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 struct extent_status *prealloc);
149 static int __es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
150 ext4_lblk_t end, int *reserved,
151 struct extent_status *prealloc);
152 static int es_reclaim_extents(struct ext4_inode_info *ei, int *nr_to_scan);
153 static int __es_shrink(struct ext4_sb_info *sbi, int nr_to_scan,
154 struct ext4_inode_info *locked_ei);
155 static void __revise_pending(struct inode *inode, ext4_lblk_t lblk,
158 int __init ext4_init_es(void)
160 ext4_es_cachep = KMEM_CACHE(extent_status, SLAB_RECLAIM_ACCOUNT);
161 if (ext4_es_cachep == NULL)
166 void ext4_exit_es(void)
168 kmem_cache_destroy(ext4_es_cachep);
171 void ext4_es_init_tree(struct ext4_es_tree *tree)
173 tree->root = RB_ROOT;
174 tree->cache_es = NULL;
178 static void ext4_es_print_tree(struct inode *inode)
180 struct ext4_es_tree *tree;
181 struct rb_node *node;
183 printk(KERN_DEBUG "status extents for inode %lu:", inode->i_ino);
184 tree = &EXT4_I(inode)->i_es_tree;
185 node = rb_first(&tree->root);
187 struct extent_status *es;
188 es = rb_entry(node, struct extent_status, rb_node);
189 printk(KERN_DEBUG " [%u/%u) %llu %x",
190 es->es_lblk, es->es_len,
191 ext4_es_pblock(es), ext4_es_status(es));
192 node = rb_next(node);
194 printk(KERN_DEBUG "\n");
197 #define ext4_es_print_tree(inode)
200 static inline ext4_lblk_t ext4_es_end(struct extent_status *es)
202 BUG_ON(es->es_lblk + es->es_len < es->es_lblk);
203 return es->es_lblk + es->es_len - 1;
207 * search through the tree for an delayed extent with a given offset. If
208 * it can't be found, try to find next extent.
210 static struct extent_status *__es_tree_search(struct rb_root *root,
213 struct rb_node *node = root->rb_node;
214 struct extent_status *es = NULL;
217 es = rb_entry(node, struct extent_status, rb_node);
218 if (lblk < es->es_lblk)
219 node = node->rb_left;
220 else if (lblk > ext4_es_end(es))
221 node = node->rb_right;
226 if (es && lblk < es->es_lblk)
229 if (es && lblk > ext4_es_end(es)) {
230 node = rb_next(&es->rb_node);
231 return node ? rb_entry(node, struct extent_status, rb_node) :
239 * ext4_es_find_extent_range - find extent with specified status within block
240 * range or next extent following block range in
241 * extents status tree
243 * @inode - file containing the range
244 * @matching_fn - pointer to function that matches extents with desired status
245 * @lblk - logical block defining start of range
246 * @end - logical block defining end of range
247 * @es - extent found, if any
249 * Find the first extent within the block range specified by @lblk and @end
250 * in the extents status tree that satisfies @matching_fn. If a match
251 * is found, it's returned in @es. If not, and a matching extent is found
252 * beyond the block range, it's returned in @es. If no match is found, an
253 * extent is returned in @es whose es_lblk, es_len, and es_pblk components
256 static void __es_find_extent_range(struct inode *inode,
257 int (*matching_fn)(struct extent_status *es),
258 ext4_lblk_t lblk, ext4_lblk_t end,
259 struct extent_status *es)
261 struct ext4_es_tree *tree = NULL;
262 struct extent_status *es1 = NULL;
263 struct rb_node *node;
268 tree = &EXT4_I(inode)->i_es_tree;
270 /* see if the extent has been cached */
271 es->es_lblk = es->es_len = es->es_pblk = 0;
272 es1 = READ_ONCE(tree->cache_es);
273 if (es1 && in_range(lblk, es1->es_lblk, es1->es_len)) {
274 es_debug("%u cached by [%u/%u) %llu %x\n",
275 lblk, es1->es_lblk, es1->es_len,
276 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 WRITE_ONCE(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);
452 * Returns true if we cannot fail to allocate memory for this extent_status
453 * entry and cannot reclaim it until its status changes.
455 static inline bool ext4_es_must_keep(struct extent_status *es)
457 /* fiemap, bigalloc, and seek_data/hole need to use it. */
458 if (ext4_es_is_delayed(es))
464 static inline struct extent_status *__es_alloc_extent(bool nofail)
467 return kmem_cache_alloc(ext4_es_cachep, GFP_ATOMIC);
469 return kmem_cache_zalloc(ext4_es_cachep, GFP_KERNEL | __GFP_NOFAIL);
472 static void ext4_es_init_extent(struct inode *inode, struct extent_status *es,
473 ext4_lblk_t lblk, ext4_lblk_t len, ext4_fsblk_t pblk)
479 /* We never try to reclaim a must kept extent, so we don't count it. */
480 if (!ext4_es_must_keep(es)) {
481 if (!EXT4_I(inode)->i_es_shk_nr++)
482 ext4_es_list_add(inode);
483 percpu_counter_inc(&EXT4_SB(inode->i_sb)->
484 s_es_stats.es_stats_shk_cnt);
487 EXT4_I(inode)->i_es_all_nr++;
488 percpu_counter_inc(&EXT4_SB(inode->i_sb)->s_es_stats.es_stats_all_cnt);
491 static inline void __es_free_extent(struct extent_status *es)
493 kmem_cache_free(ext4_es_cachep, es);
496 static void ext4_es_free_extent(struct inode *inode, struct extent_status *es)
498 EXT4_I(inode)->i_es_all_nr--;
499 percpu_counter_dec(&EXT4_SB(inode->i_sb)->s_es_stats.es_stats_all_cnt);
501 /* Decrease the shrink counter when we can reclaim the extent. */
502 if (!ext4_es_must_keep(es)) {
503 BUG_ON(EXT4_I(inode)->i_es_shk_nr == 0);
504 if (!--EXT4_I(inode)->i_es_shk_nr)
505 ext4_es_list_del(inode);
506 percpu_counter_dec(&EXT4_SB(inode->i_sb)->
507 s_es_stats.es_stats_shk_cnt);
510 __es_free_extent(es);
514 * Check whether or not two extents can be merged
516 * - logical block number is contiguous
517 * - physical block number is contiguous
520 static int ext4_es_can_be_merged(struct extent_status *es1,
521 struct extent_status *es2)
523 if (ext4_es_type(es1) != ext4_es_type(es2))
526 if (((__u64) es1->es_len) + es2->es_len > EXT_MAX_BLOCKS) {
527 pr_warn("ES assertion failed when merging extents. "
528 "The sum of lengths of es1 (%d) and es2 (%d) "
529 "is bigger than allowed file size (%d)\n",
530 es1->es_len, es2->es_len, EXT_MAX_BLOCKS);
535 if (((__u64) es1->es_lblk) + es1->es_len != es2->es_lblk)
538 if ((ext4_es_is_written(es1) || ext4_es_is_unwritten(es1)) &&
539 (ext4_es_pblock(es1) + es1->es_len == ext4_es_pblock(es2)))
542 if (ext4_es_is_hole(es1))
545 /* we need to check delayed extent is without unwritten status */
546 if (ext4_es_is_delayed(es1) && !ext4_es_is_unwritten(es1))
552 static struct extent_status *
553 ext4_es_try_to_merge_left(struct inode *inode, struct extent_status *es)
555 struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
556 struct extent_status *es1;
557 struct rb_node *node;
559 node = rb_prev(&es->rb_node);
563 es1 = rb_entry(node, struct extent_status, rb_node);
564 if (ext4_es_can_be_merged(es1, es)) {
565 es1->es_len += es->es_len;
566 if (ext4_es_is_referenced(es))
567 ext4_es_set_referenced(es1);
568 rb_erase(&es->rb_node, &tree->root);
569 ext4_es_free_extent(inode, es);
576 static struct extent_status *
577 ext4_es_try_to_merge_right(struct inode *inode, struct extent_status *es)
579 struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
580 struct extent_status *es1;
581 struct rb_node *node;
583 node = rb_next(&es->rb_node);
587 es1 = rb_entry(node, struct extent_status, rb_node);
588 if (ext4_es_can_be_merged(es, es1)) {
589 es->es_len += es1->es_len;
590 if (ext4_es_is_referenced(es1))
591 ext4_es_set_referenced(es);
592 rb_erase(node, &tree->root);
593 ext4_es_free_extent(inode, es1);
599 #ifdef ES_AGGRESSIVE_TEST
600 #include "ext4_extents.h" /* Needed when ES_AGGRESSIVE_TEST is defined */
602 static void ext4_es_insert_extent_ext_check(struct inode *inode,
603 struct extent_status *es)
605 struct ext4_ext_path *path = NULL;
606 struct ext4_extent *ex;
607 ext4_lblk_t ee_block;
608 ext4_fsblk_t ee_start;
609 unsigned short ee_len;
610 int depth, ee_status, es_status;
612 path = ext4_find_extent(inode, es->es_lblk, NULL, EXT4_EX_NOCACHE);
616 depth = ext_depth(inode);
617 ex = path[depth].p_ext;
621 ee_block = le32_to_cpu(ex->ee_block);
622 ee_start = ext4_ext_pblock(ex);
623 ee_len = ext4_ext_get_actual_len(ex);
625 ee_status = ext4_ext_is_unwritten(ex) ? 1 : 0;
626 es_status = ext4_es_is_unwritten(es) ? 1 : 0;
629 * Make sure ex and es are not overlap when we try to insert
630 * a delayed/hole extent.
632 if (!ext4_es_is_written(es) && !ext4_es_is_unwritten(es)) {
633 if (in_range(es->es_lblk, ee_block, ee_len)) {
634 pr_warn("ES insert assertion failed for "
635 "inode: %lu we can find an extent "
636 "at block [%d/%d/%llu/%c], but we "
637 "want to add a delayed/hole extent "
639 inode->i_ino, ee_block, ee_len,
640 ee_start, ee_status ? 'u' : 'w',
641 es->es_lblk, es->es_len,
642 ext4_es_pblock(es), ext4_es_status(es));
648 * We don't check ee_block == es->es_lblk, etc. because es
649 * might be a part of whole extent, vice versa.
651 if (es->es_lblk < ee_block ||
652 ext4_es_pblock(es) != ee_start + es->es_lblk - ee_block) {
653 pr_warn("ES insert assertion failed for inode: %lu "
654 "ex_status [%d/%d/%llu/%c] != "
655 "es_status [%d/%d/%llu/%c]\n", inode->i_ino,
656 ee_block, ee_len, ee_start,
657 ee_status ? 'u' : 'w', es->es_lblk, es->es_len,
658 ext4_es_pblock(es), es_status ? 'u' : 'w');
662 if (ee_status ^ es_status) {
663 pr_warn("ES insert assertion failed for inode: %lu "
664 "ex_status [%d/%d/%llu/%c] != "
665 "es_status [%d/%d/%llu/%c]\n", inode->i_ino,
666 ee_block, ee_len, ee_start,
667 ee_status ? 'u' : 'w', es->es_lblk, es->es_len,
668 ext4_es_pblock(es), es_status ? 'u' : 'w');
672 * We can't find an extent on disk. So we need to make sure
673 * that we don't want to add an written/unwritten extent.
675 if (!ext4_es_is_delayed(es) && !ext4_es_is_hole(es)) {
676 pr_warn("ES insert assertion failed for inode: %lu "
677 "can't find an extent at block %d but we want "
678 "to add a written/unwritten extent "
679 "[%d/%d/%llu/%x]\n", inode->i_ino,
680 es->es_lblk, es->es_lblk, es->es_len,
681 ext4_es_pblock(es), ext4_es_status(es));
685 ext4_free_ext_path(path);
688 static void ext4_es_insert_extent_ind_check(struct inode *inode,
689 struct extent_status *es)
691 struct ext4_map_blocks map;
695 * Here we call ext4_ind_map_blocks to lookup a block mapping because
696 * 'Indirect' structure is defined in indirect.c. So we couldn't
697 * access direct/indirect tree from outside. It is too dirty to define
698 * this function in indirect.c file.
701 map.m_lblk = es->es_lblk;
702 map.m_len = es->es_len;
704 retval = ext4_ind_map_blocks(NULL, inode, &map, 0);
706 if (ext4_es_is_delayed(es) || ext4_es_is_hole(es)) {
708 * We want to add a delayed/hole extent but this
709 * block has been allocated.
711 pr_warn("ES insert assertion failed for inode: %lu "
712 "We can find blocks but we want to add a "
713 "delayed/hole extent [%d/%d/%llu/%x]\n",
714 inode->i_ino, es->es_lblk, es->es_len,
715 ext4_es_pblock(es), ext4_es_status(es));
717 } else if (ext4_es_is_written(es)) {
718 if (retval != es->es_len) {
719 pr_warn("ES insert assertion failed for "
720 "inode: %lu retval %d != es_len %d\n",
721 inode->i_ino, retval, es->es_len);
724 if (map.m_pblk != ext4_es_pblock(es)) {
725 pr_warn("ES insert assertion failed for "
726 "inode: %lu m_pblk %llu != "
728 inode->i_ino, map.m_pblk,
734 * We don't need to check unwritten extent because
735 * indirect-based file doesn't have it.
739 } else if (retval == 0) {
740 if (ext4_es_is_written(es)) {
741 pr_warn("ES insert assertion failed for inode: %lu "
742 "We can't find the block but we want to add "
743 "a written extent [%d/%d/%llu/%x]\n",
744 inode->i_ino, es->es_lblk, es->es_len,
745 ext4_es_pblock(es), ext4_es_status(es));
751 static inline void ext4_es_insert_extent_check(struct inode *inode,
752 struct extent_status *es)
755 * We don't need to worry about the race condition because
756 * caller takes i_data_sem locking.
758 BUG_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
759 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
760 ext4_es_insert_extent_ext_check(inode, es);
762 ext4_es_insert_extent_ind_check(inode, es);
765 static inline void ext4_es_insert_extent_check(struct inode *inode,
766 struct extent_status *es)
771 static int __es_insert_extent(struct inode *inode, struct extent_status *newes,
772 struct extent_status *prealloc)
774 struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
775 struct rb_node **p = &tree->root.rb_node;
776 struct rb_node *parent = NULL;
777 struct extent_status *es;
781 es = rb_entry(parent, struct extent_status, rb_node);
783 if (newes->es_lblk < es->es_lblk) {
784 if (ext4_es_can_be_merged(newes, es)) {
786 * Here we can modify es_lblk directly
787 * because it isn't overlapped.
789 es->es_lblk = newes->es_lblk;
790 es->es_len += newes->es_len;
791 if (ext4_es_is_written(es) ||
792 ext4_es_is_unwritten(es))
793 ext4_es_store_pblock(es,
795 es = ext4_es_try_to_merge_left(inode, es);
799 } else if (newes->es_lblk > ext4_es_end(es)) {
800 if (ext4_es_can_be_merged(es, newes)) {
801 es->es_len += newes->es_len;
802 es = ext4_es_try_to_merge_right(inode, es);
815 es = __es_alloc_extent(false);
818 ext4_es_init_extent(inode, es, newes->es_lblk, newes->es_len,
821 rb_link_node(&es->rb_node, parent, p);
822 rb_insert_color(&es->rb_node, &tree->root);
830 * ext4_es_insert_extent() adds information to an inode's extent
833 void ext4_es_insert_extent(struct inode *inode, ext4_lblk_t lblk,
834 ext4_lblk_t len, ext4_fsblk_t pblk,
837 struct extent_status newes;
838 ext4_lblk_t end = lblk + len - 1;
841 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
842 struct extent_status *es1 = NULL;
843 struct extent_status *es2 = NULL;
845 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
848 es_debug("add [%u/%u) %llu %x to extent status tree of inode %lu\n",
849 lblk, len, pblk, status, inode->i_ino);
856 if ((status & EXTENT_STATUS_DELAYED) &&
857 (status & EXTENT_STATUS_WRITTEN)) {
858 ext4_warning(inode->i_sb, "Inserting extent [%u/%u] as "
859 " delayed and written which can potentially "
860 " cause data loss.", lblk, len);
864 newes.es_lblk = lblk;
866 ext4_es_store_pblock_status(&newes, pblk, status);
867 trace_ext4_es_insert_extent(inode, &newes);
869 ext4_es_insert_extent_check(inode, &newes);
873 es1 = __es_alloc_extent(true);
874 if ((err1 || err2) && !es2)
875 es2 = __es_alloc_extent(true);
876 write_lock(&EXT4_I(inode)->i_es_lock);
878 err1 = __es_remove_extent(inode, lblk, end, NULL, es1);
881 /* Free preallocated extent if it didn't get used. */
884 __es_free_extent(es1);
888 err2 = __es_insert_extent(inode, &newes, es2);
889 if (err2 == -ENOMEM && !ext4_es_must_keep(&newes))
893 /* Free preallocated extent if it didn't get used. */
896 __es_free_extent(es2);
900 if (sbi->s_cluster_ratio > 1 && test_opt(inode->i_sb, DELALLOC) &&
901 (status & EXTENT_STATUS_WRITTEN ||
902 status & EXTENT_STATUS_UNWRITTEN))
903 __revise_pending(inode, lblk, len);
905 write_unlock(&EXT4_I(inode)->i_es_lock);
909 ext4_es_print_tree(inode);
914 * ext4_es_cache_extent() inserts information into the extent status
915 * tree if and only if there isn't information about the range in
918 void ext4_es_cache_extent(struct inode *inode, ext4_lblk_t lblk,
919 ext4_lblk_t len, ext4_fsblk_t pblk,
922 struct extent_status *es;
923 struct extent_status newes;
924 ext4_lblk_t end = lblk + len - 1;
926 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
929 newes.es_lblk = lblk;
931 ext4_es_store_pblock_status(&newes, pblk, status);
932 trace_ext4_es_cache_extent(inode, &newes);
939 write_lock(&EXT4_I(inode)->i_es_lock);
941 es = __es_tree_search(&EXT4_I(inode)->i_es_tree.root, lblk);
942 if (!es || es->es_lblk > end)
943 __es_insert_extent(inode, &newes, NULL);
944 write_unlock(&EXT4_I(inode)->i_es_lock);
948 * ext4_es_lookup_extent() looks up an extent in extent status tree.
950 * ext4_es_lookup_extent is called by ext4_map_blocks/ext4_da_map_blocks.
952 * Return: 1 on found, 0 on not
954 int ext4_es_lookup_extent(struct inode *inode, ext4_lblk_t lblk,
955 ext4_lblk_t *next_lblk,
956 struct extent_status *es)
958 struct ext4_es_tree *tree;
959 struct ext4_es_stats *stats;
960 struct extent_status *es1 = NULL;
961 struct rb_node *node;
964 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
967 trace_ext4_es_lookup_extent_enter(inode, lblk);
968 es_debug("lookup extent in block %u\n", lblk);
970 tree = &EXT4_I(inode)->i_es_tree;
971 read_lock(&EXT4_I(inode)->i_es_lock);
973 /* find extent in cache firstly */
974 es->es_lblk = es->es_len = es->es_pblk = 0;
975 es1 = READ_ONCE(tree->cache_es);
976 if (es1 && in_range(lblk, es1->es_lblk, es1->es_len)) {
977 es_debug("%u cached by [%u/%u)\n",
978 lblk, es1->es_lblk, es1->es_len);
983 node = tree->root.rb_node;
985 es1 = rb_entry(node, struct extent_status, rb_node);
986 if (lblk < es1->es_lblk)
987 node = node->rb_left;
988 else if (lblk > ext4_es_end(es1))
989 node = node->rb_right;
997 stats = &EXT4_SB(inode->i_sb)->s_es_stats;
1000 es->es_lblk = es1->es_lblk;
1001 es->es_len = es1->es_len;
1002 es->es_pblk = es1->es_pblk;
1003 if (!ext4_es_is_referenced(es1))
1004 ext4_es_set_referenced(es1);
1005 percpu_counter_inc(&stats->es_stats_cache_hits);
1007 node = rb_next(&es1->rb_node);
1009 es1 = rb_entry(node, struct extent_status,
1011 *next_lblk = es1->es_lblk;
1016 percpu_counter_inc(&stats->es_stats_cache_misses);
1019 read_unlock(&EXT4_I(inode)->i_es_lock);
1021 trace_ext4_es_lookup_extent_exit(inode, es, found);
1027 bool first_do_lblk_found;
1028 ext4_lblk_t first_do_lblk;
1029 ext4_lblk_t last_do_lblk;
1030 struct extent_status *left_es;
1036 * init_rsvd - initialize reserved count data before removing block range
1037 * in file from extent status tree
1039 * @inode - file containing range
1040 * @lblk - first block in range
1041 * @es - pointer to first extent in range
1042 * @rc - pointer to reserved count data
1044 * Assumes es is not NULL
1046 static void init_rsvd(struct inode *inode, ext4_lblk_t lblk,
1047 struct extent_status *es, struct rsvd_count *rc)
1049 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1050 struct rb_node *node;
1055 * for bigalloc, note the first delonly block in the range has not
1056 * been found, record the extent containing the block to the left of
1057 * the region to be removed, if any, and note that there's no partial
1060 if (sbi->s_cluster_ratio > 1) {
1061 rc->first_do_lblk_found = false;
1062 if (lblk > es->es_lblk) {
1065 node = rb_prev(&es->rb_node);
1066 rc->left_es = node ? rb_entry(node,
1067 struct extent_status,
1070 rc->partial = false;
1075 * count_rsvd - count the clusters containing delayed and not unwritten
1076 * (delonly) blocks in a range within an extent and add to
1077 * the running tally in rsvd_count
1079 * @inode - file containing extent
1080 * @lblk - first block in range
1081 * @len - length of range in blocks
1082 * @es - pointer to extent containing clusters to be counted
1083 * @rc - pointer to reserved count data
1085 * Tracks partial clusters found at the beginning and end of extents so
1086 * they aren't overcounted when they span adjacent extents
1088 static void count_rsvd(struct inode *inode, ext4_lblk_t lblk, long len,
1089 struct extent_status *es, struct rsvd_count *rc)
1091 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1092 ext4_lblk_t i, end, nclu;
1094 if (!ext4_es_is_delonly(es))
1099 if (sbi->s_cluster_ratio == 1) {
1100 rc->ndelonly += (int) len;
1106 i = (lblk < es->es_lblk) ? es->es_lblk : lblk;
1107 end = lblk + (ext4_lblk_t) len - 1;
1108 end = (end > ext4_es_end(es)) ? ext4_es_end(es) : end;
1110 /* record the first block of the first delonly extent seen */
1111 if (!rc->first_do_lblk_found) {
1112 rc->first_do_lblk = i;
1113 rc->first_do_lblk_found = true;
1116 /* update the last lblk in the region seen so far */
1117 rc->last_do_lblk = end;
1120 * if we're tracking a partial cluster and the current extent
1121 * doesn't start with it, count it and stop tracking
1123 if (rc->partial && (rc->lclu != EXT4_B2C(sbi, i))) {
1125 rc->partial = false;
1129 * if the first cluster doesn't start on a cluster boundary but
1130 * ends on one, count it
1132 if (EXT4_LBLK_COFF(sbi, i) != 0) {
1133 if (end >= EXT4_LBLK_CFILL(sbi, i)) {
1135 rc->partial = false;
1136 i = EXT4_LBLK_CFILL(sbi, i) + 1;
1141 * if the current cluster starts on a cluster boundary, count the
1142 * number of whole delonly clusters in the extent
1144 if ((i + sbi->s_cluster_ratio - 1) <= end) {
1145 nclu = (end - i + 1) >> sbi->s_cluster_bits;
1146 rc->ndelonly += nclu;
1147 i += nclu << sbi->s_cluster_bits;
1151 * start tracking a partial cluster if there's a partial at the end
1152 * of the current extent and we're not already tracking one
1154 if (!rc->partial && i <= end) {
1156 rc->lclu = EXT4_B2C(sbi, i);
1161 * __pr_tree_search - search for a pending cluster reservation
1163 * @root - root of pending reservation tree
1164 * @lclu - logical cluster to search for
1166 * Returns the pending reservation for the cluster identified by @lclu
1167 * if found. If not, returns a reservation for the next cluster if any,
1168 * and if not, returns NULL.
1170 static struct pending_reservation *__pr_tree_search(struct rb_root *root,
1173 struct rb_node *node = root->rb_node;
1174 struct pending_reservation *pr = NULL;
1177 pr = rb_entry(node, struct pending_reservation, rb_node);
1178 if (lclu < pr->lclu)
1179 node = node->rb_left;
1180 else if (lclu > pr->lclu)
1181 node = node->rb_right;
1185 if (pr && lclu < pr->lclu)
1187 if (pr && lclu > pr->lclu) {
1188 node = rb_next(&pr->rb_node);
1189 return node ? rb_entry(node, struct pending_reservation,
1196 * get_rsvd - calculates and returns the number of cluster reservations to be
1197 * released when removing a block range from the extent status tree
1198 * and releases any pending reservations within the range
1200 * @inode - file containing block range
1201 * @end - last block in range
1202 * @right_es - pointer to extent containing next block beyond end or NULL
1203 * @rc - pointer to reserved count data
1205 * The number of reservations to be released is equal to the number of
1206 * clusters containing delayed and not unwritten (delonly) blocks within
1207 * the range, minus the number of clusters still containing delonly blocks
1208 * at the ends of the range, and minus the number of pending reservations
1211 static unsigned int get_rsvd(struct inode *inode, ext4_lblk_t end,
1212 struct extent_status *right_es,
1213 struct rsvd_count *rc)
1215 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1216 struct pending_reservation *pr;
1217 struct ext4_pending_tree *tree = &EXT4_I(inode)->i_pending_tree;
1218 struct rb_node *node;
1219 ext4_lblk_t first_lclu, last_lclu;
1220 bool left_delonly, right_delonly, count_pending;
1221 struct extent_status *es;
1223 if (sbi->s_cluster_ratio > 1) {
1224 /* count any remaining partial cluster */
1228 if (rc->ndelonly == 0)
1231 first_lclu = EXT4_B2C(sbi, rc->first_do_lblk);
1232 last_lclu = EXT4_B2C(sbi, rc->last_do_lblk);
1235 * decrease the delonly count by the number of clusters at the
1236 * ends of the range that still contain delonly blocks -
1237 * these clusters still need to be reserved
1239 left_delonly = right_delonly = false;
1242 while (es && ext4_es_end(es) >=
1243 EXT4_LBLK_CMASK(sbi, rc->first_do_lblk)) {
1244 if (ext4_es_is_delonly(es)) {
1246 left_delonly = true;
1249 node = rb_prev(&es->rb_node);
1252 es = rb_entry(node, struct extent_status, rb_node);
1254 if (right_es && (!left_delonly || first_lclu != last_lclu)) {
1255 if (end < ext4_es_end(right_es)) {
1258 node = rb_next(&right_es->rb_node);
1259 es = node ? rb_entry(node, struct extent_status,
1262 while (es && es->es_lblk <=
1263 EXT4_LBLK_CFILL(sbi, rc->last_do_lblk)) {
1264 if (ext4_es_is_delonly(es)) {
1266 right_delonly = true;
1269 node = rb_next(&es->rb_node);
1272 es = rb_entry(node, struct extent_status,
1278 * Determine the block range that should be searched for
1279 * pending reservations, if any. Clusters on the ends of the
1280 * original removed range containing delonly blocks are
1281 * excluded. They've already been accounted for and it's not
1282 * possible to determine if an associated pending reservation
1283 * should be released with the information available in the
1284 * extents status tree.
1286 if (first_lclu == last_lclu) {
1287 if (left_delonly | right_delonly)
1288 count_pending = false;
1290 count_pending = true;
1296 if (first_lclu <= last_lclu)
1297 count_pending = true;
1299 count_pending = false;
1303 * a pending reservation found between first_lclu and last_lclu
1304 * represents an allocated cluster that contained at least one
1305 * delonly block, so the delonly total must be reduced by one
1306 * for each pending reservation found and released
1308 if (count_pending) {
1309 pr = __pr_tree_search(&tree->root, first_lclu);
1310 while (pr && pr->lclu <= last_lclu) {
1312 node = rb_next(&pr->rb_node);
1313 rb_erase(&pr->rb_node, &tree->root);
1314 kmem_cache_free(ext4_pending_cachep, pr);
1317 pr = rb_entry(node, struct pending_reservation,
1322 return rc->ndelonly;
1327 * __es_remove_extent - removes block range from extent status tree
1329 * @inode - file containing range
1330 * @lblk - first block in range
1331 * @end - last block in range
1332 * @reserved - number of cluster reservations released
1333 * @prealloc - pre-allocated es to avoid memory allocation failures
1335 * If @reserved is not NULL and delayed allocation is enabled, counts
1336 * block/cluster reservations freed by removing range and if bigalloc
1337 * enabled cancels pending reservations as needed. Returns 0 on success,
1338 * error code on failure.
1340 static int __es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
1341 ext4_lblk_t end, int *reserved,
1342 struct extent_status *prealloc)
1344 struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
1345 struct rb_node *node;
1346 struct extent_status *es;
1347 struct extent_status orig_es;
1348 ext4_lblk_t len1, len2;
1351 bool count_reserved = true;
1352 struct rsvd_count rc;
1354 if (reserved == NULL || !test_opt(inode->i_sb, DELALLOC))
1355 count_reserved = false;
1357 es = __es_tree_search(&tree->root, lblk);
1360 if (es->es_lblk > end)
1363 /* Simply invalidate cache_es. */
1364 tree->cache_es = NULL;
1366 init_rsvd(inode, lblk, es, &rc);
1368 orig_es.es_lblk = es->es_lblk;
1369 orig_es.es_len = es->es_len;
1370 orig_es.es_pblk = es->es_pblk;
1372 len1 = lblk > es->es_lblk ? lblk - es->es_lblk : 0;
1373 len2 = ext4_es_end(es) > end ? ext4_es_end(es) - end : 0;
1378 struct extent_status newes;
1380 newes.es_lblk = end + 1;
1381 newes.es_len = len2;
1382 block = 0x7FDEADBEEFULL;
1383 if (ext4_es_is_written(&orig_es) ||
1384 ext4_es_is_unwritten(&orig_es))
1385 block = ext4_es_pblock(&orig_es) +
1386 orig_es.es_len - len2;
1387 ext4_es_store_pblock_status(&newes, block,
1388 ext4_es_status(&orig_es));
1389 err = __es_insert_extent(inode, &newes, prealloc);
1391 if (!ext4_es_must_keep(&newes))
1394 es->es_lblk = orig_es.es_lblk;
1395 es->es_len = orig_es.es_len;
1399 es->es_lblk = end + 1;
1401 if (ext4_es_is_written(es) ||
1402 ext4_es_is_unwritten(es)) {
1403 block = orig_es.es_pblk + orig_es.es_len - len2;
1404 ext4_es_store_pblock(es, block);
1408 count_rsvd(inode, lblk, orig_es.es_len - len1 - len2,
1410 goto out_get_reserved;
1415 count_rsvd(inode, lblk, orig_es.es_len - len1,
1417 node = rb_next(&es->rb_node);
1419 es = rb_entry(node, struct extent_status, rb_node);
1424 while (es && ext4_es_end(es) <= end) {
1426 count_rsvd(inode, es->es_lblk, es->es_len, es, &rc);
1427 node = rb_next(&es->rb_node);
1428 rb_erase(&es->rb_node, &tree->root);
1429 ext4_es_free_extent(inode, es);
1434 es = rb_entry(node, struct extent_status, rb_node);
1437 if (es && es->es_lblk < end + 1) {
1438 ext4_lblk_t orig_len = es->es_len;
1440 len1 = ext4_es_end(es) - end;
1442 count_rsvd(inode, es->es_lblk, orig_len - len1,
1444 es->es_lblk = end + 1;
1446 if (ext4_es_is_written(es) || ext4_es_is_unwritten(es)) {
1447 block = es->es_pblk + orig_len - len1;
1448 ext4_es_store_pblock(es, block);
1454 *reserved = get_rsvd(inode, end, es, &rc);
1460 * ext4_es_remove_extent - removes block range from extent status tree
1462 * @inode - file containing range
1463 * @lblk - first block in range
1464 * @len - number of blocks to remove
1466 * Reduces block/cluster reservation count and for bigalloc cancels pending
1467 * reservations as needed.
1469 void ext4_es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
1475 struct extent_status *es = NULL;
1477 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
1480 trace_ext4_es_remove_extent(inode, lblk, len);
1481 es_debug("remove [%u/%u) from extent status tree of inode %lu\n",
1482 lblk, len, inode->i_ino);
1487 end = lblk + len - 1;
1492 es = __es_alloc_extent(true);
1494 * ext4_clear_inode() depends on us taking i_es_lock unconditionally
1495 * so that we are sure __es_shrink() is done with the inode before it
1498 write_lock(&EXT4_I(inode)->i_es_lock);
1499 err = __es_remove_extent(inode, lblk, end, &reserved, es);
1500 /* Free preallocated extent if it didn't get used. */
1503 __es_free_extent(es);
1506 write_unlock(&EXT4_I(inode)->i_es_lock);
1510 ext4_es_print_tree(inode);
1511 ext4_da_release_space(inode, reserved);
1515 static int __es_shrink(struct ext4_sb_info *sbi, int nr_to_scan,
1516 struct ext4_inode_info *locked_ei)
1518 struct ext4_inode_info *ei;
1519 struct ext4_es_stats *es_stats;
1524 int retried = 0, nr_skipped = 0;
1526 es_stats = &sbi->s_es_stats;
1527 start_time = ktime_get();
1530 spin_lock(&sbi->s_es_lock);
1531 nr_to_walk = sbi->s_es_nr_inode;
1532 while (nr_to_walk-- > 0) {
1533 if (list_empty(&sbi->s_es_list)) {
1534 spin_unlock(&sbi->s_es_lock);
1537 ei = list_first_entry(&sbi->s_es_list, struct ext4_inode_info,
1539 /* Move the inode to the tail */
1540 list_move_tail(&ei->i_es_list, &sbi->s_es_list);
1543 * Normally we try hard to avoid shrinking precached inodes,
1544 * but we will as a last resort.
1546 if (!retried && ext4_test_inode_state(&ei->vfs_inode,
1547 EXT4_STATE_EXT_PRECACHED)) {
1552 if (ei == locked_ei || !write_trylock(&ei->i_es_lock)) {
1557 * Now we hold i_es_lock which protects us from inode reclaim
1558 * freeing inode under us
1560 spin_unlock(&sbi->s_es_lock);
1562 nr_shrunk += es_reclaim_extents(ei, &nr_to_scan);
1563 write_unlock(&ei->i_es_lock);
1565 if (nr_to_scan <= 0)
1567 spin_lock(&sbi->s_es_lock);
1569 spin_unlock(&sbi->s_es_lock);
1572 * If we skipped any inodes, and we weren't able to make any
1573 * forward progress, try again to scan precached inodes.
1575 if ((nr_shrunk == 0) && nr_skipped && !retried) {
1580 if (locked_ei && nr_shrunk == 0)
1581 nr_shrunk = es_reclaim_extents(locked_ei, &nr_to_scan);
1584 scan_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1585 if (likely(es_stats->es_stats_scan_time))
1586 es_stats->es_stats_scan_time = (scan_time +
1587 es_stats->es_stats_scan_time*3) / 4;
1589 es_stats->es_stats_scan_time = scan_time;
1590 if (scan_time > es_stats->es_stats_max_scan_time)
1591 es_stats->es_stats_max_scan_time = scan_time;
1592 if (likely(es_stats->es_stats_shrunk))
1593 es_stats->es_stats_shrunk = (nr_shrunk +
1594 es_stats->es_stats_shrunk*3) / 4;
1596 es_stats->es_stats_shrunk = nr_shrunk;
1598 trace_ext4_es_shrink(sbi->s_sb, nr_shrunk, scan_time,
1599 nr_skipped, retried);
1603 static unsigned long ext4_es_count(struct shrinker *shrink,
1604 struct shrink_control *sc)
1607 struct ext4_sb_info *sbi;
1609 sbi = container_of(shrink, struct ext4_sb_info, s_es_shrinker);
1610 nr = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt);
1611 trace_ext4_es_shrink_count(sbi->s_sb, sc->nr_to_scan, nr);
1615 static unsigned long ext4_es_scan(struct shrinker *shrink,
1616 struct shrink_control *sc)
1618 struct ext4_sb_info *sbi = container_of(shrink,
1619 struct ext4_sb_info, s_es_shrinker);
1620 int nr_to_scan = sc->nr_to_scan;
1623 ret = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt);
1624 trace_ext4_es_shrink_scan_enter(sbi->s_sb, nr_to_scan, ret);
1626 nr_shrunk = __es_shrink(sbi, nr_to_scan, NULL);
1628 ret = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt);
1629 trace_ext4_es_shrink_scan_exit(sbi->s_sb, nr_shrunk, ret);
1633 int ext4_seq_es_shrinker_info_show(struct seq_file *seq, void *v)
1635 struct ext4_sb_info *sbi = EXT4_SB((struct super_block *) seq->private);
1636 struct ext4_es_stats *es_stats = &sbi->s_es_stats;
1637 struct ext4_inode_info *ei, *max = NULL;
1638 unsigned int inode_cnt = 0;
1640 if (v != SEQ_START_TOKEN)
1643 /* here we just find an inode that has the max nr. of objects */
1644 spin_lock(&sbi->s_es_lock);
1645 list_for_each_entry(ei, &sbi->s_es_list, i_es_list) {
1647 if (max && max->i_es_all_nr < ei->i_es_all_nr)
1652 spin_unlock(&sbi->s_es_lock);
1654 seq_printf(seq, "stats:\n %lld objects\n %lld reclaimable objects\n",
1655 percpu_counter_sum_positive(&es_stats->es_stats_all_cnt),
1656 percpu_counter_sum_positive(&es_stats->es_stats_shk_cnt));
1657 seq_printf(seq, " %lld/%lld cache hits/misses\n",
1658 percpu_counter_sum_positive(&es_stats->es_stats_cache_hits),
1659 percpu_counter_sum_positive(&es_stats->es_stats_cache_misses));
1661 seq_printf(seq, " %d inodes on list\n", inode_cnt);
1663 seq_printf(seq, "average:\n %llu us scan time\n",
1664 div_u64(es_stats->es_stats_scan_time, 1000));
1665 seq_printf(seq, " %lu shrunk objects\n", es_stats->es_stats_shrunk);
1668 "maximum:\n %lu inode (%u objects, %u reclaimable)\n"
1669 " %llu us max scan time\n",
1670 max->vfs_inode.i_ino, max->i_es_all_nr, max->i_es_shk_nr,
1671 div_u64(es_stats->es_stats_max_scan_time, 1000));
1676 int ext4_es_register_shrinker(struct ext4_sb_info *sbi)
1680 /* Make sure we have enough bits for physical block number */
1681 BUILD_BUG_ON(ES_SHIFT < 48);
1682 INIT_LIST_HEAD(&sbi->s_es_list);
1683 sbi->s_es_nr_inode = 0;
1684 spin_lock_init(&sbi->s_es_lock);
1685 sbi->s_es_stats.es_stats_shrunk = 0;
1686 err = percpu_counter_init(&sbi->s_es_stats.es_stats_cache_hits, 0,
1690 err = percpu_counter_init(&sbi->s_es_stats.es_stats_cache_misses, 0,
1694 sbi->s_es_stats.es_stats_scan_time = 0;
1695 sbi->s_es_stats.es_stats_max_scan_time = 0;
1696 err = percpu_counter_init(&sbi->s_es_stats.es_stats_all_cnt, 0, GFP_KERNEL);
1699 err = percpu_counter_init(&sbi->s_es_stats.es_stats_shk_cnt, 0, GFP_KERNEL);
1703 sbi->s_es_shrinker.scan_objects = ext4_es_scan;
1704 sbi->s_es_shrinker.count_objects = ext4_es_count;
1705 sbi->s_es_shrinker.seeks = DEFAULT_SEEKS;
1706 err = register_shrinker(&sbi->s_es_shrinker, "ext4-es:%s",
1713 percpu_counter_destroy(&sbi->s_es_stats.es_stats_shk_cnt);
1715 percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt);
1717 percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_misses);
1719 percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_hits);
1723 void ext4_es_unregister_shrinker(struct ext4_sb_info *sbi)
1725 percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_hits);
1726 percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_misses);
1727 percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt);
1728 percpu_counter_destroy(&sbi->s_es_stats.es_stats_shk_cnt);
1729 unregister_shrinker(&sbi->s_es_shrinker);
1733 * Shrink extents in given inode from ei->i_es_shrink_lblk till end. Scan at
1734 * most *nr_to_scan extents, update *nr_to_scan accordingly.
1736 * Return 0 if we hit end of tree / interval, 1 if we exhausted nr_to_scan.
1737 * Increment *nr_shrunk by the number of reclaimed extents. Also update
1738 * ei->i_es_shrink_lblk to where we should continue scanning.
1740 static int es_do_reclaim_extents(struct ext4_inode_info *ei, ext4_lblk_t end,
1741 int *nr_to_scan, int *nr_shrunk)
1743 struct inode *inode = &ei->vfs_inode;
1744 struct ext4_es_tree *tree = &ei->i_es_tree;
1745 struct extent_status *es;
1746 struct rb_node *node;
1748 es = __es_tree_search(&tree->root, ei->i_es_shrink_lblk);
1752 while (*nr_to_scan > 0) {
1753 if (es->es_lblk > end) {
1754 ei->i_es_shrink_lblk = end + 1;
1759 node = rb_next(&es->rb_node);
1761 if (ext4_es_must_keep(es))
1763 if (ext4_es_is_referenced(es)) {
1764 ext4_es_clear_referenced(es);
1768 rb_erase(&es->rb_node, &tree->root);
1769 ext4_es_free_extent(inode, es);
1774 es = rb_entry(node, struct extent_status, rb_node);
1776 ei->i_es_shrink_lblk = es->es_lblk;
1779 ei->i_es_shrink_lblk = 0;
1783 static int es_reclaim_extents(struct ext4_inode_info *ei, int *nr_to_scan)
1785 struct inode *inode = &ei->vfs_inode;
1787 ext4_lblk_t start = ei->i_es_shrink_lblk;
1788 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
1789 DEFAULT_RATELIMIT_BURST);
1791 if (ei->i_es_shk_nr == 0)
1794 if (ext4_test_inode_state(inode, EXT4_STATE_EXT_PRECACHED) &&
1796 ext4_warning(inode->i_sb, "forced shrink of precached extents");
1798 if (!es_do_reclaim_extents(ei, EXT_MAX_BLOCKS, nr_to_scan, &nr_shrunk) &&
1800 es_do_reclaim_extents(ei, start - 1, nr_to_scan, &nr_shrunk);
1802 ei->i_es_tree.cache_es = NULL;
1807 * Called to support EXT4_IOC_CLEAR_ES_CACHE. We can only remove
1808 * discretionary entries from the extent status cache. (Some entries
1809 * must be present for proper operations.)
1811 void ext4_clear_inode_es(struct inode *inode)
1813 struct ext4_inode_info *ei = EXT4_I(inode);
1814 struct extent_status *es;
1815 struct ext4_es_tree *tree;
1816 struct rb_node *node;
1818 write_lock(&ei->i_es_lock);
1819 tree = &EXT4_I(inode)->i_es_tree;
1820 tree->cache_es = NULL;
1821 node = rb_first(&tree->root);
1823 es = rb_entry(node, struct extent_status, rb_node);
1824 node = rb_next(node);
1825 if (!ext4_es_must_keep(es)) {
1826 rb_erase(&es->rb_node, &tree->root);
1827 ext4_es_free_extent(inode, es);
1830 ext4_clear_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
1831 write_unlock(&ei->i_es_lock);
1835 static void ext4_print_pending_tree(struct inode *inode)
1837 struct ext4_pending_tree *tree;
1838 struct rb_node *node;
1839 struct pending_reservation *pr;
1841 printk(KERN_DEBUG "pending reservations for inode %lu:", inode->i_ino);
1842 tree = &EXT4_I(inode)->i_pending_tree;
1843 node = rb_first(&tree->root);
1845 pr = rb_entry(node, struct pending_reservation, rb_node);
1846 printk(KERN_DEBUG " %u", pr->lclu);
1847 node = rb_next(node);
1849 printk(KERN_DEBUG "\n");
1852 #define ext4_print_pending_tree(inode)
1855 int __init ext4_init_pending(void)
1857 ext4_pending_cachep = KMEM_CACHE(pending_reservation, SLAB_RECLAIM_ACCOUNT);
1858 if (ext4_pending_cachep == NULL)
1863 void ext4_exit_pending(void)
1865 kmem_cache_destroy(ext4_pending_cachep);
1868 void ext4_init_pending_tree(struct ext4_pending_tree *tree)
1870 tree->root = RB_ROOT;
1874 * __get_pending - retrieve a pointer to a pending reservation
1876 * @inode - file containing the pending cluster reservation
1877 * @lclu - logical cluster of interest
1879 * Returns a pointer to a pending reservation if it's a member of
1880 * the set, and NULL if not. Must be called holding i_es_lock.
1882 static struct pending_reservation *__get_pending(struct inode *inode,
1885 struct ext4_pending_tree *tree;
1886 struct rb_node *node;
1887 struct pending_reservation *pr = NULL;
1889 tree = &EXT4_I(inode)->i_pending_tree;
1890 node = (&tree->root)->rb_node;
1893 pr = rb_entry(node, struct pending_reservation, rb_node);
1894 if (lclu < pr->lclu)
1895 node = node->rb_left;
1896 else if (lclu > pr->lclu)
1897 node = node->rb_right;
1898 else if (lclu == pr->lclu)
1905 * __insert_pending - adds a pending cluster reservation to the set of
1906 * pending reservations
1908 * @inode - file containing the cluster
1909 * @lblk - logical block in the cluster to be added
1911 * Returns 0 on successful insertion and -ENOMEM on failure. If the
1912 * pending reservation is already in the set, returns successfully.
1914 static int __insert_pending(struct inode *inode, ext4_lblk_t lblk)
1916 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1917 struct ext4_pending_tree *tree = &EXT4_I(inode)->i_pending_tree;
1918 struct rb_node **p = &tree->root.rb_node;
1919 struct rb_node *parent = NULL;
1920 struct pending_reservation *pr;
1924 lclu = EXT4_B2C(sbi, lblk);
1925 /* search to find parent for insertion */
1928 pr = rb_entry(parent, struct pending_reservation, rb_node);
1930 if (lclu < pr->lclu) {
1932 } else if (lclu > pr->lclu) {
1933 p = &(*p)->rb_right;
1935 /* pending reservation already inserted */
1940 pr = kmem_cache_alloc(ext4_pending_cachep, GFP_ATOMIC);
1947 rb_link_node(&pr->rb_node, parent, p);
1948 rb_insert_color(&pr->rb_node, &tree->root);
1955 * __remove_pending - removes a pending cluster reservation from the set
1956 * of pending reservations
1958 * @inode - file containing the cluster
1959 * @lblk - logical block in the pending cluster reservation to be removed
1961 * Returns successfully if pending reservation is not a member of the set.
1963 static void __remove_pending(struct inode *inode, ext4_lblk_t lblk)
1965 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1966 struct pending_reservation *pr;
1967 struct ext4_pending_tree *tree;
1969 pr = __get_pending(inode, EXT4_B2C(sbi, lblk));
1971 tree = &EXT4_I(inode)->i_pending_tree;
1972 rb_erase(&pr->rb_node, &tree->root);
1973 kmem_cache_free(ext4_pending_cachep, pr);
1978 * ext4_remove_pending - removes a pending cluster reservation from the set
1979 * of pending reservations
1981 * @inode - file containing the cluster
1982 * @lblk - logical block in the pending cluster reservation to be removed
1984 * Locking for external use of __remove_pending.
1986 void ext4_remove_pending(struct inode *inode, ext4_lblk_t lblk)
1988 struct ext4_inode_info *ei = EXT4_I(inode);
1990 write_lock(&ei->i_es_lock);
1991 __remove_pending(inode, lblk);
1992 write_unlock(&ei->i_es_lock);
1996 * ext4_is_pending - determine whether a cluster has a pending reservation
1999 * @inode - file containing the cluster
2000 * @lblk - logical block in the cluster
2002 * Returns true if there's a pending reservation for the cluster in the
2003 * set of pending reservations, and false if not.
2005 bool ext4_is_pending(struct inode *inode, ext4_lblk_t lblk)
2007 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2008 struct ext4_inode_info *ei = EXT4_I(inode);
2011 read_lock(&ei->i_es_lock);
2012 ret = (bool)(__get_pending(inode, EXT4_B2C(sbi, lblk)) != NULL);
2013 read_unlock(&ei->i_es_lock);
2019 * ext4_es_insert_delayed_block - adds a delayed block to the extents status
2020 * tree, adding a pending reservation where
2023 * @inode - file containing the newly added block
2024 * @lblk - logical block to be added
2025 * @allocated - indicates whether a physical cluster has been allocated for
2026 * the logical cluster that contains the block
2028 void ext4_es_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk,
2031 struct extent_status newes;
2034 struct extent_status *es1 = NULL;
2035 struct extent_status *es2 = NULL;
2037 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
2040 es_debug("add [%u/1) delayed to extent status tree of inode %lu\n",
2041 lblk, inode->i_ino);
2043 newes.es_lblk = lblk;
2045 ext4_es_store_pblock_status(&newes, ~0, EXTENT_STATUS_DELAYED);
2046 trace_ext4_es_insert_delayed_block(inode, &newes, allocated);
2048 ext4_es_insert_extent_check(inode, &newes);
2052 es1 = __es_alloc_extent(true);
2053 if ((err1 || err2) && !es2)
2054 es2 = __es_alloc_extent(true);
2055 write_lock(&EXT4_I(inode)->i_es_lock);
2057 err1 = __es_remove_extent(inode, lblk, lblk, NULL, es1);
2060 /* Free preallocated extent if it didn't get used. */
2063 __es_free_extent(es1);
2067 err2 = __es_insert_extent(inode, &newes, es2);
2070 /* Free preallocated extent if it didn't get used. */
2073 __es_free_extent(es2);
2078 __insert_pending(inode, lblk);
2080 write_unlock(&EXT4_I(inode)->i_es_lock);
2084 ext4_es_print_tree(inode);
2085 ext4_print_pending_tree(inode);
2090 * __es_delayed_clu - count number of clusters containing blocks that
2093 * @inode - file containing block range
2094 * @start - logical block defining start of range
2095 * @end - logical block defining end of range
2097 * Returns the number of clusters containing only delayed (not delayed
2098 * and unwritten) blocks in the range specified by @start and @end. Any
2099 * cluster or part of a cluster within the range and containing a delayed
2100 * and not unwritten block within the range is counted as a whole cluster.
2102 static unsigned int __es_delayed_clu(struct inode *inode, ext4_lblk_t start,
2105 struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
2106 struct extent_status *es;
2107 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2108 struct rb_node *node;
2109 ext4_lblk_t first_lclu, last_lclu;
2110 unsigned long long last_counted_lclu;
2113 /* guaranteed to be unequal to any ext4_lblk_t value */
2114 last_counted_lclu = ~0ULL;
2116 es = __es_tree_search(&tree->root, start);
2118 while (es && (es->es_lblk <= end)) {
2119 if (ext4_es_is_delonly(es)) {
2120 if (es->es_lblk <= start)
2121 first_lclu = EXT4_B2C(sbi, start);
2123 first_lclu = EXT4_B2C(sbi, es->es_lblk);
2125 if (ext4_es_end(es) >= end)
2126 last_lclu = EXT4_B2C(sbi, end);
2128 last_lclu = EXT4_B2C(sbi, ext4_es_end(es));
2130 if (first_lclu == last_counted_lclu)
2131 n += last_lclu - first_lclu;
2133 n += last_lclu - first_lclu + 1;
2134 last_counted_lclu = last_lclu;
2136 node = rb_next(&es->rb_node);
2139 es = rb_entry(node, struct extent_status, rb_node);
2146 * ext4_es_delayed_clu - count number of clusters containing blocks that
2147 * are both delayed and unwritten
2149 * @inode - file containing block range
2150 * @lblk - logical block defining start of range
2151 * @len - number of blocks in range
2153 * Locking for external use of __es_delayed_clu().
2155 unsigned int ext4_es_delayed_clu(struct inode *inode, ext4_lblk_t lblk,
2158 struct ext4_inode_info *ei = EXT4_I(inode);
2165 end = lblk + len - 1;
2166 WARN_ON(end < lblk);
2168 read_lock(&ei->i_es_lock);
2170 n = __es_delayed_clu(inode, lblk, end);
2172 read_unlock(&ei->i_es_lock);
2178 * __revise_pending - makes, cancels, or leaves unchanged pending cluster
2179 * reservations for a specified block range depending
2180 * upon the presence or absence of delayed blocks
2181 * outside the range within clusters at the ends of the
2184 * @inode - file containing the range
2185 * @lblk - logical block defining the start of range
2186 * @len - length of range in blocks
2188 * Used after a newly allocated extent is added to the extents status tree.
2189 * Requires that the extents in the range have either written or unwritten
2190 * status. Must be called while holding i_es_lock.
2192 static void __revise_pending(struct inode *inode, ext4_lblk_t lblk,
2195 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2196 ext4_lblk_t end = lblk + len - 1;
2197 ext4_lblk_t first, last;
2198 bool f_del = false, l_del = false;
2204 * Two cases - block range within single cluster and block range
2205 * spanning two or more clusters. Note that a cluster belonging
2206 * to a range starting and/or ending on a cluster boundary is treated
2207 * as if it does not contain a delayed extent. The new range may
2208 * have allocated space for previously delayed blocks out to the
2209 * cluster boundary, requiring that any pre-existing pending
2210 * reservation be canceled. Because this code only looks at blocks
2211 * outside the range, it should revise pending reservations
2212 * correctly even if the extent represented by the range can't be
2213 * inserted in the extents status tree due to ENOSPC.
2216 if (EXT4_B2C(sbi, lblk) == EXT4_B2C(sbi, end)) {
2217 first = EXT4_LBLK_CMASK(sbi, lblk);
2219 f_del = __es_scan_range(inode, &ext4_es_is_delonly,
2222 __insert_pending(inode, first);
2224 last = EXT4_LBLK_CMASK(sbi, end) +
2225 sbi->s_cluster_ratio - 1;
2227 l_del = __es_scan_range(inode,
2228 &ext4_es_is_delonly,
2231 __insert_pending(inode, last);
2233 __remove_pending(inode, last);
2236 first = EXT4_LBLK_CMASK(sbi, lblk);
2238 f_del = __es_scan_range(inode, &ext4_es_is_delonly,
2241 __insert_pending(inode, first);
2243 __remove_pending(inode, first);
2245 last = EXT4_LBLK_CMASK(sbi, end) + sbi->s_cluster_ratio - 1;
2247 l_del = __es_scan_range(inode, &ext4_es_is_delonly,
2250 __insert_pending(inode, last);
2252 __remove_pending(inode, last);