xref: /openbmc/linux/fs/f2fs/extent_cache.c (revision d5a9588c)
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, struct page *ipage)
23 {
24 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
25 	struct f2fs_extent *i_ext = &F2FS_INODE(ipage)->i_ext;
26 	struct extent_info ei;
27 
28 	get_read_extent_info(&ei, i_ext);
29 
30 	if (!ei.len)
31 		return true;
32 
33 	if (!f2fs_is_valid_blkaddr(sbi, ei.blk, DATA_GENERIC_ENHANCE) ||
34 	    !f2fs_is_valid_blkaddr(sbi, ei.blk + ei.len - 1,
35 					DATA_GENERIC_ENHANCE)) {
36 		set_sbi_flag(sbi, SBI_NEED_FSCK);
37 		f2fs_warn(sbi, "%s: inode (ino=%lx) extent info [%u, %u, %u] is incorrect, run fsck to fix",
38 			  __func__, inode->i_ino,
39 			  ei.blk, ei.fofs, ei.len);
40 		return false;
41 	}
42 	return true;
43 }
44 
45 static void __set_extent_info(struct extent_info *ei,
46 				unsigned int fofs, unsigned int len,
47 				block_t blk, bool keep_clen,
48 				unsigned long age, unsigned long last_blocks,
49 				enum extent_type type)
50 {
51 	ei->fofs = fofs;
52 	ei->len = len;
53 
54 	if (type == EX_READ) {
55 		ei->blk = blk;
56 		if (keep_clen)
57 			return;
58 #ifdef CONFIG_F2FS_FS_COMPRESSION
59 		ei->c_len = 0;
60 #endif
61 	} else if (type == EX_BLOCK_AGE) {
62 		ei->age = age;
63 		ei->last_blocks = last_blocks;
64 	}
65 }
66 
67 static bool __init_may_extent_tree(struct inode *inode, enum extent_type type)
68 {
69 	if (type == EX_READ)
70 		return test_opt(F2FS_I_SB(inode), READ_EXTENT_CACHE) &&
71 			S_ISREG(inode->i_mode);
72 	if (type == EX_BLOCK_AGE)
73 		return test_opt(F2FS_I_SB(inode), AGE_EXTENT_CACHE) &&
74 			(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode));
75 	return false;
76 }
77 
78 static bool __may_extent_tree(struct inode *inode, enum extent_type type)
79 {
80 	/*
81 	 * for recovered files during mount do not create extents
82 	 * if shrinker is not registered.
83 	 */
84 	if (list_empty(&F2FS_I_SB(inode)->s_list))
85 		return false;
86 
87 	if (!__init_may_extent_tree(inode, type))
88 		return false;
89 
90 	if (type == EX_READ) {
91 		if (is_inode_flag_set(inode, FI_NO_EXTENT))
92 			return false;
93 		if (is_inode_flag_set(inode, FI_COMPRESSED_FILE) &&
94 				 !f2fs_sb_has_readonly(F2FS_I_SB(inode)))
95 			return false;
96 	} else if (type == EX_BLOCK_AGE) {
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 true;
103 }
104 
105 static void __try_update_largest_extent(struct extent_tree *et,
106 						struct extent_node *en)
107 {
108 	if (et->type != EX_READ)
109 		return;
110 	if (en->ei.len <= et->largest.len)
111 		return;
112 
113 	et->largest = en->ei;
114 	et->largest_updated = true;
115 }
116 
117 static bool __is_extent_mergeable(struct extent_info *back,
118 		struct extent_info *front, enum extent_type type)
119 {
120 	if (type == EX_READ) {
121 #ifdef CONFIG_F2FS_FS_COMPRESSION
122 		if (back->c_len && back->len != back->c_len)
123 			return false;
124 		if (front->c_len && front->len != front->c_len)
125 			return false;
126 #endif
127 		return (back->fofs + back->len == front->fofs &&
128 				back->blk + back->len == front->blk);
129 	} else if (type == EX_BLOCK_AGE) {
130 		return (back->fofs + back->len == front->fofs &&
131 			abs(back->age - front->age) <= SAME_AGE_REGION &&
132 			abs(back->last_blocks - front->last_blocks) <=
133 							SAME_AGE_REGION);
134 	}
135 	return false;
136 }
137 
138 static bool __is_back_mergeable(struct extent_info *cur,
139 		struct extent_info *back, enum extent_type type)
140 {
141 	return __is_extent_mergeable(back, cur, type);
142 }
143 
144 static bool __is_front_mergeable(struct extent_info *cur,
145 		struct extent_info *front, enum extent_type type)
146 {
147 	return __is_extent_mergeable(cur, front, type);
148 }
149 
150 static struct extent_node *__lookup_extent_node(struct rb_root_cached *root,
151 			struct extent_node *cached_en, unsigned int fofs)
152 {
153 	struct rb_node *node = root->rb_root.rb_node;
154 	struct extent_node *en;
155 
156 	/* check a cached entry */
157 	if (cached_en && cached_en->ei.fofs <= fofs &&
158 			cached_en->ei.fofs + cached_en->ei.len > fofs)
159 		return cached_en;
160 
161 	/* check rb_tree */
162 	while (node) {
163 		en = rb_entry(node, struct extent_node, rb_node);
164 
165 		if (fofs < en->ei.fofs)
166 			node = node->rb_left;
167 		else if (fofs >= en->ei.fofs + en->ei.len)
168 			node = node->rb_right;
169 		else
170 			return en;
171 	}
172 	return NULL;
173 }
174 
175 /*
176  * lookup rb entry in position of @fofs in rb-tree,
177  * if hit, return the entry, otherwise, return NULL
178  * @prev_ex: extent before fofs
179  * @next_ex: extent after fofs
180  * @insert_p: insert point for new extent at fofs
181  * in order to simplify the insertion after.
182  * tree must stay unchanged between lookup and insertion.
183  */
184 static struct extent_node *__lookup_extent_node_ret(struct rb_root_cached *root,
185 				struct extent_node *cached_en,
186 				unsigned int fofs,
187 				struct extent_node **prev_entry,
188 				struct extent_node **next_entry,
189 				struct rb_node ***insert_p,
190 				struct rb_node **insert_parent,
191 				bool *leftmost)
192 {
193 	struct rb_node **pnode = &root->rb_root.rb_node;
194 	struct rb_node *parent = NULL, *tmp_node;
195 	struct extent_node *en = cached_en;
196 
197 	*insert_p = NULL;
198 	*insert_parent = NULL;
199 	*prev_entry = NULL;
200 	*next_entry = NULL;
201 
202 	if (RB_EMPTY_ROOT(&root->rb_root))
203 		return NULL;
204 
205 	if (en && en->ei.fofs <= fofs && en->ei.fofs + en->ei.len > fofs)
206 		goto lookup_neighbors;
207 
208 	*leftmost = true;
209 
210 	while (*pnode) {
211 		parent = *pnode;
212 		en = rb_entry(*pnode, struct extent_node, rb_node);
213 
214 		if (fofs < en->ei.fofs) {
215 			pnode = &(*pnode)->rb_left;
216 		} else if (fofs >= en->ei.fofs + en->ei.len) {
217 			pnode = &(*pnode)->rb_right;
218 			*leftmost = false;
219 		} else {
220 			goto lookup_neighbors;
221 		}
222 	}
223 
224 	*insert_p = pnode;
225 	*insert_parent = parent;
226 
227 	en = rb_entry(parent, struct extent_node, rb_node);
228 	tmp_node = parent;
229 	if (parent && fofs > en->ei.fofs)
230 		tmp_node = rb_next(parent);
231 	*next_entry = rb_entry_safe(tmp_node, struct extent_node, rb_node);
232 
233 	tmp_node = parent;
234 	if (parent && fofs < en->ei.fofs)
235 		tmp_node = rb_prev(parent);
236 	*prev_entry = rb_entry_safe(tmp_node, struct extent_node, rb_node);
237 	return NULL;
238 
239 lookup_neighbors:
240 	if (fofs == en->ei.fofs) {
241 		/* lookup prev node for merging backward later */
242 		tmp_node = rb_prev(&en->rb_node);
243 		*prev_entry = rb_entry_safe(tmp_node,
244 					struct extent_node, rb_node);
245 	}
246 	if (fofs == en->ei.fofs + en->ei.len - 1) {
247 		/* lookup next node for merging frontward later */
248 		tmp_node = rb_next(&en->rb_node);
249 		*next_entry = rb_entry_safe(tmp_node,
250 					struct extent_node, rb_node);
251 	}
252 	return en;
253 }
254 
255 static struct kmem_cache *extent_tree_slab;
256 static struct kmem_cache *extent_node_slab;
257 
258 static struct extent_node *__attach_extent_node(struct f2fs_sb_info *sbi,
259 				struct extent_tree *et, struct extent_info *ei,
260 				struct rb_node *parent, struct rb_node **p,
261 				bool leftmost)
262 {
263 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
264 	struct extent_node *en;
265 
266 	en = f2fs_kmem_cache_alloc(extent_node_slab, GFP_ATOMIC, false, sbi);
267 	if (!en)
268 		return NULL;
269 
270 	en->ei = *ei;
271 	INIT_LIST_HEAD(&en->list);
272 	en->et = et;
273 
274 	rb_link_node(&en->rb_node, parent, p);
275 	rb_insert_color_cached(&en->rb_node, &et->root, leftmost);
276 	atomic_inc(&et->node_cnt);
277 	atomic_inc(&eti->total_ext_node);
278 	return en;
279 }
280 
281 static void __detach_extent_node(struct f2fs_sb_info *sbi,
282 				struct extent_tree *et, struct extent_node *en)
283 {
284 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
285 
286 	rb_erase_cached(&en->rb_node, &et->root);
287 	atomic_dec(&et->node_cnt);
288 	atomic_dec(&eti->total_ext_node);
289 
290 	if (et->cached_en == en)
291 		et->cached_en = NULL;
292 	kmem_cache_free(extent_node_slab, en);
293 }
294 
295 /*
296  * Flow to release an extent_node:
297  * 1. list_del_init
298  * 2. __detach_extent_node
299  * 3. kmem_cache_free.
300  */
301 static void __release_extent_node(struct f2fs_sb_info *sbi,
302 			struct extent_tree *et, struct extent_node *en)
303 {
304 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
305 
306 	spin_lock(&eti->extent_lock);
307 	f2fs_bug_on(sbi, list_empty(&en->list));
308 	list_del_init(&en->list);
309 	spin_unlock(&eti->extent_lock);
310 
311 	__detach_extent_node(sbi, et, en);
312 }
313 
314 static struct extent_tree *__grab_extent_tree(struct inode *inode,
315 						enum extent_type type)
316 {
317 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
318 	struct extent_tree_info *eti = &sbi->extent_tree[type];
319 	struct extent_tree *et;
320 	nid_t ino = inode->i_ino;
321 
322 	mutex_lock(&eti->extent_tree_lock);
323 	et = radix_tree_lookup(&eti->extent_tree_root, ino);
324 	if (!et) {
325 		et = f2fs_kmem_cache_alloc(extent_tree_slab,
326 					GFP_NOFS, true, NULL);
327 		f2fs_radix_tree_insert(&eti->extent_tree_root, ino, et);
328 		memset(et, 0, sizeof(struct extent_tree));
329 		et->ino = ino;
330 		et->type = type;
331 		et->root = RB_ROOT_CACHED;
332 		et->cached_en = NULL;
333 		rwlock_init(&et->lock);
334 		INIT_LIST_HEAD(&et->list);
335 		atomic_set(&et->node_cnt, 0);
336 		atomic_inc(&eti->total_ext_tree);
337 	} else {
338 		atomic_dec(&eti->total_zombie_tree);
339 		list_del_init(&et->list);
340 	}
341 	mutex_unlock(&eti->extent_tree_lock);
342 
343 	/* never died until evict_inode */
344 	F2FS_I(inode)->extent_tree[type] = et;
345 
346 	return et;
347 }
348 
349 static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi,
350 					struct extent_tree *et)
351 {
352 	struct rb_node *node, *next;
353 	struct extent_node *en;
354 	unsigned int count = atomic_read(&et->node_cnt);
355 
356 	node = rb_first_cached(&et->root);
357 	while (node) {
358 		next = rb_next(node);
359 		en = rb_entry(node, struct extent_node, rb_node);
360 		__release_extent_node(sbi, et, en);
361 		node = next;
362 	}
363 
364 	return count - atomic_read(&et->node_cnt);
365 }
366 
367 static void __drop_largest_extent(struct extent_tree *et,
368 					pgoff_t fofs, unsigned int len)
369 {
370 	if (fofs < et->largest.fofs + et->largest.len &&
371 			fofs + len > et->largest.fofs) {
372 		et->largest.len = 0;
373 		et->largest_updated = true;
374 	}
375 }
376 
377 void f2fs_init_read_extent_tree(struct inode *inode, struct page *ipage)
378 {
379 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
380 	struct extent_tree_info *eti = &sbi->extent_tree[EX_READ];
381 	struct f2fs_extent *i_ext = &F2FS_INODE(ipage)->i_ext;
382 	struct extent_tree *et;
383 	struct extent_node *en;
384 	struct extent_info ei;
385 
386 	if (!__may_extent_tree(inode, EX_READ)) {
387 		/* drop largest read extent */
388 		if (i_ext->len) {
389 			f2fs_wait_on_page_writeback(ipage, NODE, true, true);
390 			i_ext->len = 0;
391 			set_page_dirty(ipage);
392 		}
393 		set_inode_flag(inode, FI_NO_EXTENT);
394 		return;
395 	}
396 
397 	et = __grab_extent_tree(inode, EX_READ);
398 
399 	get_read_extent_info(&ei, i_ext);
400 
401 	write_lock(&et->lock);
402 	if (atomic_read(&et->node_cnt) || !ei.len)
403 		goto skip;
404 
405 	en = __attach_extent_node(sbi, et, &ei, NULL,
406 				&et->root.rb_root.rb_node, true);
407 	if (en) {
408 		et->largest = en->ei;
409 		et->cached_en = en;
410 
411 		spin_lock(&eti->extent_lock);
412 		list_add_tail(&en->list, &eti->extent_list);
413 		spin_unlock(&eti->extent_lock);
414 	}
415 skip:
416 	/* Let's drop, if checkpoint got corrupted. */
417 	if (f2fs_cp_error(sbi)) {
418 		et->largest.len = 0;
419 		et->largest_updated = true;
420 	}
421 	write_unlock(&et->lock);
422 }
423 
424 void f2fs_init_age_extent_tree(struct inode *inode)
425 {
426 	if (!__init_may_extent_tree(inode, EX_BLOCK_AGE))
427 		return;
428 	__grab_extent_tree(inode, EX_BLOCK_AGE);
429 }
430 
431 void f2fs_init_extent_tree(struct inode *inode)
432 {
433 	/* initialize read cache */
434 	if (__init_may_extent_tree(inode, EX_READ))
435 		__grab_extent_tree(inode, EX_READ);
436 
437 	/* initialize block age cache */
438 	if (__init_may_extent_tree(inode, EX_BLOCK_AGE))
439 		__grab_extent_tree(inode, EX_BLOCK_AGE);
440 }
441 
442 static bool __lookup_extent_tree(struct inode *inode, pgoff_t pgofs,
443 			struct extent_info *ei, enum extent_type type)
444 {
445 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
446 	struct extent_tree_info *eti = &sbi->extent_tree[type];
447 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
448 	struct extent_node *en;
449 	bool ret = false;
450 
451 	if (!et)
452 		return false;
453 
454 	trace_f2fs_lookup_extent_tree_start(inode, pgofs, type);
455 
456 	read_lock(&et->lock);
457 
458 	if (type == EX_READ &&
459 			et->largest.fofs <= pgofs &&
460 			et->largest.fofs + et->largest.len > pgofs) {
461 		*ei = et->largest;
462 		ret = true;
463 		stat_inc_largest_node_hit(sbi);
464 		goto out;
465 	}
466 
467 	en = __lookup_extent_node(&et->root, et->cached_en, pgofs);
468 	if (!en)
469 		goto out;
470 
471 	if (en == et->cached_en)
472 		stat_inc_cached_node_hit(sbi, type);
473 	else
474 		stat_inc_rbtree_node_hit(sbi, type);
475 
476 	*ei = en->ei;
477 	spin_lock(&eti->extent_lock);
478 	if (!list_empty(&en->list)) {
479 		list_move_tail(&en->list, &eti->extent_list);
480 		et->cached_en = en;
481 	}
482 	spin_unlock(&eti->extent_lock);
483 	ret = true;
484 out:
485 	stat_inc_total_hit(sbi, type);
486 	read_unlock(&et->lock);
487 
488 	if (type == EX_READ)
489 		trace_f2fs_lookup_read_extent_tree_end(inode, pgofs, ei);
490 	else if (type == EX_BLOCK_AGE)
491 		trace_f2fs_lookup_age_extent_tree_end(inode, pgofs, ei);
492 	return ret;
493 }
494 
495 static struct extent_node *__try_merge_extent_node(struct f2fs_sb_info *sbi,
496 				struct extent_tree *et, struct extent_info *ei,
497 				struct extent_node *prev_ex,
498 				struct extent_node *next_ex)
499 {
500 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
501 	struct extent_node *en = NULL;
502 
503 	if (prev_ex && __is_back_mergeable(ei, &prev_ex->ei, et->type)) {
504 		prev_ex->ei.len += ei->len;
505 		ei = &prev_ex->ei;
506 		en = prev_ex;
507 	}
508 
509 	if (next_ex && __is_front_mergeable(ei, &next_ex->ei, et->type)) {
510 		next_ex->ei.fofs = ei->fofs;
511 		next_ex->ei.len += ei->len;
512 		if (et->type == EX_READ)
513 			next_ex->ei.blk = ei->blk;
514 		if (en)
515 			__release_extent_node(sbi, et, prev_ex);
516 
517 		en = next_ex;
518 	}
519 
520 	if (!en)
521 		return NULL;
522 
523 	__try_update_largest_extent(et, en);
524 
525 	spin_lock(&eti->extent_lock);
526 	if (!list_empty(&en->list)) {
527 		list_move_tail(&en->list, &eti->extent_list);
528 		et->cached_en = en;
529 	}
530 	spin_unlock(&eti->extent_lock);
531 	return en;
532 }
533 
534 static struct extent_node *__insert_extent_tree(struct f2fs_sb_info *sbi,
535 				struct extent_tree *et, struct extent_info *ei,
536 				struct rb_node **insert_p,
537 				struct rb_node *insert_parent,
538 				bool leftmost)
539 {
540 	struct extent_tree_info *eti = &sbi->extent_tree[et->type];
541 	struct rb_node **p = &et->root.rb_root.rb_node;
542 	struct rb_node *parent = NULL;
543 	struct extent_node *en = NULL;
544 
545 	if (insert_p && insert_parent) {
546 		parent = insert_parent;
547 		p = insert_p;
548 		goto do_insert;
549 	}
550 
551 	leftmost = true;
552 
553 	/* look up extent_node in the rb tree */
554 	while (*p) {
555 		parent = *p;
556 		en = rb_entry(parent, struct extent_node, rb_node);
557 
558 		if (ei->fofs < en->ei.fofs) {
559 			p = &(*p)->rb_left;
560 		} else if (ei->fofs >= en->ei.fofs + en->ei.len) {
561 			p = &(*p)->rb_right;
562 			leftmost = false;
563 		} else {
564 			f2fs_bug_on(sbi, 1);
565 		}
566 	}
567 
568 do_insert:
569 	en = __attach_extent_node(sbi, et, ei, parent, p, leftmost);
570 	if (!en)
571 		return NULL;
572 
573 	__try_update_largest_extent(et, en);
574 
575 	/* update in global extent list */
576 	spin_lock(&eti->extent_lock);
577 	list_add_tail(&en->list, &eti->extent_list);
578 	et->cached_en = en;
579 	spin_unlock(&eti->extent_lock);
580 	return en;
581 }
582 
583 static void __update_extent_tree_range(struct inode *inode,
584 			struct extent_info *tei, enum extent_type type)
585 {
586 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
587 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
588 	struct extent_node *en = NULL, *en1 = NULL;
589 	struct extent_node *prev_en = NULL, *next_en = NULL;
590 	struct extent_info ei, dei, prev;
591 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
592 	unsigned int fofs = tei->fofs, len = tei->len;
593 	unsigned int end = fofs + len;
594 	bool updated = false;
595 	bool leftmost = false;
596 
597 	if (!et)
598 		return;
599 
600 	if (type == EX_READ)
601 		trace_f2fs_update_read_extent_tree_range(inode, fofs, len,
602 						tei->blk, 0);
603 	else if (type == EX_BLOCK_AGE)
604 		trace_f2fs_update_age_extent_tree_range(inode, fofs, len,
605 						tei->age, tei->last_blocks);
606 
607 	write_lock(&et->lock);
608 
609 	if (type == EX_READ) {
610 		if (is_inode_flag_set(inode, FI_NO_EXTENT)) {
611 			write_unlock(&et->lock);
612 			return;
613 		}
614 
615 		prev = et->largest;
616 		dei.len = 0;
617 
618 		/*
619 		 * drop largest extent before lookup, in case it's already
620 		 * been shrunk from extent tree
621 		 */
622 		__drop_largest_extent(et, fofs, len);
623 	}
624 
625 	/* 1. lookup first extent node in range [fofs, fofs + len - 1] */
626 	en = __lookup_extent_node_ret(&et->root,
627 					et->cached_en, fofs,
628 					&prev_en, &next_en,
629 					&insert_p, &insert_parent,
630 					&leftmost);
631 	if (!en)
632 		en = next_en;
633 
634 	/* 2. invalidate all extent nodes in range [fofs, fofs + len - 1] */
635 	while (en && en->ei.fofs < end) {
636 		unsigned int org_end;
637 		int parts = 0;	/* # of parts current extent split into */
638 
639 		next_en = en1 = NULL;
640 
641 		dei = en->ei;
642 		org_end = dei.fofs + dei.len;
643 		f2fs_bug_on(sbi, fofs >= org_end);
644 
645 		if (fofs > dei.fofs && (type != EX_READ ||
646 				fofs - dei.fofs >= F2FS_MIN_EXTENT_LEN)) {
647 			en->ei.len = fofs - en->ei.fofs;
648 			prev_en = en;
649 			parts = 1;
650 		}
651 
652 		if (end < org_end && (type != EX_READ ||
653 				org_end - end >= F2FS_MIN_EXTENT_LEN)) {
654 			if (parts) {
655 				__set_extent_info(&ei,
656 					end, org_end - end,
657 					end - dei.fofs + dei.blk, false,
658 					dei.age, dei.last_blocks,
659 					type);
660 				en1 = __insert_extent_tree(sbi, et, &ei,
661 							NULL, NULL, true);
662 				next_en = en1;
663 			} else {
664 				__set_extent_info(&en->ei,
665 					end, en->ei.len - (end - dei.fofs),
666 					en->ei.blk + (end - dei.fofs), true,
667 					dei.age, dei.last_blocks,
668 					type);
669 				next_en = en;
670 			}
671 			parts++;
672 		}
673 
674 		if (!next_en) {
675 			struct rb_node *node = rb_next(&en->rb_node);
676 
677 			next_en = rb_entry_safe(node, struct extent_node,
678 						rb_node);
679 		}
680 
681 		if (parts)
682 			__try_update_largest_extent(et, en);
683 		else
684 			__release_extent_node(sbi, et, en);
685 
686 		/*
687 		 * if original extent is split into zero or two parts, extent
688 		 * tree has been altered by deletion or insertion, therefore
689 		 * invalidate pointers regard to tree.
690 		 */
691 		if (parts != 1) {
692 			insert_p = NULL;
693 			insert_parent = NULL;
694 		}
695 		en = next_en;
696 	}
697 
698 	if (type == EX_BLOCK_AGE)
699 		goto update_age_extent_cache;
700 
701 	/* 3. update extent in read extent cache */
702 	BUG_ON(type != EX_READ);
703 
704 	if (tei->blk) {
705 		__set_extent_info(&ei, fofs, len, tei->blk, false,
706 				  0, 0, EX_READ);
707 		if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
708 			__insert_extent_tree(sbi, et, &ei,
709 					insert_p, insert_parent, leftmost);
710 
711 		/* give up extent_cache, if split and small updates happen */
712 		if (dei.len >= 1 &&
713 				prev.len < F2FS_MIN_EXTENT_LEN &&
714 				et->largest.len < F2FS_MIN_EXTENT_LEN) {
715 			et->largest.len = 0;
716 			et->largest_updated = true;
717 			set_inode_flag(inode, FI_NO_EXTENT);
718 		}
719 	}
720 
721 	if (is_inode_flag_set(inode, FI_NO_EXTENT))
722 		__free_extent_tree(sbi, et);
723 
724 	if (et->largest_updated) {
725 		et->largest_updated = false;
726 		updated = true;
727 	}
728 	goto out_read_extent_cache;
729 update_age_extent_cache:
730 	if (!tei->last_blocks)
731 		goto out_read_extent_cache;
732 
733 	__set_extent_info(&ei, fofs, len, 0, false,
734 			tei->age, tei->last_blocks, EX_BLOCK_AGE);
735 	if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
736 		__insert_extent_tree(sbi, et, &ei,
737 					insert_p, insert_parent, leftmost);
738 out_read_extent_cache:
739 	write_unlock(&et->lock);
740 
741 	if (updated)
742 		f2fs_mark_inode_dirty_sync(inode, true);
743 }
744 
745 #ifdef CONFIG_F2FS_FS_COMPRESSION
746 void f2fs_update_read_extent_tree_range_compressed(struct inode *inode,
747 				pgoff_t fofs, block_t blkaddr, unsigned int llen,
748 				unsigned int c_len)
749 {
750 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
751 	struct extent_tree *et = F2FS_I(inode)->extent_tree[EX_READ];
752 	struct extent_node *en = NULL;
753 	struct extent_node *prev_en = NULL, *next_en = NULL;
754 	struct extent_info ei;
755 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
756 	bool leftmost = false;
757 
758 	trace_f2fs_update_read_extent_tree_range(inode, fofs, llen,
759 						blkaddr, c_len);
760 
761 	/* it is safe here to check FI_NO_EXTENT w/o et->lock in ro image */
762 	if (is_inode_flag_set(inode, FI_NO_EXTENT))
763 		return;
764 
765 	write_lock(&et->lock);
766 
767 	en = __lookup_extent_node_ret(&et->root,
768 					et->cached_en, fofs,
769 					&prev_en, &next_en,
770 					&insert_p, &insert_parent,
771 					&leftmost);
772 	if (en)
773 		goto unlock_out;
774 
775 	__set_extent_info(&ei, fofs, llen, blkaddr, true, 0, 0, EX_READ);
776 	ei.c_len = c_len;
777 
778 	if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
779 		__insert_extent_tree(sbi, et, &ei,
780 				insert_p, insert_parent, leftmost);
781 unlock_out:
782 	write_unlock(&et->lock);
783 }
784 #endif
785 
786 static unsigned long long __calculate_block_age(struct f2fs_sb_info *sbi,
787 						unsigned long long new,
788 						unsigned long long old)
789 {
790 	unsigned int rem_old, rem_new;
791 	unsigned long long res;
792 	unsigned int weight = sbi->last_age_weight;
793 
794 	res = div_u64_rem(new, 100, &rem_new) * (100 - weight)
795 		+ div_u64_rem(old, 100, &rem_old) * weight;
796 
797 	if (rem_new)
798 		res += rem_new * (100 - weight) / 100;
799 	if (rem_old)
800 		res += rem_old * weight / 100;
801 
802 	return res;
803 }
804 
805 /* This returns a new age and allocated blocks in ei */
806 static int __get_new_block_age(struct inode *inode, struct extent_info *ei,
807 						block_t blkaddr)
808 {
809 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
810 	loff_t f_size = i_size_read(inode);
811 	unsigned long long cur_blocks =
812 				atomic64_read(&sbi->allocated_data_blocks);
813 	struct extent_info tei = *ei;	/* only fofs and len are valid */
814 
815 	/*
816 	 * When I/O is not aligned to a PAGE_SIZE, update will happen to the last
817 	 * file block even in seq write. So don't record age for newly last file
818 	 * block here.
819 	 */
820 	if ((f_size >> PAGE_SHIFT) == ei->fofs && f_size & (PAGE_SIZE - 1) &&
821 			blkaddr == NEW_ADDR)
822 		return -EINVAL;
823 
824 	if (__lookup_extent_tree(inode, ei->fofs, &tei, EX_BLOCK_AGE)) {
825 		unsigned long long cur_age;
826 
827 		if (cur_blocks >= tei.last_blocks)
828 			cur_age = cur_blocks - tei.last_blocks;
829 		else
830 			/* allocated_data_blocks overflow */
831 			cur_age = ULLONG_MAX - tei.last_blocks + cur_blocks;
832 
833 		if (tei.age)
834 			ei->age = __calculate_block_age(sbi, cur_age, tei.age);
835 		else
836 			ei->age = cur_age;
837 		ei->last_blocks = cur_blocks;
838 		WARN_ON(ei->age > cur_blocks);
839 		return 0;
840 	}
841 
842 	f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
843 
844 	/* the data block was allocated for the first time */
845 	if (blkaddr == NEW_ADDR)
846 		goto out;
847 
848 	if (__is_valid_data_blkaddr(blkaddr) &&
849 	    !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
850 		f2fs_bug_on(sbi, 1);
851 		return -EINVAL;
852 	}
853 out:
854 	/*
855 	 * init block age with zero, this can happen when the block age extent
856 	 * was reclaimed due to memory constraint or system reboot
857 	 */
858 	ei->age = 0;
859 	ei->last_blocks = cur_blocks;
860 	return 0;
861 }
862 
863 static void __update_extent_cache(struct dnode_of_data *dn, enum extent_type type)
864 {
865 	struct extent_info ei = {};
866 
867 	if (!__may_extent_tree(dn->inode, type))
868 		return;
869 
870 	ei.fofs = f2fs_start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
871 								dn->ofs_in_node;
872 	ei.len = 1;
873 
874 	if (type == EX_READ) {
875 		if (dn->data_blkaddr == NEW_ADDR)
876 			ei.blk = NULL_ADDR;
877 		else
878 			ei.blk = dn->data_blkaddr;
879 	} else if (type == EX_BLOCK_AGE) {
880 		if (__get_new_block_age(dn->inode, &ei, dn->data_blkaddr))
881 			return;
882 	}
883 	__update_extent_tree_range(dn->inode, &ei, type);
884 }
885 
886 static unsigned int __shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink,
887 					enum extent_type type)
888 {
889 	struct extent_tree_info *eti = &sbi->extent_tree[type];
890 	struct extent_tree *et, *next;
891 	struct extent_node *en;
892 	unsigned int node_cnt = 0, tree_cnt = 0;
893 	int remained;
894 
895 	if (!atomic_read(&eti->total_zombie_tree))
896 		goto free_node;
897 
898 	if (!mutex_trylock(&eti->extent_tree_lock))
899 		goto out;
900 
901 	/* 1. remove unreferenced extent tree */
902 	list_for_each_entry_safe(et, next, &eti->zombie_list, list) {
903 		if (atomic_read(&et->node_cnt)) {
904 			write_lock(&et->lock);
905 			node_cnt += __free_extent_tree(sbi, et);
906 			write_unlock(&et->lock);
907 		}
908 		f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
909 		list_del_init(&et->list);
910 		radix_tree_delete(&eti->extent_tree_root, et->ino);
911 		kmem_cache_free(extent_tree_slab, et);
912 		atomic_dec(&eti->total_ext_tree);
913 		atomic_dec(&eti->total_zombie_tree);
914 		tree_cnt++;
915 
916 		if (node_cnt + tree_cnt >= nr_shrink)
917 			goto unlock_out;
918 		cond_resched();
919 	}
920 	mutex_unlock(&eti->extent_tree_lock);
921 
922 free_node:
923 	/* 2. remove LRU extent entries */
924 	if (!mutex_trylock(&eti->extent_tree_lock))
925 		goto out;
926 
927 	remained = nr_shrink - (node_cnt + tree_cnt);
928 
929 	spin_lock(&eti->extent_lock);
930 	for (; remained > 0; remained--) {
931 		if (list_empty(&eti->extent_list))
932 			break;
933 		en = list_first_entry(&eti->extent_list,
934 					struct extent_node, list);
935 		et = en->et;
936 		if (!write_trylock(&et->lock)) {
937 			/* refresh this extent node's position in extent list */
938 			list_move_tail(&en->list, &eti->extent_list);
939 			continue;
940 		}
941 
942 		list_del_init(&en->list);
943 		spin_unlock(&eti->extent_lock);
944 
945 		__detach_extent_node(sbi, et, en);
946 
947 		write_unlock(&et->lock);
948 		node_cnt++;
949 		spin_lock(&eti->extent_lock);
950 	}
951 	spin_unlock(&eti->extent_lock);
952 
953 unlock_out:
954 	mutex_unlock(&eti->extent_tree_lock);
955 out:
956 	trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt, type);
957 
958 	return node_cnt + tree_cnt;
959 }
960 
961 /* read extent cache operations */
962 bool f2fs_lookup_read_extent_cache(struct inode *inode, pgoff_t pgofs,
963 				struct extent_info *ei)
964 {
965 	if (!__may_extent_tree(inode, EX_READ))
966 		return false;
967 
968 	return __lookup_extent_tree(inode, pgofs, ei, EX_READ);
969 }
970 
971 bool f2fs_lookup_read_extent_cache_block(struct inode *inode, pgoff_t index,
972 				block_t *blkaddr)
973 {
974 	struct extent_info ei = {};
975 
976 	if (!f2fs_lookup_read_extent_cache(inode, index, &ei))
977 		return false;
978 	*blkaddr = ei.blk + index - ei.fofs;
979 	return true;
980 }
981 
982 void f2fs_update_read_extent_cache(struct dnode_of_data *dn)
983 {
984 	return __update_extent_cache(dn, EX_READ);
985 }
986 
987 void f2fs_update_read_extent_cache_range(struct dnode_of_data *dn,
988 				pgoff_t fofs, block_t blkaddr, unsigned int len)
989 {
990 	struct extent_info ei = {
991 		.fofs = fofs,
992 		.len = len,
993 		.blk = blkaddr,
994 	};
995 
996 	if (!__may_extent_tree(dn->inode, EX_READ))
997 		return;
998 
999 	__update_extent_tree_range(dn->inode, &ei, EX_READ);
1000 }
1001 
1002 unsigned int f2fs_shrink_read_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
1003 {
1004 	if (!test_opt(sbi, READ_EXTENT_CACHE))
1005 		return 0;
1006 
1007 	return __shrink_extent_tree(sbi, nr_shrink, EX_READ);
1008 }
1009 
1010 /* block age extent cache operations */
1011 bool f2fs_lookup_age_extent_cache(struct inode *inode, pgoff_t pgofs,
1012 				struct extent_info *ei)
1013 {
1014 	if (!__may_extent_tree(inode, EX_BLOCK_AGE))
1015 		return false;
1016 
1017 	return __lookup_extent_tree(inode, pgofs, ei, EX_BLOCK_AGE);
1018 }
1019 
1020 void f2fs_update_age_extent_cache(struct dnode_of_data *dn)
1021 {
1022 	return __update_extent_cache(dn, EX_BLOCK_AGE);
1023 }
1024 
1025 void f2fs_update_age_extent_cache_range(struct dnode_of_data *dn,
1026 				pgoff_t fofs, unsigned int len)
1027 {
1028 	struct extent_info ei = {
1029 		.fofs = fofs,
1030 		.len = len,
1031 	};
1032 
1033 	if (!__may_extent_tree(dn->inode, EX_BLOCK_AGE))
1034 		return;
1035 
1036 	__update_extent_tree_range(dn->inode, &ei, EX_BLOCK_AGE);
1037 }
1038 
1039 unsigned int f2fs_shrink_age_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
1040 {
1041 	if (!test_opt(sbi, AGE_EXTENT_CACHE))
1042 		return 0;
1043 
1044 	return __shrink_extent_tree(sbi, nr_shrink, EX_BLOCK_AGE);
1045 }
1046 
1047 static unsigned int __destroy_extent_node(struct inode *inode,
1048 					enum extent_type type)
1049 {
1050 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1051 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
1052 	unsigned int node_cnt = 0;
1053 
1054 	if (!et || !atomic_read(&et->node_cnt))
1055 		return 0;
1056 
1057 	write_lock(&et->lock);
1058 	node_cnt = __free_extent_tree(sbi, et);
1059 	write_unlock(&et->lock);
1060 
1061 	return node_cnt;
1062 }
1063 
1064 void f2fs_destroy_extent_node(struct inode *inode)
1065 {
1066 	__destroy_extent_node(inode, EX_READ);
1067 	__destroy_extent_node(inode, EX_BLOCK_AGE);
1068 }
1069 
1070 static void __drop_extent_tree(struct inode *inode, enum extent_type type)
1071 {
1072 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1073 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
1074 	bool updated = false;
1075 
1076 	if (!__may_extent_tree(inode, type))
1077 		return;
1078 
1079 	write_lock(&et->lock);
1080 	__free_extent_tree(sbi, et);
1081 	if (type == EX_READ) {
1082 		set_inode_flag(inode, FI_NO_EXTENT);
1083 		if (et->largest.len) {
1084 			et->largest.len = 0;
1085 			updated = true;
1086 		}
1087 	}
1088 	write_unlock(&et->lock);
1089 	if (updated)
1090 		f2fs_mark_inode_dirty_sync(inode, true);
1091 }
1092 
1093 void f2fs_drop_extent_tree(struct inode *inode)
1094 {
1095 	__drop_extent_tree(inode, EX_READ);
1096 	__drop_extent_tree(inode, EX_BLOCK_AGE);
1097 }
1098 
1099 static void __destroy_extent_tree(struct inode *inode, enum extent_type type)
1100 {
1101 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1102 	struct extent_tree_info *eti = &sbi->extent_tree[type];
1103 	struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
1104 	unsigned int node_cnt = 0;
1105 
1106 	if (!et)
1107 		return;
1108 
1109 	if (inode->i_nlink && !is_bad_inode(inode) &&
1110 					atomic_read(&et->node_cnt)) {
1111 		mutex_lock(&eti->extent_tree_lock);
1112 		list_add_tail(&et->list, &eti->zombie_list);
1113 		atomic_inc(&eti->total_zombie_tree);
1114 		mutex_unlock(&eti->extent_tree_lock);
1115 		return;
1116 	}
1117 
1118 	/* free all extent info belong to this extent tree */
1119 	node_cnt = __destroy_extent_node(inode, type);
1120 
1121 	/* delete extent tree entry in radix tree */
1122 	mutex_lock(&eti->extent_tree_lock);
1123 	f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
1124 	radix_tree_delete(&eti->extent_tree_root, inode->i_ino);
1125 	kmem_cache_free(extent_tree_slab, et);
1126 	atomic_dec(&eti->total_ext_tree);
1127 	mutex_unlock(&eti->extent_tree_lock);
1128 
1129 	F2FS_I(inode)->extent_tree[type] = NULL;
1130 
1131 	trace_f2fs_destroy_extent_tree(inode, node_cnt, type);
1132 }
1133 
1134 void f2fs_destroy_extent_tree(struct inode *inode)
1135 {
1136 	__destroy_extent_tree(inode, EX_READ);
1137 	__destroy_extent_tree(inode, EX_BLOCK_AGE);
1138 }
1139 
1140 static void __init_extent_tree_info(struct extent_tree_info *eti)
1141 {
1142 	INIT_RADIX_TREE(&eti->extent_tree_root, GFP_NOIO);
1143 	mutex_init(&eti->extent_tree_lock);
1144 	INIT_LIST_HEAD(&eti->extent_list);
1145 	spin_lock_init(&eti->extent_lock);
1146 	atomic_set(&eti->total_ext_tree, 0);
1147 	INIT_LIST_HEAD(&eti->zombie_list);
1148 	atomic_set(&eti->total_zombie_tree, 0);
1149 	atomic_set(&eti->total_ext_node, 0);
1150 }
1151 
1152 void f2fs_init_extent_cache_info(struct f2fs_sb_info *sbi)
1153 {
1154 	__init_extent_tree_info(&sbi->extent_tree[EX_READ]);
1155 	__init_extent_tree_info(&sbi->extent_tree[EX_BLOCK_AGE]);
1156 
1157 	/* initialize for block age extents */
1158 	atomic64_set(&sbi->allocated_data_blocks, 0);
1159 	sbi->hot_data_age_threshold = DEF_HOT_DATA_AGE_THRESHOLD;
1160 	sbi->warm_data_age_threshold = DEF_WARM_DATA_AGE_THRESHOLD;
1161 	sbi->last_age_weight = LAST_AGE_WEIGHT;
1162 }
1163 
1164 int __init f2fs_create_extent_cache(void)
1165 {
1166 	extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree",
1167 			sizeof(struct extent_tree));
1168 	if (!extent_tree_slab)
1169 		return -ENOMEM;
1170 	extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node",
1171 			sizeof(struct extent_node));
1172 	if (!extent_node_slab) {
1173 		kmem_cache_destroy(extent_tree_slab);
1174 		return -ENOMEM;
1175 	}
1176 	return 0;
1177 }
1178 
1179 void f2fs_destroy_extent_cache(void)
1180 {
1181 	kmem_cache_destroy(extent_node_slab);
1182 	kmem_cache_destroy(extent_tree_slab);
1183 }
1184