2 * fs/ext4/extents_status.c
4 * Written by Yongqiang Yang <xiaoqiangnk@gmail.com>
6 * Allison Henderson <achender@linux.vnet.ibm.com>
7 * Hugh Dickins <hughd@google.com>
8 * Zheng Liu <wenqing.lz@taobao.com>
10 * Ext4 extents status tree core functions.
12 #include <linux/rbtree.h>
14 #include "extents_status.h"
15 #include "ext4_extents.h"
17 #include <trace/events/ext4.h>
20 * According to previous discussion in Ext4 Developer Workshop, we
21 * will introduce a new structure called io tree to track all extent
22 * status in order to solve some problems that we have met
23 * (e.g. Reservation space warning), and provide extent-level locking.
24 * Delay extent tree is the first step to achieve this goal. It is
25 * original built by Yongqiang Yang. At that time it is called delay
26 * extent tree, whose goal is only track delayed extents in memory to
27 * simplify the implementation of fiemap and bigalloc, and introduce
28 * lseek SEEK_DATA/SEEK_HOLE support. That is why it is still called
29 * delay extent tree at the first commit. But for better understand
30 * what it does, it has been rename to extent status tree.
33 * Currently the first step has been done. All delayed extents are
34 * tracked in the tree. It maintains the delayed extent when a delayed
35 * allocation is issued, and the delayed extent is written out or
36 * invalidated. Therefore the implementation of fiemap and bigalloc
37 * are simplified, and SEEK_DATA/SEEK_HOLE are introduced.
39 * The following comment describes the implemenmtation of extent
40 * status tree and future works.
43 * In this step all extent status are tracked by extent status tree.
44 * Thus, we can first try to lookup a block mapping in this tree before
45 * finding it in extent tree. Hence, single extent cache can be removed
46 * because extent status tree can do a better job. Extents in status
47 * tree are loaded on-demand. Therefore, the extent status tree may not
48 * contain all of the extents in a file. Meanwhile we define a shrinker
49 * to reclaim memory from extent status tree because fragmented extent
50 * tree will make status tree cost too much memory. written/unwritten/-
51 * hole extents in the tree will be reclaimed by this shrinker when we
52 * are under high memory pressure. Delayed extents will not be
53 * reclimed because fiemap, bigalloc, and seek_data/hole need it.
57 * Extent status tree implementation for ext4.
60 * ==========================================================================
61 * Extent status tree tracks all extent status.
63 * 1. Why we need to implement extent status tree?
65 * Without extent status tree, ext4 identifies a delayed extent by looking
66 * up page cache, this has several deficiencies - complicated, buggy,
67 * and inefficient code.
69 * FIEMAP, SEEK_HOLE/DATA, bigalloc, and writeout all need to know if a
70 * block or a range of blocks are belonged to a delayed extent.
72 * Let us have a look at how they do without extent status tree.
74 * FIEMAP looks up page cache to identify delayed allocations from holes.
77 * SEEK_HOLE/DATA has the same problem as FIEMAP.
80 * bigalloc looks up page cache to figure out if a block is
81 * already under delayed allocation or not to determine whether
82 * quota reserving is needed for the cluster.
85 * Writeout looks up whole page cache to see if a buffer is
86 * mapped, If there are not very many delayed buffers, then it is
89 * With extent status tree implementation, FIEMAP, SEEK_HOLE/DATA,
90 * bigalloc and writeout can figure out if a block or a range of
91 * blocks is under delayed allocation(belonged to a delayed extent) or
92 * not by searching the extent tree.
95 * ==========================================================================
96 * 2. Ext4 extent status tree impelmentation
99 * A extent is a range of blocks which are contiguous logically and
100 * physically. Unlike extent in extent tree, this extent in ext4 is
101 * a in-memory struct, there is no corresponding on-disk data. There
102 * is no limit on length of extent, so an extent can contain as many
103 * blocks as they are contiguous logically and physically.
105 * -- extent status tree
106 * Every inode has an extent status tree and all allocation blocks
107 * are added to the tree with different status. The extent in the
108 * tree are ordered by logical block no.
110 * -- operations on a extent status tree
111 * There are three important operations on a delayed extent tree: find
112 * next extent, adding a extent(a range of blocks) and removing a extent.
114 * -- race on a extent status tree
115 * Extent status tree is protected by inode->i_es_lock.
117 * -- memory consumption
118 * Fragmented extent tree will make extent status tree cost too much
119 * memory. Hence, we will reclaim written/unwritten/hole extents from
120 * the tree under a heavy memory pressure.
123 * ==========================================================================
124 * 3. Performance analysis
127 * 1. There is a cache extent for write access, so if writes are
128 * not very random, adding space operaions are in O(1) time.
131 * 2. Code is much simpler, more readable, more maintainable and
135 * ==========================================================================
138 * -- Refactor delayed space reservation
140 * -- Extent-level locking
143 static struct kmem_cache
*ext4_es_cachep
;
145 static int __es_insert_extent(struct inode
*inode
, struct extent_status
*newes
);
146 static int __es_remove_extent(struct inode
*inode
, ext4_lblk_t lblk
,
148 static int __es_try_to_reclaim_extents(struct ext4_inode_info
*ei
,
151 int __init
ext4_init_es(void)
153 ext4_es_cachep
= kmem_cache_create("ext4_extent_status",
154 sizeof(struct extent_status
),
155 0, (SLAB_RECLAIM_ACCOUNT
), NULL
);
156 if (ext4_es_cachep
== NULL
)
161 void ext4_exit_es(void)
164 kmem_cache_destroy(ext4_es_cachep
);
167 void ext4_es_init_tree(struct ext4_es_tree
*tree
)
169 tree
->root
= RB_ROOT
;
170 tree
->cache_es
= NULL
;
174 static void ext4_es_print_tree(struct inode
*inode
)
176 struct ext4_es_tree
*tree
;
177 struct rb_node
*node
;
179 printk(KERN_DEBUG
"status extents for inode %lu:", inode
->i_ino
);
180 tree
= &EXT4_I(inode
)->i_es_tree
;
181 node
= rb_first(&tree
->root
);
183 struct extent_status
*es
;
184 es
= rb_entry(node
, struct extent_status
, rb_node
);
185 printk(KERN_DEBUG
" [%u/%u) %llu %llx",
186 es
->es_lblk
, es
->es_len
,
187 ext4_es_pblock(es
), ext4_es_status(es
));
188 node
= rb_next(node
);
190 printk(KERN_DEBUG
"\n");
193 #define ext4_es_print_tree(inode)
196 static inline ext4_lblk_t
ext4_es_end(struct extent_status
*es
)
198 BUG_ON(es
->es_lblk
+ es
->es_len
< es
->es_lblk
);
199 return es
->es_lblk
+ es
->es_len
- 1;
203 * search through the tree for an delayed extent with a given offset. If
204 * it can't be found, try to find next extent.
206 static struct extent_status
*__es_tree_search(struct rb_root
*root
,
209 struct rb_node
*node
= root
->rb_node
;
210 struct extent_status
*es
= NULL
;
213 es
= rb_entry(node
, struct extent_status
, rb_node
);
214 if (lblk
< es
->es_lblk
)
215 node
= node
->rb_left
;
216 else if (lblk
> ext4_es_end(es
))
217 node
= node
->rb_right
;
222 if (es
&& lblk
< es
->es_lblk
)
225 if (es
&& lblk
> ext4_es_end(es
)) {
226 node
= rb_next(&es
->rb_node
);
227 return node
? rb_entry(node
, struct extent_status
, rb_node
) :
235 * ext4_es_find_delayed_extent: find the 1st delayed extent covering @es->lblk
236 * if it exists, otherwise, the next extent after @es->lblk.
238 * @inode: the inode which owns delayed extents
239 * @lblk: the offset where we start to search
240 * @es: delayed extent that we found
242 void ext4_es_find_delayed_extent(struct inode
*inode
, ext4_lblk_t lblk
,
243 struct extent_status
*es
)
245 struct ext4_es_tree
*tree
= NULL
;
246 struct extent_status
*es1
= NULL
;
247 struct rb_node
*node
;
250 trace_ext4_es_find_delayed_extent_enter(inode
, lblk
);
252 read_lock(&EXT4_I(inode
)->i_es_lock
);
253 tree
= &EXT4_I(inode
)->i_es_tree
;
255 /* find extent in cache firstly */
256 es
->es_lblk
= es
->es_len
= es
->es_pblk
= 0;
257 if (tree
->cache_es
) {
258 es1
= tree
->cache_es
;
259 if (in_range(lblk
, es1
->es_lblk
, es1
->es_len
)) {
260 es_debug("%u cached by [%u/%u) %llu %llx\n",
261 lblk
, es1
->es_lblk
, es1
->es_len
,
262 ext4_es_pblock(es1
), ext4_es_status(es1
));
267 es1
= __es_tree_search(&tree
->root
, lblk
);
270 if (es1
&& !ext4_es_is_delayed(es1
)) {
271 while ((node
= rb_next(&es1
->rb_node
)) != NULL
) {
272 es1
= rb_entry(node
, struct extent_status
, rb_node
);
273 if (ext4_es_is_delayed(es1
))
278 if (es1
&& ext4_es_is_delayed(es1
)) {
279 tree
->cache_es
= es1
;
280 es
->es_lblk
= es1
->es_lblk
;
281 es
->es_len
= es1
->es_len
;
282 es
->es_pblk
= es1
->es_pblk
;
285 read_unlock(&EXT4_I(inode
)->i_es_lock
);
287 ext4_es_lru_add(inode
);
288 trace_ext4_es_find_delayed_extent_exit(inode
, es
);
291 static struct extent_status
*
292 ext4_es_alloc_extent(struct inode
*inode
, ext4_lblk_t lblk
, ext4_lblk_t len
,
295 struct extent_status
*es
;
296 es
= kmem_cache_alloc(ext4_es_cachep
, GFP_ATOMIC
);
304 * We don't count delayed extent because we never try to reclaim them
306 if (!ext4_es_is_delayed(es
)) {
307 EXT4_I(inode
)->i_es_lru_nr
++;
308 percpu_counter_inc(&EXT4_SB(inode
->i_sb
)->s_extent_cache_cnt
);
314 static void ext4_es_free_extent(struct inode
*inode
, struct extent_status
*es
)
316 /* Decrease the lru counter when this es is not delayed */
317 if (!ext4_es_is_delayed(es
)) {
318 BUG_ON(EXT4_I(inode
)->i_es_lru_nr
== 0);
319 EXT4_I(inode
)->i_es_lru_nr
--;
320 percpu_counter_dec(&EXT4_SB(inode
->i_sb
)->s_extent_cache_cnt
);
323 kmem_cache_free(ext4_es_cachep
, es
);
327 * Check whether or not two extents can be merged
329 * - logical block number is contiguous
330 * - physical block number is contiguous
333 static int ext4_es_can_be_merged(struct extent_status
*es1
,
334 struct extent_status
*es2
)
336 if (ext4_es_status(es1
) != ext4_es_status(es2
))
339 if (((__u64
) es1
->es_len
) + es2
->es_len
> 0xFFFFFFFFULL
)
342 if (((__u64
) es1
->es_lblk
) + es1
->es_len
!= es2
->es_lblk
)
345 if ((ext4_es_is_written(es1
) || ext4_es_is_unwritten(es1
)) &&
346 (ext4_es_pblock(es1
) + es1
->es_len
== ext4_es_pblock(es2
)))
349 if (ext4_es_is_hole(es1
))
352 /* we need to check delayed extent is without unwritten status */
353 if (ext4_es_is_delayed(es1
) && !ext4_es_is_unwritten(es1
))
359 static struct extent_status
*
360 ext4_es_try_to_merge_left(struct inode
*inode
, struct extent_status
*es
)
362 struct ext4_es_tree
*tree
= &EXT4_I(inode
)->i_es_tree
;
363 struct extent_status
*es1
;
364 struct rb_node
*node
;
366 node
= rb_prev(&es
->rb_node
);
370 es1
= rb_entry(node
, struct extent_status
, rb_node
);
371 if (ext4_es_can_be_merged(es1
, es
)) {
372 es1
->es_len
+= es
->es_len
;
373 rb_erase(&es
->rb_node
, &tree
->root
);
374 ext4_es_free_extent(inode
, es
);
381 static struct extent_status
*
382 ext4_es_try_to_merge_right(struct inode
*inode
, struct extent_status
*es
)
384 struct ext4_es_tree
*tree
= &EXT4_I(inode
)->i_es_tree
;
385 struct extent_status
*es1
;
386 struct rb_node
*node
;
388 node
= rb_next(&es
->rb_node
);
392 es1
= rb_entry(node
, struct extent_status
, rb_node
);
393 if (ext4_es_can_be_merged(es
, es1
)) {
394 es
->es_len
+= es1
->es_len
;
395 rb_erase(node
, &tree
->root
);
396 ext4_es_free_extent(inode
, es1
);
402 #ifdef ES_AGGRESSIVE_TEST
403 static void ext4_es_insert_extent_ext_check(struct inode
*inode
,
404 struct extent_status
*es
)
406 struct ext4_ext_path
*path
= NULL
;
407 struct ext4_extent
*ex
;
408 ext4_lblk_t ee_block
;
409 ext4_fsblk_t ee_start
;
410 unsigned short ee_len
;
411 int depth
, ee_status
, es_status
;
413 path
= ext4_ext_find_extent(inode
, es
->es_lblk
, NULL
);
417 depth
= ext_depth(inode
);
418 ex
= path
[depth
].p_ext
;
422 ee_block
= le32_to_cpu(ex
->ee_block
);
423 ee_start
= ext4_ext_pblock(ex
);
424 ee_len
= ext4_ext_get_actual_len(ex
);
426 ee_status
= ext4_ext_is_uninitialized(ex
) ? 1 : 0;
427 es_status
= ext4_es_is_unwritten(es
) ? 1 : 0;
430 * Make sure ex and es are not overlap when we try to insert
431 * a delayed/hole extent.
433 if (!ext4_es_is_written(es
) && !ext4_es_is_unwritten(es
)) {
434 if (in_range(es
->es_lblk
, ee_block
, ee_len
)) {
435 pr_warn("ES insert assertation failed for "
436 "inode: %lu we can find an extent "
437 "at block [%d/%d/%llu/%c], but we "
438 "want to add an delayed/hole extent "
439 "[%d/%d/%llu/%llx]\n",
440 inode
->i_ino
, ee_block
, ee_len
,
441 ee_start
, ee_status
? 'u' : 'w',
442 es
->es_lblk
, es
->es_len
,
443 ext4_es_pblock(es
), ext4_es_status(es
));
449 * We don't check ee_block == es->es_lblk, etc. because es
450 * might be a part of whole extent, vice versa.
452 if (es
->es_lblk
< ee_block
||
453 ext4_es_pblock(es
) != ee_start
+ es
->es_lblk
- ee_block
) {
454 pr_warn("ES insert assertation failed for inode: %lu "
455 "ex_status [%d/%d/%llu/%c] != "
456 "es_status [%d/%d/%llu/%c]\n", inode
->i_ino
,
457 ee_block
, ee_len
, ee_start
,
458 ee_status
? 'u' : 'w', es
->es_lblk
, es
->es_len
,
459 ext4_es_pblock(es
), es_status
? 'u' : 'w');
463 if (ee_status
^ es_status
) {
464 pr_warn("ES insert assertation failed for inode: %lu "
465 "ex_status [%d/%d/%llu/%c] != "
466 "es_status [%d/%d/%llu/%c]\n", inode
->i_ino
,
467 ee_block
, ee_len
, ee_start
,
468 ee_status
? 'u' : 'w', es
->es_lblk
, es
->es_len
,
469 ext4_es_pblock(es
), es_status
? 'u' : 'w');
473 * We can't find an extent on disk. So we need to make sure
474 * that we don't want to add an written/unwritten extent.
476 if (!ext4_es_is_delayed(es
) && !ext4_es_is_hole(es
)) {
477 pr_warn("ES insert assertation failed for inode: %lu "
478 "can't find an extent at block %d but we want "
479 "to add an written/unwritten extent "
480 "[%d/%d/%llu/%llx]\n", inode
->i_ino
,
481 es
->es_lblk
, es
->es_lblk
, es
->es_len
,
482 ext4_es_pblock(es
), ext4_es_status(es
));
487 ext4_ext_drop_refs(path
);
492 static void ext4_es_insert_extent_ind_check(struct inode
*inode
,
493 struct extent_status
*es
)
495 struct ext4_map_blocks map
;
499 * Here we call ext4_ind_map_blocks to lookup a block mapping because
500 * 'Indirect' structure is defined in indirect.c. So we couldn't
501 * access direct/indirect tree from outside. It is too dirty to define
502 * this function in indirect.c file.
505 map
.m_lblk
= es
->es_lblk
;
506 map
.m_len
= es
->es_len
;
508 retval
= ext4_ind_map_blocks(NULL
, inode
, &map
, 0);
510 if (ext4_es_is_delayed(es
) || ext4_es_is_hole(es
)) {
512 * We want to add a delayed/hole extent but this
513 * block has been allocated.
515 pr_warn("ES insert assertation failed for inode: %lu "
516 "We can find blocks but we want to add a "
517 "delayed/hole extent [%d/%d/%llu/%llx]\n",
518 inode
->i_ino
, es
->es_lblk
, es
->es_len
,
519 ext4_es_pblock(es
), ext4_es_status(es
));
521 } else if (ext4_es_is_written(es
)) {
522 if (retval
!= es
->es_len
) {
523 pr_warn("ES insert assertation failed for "
524 "inode: %lu retval %d != es_len %d\n",
525 inode
->i_ino
, retval
, es
->es_len
);
528 if (map
.m_pblk
!= ext4_es_pblock(es
)) {
529 pr_warn("ES insert assertation failed for "
530 "inode: %lu m_pblk %llu != "
532 inode
->i_ino
, map
.m_pblk
,
538 * We don't need to check unwritten extent because
539 * indirect-based file doesn't have it.
543 } else if (retval
== 0) {
544 if (ext4_es_is_written(es
)) {
545 pr_warn("ES insert assertation failed for inode: %lu "
546 "We can't find the block but we want to add "
547 "an written extent [%d/%d/%llu/%llx]\n",
548 inode
->i_ino
, es
->es_lblk
, es
->es_len
,
549 ext4_es_pblock(es
), ext4_es_status(es
));
555 static inline void ext4_es_insert_extent_check(struct inode
*inode
,
556 struct extent_status
*es
)
559 * We don't need to worry about the race condition because
560 * caller takes i_data_sem locking.
562 BUG_ON(!rwsem_is_locked(&EXT4_I(inode
)->i_data_sem
));
563 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
564 ext4_es_insert_extent_ext_check(inode
, es
);
566 ext4_es_insert_extent_ind_check(inode
, es
);
569 static inline void ext4_es_insert_extent_check(struct inode
*inode
,
570 struct extent_status
*es
)
575 static int __es_insert_extent(struct inode
*inode
, struct extent_status
*newes
)
577 struct ext4_es_tree
*tree
= &EXT4_I(inode
)->i_es_tree
;
578 struct rb_node
**p
= &tree
->root
.rb_node
;
579 struct rb_node
*parent
= NULL
;
580 struct extent_status
*es
;
584 es
= rb_entry(parent
, struct extent_status
, rb_node
);
586 if (newes
->es_lblk
< es
->es_lblk
) {
587 if (ext4_es_can_be_merged(newes
, es
)) {
589 * Here we can modify es_lblk directly
590 * because it isn't overlapped.
592 es
->es_lblk
= newes
->es_lblk
;
593 es
->es_len
+= newes
->es_len
;
594 if (ext4_es_is_written(es
) ||
595 ext4_es_is_unwritten(es
))
596 ext4_es_store_pblock(es
,
598 es
= ext4_es_try_to_merge_left(inode
, es
);
602 } else if (newes
->es_lblk
> ext4_es_end(es
)) {
603 if (ext4_es_can_be_merged(es
, newes
)) {
604 es
->es_len
+= newes
->es_len
;
605 es
= ext4_es_try_to_merge_right(inode
, es
);
615 es
= ext4_es_alloc_extent(inode
, newes
->es_lblk
, newes
->es_len
,
619 rb_link_node(&es
->rb_node
, parent
, p
);
620 rb_insert_color(&es
->rb_node
, &tree
->root
);
628 * ext4_es_insert_extent() adds a space to a extent status tree.
630 * ext4_es_insert_extent is called by ext4_da_write_begin and
631 * ext4_es_remove_extent.
633 * Return 0 on success, error code on failure.
635 int ext4_es_insert_extent(struct inode
*inode
, ext4_lblk_t lblk
,
636 ext4_lblk_t len
, ext4_fsblk_t pblk
,
637 unsigned long long status
)
639 struct extent_status newes
;
640 ext4_lblk_t end
= lblk
+ len
- 1;
643 es_debug("add [%u/%u) %llu %llx to extent status tree of inode %lu\n",
644 lblk
, len
, pblk
, status
, inode
->i_ino
);
651 newes
.es_lblk
= lblk
;
653 ext4_es_store_pblock(&newes
, pblk
);
654 ext4_es_store_status(&newes
, status
);
655 trace_ext4_es_insert_extent(inode
, &newes
);
657 ext4_es_insert_extent_check(inode
, &newes
);
659 write_lock(&EXT4_I(inode
)->i_es_lock
);
660 err
= __es_remove_extent(inode
, lblk
, end
);
663 err
= __es_insert_extent(inode
, &newes
);
666 write_unlock(&EXT4_I(inode
)->i_es_lock
);
668 ext4_es_lru_add(inode
);
669 ext4_es_print_tree(inode
);
675 * ext4_es_lookup_extent() looks up an extent in extent status tree.
677 * ext4_es_lookup_extent is called by ext4_map_blocks/ext4_da_map_blocks.
679 * Return: 1 on found, 0 on not
681 int ext4_es_lookup_extent(struct inode
*inode
, ext4_lblk_t lblk
,
682 struct extent_status
*es
)
684 struct ext4_es_tree
*tree
;
685 struct extent_status
*es1
= NULL
;
686 struct rb_node
*node
;
689 trace_ext4_es_lookup_extent_enter(inode
, lblk
);
690 es_debug("lookup extent in block %u\n", lblk
);
692 tree
= &EXT4_I(inode
)->i_es_tree
;
693 read_lock(&EXT4_I(inode
)->i_es_lock
);
695 /* find extent in cache firstly */
696 es
->es_lblk
= es
->es_len
= es
->es_pblk
= 0;
697 if (tree
->cache_es
) {
698 es1
= tree
->cache_es
;
699 if (in_range(lblk
, es1
->es_lblk
, es1
->es_len
)) {
700 es_debug("%u cached by [%u/%u)\n",
701 lblk
, es1
->es_lblk
, es1
->es_len
);
707 node
= tree
->root
.rb_node
;
709 es1
= rb_entry(node
, struct extent_status
, rb_node
);
710 if (lblk
< es1
->es_lblk
)
711 node
= node
->rb_left
;
712 else if (lblk
> ext4_es_end(es1
))
713 node
= node
->rb_right
;
723 es
->es_lblk
= es1
->es_lblk
;
724 es
->es_len
= es1
->es_len
;
725 es
->es_pblk
= es1
->es_pblk
;
728 read_unlock(&EXT4_I(inode
)->i_es_lock
);
730 ext4_es_lru_add(inode
);
731 trace_ext4_es_lookup_extent_exit(inode
, es
, found
);
735 static int __es_remove_extent(struct inode
*inode
, ext4_lblk_t lblk
,
738 struct ext4_es_tree
*tree
= &EXT4_I(inode
)->i_es_tree
;
739 struct rb_node
*node
;
740 struct extent_status
*es
;
741 struct extent_status orig_es
;
742 ext4_lblk_t len1
, len2
;
746 es
= __es_tree_search(&tree
->root
, lblk
);
749 if (es
->es_lblk
> end
)
752 /* Simply invalidate cache_es. */
753 tree
->cache_es
= NULL
;
755 orig_es
.es_lblk
= es
->es_lblk
;
756 orig_es
.es_len
= es
->es_len
;
757 orig_es
.es_pblk
= es
->es_pblk
;
759 len1
= lblk
> es
->es_lblk
? lblk
- es
->es_lblk
: 0;
760 len2
= ext4_es_end(es
) > end
? ext4_es_end(es
) - end
: 0;
765 struct extent_status newes
;
767 newes
.es_lblk
= end
+ 1;
769 if (ext4_es_is_written(&orig_es
) ||
770 ext4_es_is_unwritten(&orig_es
)) {
771 block
= ext4_es_pblock(&orig_es
) +
772 orig_es
.es_len
- len2
;
773 ext4_es_store_pblock(&newes
, block
);
775 ext4_es_store_status(&newes
, ext4_es_status(&orig_es
));
776 err
= __es_insert_extent(inode
, &newes
);
778 es
->es_lblk
= orig_es
.es_lblk
;
779 es
->es_len
= orig_es
.es_len
;
783 es
->es_lblk
= end
+ 1;
785 if (ext4_es_is_written(es
) ||
786 ext4_es_is_unwritten(es
)) {
787 block
= orig_es
.es_pblk
+ orig_es
.es_len
- len2
;
788 ext4_es_store_pblock(es
, block
);
795 node
= rb_next(&es
->rb_node
);
797 es
= rb_entry(node
, struct extent_status
, rb_node
);
802 while (es
&& ext4_es_end(es
) <= end
) {
803 node
= rb_next(&es
->rb_node
);
804 rb_erase(&es
->rb_node
, &tree
->root
);
805 ext4_es_free_extent(inode
, es
);
810 es
= rb_entry(node
, struct extent_status
, rb_node
);
813 if (es
&& es
->es_lblk
< end
+ 1) {
814 ext4_lblk_t orig_len
= es
->es_len
;
816 len1
= ext4_es_end(es
) - end
;
817 es
->es_lblk
= end
+ 1;
819 if (ext4_es_is_written(es
) || ext4_es_is_unwritten(es
)) {
820 block
= es
->es_pblk
+ orig_len
- len1
;
821 ext4_es_store_pblock(es
, block
);
830 * ext4_es_remove_extent() removes a space from a extent status tree.
832 * Return 0 on success, error code on failure.
834 int ext4_es_remove_extent(struct inode
*inode
, ext4_lblk_t lblk
,
840 trace_ext4_es_remove_extent(inode
, lblk
, len
);
841 es_debug("remove [%u/%u) from extent status tree of inode %lu\n",
842 lblk
, len
, inode
->i_ino
);
847 end
= lblk
+ len
- 1;
850 write_lock(&EXT4_I(inode
)->i_es_lock
);
851 err
= __es_remove_extent(inode
, lblk
, end
);
852 write_unlock(&EXT4_I(inode
)->i_es_lock
);
853 ext4_es_print_tree(inode
);
857 int ext4_es_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
859 ext4_lblk_t ee_block
;
860 ext4_fsblk_t ee_pblock
;
863 ee_block
= le32_to_cpu(ex
->ee_block
);
864 ee_len
= ext4_ext_get_actual_len(ex
);
865 ee_pblock
= ext4_ext_pblock(ex
);
870 return ext4_es_insert_extent(inode
, ee_block
, ee_len
, ee_pblock
,
871 EXTENT_STATUS_WRITTEN
);
874 static int ext4_es_shrink(struct shrinker
*shrink
, struct shrink_control
*sc
)
876 struct ext4_sb_info
*sbi
= container_of(shrink
,
877 struct ext4_sb_info
, s_es_shrinker
);
878 struct ext4_inode_info
*ei
;
879 struct list_head
*cur
, *tmp
, scanned
;
880 int nr_to_scan
= sc
->nr_to_scan
;
881 int ret
, nr_shrunk
= 0;
883 ret
= percpu_counter_read_positive(&sbi
->s_extent_cache_cnt
);
884 trace_ext4_es_shrink_enter(sbi
->s_sb
, nr_to_scan
, ret
);
889 INIT_LIST_HEAD(&scanned
);
891 spin_lock(&sbi
->s_es_lru_lock
);
892 list_for_each_safe(cur
, tmp
, &sbi
->s_es_lru
) {
893 list_move_tail(cur
, &scanned
);
895 ei
= list_entry(cur
, struct ext4_inode_info
, i_es_lru
);
897 read_lock(&ei
->i_es_lock
);
898 if (ei
->i_es_lru_nr
== 0) {
899 read_unlock(&ei
->i_es_lock
);
902 read_unlock(&ei
->i_es_lock
);
904 write_lock(&ei
->i_es_lock
);
905 ret
= __es_try_to_reclaim_extents(ei
, nr_to_scan
);
906 write_unlock(&ei
->i_es_lock
);
913 list_splice_tail(&scanned
, &sbi
->s_es_lru
);
914 spin_unlock(&sbi
->s_es_lru_lock
);
916 ret
= percpu_counter_read_positive(&sbi
->s_extent_cache_cnt
);
917 trace_ext4_es_shrink_exit(sbi
->s_sb
, nr_shrunk
, ret
);
921 void ext4_es_register_shrinker(struct super_block
*sb
)
923 struct ext4_sb_info
*sbi
;
926 INIT_LIST_HEAD(&sbi
->s_es_lru
);
927 spin_lock_init(&sbi
->s_es_lru_lock
);
928 sbi
->s_es_shrinker
.shrink
= ext4_es_shrink
;
929 sbi
->s_es_shrinker
.seeks
= DEFAULT_SEEKS
;
930 register_shrinker(&sbi
->s_es_shrinker
);
933 void ext4_es_unregister_shrinker(struct super_block
*sb
)
935 unregister_shrinker(&EXT4_SB(sb
)->s_es_shrinker
);
938 void ext4_es_lru_add(struct inode
*inode
)
940 struct ext4_inode_info
*ei
= EXT4_I(inode
);
941 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
943 spin_lock(&sbi
->s_es_lru_lock
);
944 if (list_empty(&ei
->i_es_lru
))
945 list_add_tail(&ei
->i_es_lru
, &sbi
->s_es_lru
);
947 list_move_tail(&ei
->i_es_lru
, &sbi
->s_es_lru
);
948 spin_unlock(&sbi
->s_es_lru_lock
);
951 void ext4_es_lru_del(struct inode
*inode
)
953 struct ext4_inode_info
*ei
= EXT4_I(inode
);
954 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
956 spin_lock(&sbi
->s_es_lru_lock
);
957 if (!list_empty(&ei
->i_es_lru
))
958 list_del_init(&ei
->i_es_lru
);
959 spin_unlock(&sbi
->s_es_lru_lock
);
962 static int __es_try_to_reclaim_extents(struct ext4_inode_info
*ei
,
965 struct inode
*inode
= &ei
->vfs_inode
;
966 struct ext4_es_tree
*tree
= &ei
->i_es_tree
;
967 struct rb_node
*node
;
968 struct extent_status
*es
;
971 if (ei
->i_es_lru_nr
== 0)
974 node
= rb_first(&tree
->root
);
975 while (node
!= NULL
) {
976 es
= rb_entry(node
, struct extent_status
, rb_node
);
977 node
= rb_next(&es
->rb_node
);
979 * We can't reclaim delayed extent from status tree because
980 * fiemap, bigallic, and seek_data/hole need to use it.
982 if (!ext4_es_is_delayed(es
)) {
983 rb_erase(&es
->rb_node
, &tree
->root
);
984 ext4_es_free_extent(inode
, es
);
986 if (--nr_to_scan
== 0)
990 tree
->cache_es
= NULL
;