Loading...
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
22bool 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 int devi;
28
29 get_read_extent_info(&ei, i_ext);
30
31 if (!ei.len)
32 return true;
33
34 if (!f2fs_is_valid_blkaddr(sbi, ei.blk, DATA_GENERIC_ENHANCE) ||
35 !f2fs_is_valid_blkaddr(sbi, ei.blk + ei.len - 1,
36 DATA_GENERIC_ENHANCE)) {
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
43 if (!IS_DEVICE_ALIASING(inode))
44 return true;
45
46 for (devi = 0; devi < sbi->s_ndevs; devi++) {
47 if (FDEV(devi).start_blk != ei.blk ||
48 FDEV(devi).end_blk != ei.blk + ei.len - 1)
49 continue;
50
51 if (devi == 0) {
52 f2fs_warn(sbi,
53 "%s: inode (ino=%lx) is an alias of meta device",
54 __func__, inode->i_ino);
55 return false;
56 }
57
58 if (bdev_is_zoned(FDEV(devi).bdev)) {
59 f2fs_warn(sbi,
60 "%s: device alias inode (ino=%lx)'s extent info "
61 "[%u, %u, %u] maps to zoned block device",
62 __func__, inode->i_ino, ei.blk, ei.fofs, ei.len);
63 return false;
64 }
65 return true;
66 }
67
68 f2fs_warn(sbi, "%s: device alias inode (ino=%lx)'s extent info "
69 "[%u, %u, %u] is inconsistent w/ any devices",
70 __func__, inode->i_ino, ei.blk, ei.fofs, ei.len);
71 return false;
72}
73
74static void __set_extent_info(struct extent_info *ei,
75 unsigned int fofs, unsigned int len,
76 block_t blk, bool keep_clen,
77 unsigned long age, unsigned long last_blocks,
78 enum extent_type type)
79{
80 ei->fofs = fofs;
81 ei->len = len;
82
83 if (type == EX_READ) {
84 ei->blk = blk;
85 if (keep_clen)
86 return;
87#ifdef CONFIG_F2FS_FS_COMPRESSION
88 ei->c_len = 0;
89#endif
90 } else if (type == EX_BLOCK_AGE) {
91 ei->age = age;
92 ei->last_blocks = last_blocks;
93 }
94}
95
96static bool __init_may_extent_tree(struct inode *inode, enum extent_type type)
97{
98 if (type == EX_READ)
99 return test_opt(F2FS_I_SB(inode), READ_EXTENT_CACHE) &&
100 S_ISREG(inode->i_mode);
101 if (type == EX_BLOCK_AGE)
102 return test_opt(F2FS_I_SB(inode), AGE_EXTENT_CACHE) &&
103 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode));
104 return false;
105}
106
107static bool __may_extent_tree(struct inode *inode, enum extent_type type)
108{
109 if (IS_DEVICE_ALIASING(inode) && type == EX_READ)
110 return true;
111
112 /*
113 * for recovered files during mount do not create extents
114 * if shrinker is not registered.
115 */
116 if (list_empty(&F2FS_I_SB(inode)->s_list))
117 return false;
118
119 if (!__init_may_extent_tree(inode, type))
120 return false;
121
122 if (type == EX_READ) {
123 if (is_inode_flag_set(inode, FI_NO_EXTENT))
124 return false;
125 if (is_inode_flag_set(inode, FI_COMPRESSED_FILE) &&
126 !f2fs_sb_has_readonly(F2FS_I_SB(inode)))
127 return false;
128 } else if (type == EX_BLOCK_AGE) {
129 if (is_inode_flag_set(inode, FI_COMPRESSED_FILE))
130 return false;
131 if (file_is_cold(inode))
132 return false;
133 }
134 return true;
135}
136
137static void __try_update_largest_extent(struct extent_tree *et,
138 struct extent_node *en)
139{
140 if (et->type != EX_READ)
141 return;
142 if (en->ei.len <= et->largest.len)
143 return;
144
145 et->largest = en->ei;
146 et->largest_updated = true;
147}
148
149static bool __is_extent_mergeable(struct extent_info *back,
150 struct extent_info *front, enum extent_type type)
151{
152 if (type == EX_READ) {
153#ifdef CONFIG_F2FS_FS_COMPRESSION
154 if (back->c_len && back->len != back->c_len)
155 return false;
156 if (front->c_len && front->len != front->c_len)
157 return false;
158#endif
159 return (back->fofs + back->len == front->fofs &&
160 back->blk + back->len == front->blk);
161 } else if (type == EX_BLOCK_AGE) {
162 return (back->fofs + back->len == front->fofs &&
163 abs(back->age - front->age) <= SAME_AGE_REGION &&
164 abs(back->last_blocks - front->last_blocks) <=
165 SAME_AGE_REGION);
166 }
167 return false;
168}
169
170static bool __is_back_mergeable(struct extent_info *cur,
171 struct extent_info *back, enum extent_type type)
172{
173 return __is_extent_mergeable(back, cur, type);
174}
175
176static bool __is_front_mergeable(struct extent_info *cur,
177 struct extent_info *front, enum extent_type type)
178{
179 return __is_extent_mergeable(cur, front, type);
180}
181
182static struct extent_node *__lookup_extent_node(struct rb_root_cached *root,
183 struct extent_node *cached_en, unsigned int fofs)
184{
185 struct rb_node *node = root->rb_root.rb_node;
186 struct extent_node *en;
187
188 /* check a cached entry */
189 if (cached_en && cached_en->ei.fofs <= fofs &&
190 cached_en->ei.fofs + cached_en->ei.len > fofs)
191 return cached_en;
192
193 /* check rb_tree */
194 while (node) {
195 en = rb_entry(node, struct extent_node, rb_node);
196
197 if (fofs < en->ei.fofs)
198 node = node->rb_left;
199 else if (fofs >= en->ei.fofs + en->ei.len)
200 node = node->rb_right;
201 else
202 return en;
203 }
204 return NULL;
205}
206
207/*
208 * lookup rb entry in position of @fofs in rb-tree,
209 * if hit, return the entry, otherwise, return NULL
210 * @prev_ex: extent before fofs
211 * @next_ex: extent after fofs
212 * @insert_p: insert point for new extent at fofs
213 * in order to simplify the insertion after.
214 * tree must stay unchanged between lookup and insertion.
215 */
216static struct extent_node *__lookup_extent_node_ret(struct rb_root_cached *root,
217 struct extent_node *cached_en,
218 unsigned int fofs,
219 struct extent_node **prev_entry,
220 struct extent_node **next_entry,
221 struct rb_node ***insert_p,
222 struct rb_node **insert_parent,
223 bool *leftmost)
224{
225 struct rb_node **pnode = &root->rb_root.rb_node;
226 struct rb_node *parent = NULL, *tmp_node;
227 struct extent_node *en = cached_en;
228
229 *insert_p = NULL;
230 *insert_parent = NULL;
231 *prev_entry = NULL;
232 *next_entry = NULL;
233
234 if (RB_EMPTY_ROOT(&root->rb_root))
235 return NULL;
236
237 if (en && en->ei.fofs <= fofs && en->ei.fofs + en->ei.len > fofs)
238 goto lookup_neighbors;
239
240 *leftmost = true;
241
242 while (*pnode) {
243 parent = *pnode;
244 en = rb_entry(*pnode, struct extent_node, rb_node);
245
246 if (fofs < en->ei.fofs) {
247 pnode = &(*pnode)->rb_left;
248 } else if (fofs >= en->ei.fofs + en->ei.len) {
249 pnode = &(*pnode)->rb_right;
250 *leftmost = false;
251 } else {
252 goto lookup_neighbors;
253 }
254 }
255
256 *insert_p = pnode;
257 *insert_parent = parent;
258
259 en = rb_entry(parent, struct extent_node, rb_node);
260 tmp_node = parent;
261 if (parent && fofs > en->ei.fofs)
262 tmp_node = rb_next(parent);
263 *next_entry = rb_entry_safe(tmp_node, struct extent_node, rb_node);
264
265 tmp_node = parent;
266 if (parent && fofs < en->ei.fofs)
267 tmp_node = rb_prev(parent);
268 *prev_entry = rb_entry_safe(tmp_node, struct extent_node, rb_node);
269 return NULL;
270
271lookup_neighbors:
272 if (fofs == en->ei.fofs) {
273 /* lookup prev node for merging backward later */
274 tmp_node = rb_prev(&en->rb_node);
275 *prev_entry = rb_entry_safe(tmp_node,
276 struct extent_node, rb_node);
277 }
278 if (fofs == en->ei.fofs + en->ei.len - 1) {
279 /* lookup next node for merging frontward later */
280 tmp_node = rb_next(&en->rb_node);
281 *next_entry = rb_entry_safe(tmp_node,
282 struct extent_node, rb_node);
283 }
284 return en;
285}
286
287static struct kmem_cache *extent_tree_slab;
288static struct kmem_cache *extent_node_slab;
289
290static struct extent_node *__attach_extent_node(struct f2fs_sb_info *sbi,
291 struct extent_tree *et, struct extent_info *ei,
292 struct rb_node *parent, struct rb_node **p,
293 bool leftmost)
294{
295 struct extent_tree_info *eti = &sbi->extent_tree[et->type];
296 struct extent_node *en;
297
298 en = f2fs_kmem_cache_alloc(extent_node_slab, GFP_ATOMIC, false, sbi);
299 if (!en)
300 return NULL;
301
302 en->ei = *ei;
303 INIT_LIST_HEAD(&en->list);
304 en->et = et;
305
306 rb_link_node(&en->rb_node, parent, p);
307 rb_insert_color_cached(&en->rb_node, &et->root, leftmost);
308 atomic_inc(&et->node_cnt);
309 atomic_inc(&eti->total_ext_node);
310 return en;
311}
312
313static void __detach_extent_node(struct f2fs_sb_info *sbi,
314 struct extent_tree *et, struct extent_node *en)
315{
316 struct extent_tree_info *eti = &sbi->extent_tree[et->type];
317
318 rb_erase_cached(&en->rb_node, &et->root);
319 atomic_dec(&et->node_cnt);
320 atomic_dec(&eti->total_ext_node);
321
322 if (et->cached_en == en)
323 et->cached_en = NULL;
324 kmem_cache_free(extent_node_slab, en);
325}
326
327/*
328 * Flow to release an extent_node:
329 * 1. list_del_init
330 * 2. __detach_extent_node
331 * 3. kmem_cache_free.
332 */
333static void __release_extent_node(struct f2fs_sb_info *sbi,
334 struct extent_tree *et, struct extent_node *en)
335{
336 struct extent_tree_info *eti = &sbi->extent_tree[et->type];
337
338 spin_lock(&eti->extent_lock);
339 f2fs_bug_on(sbi, list_empty(&en->list));
340 list_del_init(&en->list);
341 spin_unlock(&eti->extent_lock);
342
343 __detach_extent_node(sbi, et, en);
344}
345
346static struct extent_tree *__grab_extent_tree(struct inode *inode,
347 enum extent_type type)
348{
349 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
350 struct extent_tree_info *eti = &sbi->extent_tree[type];
351 struct extent_tree *et;
352 nid_t ino = inode->i_ino;
353
354 mutex_lock(&eti->extent_tree_lock);
355 et = radix_tree_lookup(&eti->extent_tree_root, ino);
356 if (!et) {
357 et = f2fs_kmem_cache_alloc(extent_tree_slab,
358 GFP_NOFS, true, NULL);
359 f2fs_radix_tree_insert(&eti->extent_tree_root, ino, et);
360 memset(et, 0, sizeof(struct extent_tree));
361 et->ino = ino;
362 et->type = type;
363 et->root = RB_ROOT_CACHED;
364 et->cached_en = NULL;
365 rwlock_init(&et->lock);
366 INIT_LIST_HEAD(&et->list);
367 atomic_set(&et->node_cnt, 0);
368 atomic_inc(&eti->total_ext_tree);
369 } else {
370 atomic_dec(&eti->total_zombie_tree);
371 list_del_init(&et->list);
372 }
373 mutex_unlock(&eti->extent_tree_lock);
374
375 /* never died until evict_inode */
376 F2FS_I(inode)->extent_tree[type] = et;
377
378 return et;
379}
380
381static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi,
382 struct extent_tree *et, unsigned int nr_shrink)
383{
384 struct rb_node *node, *next;
385 struct extent_node *en;
386 unsigned int count;
387
388 node = rb_first_cached(&et->root);
389
390 for (count = 0; node && count < nr_shrink; count++) {
391 next = rb_next(node);
392 en = rb_entry(node, struct extent_node, rb_node);
393 __release_extent_node(sbi, et, en);
394 node = next;
395 }
396
397 return count;
398}
399
400static void __drop_largest_extent(struct extent_tree *et,
401 pgoff_t fofs, unsigned int len)
402{
403 if (fofs < (pgoff_t)et->largest.fofs + et->largest.len &&
404 fofs + len > et->largest.fofs) {
405 et->largest.len = 0;
406 et->largest_updated = true;
407 }
408}
409
410void f2fs_init_read_extent_tree(struct inode *inode, struct page *ipage)
411{
412 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
413 struct extent_tree_info *eti = &sbi->extent_tree[EX_READ];
414 struct f2fs_extent *i_ext = &F2FS_INODE(ipage)->i_ext;
415 struct extent_tree *et;
416 struct extent_node *en;
417 struct extent_info ei;
418
419 if (!__may_extent_tree(inode, EX_READ)) {
420 /* drop largest read extent */
421 if (i_ext->len) {
422 f2fs_wait_on_page_writeback(ipage, NODE, true, true);
423 i_ext->len = 0;
424 set_page_dirty(ipage);
425 }
426 set_inode_flag(inode, FI_NO_EXTENT);
427 return;
428 }
429
430 et = __grab_extent_tree(inode, EX_READ);
431
432 get_read_extent_info(&ei, i_ext);
433
434 write_lock(&et->lock);
435 if (atomic_read(&et->node_cnt) || !ei.len)
436 goto skip;
437
438 if (IS_DEVICE_ALIASING(inode)) {
439 et->largest = ei;
440 goto skip;
441 }
442
443 en = __attach_extent_node(sbi, et, &ei, NULL,
444 &et->root.rb_root.rb_node, true);
445 if (en) {
446 et->largest = en->ei;
447 et->cached_en = en;
448
449 spin_lock(&eti->extent_lock);
450 list_add_tail(&en->list, &eti->extent_list);
451 spin_unlock(&eti->extent_lock);
452 }
453skip:
454 /* Let's drop, if checkpoint got corrupted. */
455 if (f2fs_cp_error(sbi)) {
456 et->largest.len = 0;
457 et->largest_updated = true;
458 }
459 write_unlock(&et->lock);
460}
461
462void f2fs_init_age_extent_tree(struct inode *inode)
463{
464 if (!__init_may_extent_tree(inode, EX_BLOCK_AGE))
465 return;
466 __grab_extent_tree(inode, EX_BLOCK_AGE);
467}
468
469void f2fs_init_extent_tree(struct inode *inode)
470{
471 /* initialize read cache */
472 if (__init_may_extent_tree(inode, EX_READ))
473 __grab_extent_tree(inode, EX_READ);
474
475 /* initialize block age cache */
476 if (__init_may_extent_tree(inode, EX_BLOCK_AGE))
477 __grab_extent_tree(inode, EX_BLOCK_AGE);
478}
479
480static bool __lookup_extent_tree(struct inode *inode, pgoff_t pgofs,
481 struct extent_info *ei, enum extent_type type)
482{
483 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
484 struct extent_tree_info *eti = &sbi->extent_tree[type];
485 struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
486 struct extent_node *en;
487 bool ret = false;
488
489 if (!et)
490 return false;
491
492 trace_f2fs_lookup_extent_tree_start(inode, pgofs, type);
493
494 read_lock(&et->lock);
495
496 if (type == EX_READ &&
497 et->largest.fofs <= pgofs &&
498 (pgoff_t)et->largest.fofs + et->largest.len > pgofs) {
499 *ei = et->largest;
500 ret = true;
501 stat_inc_largest_node_hit(sbi);
502 goto out;
503 }
504
505 if (IS_DEVICE_ALIASING(inode)) {
506 ret = false;
507 goto out;
508 }
509
510 en = __lookup_extent_node(&et->root, et->cached_en, pgofs);
511 if (!en)
512 goto out;
513
514 if (en == et->cached_en)
515 stat_inc_cached_node_hit(sbi, type);
516 else
517 stat_inc_rbtree_node_hit(sbi, type);
518
519 *ei = en->ei;
520 spin_lock(&eti->extent_lock);
521 if (!list_empty(&en->list)) {
522 list_move_tail(&en->list, &eti->extent_list);
523 et->cached_en = en;
524 }
525 spin_unlock(&eti->extent_lock);
526 ret = true;
527out:
528 stat_inc_total_hit(sbi, type);
529 read_unlock(&et->lock);
530
531 if (type == EX_READ)
532 trace_f2fs_lookup_read_extent_tree_end(inode, pgofs, ei);
533 else if (type == EX_BLOCK_AGE)
534 trace_f2fs_lookup_age_extent_tree_end(inode, pgofs, ei);
535 return ret;
536}
537
538static struct extent_node *__try_merge_extent_node(struct f2fs_sb_info *sbi,
539 struct extent_tree *et, struct extent_info *ei,
540 struct extent_node *prev_ex,
541 struct extent_node *next_ex)
542{
543 struct extent_tree_info *eti = &sbi->extent_tree[et->type];
544 struct extent_node *en = NULL;
545
546 if (prev_ex && __is_back_mergeable(ei, &prev_ex->ei, et->type)) {
547 prev_ex->ei.len += ei->len;
548 ei = &prev_ex->ei;
549 en = prev_ex;
550 }
551
552 if (next_ex && __is_front_mergeable(ei, &next_ex->ei, et->type)) {
553 next_ex->ei.fofs = ei->fofs;
554 next_ex->ei.len += ei->len;
555 if (et->type == EX_READ)
556 next_ex->ei.blk = ei->blk;
557 if (en)
558 __release_extent_node(sbi, et, prev_ex);
559
560 en = next_ex;
561 }
562
563 if (!en)
564 return NULL;
565
566 __try_update_largest_extent(et, en);
567
568 spin_lock(&eti->extent_lock);
569 if (!list_empty(&en->list)) {
570 list_move_tail(&en->list, &eti->extent_list);
571 et->cached_en = en;
572 }
573 spin_unlock(&eti->extent_lock);
574 return en;
575}
576
577static struct extent_node *__insert_extent_tree(struct f2fs_sb_info *sbi,
578 struct extent_tree *et, struct extent_info *ei,
579 struct rb_node **insert_p,
580 struct rb_node *insert_parent,
581 bool leftmost)
582{
583 struct extent_tree_info *eti = &sbi->extent_tree[et->type];
584 struct rb_node **p = &et->root.rb_root.rb_node;
585 struct rb_node *parent = NULL;
586 struct extent_node *en = NULL;
587
588 if (insert_p && insert_parent) {
589 parent = insert_parent;
590 p = insert_p;
591 goto do_insert;
592 }
593
594 leftmost = true;
595
596 /* look up extent_node in the rb tree */
597 while (*p) {
598 parent = *p;
599 en = rb_entry(parent, struct extent_node, rb_node);
600
601 if (ei->fofs < en->ei.fofs) {
602 p = &(*p)->rb_left;
603 } else if (ei->fofs >= en->ei.fofs + en->ei.len) {
604 p = &(*p)->rb_right;
605 leftmost = false;
606 } else {
607 f2fs_bug_on(sbi, 1);
608 }
609 }
610
611do_insert:
612 en = __attach_extent_node(sbi, et, ei, parent, p, leftmost);
613 if (!en)
614 return NULL;
615
616 __try_update_largest_extent(et, en);
617
618 /* update in global extent list */
619 spin_lock(&eti->extent_lock);
620 list_add_tail(&en->list, &eti->extent_list);
621 et->cached_en = en;
622 spin_unlock(&eti->extent_lock);
623 return en;
624}
625
626static unsigned int __destroy_extent_node(struct inode *inode,
627 enum extent_type type)
628{
629 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
630 struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
631 unsigned int nr_shrink = type == EX_READ ?
632 READ_EXTENT_CACHE_SHRINK_NUMBER :
633 AGE_EXTENT_CACHE_SHRINK_NUMBER;
634 unsigned int node_cnt = 0;
635
636 if (!et || !atomic_read(&et->node_cnt))
637 return 0;
638
639 while (atomic_read(&et->node_cnt)) {
640 write_lock(&et->lock);
641 node_cnt += __free_extent_tree(sbi, et, nr_shrink);
642 write_unlock(&et->lock);
643 }
644
645 f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
646
647 return node_cnt;
648}
649
650static void __update_extent_tree_range(struct inode *inode,
651 struct extent_info *tei, enum extent_type type)
652{
653 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
654 struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
655 struct extent_node *en = NULL, *en1 = NULL;
656 struct extent_node *prev_en = NULL, *next_en = NULL;
657 struct extent_info ei, dei, prev;
658 struct rb_node **insert_p = NULL, *insert_parent = NULL;
659 unsigned int fofs = tei->fofs, len = tei->len;
660 unsigned int end = fofs + len;
661 bool updated = false;
662 bool leftmost = false;
663
664 if (!et)
665 return;
666
667 if (type == EX_READ)
668 trace_f2fs_update_read_extent_tree_range(inode, fofs, len,
669 tei->blk, 0);
670 else if (type == EX_BLOCK_AGE)
671 trace_f2fs_update_age_extent_tree_range(inode, fofs, len,
672 tei->age, tei->last_blocks);
673
674 write_lock(&et->lock);
675
676 if (type == EX_READ) {
677 if (is_inode_flag_set(inode, FI_NO_EXTENT)) {
678 write_unlock(&et->lock);
679 return;
680 }
681
682 prev = et->largest;
683 dei.len = 0;
684
685 /*
686 * drop largest extent before lookup, in case it's already
687 * been shrunk from extent tree
688 */
689 __drop_largest_extent(et, fofs, len);
690 }
691
692 /* 1. lookup first extent node in range [fofs, fofs + len - 1] */
693 en = __lookup_extent_node_ret(&et->root,
694 et->cached_en, fofs,
695 &prev_en, &next_en,
696 &insert_p, &insert_parent,
697 &leftmost);
698 if (!en)
699 en = next_en;
700
701 /* 2. invalidate all extent nodes in range [fofs, fofs + len - 1] */
702 while (en && en->ei.fofs < end) {
703 unsigned int org_end;
704 int parts = 0; /* # of parts current extent split into */
705
706 next_en = en1 = NULL;
707
708 dei = en->ei;
709 org_end = dei.fofs + dei.len;
710 f2fs_bug_on(sbi, fofs >= org_end);
711
712 if (fofs > dei.fofs && (type != EX_READ ||
713 fofs - dei.fofs >= F2FS_MIN_EXTENT_LEN)) {
714 en->ei.len = fofs - en->ei.fofs;
715 prev_en = en;
716 parts = 1;
717 }
718
719 if (end < org_end && (type != EX_READ ||
720 (org_end - end >= F2FS_MIN_EXTENT_LEN &&
721 atomic_read(&et->node_cnt) <
722 sbi->max_read_extent_count))) {
723 if (parts) {
724 __set_extent_info(&ei,
725 end, org_end - end,
726 end - dei.fofs + dei.blk, false,
727 dei.age, dei.last_blocks,
728 type);
729 en1 = __insert_extent_tree(sbi, et, &ei,
730 NULL, NULL, true);
731 next_en = en1;
732 } else {
733 __set_extent_info(&en->ei,
734 end, en->ei.len - (end - dei.fofs),
735 en->ei.blk + (end - dei.fofs), true,
736 dei.age, dei.last_blocks,
737 type);
738 next_en = en;
739 }
740 parts++;
741 }
742
743 if (!next_en) {
744 struct rb_node *node = rb_next(&en->rb_node);
745
746 next_en = rb_entry_safe(node, struct extent_node,
747 rb_node);
748 }
749
750 if (parts)
751 __try_update_largest_extent(et, en);
752 else
753 __release_extent_node(sbi, et, en);
754
755 /*
756 * if original extent is split into zero or two parts, extent
757 * tree has been altered by deletion or insertion, therefore
758 * invalidate pointers regard to tree.
759 */
760 if (parts != 1) {
761 insert_p = NULL;
762 insert_parent = NULL;
763 }
764 en = next_en;
765 }
766
767 if (type == EX_BLOCK_AGE)
768 goto update_age_extent_cache;
769
770 /* 3. update extent in read extent cache */
771 BUG_ON(type != EX_READ);
772
773 if (tei->blk) {
774 __set_extent_info(&ei, fofs, len, tei->blk, false,
775 0, 0, EX_READ);
776 if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
777 __insert_extent_tree(sbi, et, &ei,
778 insert_p, insert_parent, leftmost);
779
780 /* give up extent_cache, if split and small updates happen */
781 if (dei.len >= 1 &&
782 prev.len < F2FS_MIN_EXTENT_LEN &&
783 et->largest.len < F2FS_MIN_EXTENT_LEN) {
784 et->largest.len = 0;
785 et->largest_updated = true;
786 set_inode_flag(inode, FI_NO_EXTENT);
787 }
788 }
789
790 if (et->largest_updated) {
791 et->largest_updated = false;
792 updated = true;
793 }
794 goto out_read_extent_cache;
795update_age_extent_cache:
796 if (!tei->last_blocks)
797 goto out_read_extent_cache;
798
799 __set_extent_info(&ei, fofs, len, 0, false,
800 tei->age, tei->last_blocks, EX_BLOCK_AGE);
801 if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
802 __insert_extent_tree(sbi, et, &ei,
803 insert_p, insert_parent, leftmost);
804out_read_extent_cache:
805 write_unlock(&et->lock);
806
807 if (is_inode_flag_set(inode, FI_NO_EXTENT))
808 __destroy_extent_node(inode, EX_READ);
809
810 if (updated)
811 f2fs_mark_inode_dirty_sync(inode, true);
812}
813
814#ifdef CONFIG_F2FS_FS_COMPRESSION
815void f2fs_update_read_extent_tree_range_compressed(struct inode *inode,
816 pgoff_t fofs, block_t blkaddr, unsigned int llen,
817 unsigned int c_len)
818{
819 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
820 struct extent_tree *et = F2FS_I(inode)->extent_tree[EX_READ];
821 struct extent_node *en = NULL;
822 struct extent_node *prev_en = NULL, *next_en = NULL;
823 struct extent_info ei;
824 struct rb_node **insert_p = NULL, *insert_parent = NULL;
825 bool leftmost = false;
826
827 trace_f2fs_update_read_extent_tree_range(inode, fofs, llen,
828 blkaddr, c_len);
829
830 /* it is safe here to check FI_NO_EXTENT w/o et->lock in ro image */
831 if (is_inode_flag_set(inode, FI_NO_EXTENT))
832 return;
833
834 write_lock(&et->lock);
835
836 en = __lookup_extent_node_ret(&et->root,
837 et->cached_en, fofs,
838 &prev_en, &next_en,
839 &insert_p, &insert_parent,
840 &leftmost);
841 if (en)
842 goto unlock_out;
843
844 __set_extent_info(&ei, fofs, llen, blkaddr, true, 0, 0, EX_READ);
845 ei.c_len = c_len;
846
847 if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
848 __insert_extent_tree(sbi, et, &ei,
849 insert_p, insert_parent, leftmost);
850unlock_out:
851 write_unlock(&et->lock);
852}
853#endif
854
855static unsigned long long __calculate_block_age(struct f2fs_sb_info *sbi,
856 unsigned long long new,
857 unsigned long long old)
858{
859 unsigned int rem_old, rem_new;
860 unsigned long long res;
861 unsigned int weight = sbi->last_age_weight;
862
863 res = div_u64_rem(new, 100, &rem_new) * (100 - weight)
864 + div_u64_rem(old, 100, &rem_old) * weight;
865
866 if (rem_new)
867 res += rem_new * (100 - weight) / 100;
868 if (rem_old)
869 res += rem_old * weight / 100;
870
871 return res;
872}
873
874/* This returns a new age and allocated blocks in ei */
875static int __get_new_block_age(struct inode *inode, struct extent_info *ei,
876 block_t blkaddr)
877{
878 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
879 loff_t f_size = i_size_read(inode);
880 unsigned long long cur_blocks =
881 atomic64_read(&sbi->allocated_data_blocks);
882 struct extent_info tei = *ei; /* only fofs and len are valid */
883
884 /*
885 * When I/O is not aligned to a PAGE_SIZE, update will happen to the last
886 * file block even in seq write. So don't record age for newly last file
887 * block here.
888 */
889 if ((f_size >> PAGE_SHIFT) == ei->fofs && f_size & (PAGE_SIZE - 1) &&
890 blkaddr == NEW_ADDR)
891 return -EINVAL;
892
893 if (__lookup_extent_tree(inode, ei->fofs, &tei, EX_BLOCK_AGE)) {
894 unsigned long long cur_age;
895
896 if (cur_blocks >= tei.last_blocks)
897 cur_age = cur_blocks - tei.last_blocks;
898 else
899 /* allocated_data_blocks overflow */
900 cur_age = ULLONG_MAX - tei.last_blocks + cur_blocks;
901
902 if (tei.age)
903 ei->age = __calculate_block_age(sbi, cur_age, tei.age);
904 else
905 ei->age = cur_age;
906 ei->last_blocks = cur_blocks;
907 WARN_ON(ei->age > cur_blocks);
908 return 0;
909 }
910
911 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
912
913 /* the data block was allocated for the first time */
914 if (blkaddr == NEW_ADDR)
915 goto out;
916
917 if (__is_valid_data_blkaddr(blkaddr) &&
918 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
919 return -EINVAL;
920out:
921 /*
922 * init block age with zero, this can happen when the block age extent
923 * was reclaimed due to memory constraint or system reboot
924 */
925 ei->age = 0;
926 ei->last_blocks = cur_blocks;
927 return 0;
928}
929
930static void __update_extent_cache(struct dnode_of_data *dn, enum extent_type type)
931{
932 struct extent_info ei = {};
933
934 if (!__may_extent_tree(dn->inode, type))
935 return;
936
937 ei.fofs = f2fs_start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
938 dn->ofs_in_node;
939 ei.len = 1;
940
941 if (type == EX_READ) {
942 if (dn->data_blkaddr == NEW_ADDR)
943 ei.blk = NULL_ADDR;
944 else
945 ei.blk = dn->data_blkaddr;
946 } else if (type == EX_BLOCK_AGE) {
947 if (__get_new_block_age(dn->inode, &ei, dn->data_blkaddr))
948 return;
949 }
950 __update_extent_tree_range(dn->inode, &ei, type);
951}
952
953static unsigned int __shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink,
954 enum extent_type type)
955{
956 struct extent_tree_info *eti = &sbi->extent_tree[type];
957 struct extent_tree *et, *next;
958 struct extent_node *en;
959 unsigned int node_cnt = 0, tree_cnt = 0;
960 int remained;
961
962 if (!atomic_read(&eti->total_zombie_tree))
963 goto free_node;
964
965 if (!mutex_trylock(&eti->extent_tree_lock))
966 goto out;
967
968 /* 1. remove unreferenced extent tree */
969 list_for_each_entry_safe(et, next, &eti->zombie_list, list) {
970 if (atomic_read(&et->node_cnt)) {
971 write_lock(&et->lock);
972 node_cnt += __free_extent_tree(sbi, et,
973 nr_shrink - node_cnt - tree_cnt);
974 write_unlock(&et->lock);
975 }
976
977 if (atomic_read(&et->node_cnt))
978 goto unlock_out;
979
980 list_del_init(&et->list);
981 radix_tree_delete(&eti->extent_tree_root, et->ino);
982 kmem_cache_free(extent_tree_slab, et);
983 atomic_dec(&eti->total_ext_tree);
984 atomic_dec(&eti->total_zombie_tree);
985 tree_cnt++;
986
987 if (node_cnt + tree_cnt >= nr_shrink)
988 goto unlock_out;
989 cond_resched();
990 }
991 mutex_unlock(&eti->extent_tree_lock);
992
993free_node:
994 /* 2. remove LRU extent entries */
995 if (!mutex_trylock(&eti->extent_tree_lock))
996 goto out;
997
998 remained = nr_shrink - (node_cnt + tree_cnt);
999
1000 spin_lock(&eti->extent_lock);
1001 for (; remained > 0; remained--) {
1002 if (list_empty(&eti->extent_list))
1003 break;
1004 en = list_first_entry(&eti->extent_list,
1005 struct extent_node, list);
1006 et = en->et;
1007 if (!write_trylock(&et->lock)) {
1008 /* refresh this extent node's position in extent list */
1009 list_move_tail(&en->list, &eti->extent_list);
1010 continue;
1011 }
1012
1013 list_del_init(&en->list);
1014 spin_unlock(&eti->extent_lock);
1015
1016 __detach_extent_node(sbi, et, en);
1017
1018 write_unlock(&et->lock);
1019 node_cnt++;
1020 spin_lock(&eti->extent_lock);
1021 }
1022 spin_unlock(&eti->extent_lock);
1023
1024unlock_out:
1025 mutex_unlock(&eti->extent_tree_lock);
1026out:
1027 trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt, type);
1028
1029 return node_cnt + tree_cnt;
1030}
1031
1032/* read extent cache operations */
1033bool f2fs_lookup_read_extent_cache(struct inode *inode, pgoff_t pgofs,
1034 struct extent_info *ei)
1035{
1036 if (!__may_extent_tree(inode, EX_READ))
1037 return false;
1038
1039 return __lookup_extent_tree(inode, pgofs, ei, EX_READ);
1040}
1041
1042bool f2fs_lookup_read_extent_cache_block(struct inode *inode, pgoff_t index,
1043 block_t *blkaddr)
1044{
1045 struct extent_info ei = {};
1046
1047 if (!f2fs_lookup_read_extent_cache(inode, index, &ei))
1048 return false;
1049 *blkaddr = ei.blk + index - ei.fofs;
1050 return true;
1051}
1052
1053void f2fs_update_read_extent_cache(struct dnode_of_data *dn)
1054{
1055 return __update_extent_cache(dn, EX_READ);
1056}
1057
1058void f2fs_update_read_extent_cache_range(struct dnode_of_data *dn,
1059 pgoff_t fofs, block_t blkaddr, unsigned int len)
1060{
1061 struct extent_info ei = {
1062 .fofs = fofs,
1063 .len = len,
1064 .blk = blkaddr,
1065 };
1066
1067 if (!__may_extent_tree(dn->inode, EX_READ))
1068 return;
1069
1070 __update_extent_tree_range(dn->inode, &ei, EX_READ);
1071}
1072
1073unsigned int f2fs_shrink_read_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
1074{
1075 if (!test_opt(sbi, READ_EXTENT_CACHE))
1076 return 0;
1077
1078 return __shrink_extent_tree(sbi, nr_shrink, EX_READ);
1079}
1080
1081/* block age extent cache operations */
1082bool f2fs_lookup_age_extent_cache(struct inode *inode, pgoff_t pgofs,
1083 struct extent_info *ei)
1084{
1085 if (!__may_extent_tree(inode, EX_BLOCK_AGE))
1086 return false;
1087
1088 return __lookup_extent_tree(inode, pgofs, ei, EX_BLOCK_AGE);
1089}
1090
1091void f2fs_update_age_extent_cache(struct dnode_of_data *dn)
1092{
1093 return __update_extent_cache(dn, EX_BLOCK_AGE);
1094}
1095
1096void f2fs_update_age_extent_cache_range(struct dnode_of_data *dn,
1097 pgoff_t fofs, unsigned int len)
1098{
1099 struct extent_info ei = {
1100 .fofs = fofs,
1101 .len = len,
1102 };
1103
1104 if (!__may_extent_tree(dn->inode, EX_BLOCK_AGE))
1105 return;
1106
1107 __update_extent_tree_range(dn->inode, &ei, EX_BLOCK_AGE);
1108}
1109
1110unsigned int f2fs_shrink_age_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
1111{
1112 if (!test_opt(sbi, AGE_EXTENT_CACHE))
1113 return 0;
1114
1115 return __shrink_extent_tree(sbi, nr_shrink, EX_BLOCK_AGE);
1116}
1117
1118void f2fs_destroy_extent_node(struct inode *inode)
1119{
1120 __destroy_extent_node(inode, EX_READ);
1121 __destroy_extent_node(inode, EX_BLOCK_AGE);
1122}
1123
1124static void __drop_extent_tree(struct inode *inode, enum extent_type type)
1125{
1126 struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
1127 bool updated = false;
1128
1129 if (!__may_extent_tree(inode, type))
1130 return;
1131
1132 write_lock(&et->lock);
1133 if (type == EX_READ) {
1134 set_inode_flag(inode, FI_NO_EXTENT);
1135 if (et->largest.len) {
1136 et->largest.len = 0;
1137 updated = true;
1138 }
1139 }
1140 write_unlock(&et->lock);
1141
1142 __destroy_extent_node(inode, type);
1143
1144 if (updated)
1145 f2fs_mark_inode_dirty_sync(inode, true);
1146}
1147
1148void f2fs_drop_extent_tree(struct inode *inode)
1149{
1150 __drop_extent_tree(inode, EX_READ);
1151 __drop_extent_tree(inode, EX_BLOCK_AGE);
1152}
1153
1154static void __destroy_extent_tree(struct inode *inode, enum extent_type type)
1155{
1156 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1157 struct extent_tree_info *eti = &sbi->extent_tree[type];
1158 struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
1159 unsigned int node_cnt = 0;
1160
1161 if (!et)
1162 return;
1163
1164 if (inode->i_nlink && !is_bad_inode(inode) &&
1165 atomic_read(&et->node_cnt)) {
1166 mutex_lock(&eti->extent_tree_lock);
1167 list_add_tail(&et->list, &eti->zombie_list);
1168 atomic_inc(&eti->total_zombie_tree);
1169 mutex_unlock(&eti->extent_tree_lock);
1170 return;
1171 }
1172
1173 /* free all extent info belong to this extent tree */
1174 node_cnt = __destroy_extent_node(inode, type);
1175
1176 /* delete extent tree entry in radix tree */
1177 mutex_lock(&eti->extent_tree_lock);
1178 f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
1179 radix_tree_delete(&eti->extent_tree_root, inode->i_ino);
1180 kmem_cache_free(extent_tree_slab, et);
1181 atomic_dec(&eti->total_ext_tree);
1182 mutex_unlock(&eti->extent_tree_lock);
1183
1184 F2FS_I(inode)->extent_tree[type] = NULL;
1185
1186 trace_f2fs_destroy_extent_tree(inode, node_cnt, type);
1187}
1188
1189void f2fs_destroy_extent_tree(struct inode *inode)
1190{
1191 __destroy_extent_tree(inode, EX_READ);
1192 __destroy_extent_tree(inode, EX_BLOCK_AGE);
1193}
1194
1195static void __init_extent_tree_info(struct extent_tree_info *eti)
1196{
1197 INIT_RADIX_TREE(&eti->extent_tree_root, GFP_NOIO);
1198 mutex_init(&eti->extent_tree_lock);
1199 INIT_LIST_HEAD(&eti->extent_list);
1200 spin_lock_init(&eti->extent_lock);
1201 atomic_set(&eti->total_ext_tree, 0);
1202 INIT_LIST_HEAD(&eti->zombie_list);
1203 atomic_set(&eti->total_zombie_tree, 0);
1204 atomic_set(&eti->total_ext_node, 0);
1205}
1206
1207void f2fs_init_extent_cache_info(struct f2fs_sb_info *sbi)
1208{
1209 __init_extent_tree_info(&sbi->extent_tree[EX_READ]);
1210 __init_extent_tree_info(&sbi->extent_tree[EX_BLOCK_AGE]);
1211
1212 /* initialize for block age extents */
1213 atomic64_set(&sbi->allocated_data_blocks, 0);
1214 sbi->hot_data_age_threshold = DEF_HOT_DATA_AGE_THRESHOLD;
1215 sbi->warm_data_age_threshold = DEF_WARM_DATA_AGE_THRESHOLD;
1216 sbi->last_age_weight = LAST_AGE_WEIGHT;
1217 sbi->max_read_extent_count = DEF_MAX_READ_EXTENT_COUNT;
1218}
1219
1220int __init f2fs_create_extent_cache(void)
1221{
1222 extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree",
1223 sizeof(struct extent_tree));
1224 if (!extent_tree_slab)
1225 return -ENOMEM;
1226 extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node",
1227 sizeof(struct extent_node));
1228 if (!extent_node_slab) {
1229 kmem_cache_destroy(extent_tree_slab);
1230 return -ENOMEM;
1231 }
1232 return 0;
1233}
1234
1235void f2fs_destroy_extent_cache(void)
1236{
1237 kmem_cache_destroy(extent_node_slab);
1238 kmem_cache_destroy(extent_tree_slab);
1239}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * f2fs extent cache support
4 *
5 * Copyright (c) 2015 Motorola Mobility
6 * Copyright (c) 2015 Samsung Electronics
7 * Authors: Jaegeuk Kim <jaegeuk@kernel.org>
8 * Chao Yu <chao2.yu@samsung.com>
9 */
10
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13
14#include "f2fs.h"
15#include "node.h"
16#include <trace/events/f2fs.h>
17
18static struct rb_entry *__lookup_rb_tree_fast(struct rb_entry *cached_re,
19 unsigned int ofs)
20{
21 if (cached_re) {
22 if (cached_re->ofs <= ofs &&
23 cached_re->ofs + cached_re->len > ofs) {
24 return cached_re;
25 }
26 }
27 return NULL;
28}
29
30static struct rb_entry *__lookup_rb_tree_slow(struct rb_root_cached *root,
31 unsigned int ofs)
32{
33 struct rb_node *node = root->rb_root.rb_node;
34 struct rb_entry *re;
35
36 while (node) {
37 re = rb_entry(node, struct rb_entry, rb_node);
38
39 if (ofs < re->ofs)
40 node = node->rb_left;
41 else if (ofs >= re->ofs + re->len)
42 node = node->rb_right;
43 else
44 return re;
45 }
46 return NULL;
47}
48
49struct rb_entry *f2fs_lookup_rb_tree(struct rb_root_cached *root,
50 struct rb_entry *cached_re, unsigned int ofs)
51{
52 struct rb_entry *re;
53
54 re = __lookup_rb_tree_fast(cached_re, ofs);
55 if (!re)
56 return __lookup_rb_tree_slow(root, ofs);
57
58 return re;
59}
60
61struct rb_node **f2fs_lookup_rb_tree_ext(struct f2fs_sb_info *sbi,
62 struct rb_root_cached *root,
63 struct rb_node **parent,
64 unsigned long long key, bool *leftmost)
65{
66 struct rb_node **p = &root->rb_root.rb_node;
67 struct rb_entry *re;
68
69 while (*p) {
70 *parent = *p;
71 re = rb_entry(*parent, struct rb_entry, rb_node);
72
73 if (key < re->key) {
74 p = &(*p)->rb_left;
75 } else {
76 p = &(*p)->rb_right;
77 *leftmost = false;
78 }
79 }
80
81 return p;
82}
83
84struct rb_node **f2fs_lookup_rb_tree_for_insert(struct f2fs_sb_info *sbi,
85 struct rb_root_cached *root,
86 struct rb_node **parent,
87 unsigned int ofs, bool *leftmost)
88{
89 struct rb_node **p = &root->rb_root.rb_node;
90 struct rb_entry *re;
91
92 while (*p) {
93 *parent = *p;
94 re = rb_entry(*parent, struct rb_entry, rb_node);
95
96 if (ofs < re->ofs) {
97 p = &(*p)->rb_left;
98 } else if (ofs >= re->ofs + re->len) {
99 p = &(*p)->rb_right;
100 *leftmost = false;
101 } else {
102 f2fs_bug_on(sbi, 1);
103 }
104 }
105
106 return p;
107}
108
109/*
110 * lookup rb entry in position of @ofs in rb-tree,
111 * if hit, return the entry, otherwise, return NULL
112 * @prev_ex: extent before ofs
113 * @next_ex: extent after ofs
114 * @insert_p: insert point for new extent at ofs
115 * in order to simpfy the insertion after.
116 * tree must stay unchanged between lookup and insertion.
117 */
118struct rb_entry *f2fs_lookup_rb_tree_ret(struct rb_root_cached *root,
119 struct rb_entry *cached_re,
120 unsigned int ofs,
121 struct rb_entry **prev_entry,
122 struct rb_entry **next_entry,
123 struct rb_node ***insert_p,
124 struct rb_node **insert_parent,
125 bool force, bool *leftmost)
126{
127 struct rb_node **pnode = &root->rb_root.rb_node;
128 struct rb_node *parent = NULL, *tmp_node;
129 struct rb_entry *re = cached_re;
130
131 *insert_p = NULL;
132 *insert_parent = NULL;
133 *prev_entry = NULL;
134 *next_entry = NULL;
135
136 if (RB_EMPTY_ROOT(&root->rb_root))
137 return NULL;
138
139 if (re) {
140 if (re->ofs <= ofs && re->ofs + re->len > ofs)
141 goto lookup_neighbors;
142 }
143
144 if (leftmost)
145 *leftmost = true;
146
147 while (*pnode) {
148 parent = *pnode;
149 re = rb_entry(*pnode, struct rb_entry, rb_node);
150
151 if (ofs < re->ofs) {
152 pnode = &(*pnode)->rb_left;
153 } else if (ofs >= re->ofs + re->len) {
154 pnode = &(*pnode)->rb_right;
155 if (leftmost)
156 *leftmost = false;
157 } else {
158 goto lookup_neighbors;
159 }
160 }
161
162 *insert_p = pnode;
163 *insert_parent = parent;
164
165 re = rb_entry(parent, struct rb_entry, rb_node);
166 tmp_node = parent;
167 if (parent && ofs > re->ofs)
168 tmp_node = rb_next(parent);
169 *next_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
170
171 tmp_node = parent;
172 if (parent && ofs < re->ofs)
173 tmp_node = rb_prev(parent);
174 *prev_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
175 return NULL;
176
177lookup_neighbors:
178 if (ofs == re->ofs || force) {
179 /* lookup prev node for merging backward later */
180 tmp_node = rb_prev(&re->rb_node);
181 *prev_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
182 }
183 if (ofs == re->ofs + re->len - 1 || force) {
184 /* lookup next node for merging frontward later */
185 tmp_node = rb_next(&re->rb_node);
186 *next_entry = rb_entry_safe(tmp_node, struct rb_entry, rb_node);
187 }
188 return re;
189}
190
191bool f2fs_check_rb_tree_consistence(struct f2fs_sb_info *sbi,
192 struct rb_root_cached *root, bool check_key)
193{
194#ifdef CONFIG_F2FS_CHECK_FS
195 struct rb_node *cur = rb_first_cached(root), *next;
196 struct rb_entry *cur_re, *next_re;
197
198 if (!cur)
199 return true;
200
201 while (cur) {
202 next = rb_next(cur);
203 if (!next)
204 return true;
205
206 cur_re = rb_entry(cur, struct rb_entry, rb_node);
207 next_re = rb_entry(next, struct rb_entry, rb_node);
208
209 if (check_key) {
210 if (cur_re->key > next_re->key) {
211 f2fs_info(sbi, "inconsistent rbtree, "
212 "cur(%llu) next(%llu)",
213 cur_re->key, next_re->key);
214 return false;
215 }
216 goto next;
217 }
218
219 if (cur_re->ofs + cur_re->len > next_re->ofs) {
220 f2fs_info(sbi, "inconsistent rbtree, cur(%u, %u) next(%u, %u)",
221 cur_re->ofs, cur_re->len,
222 next_re->ofs, next_re->len);
223 return false;
224 }
225next:
226 cur = next;
227 }
228#endif
229 return true;
230}
231
232static struct kmem_cache *extent_tree_slab;
233static struct kmem_cache *extent_node_slab;
234
235static struct extent_node *__attach_extent_node(struct f2fs_sb_info *sbi,
236 struct extent_tree *et, struct extent_info *ei,
237 struct rb_node *parent, struct rb_node **p,
238 bool leftmost)
239{
240 struct extent_node *en;
241
242 en = kmem_cache_alloc(extent_node_slab, GFP_ATOMIC);
243 if (!en)
244 return NULL;
245
246 en->ei = *ei;
247 INIT_LIST_HEAD(&en->list);
248 en->et = et;
249
250 rb_link_node(&en->rb_node, parent, p);
251 rb_insert_color_cached(&en->rb_node, &et->root, leftmost);
252 atomic_inc(&et->node_cnt);
253 atomic_inc(&sbi->total_ext_node);
254 return en;
255}
256
257static void __detach_extent_node(struct f2fs_sb_info *sbi,
258 struct extent_tree *et, struct extent_node *en)
259{
260 rb_erase_cached(&en->rb_node, &et->root);
261 atomic_dec(&et->node_cnt);
262 atomic_dec(&sbi->total_ext_node);
263
264 if (et->cached_en == en)
265 et->cached_en = NULL;
266 kmem_cache_free(extent_node_slab, en);
267}
268
269/*
270 * Flow to release an extent_node:
271 * 1. list_del_init
272 * 2. __detach_extent_node
273 * 3. kmem_cache_free.
274 */
275static void __release_extent_node(struct f2fs_sb_info *sbi,
276 struct extent_tree *et, struct extent_node *en)
277{
278 spin_lock(&sbi->extent_lock);
279 f2fs_bug_on(sbi, list_empty(&en->list));
280 list_del_init(&en->list);
281 spin_unlock(&sbi->extent_lock);
282
283 __detach_extent_node(sbi, et, en);
284}
285
286static struct extent_tree *__grab_extent_tree(struct inode *inode)
287{
288 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
289 struct extent_tree *et;
290 nid_t ino = inode->i_ino;
291
292 mutex_lock(&sbi->extent_tree_lock);
293 et = radix_tree_lookup(&sbi->extent_tree_root, ino);
294 if (!et) {
295 et = f2fs_kmem_cache_alloc(extent_tree_slab, GFP_NOFS);
296 f2fs_radix_tree_insert(&sbi->extent_tree_root, ino, et);
297 memset(et, 0, sizeof(struct extent_tree));
298 et->ino = ino;
299 et->root = RB_ROOT_CACHED;
300 et->cached_en = NULL;
301 rwlock_init(&et->lock);
302 INIT_LIST_HEAD(&et->list);
303 atomic_set(&et->node_cnt, 0);
304 atomic_inc(&sbi->total_ext_tree);
305 } else {
306 atomic_dec(&sbi->total_zombie_tree);
307 list_del_init(&et->list);
308 }
309 mutex_unlock(&sbi->extent_tree_lock);
310
311 /* never died until evict_inode */
312 F2FS_I(inode)->extent_tree = et;
313
314 return et;
315}
316
317static struct extent_node *__init_extent_tree(struct f2fs_sb_info *sbi,
318 struct extent_tree *et, struct extent_info *ei)
319{
320 struct rb_node **p = &et->root.rb_root.rb_node;
321 struct extent_node *en;
322
323 en = __attach_extent_node(sbi, et, ei, NULL, p, true);
324 if (!en)
325 return NULL;
326
327 et->largest = en->ei;
328 et->cached_en = en;
329 return en;
330}
331
332static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi,
333 struct extent_tree *et)
334{
335 struct rb_node *node, *next;
336 struct extent_node *en;
337 unsigned int count = atomic_read(&et->node_cnt);
338
339 node = rb_first_cached(&et->root);
340 while (node) {
341 next = rb_next(node);
342 en = rb_entry(node, struct extent_node, rb_node);
343 __release_extent_node(sbi, et, en);
344 node = next;
345 }
346
347 return count - atomic_read(&et->node_cnt);
348}
349
350static void __drop_largest_extent(struct extent_tree *et,
351 pgoff_t fofs, unsigned int len)
352{
353 if (fofs < et->largest.fofs + et->largest.len &&
354 fofs + len > et->largest.fofs) {
355 et->largest.len = 0;
356 et->largest_updated = true;
357 }
358}
359
360/* return true, if inode page is changed */
361static void __f2fs_init_extent_tree(struct inode *inode, struct page *ipage)
362{
363 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
364 struct f2fs_extent *i_ext = ipage ? &F2FS_INODE(ipage)->i_ext : NULL;
365 struct extent_tree *et;
366 struct extent_node *en;
367 struct extent_info ei;
368
369 if (!f2fs_may_extent_tree(inode)) {
370 /* drop largest extent */
371 if (i_ext && i_ext->len) {
372 f2fs_wait_on_page_writeback(ipage, NODE, true, true);
373 i_ext->len = 0;
374 set_page_dirty(ipage);
375 return;
376 }
377 return;
378 }
379
380 et = __grab_extent_tree(inode);
381
382 if (!i_ext || !i_ext->len)
383 return;
384
385 get_extent_info(&ei, i_ext);
386
387 write_lock(&et->lock);
388 if (atomic_read(&et->node_cnt))
389 goto out;
390
391 en = __init_extent_tree(sbi, et, &ei);
392 if (en) {
393 spin_lock(&sbi->extent_lock);
394 list_add_tail(&en->list, &sbi->extent_list);
395 spin_unlock(&sbi->extent_lock);
396 }
397out:
398 write_unlock(&et->lock);
399}
400
401void f2fs_init_extent_tree(struct inode *inode, struct page *ipage)
402{
403 __f2fs_init_extent_tree(inode, ipage);
404
405 if (!F2FS_I(inode)->extent_tree)
406 set_inode_flag(inode, FI_NO_EXTENT);
407}
408
409static bool f2fs_lookup_extent_tree(struct inode *inode, pgoff_t pgofs,
410 struct extent_info *ei)
411{
412 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
413 struct extent_tree *et = F2FS_I(inode)->extent_tree;
414 struct extent_node *en;
415 bool ret = false;
416
417 f2fs_bug_on(sbi, !et);
418
419 trace_f2fs_lookup_extent_tree_start(inode, pgofs);
420
421 read_lock(&et->lock);
422
423 if (et->largest.fofs <= pgofs &&
424 et->largest.fofs + et->largest.len > pgofs) {
425 *ei = et->largest;
426 ret = true;
427 stat_inc_largest_node_hit(sbi);
428 goto out;
429 }
430
431 en = (struct extent_node *)f2fs_lookup_rb_tree(&et->root,
432 (struct rb_entry *)et->cached_en, pgofs);
433 if (!en)
434 goto out;
435
436 if (en == et->cached_en)
437 stat_inc_cached_node_hit(sbi);
438 else
439 stat_inc_rbtree_node_hit(sbi);
440
441 *ei = en->ei;
442 spin_lock(&sbi->extent_lock);
443 if (!list_empty(&en->list)) {
444 list_move_tail(&en->list, &sbi->extent_list);
445 et->cached_en = en;
446 }
447 spin_unlock(&sbi->extent_lock);
448 ret = true;
449out:
450 stat_inc_total_hit(sbi);
451 read_unlock(&et->lock);
452
453 trace_f2fs_lookup_extent_tree_end(inode, pgofs, ei);
454 return ret;
455}
456
457static struct extent_node *__try_merge_extent_node(struct f2fs_sb_info *sbi,
458 struct extent_tree *et, struct extent_info *ei,
459 struct extent_node *prev_ex,
460 struct extent_node *next_ex)
461{
462 struct extent_node *en = NULL;
463
464 if (prev_ex && __is_back_mergeable(ei, &prev_ex->ei)) {
465 prev_ex->ei.len += ei->len;
466 ei = &prev_ex->ei;
467 en = prev_ex;
468 }
469
470 if (next_ex && __is_front_mergeable(ei, &next_ex->ei)) {
471 next_ex->ei.fofs = ei->fofs;
472 next_ex->ei.blk = ei->blk;
473 next_ex->ei.len += ei->len;
474 if (en)
475 __release_extent_node(sbi, et, prev_ex);
476
477 en = next_ex;
478 }
479
480 if (!en)
481 return NULL;
482
483 __try_update_largest_extent(et, en);
484
485 spin_lock(&sbi->extent_lock);
486 if (!list_empty(&en->list)) {
487 list_move_tail(&en->list, &sbi->extent_list);
488 et->cached_en = en;
489 }
490 spin_unlock(&sbi->extent_lock);
491 return en;
492}
493
494static struct extent_node *__insert_extent_tree(struct f2fs_sb_info *sbi,
495 struct extent_tree *et, struct extent_info *ei,
496 struct rb_node **insert_p,
497 struct rb_node *insert_parent,
498 bool leftmost)
499{
500 struct rb_node **p;
501 struct rb_node *parent = NULL;
502 struct extent_node *en = NULL;
503
504 if (insert_p && insert_parent) {
505 parent = insert_parent;
506 p = insert_p;
507 goto do_insert;
508 }
509
510 leftmost = true;
511
512 p = f2fs_lookup_rb_tree_for_insert(sbi, &et->root, &parent,
513 ei->fofs, &leftmost);
514do_insert:
515 en = __attach_extent_node(sbi, et, ei, parent, p, leftmost);
516 if (!en)
517 return NULL;
518
519 __try_update_largest_extent(et, en);
520
521 /* update in global extent list */
522 spin_lock(&sbi->extent_lock);
523 list_add_tail(&en->list, &sbi->extent_list);
524 et->cached_en = en;
525 spin_unlock(&sbi->extent_lock);
526 return en;
527}
528
529static void f2fs_update_extent_tree_range(struct inode *inode,
530 pgoff_t fofs, block_t blkaddr, unsigned int len)
531{
532 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
533 struct extent_tree *et = F2FS_I(inode)->extent_tree;
534 struct extent_node *en = NULL, *en1 = NULL;
535 struct extent_node *prev_en = NULL, *next_en = NULL;
536 struct extent_info ei, dei, prev;
537 struct rb_node **insert_p = NULL, *insert_parent = NULL;
538 unsigned int end = fofs + len;
539 unsigned int pos = (unsigned int)fofs;
540 bool updated = false;
541 bool leftmost = false;
542
543 if (!et)
544 return;
545
546 trace_f2fs_update_extent_tree_range(inode, fofs, blkaddr, len);
547
548 write_lock(&et->lock);
549
550 if (is_inode_flag_set(inode, FI_NO_EXTENT)) {
551 write_unlock(&et->lock);
552 return;
553 }
554
555 prev = et->largest;
556 dei.len = 0;
557
558 /*
559 * drop largest extent before lookup, in case it's already
560 * been shrunk from extent tree
561 */
562 __drop_largest_extent(et, fofs, len);
563
564 /* 1. lookup first extent node in range [fofs, fofs + len - 1] */
565 en = (struct extent_node *)f2fs_lookup_rb_tree_ret(&et->root,
566 (struct rb_entry *)et->cached_en, fofs,
567 (struct rb_entry **)&prev_en,
568 (struct rb_entry **)&next_en,
569 &insert_p, &insert_parent, false,
570 &leftmost);
571 if (!en)
572 en = next_en;
573
574 /* 2. invlidate all extent nodes in range [fofs, fofs + len - 1] */
575 while (en && en->ei.fofs < end) {
576 unsigned int org_end;
577 int parts = 0; /* # of parts current extent split into */
578
579 next_en = en1 = NULL;
580
581 dei = en->ei;
582 org_end = dei.fofs + dei.len;
583 f2fs_bug_on(sbi, pos >= org_end);
584
585 if (pos > dei.fofs && pos - dei.fofs >= F2FS_MIN_EXTENT_LEN) {
586 en->ei.len = pos - en->ei.fofs;
587 prev_en = en;
588 parts = 1;
589 }
590
591 if (end < org_end && org_end - end >= F2FS_MIN_EXTENT_LEN) {
592 if (parts) {
593 set_extent_info(&ei, end,
594 end - dei.fofs + dei.blk,
595 org_end - end);
596 en1 = __insert_extent_tree(sbi, et, &ei,
597 NULL, NULL, true);
598 next_en = en1;
599 } else {
600 en->ei.fofs = end;
601 en->ei.blk += end - dei.fofs;
602 en->ei.len -= end - dei.fofs;
603 next_en = en;
604 }
605 parts++;
606 }
607
608 if (!next_en) {
609 struct rb_node *node = rb_next(&en->rb_node);
610
611 next_en = rb_entry_safe(node, struct extent_node,
612 rb_node);
613 }
614
615 if (parts)
616 __try_update_largest_extent(et, en);
617 else
618 __release_extent_node(sbi, et, en);
619
620 /*
621 * if original extent is split into zero or two parts, extent
622 * tree has been altered by deletion or insertion, therefore
623 * invalidate pointers regard to tree.
624 */
625 if (parts != 1) {
626 insert_p = NULL;
627 insert_parent = NULL;
628 }
629 en = next_en;
630 }
631
632 /* 3. update extent in extent cache */
633 if (blkaddr) {
634
635 set_extent_info(&ei, fofs, blkaddr, len);
636 if (!__try_merge_extent_node(sbi, et, &ei, prev_en, next_en))
637 __insert_extent_tree(sbi, et, &ei,
638 insert_p, insert_parent, leftmost);
639
640 /* give up extent_cache, if split and small updates happen */
641 if (dei.len >= 1 &&
642 prev.len < F2FS_MIN_EXTENT_LEN &&
643 et->largest.len < F2FS_MIN_EXTENT_LEN) {
644 et->largest.len = 0;
645 et->largest_updated = true;
646 set_inode_flag(inode, FI_NO_EXTENT);
647 }
648 }
649
650 if (is_inode_flag_set(inode, FI_NO_EXTENT))
651 __free_extent_tree(sbi, et);
652
653 if (et->largest_updated) {
654 et->largest_updated = false;
655 updated = true;
656 }
657
658 write_unlock(&et->lock);
659
660 if (updated)
661 f2fs_mark_inode_dirty_sync(inode, true);
662}
663
664unsigned int f2fs_shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
665{
666 struct extent_tree *et, *next;
667 struct extent_node *en;
668 unsigned int node_cnt = 0, tree_cnt = 0;
669 int remained;
670
671 if (!test_opt(sbi, EXTENT_CACHE))
672 return 0;
673
674 if (!atomic_read(&sbi->total_zombie_tree))
675 goto free_node;
676
677 if (!mutex_trylock(&sbi->extent_tree_lock))
678 goto out;
679
680 /* 1. remove unreferenced extent tree */
681 list_for_each_entry_safe(et, next, &sbi->zombie_list, list) {
682 if (atomic_read(&et->node_cnt)) {
683 write_lock(&et->lock);
684 node_cnt += __free_extent_tree(sbi, et);
685 write_unlock(&et->lock);
686 }
687 f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
688 list_del_init(&et->list);
689 radix_tree_delete(&sbi->extent_tree_root, et->ino);
690 kmem_cache_free(extent_tree_slab, et);
691 atomic_dec(&sbi->total_ext_tree);
692 atomic_dec(&sbi->total_zombie_tree);
693 tree_cnt++;
694
695 if (node_cnt + tree_cnt >= nr_shrink)
696 goto unlock_out;
697 cond_resched();
698 }
699 mutex_unlock(&sbi->extent_tree_lock);
700
701free_node:
702 /* 2. remove LRU extent entries */
703 if (!mutex_trylock(&sbi->extent_tree_lock))
704 goto out;
705
706 remained = nr_shrink - (node_cnt + tree_cnt);
707
708 spin_lock(&sbi->extent_lock);
709 for (; remained > 0; remained--) {
710 if (list_empty(&sbi->extent_list))
711 break;
712 en = list_first_entry(&sbi->extent_list,
713 struct extent_node, list);
714 et = en->et;
715 if (!write_trylock(&et->lock)) {
716 /* refresh this extent node's position in extent list */
717 list_move_tail(&en->list, &sbi->extent_list);
718 continue;
719 }
720
721 list_del_init(&en->list);
722 spin_unlock(&sbi->extent_lock);
723
724 __detach_extent_node(sbi, et, en);
725
726 write_unlock(&et->lock);
727 node_cnt++;
728 spin_lock(&sbi->extent_lock);
729 }
730 spin_unlock(&sbi->extent_lock);
731
732unlock_out:
733 mutex_unlock(&sbi->extent_tree_lock);
734out:
735 trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt);
736
737 return node_cnt + tree_cnt;
738}
739
740unsigned int f2fs_destroy_extent_node(struct inode *inode)
741{
742 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
743 struct extent_tree *et = F2FS_I(inode)->extent_tree;
744 unsigned int node_cnt = 0;
745
746 if (!et || !atomic_read(&et->node_cnt))
747 return 0;
748
749 write_lock(&et->lock);
750 node_cnt = __free_extent_tree(sbi, et);
751 write_unlock(&et->lock);
752
753 return node_cnt;
754}
755
756void f2fs_drop_extent_tree(struct inode *inode)
757{
758 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
759 struct extent_tree *et = F2FS_I(inode)->extent_tree;
760 bool updated = false;
761
762 if (!f2fs_may_extent_tree(inode))
763 return;
764
765 set_inode_flag(inode, FI_NO_EXTENT);
766
767 write_lock(&et->lock);
768 __free_extent_tree(sbi, et);
769 if (et->largest.len) {
770 et->largest.len = 0;
771 updated = true;
772 }
773 write_unlock(&et->lock);
774 if (updated)
775 f2fs_mark_inode_dirty_sync(inode, true);
776}
777
778void f2fs_destroy_extent_tree(struct inode *inode)
779{
780 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
781 struct extent_tree *et = F2FS_I(inode)->extent_tree;
782 unsigned int node_cnt = 0;
783
784 if (!et)
785 return;
786
787 if (inode->i_nlink && !is_bad_inode(inode) &&
788 atomic_read(&et->node_cnt)) {
789 mutex_lock(&sbi->extent_tree_lock);
790 list_add_tail(&et->list, &sbi->zombie_list);
791 atomic_inc(&sbi->total_zombie_tree);
792 mutex_unlock(&sbi->extent_tree_lock);
793 return;
794 }
795
796 /* free all extent info belong to this extent tree */
797 node_cnt = f2fs_destroy_extent_node(inode);
798
799 /* delete extent tree entry in radix tree */
800 mutex_lock(&sbi->extent_tree_lock);
801 f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
802 radix_tree_delete(&sbi->extent_tree_root, inode->i_ino);
803 kmem_cache_free(extent_tree_slab, et);
804 atomic_dec(&sbi->total_ext_tree);
805 mutex_unlock(&sbi->extent_tree_lock);
806
807 F2FS_I(inode)->extent_tree = NULL;
808
809 trace_f2fs_destroy_extent_tree(inode, node_cnt);
810}
811
812bool f2fs_lookup_extent_cache(struct inode *inode, pgoff_t pgofs,
813 struct extent_info *ei)
814{
815 if (!f2fs_may_extent_tree(inode))
816 return false;
817
818 return f2fs_lookup_extent_tree(inode, pgofs, ei);
819}
820
821void f2fs_update_extent_cache(struct dnode_of_data *dn)
822{
823 pgoff_t fofs;
824 block_t blkaddr;
825
826 if (!f2fs_may_extent_tree(dn->inode))
827 return;
828
829 if (dn->data_blkaddr == NEW_ADDR)
830 blkaddr = NULL_ADDR;
831 else
832 blkaddr = dn->data_blkaddr;
833
834 fofs = f2fs_start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
835 dn->ofs_in_node;
836 f2fs_update_extent_tree_range(dn->inode, fofs, blkaddr, 1);
837}
838
839void f2fs_update_extent_cache_range(struct dnode_of_data *dn,
840 pgoff_t fofs, block_t blkaddr, unsigned int len)
841
842{
843 if (!f2fs_may_extent_tree(dn->inode))
844 return;
845
846 f2fs_update_extent_tree_range(dn->inode, fofs, blkaddr, len);
847}
848
849void f2fs_init_extent_cache_info(struct f2fs_sb_info *sbi)
850{
851 INIT_RADIX_TREE(&sbi->extent_tree_root, GFP_NOIO);
852 mutex_init(&sbi->extent_tree_lock);
853 INIT_LIST_HEAD(&sbi->extent_list);
854 spin_lock_init(&sbi->extent_lock);
855 atomic_set(&sbi->total_ext_tree, 0);
856 INIT_LIST_HEAD(&sbi->zombie_list);
857 atomic_set(&sbi->total_zombie_tree, 0);
858 atomic_set(&sbi->total_ext_node, 0);
859}
860
861int __init f2fs_create_extent_cache(void)
862{
863 extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree",
864 sizeof(struct extent_tree));
865 if (!extent_tree_slab)
866 return -ENOMEM;
867 extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node",
868 sizeof(struct extent_node));
869 if (!extent_node_slab) {
870 kmem_cache_destroy(extent_tree_slab);
871 return -ENOMEM;
872 }
873 return 0;
874}
875
876void f2fs_destroy_extent_cache(void)
877{
878 kmem_cache_destroy(extent_node_slab);
879 kmem_cache_destroy(extent_tree_slab);
880}