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
,
151 static int __ext4_es_shrink(struct ext4_sb_info
*sbi
, int nr_to_scan
,
152 struct ext4_inode_info
*locked_ei
);
154 int __init
ext4_init_es(void)
156 ext4_es_cachep
= kmem_cache_create("ext4_extent_status",
157 sizeof(struct extent_status
),
158 0, (SLAB_RECLAIM_ACCOUNT
), NULL
);
159 if (ext4_es_cachep
== NULL
)
164 void ext4_exit_es(void)
167 kmem_cache_destroy(ext4_es_cachep
);
170 void ext4_es_init_tree(struct ext4_es_tree
*tree
)
172 tree
->root
= RB_ROOT
;
173 tree
->cache_es
= NULL
;
177 static void ext4_es_print_tree(struct inode
*inode
)
179 struct ext4_es_tree
*tree
;
180 struct rb_node
*node
;
182 printk(KERN_DEBUG
"status extents for inode %lu:", inode
->i_ino
);
183 tree
= &EXT4_I(inode
)->i_es_tree
;
184 node
= rb_first(&tree
->root
);
186 struct extent_status
*es
;
187 es
= rb_entry(node
, struct extent_status
, rb_node
);
188 printk(KERN_DEBUG
" [%u/%u) %llu %llx",
189 es
->es_lblk
, es
->es_len
,
190 ext4_es_pblock(es
), ext4_es_status(es
));
191 node
= rb_next(node
);
193 printk(KERN_DEBUG
"\n");
196 #define ext4_es_print_tree(inode)
199 static inline ext4_lblk_t
ext4_es_end(struct extent_status
*es
)
201 BUG_ON(es
->es_lblk
+ es
->es_len
< es
->es_lblk
);
202 return es
->es_lblk
+ es
->es_len
- 1;
206 * search through the tree for an delayed extent with a given offset. If
207 * it can't be found, try to find next extent.
209 static struct extent_status
*__es_tree_search(struct rb_root
*root
,
212 struct rb_node
*node
= root
->rb_node
;
213 struct extent_status
*es
= NULL
;
216 es
= rb_entry(node
, struct extent_status
, rb_node
);
217 if (lblk
< es
->es_lblk
)
218 node
= node
->rb_left
;
219 else if (lblk
> ext4_es_end(es
))
220 node
= node
->rb_right
;
225 if (es
&& lblk
< es
->es_lblk
)
228 if (es
&& lblk
> ext4_es_end(es
)) {
229 node
= rb_next(&es
->rb_node
);
230 return node
? rb_entry(node
, struct extent_status
, rb_node
) :
238 * ext4_es_find_delayed_extent_range: find the 1st delayed extent covering
239 * @es->lblk if it exists, otherwise, the next extent after @es->lblk.
241 * @inode: the inode which owns delayed extents
242 * @lblk: the offset where we start to search
243 * @end: the offset where we stop to search
244 * @es: delayed extent that we found
246 void ext4_es_find_delayed_extent_range(struct inode
*inode
,
247 ext4_lblk_t lblk
, ext4_lblk_t end
,
248 struct extent_status
*es
)
250 struct ext4_es_tree
*tree
= NULL
;
251 struct extent_status
*es1
= NULL
;
252 struct rb_node
*node
;
256 trace_ext4_es_find_delayed_extent_range_enter(inode
, lblk
);
258 read_lock(&EXT4_I(inode
)->i_es_lock
);
259 tree
= &EXT4_I(inode
)->i_es_tree
;
261 /* find extent in cache firstly */
262 es
->es_lblk
= es
->es_len
= es
->es_pblk
= 0;
263 if (tree
->cache_es
) {
264 es1
= tree
->cache_es
;
265 if (in_range(lblk
, es1
->es_lblk
, es1
->es_len
)) {
266 es_debug("%u cached by [%u/%u) %llu %llx\n",
267 lblk
, es1
->es_lblk
, es1
->es_len
,
268 ext4_es_pblock(es1
), ext4_es_status(es1
));
273 es1
= __es_tree_search(&tree
->root
, lblk
);
276 if (es1
&& !ext4_es_is_delayed(es1
)) {
277 while ((node
= rb_next(&es1
->rb_node
)) != NULL
) {
278 es1
= rb_entry(node
, struct extent_status
, rb_node
);
279 if (es1
->es_lblk
> end
) {
283 if (ext4_es_is_delayed(es1
))
288 if (es1
&& ext4_es_is_delayed(es1
)) {
289 tree
->cache_es
= es1
;
290 es
->es_lblk
= es1
->es_lblk
;
291 es
->es_len
= es1
->es_len
;
292 es
->es_pblk
= es1
->es_pblk
;
295 read_unlock(&EXT4_I(inode
)->i_es_lock
);
297 trace_ext4_es_find_delayed_extent_range_exit(inode
, es
);
300 static struct extent_status
*
301 ext4_es_alloc_extent(struct inode
*inode
, ext4_lblk_t lblk
, ext4_lblk_t len
,
304 struct extent_status
*es
;
305 es
= kmem_cache_alloc(ext4_es_cachep
, GFP_ATOMIC
);
313 * We don't count delayed extent because we never try to reclaim them
315 if (!ext4_es_is_delayed(es
)) {
316 EXT4_I(inode
)->i_es_lru_nr
++;
317 percpu_counter_inc(&EXT4_SB(inode
->i_sb
)->s_extent_cache_cnt
);
323 static void ext4_es_free_extent(struct inode
*inode
, struct extent_status
*es
)
325 /* Decrease the lru counter when this es is not delayed */
326 if (!ext4_es_is_delayed(es
)) {
327 BUG_ON(EXT4_I(inode
)->i_es_lru_nr
== 0);
328 EXT4_I(inode
)->i_es_lru_nr
--;
329 percpu_counter_dec(&EXT4_SB(inode
->i_sb
)->s_extent_cache_cnt
);
332 kmem_cache_free(ext4_es_cachep
, es
);
336 * Check whether or not two extents can be merged
338 * - logical block number is contiguous
339 * - physical block number is contiguous
342 static int ext4_es_can_be_merged(struct extent_status
*es1
,
343 struct extent_status
*es2
)
345 if (ext4_es_status(es1
) != ext4_es_status(es2
))
348 if (((__u64
) es1
->es_len
) + es2
->es_len
> 0xFFFFFFFFULL
)
351 if (((__u64
) es1
->es_lblk
) + es1
->es_len
!= es2
->es_lblk
)
354 if ((ext4_es_is_written(es1
) || ext4_es_is_unwritten(es1
)) &&
355 (ext4_es_pblock(es1
) + es1
->es_len
== ext4_es_pblock(es2
)))
358 if (ext4_es_is_hole(es1
))
361 /* we need to check delayed extent is without unwritten status */
362 if (ext4_es_is_delayed(es1
) && !ext4_es_is_unwritten(es1
))
368 static struct extent_status
*
369 ext4_es_try_to_merge_left(struct inode
*inode
, struct extent_status
*es
)
371 struct ext4_es_tree
*tree
= &EXT4_I(inode
)->i_es_tree
;
372 struct extent_status
*es1
;
373 struct rb_node
*node
;
375 node
= rb_prev(&es
->rb_node
);
379 es1
= rb_entry(node
, struct extent_status
, rb_node
);
380 if (ext4_es_can_be_merged(es1
, es
)) {
381 es1
->es_len
+= es
->es_len
;
382 rb_erase(&es
->rb_node
, &tree
->root
);
383 ext4_es_free_extent(inode
, es
);
390 static struct extent_status
*
391 ext4_es_try_to_merge_right(struct inode
*inode
, struct extent_status
*es
)
393 struct ext4_es_tree
*tree
= &EXT4_I(inode
)->i_es_tree
;
394 struct extent_status
*es1
;
395 struct rb_node
*node
;
397 node
= rb_next(&es
->rb_node
);
401 es1
= rb_entry(node
, struct extent_status
, rb_node
);
402 if (ext4_es_can_be_merged(es
, es1
)) {
403 es
->es_len
+= es1
->es_len
;
404 rb_erase(node
, &tree
->root
);
405 ext4_es_free_extent(inode
, es1
);
411 #ifdef ES_AGGRESSIVE_TEST
412 static void ext4_es_insert_extent_ext_check(struct inode
*inode
,
413 struct extent_status
*es
)
415 struct ext4_ext_path
*path
= NULL
;
416 struct ext4_extent
*ex
;
417 ext4_lblk_t ee_block
;
418 ext4_fsblk_t ee_start
;
419 unsigned short ee_len
;
420 int depth
, ee_status
, es_status
;
422 path
= ext4_ext_find_extent(inode
, es
->es_lblk
, NULL
);
426 depth
= ext_depth(inode
);
427 ex
= path
[depth
].p_ext
;
431 ee_block
= le32_to_cpu(ex
->ee_block
);
432 ee_start
= ext4_ext_pblock(ex
);
433 ee_len
= ext4_ext_get_actual_len(ex
);
435 ee_status
= ext4_ext_is_uninitialized(ex
) ? 1 : 0;
436 es_status
= ext4_es_is_unwritten(es
) ? 1 : 0;
439 * Make sure ex and es are not overlap when we try to insert
440 * a delayed/hole extent.
442 if (!ext4_es_is_written(es
) && !ext4_es_is_unwritten(es
)) {
443 if (in_range(es
->es_lblk
, ee_block
, ee_len
)) {
444 pr_warn("ES insert assertion failed for "
445 "inode: %lu we can find an extent "
446 "at block [%d/%d/%llu/%c], but we "
447 "want to add an delayed/hole extent "
448 "[%d/%d/%llu/%llx]\n",
449 inode
->i_ino
, ee_block
, ee_len
,
450 ee_start
, ee_status
? 'u' : 'w',
451 es
->es_lblk
, es
->es_len
,
452 ext4_es_pblock(es
), ext4_es_status(es
));
458 * We don't check ee_block == es->es_lblk, etc. because es
459 * might be a part of whole extent, vice versa.
461 if (es
->es_lblk
< ee_block
||
462 ext4_es_pblock(es
) != ee_start
+ es
->es_lblk
- ee_block
) {
463 pr_warn("ES insert assertion failed for inode: %lu "
464 "ex_status [%d/%d/%llu/%c] != "
465 "es_status [%d/%d/%llu/%c]\n", inode
->i_ino
,
466 ee_block
, ee_len
, ee_start
,
467 ee_status
? 'u' : 'w', es
->es_lblk
, es
->es_len
,
468 ext4_es_pblock(es
), es_status
? 'u' : 'w');
472 if (ee_status
^ es_status
) {
473 pr_warn("ES insert assertion failed for inode: %lu "
474 "ex_status [%d/%d/%llu/%c] != "
475 "es_status [%d/%d/%llu/%c]\n", inode
->i_ino
,
476 ee_block
, ee_len
, ee_start
,
477 ee_status
? 'u' : 'w', es
->es_lblk
, es
->es_len
,
478 ext4_es_pblock(es
), es_status
? 'u' : 'w');
482 * We can't find an extent on disk. So we need to make sure
483 * that we don't want to add an written/unwritten extent.
485 if (!ext4_es_is_delayed(es
) && !ext4_es_is_hole(es
)) {
486 pr_warn("ES insert assertion failed for inode: %lu "
487 "can't find an extent at block %d but we want "
488 "to add an written/unwritten extent "
489 "[%d/%d/%llu/%llx]\n", inode
->i_ino
,
490 es
->es_lblk
, es
->es_lblk
, es
->es_len
,
491 ext4_es_pblock(es
), ext4_es_status(es
));
496 ext4_ext_drop_refs(path
);
501 static void ext4_es_insert_extent_ind_check(struct inode
*inode
,
502 struct extent_status
*es
)
504 struct ext4_map_blocks map
;
508 * Here we call ext4_ind_map_blocks to lookup a block mapping because
509 * 'Indirect' structure is defined in indirect.c. So we couldn't
510 * access direct/indirect tree from outside. It is too dirty to define
511 * this function in indirect.c file.
514 map
.m_lblk
= es
->es_lblk
;
515 map
.m_len
= es
->es_len
;
517 retval
= ext4_ind_map_blocks(NULL
, inode
, &map
, 0);
519 if (ext4_es_is_delayed(es
) || ext4_es_is_hole(es
)) {
521 * We want to add a delayed/hole extent but this
522 * block has been allocated.
524 pr_warn("ES insert assertion failed for inode: %lu "
525 "We can find blocks but we want to add a "
526 "delayed/hole extent [%d/%d/%llu/%llx]\n",
527 inode
->i_ino
, es
->es_lblk
, es
->es_len
,
528 ext4_es_pblock(es
), ext4_es_status(es
));
530 } else if (ext4_es_is_written(es
)) {
531 if (retval
!= es
->es_len
) {
532 pr_warn("ES insert assertion failed for "
533 "inode: %lu retval %d != es_len %d\n",
534 inode
->i_ino
, retval
, es
->es_len
);
537 if (map
.m_pblk
!= ext4_es_pblock(es
)) {
538 pr_warn("ES insert assertion failed for "
539 "inode: %lu m_pblk %llu != "
541 inode
->i_ino
, map
.m_pblk
,
547 * We don't need to check unwritten extent because
548 * indirect-based file doesn't have it.
552 } else if (retval
== 0) {
553 if (ext4_es_is_written(es
)) {
554 pr_warn("ES insert assertion failed for inode: %lu "
555 "We can't find the block but we want to add "
556 "an written extent [%d/%d/%llu/%llx]\n",
557 inode
->i_ino
, es
->es_lblk
, es
->es_len
,
558 ext4_es_pblock(es
), ext4_es_status(es
));
564 static inline void ext4_es_insert_extent_check(struct inode
*inode
,
565 struct extent_status
*es
)
568 * We don't need to worry about the race condition because
569 * caller takes i_data_sem locking.
571 BUG_ON(!rwsem_is_locked(&EXT4_I(inode
)->i_data_sem
));
572 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
573 ext4_es_insert_extent_ext_check(inode
, es
);
575 ext4_es_insert_extent_ind_check(inode
, es
);
578 static inline void ext4_es_insert_extent_check(struct inode
*inode
,
579 struct extent_status
*es
)
584 static int __es_insert_extent(struct inode
*inode
, struct extent_status
*newes
)
586 struct ext4_es_tree
*tree
= &EXT4_I(inode
)->i_es_tree
;
587 struct rb_node
**p
= &tree
->root
.rb_node
;
588 struct rb_node
*parent
= NULL
;
589 struct extent_status
*es
;
593 es
= rb_entry(parent
, struct extent_status
, rb_node
);
595 if (newes
->es_lblk
< es
->es_lblk
) {
596 if (ext4_es_can_be_merged(newes
, es
)) {
598 * Here we can modify es_lblk directly
599 * because it isn't overlapped.
601 es
->es_lblk
= newes
->es_lblk
;
602 es
->es_len
+= newes
->es_len
;
603 if (ext4_es_is_written(es
) ||
604 ext4_es_is_unwritten(es
))
605 ext4_es_store_pblock(es
,
607 es
= ext4_es_try_to_merge_left(inode
, es
);
611 } else if (newes
->es_lblk
> ext4_es_end(es
)) {
612 if (ext4_es_can_be_merged(es
, newes
)) {
613 es
->es_len
+= newes
->es_len
;
614 es
= ext4_es_try_to_merge_right(inode
, es
);
624 es
= ext4_es_alloc_extent(inode
, newes
->es_lblk
, newes
->es_len
,
628 rb_link_node(&es
->rb_node
, parent
, p
);
629 rb_insert_color(&es
->rb_node
, &tree
->root
);
637 * ext4_es_insert_extent() adds information to an inode's 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
);
671 err
= __es_insert_extent(inode
, &newes
);
672 if (err
== -ENOMEM
&& __ext4_es_shrink(EXT4_SB(inode
->i_sb
), 1,
675 if (err
== -ENOMEM
&& !ext4_es_is_delayed(&newes
))
679 write_unlock(&EXT4_I(inode
)->i_es_lock
);
681 ext4_es_print_tree(inode
);
687 * ext4_es_lookup_extent() looks up an extent in extent status tree.
689 * ext4_es_lookup_extent is called by ext4_map_blocks/ext4_da_map_blocks.
691 * Return: 1 on found, 0 on not
693 int ext4_es_lookup_extent(struct inode
*inode
, ext4_lblk_t lblk
,
694 struct extent_status
*es
)
696 struct ext4_es_tree
*tree
;
697 struct extent_status
*es1
= NULL
;
698 struct rb_node
*node
;
701 trace_ext4_es_lookup_extent_enter(inode
, lblk
);
702 es_debug("lookup extent in block %u\n", lblk
);
704 tree
= &EXT4_I(inode
)->i_es_tree
;
705 read_lock(&EXT4_I(inode
)->i_es_lock
);
707 /* find extent in cache firstly */
708 es
->es_lblk
= es
->es_len
= es
->es_pblk
= 0;
709 if (tree
->cache_es
) {
710 es1
= tree
->cache_es
;
711 if (in_range(lblk
, es1
->es_lblk
, es1
->es_len
)) {
712 es_debug("%u cached by [%u/%u)\n",
713 lblk
, es1
->es_lblk
, es1
->es_len
);
719 node
= tree
->root
.rb_node
;
721 es1
= rb_entry(node
, struct extent_status
, rb_node
);
722 if (lblk
< es1
->es_lblk
)
723 node
= node
->rb_left
;
724 else if (lblk
> ext4_es_end(es1
))
725 node
= node
->rb_right
;
735 es
->es_lblk
= es1
->es_lblk
;
736 es
->es_len
= es1
->es_len
;
737 es
->es_pblk
= es1
->es_pblk
;
740 read_unlock(&EXT4_I(inode
)->i_es_lock
);
742 trace_ext4_es_lookup_extent_exit(inode
, es
, found
);
746 static int __es_remove_extent(struct inode
*inode
, ext4_lblk_t lblk
,
749 struct ext4_es_tree
*tree
= &EXT4_I(inode
)->i_es_tree
;
750 struct rb_node
*node
;
751 struct extent_status
*es
;
752 struct extent_status orig_es
;
753 ext4_lblk_t len1
, len2
;
759 es
= __es_tree_search(&tree
->root
, lblk
);
762 if (es
->es_lblk
> end
)
765 /* Simply invalidate cache_es. */
766 tree
->cache_es
= NULL
;
768 orig_es
.es_lblk
= es
->es_lblk
;
769 orig_es
.es_len
= es
->es_len
;
770 orig_es
.es_pblk
= es
->es_pblk
;
772 len1
= lblk
> es
->es_lblk
? lblk
- es
->es_lblk
: 0;
773 len2
= ext4_es_end(es
) > end
? ext4_es_end(es
) - end
: 0;
778 struct extent_status newes
;
780 newes
.es_lblk
= end
+ 1;
782 if (ext4_es_is_written(&orig_es
) ||
783 ext4_es_is_unwritten(&orig_es
)) {
784 block
= ext4_es_pblock(&orig_es
) +
785 orig_es
.es_len
- len2
;
786 ext4_es_store_pblock(&newes
, block
);
788 ext4_es_store_status(&newes
, ext4_es_status(&orig_es
));
789 err
= __es_insert_extent(inode
, &newes
);
791 es
->es_lblk
= orig_es
.es_lblk
;
792 es
->es_len
= orig_es
.es_len
;
793 if ((err
== -ENOMEM
) &&
794 __ext4_es_shrink(EXT4_SB(inode
->i_sb
), 1,
800 es
->es_lblk
= end
+ 1;
802 if (ext4_es_is_written(es
) ||
803 ext4_es_is_unwritten(es
)) {
804 block
= orig_es
.es_pblk
+ orig_es
.es_len
- len2
;
805 ext4_es_store_pblock(es
, block
);
812 node
= rb_next(&es
->rb_node
);
814 es
= rb_entry(node
, struct extent_status
, rb_node
);
819 while (es
&& ext4_es_end(es
) <= end
) {
820 node
= rb_next(&es
->rb_node
);
821 rb_erase(&es
->rb_node
, &tree
->root
);
822 ext4_es_free_extent(inode
, es
);
827 es
= rb_entry(node
, struct extent_status
, rb_node
);
830 if (es
&& es
->es_lblk
< end
+ 1) {
831 ext4_lblk_t orig_len
= es
->es_len
;
833 len1
= ext4_es_end(es
) - end
;
834 es
->es_lblk
= end
+ 1;
836 if (ext4_es_is_written(es
) || ext4_es_is_unwritten(es
)) {
837 block
= es
->es_pblk
+ orig_len
- len1
;
838 ext4_es_store_pblock(es
, block
);
847 * ext4_es_remove_extent() removes a space from a extent status tree.
849 * Return 0 on success, error code on failure.
851 int ext4_es_remove_extent(struct inode
*inode
, ext4_lblk_t lblk
,
857 trace_ext4_es_remove_extent(inode
, lblk
, len
);
858 es_debug("remove [%u/%u) from extent status tree of inode %lu\n",
859 lblk
, len
, inode
->i_ino
);
864 end
= lblk
+ len
- 1;
867 write_lock(&EXT4_I(inode
)->i_es_lock
);
868 err
= __es_remove_extent(inode
, lblk
, end
);
869 write_unlock(&EXT4_I(inode
)->i_es_lock
);
870 ext4_es_print_tree(inode
);
874 int ext4_es_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
876 ext4_lblk_t ee_block
;
877 ext4_fsblk_t ee_pblock
;
880 ee_block
= le32_to_cpu(ex
->ee_block
);
881 ee_len
= ext4_ext_get_actual_len(ex
);
882 ee_pblock
= ext4_ext_pblock(ex
);
887 return ext4_es_insert_extent(inode
, ee_block
, ee_len
, ee_pblock
,
888 EXTENT_STATUS_WRITTEN
);
891 static int ext4_inode_touch_time_cmp(void *priv
, struct list_head
*a
,
894 struct ext4_inode_info
*eia
, *eib
;
895 eia
= list_entry(a
, struct ext4_inode_info
, i_es_lru
);
896 eib
= list_entry(b
, struct ext4_inode_info
, i_es_lru
);
898 if (eia
->i_touch_when
== eib
->i_touch_when
)
900 if (time_after(eia
->i_touch_when
, eib
->i_touch_when
))
906 static int __ext4_es_shrink(struct ext4_sb_info
*sbi
, int nr_to_scan
,
907 struct ext4_inode_info
*locked_ei
)
909 struct ext4_inode_info
*ei
;
910 struct list_head
*cur
, *tmp
;
912 int ret
, nr_shrunk
= 0;
914 spin_lock(&sbi
->s_es_lru_lock
);
917 * If the inode that is at the head of LRU list is newer than
918 * last_sorted time, that means that we need to sort this list.
920 ei
= list_first_entry(&sbi
->s_es_lru
, struct ext4_inode_info
, i_es_lru
);
921 if (sbi
->s_es_last_sorted
< ei
->i_touch_when
) {
922 list_sort(NULL
, &sbi
->s_es_lru
, ext4_inode_touch_time_cmp
);
923 sbi
->s_es_last_sorted
= jiffies
;
926 list_for_each_safe(cur
, tmp
, &sbi
->s_es_lru
) {
928 * If we have already reclaimed all extents from extent
929 * status tree, just stop the loop immediately.
931 if (percpu_counter_read_positive(&sbi
->s_extent_cache_cnt
) == 0)
934 ei
= list_entry(cur
, struct ext4_inode_info
, i_es_lru
);
936 /* Skip the inode that is newer than the last_sorted time */
937 if (sbi
->s_es_last_sorted
< ei
->i_touch_when
) {
938 list_move_tail(cur
, &skiped
);
942 if (ei
->i_es_lru_nr
== 0 || ei
== locked_ei
)
945 write_lock(&ei
->i_es_lock
);
946 ret
= __es_try_to_reclaim_extents(ei
, nr_to_scan
);
947 if (ei
->i_es_lru_nr
== 0)
948 list_del_init(&ei
->i_es_lru
);
949 write_unlock(&ei
->i_es_lock
);
957 /* Move the newer inodes into the tail of the LRU list. */
958 list_splice_tail(&skiped
, &sbi
->s_es_lru
);
959 spin_unlock(&sbi
->s_es_lru_lock
);
961 if (locked_ei
&& nr_shrunk
== 0)
962 nr_shrunk
= __es_try_to_reclaim_extents(ei
, nr_to_scan
);
967 static int ext4_es_shrink(struct shrinker
*shrink
, struct shrink_control
*sc
)
969 struct ext4_sb_info
*sbi
= container_of(shrink
,
970 struct ext4_sb_info
, s_es_shrinker
);
971 int nr_to_scan
= sc
->nr_to_scan
;
974 ret
= percpu_counter_read_positive(&sbi
->s_extent_cache_cnt
);
975 trace_ext4_es_shrink_enter(sbi
->s_sb
, nr_to_scan
, ret
);
980 nr_shrunk
= __ext4_es_shrink(sbi
, nr_to_scan
, NULL
);
982 ret
= percpu_counter_read_positive(&sbi
->s_extent_cache_cnt
);
983 trace_ext4_es_shrink_exit(sbi
->s_sb
, nr_shrunk
, ret
);
987 void ext4_es_register_shrinker(struct ext4_sb_info
*sbi
)
989 INIT_LIST_HEAD(&sbi
->s_es_lru
);
990 spin_lock_init(&sbi
->s_es_lru_lock
);
991 sbi
->s_es_last_sorted
= 0;
992 sbi
->s_es_shrinker
.shrink
= ext4_es_shrink
;
993 sbi
->s_es_shrinker
.seeks
= DEFAULT_SEEKS
;
994 register_shrinker(&sbi
->s_es_shrinker
);
997 void ext4_es_unregister_shrinker(struct ext4_sb_info
*sbi
)
999 unregister_shrinker(&sbi
->s_es_shrinker
);
1002 void ext4_es_lru_add(struct inode
*inode
)
1004 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1005 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1007 ei
->i_touch_when
= jiffies
;
1009 if (!list_empty(&ei
->i_es_lru
))
1012 spin_lock(&sbi
->s_es_lru_lock
);
1013 if (list_empty(&ei
->i_es_lru
))
1014 list_add_tail(&ei
->i_es_lru
, &sbi
->s_es_lru
);
1015 spin_unlock(&sbi
->s_es_lru_lock
);
1018 void ext4_es_lru_del(struct inode
*inode
)
1020 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1021 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1023 spin_lock(&sbi
->s_es_lru_lock
);
1024 if (!list_empty(&ei
->i_es_lru
))
1025 list_del_init(&ei
->i_es_lru
);
1026 spin_unlock(&sbi
->s_es_lru_lock
);
1029 static int __es_try_to_reclaim_extents(struct ext4_inode_info
*ei
,
1032 struct inode
*inode
= &ei
->vfs_inode
;
1033 struct ext4_es_tree
*tree
= &ei
->i_es_tree
;
1034 struct rb_node
*node
;
1035 struct extent_status
*es
;
1038 if (ei
->i_es_lru_nr
== 0)
1041 node
= rb_first(&tree
->root
);
1042 while (node
!= NULL
) {
1043 es
= rb_entry(node
, struct extent_status
, rb_node
);
1044 node
= rb_next(&es
->rb_node
);
1046 * We can't reclaim delayed extent from status tree because
1047 * fiemap, bigallic, and seek_data/hole need to use it.
1049 if (!ext4_es_is_delayed(es
)) {
1050 rb_erase(&es
->rb_node
, &tree
->root
);
1051 ext4_es_free_extent(inode
, es
);
1053 if (--nr_to_scan
== 0)
1057 tree
->cache_es
= NULL
;