xref: /linux/fs/f2fs/extent_cache.c (revision 9f7861c56b51b84d30114e7fea9d744a9d5ba9b7)
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 bool sanity_check_extent_cache(struct inode *inode)
23 {
24 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
25 	struct f2fs_inode_info *fi = F2FS_I(inode);
26 	struct extent_tree *et = fi->extent_tree[EX_READ];
27 	struct extent_info *ei;
28 
29 	if (!et)
30 		return true;
31 
32 	ei = &et->largest;
33 	if (!ei->len)
34 		return true;
35 
36 	/* Let's drop, if checkpoint got corrupted. */
37 	if (is_set_ckpt_flags(sbi, CP_ERROR_FLAG)) {
38 		ei->len = 0;
39 		et->largest_updated = true;
40 		return true;
41 	}
42 
43 	if (!f2fs_is_valid_blkaddr(sbi, ei->blk, DATA_GENERIC_ENHANCE) ||
44 	    !f2fs_is_valid_blkaddr(sbi, ei->blk + ei->len - 1,
45 					DATA_GENERIC_ENHANCE)) {
46 		set_sbi_flag(sbi, SBI_NEED_FSCK);
47 		f2fs_warn(sbi, "%s: inode (ino=%lx) extent info [%u, %u, %u] is incorrect, run fsck to fix",
48 			  __func__, inode->i_ino,
49 			  ei->blk, ei->fofs, ei->len);
50 		return false;
51 	}
52 	return true;
53 }
54 
55 static void __set_extent_info(struct extent_info *ei,
56 				unsigned int fofs, unsigned int len,
57 				block_t blk, bool keep_clen,
58 				unsigned long age, unsigned long last_blocks,
59 				enum extent_type type)
60 {
61 	ei->fofs = fofs;
62 	ei->len = len;
63 
64 	if (type == EX_READ) {
65 		ei->blk = blk;
66 		if (keep_clen)
67 			return;
68 #ifdef CONFIG_F2FS_FS_COMPRESSION
69 		ei->c_len = 0;
70 #endif
71 	} else if (type == EX_BLOCK_AGE) {
72 		ei->age = age;
73 		ei->last_blocks = last_blocks;
74 	}
75 }
76 
77 static bool __init_may_extent_tree(struct inode *inode, enum extent_type type)
78 {
79 	if (type == EX_READ)
80 		return test_opt(F2FS_I_SB(inode), READ_EXTENT_CACHE) &&
81 			S_ISREG(inode->i_mode);
82 	if (type == EX_BLOCK_AGE)
83 		return test_opt(F2FS_I_SB(inode), AGE_EXTENT_CACHE) &&
84 			(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode));
85 	return false;
86 }
87 
88 static bool __may_extent_tree(struct inode *inode, enum extent_type type)
89 {
90 	/*
91 	 * for recovered files during mount do not create extents
92 	 * if shrinker is not registered.
93 	 */
94 	if (list_empty(&F2FS_I_SB(inode)->s_list))
95 		return false;
96 
97 	if (!__init_may_extent_tree(inode, type))
98 		return false;
99 
100 	if (type == EX_READ) {
101 		if (is_inode_flag_set(inode, FI_NO_EXTENT))
102 			return false;
103 		if (is_inode_flag_set(inode, FI_COMPRESSED_FILE) &&
104 				 !f2fs_sb_has_readonly(F2FS_I_SB(inode)))
105 			return false;
106 	} else if (type == EX_BLOCK_AGE) {
107 		if (is_inode_flag_set(inode, FI_COMPRESSED_FILE))
108 			return false;
109 		if (file_is_cold(inode))
110 			return false;
111 	}
112 	return true;
113 }
114 
115 static void __try_update_largest_extent(struct extent_tree *et,
116 						struct extent_node *en)
117 {
118 	if (et->type != EX_READ)
119 		return;
120 	if (en->ei.len <= et->largest.len)
121 		return;
122 
123 	et->largest = en->ei;
124 	et->largest_updated = true;
125 }
126 
127 static bool __is_extent_mergeable(struct extent_info *back,
128 		struct extent_info *front, enum extent_type type)
129 {
130 	if (type == EX_READ) {
131 #ifdef CONFIG_F2FS_FS_COMPRESSION
132 		if (back->c_len && back->len != back->c_len)
133 			return false;
134 		if (front->c_len && front->len != front->c_len)
135 			return false;
136 #endif
137 		return (back->fofs + back->len == front->fofs &&
138 				back->blk + back->len == front->blk);
139 	} else if (type == EX_BLOCK_AGE) {
140 		return (back->fofs + back->len == front->fofs &&
141 			abs(back->age - front->age) <= SAME_AGE_REGION &&
142 			abs(back->last_blocks - front->last_blocks) <=
143 							SAME_AGE_REGION);
144 	}
145 	return false;
146 }
147 
148 static bool __is_back_mergeable(struct extent_info *cur,
149 		struct extent_info *back, enum extent_type type)
150 {
151 	return __is_extent_mergeable(back, cur, type);
152 }
153 
154 static bool __is_front_mergeable(struct extent_info *cur,
155 		struct extent_info *front, enum extent_type type)
156 {
157 	return __is_extent_mergeable(cur, front, type);
158 }
159 
160 static struct extent_node *__lookup_extent_node(struct rb_root_cached *root,
161 			struct extent_node *cached_en, unsigned int fofs)
162 {
163 	struct rb_node *node = root->rb_root.rb_node;
164 	struct extent_node *en;
165 
166 	/* check a cached entry */
167 	if (cached_en && cached_en->ei.fofs <= fofs &&
168 			cached_en->ei.fofs + cached_en->ei.len > fofs)
169 		return cached_en;
170 
171 	/* check rb_tree */
172 	while (node) {
173 		en = rb_entry(node, struct extent_node, rb_node);
174 
175 		if (fofs < en->ei.fofs)
176 			node = node->rb_left;
177 		else if (fofs >= en->ei.fofs + en->ei.len)
178 			node = node->rb_right;
179 		else
180 			return en;
181 	}
182 	return NULL;
183 }
184 
185 /*
186  * lookup rb entry in position of @fofs in rb-tree,
187  * if hit, return the entry, otherwise, return NULL
188  * @prev_ex: extent before fofs
189  * @next_ex: extent after fofs
190  * @insert_p: insert point for new extent at fofs
191  * in order to simplify the insertion after.
192  * tree must stay unchanged between lookup and insertion.
193  */
194 static struct extent_node *__lookup_extent_node_ret(struct rb_root_cached *root,
195 				struct extent_node *cached_en,
196 				unsigned int fofs,
197 				struct extent_node **prev_entry,
198 				struct extent_node **next_entry,
199 				struct rb_node ***insert_p,
200 				struct rb_node **insert_parent,
201 				bool *leftmost)
202 {
203 	struct rb_node **pnode = &root->rb_root.rb_node;
204 	struct rb_node *parent = NULL, *tmp_node;
205 	struct extent_node *en = cached_en;
206 
207 	*insert_p = NULL;
208 	*insert_parent = NULL;
209 	*prev_entry = NULL;
210 	*next_entry = NULL;
211 
212 	if (RB_EMPTY_ROOT(&root->rb_root))
213 		return NULL;
214 
215 	if (en && en->ei.fofs <= fofs && en->ei.fofs + en->ei.len > fofs)
216 		goto lookup_neighbors;
217 
218 	*leftmost = true;
219 
220 	while (*pnode) {
221 		parent = *pnode;
222 		en = rb_entry(*pnode, struct extent_node, rb_node);
223 
224 		if (fofs < en->ei.fofs) {
225 			pnode = &(*pnode)->rb_left;
226 		} else if (fofs >= en->ei.fofs + en->ei.len) {
227 			pnode = &(*pnode)->rb_right;
228 			*leftmost = false;
229 		} else {
230 			goto lookup_neighbors;
231 		}
232 	}
233 
234 	*insert_p = pnode;
235 	*insert_parent = parent;
236 
237 	en = rb_entry(parent, struct extent_node, rb_node);
238 	tmp_node = parent;
239 	if (parent && fofs > en->ei.fofs)
240 		tmp_node = rb_next(parent);
241 	*next_entry = rb_entry_safe(tmp_node, struct extent_node, rb_node);
242 
243 	tmp_node = parent;
244 	if (parent && fofs < en->ei.fofs)
245 		tmp_node = rb_prev(parent);
246 	*prev_entry = rb_entry_safe(tmp_node, struct extent_node, rb_node);
247 	return NULL;
248 
249 lookup_neighbors:
250 	if (fofs == en->ei.fofs) {
251 		/* lookup prev node for merging backward later */
252 		tmp_node = rb_prev(&en->rb_node);
253 		*prev_entry = rb_entry_safe(tmp_node,
254 					struct extent_node, rb_node);
255 	}
256 	if (fofs == en->ei.fofs + en->ei.len - 1) {
257 		/* lookup next node for merging frontward later */
258 		tmp_node = rb_next(&en->rb_node);
259 		*next_entry = rb_entry_safe(tmp_node,
260 					struct extent_node, rb_node);
261 	}
262 	return en;
263 }
264 
265 static struct kmem_cache *extent_tree_slab;
266 static struct kmem_cache *extent_node_slab;
267 
268 static struct extent_node *__attach_extent_node(struct f2fs_sb_info *sbi,
269 				struct extent_tree *et, struct extent_info *ei,
270 				struct rb_node *parent, struct rb_node **p,
271 				bool leftmost)
272 {
273 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
274 	struct extent_node *en;
275 
276 	en = f2fs_kmem_cache_alloc(extent_node_slab, GFP_ATOMIC, false, sbi);
277 	if (!en)
278 		return NULL;
279 
280 	en->ei = *ei;
281 	INIT_LIST_HEAD(&en->list);
282 	en->et = et;
283 
284 	rb_link_node(&en->rb_node, parent, p);
285 	rb_insert_color_cached(&en->rb_node, &et->root, leftmost);
286 	atomic_inc(&et->node_cnt);
287 	atomic_inc(&eti->total_ext_node);
288 	return en;
289 }
290 
291 static void __detach_extent_node(struct f2fs_sb_info *sbi,
292 				struct extent_tree *et, struct extent_node *en)
293 {
294 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
295 
296 	rb_erase_cached(&en->rb_node, &et->root);
297 	atomic_dec(&et->node_cnt);
298 	atomic_dec(&eti->total_ext_node);
299 
300 	if (et->cached_en == en)
301 		et->cached_en = NULL;
302 	kmem_cache_free(extent_node_slab, en);
303 }
304 
305 /*
306  * Flow to release an extent_node:
307  * 1. list_del_init
308  * 2. __detach_extent_node
309  * 3. kmem_cache_free.
310  */
311 static void __release_extent_node(struct f2fs_sb_info *sbi,
312 			struct extent_tree *et, struct extent_node *en)
313 {
314 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
315 
316 	spin_lock(&eti->extent_lock);
317 	f2fs_bug_on(sbi, list_empty(&en->list));
318 	list_del_init(&en->list);
319 	spin_unlock(&eti->extent_lock);
320 
321 	__detach_extent_node(sbi, et, en);
322 }
323 
324 static struct extent_tree *__grab_extent_tree(struct inode *inode,
325 						enum extent_type type)
326 {
327 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
328 	struct extent_tree_info *eti = &sbi->extent_tree[type];
329 	struct extent_tree *et;
330 	nid_t ino = inode->i_ino;
331 
332 	mutex_lock(&eti->extent_tree_lock);
333 	et = radix_tree_lookup(&eti->extent_tree_root, ino);
334 	if (!et) {
335 		et = f2fs_kmem_cache_alloc(extent_tree_slab,
336 					GFP_NOFS, true, NULL);
337 		f2fs_radix_tree_insert(&eti->extent_tree_root, ino, et);
338 		memset(et, 0, sizeof(struct extent_tree));
339 		et->ino = ino;
340 		et->type = type;
341 		et->root = RB_ROOT_CACHED;
342 		et->cached_en = NULL;
343 		rwlock_init(&et->lock);
344 		INIT_LIST_HEAD(&et->list);
345 		atomic_set(&et->node_cnt, 0);
346 		atomic_inc(&eti->total_ext_tree);
347 	} else {
348 		atomic_dec(&eti->total_zombie_tree);
349 		list_del_init(&et->list);
350 	}
351 	mutex_unlock(&eti->extent_tree_lock);
352 
353 	/* never died until evict_inode */
354 	F2FS_I(inode)->extent_tree[type] = et;
355 
356 	return et;
357 }
358 
359 static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi,
360 					struct extent_tree *et)
361 {
362 	struct rb_node *node, *next;
363 	struct extent_node *en;
364 	unsigned int count = atomic_read(&et->node_cnt);
365 
366 	node = rb_first_cached(&et->root);
367 	while (node) {
368 		next = rb_next(node);
369 		en = rb_entry(node, struct extent_node, rb_node);
370 		__release_extent_node(sbi, et, en);
371 		node = next;
372 	}
373 
374 	return count - atomic_read(&et->node_cnt);
375 }
376 
377 static void __drop_largest_extent(struct extent_tree *et,
378 					pgoff_t fofs, unsigned int len)
379 {
380 	if (fofs < et->largest.fofs + et->largest.len &&
381 			fofs + len > et->largest.fofs) {
382 		et->largest.len = 0;
383 		et->largest_updated = true;
384 	}
385 }
386 
387 void f2fs_init_read_extent_tree(struct inode *inode, struct page *ipage)
388 {
389 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
390 	struct extent_tree_info *eti = &sbi->extent_tree[EX_READ];
391 	struct f2fs_extent *i_ext = &F2FS_INODE(ipage)->i_ext;
392 	struct extent_tree *et;
393 	struct extent_node *en;
394 	struct extent_info ei;
395 
396 	if (!__may_extent_tree(inode, EX_READ)) {
397 		/* drop largest read extent */
398 		if (i_ext && i_ext->len) {
399 			f2fs_wait_on_page_writeback(ipage, NODE, true, true);
400 			i_ext->len = 0;
401 			set_page_dirty(ipage);
402 		}
403 		goto out;
404 	}
405 
406 	et = __grab_extent_tree(inode, EX_READ);
407 
408 	if (!i_ext || !i_ext->len)
409 		goto out;
410 
411 	get_read_extent_info(&ei, i_ext);
412 
413 	write_lock(&et->lock);
414 	if (atomic_read(&et->node_cnt))
415 		goto unlock_out;
416 
417 	en = __attach_extent_node(sbi, et, &ei, NULL,
418 				&et->root.rb_root.rb_node, true);
419 	if (en) {
420 		et->largest = en->ei;
421 		et->cached_en = en;
422 
423 		spin_lock(&eti->extent_lock);
424 		list_add_tail(&en->list, &eti->extent_list);
425 		spin_unlock(&eti->extent_lock);
426 	}
427 unlock_out:
428 	write_unlock(&et->lock);
429 out:
430 	if (!F2FS_I(inode)->extent_tree[EX_READ])
431 		set_inode_flag(inode, FI_NO_EXTENT);
432 }
433 
434 void f2fs_init_age_extent_tree(struct inode *inode)
435 {
436 	if (!__init_may_extent_tree(inode, EX_BLOCK_AGE))
437 		return;
438 	__grab_extent_tree(inode, EX_BLOCK_AGE);
439 }
440 
441 void f2fs_init_extent_tree(struct inode *inode)
442 {
443 	/* initialize read cache */
444 	if (__init_may_extent_tree(inode, EX_READ))
445 		__grab_extent_tree(inode, EX_READ);
446 
447 	/* initialize block age cache */
448 	if (__init_may_extent_tree(inode, EX_BLOCK_AGE))
449 		__grab_extent_tree(inode, EX_BLOCK_AGE);
450 }
451 
452 static bool __lookup_extent_tree(struct inode *inode, pgoff_t pgofs,
453 			struct extent_info *ei, enum extent_type type)
454 {
455 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
456 	struct extent_tree_info *eti = &sbi->extent_tree[type];
457 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
458 	struct extent_node *en;
459 	bool ret = false;
460 
461 	if (!et)
462 		return false;
463 
464 	trace_f2fs_lookup_extent_tree_start(inode, pgofs, type);
465 
466 	read_lock(&et->lock);
467 
468 	if (type == EX_READ &&
469 			et->largest.fofs <= pgofs &&
470 			et->largest.fofs + et->largest.len > pgofs) {
471 		*ei = et->largest;
472 		ret = true;
473 		stat_inc_largest_node_hit(sbi);
474 		goto out;
475 	}
476 
477 	en = __lookup_extent_node(&et->root, et->cached_en, pgofs);
478 	if (!en)
479 		goto out;
480 
481 	if (en == et->cached_en)
482 		stat_inc_cached_node_hit(sbi, type);
483 	else
484 		stat_inc_rbtree_node_hit(sbi, type);
485 
486 	*ei = en->ei;
487 	spin_lock(&eti->extent_lock);
488 	if (!list_empty(&en->list)) {
489 		list_move_tail(&en->list, &eti->extent_list);
490 		et->cached_en = en;
491 	}
492 	spin_unlock(&eti->extent_lock);
493 	ret = true;
494 out:
495 	stat_inc_total_hit(sbi, type);
496 	read_unlock(&et->lock);
497 
498 	if (type == EX_READ)
499 		trace_f2fs_lookup_read_extent_tree_end(inode, pgofs, ei);
500 	else if (type == EX_BLOCK_AGE)
501 		trace_f2fs_lookup_age_extent_tree_end(inode, pgofs, ei);
502 	return ret;
503 }
504 
505 static struct extent_node *__try_merge_extent_node(struct f2fs_sb_info *sbi,
506 				struct extent_tree *et, struct extent_info *ei,
507 				struct extent_node *prev_ex,
508 				struct extent_node *next_ex)
509 {
510 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
511 	struct extent_node *en = NULL;
512 
513 	if (prev_ex && __is_back_mergeable(ei, &prev_ex->ei, et->type)) {
514 		prev_ex->ei.len += ei->len;
515 		ei = &prev_ex->ei;
516 		en = prev_ex;
517 	}
518 
519 	if (next_ex && __is_front_mergeable(ei, &next_ex->ei, et->type)) {
520 		next_ex->ei.fofs = ei->fofs;
521 		next_ex->ei.len += ei->len;
522 		if (et->type == EX_READ)
523 			next_ex->ei.blk = ei->blk;
524 		if (en)
525 			__release_extent_node(sbi, et, prev_ex);
526 
527 		en = next_ex;
528 	}
529 
530 	if (!en)
531 		return NULL;
532 
533 	__try_update_largest_extent(et, en);
534 
535 	spin_lock(&eti->extent_lock);
536 	if (!list_empty(&en->list)) {
537 		list_move_tail(&en->list, &eti->extent_list);
538 		et->cached_en = en;
539 	}
540 	spin_unlock(&eti->extent_lock);
541 	return en;
542 }
543 
544 static struct extent_node *__insert_extent_tree(struct f2fs_sb_info *sbi,
545 				struct extent_tree *et, struct extent_info *ei,
546 				struct rb_node **insert_p,
547 				struct rb_node *insert_parent,
548 				bool leftmost)
549 {
550 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
551 	struct rb_node **p = &et->root.rb_root.rb_node;
552 	struct rb_node *parent = NULL;
553 	struct extent_node *en = NULL;
554 
555 	if (insert_p && insert_parent) {
556 		parent = insert_parent;
557 		p = insert_p;
558 		goto do_insert;
559 	}
560 
561 	leftmost = true;
562 
563 	/* look up extent_node in the rb tree */
564 	while (*p) {
565 		parent = *p;
566 		en = rb_entry(parent, struct extent_node, rb_node);
567 
568 		if (ei->fofs < en->ei.fofs) {
569 			p = &(*p)->rb_left;
570 		} else if (ei->fofs >= en->ei.fofs + en->ei.len) {
571 			p = &(*p)->rb_right;
572 			leftmost = false;
573 		} else {
574 			f2fs_bug_on(sbi, 1);
575 		}
576 	}
577 
578 do_insert:
579 	en = __attach_extent_node(sbi, et, ei, parent, p, leftmost);
580 	if (!en)
581 		return NULL;
582 
583 	__try_update_largest_extent(et, en);
584 
585 	/* update in global extent list */
586 	spin_lock(&eti->extent_lock);
587 	list_add_tail(&en->list, &eti->extent_list);
588 	et->cached_en = en;
589 	spin_unlock(&eti->extent_lock);
590 	return en;
591 }
592 
593 static void __update_extent_tree_range(struct inode *inode,
594 			struct extent_info *tei, enum extent_type type)
595 {
596 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
597 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
598 	struct extent_node *en = NULL, *en1 = NULL;
599 	struct extent_node *prev_en = NULL, *next_en = NULL;
600 	struct extent_info ei, dei, prev;
601 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
602 	unsigned int fofs = tei->fofs, len = tei->len;
603 	unsigned int end = fofs + len;
604 	bool updated = false;
605 	bool leftmost = false;
606 
607 	if (!et)
608 		return;
609 
610 	if (type == EX_READ)
611 		trace_f2fs_update_read_extent_tree_range(inode, fofs, len,
612 						tei->blk, 0);
613 	else if (type == EX_BLOCK_AGE)
614 		trace_f2fs_update_age_extent_tree_range(inode, fofs, len,
615 						tei->age, tei->last_blocks);
616 
617 	write_lock(&et->lock);
618 
619 	if (type == EX_READ) {
620 		if (is_inode_flag_set(inode, FI_NO_EXTENT)) {
621 			write_unlock(&et->lock);
622 			return;
623 		}
624 
625 		prev = et->largest;
626 		dei.len = 0;
627 
628 		/*
629 		 * drop largest extent before lookup, in case it's already
630 		 * been shrunk from extent tree
631 		 */
632 		__drop_largest_extent(et, fofs, len);
633 	}
634 
635 	/* 1. lookup first extent node in range [fofs, fofs + len - 1] */
636 	en = __lookup_extent_node_ret(&et->root,
637 					et->cached_en, fofs,
638 					&prev_en, &next_en,
639 					&insert_p, &insert_parent,
640 					&leftmost);
641 	if (!en)
642 		en = next_en;
643 
644 	/* 2. invalidate all extent nodes in range [fofs, fofs + len - 1] */
645 	while (en && en->ei.fofs < end) {
646 		unsigned int org_end;
647 		int parts = 0;	/* # of parts current extent split into */
648 
649 		next_en = en1 = NULL;
650 
651 		dei = en->ei;
652 		org_end = dei.fofs + dei.len;
653 		f2fs_bug_on(sbi, fofs >= org_end);
654 
655 		if (fofs > dei.fofs && (type != EX_READ ||
656 				fofs - dei.fofs >= F2FS_MIN_EXTENT_LEN)) {
657 			en->ei.len = fofs - en->ei.fofs;
658 			prev_en = en;
659 			parts = 1;
660 		}
661 
662 		if (end < org_end && (type != EX_READ ||
663 				org_end - end >= F2FS_MIN_EXTENT_LEN)) {
664 			if (parts) {
665 				__set_extent_info(&ei,
666 					end, org_end - end,
667 					end - dei.fofs + dei.blk, false,
668 					dei.age, dei.last_blocks,
669 					type);
670 				en1 = __insert_extent_tree(sbi, et, &ei,
671 							NULL, NULL, true);
672 				next_en = en1;
673 			} else {
674 				__set_extent_info(&en->ei,
675 					end, en->ei.len - (end - dei.fofs),
676 					en->ei.blk + (end - dei.fofs), true,
677 					dei.age, dei.last_blocks,
678 					type);
679 				next_en = en;
680 			}
681 			parts++;
682 		}
683 
684 		if (!next_en) {
685 			struct rb_node *node = rb_next(&en->rb_node);
686 
687 			next_en = rb_entry_safe(node, struct extent_node,
688 						rb_node);
689 		}
690 
691 		if (parts)
692 			__try_update_largest_extent(et, en);
693 		else
694 			__release_extent_node(sbi, et, en);
695 
696 		/*
697 		 * if original extent is split into zero or two parts, extent
698 		 * tree has been altered by deletion or insertion, therefore
699 		 * invalidate pointers regard to tree.
700 		 */
701 		if (parts != 1) {
702 			insert_p = NULL;
703 			insert_parent = NULL;
704 		}
705 		en = next_en;
706 	}
707 
708 	if (type == EX_BLOCK_AGE)
709 		goto update_age_extent_cache;
710 
711 	/* 3. update extent in read extent cache */
712 	BUG_ON(type != EX_READ);
713 
714 	if (tei->blk) {
715 		__set_extent_info(&ei, fofs, len, tei->blk, false,
716 				  0, 0, EX_READ);
717 		if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
718 			__insert_extent_tree(sbi, et, &ei,
719 					insert_p, insert_parent, leftmost);
720 
721 		/* give up extent_cache, if split and small updates happen */
722 		if (dei.len >= 1 &&
723 				prev.len < F2FS_MIN_EXTENT_LEN &&
724 				et->largest.len < F2FS_MIN_EXTENT_LEN) {
725 			et->largest.len = 0;
726 			et->largest_updated = true;
727 			set_inode_flag(inode, FI_NO_EXTENT);
728 		}
729 	}
730 
731 	if (is_inode_flag_set(inode, FI_NO_EXTENT))
732 		__free_extent_tree(sbi, et);
733 
734 	if (et->largest_updated) {
735 		et->largest_updated = false;
736 		updated = true;
737 	}
738 	goto out_read_extent_cache;
739 update_age_extent_cache:
740 	if (!tei->last_blocks)
741 		goto out_read_extent_cache;
742 
743 	__set_extent_info(&ei, fofs, len, 0, false,
744 			tei->age, tei->last_blocks, EX_BLOCK_AGE);
745 	if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
746 		__insert_extent_tree(sbi, et, &ei,
747 					insert_p, insert_parent, leftmost);
748 out_read_extent_cache:
749 	write_unlock(&et->lock);
750 
751 	if (updated)
752 		f2fs_mark_inode_dirty_sync(inode, true);
753 }
754 
755 #ifdef CONFIG_F2FS_FS_COMPRESSION
756 void f2fs_update_read_extent_tree_range_compressed(struct inode *inode,
757 				pgoff_t fofs, block_t blkaddr, unsigned int llen,
758 				unsigned int c_len)
759 {
760 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
761 	struct extent_tree *et = F2FS_I(inode)->extent_tree[EX_READ];
762 	struct extent_node *en = NULL;
763 	struct extent_node *prev_en = NULL, *next_en = NULL;
764 	struct extent_info ei;
765 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
766 	bool leftmost = false;
767 
768 	trace_f2fs_update_read_extent_tree_range(inode, fofs, llen,
769 						blkaddr, c_len);
770 
771 	/* it is safe here to check FI_NO_EXTENT w/o et->lock in ro image */
772 	if (is_inode_flag_set(inode, FI_NO_EXTENT))
773 		return;
774 
775 	write_lock(&et->lock);
776 
777 	en = __lookup_extent_node_ret(&et->root,
778 					et->cached_en, fofs,
779 					&prev_en, &next_en,
780 					&insert_p, &insert_parent,
781 					&leftmost);
782 	if (en)
783 		goto unlock_out;
784 
785 	__set_extent_info(&ei, fofs, llen, blkaddr, true, 0, 0, EX_READ);
786 	ei.c_len = c_len;
787 
788 	if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
789 		__insert_extent_tree(sbi, et, &ei,
790 				insert_p, insert_parent, leftmost);
791 unlock_out:
792 	write_unlock(&et->lock);
793 }
794 #endif
795 
796 static unsigned long long __calculate_block_age(struct f2fs_sb_info *sbi,
797 						unsigned long long new,
798 						unsigned long long old)
799 {
800 	unsigned int rem_old, rem_new;
801 	unsigned long long res;
802 	unsigned int weight = sbi->last_age_weight;
803 
804 	res = div_u64_rem(new, 100, &rem_new) * (100 - weight)
805 		+ div_u64_rem(old, 100, &rem_old) * weight;
806 
807 	if (rem_new)
808 		res += rem_new * (100 - weight) / 100;
809 	if (rem_old)
810 		res += rem_old * weight / 100;
811 
812 	return res;
813 }
814 
815 /* This returns a new age and allocated blocks in ei */
816 static int __get_new_block_age(struct inode *inode, struct extent_info *ei,
817 						block_t blkaddr)
818 {
819 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
820 	loff_t f_size = i_size_read(inode);
821 	unsigned long long cur_blocks =
822 				atomic64_read(&sbi->allocated_data_blocks);
823 	struct extent_info tei = *ei;	/* only fofs and len are valid */
824 
825 	/*
826 	 * When I/O is not aligned to a PAGE_SIZE, update will happen to the last
827 	 * file block even in seq write. So don't record age for newly last file
828 	 * block here.
829 	 */
830 	if ((f_size >> PAGE_SHIFT) == ei->fofs && f_size & (PAGE_SIZE - 1) &&
831 			blkaddr == NEW_ADDR)
832 		return -EINVAL;
833 
834 	if (__lookup_extent_tree(inode, ei->fofs, &tei, EX_BLOCK_AGE)) {
835 		unsigned long long cur_age;
836 
837 		if (cur_blocks >= tei.last_blocks)
838 			cur_age = cur_blocks - tei.last_blocks;
839 		else
840 			/* allocated_data_blocks overflow */
841 			cur_age = ULLONG_MAX - tei.last_blocks + cur_blocks;
842 
843 		if (tei.age)
844 			ei->age = __calculate_block_age(sbi, cur_age, tei.age);
845 		else
846 			ei->age = cur_age;
847 		ei->last_blocks = cur_blocks;
848 		WARN_ON(ei->age > cur_blocks);
849 		return 0;
850 	}
851 
852 	f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
853 
854 	/* the data block was allocated for the first time */
855 	if (blkaddr == NEW_ADDR)
856 		goto out;
857 
858 	if (__is_valid_data_blkaddr(blkaddr) &&
859 	    !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
860 		f2fs_bug_on(sbi, 1);
861 		return -EINVAL;
862 	}
863 out:
864 	/*
865 	 * init block age with zero, this can happen when the block age extent
866 	 * was reclaimed due to memory constraint or system reboot
867 	 */
868 	ei->age = 0;
869 	ei->last_blocks = cur_blocks;
870 	return 0;
871 }
872 
873 static void __update_extent_cache(struct dnode_of_data *dn, enum extent_type type)
874 {
875 	struct extent_info ei = {};
876 
877 	if (!__may_extent_tree(dn->inode, type))
878 		return;
879 
880 	ei.fofs = f2fs_start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
881 								dn->ofs_in_node;
882 	ei.len = 1;
883 
884 	if (type == EX_READ) {
885 		if (dn->data_blkaddr == NEW_ADDR)
886 			ei.blk = NULL_ADDR;
887 		else
888 			ei.blk = dn->data_blkaddr;
889 	} else if (type == EX_BLOCK_AGE) {
890 		if (__get_new_block_age(dn->inode, &ei, dn->data_blkaddr))
891 			return;
892 	}
893 	__update_extent_tree_range(dn->inode, &ei, type);
894 }
895 
896 static unsigned int __shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink,
897 					enum extent_type type)
898 {
899 	struct extent_tree_info *eti = &sbi->extent_tree[type];
900 	struct extent_tree *et, *next;
901 	struct extent_node *en;
902 	unsigned int node_cnt = 0, tree_cnt = 0;
903 	int remained;
904 
905 	if (!atomic_read(&eti->total_zombie_tree))
906 		goto free_node;
907 
908 	if (!mutex_trylock(&eti->extent_tree_lock))
909 		goto out;
910 
911 	/* 1. remove unreferenced extent tree */
912 	list_for_each_entry_safe(et, next, &eti->zombie_list, list) {
913 		if (atomic_read(&et->node_cnt)) {
914 			write_lock(&et->lock);
915 			node_cnt += __free_extent_tree(sbi, et);
916 			write_unlock(&et->lock);
917 		}
918 		f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
919 		list_del_init(&et->list);
920 		radix_tree_delete(&eti->extent_tree_root, et->ino);
921 		kmem_cache_free(extent_tree_slab, et);
922 		atomic_dec(&eti->total_ext_tree);
923 		atomic_dec(&eti->total_zombie_tree);
924 		tree_cnt++;
925 
926 		if (node_cnt + tree_cnt >= nr_shrink)
927 			goto unlock_out;
928 		cond_resched();
929 	}
930 	mutex_unlock(&eti->extent_tree_lock);
931 
932 free_node:
933 	/* 2. remove LRU extent entries */
934 	if (!mutex_trylock(&eti->extent_tree_lock))
935 		goto out;
936 
937 	remained = nr_shrink - (node_cnt + tree_cnt);
938 
939 	spin_lock(&eti->extent_lock);
940 	for (; remained > 0; remained--) {
941 		if (list_empty(&eti->extent_list))
942 			break;
943 		en = list_first_entry(&eti->extent_list,
944 					struct extent_node, list);
945 		et = en->et;
946 		if (!write_trylock(&et->lock)) {
947 			/* refresh this extent node's position in extent list */
948 			list_move_tail(&en->list, &eti->extent_list);
949 			continue;
950 		}
951 
952 		list_del_init(&en->list);
953 		spin_unlock(&eti->extent_lock);
954 
955 		__detach_extent_node(sbi, et, en);
956 
957 		write_unlock(&et->lock);
958 		node_cnt++;
959 		spin_lock(&eti->extent_lock);
960 	}
961 	spin_unlock(&eti->extent_lock);
962 
963 unlock_out:
964 	mutex_unlock(&eti->extent_tree_lock);
965 out:
966 	trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt, type);
967 
968 	return node_cnt + tree_cnt;
969 }
970 
971 /* read extent cache operations */
972 bool f2fs_lookup_read_extent_cache(struct inode *inode, pgoff_t pgofs,
973 				struct extent_info *ei)
974 {
975 	if (!__may_extent_tree(inode, EX_READ))
976 		return false;
977 
978 	return __lookup_extent_tree(inode, pgofs, ei, EX_READ);
979 }
980 
981 bool f2fs_lookup_read_extent_cache_block(struct inode *inode, pgoff_t index,
982 				block_t *blkaddr)
983 {
984 	struct extent_info ei = {};
985 
986 	if (!f2fs_lookup_read_extent_cache(inode, index, &ei))
987 		return false;
988 	*blkaddr = ei.blk + index - ei.fofs;
989 	return true;
990 }
991 
992 void f2fs_update_read_extent_cache(struct dnode_of_data *dn)
993 {
994 	return __update_extent_cache(dn, EX_READ);
995 }
996 
997 void f2fs_update_read_extent_cache_range(struct dnode_of_data *dn,
998 				pgoff_t fofs, block_t blkaddr, unsigned int len)
999 {
1000 	struct extent_info ei = {
1001 		.fofs = fofs,
1002 		.len = len,
1003 		.blk = blkaddr,
1004 	};
1005 
1006 	if (!__may_extent_tree(dn->inode, EX_READ))
1007 		return;
1008 
1009 	__update_extent_tree_range(dn->inode, &ei, EX_READ);
1010 }
1011 
1012 unsigned int f2fs_shrink_read_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
1013 {
1014 	if (!test_opt(sbi, READ_EXTENT_CACHE))
1015 		return 0;
1016 
1017 	return __shrink_extent_tree(sbi, nr_shrink, EX_READ);
1018 }
1019 
1020 /* block age extent cache operations */
1021 bool f2fs_lookup_age_extent_cache(struct inode *inode, pgoff_t pgofs,
1022 				struct extent_info *ei)
1023 {
1024 	if (!__may_extent_tree(inode, EX_BLOCK_AGE))
1025 		return false;
1026 
1027 	return __lookup_extent_tree(inode, pgofs, ei, EX_BLOCK_AGE);
1028 }
1029 
1030 void f2fs_update_age_extent_cache(struct dnode_of_data *dn)
1031 {
1032 	return __update_extent_cache(dn, EX_BLOCK_AGE);
1033 }
1034 
1035 void f2fs_update_age_extent_cache_range(struct dnode_of_data *dn,
1036 				pgoff_t fofs, unsigned int len)
1037 {
1038 	struct extent_info ei = {
1039 		.fofs = fofs,
1040 		.len = len,
1041 	};
1042 
1043 	if (!__may_extent_tree(dn->inode, EX_BLOCK_AGE))
1044 		return;
1045 
1046 	__update_extent_tree_range(dn->inode, &ei, EX_BLOCK_AGE);
1047 }
1048 
1049 unsigned int f2fs_shrink_age_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
1050 {
1051 	if (!test_opt(sbi, AGE_EXTENT_CACHE))
1052 		return 0;
1053 
1054 	return __shrink_extent_tree(sbi, nr_shrink, EX_BLOCK_AGE);
1055 }
1056 
1057 static unsigned int __destroy_extent_node(struct inode *inode,
1058 					enum extent_type type)
1059 {
1060 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1061 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
1062 	unsigned int node_cnt = 0;
1063 
1064 	if (!et || !atomic_read(&et->node_cnt))
1065 		return 0;
1066 
1067 	write_lock(&et->lock);
1068 	node_cnt = __free_extent_tree(sbi, et);
1069 	write_unlock(&et->lock);
1070 
1071 	return node_cnt;
1072 }
1073 
1074 void f2fs_destroy_extent_node(struct inode *inode)
1075 {
1076 	__destroy_extent_node(inode, EX_READ);
1077 	__destroy_extent_node(inode, EX_BLOCK_AGE);
1078 }
1079 
1080 static void __drop_extent_tree(struct inode *inode, enum extent_type type)
1081 {
1082 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1083 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
1084 	bool updated = false;
1085 
1086 	if (!__may_extent_tree(inode, type))
1087 		return;
1088 
1089 	write_lock(&et->lock);
1090 	__free_extent_tree(sbi, et);
1091 	if (type == EX_READ) {
1092 		set_inode_flag(inode, FI_NO_EXTENT);
1093 		if (et->largest.len) {
1094 			et->largest.len = 0;
1095 			updated = true;
1096 		}
1097 	}
1098 	write_unlock(&et->lock);
1099 	if (updated)
1100 		f2fs_mark_inode_dirty_sync(inode, true);
1101 }
1102 
1103 void f2fs_drop_extent_tree(struct inode *inode)
1104 {
1105 	__drop_extent_tree(inode, EX_READ);
1106 	__drop_extent_tree(inode, EX_BLOCK_AGE);
1107 }
1108 
1109 static void __destroy_extent_tree(struct inode *inode, enum extent_type type)
1110 {
1111 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1112 	struct extent_tree_info *eti = &sbi->extent_tree[type];
1113 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
1114 	unsigned int node_cnt = 0;
1115 
1116 	if (!et)
1117 		return;
1118 
1119 	if (inode->i_nlink && !is_bad_inode(inode) &&
1120 					atomic_read(&et->node_cnt)) {
1121 		mutex_lock(&eti->extent_tree_lock);
1122 		list_add_tail(&et->list, &eti->zombie_list);
1123 		atomic_inc(&eti->total_zombie_tree);
1124 		mutex_unlock(&eti->extent_tree_lock);
1125 		return;
1126 	}
1127 
1128 	/* free all extent info belong to this extent tree */
1129 	node_cnt = __destroy_extent_node(inode, type);
1130 
1131 	/* delete extent tree entry in radix tree */
1132 	mutex_lock(&eti->extent_tree_lock);
1133 	f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
1134 	radix_tree_delete(&eti->extent_tree_root, inode->i_ino);
1135 	kmem_cache_free(extent_tree_slab, et);
1136 	atomic_dec(&eti->total_ext_tree);
1137 	mutex_unlock(&eti->extent_tree_lock);
1138 
1139 	F2FS_I(inode)->extent_tree[type] = NULL;
1140 
1141 	trace_f2fs_destroy_extent_tree(inode, node_cnt, type);
1142 }
1143 
1144 void f2fs_destroy_extent_tree(struct inode *inode)
1145 {
1146 	__destroy_extent_tree(inode, EX_READ);
1147 	__destroy_extent_tree(inode, EX_BLOCK_AGE);
1148 }
1149 
1150 static void __init_extent_tree_info(struct extent_tree_info *eti)
1151 {
1152 	INIT_RADIX_TREE(&eti->extent_tree_root, GFP_NOIO);
1153 	mutex_init(&eti->extent_tree_lock);
1154 	INIT_LIST_HEAD(&eti->extent_list);
1155 	spin_lock_init(&eti->extent_lock);
1156 	atomic_set(&eti->total_ext_tree, 0);
1157 	INIT_LIST_HEAD(&eti->zombie_list);
1158 	atomic_set(&eti->total_zombie_tree, 0);
1159 	atomic_set(&eti->total_ext_node, 0);
1160 }
1161 
1162 void f2fs_init_extent_cache_info(struct f2fs_sb_info *sbi)
1163 {
1164 	__init_extent_tree_info(&sbi->extent_tree[EX_READ]);
1165 	__init_extent_tree_info(&sbi->extent_tree[EX_BLOCK_AGE]);
1166 
1167 	/* initialize for block age extents */
1168 	atomic64_set(&sbi->allocated_data_blocks, 0);
1169 	sbi->hot_data_age_threshold = DEF_HOT_DATA_AGE_THRESHOLD;
1170 	sbi->warm_data_age_threshold = DEF_WARM_DATA_AGE_THRESHOLD;
1171 	sbi->last_age_weight = LAST_AGE_WEIGHT;
1172 }
1173 
1174 int __init f2fs_create_extent_cache(void)
1175 {
1176 	extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree",
1177 			sizeof(struct extent_tree));
1178 	if (!extent_tree_slab)
1179 		return -ENOMEM;
1180 	extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node",
1181 			sizeof(struct extent_node));
1182 	if (!extent_node_slab) {
1183 		kmem_cache_destroy(extent_tree_slab);
1184 		return -ENOMEM;
1185 	}
1186 	return 0;
1187 }
1188 
1189 void f2fs_destroy_extent_cache(void)
1190 {
1191 	kmem_cache_destroy(extent_node_slab);
1192 	kmem_cache_destroy(extent_tree_slab);
1193 }
1194