1 // SPDX-License-Identifier: GPL-2.0
3 * fs/ext4/extents_status.c
5 * Written by Yongqiang Yang <xiaoqiangnk@gmail.com>
7 * Allison Henderson <achender@linux.vnet.ibm.com>
8 * Hugh Dickins <hughd@google.com>
9 * Zheng Liu <wenqing.lz@taobao.com>
11 * Ext4 extents status tree core functions.
13 #include <linux/list_sort.h>
14 #include <linux/proc_fs.h>
15 #include <linux/seq_file.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;
145 static struct kmem_cache *ext4_pending_cachep;
147 static int __es_insert_extent(struct inode *inode, struct extent_status *newes);
148 static int __es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
149 ext4_lblk_t end, int *reserved);
150 static int es_reclaim_extents(struct ext4_inode_info *ei, int *nr_to_scan);
151 static int __es_shrink(struct ext4_sb_info *sbi, int nr_to_scan,
152 struct ext4_inode_info *locked_ei);
153 static void __revise_pending(struct inode *inode, ext4_lblk_t lblk,
156 int __init ext4_init_es(void)
158 ext4_es_cachep = KMEM_CACHE(extent_status, SLAB_RECLAIM_ACCOUNT);
159 if (ext4_es_cachep == NULL)
164 void ext4_exit_es(void)
166 kmem_cache_destroy(ext4_es_cachep);
169 void ext4_es_init_tree(struct ext4_es_tree *tree)
171 tree->root = RB_ROOT;
172 tree->cache_es = NULL;
176 static void ext4_es_print_tree(struct inode *inode)
178 struct ext4_es_tree *tree;
179 struct rb_node *node;
181 printk(KERN_DEBUG "status extents for inode %lu:", inode->i_ino);
182 tree = &EXT4_I(inode)->i_es_tree;
183 node = rb_first(&tree->root);
185 struct extent_status *es;
186 es = rb_entry(node, struct extent_status, rb_node);
187 printk(KERN_DEBUG " [%u/%u) %llu %x",
188 es->es_lblk, es->es_len,
189 ext4_es_pblock(es), ext4_es_status(es));
190 node = rb_next(node);
192 printk(KERN_DEBUG "\n");
195 #define ext4_es_print_tree(inode)
198 static inline ext4_lblk_t ext4_es_end(struct extent_status *es)
200 BUG_ON(es->es_lblk + es->es_len < es->es_lblk);
201 return es->es_lblk + es->es_len - 1;
205 * search through the tree for an delayed extent with a given offset. If
206 * it can't be found, try to find next extent.
208 static struct extent_status *__es_tree_search(struct rb_root *root,
211 struct rb_node *node = root->rb_node;
212 struct extent_status *es = NULL;
215 es = rb_entry(node, struct extent_status, rb_node);
216 if (lblk < es->es_lblk)
217 node = node->rb_left;
218 else if (lblk > ext4_es_end(es))
219 node = node->rb_right;
224 if (es && lblk < es->es_lblk)
227 if (es && lblk > ext4_es_end(es)) {
228 node = rb_next(&es->rb_node);
229 return node ? rb_entry(node, struct extent_status, rb_node) :
237 * ext4_es_find_extent_range - find extent with specified status within block
238 * range or next extent following block range in
239 * extents status tree
241 * @inode - file containing the range
242 * @matching_fn - pointer to function that matches extents with desired status
243 * @lblk - logical block defining start of range
244 * @end - logical block defining end of range
245 * @es - extent found, if any
247 * Find the first extent within the block range specified by @lblk and @end
248 * in the extents status tree that satisfies @matching_fn. If a match
249 * is found, it's returned in @es. If not, and a matching extent is found
250 * beyond the block range, it's returned in @es. If no match is found, an
251 * extent is returned in @es whose es_lblk, es_len, and es_pblk components
254 static void __es_find_extent_range(struct inode *inode,
255 int (*matching_fn)(struct extent_status *es),
256 ext4_lblk_t lblk, ext4_lblk_t end,
257 struct extent_status *es)
259 struct ext4_es_tree *tree = NULL;
260 struct extent_status *es1 = NULL;
261 struct rb_node *node;
266 tree = &EXT4_I(inode)->i_es_tree;
268 /* see if the extent has been cached */
269 es->es_lblk = es->es_len = es->es_pblk = 0;
270 if (tree->cache_es) {
271 es1 = tree->cache_es;
272 if (in_range(lblk, es1->es_lblk, es1->es_len)) {
273 es_debug("%u cached by [%u/%u) %llu %x\n",
274 lblk, es1->es_lblk, es1->es_len,
275 ext4_es_pblock(es1), ext4_es_status(es1));
280 es1 = __es_tree_search(&tree->root, lblk);
283 if (es1 && !matching_fn(es1)) {
284 while ((node = rb_next(&es1->rb_node)) != NULL) {
285 es1 = rb_entry(node, struct extent_status, rb_node);
286 if (es1->es_lblk > end) {
290 if (matching_fn(es1))
295 if (es1 && matching_fn(es1)) {
296 tree->cache_es = es1;
297 es->es_lblk = es1->es_lblk;
298 es->es_len = es1->es_len;
299 es->es_pblk = es1->es_pblk;
305 * Locking for __es_find_extent_range() for external use
307 void ext4_es_find_extent_range(struct inode *inode,
308 int (*matching_fn)(struct extent_status *es),
309 ext4_lblk_t lblk, ext4_lblk_t end,
310 struct extent_status *es)
312 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
315 trace_ext4_es_find_extent_range_enter(inode, lblk);
317 read_lock(&EXT4_I(inode)->i_es_lock);
318 __es_find_extent_range(inode, matching_fn, lblk, end, es);
319 read_unlock(&EXT4_I(inode)->i_es_lock);
321 trace_ext4_es_find_extent_range_exit(inode, es);
325 * __es_scan_range - search block range for block with specified status
326 * in extents status tree
328 * @inode - file containing the range
329 * @matching_fn - pointer to function that matches extents with desired status
330 * @lblk - logical block defining start of range
331 * @end - logical block defining end of range
333 * Returns true if at least one block in the specified block range satisfies
334 * the criterion specified by @matching_fn, and false if not. If at least
335 * one extent has the specified status, then there is at least one block
336 * in the cluster with that status. Should only be called by code that has
339 static bool __es_scan_range(struct inode *inode,
340 int (*matching_fn)(struct extent_status *es),
341 ext4_lblk_t start, ext4_lblk_t end)
343 struct extent_status es;
345 __es_find_extent_range(inode, matching_fn, start, end, &es);
347 return false; /* no matching extent in the tree */
348 else if (es.es_lblk <= start &&
349 start < es.es_lblk + es.es_len)
351 else if (start <= es.es_lblk && es.es_lblk <= end)
357 * Locking for __es_scan_range() for external use
359 bool ext4_es_scan_range(struct inode *inode,
360 int (*matching_fn)(struct extent_status *es),
361 ext4_lblk_t lblk, ext4_lblk_t end)
365 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
368 read_lock(&EXT4_I(inode)->i_es_lock);
369 ret = __es_scan_range(inode, matching_fn, lblk, end);
370 read_unlock(&EXT4_I(inode)->i_es_lock);
376 * __es_scan_clu - search cluster for block with specified status in
377 * extents status tree
379 * @inode - file containing the cluster
380 * @matching_fn - pointer to function that matches extents with desired status
381 * @lblk - logical block in cluster to be searched
383 * Returns true if at least one extent in the cluster containing @lblk
384 * satisfies the criterion specified by @matching_fn, and false if not. If at
385 * least one extent has the specified status, then there is at least one block
386 * in the cluster with that status. Should only be called by code that has
389 static bool __es_scan_clu(struct inode *inode,
390 int (*matching_fn)(struct extent_status *es),
393 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
394 ext4_lblk_t lblk_start, lblk_end;
396 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
397 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
399 return __es_scan_range(inode, matching_fn, lblk_start, lblk_end);
403 * Locking for __es_scan_clu() for external use
405 bool ext4_es_scan_clu(struct inode *inode,
406 int (*matching_fn)(struct extent_status *es),
411 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
414 read_lock(&EXT4_I(inode)->i_es_lock);
415 ret = __es_scan_clu(inode, matching_fn, lblk);
416 read_unlock(&EXT4_I(inode)->i_es_lock);
421 static void ext4_es_list_add(struct inode *inode)
423 struct ext4_inode_info *ei = EXT4_I(inode);
424 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
426 if (!list_empty(&ei->i_es_list))
429 spin_lock(&sbi->s_es_lock);
430 if (list_empty(&ei->i_es_list)) {
431 list_add_tail(&ei->i_es_list, &sbi->s_es_list);
432 sbi->s_es_nr_inode++;
434 spin_unlock(&sbi->s_es_lock);
437 static void ext4_es_list_del(struct inode *inode)
439 struct ext4_inode_info *ei = EXT4_I(inode);
440 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
442 spin_lock(&sbi->s_es_lock);
443 if (!list_empty(&ei->i_es_list)) {
444 list_del_init(&ei->i_es_list);
445 sbi->s_es_nr_inode--;
446 WARN_ON_ONCE(sbi->s_es_nr_inode < 0);
448 spin_unlock(&sbi->s_es_lock);
451 static struct extent_status *
452 ext4_es_alloc_extent(struct inode *inode, ext4_lblk_t lblk, ext4_lblk_t len,
455 struct extent_status *es;
456 es = kmem_cache_alloc(ext4_es_cachep, GFP_ATOMIC);
464 * We don't count delayed extent because we never try to reclaim them
466 if (!ext4_es_is_delayed(es)) {
467 if (!EXT4_I(inode)->i_es_shk_nr++)
468 ext4_es_list_add(inode);
469 percpu_counter_inc(&EXT4_SB(inode->i_sb)->
470 s_es_stats.es_stats_shk_cnt);
473 EXT4_I(inode)->i_es_all_nr++;
474 percpu_counter_inc(&EXT4_SB(inode->i_sb)->s_es_stats.es_stats_all_cnt);
479 static void ext4_es_free_extent(struct inode *inode, struct extent_status *es)
481 EXT4_I(inode)->i_es_all_nr--;
482 percpu_counter_dec(&EXT4_SB(inode->i_sb)->s_es_stats.es_stats_all_cnt);
484 /* Decrease the shrink counter when this es is not delayed */
485 if (!ext4_es_is_delayed(es)) {
486 BUG_ON(EXT4_I(inode)->i_es_shk_nr == 0);
487 if (!--EXT4_I(inode)->i_es_shk_nr)
488 ext4_es_list_del(inode);
489 percpu_counter_dec(&EXT4_SB(inode->i_sb)->
490 s_es_stats.es_stats_shk_cnt);
493 kmem_cache_free(ext4_es_cachep, es);
497 * Check whether or not two extents can be merged
499 * - logical block number is contiguous
500 * - physical block number is contiguous
503 static int ext4_es_can_be_merged(struct extent_status *es1,
504 struct extent_status *es2)
506 if (ext4_es_type(es1) != ext4_es_type(es2))
509 if (((__u64) es1->es_len) + es2->es_len > EXT_MAX_BLOCKS) {
510 pr_warn("ES assertion failed when merging extents. "
511 "The sum of lengths of es1 (%d) and es2 (%d) "
512 "is bigger than allowed file size (%d)\n",
513 es1->es_len, es2->es_len, EXT_MAX_BLOCKS);
518 if (((__u64) es1->es_lblk) + es1->es_len != es2->es_lblk)
521 if ((ext4_es_is_written(es1) || ext4_es_is_unwritten(es1)) &&
522 (ext4_es_pblock(es1) + es1->es_len == ext4_es_pblock(es2)))
525 if (ext4_es_is_hole(es1))
528 /* we need to check delayed extent is without unwritten status */
529 if (ext4_es_is_delayed(es1) && !ext4_es_is_unwritten(es1))
535 static struct extent_status *
536 ext4_es_try_to_merge_left(struct inode *inode, struct extent_status *es)
538 struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
539 struct extent_status *es1;
540 struct rb_node *node;
542 node = rb_prev(&es->rb_node);
546 es1 = rb_entry(node, struct extent_status, rb_node);
547 if (ext4_es_can_be_merged(es1, es)) {
548 es1->es_len += es->es_len;
549 if (ext4_es_is_referenced(es))
550 ext4_es_set_referenced(es1);
551 rb_erase(&es->rb_node, &tree->root);
552 ext4_es_free_extent(inode, es);
559 static struct extent_status *
560 ext4_es_try_to_merge_right(struct inode *inode, struct extent_status *es)
562 struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
563 struct extent_status *es1;
564 struct rb_node *node;
566 node = rb_next(&es->rb_node);
570 es1 = rb_entry(node, struct extent_status, rb_node);
571 if (ext4_es_can_be_merged(es, es1)) {
572 es->es_len += es1->es_len;
573 if (ext4_es_is_referenced(es1))
574 ext4_es_set_referenced(es);
575 rb_erase(node, &tree->root);
576 ext4_es_free_extent(inode, es1);
582 #ifdef ES_AGGRESSIVE_TEST
583 #include "ext4_extents.h" /* Needed when ES_AGGRESSIVE_TEST is defined */
585 static void ext4_es_insert_extent_ext_check(struct inode *inode,
586 struct extent_status *es)
588 struct ext4_ext_path *path = NULL;
589 struct ext4_extent *ex;
590 ext4_lblk_t ee_block;
591 ext4_fsblk_t ee_start;
592 unsigned short ee_len;
593 int depth, ee_status, es_status;
595 path = ext4_find_extent(inode, es->es_lblk, NULL, EXT4_EX_NOCACHE);
599 depth = ext_depth(inode);
600 ex = path[depth].p_ext;
604 ee_block = le32_to_cpu(ex->ee_block);
605 ee_start = ext4_ext_pblock(ex);
606 ee_len = ext4_ext_get_actual_len(ex);
608 ee_status = ext4_ext_is_unwritten(ex) ? 1 : 0;
609 es_status = ext4_es_is_unwritten(es) ? 1 : 0;
612 * Make sure ex and es are not overlap when we try to insert
613 * a delayed/hole extent.
615 if (!ext4_es_is_written(es) && !ext4_es_is_unwritten(es)) {
616 if (in_range(es->es_lblk, ee_block, ee_len)) {
617 pr_warn("ES insert assertion failed for "
618 "inode: %lu we can find an extent "
619 "at block [%d/%d/%llu/%c], but we "
620 "want to add a delayed/hole extent "
622 inode->i_ino, ee_block, ee_len,
623 ee_start, ee_status ? 'u' : 'w',
624 es->es_lblk, es->es_len,
625 ext4_es_pblock(es), ext4_es_status(es));
631 * We don't check ee_block == es->es_lblk, etc. because es
632 * might be a part of whole extent, vice versa.
634 if (es->es_lblk < ee_block ||
635 ext4_es_pblock(es) != ee_start + es->es_lblk - ee_block) {
636 pr_warn("ES insert assertion failed for inode: %lu "
637 "ex_status [%d/%d/%llu/%c] != "
638 "es_status [%d/%d/%llu/%c]\n", inode->i_ino,
639 ee_block, ee_len, ee_start,
640 ee_status ? 'u' : 'w', es->es_lblk, es->es_len,
641 ext4_es_pblock(es), es_status ? 'u' : 'w');
645 if (ee_status ^ es_status) {
646 pr_warn("ES insert assertion failed for inode: %lu "
647 "ex_status [%d/%d/%llu/%c] != "
648 "es_status [%d/%d/%llu/%c]\n", inode->i_ino,
649 ee_block, ee_len, ee_start,
650 ee_status ? 'u' : 'w', es->es_lblk, es->es_len,
651 ext4_es_pblock(es), es_status ? 'u' : 'w');
655 * We can't find an extent on disk. So we need to make sure
656 * that we don't want to add an written/unwritten extent.
658 if (!ext4_es_is_delayed(es) && !ext4_es_is_hole(es)) {
659 pr_warn("ES insert assertion failed for inode: %lu "
660 "can't find an extent at block %d but we want "
661 "to add a written/unwritten extent "
662 "[%d/%d/%llu/%x]\n", inode->i_ino,
663 es->es_lblk, es->es_lblk, es->es_len,
664 ext4_es_pblock(es), ext4_es_status(es));
668 ext4_free_ext_path(path);
671 static void ext4_es_insert_extent_ind_check(struct inode *inode,
672 struct extent_status *es)
674 struct ext4_map_blocks map;
678 * Here we call ext4_ind_map_blocks to lookup a block mapping because
679 * 'Indirect' structure is defined in indirect.c. So we couldn't
680 * access direct/indirect tree from outside. It is too dirty to define
681 * this function in indirect.c file.
684 map.m_lblk = es->es_lblk;
685 map.m_len = es->es_len;
687 retval = ext4_ind_map_blocks(NULL, inode, &map, 0);
689 if (ext4_es_is_delayed(es) || ext4_es_is_hole(es)) {
691 * We want to add a delayed/hole extent but this
692 * block has been allocated.
694 pr_warn("ES insert assertion failed for inode: %lu "
695 "We can find blocks but we want to add a "
696 "delayed/hole extent [%d/%d/%llu/%x]\n",
697 inode->i_ino, es->es_lblk, es->es_len,
698 ext4_es_pblock(es), ext4_es_status(es));
700 } else if (ext4_es_is_written(es)) {
701 if (retval != es->es_len) {
702 pr_warn("ES insert assertion failed for "
703 "inode: %lu retval %d != es_len %d\n",
704 inode->i_ino, retval, es->es_len);
707 if (map.m_pblk != ext4_es_pblock(es)) {
708 pr_warn("ES insert assertion failed for "
709 "inode: %lu m_pblk %llu != "
711 inode->i_ino, map.m_pblk,
717 * We don't need to check unwritten extent because
718 * indirect-based file doesn't have it.
722 } else if (retval == 0) {
723 if (ext4_es_is_written(es)) {
724 pr_warn("ES insert assertion failed for inode: %lu "
725 "We can't find the block but we want to add "
726 "a written extent [%d/%d/%llu/%x]\n",
727 inode->i_ino, es->es_lblk, es->es_len,
728 ext4_es_pblock(es), ext4_es_status(es));
734 static inline void ext4_es_insert_extent_check(struct inode *inode,
735 struct extent_status *es)
738 * We don't need to worry about the race condition because
739 * caller takes i_data_sem locking.
741 BUG_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
742 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
743 ext4_es_insert_extent_ext_check(inode, es);
745 ext4_es_insert_extent_ind_check(inode, es);
748 static inline void ext4_es_insert_extent_check(struct inode *inode,
749 struct extent_status *es)
754 static int __es_insert_extent(struct inode *inode, struct extent_status *newes)
756 struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
757 struct rb_node **p = &tree->root.rb_node;
758 struct rb_node *parent = NULL;
759 struct extent_status *es;
763 es = rb_entry(parent, struct extent_status, rb_node);
765 if (newes->es_lblk < es->es_lblk) {
766 if (ext4_es_can_be_merged(newes, es)) {
768 * Here we can modify es_lblk directly
769 * because it isn't overlapped.
771 es->es_lblk = newes->es_lblk;
772 es->es_len += newes->es_len;
773 if (ext4_es_is_written(es) ||
774 ext4_es_is_unwritten(es))
775 ext4_es_store_pblock(es,
777 es = ext4_es_try_to_merge_left(inode, es);
781 } else if (newes->es_lblk > ext4_es_end(es)) {
782 if (ext4_es_can_be_merged(es, newes)) {
783 es->es_len += newes->es_len;
784 es = ext4_es_try_to_merge_right(inode, es);
794 es = ext4_es_alloc_extent(inode, newes->es_lblk, newes->es_len,
798 rb_link_node(&es->rb_node, parent, p);
799 rb_insert_color(&es->rb_node, &tree->root);
807 * ext4_es_insert_extent() adds information to an inode's extent
810 * Return 0 on success, error code on failure.
812 int ext4_es_insert_extent(struct inode *inode, ext4_lblk_t lblk,
813 ext4_lblk_t len, ext4_fsblk_t pblk,
816 struct extent_status newes;
817 ext4_lblk_t end = lblk + len - 1;
819 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
821 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
824 es_debug("add [%u/%u) %llu %x to extent status tree of inode %lu\n",
825 lblk, len, pblk, status, inode->i_ino);
832 if ((status & EXTENT_STATUS_DELAYED) &&
833 (status & EXTENT_STATUS_WRITTEN)) {
834 ext4_warning(inode->i_sb, "Inserting extent [%u/%u] as "
835 " delayed and written which can potentially "
836 " cause data loss.", lblk, len);
840 newes.es_lblk = lblk;
842 ext4_es_store_pblock_status(&newes, pblk, status);
843 trace_ext4_es_insert_extent(inode, &newes);
845 ext4_es_insert_extent_check(inode, &newes);
847 write_lock(&EXT4_I(inode)->i_es_lock);
848 err = __es_remove_extent(inode, lblk, end, NULL);
852 err = __es_insert_extent(inode, &newes);
853 if (err == -ENOMEM && __es_shrink(EXT4_SB(inode->i_sb),
856 if (err == -ENOMEM && !ext4_es_is_delayed(&newes))
859 if (sbi->s_cluster_ratio > 1 && test_opt(inode->i_sb, DELALLOC) &&
860 (status & EXTENT_STATUS_WRITTEN ||
861 status & EXTENT_STATUS_UNWRITTEN))
862 __revise_pending(inode, lblk, len);
865 write_unlock(&EXT4_I(inode)->i_es_lock);
867 ext4_es_print_tree(inode);
873 * ext4_es_cache_extent() inserts information into the extent status
874 * tree if and only if there isn't information about the range in
877 void ext4_es_cache_extent(struct inode *inode, ext4_lblk_t lblk,
878 ext4_lblk_t len, ext4_fsblk_t pblk,
881 struct extent_status *es;
882 struct extent_status newes;
883 ext4_lblk_t end = lblk + len - 1;
885 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
888 newes.es_lblk = lblk;
890 ext4_es_store_pblock_status(&newes, pblk, status);
891 trace_ext4_es_cache_extent(inode, &newes);
898 write_lock(&EXT4_I(inode)->i_es_lock);
900 es = __es_tree_search(&EXT4_I(inode)->i_es_tree.root, lblk);
901 if (!es || es->es_lblk > end)
902 __es_insert_extent(inode, &newes);
903 write_unlock(&EXT4_I(inode)->i_es_lock);
907 * ext4_es_lookup_extent() looks up an extent in extent status tree.
909 * ext4_es_lookup_extent is called by ext4_map_blocks/ext4_da_map_blocks.
911 * Return: 1 on found, 0 on not
913 int ext4_es_lookup_extent(struct inode *inode, ext4_lblk_t lblk,
914 ext4_lblk_t *next_lblk,
915 struct extent_status *es)
917 struct ext4_es_tree *tree;
918 struct ext4_es_stats *stats;
919 struct extent_status *es1 = NULL;
920 struct rb_node *node;
923 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
926 trace_ext4_es_lookup_extent_enter(inode, lblk);
927 es_debug("lookup extent in block %u\n", lblk);
929 tree = &EXT4_I(inode)->i_es_tree;
930 read_lock(&EXT4_I(inode)->i_es_lock);
932 /* find extent in cache firstly */
933 es->es_lblk = es->es_len = es->es_pblk = 0;
934 if (tree->cache_es) {
935 es1 = tree->cache_es;
936 if (in_range(lblk, es1->es_lblk, es1->es_len)) {
937 es_debug("%u cached by [%u/%u)\n",
938 lblk, es1->es_lblk, es1->es_len);
944 node = tree->root.rb_node;
946 es1 = rb_entry(node, struct extent_status, rb_node);
947 if (lblk < es1->es_lblk)
948 node = node->rb_left;
949 else if (lblk > ext4_es_end(es1))
950 node = node->rb_right;
958 stats = &EXT4_SB(inode->i_sb)->s_es_stats;
961 es->es_lblk = es1->es_lblk;
962 es->es_len = es1->es_len;
963 es->es_pblk = es1->es_pblk;
964 if (!ext4_es_is_referenced(es1))
965 ext4_es_set_referenced(es1);
966 percpu_counter_inc(&stats->es_stats_cache_hits);
968 node = rb_next(&es1->rb_node);
970 es1 = rb_entry(node, struct extent_status,
972 *next_lblk = es1->es_lblk;
977 percpu_counter_inc(&stats->es_stats_cache_misses);
980 read_unlock(&EXT4_I(inode)->i_es_lock);
982 trace_ext4_es_lookup_extent_exit(inode, es, found);
988 bool first_do_lblk_found;
989 ext4_lblk_t first_do_lblk;
990 ext4_lblk_t last_do_lblk;
991 struct extent_status *left_es;
997 * init_rsvd - initialize reserved count data before removing block range
998 * in file from extent status tree
1000 * @inode - file containing range
1001 * @lblk - first block in range
1002 * @es - pointer to first extent in range
1003 * @rc - pointer to reserved count data
1005 * Assumes es is not NULL
1007 static void init_rsvd(struct inode *inode, ext4_lblk_t lblk,
1008 struct extent_status *es, struct rsvd_count *rc)
1010 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1011 struct rb_node *node;
1016 * for bigalloc, note the first delonly block in the range has not
1017 * been found, record the extent containing the block to the left of
1018 * the region to be removed, if any, and note that there's no partial
1021 if (sbi->s_cluster_ratio > 1) {
1022 rc->first_do_lblk_found = false;
1023 if (lblk > es->es_lblk) {
1026 node = rb_prev(&es->rb_node);
1027 rc->left_es = node ? rb_entry(node,
1028 struct extent_status,
1031 rc->partial = false;
1036 * count_rsvd - count the clusters containing delayed and not unwritten
1037 * (delonly) blocks in a range within an extent and add to
1038 * the running tally in rsvd_count
1040 * @inode - file containing extent
1041 * @lblk - first block in range
1042 * @len - length of range in blocks
1043 * @es - pointer to extent containing clusters to be counted
1044 * @rc - pointer to reserved count data
1046 * Tracks partial clusters found at the beginning and end of extents so
1047 * they aren't overcounted when they span adjacent extents
1049 static void count_rsvd(struct inode *inode, ext4_lblk_t lblk, long len,
1050 struct extent_status *es, struct rsvd_count *rc)
1052 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1053 ext4_lblk_t i, end, nclu;
1055 if (!ext4_es_is_delonly(es))
1060 if (sbi->s_cluster_ratio == 1) {
1061 rc->ndelonly += (int) len;
1067 i = (lblk < es->es_lblk) ? es->es_lblk : lblk;
1068 end = lblk + (ext4_lblk_t) len - 1;
1069 end = (end > ext4_es_end(es)) ? ext4_es_end(es) : end;
1071 /* record the first block of the first delonly extent seen */
1072 if (!rc->first_do_lblk_found) {
1073 rc->first_do_lblk = i;
1074 rc->first_do_lblk_found = true;
1077 /* update the last lblk in the region seen so far */
1078 rc->last_do_lblk = end;
1081 * if we're tracking a partial cluster and the current extent
1082 * doesn't start with it, count it and stop tracking
1084 if (rc->partial && (rc->lclu != EXT4_B2C(sbi, i))) {
1086 rc->partial = false;
1090 * if the first cluster doesn't start on a cluster boundary but
1091 * ends on one, count it
1093 if (EXT4_LBLK_COFF(sbi, i) != 0) {
1094 if (end >= EXT4_LBLK_CFILL(sbi, i)) {
1096 rc->partial = false;
1097 i = EXT4_LBLK_CFILL(sbi, i) + 1;
1102 * if the current cluster starts on a cluster boundary, count the
1103 * number of whole delonly clusters in the extent
1105 if ((i + sbi->s_cluster_ratio - 1) <= end) {
1106 nclu = (end - i + 1) >> sbi->s_cluster_bits;
1107 rc->ndelonly += nclu;
1108 i += nclu << sbi->s_cluster_bits;
1112 * start tracking a partial cluster if there's a partial at the end
1113 * of the current extent and we're not already tracking one
1115 if (!rc->partial && i <= end) {
1117 rc->lclu = EXT4_B2C(sbi, i);
1122 * __pr_tree_search - search for a pending cluster reservation
1124 * @root - root of pending reservation tree
1125 * @lclu - logical cluster to search for
1127 * Returns the pending reservation for the cluster identified by @lclu
1128 * if found. If not, returns a reservation for the next cluster if any,
1129 * and if not, returns NULL.
1131 static struct pending_reservation *__pr_tree_search(struct rb_root *root,
1134 struct rb_node *node = root->rb_node;
1135 struct pending_reservation *pr = NULL;
1138 pr = rb_entry(node, struct pending_reservation, rb_node);
1139 if (lclu < pr->lclu)
1140 node = node->rb_left;
1141 else if (lclu > pr->lclu)
1142 node = node->rb_right;
1146 if (pr && lclu < pr->lclu)
1148 if (pr && lclu > pr->lclu) {
1149 node = rb_next(&pr->rb_node);
1150 return node ? rb_entry(node, struct pending_reservation,
1157 * get_rsvd - calculates and returns the number of cluster reservations to be
1158 * released when removing a block range from the extent status tree
1159 * and releases any pending reservations within the range
1161 * @inode - file containing block range
1162 * @end - last block in range
1163 * @right_es - pointer to extent containing next block beyond end or NULL
1164 * @rc - pointer to reserved count data
1166 * The number of reservations to be released is equal to the number of
1167 * clusters containing delayed and not unwritten (delonly) blocks within
1168 * the range, minus the number of clusters still containing delonly blocks
1169 * at the ends of the range, and minus the number of pending reservations
1172 static unsigned int get_rsvd(struct inode *inode, ext4_lblk_t end,
1173 struct extent_status *right_es,
1174 struct rsvd_count *rc)
1176 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1177 struct pending_reservation *pr;
1178 struct ext4_pending_tree *tree = &EXT4_I(inode)->i_pending_tree;
1179 struct rb_node *node;
1180 ext4_lblk_t first_lclu, last_lclu;
1181 bool left_delonly, right_delonly, count_pending;
1182 struct extent_status *es;
1184 if (sbi->s_cluster_ratio > 1) {
1185 /* count any remaining partial cluster */
1189 if (rc->ndelonly == 0)
1192 first_lclu = EXT4_B2C(sbi, rc->first_do_lblk);
1193 last_lclu = EXT4_B2C(sbi, rc->last_do_lblk);
1196 * decrease the delonly count by the number of clusters at the
1197 * ends of the range that still contain delonly blocks -
1198 * these clusters still need to be reserved
1200 left_delonly = right_delonly = false;
1203 while (es && ext4_es_end(es) >=
1204 EXT4_LBLK_CMASK(sbi, rc->first_do_lblk)) {
1205 if (ext4_es_is_delonly(es)) {
1207 left_delonly = true;
1210 node = rb_prev(&es->rb_node);
1213 es = rb_entry(node, struct extent_status, rb_node);
1215 if (right_es && (!left_delonly || first_lclu != last_lclu)) {
1216 if (end < ext4_es_end(right_es)) {
1219 node = rb_next(&right_es->rb_node);
1220 es = node ? rb_entry(node, struct extent_status,
1223 while (es && es->es_lblk <=
1224 EXT4_LBLK_CFILL(sbi, rc->last_do_lblk)) {
1225 if (ext4_es_is_delonly(es)) {
1227 right_delonly = true;
1230 node = rb_next(&es->rb_node);
1233 es = rb_entry(node, struct extent_status,
1239 * Determine the block range that should be searched for
1240 * pending reservations, if any. Clusters on the ends of the
1241 * original removed range containing delonly blocks are
1242 * excluded. They've already been accounted for and it's not
1243 * possible to determine if an associated pending reservation
1244 * should be released with the information available in the
1245 * extents status tree.
1247 if (first_lclu == last_lclu) {
1248 if (left_delonly | right_delonly)
1249 count_pending = false;
1251 count_pending = true;
1257 if (first_lclu <= last_lclu)
1258 count_pending = true;
1260 count_pending = false;
1264 * a pending reservation found between first_lclu and last_lclu
1265 * represents an allocated cluster that contained at least one
1266 * delonly block, so the delonly total must be reduced by one
1267 * for each pending reservation found and released
1269 if (count_pending) {
1270 pr = __pr_tree_search(&tree->root, first_lclu);
1271 while (pr && pr->lclu <= last_lclu) {
1273 node = rb_next(&pr->rb_node);
1274 rb_erase(&pr->rb_node, &tree->root);
1275 kmem_cache_free(ext4_pending_cachep, pr);
1278 pr = rb_entry(node, struct pending_reservation,
1283 return rc->ndelonly;
1288 * __es_remove_extent - removes block range from extent status tree
1290 * @inode - file containing range
1291 * @lblk - first block in range
1292 * @end - last block in range
1293 * @reserved - number of cluster reservations released
1295 * If @reserved is not NULL and delayed allocation is enabled, counts
1296 * block/cluster reservations freed by removing range and if bigalloc
1297 * enabled cancels pending reservations as needed. Returns 0 on success,
1298 * error code on failure.
1300 static int __es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
1301 ext4_lblk_t end, int *reserved)
1303 struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
1304 struct rb_node *node;
1305 struct extent_status *es;
1306 struct extent_status orig_es;
1307 ext4_lblk_t len1, len2;
1310 bool count_reserved = true;
1311 struct rsvd_count rc;
1313 if (reserved == NULL || !test_opt(inode->i_sb, DELALLOC))
1314 count_reserved = false;
1318 es = __es_tree_search(&tree->root, lblk);
1321 if (es->es_lblk > end)
1324 /* Simply invalidate cache_es. */
1325 tree->cache_es = NULL;
1327 init_rsvd(inode, lblk, es, &rc);
1329 orig_es.es_lblk = es->es_lblk;
1330 orig_es.es_len = es->es_len;
1331 orig_es.es_pblk = es->es_pblk;
1333 len1 = lblk > es->es_lblk ? lblk - es->es_lblk : 0;
1334 len2 = ext4_es_end(es) > end ? ext4_es_end(es) - end : 0;
1339 struct extent_status newes;
1341 newes.es_lblk = end + 1;
1342 newes.es_len = len2;
1343 block = 0x7FDEADBEEFULL;
1344 if (ext4_es_is_written(&orig_es) ||
1345 ext4_es_is_unwritten(&orig_es))
1346 block = ext4_es_pblock(&orig_es) +
1347 orig_es.es_len - len2;
1348 ext4_es_store_pblock_status(&newes, block,
1349 ext4_es_status(&orig_es));
1350 err = __es_insert_extent(inode, &newes);
1352 es->es_lblk = orig_es.es_lblk;
1353 es->es_len = orig_es.es_len;
1354 if ((err == -ENOMEM) &&
1355 __es_shrink(EXT4_SB(inode->i_sb),
1356 128, EXT4_I(inode)))
1361 es->es_lblk = end + 1;
1363 if (ext4_es_is_written(es) ||
1364 ext4_es_is_unwritten(es)) {
1365 block = orig_es.es_pblk + orig_es.es_len - len2;
1366 ext4_es_store_pblock(es, block);
1370 count_rsvd(inode, lblk, orig_es.es_len - len1 - len2,
1372 goto out_get_reserved;
1377 count_rsvd(inode, lblk, orig_es.es_len - len1,
1379 node = rb_next(&es->rb_node);
1381 es = rb_entry(node, struct extent_status, rb_node);
1386 while (es && ext4_es_end(es) <= end) {
1388 count_rsvd(inode, es->es_lblk, es->es_len, es, &rc);
1389 node = rb_next(&es->rb_node);
1390 rb_erase(&es->rb_node, &tree->root);
1391 ext4_es_free_extent(inode, es);
1396 es = rb_entry(node, struct extent_status, rb_node);
1399 if (es && es->es_lblk < end + 1) {
1400 ext4_lblk_t orig_len = es->es_len;
1402 len1 = ext4_es_end(es) - end;
1404 count_rsvd(inode, es->es_lblk, orig_len - len1,
1406 es->es_lblk = end + 1;
1408 if (ext4_es_is_written(es) || ext4_es_is_unwritten(es)) {
1409 block = es->es_pblk + orig_len - len1;
1410 ext4_es_store_pblock(es, block);
1416 *reserved = get_rsvd(inode, end, es, &rc);
1422 * ext4_es_remove_extent - removes block range from extent status tree
1424 * @inode - file containing range
1425 * @lblk - first block in range
1426 * @len - number of blocks to remove
1428 * Reduces block/cluster reservation count and for bigalloc cancels pending
1429 * reservations as needed. Returns 0 on success, error code on failure.
1431 int ext4_es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
1438 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
1441 trace_ext4_es_remove_extent(inode, lblk, len);
1442 es_debug("remove [%u/%u) from extent status tree of inode %lu\n",
1443 lblk, len, inode->i_ino);
1448 end = lblk + len - 1;
1452 * ext4_clear_inode() depends on us taking i_es_lock unconditionally
1453 * so that we are sure __es_shrink() is done with the inode before it
1456 write_lock(&EXT4_I(inode)->i_es_lock);
1457 err = __es_remove_extent(inode, lblk, end, &reserved);
1458 write_unlock(&EXT4_I(inode)->i_es_lock);
1459 ext4_es_print_tree(inode);
1460 ext4_da_release_space(inode, reserved);
1464 static int __es_shrink(struct ext4_sb_info *sbi, int nr_to_scan,
1465 struct ext4_inode_info *locked_ei)
1467 struct ext4_inode_info *ei;
1468 struct ext4_es_stats *es_stats;
1473 int retried = 0, nr_skipped = 0;
1475 es_stats = &sbi->s_es_stats;
1476 start_time = ktime_get();
1479 spin_lock(&sbi->s_es_lock);
1480 nr_to_walk = sbi->s_es_nr_inode;
1481 while (nr_to_walk-- > 0) {
1482 if (list_empty(&sbi->s_es_list)) {
1483 spin_unlock(&sbi->s_es_lock);
1486 ei = list_first_entry(&sbi->s_es_list, struct ext4_inode_info,
1488 /* Move the inode to the tail */
1489 list_move_tail(&ei->i_es_list, &sbi->s_es_list);
1492 * Normally we try hard to avoid shrinking precached inodes,
1493 * but we will as a last resort.
1495 if (!retried && ext4_test_inode_state(&ei->vfs_inode,
1496 EXT4_STATE_EXT_PRECACHED)) {
1501 if (ei == locked_ei || !write_trylock(&ei->i_es_lock)) {
1506 * Now we hold i_es_lock which protects us from inode reclaim
1507 * freeing inode under us
1509 spin_unlock(&sbi->s_es_lock);
1511 nr_shrunk += es_reclaim_extents(ei, &nr_to_scan);
1512 write_unlock(&ei->i_es_lock);
1514 if (nr_to_scan <= 0)
1516 spin_lock(&sbi->s_es_lock);
1518 spin_unlock(&sbi->s_es_lock);
1521 * If we skipped any inodes, and we weren't able to make any
1522 * forward progress, try again to scan precached inodes.
1524 if ((nr_shrunk == 0) && nr_skipped && !retried) {
1529 if (locked_ei && nr_shrunk == 0)
1530 nr_shrunk = es_reclaim_extents(locked_ei, &nr_to_scan);
1533 scan_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1534 if (likely(es_stats->es_stats_scan_time))
1535 es_stats->es_stats_scan_time = (scan_time +
1536 es_stats->es_stats_scan_time*3) / 4;
1538 es_stats->es_stats_scan_time = scan_time;
1539 if (scan_time > es_stats->es_stats_max_scan_time)
1540 es_stats->es_stats_max_scan_time = scan_time;
1541 if (likely(es_stats->es_stats_shrunk))
1542 es_stats->es_stats_shrunk = (nr_shrunk +
1543 es_stats->es_stats_shrunk*3) / 4;
1545 es_stats->es_stats_shrunk = nr_shrunk;
1547 trace_ext4_es_shrink(sbi->s_sb, nr_shrunk, scan_time,
1548 nr_skipped, retried);
1552 static unsigned long ext4_es_count(struct shrinker *shrink,
1553 struct shrink_control *sc)
1556 struct ext4_sb_info *sbi;
1558 sbi = container_of(shrink, struct ext4_sb_info, s_es_shrinker);
1559 nr = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt);
1560 trace_ext4_es_shrink_count(sbi->s_sb, sc->nr_to_scan, nr);
1564 static unsigned long ext4_es_scan(struct shrinker *shrink,
1565 struct shrink_control *sc)
1567 struct ext4_sb_info *sbi = container_of(shrink,
1568 struct ext4_sb_info, s_es_shrinker);
1569 int nr_to_scan = sc->nr_to_scan;
1572 ret = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt);
1573 trace_ext4_es_shrink_scan_enter(sbi->s_sb, nr_to_scan, ret);
1575 nr_shrunk = __es_shrink(sbi, nr_to_scan, NULL);
1577 ret = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt);
1578 trace_ext4_es_shrink_scan_exit(sbi->s_sb, nr_shrunk, ret);
1582 int ext4_seq_es_shrinker_info_show(struct seq_file *seq, void *v)
1584 struct ext4_sb_info *sbi = EXT4_SB((struct super_block *) seq->private);
1585 struct ext4_es_stats *es_stats = &sbi->s_es_stats;
1586 struct ext4_inode_info *ei, *max = NULL;
1587 unsigned int inode_cnt = 0;
1589 if (v != SEQ_START_TOKEN)
1592 /* here we just find an inode that has the max nr. of objects */
1593 spin_lock(&sbi->s_es_lock);
1594 list_for_each_entry(ei, &sbi->s_es_list, i_es_list) {
1596 if (max && max->i_es_all_nr < ei->i_es_all_nr)
1601 spin_unlock(&sbi->s_es_lock);
1603 seq_printf(seq, "stats:\n %lld objects\n %lld reclaimable objects\n",
1604 percpu_counter_sum_positive(&es_stats->es_stats_all_cnt),
1605 percpu_counter_sum_positive(&es_stats->es_stats_shk_cnt));
1606 seq_printf(seq, " %lld/%lld cache hits/misses\n",
1607 percpu_counter_sum_positive(&es_stats->es_stats_cache_hits),
1608 percpu_counter_sum_positive(&es_stats->es_stats_cache_misses));
1610 seq_printf(seq, " %d inodes on list\n", inode_cnt);
1612 seq_printf(seq, "average:\n %llu us scan time\n",
1613 div_u64(es_stats->es_stats_scan_time, 1000));
1614 seq_printf(seq, " %lu shrunk objects\n", es_stats->es_stats_shrunk);
1617 "maximum:\n %lu inode (%u objects, %u reclaimable)\n"
1618 " %llu us max scan time\n",
1619 max->vfs_inode.i_ino, max->i_es_all_nr, max->i_es_shk_nr,
1620 div_u64(es_stats->es_stats_max_scan_time, 1000));
1625 int ext4_es_register_shrinker(struct ext4_sb_info *sbi)
1629 /* Make sure we have enough bits for physical block number */
1630 BUILD_BUG_ON(ES_SHIFT < 48);
1631 INIT_LIST_HEAD(&sbi->s_es_list);
1632 sbi->s_es_nr_inode = 0;
1633 spin_lock_init(&sbi->s_es_lock);
1634 sbi->s_es_stats.es_stats_shrunk = 0;
1635 err = percpu_counter_init(&sbi->s_es_stats.es_stats_cache_hits, 0,
1639 err = percpu_counter_init(&sbi->s_es_stats.es_stats_cache_misses, 0,
1643 sbi->s_es_stats.es_stats_scan_time = 0;
1644 sbi->s_es_stats.es_stats_max_scan_time = 0;
1645 err = percpu_counter_init(&sbi->s_es_stats.es_stats_all_cnt, 0, GFP_KERNEL);
1648 err = percpu_counter_init(&sbi->s_es_stats.es_stats_shk_cnt, 0, GFP_KERNEL);
1652 sbi->s_es_shrinker.scan_objects = ext4_es_scan;
1653 sbi->s_es_shrinker.count_objects = ext4_es_count;
1654 sbi->s_es_shrinker.seeks = DEFAULT_SEEKS;
1655 err = register_shrinker(&sbi->s_es_shrinker, "ext4-es:%s",
1662 percpu_counter_destroy(&sbi->s_es_stats.es_stats_shk_cnt);
1664 percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt);
1666 percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_misses);
1668 percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_hits);
1672 void ext4_es_unregister_shrinker(struct ext4_sb_info *sbi)
1674 percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_hits);
1675 percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_misses);
1676 percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt);
1677 percpu_counter_destroy(&sbi->s_es_stats.es_stats_shk_cnt);
1678 unregister_shrinker(&sbi->s_es_shrinker);
1682 * Shrink extents in given inode from ei->i_es_shrink_lblk till end. Scan at
1683 * most *nr_to_scan extents, update *nr_to_scan accordingly.
1685 * Return 0 if we hit end of tree / interval, 1 if we exhausted nr_to_scan.
1686 * Increment *nr_shrunk by the number of reclaimed extents. Also update
1687 * ei->i_es_shrink_lblk to where we should continue scanning.
1689 static int es_do_reclaim_extents(struct ext4_inode_info *ei, ext4_lblk_t end,
1690 int *nr_to_scan, int *nr_shrunk)
1692 struct inode *inode = &ei->vfs_inode;
1693 struct ext4_es_tree *tree = &ei->i_es_tree;
1694 struct extent_status *es;
1695 struct rb_node *node;
1697 es = __es_tree_search(&tree->root, ei->i_es_shrink_lblk);
1701 while (*nr_to_scan > 0) {
1702 if (es->es_lblk > end) {
1703 ei->i_es_shrink_lblk = end + 1;
1708 node = rb_next(&es->rb_node);
1710 * We can't reclaim delayed extent from status tree because
1711 * fiemap, bigallic, and seek_data/hole need to use it.
1713 if (ext4_es_is_delayed(es))
1715 if (ext4_es_is_referenced(es)) {
1716 ext4_es_clear_referenced(es);
1720 rb_erase(&es->rb_node, &tree->root);
1721 ext4_es_free_extent(inode, es);
1726 es = rb_entry(node, struct extent_status, rb_node);
1728 ei->i_es_shrink_lblk = es->es_lblk;
1731 ei->i_es_shrink_lblk = 0;
1735 static int es_reclaim_extents(struct ext4_inode_info *ei, int *nr_to_scan)
1737 struct inode *inode = &ei->vfs_inode;
1739 ext4_lblk_t start = ei->i_es_shrink_lblk;
1740 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
1741 DEFAULT_RATELIMIT_BURST);
1743 if (ei->i_es_shk_nr == 0)
1746 if (ext4_test_inode_state(inode, EXT4_STATE_EXT_PRECACHED) &&
1748 ext4_warning(inode->i_sb, "forced shrink of precached extents");
1750 if (!es_do_reclaim_extents(ei, EXT_MAX_BLOCKS, nr_to_scan, &nr_shrunk) &&
1752 es_do_reclaim_extents(ei, start - 1, nr_to_scan, &nr_shrunk);
1754 ei->i_es_tree.cache_es = NULL;
1759 * Called to support EXT4_IOC_CLEAR_ES_CACHE. We can only remove
1760 * discretionary entries from the extent status cache. (Some entries
1761 * must be present for proper operations.)
1763 void ext4_clear_inode_es(struct inode *inode)
1765 struct ext4_inode_info *ei = EXT4_I(inode);
1766 struct extent_status *es;
1767 struct ext4_es_tree *tree;
1768 struct rb_node *node;
1770 write_lock(&ei->i_es_lock);
1771 tree = &EXT4_I(inode)->i_es_tree;
1772 tree->cache_es = NULL;
1773 node = rb_first(&tree->root);
1775 es = rb_entry(node, struct extent_status, rb_node);
1776 node = rb_next(node);
1777 if (!ext4_es_is_delayed(es)) {
1778 rb_erase(&es->rb_node, &tree->root);
1779 ext4_es_free_extent(inode, es);
1782 ext4_clear_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
1783 write_unlock(&ei->i_es_lock);
1787 static void ext4_print_pending_tree(struct inode *inode)
1789 struct ext4_pending_tree *tree;
1790 struct rb_node *node;
1791 struct pending_reservation *pr;
1793 printk(KERN_DEBUG "pending reservations for inode %lu:", inode->i_ino);
1794 tree = &EXT4_I(inode)->i_pending_tree;
1795 node = rb_first(&tree->root);
1797 pr = rb_entry(node, struct pending_reservation, rb_node);
1798 printk(KERN_DEBUG " %u", pr->lclu);
1799 node = rb_next(node);
1801 printk(KERN_DEBUG "\n");
1804 #define ext4_print_pending_tree(inode)
1807 int __init ext4_init_pending(void)
1809 ext4_pending_cachep = KMEM_CACHE(pending_reservation, SLAB_RECLAIM_ACCOUNT);
1810 if (ext4_pending_cachep == NULL)
1815 void ext4_exit_pending(void)
1817 kmem_cache_destroy(ext4_pending_cachep);
1820 void ext4_init_pending_tree(struct ext4_pending_tree *tree)
1822 tree->root = RB_ROOT;
1826 * __get_pending - retrieve a pointer to a pending reservation
1828 * @inode - file containing the pending cluster reservation
1829 * @lclu - logical cluster of interest
1831 * Returns a pointer to a pending reservation if it's a member of
1832 * the set, and NULL if not. Must be called holding i_es_lock.
1834 static struct pending_reservation *__get_pending(struct inode *inode,
1837 struct ext4_pending_tree *tree;
1838 struct rb_node *node;
1839 struct pending_reservation *pr = NULL;
1841 tree = &EXT4_I(inode)->i_pending_tree;
1842 node = (&tree->root)->rb_node;
1845 pr = rb_entry(node, struct pending_reservation, rb_node);
1846 if (lclu < pr->lclu)
1847 node = node->rb_left;
1848 else if (lclu > pr->lclu)
1849 node = node->rb_right;
1850 else if (lclu == pr->lclu)
1857 * __insert_pending - adds a pending cluster reservation to the set of
1858 * pending reservations
1860 * @inode - file containing the cluster
1861 * @lblk - logical block in the cluster to be added
1863 * Returns 0 on successful insertion and -ENOMEM on failure. If the
1864 * pending reservation is already in the set, returns successfully.
1866 static int __insert_pending(struct inode *inode, ext4_lblk_t lblk)
1868 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1869 struct ext4_pending_tree *tree = &EXT4_I(inode)->i_pending_tree;
1870 struct rb_node **p = &tree->root.rb_node;
1871 struct rb_node *parent = NULL;
1872 struct pending_reservation *pr;
1876 lclu = EXT4_B2C(sbi, lblk);
1877 /* search to find parent for insertion */
1880 pr = rb_entry(parent, struct pending_reservation, rb_node);
1882 if (lclu < pr->lclu) {
1884 } else if (lclu > pr->lclu) {
1885 p = &(*p)->rb_right;
1887 /* pending reservation already inserted */
1892 pr = kmem_cache_alloc(ext4_pending_cachep, GFP_ATOMIC);
1899 rb_link_node(&pr->rb_node, parent, p);
1900 rb_insert_color(&pr->rb_node, &tree->root);
1907 * __remove_pending - removes a pending cluster reservation from the set
1908 * of pending reservations
1910 * @inode - file containing the cluster
1911 * @lblk - logical block in the pending cluster reservation to be removed
1913 * Returns successfully if pending reservation is not a member of the set.
1915 static void __remove_pending(struct inode *inode, ext4_lblk_t lblk)
1917 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1918 struct pending_reservation *pr;
1919 struct ext4_pending_tree *tree;
1921 pr = __get_pending(inode, EXT4_B2C(sbi, lblk));
1923 tree = &EXT4_I(inode)->i_pending_tree;
1924 rb_erase(&pr->rb_node, &tree->root);
1925 kmem_cache_free(ext4_pending_cachep, pr);
1930 * ext4_remove_pending - removes a pending cluster reservation from the set
1931 * of pending reservations
1933 * @inode - file containing the cluster
1934 * @lblk - logical block in the pending cluster reservation to be removed
1936 * Locking for external use of __remove_pending.
1938 void ext4_remove_pending(struct inode *inode, ext4_lblk_t lblk)
1940 struct ext4_inode_info *ei = EXT4_I(inode);
1942 write_lock(&ei->i_es_lock);
1943 __remove_pending(inode, lblk);
1944 write_unlock(&ei->i_es_lock);
1948 * ext4_is_pending - determine whether a cluster has a pending reservation
1951 * @inode - file containing the cluster
1952 * @lblk - logical block in the cluster
1954 * Returns true if there's a pending reservation for the cluster in the
1955 * set of pending reservations, and false if not.
1957 bool ext4_is_pending(struct inode *inode, ext4_lblk_t lblk)
1959 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1960 struct ext4_inode_info *ei = EXT4_I(inode);
1963 read_lock(&ei->i_es_lock);
1964 ret = (bool)(__get_pending(inode, EXT4_B2C(sbi, lblk)) != NULL);
1965 read_unlock(&ei->i_es_lock);
1971 * ext4_es_insert_delayed_block - adds a delayed block to the extents status
1972 * tree, adding a pending reservation where
1975 * @inode - file containing the newly added block
1976 * @lblk - logical block to be added
1977 * @allocated - indicates whether a physical cluster has been allocated for
1978 * the logical cluster that contains the block
1980 * Returns 0 on success, negative error code on failure.
1982 int ext4_es_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk,
1985 struct extent_status newes;
1988 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
1991 es_debug("add [%u/1) delayed to extent status tree of inode %lu\n",
1992 lblk, inode->i_ino);
1994 newes.es_lblk = lblk;
1996 ext4_es_store_pblock_status(&newes, ~0, EXTENT_STATUS_DELAYED);
1997 trace_ext4_es_insert_delayed_block(inode, &newes, allocated);
1999 ext4_es_insert_extent_check(inode, &newes);
2001 write_lock(&EXT4_I(inode)->i_es_lock);
2003 err = __es_remove_extent(inode, lblk, lblk, NULL);
2007 err = __es_insert_extent(inode, &newes);
2008 if (err == -ENOMEM && __es_shrink(EXT4_SB(inode->i_sb),
2009 128, EXT4_I(inode)))
2015 __insert_pending(inode, lblk);
2018 write_unlock(&EXT4_I(inode)->i_es_lock);
2020 ext4_es_print_tree(inode);
2021 ext4_print_pending_tree(inode);
2027 * __es_delayed_clu - count number of clusters containing blocks that
2030 * @inode - file containing block range
2031 * @start - logical block defining start of range
2032 * @end - logical block defining end of range
2034 * Returns the number of clusters containing only delayed (not delayed
2035 * and unwritten) blocks in the range specified by @start and @end. Any
2036 * cluster or part of a cluster within the range and containing a delayed
2037 * and not unwritten block within the range is counted as a whole cluster.
2039 static unsigned int __es_delayed_clu(struct inode *inode, ext4_lblk_t start,
2042 struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
2043 struct extent_status *es;
2044 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2045 struct rb_node *node;
2046 ext4_lblk_t first_lclu, last_lclu;
2047 unsigned long long last_counted_lclu;
2050 /* guaranteed to be unequal to any ext4_lblk_t value */
2051 last_counted_lclu = ~0ULL;
2053 es = __es_tree_search(&tree->root, start);
2055 while (es && (es->es_lblk <= end)) {
2056 if (ext4_es_is_delonly(es)) {
2057 if (es->es_lblk <= start)
2058 first_lclu = EXT4_B2C(sbi, start);
2060 first_lclu = EXT4_B2C(sbi, es->es_lblk);
2062 if (ext4_es_end(es) >= end)
2063 last_lclu = EXT4_B2C(sbi, end);
2065 last_lclu = EXT4_B2C(sbi, ext4_es_end(es));
2067 if (first_lclu == last_counted_lclu)
2068 n += last_lclu - first_lclu;
2070 n += last_lclu - first_lclu + 1;
2071 last_counted_lclu = last_lclu;
2073 node = rb_next(&es->rb_node);
2076 es = rb_entry(node, struct extent_status, rb_node);
2083 * ext4_es_delayed_clu - count number of clusters containing blocks that
2084 * are both delayed and unwritten
2086 * @inode - file containing block range
2087 * @lblk - logical block defining start of range
2088 * @len - number of blocks in range
2090 * Locking for external use of __es_delayed_clu().
2092 unsigned int ext4_es_delayed_clu(struct inode *inode, ext4_lblk_t lblk,
2095 struct ext4_inode_info *ei = EXT4_I(inode);
2102 end = lblk + len - 1;
2103 WARN_ON(end < lblk);
2105 read_lock(&ei->i_es_lock);
2107 n = __es_delayed_clu(inode, lblk, end);
2109 read_unlock(&ei->i_es_lock);
2115 * __revise_pending - makes, cancels, or leaves unchanged pending cluster
2116 * reservations for a specified block range depending
2117 * upon the presence or absence of delayed blocks
2118 * outside the range within clusters at the ends of the
2121 * @inode - file containing the range
2122 * @lblk - logical block defining the start of range
2123 * @len - length of range in blocks
2125 * Used after a newly allocated extent is added to the extents status tree.
2126 * Requires that the extents in the range have either written or unwritten
2127 * status. Must be called while holding i_es_lock.
2129 static void __revise_pending(struct inode *inode, ext4_lblk_t lblk,
2132 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2133 ext4_lblk_t end = lblk + len - 1;
2134 ext4_lblk_t first, last;
2135 bool f_del = false, l_del = false;
2141 * Two cases - block range within single cluster and block range
2142 * spanning two or more clusters. Note that a cluster belonging
2143 * to a range starting and/or ending on a cluster boundary is treated
2144 * as if it does not contain a delayed extent. The new range may
2145 * have allocated space for previously delayed blocks out to the
2146 * cluster boundary, requiring that any pre-existing pending
2147 * reservation be canceled. Because this code only looks at blocks
2148 * outside the range, it should revise pending reservations
2149 * correctly even if the extent represented by the range can't be
2150 * inserted in the extents status tree due to ENOSPC.
2153 if (EXT4_B2C(sbi, lblk) == EXT4_B2C(sbi, end)) {
2154 first = EXT4_LBLK_CMASK(sbi, lblk);
2156 f_del = __es_scan_range(inode, &ext4_es_is_delonly,
2159 __insert_pending(inode, first);
2161 last = EXT4_LBLK_CMASK(sbi, end) +
2162 sbi->s_cluster_ratio - 1;
2164 l_del = __es_scan_range(inode,
2165 &ext4_es_is_delonly,
2168 __insert_pending(inode, last);
2170 __remove_pending(inode, last);
2173 first = EXT4_LBLK_CMASK(sbi, lblk);
2175 f_del = __es_scan_range(inode, &ext4_es_is_delonly,
2178 __insert_pending(inode, first);
2180 __remove_pending(inode, first);
2182 last = EXT4_LBLK_CMASK(sbi, end) + sbi->s_cluster_ratio - 1;
2184 l_del = __es_scan_range(inode, &ext4_es_is_delonly,
2187 __insert_pending(inode, last);
2189 __remove_pending(inode, last);