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>
13 #include <linux/list_sort.h>
15 #include "extents_status.h"
16 #include "ext4_extents.h"
18 #include <trace/events/ext4.h>
21 * According to previous discussion in Ext4 Developer Workshop, we
22 * will introduce a new structure called io tree to track all extent
23 * status in order to solve some problems that we have met
24 * (e.g. Reservation space warning), and provide extent-level locking.
25 * Delay extent tree is the first step to achieve this goal. It is
26 * original built by Yongqiang Yang. At that time it is called delay
27 * extent tree, whose goal is only track delayed extents in memory to
28 * simplify the implementation of fiemap and bigalloc, and introduce
29 * lseek SEEK_DATA/SEEK_HOLE support. That is why it is still called
30 * delay extent tree at the first commit. But for better understand
31 * what it does, it has been rename to extent status tree.
34 * Currently the first step has been done. All delayed extents are
35 * tracked in the tree. It maintains the delayed extent when a delayed
36 * allocation is issued, and the delayed extent is written out or
37 * invalidated. Therefore the implementation of fiemap and bigalloc
38 * are simplified, and SEEK_DATA/SEEK_HOLE are introduced.
40 * The following comment describes the implemenmtation of extent
41 * status tree and future works.
44 * In this step all extent status are tracked by extent status tree.
45 * Thus, we can first try to lookup a block mapping in this tree before
46 * finding it in extent tree. Hence, single extent cache can be removed
47 * because extent status tree can do a better job. Extents in status
48 * tree are loaded on-demand. Therefore, the extent status tree may not
49 * contain all of the extents in a file. Meanwhile we define a shrinker
50 * to reclaim memory from extent status tree because fragmented extent
51 * tree will make status tree cost too much memory. written/unwritten/-
52 * hole extents in the tree will be reclaimed by this shrinker when we
53 * are under high memory pressure. Delayed extents will not be
54 * reclimed because fiemap, bigalloc, and seek_data/hole need it.
58 * Extent status tree implementation for ext4.
61 * ==========================================================================
62 * Extent status tree tracks all extent status.
64 * 1. Why we need to implement extent status tree?
66 * Without extent status tree, ext4 identifies a delayed extent by looking
67 * up page cache, this has several deficiencies - complicated, buggy,
68 * and inefficient code.
70 * FIEMAP, SEEK_HOLE/DATA, bigalloc, and writeout all need to know if a
71 * block or a range of blocks are belonged to a delayed extent.
73 * Let us have a look at how they do without extent status tree.
75 * FIEMAP looks up page cache to identify delayed allocations from holes.
78 * SEEK_HOLE/DATA has the same problem as FIEMAP.
81 * bigalloc looks up page cache to figure out if a block is
82 * already under delayed allocation or not to determine whether
83 * quota reserving is needed for the cluster.
86 * Writeout looks up whole page cache to see if a buffer is
87 * mapped, If there are not very many delayed buffers, then it is
90 * With extent status tree implementation, FIEMAP, SEEK_HOLE/DATA,
91 * bigalloc and writeout can figure out if a block or a range of
92 * blocks is under delayed allocation(belonged to a delayed extent) or
93 * not by searching the extent tree.
96 * ==========================================================================
97 * 2. Ext4 extent status tree impelmentation
100 * A extent is a range of blocks which are contiguous logically and
101 * physically. Unlike extent in extent tree, this extent in ext4 is
102 * a in-memory struct, there is no corresponding on-disk data. There
103 * is no limit on length of extent, so an extent can contain as many
104 * blocks as they are contiguous logically and physically.
106 * -- extent status tree
107 * Every inode has an extent status tree and all allocation blocks
108 * are added to the tree with different status. The extent in the
109 * tree are ordered by logical block no.
111 * -- operations on a extent status tree
112 * There are three important operations on a delayed extent tree: find
113 * next extent, adding a extent(a range of blocks) and removing a extent.
115 * -- race on a extent status tree
116 * Extent status tree is protected by inode->i_es_lock.
118 * -- memory consumption
119 * Fragmented extent tree will make extent status tree cost too much
120 * memory. Hence, we will reclaim written/unwritten/hole extents from
121 * the tree under a heavy memory pressure.
124 * ==========================================================================
125 * 3. Performance analysis
128 * 1. There is a cache extent for write access, so if writes are
129 * not very random, adding space operaions are in O(1) time.
132 * 2. Code is much simpler, more readable, more maintainable and
136 * ==========================================================================
139 * -- Refactor delayed space reservation
141 * -- Extent-level locking
144 static struct kmem_cache
*ext4_es_cachep
;
146 static int __es_insert_extent(struct inode
*inode
, struct extent_status
*newes
);
147 static int __es_remove_extent(struct inode
*inode
, ext4_lblk_t lblk
,
149 static int __es_try_to_reclaim_extents(struct ext4_inode_info
*ei
,
152 int __init
ext4_init_es(void)
154 ext4_es_cachep
= kmem_cache_create("ext4_extent_status",
155 sizeof(struct extent_status
),
156 0, (SLAB_RECLAIM_ACCOUNT
), NULL
);
157 if (ext4_es_cachep
== NULL
)
162 void ext4_exit_es(void)
165 kmem_cache_destroy(ext4_es_cachep
);
168 void ext4_es_init_tree(struct ext4_es_tree
*tree
)
170 tree
->root
= RB_ROOT
;
171 tree
->cache_es
= NULL
;
175 static void ext4_es_print_tree(struct inode
*inode
)
177 struct ext4_es_tree
*tree
;
178 struct rb_node
*node
;
180 printk(KERN_DEBUG
"status extents for inode %lu:", inode
->i_ino
);
181 tree
= &EXT4_I(inode
)->i_es_tree
;
182 node
= rb_first(&tree
->root
);
184 struct extent_status
*es
;
185 es
= rb_entry(node
, struct extent_status
, rb_node
);
186 printk(KERN_DEBUG
" [%u/%u) %llu %llx",
187 es
->es_lblk
, es
->es_len
,
188 ext4_es_pblock(es
), ext4_es_status(es
));
189 node
= rb_next(node
);
191 printk(KERN_DEBUG
"\n");
194 #define ext4_es_print_tree(inode)
197 static inline ext4_lblk_t
ext4_es_end(struct extent_status
*es
)
199 BUG_ON(es
->es_lblk
+ es
->es_len
< es
->es_lblk
);
200 return es
->es_lblk
+ es
->es_len
- 1;
204 * search through the tree for an delayed extent with a given offset. If
205 * it can't be found, try to find next extent.
207 static struct extent_status
*__es_tree_search(struct rb_root
*root
,
210 struct rb_node
*node
= root
->rb_node
;
211 struct extent_status
*es
= NULL
;
214 es
= rb_entry(node
, struct extent_status
, rb_node
);
215 if (lblk
< es
->es_lblk
)
216 node
= node
->rb_left
;
217 else if (lblk
> ext4_es_end(es
))
218 node
= node
->rb_right
;
223 if (es
&& lblk
< es
->es_lblk
)
226 if (es
&& lblk
> ext4_es_end(es
)) {
227 node
= rb_next(&es
->rb_node
);
228 return node
? rb_entry(node
, struct extent_status
, rb_node
) :
236 * ext4_es_find_delayed_extent_range: find the 1st delayed extent covering
237 * @es->lblk if it exists, otherwise, the next extent after @es->lblk.
239 * @inode: the inode which owns delayed extents
240 * @lblk: the offset where we start to search
241 * @end: the offset where we stop to search
242 * @es: delayed extent that we found
244 void ext4_es_find_delayed_extent_range(struct inode
*inode
,
245 ext4_lblk_t lblk
, ext4_lblk_t end
,
246 struct extent_status
*es
)
248 struct ext4_es_tree
*tree
= NULL
;
249 struct extent_status
*es1
= NULL
;
250 struct rb_node
*node
;
254 trace_ext4_es_find_delayed_extent_range_enter(inode
, lblk
);
256 read_lock(&EXT4_I(inode
)->i_es_lock
);
257 tree
= &EXT4_I(inode
)->i_es_tree
;
259 /* find extent in cache firstly */
260 es
->es_lblk
= es
->es_len
= es
->es_pblk
= 0;
261 if (tree
->cache_es
) {
262 es1
= tree
->cache_es
;
263 if (in_range(lblk
, es1
->es_lblk
, es1
->es_len
)) {
264 es_debug("%u cached by [%u/%u) %llu %llx\n",
265 lblk
, es1
->es_lblk
, es1
->es_len
,
266 ext4_es_pblock(es1
), ext4_es_status(es1
));
271 es1
= __es_tree_search(&tree
->root
, lblk
);
274 if (es1
&& !ext4_es_is_delayed(es1
)) {
275 while ((node
= rb_next(&es1
->rb_node
)) != NULL
) {
276 es1
= rb_entry(node
, struct extent_status
, rb_node
);
277 if (es1
->es_lblk
> end
) {
281 if (ext4_es_is_delayed(es1
))
286 if (es1
&& ext4_es_is_delayed(es1
)) {
287 tree
->cache_es
= es1
;
288 es
->es_lblk
= es1
->es_lblk
;
289 es
->es_len
= es1
->es_len
;
290 es
->es_pblk
= es1
->es_pblk
;
293 read_unlock(&EXT4_I(inode
)->i_es_lock
);
295 trace_ext4_es_find_delayed_extent_range_exit(inode
, es
);
298 static struct extent_status
*
299 ext4_es_alloc_extent(struct inode
*inode
, ext4_lblk_t lblk
, ext4_lblk_t len
,
302 struct extent_status
*es
;
303 es
= kmem_cache_alloc(ext4_es_cachep
, GFP_ATOMIC
);
311 * We don't count delayed extent because we never try to reclaim them
313 if (!ext4_es_is_delayed(es
)) {
314 EXT4_I(inode
)->i_es_lru_nr
++;
315 percpu_counter_inc(&EXT4_SB(inode
->i_sb
)->s_extent_cache_cnt
);
321 static void ext4_es_free_extent(struct inode
*inode
, struct extent_status
*es
)
323 /* Decrease the lru counter when this es is not delayed */
324 if (!ext4_es_is_delayed(es
)) {
325 BUG_ON(EXT4_I(inode
)->i_es_lru_nr
== 0);
326 EXT4_I(inode
)->i_es_lru_nr
--;
327 percpu_counter_dec(&EXT4_SB(inode
->i_sb
)->s_extent_cache_cnt
);
330 kmem_cache_free(ext4_es_cachep
, es
);
334 * Check whether or not two extents can be merged
336 * - logical block number is contiguous
337 * - physical block number is contiguous
340 static int ext4_es_can_be_merged(struct extent_status
*es1
,
341 struct extent_status
*es2
)
343 if (ext4_es_status(es1
) != ext4_es_status(es2
))
346 if (((__u64
) es1
->es_len
) + es2
->es_len
> 0xFFFFFFFFULL
)
349 if (((__u64
) es1
->es_lblk
) + es1
->es_len
!= es2
->es_lblk
)
352 if ((ext4_es_is_written(es1
) || ext4_es_is_unwritten(es1
)) &&
353 (ext4_es_pblock(es1
) + es1
->es_len
== ext4_es_pblock(es2
)))
356 if (ext4_es_is_hole(es1
))
359 /* we need to check delayed extent is without unwritten status */
360 if (ext4_es_is_delayed(es1
) && !ext4_es_is_unwritten(es1
))
366 static struct extent_status
*
367 ext4_es_try_to_merge_left(struct inode
*inode
, struct extent_status
*es
)
369 struct ext4_es_tree
*tree
= &EXT4_I(inode
)->i_es_tree
;
370 struct extent_status
*es1
;
371 struct rb_node
*node
;
373 node
= rb_prev(&es
->rb_node
);
377 es1
= rb_entry(node
, struct extent_status
, rb_node
);
378 if (ext4_es_can_be_merged(es1
, es
)) {
379 es1
->es_len
+= es
->es_len
;
380 rb_erase(&es
->rb_node
, &tree
->root
);
381 ext4_es_free_extent(inode
, es
);
388 static struct extent_status
*
389 ext4_es_try_to_merge_right(struct inode
*inode
, struct extent_status
*es
)
391 struct ext4_es_tree
*tree
= &EXT4_I(inode
)->i_es_tree
;
392 struct extent_status
*es1
;
393 struct rb_node
*node
;
395 node
= rb_next(&es
->rb_node
);
399 es1
= rb_entry(node
, struct extent_status
, rb_node
);
400 if (ext4_es_can_be_merged(es
, es1
)) {
401 es
->es_len
+= es1
->es_len
;
402 rb_erase(node
, &tree
->root
);
403 ext4_es_free_extent(inode
, es1
);
409 #ifdef ES_AGGRESSIVE_TEST
410 static void ext4_es_insert_extent_ext_check(struct inode
*inode
,
411 struct extent_status
*es
)
413 struct ext4_ext_path
*path
= NULL
;
414 struct ext4_extent
*ex
;
415 ext4_lblk_t ee_block
;
416 ext4_fsblk_t ee_start
;
417 unsigned short ee_len
;
418 int depth
, ee_status
, es_status
;
420 path
= ext4_ext_find_extent(inode
, es
->es_lblk
, NULL
);
424 depth
= ext_depth(inode
);
425 ex
= path
[depth
].p_ext
;
429 ee_block
= le32_to_cpu(ex
->ee_block
);
430 ee_start
= ext4_ext_pblock(ex
);
431 ee_len
= ext4_ext_get_actual_len(ex
);
433 ee_status
= ext4_ext_is_uninitialized(ex
) ? 1 : 0;
434 es_status
= ext4_es_is_unwritten(es
) ? 1 : 0;
437 * Make sure ex and es are not overlap when we try to insert
438 * a delayed/hole extent.
440 if (!ext4_es_is_written(es
) && !ext4_es_is_unwritten(es
)) {
441 if (in_range(es
->es_lblk
, ee_block
, ee_len
)) {
442 pr_warn("ES insert assertation failed for "
443 "inode: %lu we can find an extent "
444 "at block [%d/%d/%llu/%c], but we "
445 "want to add an delayed/hole extent "
446 "[%d/%d/%llu/%llx]\n",
447 inode
->i_ino
, ee_block
, ee_len
,
448 ee_start
, ee_status
? 'u' : 'w',
449 es
->es_lblk
, es
->es_len
,
450 ext4_es_pblock(es
), ext4_es_status(es
));
456 * We don't check ee_block == es->es_lblk, etc. because es
457 * might be a part of whole extent, vice versa.
459 if (es
->es_lblk
< ee_block
||
460 ext4_es_pblock(es
) != ee_start
+ es
->es_lblk
- ee_block
) {
461 pr_warn("ES insert assertation failed for inode: %lu "
462 "ex_status [%d/%d/%llu/%c] != "
463 "es_status [%d/%d/%llu/%c]\n", inode
->i_ino
,
464 ee_block
, ee_len
, ee_start
,
465 ee_status
? 'u' : 'w', es
->es_lblk
, es
->es_len
,
466 ext4_es_pblock(es
), es_status
? 'u' : 'w');
470 if (ee_status
^ es_status
) {
471 pr_warn("ES insert assertation failed for inode: %lu "
472 "ex_status [%d/%d/%llu/%c] != "
473 "es_status [%d/%d/%llu/%c]\n", inode
->i_ino
,
474 ee_block
, ee_len
, ee_start
,
475 ee_status
? 'u' : 'w', es
->es_lblk
, es
->es_len
,
476 ext4_es_pblock(es
), es_status
? 'u' : 'w');
480 * We can't find an extent on disk. So we need to make sure
481 * that we don't want to add an written/unwritten extent.
483 if (!ext4_es_is_delayed(es
) && !ext4_es_is_hole(es
)) {
484 pr_warn("ES insert assertation failed for inode: %lu "
485 "can't find an extent at block %d but we want "
486 "to add an written/unwritten extent "
487 "[%d/%d/%llu/%llx]\n", inode
->i_ino
,
488 es
->es_lblk
, es
->es_lblk
, es
->es_len
,
489 ext4_es_pblock(es
), ext4_es_status(es
));
494 ext4_ext_drop_refs(path
);
499 static void ext4_es_insert_extent_ind_check(struct inode
*inode
,
500 struct extent_status
*es
)
502 struct ext4_map_blocks map
;
506 * Here we call ext4_ind_map_blocks to lookup a block mapping because
507 * 'Indirect' structure is defined in indirect.c. So we couldn't
508 * access direct/indirect tree from outside. It is too dirty to define
509 * this function in indirect.c file.
512 map
.m_lblk
= es
->es_lblk
;
513 map
.m_len
= es
->es_len
;
515 retval
= ext4_ind_map_blocks(NULL
, inode
, &map
, 0);
517 if (ext4_es_is_delayed(es
) || ext4_es_is_hole(es
)) {
519 * We want to add a delayed/hole extent but this
520 * block has been allocated.
522 pr_warn("ES insert assertation failed for inode: %lu "
523 "We can find blocks but we want to add a "
524 "delayed/hole extent [%d/%d/%llu/%llx]\n",
525 inode
->i_ino
, es
->es_lblk
, es
->es_len
,
526 ext4_es_pblock(es
), ext4_es_status(es
));
528 } else if (ext4_es_is_written(es
)) {
529 if (retval
!= es
->es_len
) {
530 pr_warn("ES insert assertation failed for "
531 "inode: %lu retval %d != es_len %d\n",
532 inode
->i_ino
, retval
, es
->es_len
);
535 if (map
.m_pblk
!= ext4_es_pblock(es
)) {
536 pr_warn("ES insert assertation failed for "
537 "inode: %lu m_pblk %llu != "
539 inode
->i_ino
, map
.m_pblk
,
545 * We don't need to check unwritten extent because
546 * indirect-based file doesn't have it.
550 } else if (retval
== 0) {
551 if (ext4_es_is_written(es
)) {
552 pr_warn("ES insert assertation failed for inode: %lu "
553 "We can't find the block but we want to add "
554 "an written extent [%d/%d/%llu/%llx]\n",
555 inode
->i_ino
, es
->es_lblk
, es
->es_len
,
556 ext4_es_pblock(es
), ext4_es_status(es
));
562 static inline void ext4_es_insert_extent_check(struct inode
*inode
,
563 struct extent_status
*es
)
566 * We don't need to worry about the race condition because
567 * caller takes i_data_sem locking.
569 BUG_ON(!rwsem_is_locked(&EXT4_I(inode
)->i_data_sem
));
570 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
571 ext4_es_insert_extent_ext_check(inode
, es
);
573 ext4_es_insert_extent_ind_check(inode
, es
);
576 static inline void ext4_es_insert_extent_check(struct inode
*inode
,
577 struct extent_status
*es
)
582 static int __es_insert_extent(struct inode
*inode
, struct extent_status
*newes
)
584 struct ext4_es_tree
*tree
= &EXT4_I(inode
)->i_es_tree
;
585 struct rb_node
**p
= &tree
->root
.rb_node
;
586 struct rb_node
*parent
= NULL
;
587 struct extent_status
*es
;
591 es
= rb_entry(parent
, struct extent_status
, rb_node
);
593 if (newes
->es_lblk
< es
->es_lblk
) {
594 if (ext4_es_can_be_merged(newes
, es
)) {
596 * Here we can modify es_lblk directly
597 * because it isn't overlapped.
599 es
->es_lblk
= newes
->es_lblk
;
600 es
->es_len
+= newes
->es_len
;
601 if (ext4_es_is_written(es
) ||
602 ext4_es_is_unwritten(es
))
603 ext4_es_store_pblock(es
,
605 es
= ext4_es_try_to_merge_left(inode
, es
);
609 } else if (newes
->es_lblk
> ext4_es_end(es
)) {
610 if (ext4_es_can_be_merged(es
, newes
)) {
611 es
->es_len
+= newes
->es_len
;
612 es
= ext4_es_try_to_merge_right(inode
, es
);
622 es
= ext4_es_alloc_extent(inode
, newes
->es_lblk
, newes
->es_len
,
626 rb_link_node(&es
->rb_node
, parent
, p
);
627 rb_insert_color(&es
->rb_node
, &tree
->root
);
635 * ext4_es_insert_extent() adds a space to a extent status tree.
637 * ext4_es_insert_extent is called by ext4_da_write_begin and
638 * ext4_es_remove_extent.
640 * Return 0 on success, error code on failure.
642 int ext4_es_insert_extent(struct inode
*inode
, ext4_lblk_t lblk
,
643 ext4_lblk_t len
, ext4_fsblk_t pblk
,
644 unsigned long long status
)
646 struct extent_status newes
;
647 ext4_lblk_t end
= lblk
+ len
- 1;
650 es_debug("add [%u/%u) %llu %llx to extent status tree of inode %lu\n",
651 lblk
, len
, pblk
, status
, inode
->i_ino
);
658 newes
.es_lblk
= lblk
;
660 ext4_es_store_pblock(&newes
, pblk
);
661 ext4_es_store_status(&newes
, status
);
662 trace_ext4_es_insert_extent(inode
, &newes
);
664 ext4_es_insert_extent_check(inode
, &newes
);
666 write_lock(&EXT4_I(inode
)->i_es_lock
);
667 err
= __es_remove_extent(inode
, lblk
, end
);
670 err
= __es_insert_extent(inode
, &newes
);
673 write_unlock(&EXT4_I(inode
)->i_es_lock
);
675 ext4_es_print_tree(inode
);
681 * ext4_es_lookup_extent() looks up an extent in extent status tree.
683 * ext4_es_lookup_extent is called by ext4_map_blocks/ext4_da_map_blocks.
685 * Return: 1 on found, 0 on not
687 int ext4_es_lookup_extent(struct inode
*inode
, ext4_lblk_t lblk
,
688 struct extent_status
*es
)
690 struct ext4_es_tree
*tree
;
691 struct extent_status
*es1
= NULL
;
692 struct rb_node
*node
;
695 trace_ext4_es_lookup_extent_enter(inode
, lblk
);
696 es_debug("lookup extent in block %u\n", lblk
);
698 tree
= &EXT4_I(inode
)->i_es_tree
;
699 read_lock(&EXT4_I(inode
)->i_es_lock
);
701 /* find extent in cache firstly */
702 es
->es_lblk
= es
->es_len
= es
->es_pblk
= 0;
703 if (tree
->cache_es
) {
704 es1
= tree
->cache_es
;
705 if (in_range(lblk
, es1
->es_lblk
, es1
->es_len
)) {
706 es_debug("%u cached by [%u/%u)\n",
707 lblk
, es1
->es_lblk
, es1
->es_len
);
713 node
= tree
->root
.rb_node
;
715 es1
= rb_entry(node
, struct extent_status
, rb_node
);
716 if (lblk
< es1
->es_lblk
)
717 node
= node
->rb_left
;
718 else if (lblk
> ext4_es_end(es1
))
719 node
= node
->rb_right
;
729 es
->es_lblk
= es1
->es_lblk
;
730 es
->es_len
= es1
->es_len
;
731 es
->es_pblk
= es1
->es_pblk
;
734 read_unlock(&EXT4_I(inode
)->i_es_lock
);
736 trace_ext4_es_lookup_extent_exit(inode
, es
, found
);
740 static int __es_remove_extent(struct inode
*inode
, ext4_lblk_t lblk
,
743 struct ext4_es_tree
*tree
= &EXT4_I(inode
)->i_es_tree
;
744 struct rb_node
*node
;
745 struct extent_status
*es
;
746 struct extent_status orig_es
;
747 ext4_lblk_t len1
, len2
;
751 es
= __es_tree_search(&tree
->root
, lblk
);
754 if (es
->es_lblk
> end
)
757 /* Simply invalidate cache_es. */
758 tree
->cache_es
= NULL
;
760 orig_es
.es_lblk
= es
->es_lblk
;
761 orig_es
.es_len
= es
->es_len
;
762 orig_es
.es_pblk
= es
->es_pblk
;
764 len1
= lblk
> es
->es_lblk
? lblk
- es
->es_lblk
: 0;
765 len2
= ext4_es_end(es
) > end
? ext4_es_end(es
) - end
: 0;
770 struct extent_status newes
;
772 newes
.es_lblk
= end
+ 1;
774 if (ext4_es_is_written(&orig_es
) ||
775 ext4_es_is_unwritten(&orig_es
)) {
776 block
= ext4_es_pblock(&orig_es
) +
777 orig_es
.es_len
- len2
;
778 ext4_es_store_pblock(&newes
, block
);
780 ext4_es_store_status(&newes
, ext4_es_status(&orig_es
));
781 err
= __es_insert_extent(inode
, &newes
);
783 es
->es_lblk
= orig_es
.es_lblk
;
784 es
->es_len
= orig_es
.es_len
;
788 es
->es_lblk
= end
+ 1;
790 if (ext4_es_is_written(es
) ||
791 ext4_es_is_unwritten(es
)) {
792 block
= orig_es
.es_pblk
+ orig_es
.es_len
- len2
;
793 ext4_es_store_pblock(es
, block
);
800 node
= rb_next(&es
->rb_node
);
802 es
= rb_entry(node
, struct extent_status
, rb_node
);
807 while (es
&& ext4_es_end(es
) <= end
) {
808 node
= rb_next(&es
->rb_node
);
809 rb_erase(&es
->rb_node
, &tree
->root
);
810 ext4_es_free_extent(inode
, es
);
815 es
= rb_entry(node
, struct extent_status
, rb_node
);
818 if (es
&& es
->es_lblk
< end
+ 1) {
819 ext4_lblk_t orig_len
= es
->es_len
;
821 len1
= ext4_es_end(es
) - end
;
822 es
->es_lblk
= end
+ 1;
824 if (ext4_es_is_written(es
) || ext4_es_is_unwritten(es
)) {
825 block
= es
->es_pblk
+ orig_len
- len1
;
826 ext4_es_store_pblock(es
, block
);
835 * ext4_es_remove_extent() removes a space from a extent status tree.
837 * Return 0 on success, error code on failure.
839 int ext4_es_remove_extent(struct inode
*inode
, ext4_lblk_t lblk
,
845 trace_ext4_es_remove_extent(inode
, lblk
, len
);
846 es_debug("remove [%u/%u) from extent status tree of inode %lu\n",
847 lblk
, len
, inode
->i_ino
);
852 end
= lblk
+ len
- 1;
855 write_lock(&EXT4_I(inode
)->i_es_lock
);
856 err
= __es_remove_extent(inode
, lblk
, end
);
857 write_unlock(&EXT4_I(inode
)->i_es_lock
);
858 ext4_es_print_tree(inode
);
862 int ext4_es_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
864 ext4_lblk_t ee_block
;
865 ext4_fsblk_t ee_pblock
;
868 ee_block
= le32_to_cpu(ex
->ee_block
);
869 ee_len
= ext4_ext_get_actual_len(ex
);
870 ee_pblock
= ext4_ext_pblock(ex
);
875 return ext4_es_insert_extent(inode
, ee_block
, ee_len
, ee_pblock
,
876 EXTENT_STATUS_WRITTEN
);
879 static int ext4_inode_touch_time_cmp(void *priv
, struct list_head
*a
,
882 struct ext4_inode_info
*eia
, *eib
;
883 eia
= list_entry(a
, struct ext4_inode_info
, i_es_lru
);
884 eib
= list_entry(b
, struct ext4_inode_info
, i_es_lru
);
886 if (eia
->i_touch_when
== eib
->i_touch_when
)
888 if (time_after(eia
->i_touch_when
, eib
->i_touch_when
))
894 static int ext4_es_shrink(struct shrinker
*shrink
, struct shrink_control
*sc
)
896 struct ext4_sb_info
*sbi
= container_of(shrink
,
897 struct ext4_sb_info
, s_es_shrinker
);
898 struct ext4_inode_info
*ei
;
899 struct list_head
*cur
, *tmp
;
901 int nr_to_scan
= sc
->nr_to_scan
;
902 int ret
, nr_shrunk
= 0;
904 ret
= percpu_counter_read_positive(&sbi
->s_extent_cache_cnt
);
905 trace_ext4_es_shrink_enter(sbi
->s_sb
, nr_to_scan
, ret
);
910 spin_lock(&sbi
->s_es_lru_lock
);
913 * If the inode that is at the head of LRU list is newer than
914 * last_sorted time, that means that we need to sort this list.
916 ei
= list_first_entry(&sbi
->s_es_lru
, struct ext4_inode_info
, i_es_lru
);
917 if (sbi
->s_es_last_sorted
< ei
->i_touch_when
) {
918 list_sort(NULL
, &sbi
->s_es_lru
, ext4_inode_touch_time_cmp
);
919 sbi
->s_es_last_sorted
= jiffies
;
922 list_for_each_safe(cur
, tmp
, &sbi
->s_es_lru
) {
924 * If we have already reclaimed all extents from extent
925 * status tree, just stop the loop immediately.
927 if (percpu_counter_read_positive(&sbi
->s_extent_cache_cnt
) == 0)
930 ei
= list_entry(cur
, struct ext4_inode_info
, i_es_lru
);
932 /* Skip the inode that is newer than the last_sorted time */
933 if (sbi
->s_es_last_sorted
< ei
->i_touch_when
) {
934 list_move_tail(cur
, &skiped
);
938 if (ei
->i_es_lru_nr
== 0)
941 write_lock(&ei
->i_es_lock
);
942 ret
= __es_try_to_reclaim_extents(ei
, nr_to_scan
);
943 if (ei
->i_es_lru_nr
== 0)
944 list_del_init(&ei
->i_es_lru
);
945 write_unlock(&ei
->i_es_lock
);
953 /* Move the newer inodes into the tail of the LRU list. */
954 list_splice_tail(&skiped
, &sbi
->s_es_lru
);
955 spin_unlock(&sbi
->s_es_lru_lock
);
957 ret
= percpu_counter_read_positive(&sbi
->s_extent_cache_cnt
);
958 trace_ext4_es_shrink_exit(sbi
->s_sb
, nr_shrunk
, ret
);
962 void ext4_es_register_shrinker(struct ext4_sb_info
*sbi
)
964 INIT_LIST_HEAD(&sbi
->s_es_lru
);
965 spin_lock_init(&sbi
->s_es_lru_lock
);
966 sbi
->s_es_last_sorted
= 0;
967 sbi
->s_es_shrinker
.shrink
= ext4_es_shrink
;
968 sbi
->s_es_shrinker
.seeks
= DEFAULT_SEEKS
;
969 register_shrinker(&sbi
->s_es_shrinker
);
972 void ext4_es_unregister_shrinker(struct ext4_sb_info
*sbi
)
974 unregister_shrinker(&sbi
->s_es_shrinker
);
977 void ext4_es_lru_add(struct inode
*inode
)
979 struct ext4_inode_info
*ei
= EXT4_I(inode
);
980 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
982 ei
->i_touch_when
= jiffies
;
984 if (!list_empty(&ei
->i_es_lru
))
987 spin_lock(&sbi
->s_es_lru_lock
);
988 if (list_empty(&ei
->i_es_lru
))
989 list_add_tail(&ei
->i_es_lru
, &sbi
->s_es_lru
);
990 spin_unlock(&sbi
->s_es_lru_lock
);
993 void ext4_es_lru_del(struct inode
*inode
)
995 struct ext4_inode_info
*ei
= EXT4_I(inode
);
996 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
998 spin_lock(&sbi
->s_es_lru_lock
);
999 if (!list_empty(&ei
->i_es_lru
))
1000 list_del_init(&ei
->i_es_lru
);
1001 spin_unlock(&sbi
->s_es_lru_lock
);
1004 static int __es_try_to_reclaim_extents(struct ext4_inode_info
*ei
,
1007 struct inode
*inode
= &ei
->vfs_inode
;
1008 struct ext4_es_tree
*tree
= &ei
->i_es_tree
;
1009 struct rb_node
*node
;
1010 struct extent_status
*es
;
1013 if (ei
->i_es_lru_nr
== 0)
1016 node
= rb_first(&tree
->root
);
1017 while (node
!= NULL
) {
1018 es
= rb_entry(node
, struct extent_status
, rb_node
);
1019 node
= rb_next(&es
->rb_node
);
1021 * We can't reclaim delayed extent from status tree because
1022 * fiemap, bigallic, and seek_data/hole need to use it.
1024 if (!ext4_es_is_delayed(es
)) {
1025 rb_erase(&es
->rb_node
, &tree
->root
);
1026 ext4_es_free_extent(inode
, es
);
1028 if (--nr_to_scan
== 0)
1032 tree
->cache_es
= NULL
;