xref: /openbmc/linux/fs/f2fs/extent_cache.c (revision 68f436a8)
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 __may_read_extent_tree(struct inode *inode)
78 {
79 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
80 
81 	if (!test_opt(sbi, READ_EXTENT_CACHE))
82 		return false;
83 	if (is_inode_flag_set(inode, FI_NO_EXTENT))
84 		return false;
85 	if (is_inode_flag_set(inode, FI_COMPRESSED_FILE) &&
86 			 !f2fs_sb_has_readonly(sbi))
87 		return false;
88 	return S_ISREG(inode->i_mode);
89 }
90 
91 static bool __may_age_extent_tree(struct inode *inode)
92 {
93 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
94 
95 	if (!test_opt(sbi, AGE_EXTENT_CACHE))
96 		return false;
97 	if (is_inode_flag_set(inode, FI_COMPRESSED_FILE))
98 		return false;
99 	if (file_is_cold(inode))
100 		return false;
101 
102 	return S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode);
103 }
104 
105 static bool __init_may_extent_tree(struct inode *inode, enum extent_type type)
106 {
107 	if (type == EX_READ)
108 		return __may_read_extent_tree(inode);
109 	else if (type == EX_BLOCK_AGE)
110 		return __may_age_extent_tree(inode);
111 	return false;
112 }
113 
114 static bool __may_extent_tree(struct inode *inode, enum extent_type type)
115 {
116 	/*
117 	 * for recovered files during mount do not create extents
118 	 * if shrinker is not registered.
119 	 */
120 	if (list_empty(&F2FS_I_SB(inode)->s_list))
121 		return false;
122 
123 	return __init_may_extent_tree(inode, type);
124 }
125 
126 static void __try_update_largest_extent(struct extent_tree *et,
127 						struct extent_node *en)
128 {
129 	if (et->type != EX_READ)
130 		return;
131 	if (en->ei.len <= et->largest.len)
132 		return;
133 
134 	et->largest = en->ei;
135 	et->largest_updated = true;
136 }
137 
138 static bool __is_extent_mergeable(struct extent_info *back,
139 		struct extent_info *front, enum extent_type type)
140 {
141 	if (type == EX_READ) {
142 #ifdef CONFIG_F2FS_FS_COMPRESSION
143 		if (back->c_len && back->len != back->c_len)
144 			return false;
145 		if (front->c_len && front->len != front->c_len)
146 			return false;
147 #endif
148 		return (back->fofs + back->len == front->fofs &&
149 				back->blk + back->len == front->blk);
150 	} else if (type == EX_BLOCK_AGE) {
151 		return (back->fofs + back->len == front->fofs &&
152 			abs(back->age - front->age) <= SAME_AGE_REGION &&
153 			abs(back->last_blocks - front->last_blocks) <=
154 							SAME_AGE_REGION);
155 	}
156 	return false;
157 }
158 
159 static bool __is_back_mergeable(struct extent_info *cur,
160 		struct extent_info *back, enum extent_type type)
161 {
162 	return __is_extent_mergeable(back, cur, type);
163 }
164 
165 static bool __is_front_mergeable(struct extent_info *cur,
166 		struct extent_info *front, enum extent_type type)
167 {
168 	return __is_extent_mergeable(cur, front, type);
169 }
170 
171 static struct extent_node *__lookup_extent_node(struct rb_root_cached *root,
172 			struct extent_node *cached_en, unsigned int fofs)
173 {
174 	struct rb_node *node = root->rb_root.rb_node;
175 	struct extent_node *en;
176 
177 	/* check a cached entry */
178 	if (cached_en && cached_en->ei.fofs <= fofs &&
179 			cached_en->ei.fofs + cached_en->ei.len > fofs)
180 		return cached_en;
181 
182 	/* check rb_tree */
183 	while (node) {
184 		en = rb_entry(node, struct extent_node, rb_node);
185 
186 		if (fofs < en->ei.fofs)
187 			node = node->rb_left;
188 		else if (fofs >= en->ei.fofs + en->ei.len)
189 			node = node->rb_right;
190 		else
191 			return en;
192 	}
193 	return NULL;
194 }
195 
196 /*
197  * lookup rb entry in position of @fofs in rb-tree,
198  * if hit, return the entry, otherwise, return NULL
199  * @prev_ex: extent before fofs
200  * @next_ex: extent after fofs
201  * @insert_p: insert point for new extent at fofs
202  * in order to simplify the insertion after.
203  * tree must stay unchanged between lookup and insertion.
204  */
205 static struct extent_node *__lookup_extent_node_ret(struct rb_root_cached *root,
206 				struct extent_node *cached_en,
207 				unsigned int fofs,
208 				struct extent_node **prev_entry,
209 				struct extent_node **next_entry,
210 				struct rb_node ***insert_p,
211 				struct rb_node **insert_parent,
212 				bool *leftmost)
213 {
214 	struct rb_node **pnode = &root->rb_root.rb_node;
215 	struct rb_node *parent = NULL, *tmp_node;
216 	struct extent_node *en = cached_en;
217 
218 	*insert_p = NULL;
219 	*insert_parent = NULL;
220 	*prev_entry = NULL;
221 	*next_entry = NULL;
222 
223 	if (RB_EMPTY_ROOT(&root->rb_root))
224 		return NULL;
225 
226 	if (en && en->ei.fofs <= fofs && en->ei.fofs + en->ei.len > fofs)
227 		goto lookup_neighbors;
228 
229 	*leftmost = true;
230 
231 	while (*pnode) {
232 		parent = *pnode;
233 		en = rb_entry(*pnode, struct extent_node, rb_node);
234 
235 		if (fofs < en->ei.fofs) {
236 			pnode = &(*pnode)->rb_left;
237 		} else if (fofs >= en->ei.fofs + en->ei.len) {
238 			pnode = &(*pnode)->rb_right;
239 			*leftmost = false;
240 		} else {
241 			goto lookup_neighbors;
242 		}
243 	}
244 
245 	*insert_p = pnode;
246 	*insert_parent = parent;
247 
248 	en = rb_entry(parent, struct extent_node, rb_node);
249 	tmp_node = parent;
250 	if (parent && fofs > en->ei.fofs)
251 		tmp_node = rb_next(parent);
252 	*next_entry = rb_entry_safe(tmp_node, struct extent_node, rb_node);
253 
254 	tmp_node = parent;
255 	if (parent && fofs < en->ei.fofs)
256 		tmp_node = rb_prev(parent);
257 	*prev_entry = rb_entry_safe(tmp_node, struct extent_node, rb_node);
258 	return NULL;
259 
260 lookup_neighbors:
261 	if (fofs == en->ei.fofs) {
262 		/* lookup prev node for merging backward later */
263 		tmp_node = rb_prev(&en->rb_node);
264 		*prev_entry = rb_entry_safe(tmp_node,
265 					struct extent_node, rb_node);
266 	}
267 	if (fofs == en->ei.fofs + en->ei.len - 1) {
268 		/* lookup next node for merging frontward later */
269 		tmp_node = rb_next(&en->rb_node);
270 		*next_entry = rb_entry_safe(tmp_node,
271 					struct extent_node, rb_node);
272 	}
273 	return en;
274 }
275 
276 static struct kmem_cache *extent_tree_slab;
277 static struct kmem_cache *extent_node_slab;
278 
279 static struct extent_node *__attach_extent_node(struct f2fs_sb_info *sbi,
280 				struct extent_tree *et, struct extent_info *ei,
281 				struct rb_node *parent, struct rb_node **p,
282 				bool leftmost)
283 {
284 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
285 	struct extent_node *en;
286 
287 	en = f2fs_kmem_cache_alloc(extent_node_slab, GFP_ATOMIC, false, sbi);
288 	if (!en)
289 		return NULL;
290 
291 	en->ei = *ei;
292 	INIT_LIST_HEAD(&en->list);
293 	en->et = et;
294 
295 	rb_link_node(&en->rb_node, parent, p);
296 	rb_insert_color_cached(&en->rb_node, &et->root, leftmost);
297 	atomic_inc(&et->node_cnt);
298 	atomic_inc(&eti->total_ext_node);
299 	return en;
300 }
301 
302 static void __detach_extent_node(struct f2fs_sb_info *sbi,
303 				struct extent_tree *et, struct extent_node *en)
304 {
305 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
306 
307 	rb_erase_cached(&en->rb_node, &et->root);
308 	atomic_dec(&et->node_cnt);
309 	atomic_dec(&eti->total_ext_node);
310 
311 	if (et->cached_en == en)
312 		et->cached_en = NULL;
313 	kmem_cache_free(extent_node_slab, en);
314 }
315 
316 /*
317  * Flow to release an extent_node:
318  * 1. list_del_init
319  * 2. __detach_extent_node
320  * 3. kmem_cache_free.
321  */
322 static void __release_extent_node(struct f2fs_sb_info *sbi,
323 			struct extent_tree *et, struct extent_node *en)
324 {
325 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
326 
327 	spin_lock(&eti->extent_lock);
328 	f2fs_bug_on(sbi, list_empty(&en->list));
329 	list_del_init(&en->list);
330 	spin_unlock(&eti->extent_lock);
331 
332 	__detach_extent_node(sbi, et, en);
333 }
334 
335 static struct extent_tree *__grab_extent_tree(struct inode *inode,
336 						enum extent_type type)
337 {
338 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
339 	struct extent_tree_info *eti = &sbi->extent_tree[type];
340 	struct extent_tree *et;
341 	nid_t ino = inode->i_ino;
342 
343 	mutex_lock(&eti->extent_tree_lock);
344 	et = radix_tree_lookup(&eti->extent_tree_root, ino);
345 	if (!et) {
346 		et = f2fs_kmem_cache_alloc(extent_tree_slab,
347 					GFP_NOFS, true, NULL);
348 		f2fs_radix_tree_insert(&eti->extent_tree_root, ino, et);
349 		memset(et, 0, sizeof(struct extent_tree));
350 		et->ino = ino;
351 		et->type = type;
352 		et->root = RB_ROOT_CACHED;
353 		et->cached_en = NULL;
354 		rwlock_init(&et->lock);
355 		INIT_LIST_HEAD(&et->list);
356 		atomic_set(&et->node_cnt, 0);
357 		atomic_inc(&eti->total_ext_tree);
358 	} else {
359 		atomic_dec(&eti->total_zombie_tree);
360 		list_del_init(&et->list);
361 	}
362 	mutex_unlock(&eti->extent_tree_lock);
363 
364 	/* never died until evict_inode */
365 	F2FS_I(inode)->extent_tree[type] = et;
366 
367 	return et;
368 }
369 
370 static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi,
371 					struct extent_tree *et)
372 {
373 	struct rb_node *node, *next;
374 	struct extent_node *en;
375 	unsigned int count = atomic_read(&et->node_cnt);
376 
377 	node = rb_first_cached(&et->root);
378 	while (node) {
379 		next = rb_next(node);
380 		en = rb_entry(node, struct extent_node, rb_node);
381 		__release_extent_node(sbi, et, en);
382 		node = next;
383 	}
384 
385 	return count - atomic_read(&et->node_cnt);
386 }
387 
388 static void __drop_largest_extent(struct extent_tree *et,
389 					pgoff_t fofs, unsigned int len)
390 {
391 	if (fofs < et->largest.fofs + et->largest.len &&
392 			fofs + len > et->largest.fofs) {
393 		et->largest.len = 0;
394 		et->largest_updated = true;
395 	}
396 }
397 
398 void f2fs_init_read_extent_tree(struct inode *inode, struct page *ipage)
399 {
400 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
401 	struct extent_tree_info *eti = &sbi->extent_tree[EX_READ];
402 	struct f2fs_extent *i_ext = &F2FS_INODE(ipage)->i_ext;
403 	struct extent_tree *et;
404 	struct extent_node *en;
405 	struct extent_info ei;
406 
407 	if (!__may_extent_tree(inode, EX_READ)) {
408 		/* drop largest read extent */
409 		if (i_ext && i_ext->len) {
410 			f2fs_wait_on_page_writeback(ipage, NODE, true, true);
411 			i_ext->len = 0;
412 			set_page_dirty(ipage);
413 		}
414 		goto out;
415 	}
416 
417 	et = __grab_extent_tree(inode, EX_READ);
418 
419 	if (!i_ext || !i_ext->len)
420 		goto out;
421 
422 	get_read_extent_info(&ei, i_ext);
423 
424 	write_lock(&et->lock);
425 	if (atomic_read(&et->node_cnt))
426 		goto unlock_out;
427 
428 	en = __attach_extent_node(sbi, et, &ei, NULL,
429 				&et->root.rb_root.rb_node, true);
430 	if (en) {
431 		et->largest = en->ei;
432 		et->cached_en = en;
433 
434 		spin_lock(&eti->extent_lock);
435 		list_add_tail(&en->list, &eti->extent_list);
436 		spin_unlock(&eti->extent_lock);
437 	}
438 unlock_out:
439 	write_unlock(&et->lock);
440 out:
441 	if (!F2FS_I(inode)->extent_tree[EX_READ])
442 		set_inode_flag(inode, FI_NO_EXTENT);
443 }
444 
445 void f2fs_init_age_extent_tree(struct inode *inode)
446 {
447 	if (!__init_may_extent_tree(inode, EX_BLOCK_AGE))
448 		return;
449 	__grab_extent_tree(inode, EX_BLOCK_AGE);
450 }
451 
452 void f2fs_init_extent_tree(struct inode *inode)
453 {
454 	/* initialize read cache */
455 	if (__init_may_extent_tree(inode, EX_READ))
456 		__grab_extent_tree(inode, EX_READ);
457 
458 	/* initialize block age cache */
459 	if (__init_may_extent_tree(inode, EX_BLOCK_AGE))
460 		__grab_extent_tree(inode, EX_BLOCK_AGE);
461 }
462 
463 static bool __lookup_extent_tree(struct inode *inode, pgoff_t pgofs,
464 			struct extent_info *ei, enum extent_type type)
465 {
466 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
467 	struct extent_tree_info *eti = &sbi->extent_tree[type];
468 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
469 	struct extent_node *en;
470 	bool ret = false;
471 
472 	if (!et)
473 		return false;
474 
475 	trace_f2fs_lookup_extent_tree_start(inode, pgofs, type);
476 
477 	read_lock(&et->lock);
478 
479 	if (type == EX_READ &&
480 			et->largest.fofs <= pgofs &&
481 			et->largest.fofs + et->largest.len > pgofs) {
482 		*ei = et->largest;
483 		ret = true;
484 		stat_inc_largest_node_hit(sbi);
485 		goto out;
486 	}
487 
488 	en = __lookup_extent_node(&et->root, et->cached_en, pgofs);
489 	if (!en)
490 		goto out;
491 
492 	if (en == et->cached_en)
493 		stat_inc_cached_node_hit(sbi, type);
494 	else
495 		stat_inc_rbtree_node_hit(sbi, type);
496 
497 	*ei = en->ei;
498 	spin_lock(&eti->extent_lock);
499 	if (!list_empty(&en->list)) {
500 		list_move_tail(&en->list, &eti->extent_list);
501 		et->cached_en = en;
502 	}
503 	spin_unlock(&eti->extent_lock);
504 	ret = true;
505 out:
506 	stat_inc_total_hit(sbi, type);
507 	read_unlock(&et->lock);
508 
509 	if (type == EX_READ)
510 		trace_f2fs_lookup_read_extent_tree_end(inode, pgofs, ei);
511 	else if (type == EX_BLOCK_AGE)
512 		trace_f2fs_lookup_age_extent_tree_end(inode, pgofs, ei);
513 	return ret;
514 }
515 
516 static struct extent_node *__try_merge_extent_node(struct f2fs_sb_info *sbi,
517 				struct extent_tree *et, struct extent_info *ei,
518 				struct extent_node *prev_ex,
519 				struct extent_node *next_ex)
520 {
521 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
522 	struct extent_node *en = NULL;
523 
524 	if (prev_ex && __is_back_mergeable(ei, &prev_ex->ei, et->type)) {
525 		prev_ex->ei.len += ei->len;
526 		ei = &prev_ex->ei;
527 		en = prev_ex;
528 	}
529 
530 	if (next_ex && __is_front_mergeable(ei, &next_ex->ei, et->type)) {
531 		next_ex->ei.fofs = ei->fofs;
532 		next_ex->ei.len += ei->len;
533 		if (et->type == EX_READ)
534 			next_ex->ei.blk = ei->blk;
535 		if (en)
536 			__release_extent_node(sbi, et, prev_ex);
537 
538 		en = next_ex;
539 	}
540 
541 	if (!en)
542 		return NULL;
543 
544 	__try_update_largest_extent(et, en);
545 
546 	spin_lock(&eti->extent_lock);
547 	if (!list_empty(&en->list)) {
548 		list_move_tail(&en->list, &eti->extent_list);
549 		et->cached_en = en;
550 	}
551 	spin_unlock(&eti->extent_lock);
552 	return en;
553 }
554 
555 static struct extent_node *__insert_extent_tree(struct f2fs_sb_info *sbi,
556 				struct extent_tree *et, struct extent_info *ei,
557 				struct rb_node **insert_p,
558 				struct rb_node *insert_parent,
559 				bool leftmost)
560 {
561 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
562 	struct rb_node **p = &et->root.rb_root.rb_node;
563 	struct rb_node *parent = NULL;
564 	struct extent_node *en = NULL;
565 
566 	if (insert_p && insert_parent) {
567 		parent = insert_parent;
568 		p = insert_p;
569 		goto do_insert;
570 	}
571 
572 	leftmost = true;
573 
574 	/* look up extent_node in the rb tree */
575 	while (*p) {
576 		parent = *p;
577 		en = rb_entry(parent, struct extent_node, rb_node);
578 
579 		if (ei->fofs < en->ei.fofs) {
580 			p = &(*p)->rb_left;
581 		} else if (ei->fofs >= en->ei.fofs + en->ei.len) {
582 			p = &(*p)->rb_right;
583 			leftmost = false;
584 		} else {
585 			f2fs_bug_on(sbi, 1);
586 		}
587 	}
588 
589 do_insert:
590 	en = __attach_extent_node(sbi, et, ei, parent, p, leftmost);
591 	if (!en)
592 		return NULL;
593 
594 	__try_update_largest_extent(et, en);
595 
596 	/* update in global extent list */
597 	spin_lock(&eti->extent_lock);
598 	list_add_tail(&en->list, &eti->extent_list);
599 	et->cached_en = en;
600 	spin_unlock(&eti->extent_lock);
601 	return en;
602 }
603 
604 static void __update_extent_tree_range(struct inode *inode,
605 			struct extent_info *tei, enum extent_type type)
606 {
607 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
608 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
609 	struct extent_node *en = NULL, *en1 = NULL;
610 	struct extent_node *prev_en = NULL, *next_en = NULL;
611 	struct extent_info ei, dei, prev;
612 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
613 	unsigned int fofs = tei->fofs, len = tei->len;
614 	unsigned int end = fofs + len;
615 	bool updated = false;
616 	bool leftmost = false;
617 
618 	if (!et)
619 		return;
620 
621 	if (type == EX_READ)
622 		trace_f2fs_update_read_extent_tree_range(inode, fofs, len,
623 						tei->blk, 0);
624 	else if (type == EX_BLOCK_AGE)
625 		trace_f2fs_update_age_extent_tree_range(inode, fofs, len,
626 						tei->age, tei->last_blocks);
627 
628 	write_lock(&et->lock);
629 
630 	if (type == EX_READ) {
631 		if (is_inode_flag_set(inode, FI_NO_EXTENT)) {
632 			write_unlock(&et->lock);
633 			return;
634 		}
635 
636 		prev = et->largest;
637 		dei.len = 0;
638 
639 		/*
640 		 * drop largest extent before lookup, in case it's already
641 		 * been shrunk from extent tree
642 		 */
643 		__drop_largest_extent(et, fofs, len);
644 	}
645 
646 	/* 1. lookup first extent node in range [fofs, fofs + len - 1] */
647 	en = __lookup_extent_node_ret(&et->root,
648 					et->cached_en, fofs,
649 					&prev_en, &next_en,
650 					&insert_p, &insert_parent,
651 					&leftmost);
652 	if (!en)
653 		en = next_en;
654 
655 	/* 2. invalidate all extent nodes in range [fofs, fofs + len - 1] */
656 	while (en && en->ei.fofs < end) {
657 		unsigned int org_end;
658 		int parts = 0;	/* # of parts current extent split into */
659 
660 		next_en = en1 = NULL;
661 
662 		dei = en->ei;
663 		org_end = dei.fofs + dei.len;
664 		f2fs_bug_on(sbi, fofs >= org_end);
665 
666 		if (fofs > dei.fofs && (type != EX_READ ||
667 				fofs - dei.fofs >= F2FS_MIN_EXTENT_LEN)) {
668 			en->ei.len = fofs - en->ei.fofs;
669 			prev_en = en;
670 			parts = 1;
671 		}
672 
673 		if (end < org_end && (type != EX_READ ||
674 				org_end - end >= F2FS_MIN_EXTENT_LEN)) {
675 			if (parts) {
676 				__set_extent_info(&ei,
677 					end, org_end - end,
678 					end - dei.fofs + dei.blk, false,
679 					dei.age, dei.last_blocks,
680 					type);
681 				en1 = __insert_extent_tree(sbi, et, &ei,
682 							NULL, NULL, true);
683 				next_en = en1;
684 			} else {
685 				__set_extent_info(&en->ei,
686 					end, en->ei.len - (end - dei.fofs),
687 					en->ei.blk + (end - dei.fofs), true,
688 					dei.age, dei.last_blocks,
689 					type);
690 				next_en = en;
691 			}
692 			parts++;
693 		}
694 
695 		if (!next_en) {
696 			struct rb_node *node = rb_next(&en->rb_node);
697 
698 			next_en = rb_entry_safe(node, struct extent_node,
699 						rb_node);
700 		}
701 
702 		if (parts)
703 			__try_update_largest_extent(et, en);
704 		else
705 			__release_extent_node(sbi, et, en);
706 
707 		/*
708 		 * if original extent is split into zero or two parts, extent
709 		 * tree has been altered by deletion or insertion, therefore
710 		 * invalidate pointers regard to tree.
711 		 */
712 		if (parts != 1) {
713 			insert_p = NULL;
714 			insert_parent = NULL;
715 		}
716 		en = next_en;
717 	}
718 
719 	if (type == EX_BLOCK_AGE)
720 		goto update_age_extent_cache;
721 
722 	/* 3. update extent in read extent cache */
723 	BUG_ON(type != EX_READ);
724 
725 	if (tei->blk) {
726 		__set_extent_info(&ei, fofs, len, tei->blk, false,
727 				  0, 0, EX_READ);
728 		if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
729 			__insert_extent_tree(sbi, et, &ei,
730 					insert_p, insert_parent, leftmost);
731 
732 		/* give up extent_cache, if split and small updates happen */
733 		if (dei.len >= 1 &&
734 				prev.len < F2FS_MIN_EXTENT_LEN &&
735 				et->largest.len < F2FS_MIN_EXTENT_LEN) {
736 			et->largest.len = 0;
737 			et->largest_updated = true;
738 			set_inode_flag(inode, FI_NO_EXTENT);
739 		}
740 	}
741 
742 	if (is_inode_flag_set(inode, FI_NO_EXTENT))
743 		__free_extent_tree(sbi, et);
744 
745 	if (et->largest_updated) {
746 		et->largest_updated = false;
747 		updated = true;
748 	}
749 	goto out_read_extent_cache;
750 update_age_extent_cache:
751 	if (!tei->last_blocks)
752 		goto out_read_extent_cache;
753 
754 	__set_extent_info(&ei, fofs, len, 0, false,
755 			tei->age, tei->last_blocks, EX_BLOCK_AGE);
756 	if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
757 		__insert_extent_tree(sbi, et, &ei,
758 					insert_p, insert_parent, leftmost);
759 out_read_extent_cache:
760 	write_unlock(&et->lock);
761 
762 	if (updated)
763 		f2fs_mark_inode_dirty_sync(inode, true);
764 }
765 
766 #ifdef CONFIG_F2FS_FS_COMPRESSION
767 void f2fs_update_read_extent_tree_range_compressed(struct inode *inode,
768 				pgoff_t fofs, block_t blkaddr, unsigned int llen,
769 				unsigned int c_len)
770 {
771 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
772 	struct extent_tree *et = F2FS_I(inode)->extent_tree[EX_READ];
773 	struct extent_node *en = NULL;
774 	struct extent_node *prev_en = NULL, *next_en = NULL;
775 	struct extent_info ei;
776 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
777 	bool leftmost = false;
778 
779 	trace_f2fs_update_read_extent_tree_range(inode, fofs, llen,
780 						blkaddr, c_len);
781 
782 	/* it is safe here to check FI_NO_EXTENT w/o et->lock in ro image */
783 	if (is_inode_flag_set(inode, FI_NO_EXTENT))
784 		return;
785 
786 	write_lock(&et->lock);
787 
788 	en = __lookup_extent_node_ret(&et->root,
789 					et->cached_en, fofs,
790 					&prev_en, &next_en,
791 					&insert_p, &insert_parent,
792 					&leftmost);
793 	if (en)
794 		goto unlock_out;
795 
796 	__set_extent_info(&ei, fofs, llen, blkaddr, true, 0, 0, EX_READ);
797 	ei.c_len = c_len;
798 
799 	if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
800 		__insert_extent_tree(sbi, et, &ei,
801 				insert_p, insert_parent, leftmost);
802 unlock_out:
803 	write_unlock(&et->lock);
804 }
805 #endif
806 
807 static unsigned long long __calculate_block_age(struct f2fs_sb_info *sbi,
808 						unsigned long long new,
809 						unsigned long long old)
810 {
811 	unsigned int rem_old, rem_new;
812 	unsigned long long res;
813 	unsigned int weight = sbi->last_age_weight;
814 
815 	res = div_u64_rem(new, 100, &rem_new) * (100 - weight)
816 		+ div_u64_rem(old, 100, &rem_old) * weight;
817 
818 	if (rem_new)
819 		res += rem_new * (100 - weight) / 100;
820 	if (rem_old)
821 		res += rem_old * weight / 100;
822 
823 	return res;
824 }
825 
826 /* This returns a new age and allocated blocks in ei */
827 static int __get_new_block_age(struct inode *inode, struct extent_info *ei,
828 						block_t blkaddr)
829 {
830 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
831 	loff_t f_size = i_size_read(inode);
832 	unsigned long long cur_blocks =
833 				atomic64_read(&sbi->allocated_data_blocks);
834 	struct extent_info tei = *ei;	/* only fofs and len are valid */
835 
836 	/*
837 	 * When I/O is not aligned to a PAGE_SIZE, update will happen to the last
838 	 * file block even in seq write. So don't record age for newly last file
839 	 * block here.
840 	 */
841 	if ((f_size >> PAGE_SHIFT) == ei->fofs && f_size & (PAGE_SIZE - 1) &&
842 			blkaddr == NEW_ADDR)
843 		return -EINVAL;
844 
845 	if (__lookup_extent_tree(inode, ei->fofs, &tei, EX_BLOCK_AGE)) {
846 		unsigned long long cur_age;
847 
848 		if (cur_blocks >= tei.last_blocks)
849 			cur_age = cur_blocks - tei.last_blocks;
850 		else
851 			/* allocated_data_blocks overflow */
852 			cur_age = ULLONG_MAX - tei.last_blocks + cur_blocks;
853 
854 		if (tei.age)
855 			ei->age = __calculate_block_age(sbi, cur_age, tei.age);
856 		else
857 			ei->age = cur_age;
858 		ei->last_blocks = cur_blocks;
859 		WARN_ON(ei->age > cur_blocks);
860 		return 0;
861 	}
862 
863 	f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
864 
865 	/* the data block was allocated for the first time */
866 	if (blkaddr == NEW_ADDR)
867 		goto out;
868 
869 	if (__is_valid_data_blkaddr(blkaddr) &&
870 	    !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
871 		f2fs_bug_on(sbi, 1);
872 		return -EINVAL;
873 	}
874 out:
875 	/*
876 	 * init block age with zero, this can happen when the block age extent
877 	 * was reclaimed due to memory constraint or system reboot
878 	 */
879 	ei->age = 0;
880 	ei->last_blocks = cur_blocks;
881 	return 0;
882 }
883 
884 static void __update_extent_cache(struct dnode_of_data *dn, enum extent_type type)
885 {
886 	struct extent_info ei = {};
887 
888 	if (!__may_extent_tree(dn->inode, type))
889 		return;
890 
891 	ei.fofs = f2fs_start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
892 								dn->ofs_in_node;
893 	ei.len = 1;
894 
895 	if (type == EX_READ) {
896 		if (dn->data_blkaddr == NEW_ADDR)
897 			ei.blk = NULL_ADDR;
898 		else
899 			ei.blk = dn->data_blkaddr;
900 	} else if (type == EX_BLOCK_AGE) {
901 		if (__get_new_block_age(dn->inode, &ei, dn->data_blkaddr))
902 			return;
903 	}
904 	__update_extent_tree_range(dn->inode, &ei, type);
905 }
906 
907 static unsigned int __shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink,
908 					enum extent_type type)
909 {
910 	struct extent_tree_info *eti = &sbi->extent_tree[type];
911 	struct extent_tree *et, *next;
912 	struct extent_node *en;
913 	unsigned int node_cnt = 0, tree_cnt = 0;
914 	int remained;
915 
916 	if (!atomic_read(&eti->total_zombie_tree))
917 		goto free_node;
918 
919 	if (!mutex_trylock(&eti->extent_tree_lock))
920 		goto out;
921 
922 	/* 1. remove unreferenced extent tree */
923 	list_for_each_entry_safe(et, next, &eti->zombie_list, list) {
924 		if (atomic_read(&et->node_cnt)) {
925 			write_lock(&et->lock);
926 			node_cnt += __free_extent_tree(sbi, et);
927 			write_unlock(&et->lock);
928 		}
929 		f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
930 		list_del_init(&et->list);
931 		radix_tree_delete(&eti->extent_tree_root, et->ino);
932 		kmem_cache_free(extent_tree_slab, et);
933 		atomic_dec(&eti->total_ext_tree);
934 		atomic_dec(&eti->total_zombie_tree);
935 		tree_cnt++;
936 
937 		if (node_cnt + tree_cnt >= nr_shrink)
938 			goto unlock_out;
939 		cond_resched();
940 	}
941 	mutex_unlock(&eti->extent_tree_lock);
942 
943 free_node:
944 	/* 2. remove LRU extent entries */
945 	if (!mutex_trylock(&eti->extent_tree_lock))
946 		goto out;
947 
948 	remained = nr_shrink - (node_cnt + tree_cnt);
949 
950 	spin_lock(&eti->extent_lock);
951 	for (; remained > 0; remained--) {
952 		if (list_empty(&eti->extent_list))
953 			break;
954 		en = list_first_entry(&eti->extent_list,
955 					struct extent_node, list);
956 		et = en->et;
957 		if (!write_trylock(&et->lock)) {
958 			/* refresh this extent node's position in extent list */
959 			list_move_tail(&en->list, &eti->extent_list);
960 			continue;
961 		}
962 
963 		list_del_init(&en->list);
964 		spin_unlock(&eti->extent_lock);
965 
966 		__detach_extent_node(sbi, et, en);
967 
968 		write_unlock(&et->lock);
969 		node_cnt++;
970 		spin_lock(&eti->extent_lock);
971 	}
972 	spin_unlock(&eti->extent_lock);
973 
974 unlock_out:
975 	mutex_unlock(&eti->extent_tree_lock);
976 out:
977 	trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt, type);
978 
979 	return node_cnt + tree_cnt;
980 }
981 
982 /* read extent cache operations */
983 bool f2fs_lookup_read_extent_cache(struct inode *inode, pgoff_t pgofs,
984 				struct extent_info *ei)
985 {
986 	if (!__may_extent_tree(inode, EX_READ))
987 		return false;
988 
989 	return __lookup_extent_tree(inode, pgofs, ei, EX_READ);
990 }
991 
992 bool f2fs_lookup_read_extent_cache_block(struct inode *inode, pgoff_t index,
993 				block_t *blkaddr)
994 {
995 	struct extent_info ei = {};
996 
997 	if (!f2fs_lookup_read_extent_cache(inode, index, &ei))
998 		return false;
999 	*blkaddr = ei.blk + index - ei.fofs;
1000 	return true;
1001 }
1002 
1003 void f2fs_update_read_extent_cache(struct dnode_of_data *dn)
1004 {
1005 	return __update_extent_cache(dn, EX_READ);
1006 }
1007 
1008 void f2fs_update_read_extent_cache_range(struct dnode_of_data *dn,
1009 				pgoff_t fofs, block_t blkaddr, unsigned int len)
1010 {
1011 	struct extent_info ei = {
1012 		.fofs = fofs,
1013 		.len = len,
1014 		.blk = blkaddr,
1015 	};
1016 
1017 	if (!__may_extent_tree(dn->inode, EX_READ))
1018 		return;
1019 
1020 	__update_extent_tree_range(dn->inode, &ei, EX_READ);
1021 }
1022 
1023 unsigned int f2fs_shrink_read_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
1024 {
1025 	if (!test_opt(sbi, READ_EXTENT_CACHE))
1026 		return 0;
1027 
1028 	return __shrink_extent_tree(sbi, nr_shrink, EX_READ);
1029 }
1030 
1031 /* block age extent cache operations */
1032 bool f2fs_lookup_age_extent_cache(struct inode *inode, pgoff_t pgofs,
1033 				struct extent_info *ei)
1034 {
1035 	if (!__may_extent_tree(inode, EX_BLOCK_AGE))
1036 		return false;
1037 
1038 	return __lookup_extent_tree(inode, pgofs, ei, EX_BLOCK_AGE);
1039 }
1040 
1041 void f2fs_update_age_extent_cache(struct dnode_of_data *dn)
1042 {
1043 	return __update_extent_cache(dn, EX_BLOCK_AGE);
1044 }
1045 
1046 void f2fs_update_age_extent_cache_range(struct dnode_of_data *dn,
1047 				pgoff_t fofs, unsigned int len)
1048 {
1049 	struct extent_info ei = {
1050 		.fofs = fofs,
1051 		.len = len,
1052 	};
1053 
1054 	if (!__may_extent_tree(dn->inode, EX_BLOCK_AGE))
1055 		return;
1056 
1057 	__update_extent_tree_range(dn->inode, &ei, EX_BLOCK_AGE);
1058 }
1059 
1060 unsigned int f2fs_shrink_age_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
1061 {
1062 	if (!test_opt(sbi, AGE_EXTENT_CACHE))
1063 		return 0;
1064 
1065 	return __shrink_extent_tree(sbi, nr_shrink, EX_BLOCK_AGE);
1066 }
1067 
1068 static unsigned int __destroy_extent_node(struct inode *inode,
1069 					enum extent_type type)
1070 {
1071 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1072 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
1073 	unsigned int node_cnt = 0;
1074 
1075 	if (!et || !atomic_read(&et->node_cnt))
1076 		return 0;
1077 
1078 	write_lock(&et->lock);
1079 	node_cnt = __free_extent_tree(sbi, et);
1080 	write_unlock(&et->lock);
1081 
1082 	return node_cnt;
1083 }
1084 
1085 void f2fs_destroy_extent_node(struct inode *inode)
1086 {
1087 	__destroy_extent_node(inode, EX_READ);
1088 	__destroy_extent_node(inode, EX_BLOCK_AGE);
1089 }
1090 
1091 static void __drop_extent_tree(struct inode *inode, enum extent_type type)
1092 {
1093 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1094 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
1095 	bool updated = false;
1096 
1097 	if (!__may_extent_tree(inode, type))
1098 		return;
1099 
1100 	write_lock(&et->lock);
1101 	__free_extent_tree(sbi, et);
1102 	if (type == EX_READ) {
1103 		set_inode_flag(inode, FI_NO_EXTENT);
1104 		if (et->largest.len) {
1105 			et->largest.len = 0;
1106 			updated = true;
1107 		}
1108 	}
1109 	write_unlock(&et->lock);
1110 	if (updated)
1111 		f2fs_mark_inode_dirty_sync(inode, true);
1112 }
1113 
1114 void f2fs_drop_extent_tree(struct inode *inode)
1115 {
1116 	__drop_extent_tree(inode, EX_READ);
1117 	__drop_extent_tree(inode, EX_BLOCK_AGE);
1118 }
1119 
1120 static void __destroy_extent_tree(struct inode *inode, enum extent_type type)
1121 {
1122 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1123 	struct extent_tree_info *eti = &sbi->extent_tree[type];
1124 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
1125 	unsigned int node_cnt = 0;
1126 
1127 	if (!et)
1128 		return;
1129 
1130 	if (inode->i_nlink && !is_bad_inode(inode) &&
1131 					atomic_read(&et->node_cnt)) {
1132 		mutex_lock(&eti->extent_tree_lock);
1133 		list_add_tail(&et->list, &eti->zombie_list);
1134 		atomic_inc(&eti->total_zombie_tree);
1135 		mutex_unlock(&eti->extent_tree_lock);
1136 		return;
1137 	}
1138 
1139 	/* free all extent info belong to this extent tree */
1140 	node_cnt = __destroy_extent_node(inode, type);
1141 
1142 	/* delete extent tree entry in radix tree */
1143 	mutex_lock(&eti->extent_tree_lock);
1144 	f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
1145 	radix_tree_delete(&eti->extent_tree_root, inode->i_ino);
1146 	kmem_cache_free(extent_tree_slab, et);
1147 	atomic_dec(&eti->total_ext_tree);
1148 	mutex_unlock(&eti->extent_tree_lock);
1149 
1150 	F2FS_I(inode)->extent_tree[type] = NULL;
1151 
1152 	trace_f2fs_destroy_extent_tree(inode, node_cnt, type);
1153 }
1154 
1155 void f2fs_destroy_extent_tree(struct inode *inode)
1156 {
1157 	__destroy_extent_tree(inode, EX_READ);
1158 	__destroy_extent_tree(inode, EX_BLOCK_AGE);
1159 }
1160 
1161 static void __init_extent_tree_info(struct extent_tree_info *eti)
1162 {
1163 	INIT_RADIX_TREE(&eti->extent_tree_root, GFP_NOIO);
1164 	mutex_init(&eti->extent_tree_lock);
1165 	INIT_LIST_HEAD(&eti->extent_list);
1166 	spin_lock_init(&eti->extent_lock);
1167 	atomic_set(&eti->total_ext_tree, 0);
1168 	INIT_LIST_HEAD(&eti->zombie_list);
1169 	atomic_set(&eti->total_zombie_tree, 0);
1170 	atomic_set(&eti->total_ext_node, 0);
1171 }
1172 
1173 void f2fs_init_extent_cache_info(struct f2fs_sb_info *sbi)
1174 {
1175 	__init_extent_tree_info(&sbi->extent_tree[EX_READ]);
1176 	__init_extent_tree_info(&sbi->extent_tree[EX_BLOCK_AGE]);
1177 
1178 	/* initialize for block age extents */
1179 	atomic64_set(&sbi->allocated_data_blocks, 0);
1180 	sbi->hot_data_age_threshold = DEF_HOT_DATA_AGE_THRESHOLD;
1181 	sbi->warm_data_age_threshold = DEF_WARM_DATA_AGE_THRESHOLD;
1182 	sbi->last_age_weight = LAST_AGE_WEIGHT;
1183 }
1184 
1185 int __init f2fs_create_extent_cache(void)
1186 {
1187 	extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree",
1188 			sizeof(struct extent_tree));
1189 	if (!extent_tree_slab)
1190 		return -ENOMEM;
1191 	extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node",
1192 			sizeof(struct extent_node));
1193 	if (!extent_node_slab) {
1194 		kmem_cache_destroy(extent_tree_slab);
1195 		return -ENOMEM;
1196 	}
1197 	return 0;
1198 }
1199 
1200 void f2fs_destroy_extent_cache(void)
1201 {
1202 	kmem_cache_destroy(extent_node_slab);
1203 	kmem_cache_destroy(extent_tree_slab);
1204 }
1205