
1011 					struct extent_node, list);
1012 		et = en->et;
1013 		if (!write_trylock(&et->lock)) {
1014 			/* refresh this extent node's position in extent list */
1015 			list_move_tail(&en->list, &eti->extent_list);
1016 			continue;
1017 		}
1018 
1019 		list_del_init(&en->list);
1020 		spin_unlock(&eti->extent_lock);
1021 
1022 		__detach_extent_node(sbi, et, en);
1023 
1024 		write_unlock(&et->lock);
1025 		node_cnt++;
1026 		spin_lock(&eti->extent_lock);
1027 	}
1028 	spin_unlock(&eti->extent_lock);
1029 
1030 unlock_out:
1031 	mutex_unlock(&eti->extent_tree_lock);
1032 out:
1033 	trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt, type);
1034 
1035 	return node_cnt + tree_cnt;
1036 }
1037 
1038 /* read extent cache operations */
1039 bool f2fs_lookup_read_extent_cache(struct inode *inode, pgoff_t pgofs,
1040 				struct extent_info *ei)
1041 {
1042 	if (!__may_extent_tree(inode, EX_READ))
1043 		return false;
1044 
1045 	return __lookup_extent_tree(inode, pgofs, ei, EX_READ);
1046 }
1047 
1048 void f2fs_update_read_extent_cache(struct dnode_of_data *dn)
1049 {
1050 	return __update_extent_cache(dn, EX_READ);
1051 }
1052 
1053 void f2fs_update_read_extent_cache_range(struct dnode_of_data *dn,
1054 				pgoff_t fofs, block_t blkaddr, unsigned int len)
1055 {
1056 	struct extent_info ei = {
1057 		.fofs = fofs,
1058 		.len = len,
1059 		.blk = blkaddr,
1060 	};
1061 
1062 	if (!__may_extent_tree(dn->inode, EX_READ))
1063 		return;
1064 
1065 	__update_extent_tree_range(dn->inode, &ei, EX_READ);
1066 }
1067 
1068 unsigned int f2fs_shrink_read_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
1069 {
1070 	if (!test_opt(sbi, READ_EXTENT_CACHE))
1071 		return 0;
1072 
1073 	return __shrink_extent_tree(sbi, nr_shrink, EX_READ);
1074 }
1075 
1076 /* block age extent cache operations */
1077 bool f2fs_lookup_age_extent_cache(struct inode *inode, pgoff_t pgofs,
1078 				struct extent_info *ei)
1079 {
1080 	if (!__may_extent_tree(inode, EX_BLOCK_AGE))
1081 		return false;
1082 
1083 	return __lookup_extent_tree(inode, pgofs, ei, EX_BLOCK_AGE);
1084 }
1085 
1086 void f2fs_update_age_extent_cache(struct dnode_of_data *dn)
1087 {
1088 	return __update_extent_cache(dn, EX_BLOCK_AGE);
1089 }
1090 
1091 void f2fs_update_age_extent_cache_range(struct dnode_of_data *dn,
1092 				pgoff_t fofs, unsigned int len)
1093 {
1094 	struct extent_info ei = {
1095 		.fofs = fofs,
1096 		.len = len,
1097 	};
1098 
1099 	if (!__may_extent_tree(dn->inode, EX_BLOCK_AGE))
1100 		return;
1101 
1102 	__update_extent_tree_range(dn->inode, &ei, EX_BLOCK_AGE);
1103 }
1104 
1105 unsigned int f2fs_shrink_age_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
1106 {
1107 	if (!test_opt(sbi, AGE_EXTENT_CACHE))
1108 		return 0;
1109 
1110 	return __shrink_extent_tree(sbi, nr_shrink, EX_BLOCK_AGE);
1111 }
1112 
1113 static unsigned int __destroy_extent_node(struct inode *inode,
1114 					enum extent_type type)
1115 {
1116 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1117 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
1118 	unsigned int node_cnt = 0;
1119 
1120 	if (!et || !atomic_read(&et->node_cnt))
1121 		return 0;
1122 
1123 	write_lock(&et->lock);
1124 	node_cnt = __free_extent_tree(sbi, et);
1125 	write_unlock(&et->lock);
1126 
1127 	return node_cnt;
1128 }
1129 
1130 void f2fs_destroy_extent_node(struct inode *inode)
1131 {
1132 	__destroy_extent_node(inode, EX_READ);
1133 	__destroy_extent_node(inode, EX_BLOCK_AGE);
1134 }
1135 
1136 static void __drop_extent_tree(struct inode *inode, enum extent_type type)
1137 {
1138 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1139 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
1140 	bool updated = false;
1141 
1142 	if (!__may_extent_tree(inode, type))
1143 		return;
1144 
1145 	write_lock(&et->lock);
1146 	__free_extent_tree(sbi, et);
1147 	if (type == EX_READ) {
1148 		set_inode_flag(inode, FI_NO_EXTENT);
1149 		if (et->largest.len) {
1150 			et->largest.len = 0;
1151 			updated = true;
1152 		}
1153 	}
1154 	write_unlock(&et->lock);
1155 	if (updated)
1156 		f2fs_mark_inode_dirty_sync(inode, true);
1157 }
1158 
1159 void f2fs_drop_extent_tree(struct inode *inode)
1160 {
1161 	__drop_extent_tree(inode, EX_READ);
1162 	__drop_extent_tree(inode, EX_BLOCK_AGE);
1163 }
1164 
1165 static void __destroy_extent_tree(struct inode *inode, enum extent_type type)
1166 {
1167 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1168 	struct extent_tree_info *eti = &sbi->extent_tree[type];
1169 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
1170 	unsigned int node_cnt = 0;
1171 
1172 	if (!et)
1173 		return;
1174 
1175 	if (inode->i_nlink && !is_bad_inode(inode) &&
1176 					atomic_read(&et->node_cnt)) {
1177 		mutex_lock(&eti->extent_tree_lock);
1178 		list_add_tail(&et->list, &eti->zombie_list);
1179 		atomic_inc(&eti->total_zombie_tree);
1180 		mutex_unlock(&eti->extent_tree_lock);
1181 		return;
1182 	}
1183 
1184 	/* free all extent info belong to this extent tree */
1185 	node_cnt = __destroy_extent_node(inode, type);
1186 
1187 	/* delete extent tree entry in radix tree */
1188 	mutex_lock(&eti->extent_tree_lock);
1189 	f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
1190 	radix_tree_delete(&eti->extent_tree_root, inode->i_ino);
1191 	kmem_cache_free(extent_tree_slab, et);
1192 	atomic_dec(&eti->total_ext_tree);
1193 	mutex_unlock(&eti->extent_tree_lock);
1194 
1195 	F2FS_I(inode)->extent_tree[type] = NULL;
1196 
1197 	trace_f2fs_destroy_extent_tree(inode, node_cnt, type);
1198 }
1199 
1200 void f2fs_destroy_extent_tree(struct inode *inode)
1201 {
1202 	__destroy_extent_tree(inode, EX_READ);
1203 	__destroy_extent_tree(inode, EX_BLOCK_AGE);
1204 }
1205 
1206 static void __init_extent_tree_info(struct extent_tree_info *eti)
1207 {
1208 	INIT_RADIX_TREE(&eti->extent_tree_root, GFP_NOIO);
1209 	mutex_init(&eti->extent_tree_lock);
1210 	INIT_LIST_HEAD(&eti->extent_list);
1211 	spin_lock_init(&eti->extent_lock);
1212 	atomic_set(&eti->total_ext_tree, 0);
1213 	INIT_LIST_HEAD(&eti->zombie_list);
1214 	atomic_set(&eti->total_zombie_tree, 0);
1215 	atomic_set(&eti->total_ext_node, 0);
1216 }
1217 
1218 void f2fs_init_extent_cache_info(struct f2fs_sb_info *sbi)
1219 {
1220 	__init_extent_tree_info(&sbi->extent_tree[EX_READ]);
1221 	__init_extent_tree_info(&sbi->extent_tree[EX_BLOCK_AGE]);
1222 
1223 	/* initialize for block age extents */
1224 	atomic64_set(&sbi->allocated_data_blocks, 0);
1225 	sbi->hot_data_age_threshold = DEF_HOT_DATA_AGE_THRESHOLD;
1226 	sbi->warm_data_age_threshold = DEF_WARM_DATA_AGE_THRESHOLD;
1227 }
1228 
1229 int __init f2fs_create_extent_cache(void)
1230 {
1231 	extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree",
1232 			sizeof(struct extent_tree));
1233 	if (!extent_tree_slab)
1234 		return -ENOMEM;
1235 	extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node",
1236 			sizeof(struct extent_node));
1237 	if (!extent_node_slab) {
1238 		kmem_cache_destroy(extent_tree_slab);
1239 		return -ENOMEM;
1240 	}
1241 	return 0;
1242 }
1243 
1244 void f2fs_destroy_extent_cache(void)
1245 {
1246 	kmem_cache_destroy(extent_node_slab);
1247 	kmem_cache_destroy(extent_tree_slab);
1248 }
1249