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_create("ext4_extent_status",
159 sizeof(struct extent_status),
160 0, (SLAB_RECLAIM_ACCOUNT), NULL);
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 if (tree->cache_es) {
273 es1 = tree->cache_es;
274 if (in_range(lblk, es1->es_lblk, es1->es_len)) {
275 es_debug("%u cached by [%u/%u) %llu %x\n",
276 lblk, es1->es_lblk, es1->es_len,
277 ext4_es_pblock(es1), ext4_es_status(es1));
282 es1 = __es_tree_search(&tree->root, lblk);
285 if (es1 && !matching_fn(es1)) {
286 while ((node = rb_next(&es1->rb_node)) != NULL) {
287 es1 = rb_entry(node, struct extent_status, rb_node);
288 if (es1->es_lblk > end) {
292 if (matching_fn(es1))
297 if (es1 && matching_fn(es1)) {
298 tree->cache_es = es1;
299 es->es_lblk = es1->es_lblk;
300 es->es_len = es1->es_len;
301 es->es_pblk = es1->es_pblk;
307 * Locking for __es_find_extent_range() for external use
309 void ext4_es_find_extent_range(struct inode *inode,
310 int (*matching_fn)(struct extent_status *es),
311 ext4_lblk_t lblk, ext4_lblk_t end,
312 struct extent_status *es)
314 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
317 trace_ext4_es_find_extent_range_enter(inode, lblk);
319 read_lock(&EXT4_I(inode)->i_es_lock);
320 __es_find_extent_range(inode, matching_fn, lblk, end, es);
321 read_unlock(&EXT4_I(inode)->i_es_lock);
323 trace_ext4_es_find_extent_range_exit(inode, es);
327 * __es_scan_range - search block range for block with specified status
328 * in extents status tree
330 * @inode - file containing the range
331 * @matching_fn - pointer to function that matches extents with desired status
332 * @lblk - logical block defining start of range
333 * @end - logical block defining end of range
335 * Returns true if at least one block in the specified block range satisfies
336 * the criterion specified by @matching_fn, and false if not. If at least
337 * one extent has the specified status, then there is at least one block
338 * in the cluster with that status. Should only be called by code that has
341 static bool __es_scan_range(struct inode *inode,
342 int (*matching_fn)(struct extent_status *es),
343 ext4_lblk_t start, ext4_lblk_t end)
345 struct extent_status es;
347 __es_find_extent_range(inode, matching_fn, start, end, &es);
349 return false; /* no matching extent in the tree */
350 else if (es.es_lblk <= start &&
351 start < es.es_lblk + es.es_len)
353 else if (start <= es.es_lblk && es.es_lblk <= end)
359 * Locking for __es_scan_range() for external use
361 bool ext4_es_scan_range(struct inode *inode,
362 int (*matching_fn)(struct extent_status *es),
363 ext4_lblk_t lblk, ext4_lblk_t end)
367 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
370 read_lock(&EXT4_I(inode)->i_es_lock);
371 ret = __es_scan_range(inode, matching_fn, lblk, end);
372 read_unlock(&EXT4_I(inode)->i_es_lock);
378 * __es_scan_clu - search cluster for block with specified status in
379 * extents status tree
381 * @inode - file containing the cluster
382 * @matching_fn - pointer to function that matches extents with desired status
383 * @lblk - logical block in cluster to be searched
385 * Returns true if at least one extent in the cluster containing @lblk
386 * satisfies the criterion specified by @matching_fn, and false if not. If at
387 * least one extent has the specified status, then there is at least one block
388 * in the cluster with that status. Should only be called by code that has
391 static bool __es_scan_clu(struct inode *inode,
392 int (*matching_fn)(struct extent_status *es),
395 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
396 ext4_lblk_t lblk_start, lblk_end;
398 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
399 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
401 return __es_scan_range(inode, matching_fn, lblk_start, lblk_end);
405 * Locking for __es_scan_clu() for external use
407 bool ext4_es_scan_clu(struct inode *inode,
408 int (*matching_fn)(struct extent_status *es),
413 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
416 read_lock(&EXT4_I(inode)->i_es_lock);
417 ret = __es_scan_clu(inode, matching_fn, lblk);
418 read_unlock(&EXT4_I(inode)->i_es_lock);
423 static void ext4_es_list_add(struct inode *inode)
425 struct ext4_inode_info *ei = EXT4_I(inode);
426 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
428 if (!list_empty(&ei->i_es_list))
431 spin_lock(&sbi->s_es_lock);
432 if (list_empty(&ei->i_es_list)) {
433 list_add_tail(&ei->i_es_list, &sbi->s_es_list);
434 sbi->s_es_nr_inode++;
436 spin_unlock(&sbi->s_es_lock);
439 static void ext4_es_list_del(struct inode *inode)
441 struct ext4_inode_info *ei = EXT4_I(inode);
442 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
444 spin_lock(&sbi->s_es_lock);
445 if (!list_empty(&ei->i_es_list)) {
446 list_del_init(&ei->i_es_list);
447 sbi->s_es_nr_inode--;
448 WARN_ON_ONCE(sbi->s_es_nr_inode < 0);
450 spin_unlock(&sbi->s_es_lock);
453 static struct extent_status *
454 ext4_es_alloc_extent(struct inode *inode, ext4_lblk_t lblk, ext4_lblk_t len,
457 struct extent_status *es;
458 es = kmem_cache_alloc(ext4_es_cachep, GFP_ATOMIC);
466 * We don't count delayed extent because we never try to reclaim them
468 if (!ext4_es_is_delayed(es)) {
469 if (!EXT4_I(inode)->i_es_shk_nr++)
470 ext4_es_list_add(inode);
471 percpu_counter_inc(&EXT4_SB(inode->i_sb)->
472 s_es_stats.es_stats_shk_cnt);
475 EXT4_I(inode)->i_es_all_nr++;
476 percpu_counter_inc(&EXT4_SB(inode->i_sb)->s_es_stats.es_stats_all_cnt);
481 static void ext4_es_free_extent(struct inode *inode, struct extent_status *es)
483 EXT4_I(inode)->i_es_all_nr--;
484 percpu_counter_dec(&EXT4_SB(inode->i_sb)->s_es_stats.es_stats_all_cnt);
486 /* Decrease the shrink counter when this es is not delayed */
487 if (!ext4_es_is_delayed(es)) {
488 BUG_ON(EXT4_I(inode)->i_es_shk_nr == 0);
489 if (!--EXT4_I(inode)->i_es_shk_nr)
490 ext4_es_list_del(inode);
491 percpu_counter_dec(&EXT4_SB(inode->i_sb)->
492 s_es_stats.es_stats_shk_cnt);
495 kmem_cache_free(ext4_es_cachep, es);
499 * Check whether or not two extents can be merged
501 * - logical block number is contiguous
502 * - physical block number is contiguous
505 static int ext4_es_can_be_merged(struct extent_status *es1,
506 struct extent_status *es2)
508 if (ext4_es_type(es1) != ext4_es_type(es2))
511 if (((__u64) es1->es_len) + es2->es_len > EXT_MAX_BLOCKS) {
512 pr_warn("ES assertion failed when merging extents. "
513 "The sum of lengths of es1 (%d) and es2 (%d) "
514 "is bigger than allowed file size (%d)\n",
515 es1->es_len, es2->es_len, EXT_MAX_BLOCKS);
520 if (((__u64) es1->es_lblk) + es1->es_len != es2->es_lblk)
523 if ((ext4_es_is_written(es1) || ext4_es_is_unwritten(es1)) &&
524 (ext4_es_pblock(es1) + es1->es_len == ext4_es_pblock(es2)))
527 if (ext4_es_is_hole(es1))
530 /* we need to check delayed extent is without unwritten status */
531 if (ext4_es_is_delayed(es1) && !ext4_es_is_unwritten(es1))
537 static struct extent_status *
538 ext4_es_try_to_merge_left(struct inode *inode, struct extent_status *es)
540 struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
541 struct extent_status *es1;
542 struct rb_node *node;
544 node = rb_prev(&es->rb_node);
548 es1 = rb_entry(node, struct extent_status, rb_node);
549 if (ext4_es_can_be_merged(es1, es)) {
550 es1->es_len += es->es_len;
551 if (ext4_es_is_referenced(es))
552 ext4_es_set_referenced(es1);
553 rb_erase(&es->rb_node, &tree->root);
554 ext4_es_free_extent(inode, es);
561 static struct extent_status *
562 ext4_es_try_to_merge_right(struct inode *inode, struct extent_status *es)
564 struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
565 struct extent_status *es1;
566 struct rb_node *node;
568 node = rb_next(&es->rb_node);
572 es1 = rb_entry(node, struct extent_status, rb_node);
573 if (ext4_es_can_be_merged(es, es1)) {
574 es->es_len += es1->es_len;
575 if (ext4_es_is_referenced(es1))
576 ext4_es_set_referenced(es);
577 rb_erase(node, &tree->root);
578 ext4_es_free_extent(inode, es1);
584 #ifdef ES_AGGRESSIVE_TEST
585 #include "ext4_extents.h" /* Needed when ES_AGGRESSIVE_TEST is defined */
587 static void ext4_es_insert_extent_ext_check(struct inode *inode,
588 struct extent_status *es)
590 struct ext4_ext_path *path = NULL;
591 struct ext4_extent *ex;
592 ext4_lblk_t ee_block;
593 ext4_fsblk_t ee_start;
594 unsigned short ee_len;
595 int depth, ee_status, es_status;
597 path = ext4_find_extent(inode, es->es_lblk, NULL, EXT4_EX_NOCACHE);
601 depth = ext_depth(inode);
602 ex = path[depth].p_ext;
606 ee_block = le32_to_cpu(ex->ee_block);
607 ee_start = ext4_ext_pblock(ex);
608 ee_len = ext4_ext_get_actual_len(ex);
610 ee_status = ext4_ext_is_unwritten(ex) ? 1 : 0;
611 es_status = ext4_es_is_unwritten(es) ? 1 : 0;
614 * Make sure ex and es are not overlap when we try to insert
615 * a delayed/hole extent.
617 if (!ext4_es_is_written(es) && !ext4_es_is_unwritten(es)) {
618 if (in_range(es->es_lblk, ee_block, ee_len)) {
619 pr_warn("ES insert assertion failed for "
620 "inode: %lu we can find an extent "
621 "at block [%d/%d/%llu/%c], but we "
622 "want to add a delayed/hole extent "
624 inode->i_ino, ee_block, ee_len,
625 ee_start, ee_status ? 'u' : 'w',
626 es->es_lblk, es->es_len,
627 ext4_es_pblock(es), ext4_es_status(es));
633 * We don't check ee_block == es->es_lblk, etc. because es
634 * might be a part of whole extent, vice versa.
636 if (es->es_lblk < ee_block ||
637 ext4_es_pblock(es) != ee_start + es->es_lblk - ee_block) {
638 pr_warn("ES insert assertion failed for inode: %lu "
639 "ex_status [%d/%d/%llu/%c] != "
640 "es_status [%d/%d/%llu/%c]\n", inode->i_ino,
641 ee_block, ee_len, ee_start,
642 ee_status ? 'u' : 'w', es->es_lblk, es->es_len,
643 ext4_es_pblock(es), es_status ? 'u' : 'w');
647 if (ee_status ^ es_status) {
648 pr_warn("ES insert assertion failed for inode: %lu "
649 "ex_status [%d/%d/%llu/%c] != "
650 "es_status [%d/%d/%llu/%c]\n", inode->i_ino,
651 ee_block, ee_len, ee_start,
652 ee_status ? 'u' : 'w', es->es_lblk, es->es_len,
653 ext4_es_pblock(es), es_status ? 'u' : 'w');
657 * We can't find an extent on disk. So we need to make sure
658 * that we don't want to add an written/unwritten extent.
660 if (!ext4_es_is_delayed(es) && !ext4_es_is_hole(es)) {
661 pr_warn("ES insert assertion failed for inode: %lu "
662 "can't find an extent at block %d but we want "
663 "to add a written/unwritten extent "
664 "[%d/%d/%llu/%x]\n", inode->i_ino,
665 es->es_lblk, es->es_lblk, es->es_len,
666 ext4_es_pblock(es), ext4_es_status(es));
670 ext4_free_ext_path(path);
673 static void ext4_es_insert_extent_ind_check(struct inode *inode,
674 struct extent_status *es)
676 struct ext4_map_blocks map;
680 * Here we call ext4_ind_map_blocks to lookup a block mapping because
681 * 'Indirect' structure is defined in indirect.c. So we couldn't
682 * access direct/indirect tree from outside. It is too dirty to define
683 * this function in indirect.c file.
686 map.m_lblk = es->es_lblk;
687 map.m_len = es->es_len;
689 retval = ext4_ind_map_blocks(NULL, inode, &map, 0);
691 if (ext4_es_is_delayed(es) || ext4_es_is_hole(es)) {
693 * We want to add a delayed/hole extent but this
694 * block has been allocated.
696 pr_warn("ES insert assertion failed for inode: %lu "
697 "We can find blocks but we want to add a "
698 "delayed/hole extent [%d/%d/%llu/%x]\n",
699 inode->i_ino, es->es_lblk, es->es_len,
700 ext4_es_pblock(es), ext4_es_status(es));
702 } else if (ext4_es_is_written(es)) {
703 if (retval != es->es_len) {
704 pr_warn("ES insert assertion failed for "
705 "inode: %lu retval %d != es_len %d\n",
706 inode->i_ino, retval, es->es_len);
709 if (map.m_pblk != ext4_es_pblock(es)) {
710 pr_warn("ES insert assertion failed for "
711 "inode: %lu m_pblk %llu != "
713 inode->i_ino, map.m_pblk,
719 * We don't need to check unwritten extent because
720 * indirect-based file doesn't have it.
724 } else if (retval == 0) {
725 if (ext4_es_is_written(es)) {
726 pr_warn("ES insert assertion failed for inode: %lu "
727 "We can't find the block but we want to add "
728 "a written extent [%d/%d/%llu/%x]\n",
729 inode->i_ino, es->es_lblk, es->es_len,
730 ext4_es_pblock(es), ext4_es_status(es));
736 static inline void ext4_es_insert_extent_check(struct inode *inode,
737 struct extent_status *es)
740 * We don't need to worry about the race condition because
741 * caller takes i_data_sem locking.
743 BUG_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
744 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
745 ext4_es_insert_extent_ext_check(inode, es);
747 ext4_es_insert_extent_ind_check(inode, es);
750 static inline void ext4_es_insert_extent_check(struct inode *inode,
751 struct extent_status *es)
756 static int __es_insert_extent(struct inode *inode, struct extent_status *newes)
758 struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
759 struct rb_node **p = &tree->root.rb_node;
760 struct rb_node *parent = NULL;
761 struct extent_status *es;
765 es = rb_entry(parent, struct extent_status, rb_node);
767 if (newes->es_lblk < es->es_lblk) {
768 if (ext4_es_can_be_merged(newes, es)) {
770 * Here we can modify es_lblk directly
771 * because it isn't overlapped.
773 es->es_lblk = newes->es_lblk;
774 es->es_len += newes->es_len;
775 if (ext4_es_is_written(es) ||
776 ext4_es_is_unwritten(es))
777 ext4_es_store_pblock(es,
779 es = ext4_es_try_to_merge_left(inode, es);
783 } else if (newes->es_lblk > ext4_es_end(es)) {
784 if (ext4_es_can_be_merged(es, newes)) {
785 es->es_len += newes->es_len;
786 es = ext4_es_try_to_merge_right(inode, es);
796 es = ext4_es_alloc_extent(inode, newes->es_lblk, newes->es_len,
800 rb_link_node(&es->rb_node, parent, p);
801 rb_insert_color(&es->rb_node, &tree->root);
809 * ext4_es_insert_extent() adds information to an inode's extent
812 * Return 0 on success, error code on failure.
814 int ext4_es_insert_extent(struct inode *inode, ext4_lblk_t lblk,
815 ext4_lblk_t len, ext4_fsblk_t pblk,
818 struct extent_status newes;
819 ext4_lblk_t end = lblk + len - 1;
821 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
823 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
826 es_debug("add [%u/%u) %llu %x to extent status tree of inode %lu\n",
827 lblk, len, pblk, status, inode->i_ino);
834 if ((status & EXTENT_STATUS_DELAYED) &&
835 (status & EXTENT_STATUS_WRITTEN)) {
836 ext4_warning(inode->i_sb, "Inserting extent [%u/%u] as "
837 " delayed and written which can potentially "
838 " cause data loss.", lblk, len);
842 newes.es_lblk = lblk;
844 ext4_es_store_pblock_status(&newes, pblk, status);
845 trace_ext4_es_insert_extent(inode, &newes);
847 ext4_es_insert_extent_check(inode, &newes);
849 write_lock(&EXT4_I(inode)->i_es_lock);
850 err = __es_remove_extent(inode, lblk, end, NULL);
854 err = __es_insert_extent(inode, &newes);
855 if (err == -ENOMEM && __es_shrink(EXT4_SB(inode->i_sb),
858 if (err == -ENOMEM && !ext4_es_is_delayed(&newes))
861 if (sbi->s_cluster_ratio > 1 && test_opt(inode->i_sb, DELALLOC) &&
862 (status & EXTENT_STATUS_WRITTEN ||
863 status & EXTENT_STATUS_UNWRITTEN))
864 __revise_pending(inode, lblk, len);
867 write_unlock(&EXT4_I(inode)->i_es_lock);
869 ext4_es_print_tree(inode);
875 * ext4_es_cache_extent() inserts information into the extent status
876 * tree if and only if there isn't information about the range in
879 void ext4_es_cache_extent(struct inode *inode, ext4_lblk_t lblk,
880 ext4_lblk_t len, ext4_fsblk_t pblk,
883 struct extent_status *es;
884 struct extent_status newes;
885 ext4_lblk_t end = lblk + len - 1;
887 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
890 newes.es_lblk = lblk;
892 ext4_es_store_pblock_status(&newes, pblk, status);
893 trace_ext4_es_cache_extent(inode, &newes);
900 write_lock(&EXT4_I(inode)->i_es_lock);
902 es = __es_tree_search(&EXT4_I(inode)->i_es_tree.root, lblk);
903 if (!es || es->es_lblk > end)
904 __es_insert_extent(inode, &newes);
905 write_unlock(&EXT4_I(inode)->i_es_lock);
909 * ext4_es_lookup_extent() looks up an extent in extent status tree.
911 * ext4_es_lookup_extent is called by ext4_map_blocks/ext4_da_map_blocks.
913 * Return: 1 on found, 0 on not
915 int ext4_es_lookup_extent(struct inode *inode, ext4_lblk_t lblk,
916 ext4_lblk_t *next_lblk,
917 struct extent_status *es)
919 struct ext4_es_tree *tree;
920 struct ext4_es_stats *stats;
921 struct extent_status *es1 = NULL;
922 struct rb_node *node;
925 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
928 trace_ext4_es_lookup_extent_enter(inode, lblk);
929 es_debug("lookup extent in block %u\n", lblk);
931 tree = &EXT4_I(inode)->i_es_tree;
932 read_lock(&EXT4_I(inode)->i_es_lock);
934 /* find extent in cache firstly */
935 es->es_lblk = es->es_len = es->es_pblk = 0;
936 if (tree->cache_es) {
937 es1 = tree->cache_es;
938 if (in_range(lblk, es1->es_lblk, es1->es_len)) {
939 es_debug("%u cached by [%u/%u)\n",
940 lblk, es1->es_lblk, es1->es_len);
946 node = tree->root.rb_node;
948 es1 = rb_entry(node, struct extent_status, rb_node);
949 if (lblk < es1->es_lblk)
950 node = node->rb_left;
951 else if (lblk > ext4_es_end(es1))
952 node = node->rb_right;
960 stats = &EXT4_SB(inode->i_sb)->s_es_stats;
963 es->es_lblk = es1->es_lblk;
964 es->es_len = es1->es_len;
965 es->es_pblk = es1->es_pblk;
966 if (!ext4_es_is_referenced(es1))
967 ext4_es_set_referenced(es1);
968 percpu_counter_inc(&stats->es_stats_cache_hits);
970 node = rb_next(&es1->rb_node);
972 es1 = rb_entry(node, struct extent_status,
974 *next_lblk = es1->es_lblk;
979 percpu_counter_inc(&stats->es_stats_cache_misses);
982 read_unlock(&EXT4_I(inode)->i_es_lock);
984 trace_ext4_es_lookup_extent_exit(inode, es, found);
990 bool first_do_lblk_found;
991 ext4_lblk_t first_do_lblk;
992 ext4_lblk_t last_do_lblk;
993 struct extent_status *left_es;
999 * init_rsvd - initialize reserved count data before removing block range
1000 * in file from extent status tree
1002 * @inode - file containing range
1003 * @lblk - first block in range
1004 * @es - pointer to first extent in range
1005 * @rc - pointer to reserved count data
1007 * Assumes es is not NULL
1009 static void init_rsvd(struct inode *inode, ext4_lblk_t lblk,
1010 struct extent_status *es, struct rsvd_count *rc)
1012 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1013 struct rb_node *node;
1018 * for bigalloc, note the first delonly block in the range has not
1019 * been found, record the extent containing the block to the left of
1020 * the region to be removed, if any, and note that there's no partial
1023 if (sbi->s_cluster_ratio > 1) {
1024 rc->first_do_lblk_found = false;
1025 if (lblk > es->es_lblk) {
1028 node = rb_prev(&es->rb_node);
1029 rc->left_es = node ? rb_entry(node,
1030 struct extent_status,
1033 rc->partial = false;
1038 * count_rsvd - count the clusters containing delayed and not unwritten
1039 * (delonly) blocks in a range within an extent and add to
1040 * the running tally in rsvd_count
1042 * @inode - file containing extent
1043 * @lblk - first block in range
1044 * @len - length of range in blocks
1045 * @es - pointer to extent containing clusters to be counted
1046 * @rc - pointer to reserved count data
1048 * Tracks partial clusters found at the beginning and end of extents so
1049 * they aren't overcounted when they span adjacent extents
1051 static void count_rsvd(struct inode *inode, ext4_lblk_t lblk, long len,
1052 struct extent_status *es, struct rsvd_count *rc)
1054 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1055 ext4_lblk_t i, end, nclu;
1057 if (!ext4_es_is_delonly(es))
1062 if (sbi->s_cluster_ratio == 1) {
1063 rc->ndelonly += (int) len;
1069 i = (lblk < es->es_lblk) ? es->es_lblk : lblk;
1070 end = lblk + (ext4_lblk_t) len - 1;
1071 end = (end > ext4_es_end(es)) ? ext4_es_end(es) : end;
1073 /* record the first block of the first delonly extent seen */
1074 if (!rc->first_do_lblk_found) {
1075 rc->first_do_lblk = i;
1076 rc->first_do_lblk_found = true;
1079 /* update the last lblk in the region seen so far */
1080 rc->last_do_lblk = end;
1083 * if we're tracking a partial cluster and the current extent
1084 * doesn't start with it, count it and stop tracking
1086 if (rc->partial && (rc->lclu != EXT4_B2C(sbi, i))) {
1088 rc->partial = false;
1092 * if the first cluster doesn't start on a cluster boundary but
1093 * ends on one, count it
1095 if (EXT4_LBLK_COFF(sbi, i) != 0) {
1096 if (end >= EXT4_LBLK_CFILL(sbi, i)) {
1098 rc->partial = false;
1099 i = EXT4_LBLK_CFILL(sbi, i) + 1;
1104 * if the current cluster starts on a cluster boundary, count the
1105 * number of whole delonly clusters in the extent
1107 if ((i + sbi->s_cluster_ratio - 1) <= end) {
1108 nclu = (end - i + 1) >> sbi->s_cluster_bits;
1109 rc->ndelonly += nclu;
1110 i += nclu << sbi->s_cluster_bits;
1114 * start tracking a partial cluster if there's a partial at the end
1115 * of the current extent and we're not already tracking one
1117 if (!rc->partial && i <= end) {
1119 rc->lclu = EXT4_B2C(sbi, i);
1124 * __pr_tree_search - search for a pending cluster reservation
1126 * @root - root of pending reservation tree
1127 * @lclu - logical cluster to search for
1129 * Returns the pending reservation for the cluster identified by @lclu
1130 * if found. If not, returns a reservation for the next cluster if any,
1131 * and if not, returns NULL.
1133 static struct pending_reservation *__pr_tree_search(struct rb_root *root,
1136 struct rb_node *node = root->rb_node;
1137 struct pending_reservation *pr = NULL;
1140 pr = rb_entry(node, struct pending_reservation, rb_node);
1141 if (lclu < pr->lclu)
1142 node = node->rb_left;
1143 else if (lclu > pr->lclu)
1144 node = node->rb_right;
1148 if (pr && lclu < pr->lclu)
1150 if (pr && lclu > pr->lclu) {
1151 node = rb_next(&pr->rb_node);
1152 return node ? rb_entry(node, struct pending_reservation,
1159 * get_rsvd - calculates and returns the number of cluster reservations to be
1160 * released when removing a block range from the extent status tree
1161 * and releases any pending reservations within the range
1163 * @inode - file containing block range
1164 * @end - last block in range
1165 * @right_es - pointer to extent containing next block beyond end or NULL
1166 * @rc - pointer to reserved count data
1168 * The number of reservations to be released is equal to the number of
1169 * clusters containing delayed and not unwritten (delonly) blocks within
1170 * the range, minus the number of clusters still containing delonly blocks
1171 * at the ends of the range, and minus the number of pending reservations
1174 static unsigned int get_rsvd(struct inode *inode, ext4_lblk_t end,
1175 struct extent_status *right_es,
1176 struct rsvd_count *rc)
1178 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1179 struct pending_reservation *pr;
1180 struct ext4_pending_tree *tree = &EXT4_I(inode)->i_pending_tree;
1181 struct rb_node *node;
1182 ext4_lblk_t first_lclu, last_lclu;
1183 bool left_delonly, right_delonly, count_pending;
1184 struct extent_status *es;
1186 if (sbi->s_cluster_ratio > 1) {
1187 /* count any remaining partial cluster */
1191 if (rc->ndelonly == 0)
1194 first_lclu = EXT4_B2C(sbi, rc->first_do_lblk);
1195 last_lclu = EXT4_B2C(sbi, rc->last_do_lblk);
1198 * decrease the delonly count by the number of clusters at the
1199 * ends of the range that still contain delonly blocks -
1200 * these clusters still need to be reserved
1202 left_delonly = right_delonly = false;
1205 while (es && ext4_es_end(es) >=
1206 EXT4_LBLK_CMASK(sbi, rc->first_do_lblk)) {
1207 if (ext4_es_is_delonly(es)) {
1209 left_delonly = true;
1212 node = rb_prev(&es->rb_node);
1215 es = rb_entry(node, struct extent_status, rb_node);
1217 if (right_es && (!left_delonly || first_lclu != last_lclu)) {
1218 if (end < ext4_es_end(right_es)) {
1221 node = rb_next(&right_es->rb_node);
1222 es = node ? rb_entry(node, struct extent_status,
1225 while (es && es->es_lblk <=
1226 EXT4_LBLK_CFILL(sbi, rc->last_do_lblk)) {
1227 if (ext4_es_is_delonly(es)) {
1229 right_delonly = true;
1232 node = rb_next(&es->rb_node);
1235 es = rb_entry(node, struct extent_status,
1241 * Determine the block range that should be searched for
1242 * pending reservations, if any. Clusters on the ends of the
1243 * original removed range containing delonly blocks are
1244 * excluded. They've already been accounted for and it's not
1245 * possible to determine if an associated pending reservation
1246 * should be released with the information available in the
1247 * extents status tree.
1249 if (first_lclu == last_lclu) {
1250 if (left_delonly | right_delonly)
1251 count_pending = false;
1253 count_pending = true;
1259 if (first_lclu <= last_lclu)
1260 count_pending = true;
1262 count_pending = false;
1266 * a pending reservation found between first_lclu and last_lclu
1267 * represents an allocated cluster that contained at least one
1268 * delonly block, so the delonly total must be reduced by one
1269 * for each pending reservation found and released
1271 if (count_pending) {
1272 pr = __pr_tree_search(&tree->root, first_lclu);
1273 while (pr && pr->lclu <= last_lclu) {
1275 node = rb_next(&pr->rb_node);
1276 rb_erase(&pr->rb_node, &tree->root);
1277 kmem_cache_free(ext4_pending_cachep, pr);
1280 pr = rb_entry(node, struct pending_reservation,
1285 return rc->ndelonly;
1290 * __es_remove_extent - removes block range from extent status tree
1292 * @inode - file containing range
1293 * @lblk - first block in range
1294 * @end - last block in range
1295 * @reserved - number of cluster reservations released
1297 * If @reserved is not NULL and delayed allocation is enabled, counts
1298 * block/cluster reservations freed by removing range and if bigalloc
1299 * enabled cancels pending reservations as needed. Returns 0 on success,
1300 * error code on failure.
1302 static int __es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
1303 ext4_lblk_t end, int *reserved)
1305 struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
1306 struct rb_node *node;
1307 struct extent_status *es;
1308 struct extent_status orig_es;
1309 ext4_lblk_t len1, len2;
1312 bool count_reserved = true;
1313 struct rsvd_count rc;
1315 if (reserved == NULL || !test_opt(inode->i_sb, DELALLOC))
1316 count_reserved = false;
1320 es = __es_tree_search(&tree->root, lblk);
1323 if (es->es_lblk > end)
1326 /* Simply invalidate cache_es. */
1327 tree->cache_es = NULL;
1329 init_rsvd(inode, lblk, es, &rc);
1331 orig_es.es_lblk = es->es_lblk;
1332 orig_es.es_len = es->es_len;
1333 orig_es.es_pblk = es->es_pblk;
1335 len1 = lblk > es->es_lblk ? lblk - es->es_lblk : 0;
1336 len2 = ext4_es_end(es) > end ? ext4_es_end(es) - end : 0;
1341 struct extent_status newes;
1343 newes.es_lblk = end + 1;
1344 newes.es_len = len2;
1345 block = 0x7FDEADBEEFULL;
1346 if (ext4_es_is_written(&orig_es) ||
1347 ext4_es_is_unwritten(&orig_es))
1348 block = ext4_es_pblock(&orig_es) +
1349 orig_es.es_len - len2;
1350 ext4_es_store_pblock_status(&newes, block,
1351 ext4_es_status(&orig_es));
1352 err = __es_insert_extent(inode, &newes);
1354 es->es_lblk = orig_es.es_lblk;
1355 es->es_len = orig_es.es_len;
1356 if ((err == -ENOMEM) &&
1357 __es_shrink(EXT4_SB(inode->i_sb),
1358 128, EXT4_I(inode)))
1363 es->es_lblk = end + 1;
1365 if (ext4_es_is_written(es) ||
1366 ext4_es_is_unwritten(es)) {
1367 block = orig_es.es_pblk + orig_es.es_len - len2;
1368 ext4_es_store_pblock(es, block);
1372 count_rsvd(inode, lblk, orig_es.es_len - len1 - len2,
1374 goto out_get_reserved;
1379 count_rsvd(inode, lblk, orig_es.es_len - len1,
1381 node = rb_next(&es->rb_node);
1383 es = rb_entry(node, struct extent_status, rb_node);
1388 while (es && ext4_es_end(es) <= end) {
1390 count_rsvd(inode, es->es_lblk, es->es_len, es, &rc);
1391 node = rb_next(&es->rb_node);
1392 rb_erase(&es->rb_node, &tree->root);
1393 ext4_es_free_extent(inode, es);
1398 es = rb_entry(node, struct extent_status, rb_node);
1401 if (es && es->es_lblk < end + 1) {
1402 ext4_lblk_t orig_len = es->es_len;
1404 len1 = ext4_es_end(es) - end;
1406 count_rsvd(inode, es->es_lblk, orig_len - len1,
1408 es->es_lblk = end + 1;
1410 if (ext4_es_is_written(es) || ext4_es_is_unwritten(es)) {
1411 block = es->es_pblk + orig_len - len1;
1412 ext4_es_store_pblock(es, block);
1418 *reserved = get_rsvd(inode, end, es, &rc);
1424 * ext4_es_remove_extent - removes block range from extent status tree
1426 * @inode - file containing range
1427 * @lblk - first block in range
1428 * @len - number of blocks to remove
1430 * Reduces block/cluster reservation count and for bigalloc cancels pending
1431 * reservations as needed. Returns 0 on success, error code on failure.
1433 int ext4_es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
1440 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
1443 trace_ext4_es_remove_extent(inode, lblk, len);
1444 es_debug("remove [%u/%u) from extent status tree of inode %lu\n",
1445 lblk, len, inode->i_ino);
1450 end = lblk + len - 1;
1454 * ext4_clear_inode() depends on us taking i_es_lock unconditionally
1455 * so that we are sure __es_shrink() is done with the inode before it
1458 write_lock(&EXT4_I(inode)->i_es_lock);
1459 err = __es_remove_extent(inode, lblk, end, &reserved);
1460 write_unlock(&EXT4_I(inode)->i_es_lock);
1461 ext4_es_print_tree(inode);
1462 ext4_da_release_space(inode, reserved);
1466 static int __es_shrink(struct ext4_sb_info *sbi, int nr_to_scan,
1467 struct ext4_inode_info *locked_ei)
1469 struct ext4_inode_info *ei;
1470 struct ext4_es_stats *es_stats;
1475 int retried = 0, nr_skipped = 0;
1477 es_stats = &sbi->s_es_stats;
1478 start_time = ktime_get();
1481 spin_lock(&sbi->s_es_lock);
1482 nr_to_walk = sbi->s_es_nr_inode;
1483 while (nr_to_walk-- > 0) {
1484 if (list_empty(&sbi->s_es_list)) {
1485 spin_unlock(&sbi->s_es_lock);
1488 ei = list_first_entry(&sbi->s_es_list, struct ext4_inode_info,
1490 /* Move the inode to the tail */
1491 list_move_tail(&ei->i_es_list, &sbi->s_es_list);
1494 * Normally we try hard to avoid shrinking precached inodes,
1495 * but we will as a last resort.
1497 if (!retried && ext4_test_inode_state(&ei->vfs_inode,
1498 EXT4_STATE_EXT_PRECACHED)) {
1503 if (ei == locked_ei || !write_trylock(&ei->i_es_lock)) {
1508 * Now we hold i_es_lock which protects us from inode reclaim
1509 * freeing inode under us
1511 spin_unlock(&sbi->s_es_lock);
1513 nr_shrunk += es_reclaim_extents(ei, &nr_to_scan);
1514 write_unlock(&ei->i_es_lock);
1516 if (nr_to_scan <= 0)
1518 spin_lock(&sbi->s_es_lock);
1520 spin_unlock(&sbi->s_es_lock);
1523 * If we skipped any inodes, and we weren't able to make any
1524 * forward progress, try again to scan precached inodes.
1526 if ((nr_shrunk == 0) && nr_skipped && !retried) {
1531 if (locked_ei && nr_shrunk == 0)
1532 nr_shrunk = es_reclaim_extents(locked_ei, &nr_to_scan);
1535 scan_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1536 if (likely(es_stats->es_stats_scan_time))
1537 es_stats->es_stats_scan_time = (scan_time +
1538 es_stats->es_stats_scan_time*3) / 4;
1540 es_stats->es_stats_scan_time = scan_time;
1541 if (scan_time > es_stats->es_stats_max_scan_time)
1542 es_stats->es_stats_max_scan_time = scan_time;
1543 if (likely(es_stats->es_stats_shrunk))
1544 es_stats->es_stats_shrunk = (nr_shrunk +
1545 es_stats->es_stats_shrunk*3) / 4;
1547 es_stats->es_stats_shrunk = nr_shrunk;
1549 trace_ext4_es_shrink(sbi->s_sb, nr_shrunk, scan_time,
1550 nr_skipped, retried);
1554 static unsigned long ext4_es_count(struct shrinker *shrink,
1555 struct shrink_control *sc)
1558 struct ext4_sb_info *sbi;
1560 sbi = container_of(shrink, struct ext4_sb_info, s_es_shrinker);
1561 nr = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt);
1562 trace_ext4_es_shrink_count(sbi->s_sb, sc->nr_to_scan, nr);
1566 static unsigned long ext4_es_scan(struct shrinker *shrink,
1567 struct shrink_control *sc)
1569 struct ext4_sb_info *sbi = container_of(shrink,
1570 struct ext4_sb_info, s_es_shrinker);
1571 int nr_to_scan = sc->nr_to_scan;
1574 ret = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt);
1575 trace_ext4_es_shrink_scan_enter(sbi->s_sb, nr_to_scan, ret);
1577 nr_shrunk = __es_shrink(sbi, nr_to_scan, NULL);
1579 ret = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt);
1580 trace_ext4_es_shrink_scan_exit(sbi->s_sb, nr_shrunk, ret);
1584 int ext4_seq_es_shrinker_info_show(struct seq_file *seq, void *v)
1586 struct ext4_sb_info *sbi = EXT4_SB((struct super_block *) seq->private);
1587 struct ext4_es_stats *es_stats = &sbi->s_es_stats;
1588 struct ext4_inode_info *ei, *max = NULL;
1589 unsigned int inode_cnt = 0;
1591 if (v != SEQ_START_TOKEN)
1594 /* here we just find an inode that has the max nr. of objects */
1595 spin_lock(&sbi->s_es_lock);
1596 list_for_each_entry(ei, &sbi->s_es_list, i_es_list) {
1598 if (max && max->i_es_all_nr < ei->i_es_all_nr)
1603 spin_unlock(&sbi->s_es_lock);
1605 seq_printf(seq, "stats:\n %lld objects\n %lld reclaimable objects\n",
1606 percpu_counter_sum_positive(&es_stats->es_stats_all_cnt),
1607 percpu_counter_sum_positive(&es_stats->es_stats_shk_cnt));
1608 seq_printf(seq, " %lld/%lld cache hits/misses\n",
1609 percpu_counter_sum_positive(&es_stats->es_stats_cache_hits),
1610 percpu_counter_sum_positive(&es_stats->es_stats_cache_misses));
1612 seq_printf(seq, " %d inodes on list\n", inode_cnt);
1614 seq_printf(seq, "average:\n %llu us scan time\n",
1615 div_u64(es_stats->es_stats_scan_time, 1000));
1616 seq_printf(seq, " %lu shrunk objects\n", es_stats->es_stats_shrunk);
1619 "maximum:\n %lu inode (%u objects, %u reclaimable)\n"
1620 " %llu us max scan time\n",
1621 max->vfs_inode.i_ino, max->i_es_all_nr, max->i_es_shk_nr,
1622 div_u64(es_stats->es_stats_max_scan_time, 1000));
1627 int ext4_es_register_shrinker(struct ext4_sb_info *sbi)
1631 /* Make sure we have enough bits for physical block number */
1632 BUILD_BUG_ON(ES_SHIFT < 48);
1633 INIT_LIST_HEAD(&sbi->s_es_list);
1634 sbi->s_es_nr_inode = 0;
1635 spin_lock_init(&sbi->s_es_lock);
1636 sbi->s_es_stats.es_stats_shrunk = 0;
1637 err = percpu_counter_init(&sbi->s_es_stats.es_stats_cache_hits, 0,
1641 err = percpu_counter_init(&sbi->s_es_stats.es_stats_cache_misses, 0,
1645 sbi->s_es_stats.es_stats_scan_time = 0;
1646 sbi->s_es_stats.es_stats_max_scan_time = 0;
1647 err = percpu_counter_init(&sbi->s_es_stats.es_stats_all_cnt, 0, GFP_KERNEL);
1650 err = percpu_counter_init(&sbi->s_es_stats.es_stats_shk_cnt, 0, GFP_KERNEL);
1654 sbi->s_es_shrinker.scan_objects = ext4_es_scan;
1655 sbi->s_es_shrinker.count_objects = ext4_es_count;
1656 sbi->s_es_shrinker.seeks = DEFAULT_SEEKS;
1657 err = register_shrinker(&sbi->s_es_shrinker, "ext4-es:%s",
1664 percpu_counter_destroy(&sbi->s_es_stats.es_stats_shk_cnt);
1666 percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt);
1668 percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_misses);
1670 percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_hits);
1674 void ext4_es_unregister_shrinker(struct ext4_sb_info *sbi)
1676 percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_hits);
1677 percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_misses);
1678 percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt);
1679 percpu_counter_destroy(&sbi->s_es_stats.es_stats_shk_cnt);
1680 unregister_shrinker(&sbi->s_es_shrinker);
1684 * Shrink extents in given inode from ei->i_es_shrink_lblk till end. Scan at
1685 * most *nr_to_scan extents, update *nr_to_scan accordingly.
1687 * Return 0 if we hit end of tree / interval, 1 if we exhausted nr_to_scan.
1688 * Increment *nr_shrunk by the number of reclaimed extents. Also update
1689 * ei->i_es_shrink_lblk to where we should continue scanning.
1691 static int es_do_reclaim_extents(struct ext4_inode_info *ei, ext4_lblk_t end,
1692 int *nr_to_scan, int *nr_shrunk)
1694 struct inode *inode = &ei->vfs_inode;
1695 struct ext4_es_tree *tree = &ei->i_es_tree;
1696 struct extent_status *es;
1697 struct rb_node *node;
1699 es = __es_tree_search(&tree->root, ei->i_es_shrink_lblk);
1703 while (*nr_to_scan > 0) {
1704 if (es->es_lblk > end) {
1705 ei->i_es_shrink_lblk = end + 1;
1710 node = rb_next(&es->rb_node);
1712 * We can't reclaim delayed extent from status tree because
1713 * fiemap, bigallic, and seek_data/hole need to use it.
1715 if (ext4_es_is_delayed(es))
1717 if (ext4_es_is_referenced(es)) {
1718 ext4_es_clear_referenced(es);
1722 rb_erase(&es->rb_node, &tree->root);
1723 ext4_es_free_extent(inode, es);
1728 es = rb_entry(node, struct extent_status, rb_node);
1730 ei->i_es_shrink_lblk = es->es_lblk;
1733 ei->i_es_shrink_lblk = 0;
1737 static int es_reclaim_extents(struct ext4_inode_info *ei, int *nr_to_scan)
1739 struct inode *inode = &ei->vfs_inode;
1741 ext4_lblk_t start = ei->i_es_shrink_lblk;
1742 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
1743 DEFAULT_RATELIMIT_BURST);
1745 if (ei->i_es_shk_nr == 0)
1748 if (ext4_test_inode_state(inode, EXT4_STATE_EXT_PRECACHED) &&
1750 ext4_warning(inode->i_sb, "forced shrink of precached extents");
1752 if (!es_do_reclaim_extents(ei, EXT_MAX_BLOCKS, nr_to_scan, &nr_shrunk) &&
1754 es_do_reclaim_extents(ei, start - 1, nr_to_scan, &nr_shrunk);
1756 ei->i_es_tree.cache_es = NULL;
1761 * Called to support EXT4_IOC_CLEAR_ES_CACHE. We can only remove
1762 * discretionary entries from the extent status cache. (Some entries
1763 * must be present for proper operations.)
1765 void ext4_clear_inode_es(struct inode *inode)
1767 struct ext4_inode_info *ei = EXT4_I(inode);
1768 struct extent_status *es;
1769 struct ext4_es_tree *tree;
1770 struct rb_node *node;
1772 write_lock(&ei->i_es_lock);
1773 tree = &EXT4_I(inode)->i_es_tree;
1774 tree->cache_es = NULL;
1775 node = rb_first(&tree->root);
1777 es = rb_entry(node, struct extent_status, rb_node);
1778 node = rb_next(node);
1779 if (!ext4_es_is_delayed(es)) {
1780 rb_erase(&es->rb_node, &tree->root);
1781 ext4_es_free_extent(inode, es);
1784 ext4_clear_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
1785 write_unlock(&ei->i_es_lock);
1789 static void ext4_print_pending_tree(struct inode *inode)
1791 struct ext4_pending_tree *tree;
1792 struct rb_node *node;
1793 struct pending_reservation *pr;
1795 printk(KERN_DEBUG "pending reservations for inode %lu:", inode->i_ino);
1796 tree = &EXT4_I(inode)->i_pending_tree;
1797 node = rb_first(&tree->root);
1799 pr = rb_entry(node, struct pending_reservation, rb_node);
1800 printk(KERN_DEBUG " %u", pr->lclu);
1801 node = rb_next(node);
1803 printk(KERN_DEBUG "\n");
1806 #define ext4_print_pending_tree(inode)
1809 int __init ext4_init_pending(void)
1811 ext4_pending_cachep = kmem_cache_create("ext4_pending_reservation",
1812 sizeof(struct pending_reservation),
1813 0, (SLAB_RECLAIM_ACCOUNT), NULL);
1814 if (ext4_pending_cachep == NULL)
1819 void ext4_exit_pending(void)
1821 kmem_cache_destroy(ext4_pending_cachep);
1824 void ext4_init_pending_tree(struct ext4_pending_tree *tree)
1826 tree->root = RB_ROOT;
1830 * __get_pending - retrieve a pointer to a pending reservation
1832 * @inode - file containing the pending cluster reservation
1833 * @lclu - logical cluster of interest
1835 * Returns a pointer to a pending reservation if it's a member of
1836 * the set, and NULL if not. Must be called holding i_es_lock.
1838 static struct pending_reservation *__get_pending(struct inode *inode,
1841 struct ext4_pending_tree *tree;
1842 struct rb_node *node;
1843 struct pending_reservation *pr = NULL;
1845 tree = &EXT4_I(inode)->i_pending_tree;
1846 node = (&tree->root)->rb_node;
1849 pr = rb_entry(node, struct pending_reservation, rb_node);
1850 if (lclu < pr->lclu)
1851 node = node->rb_left;
1852 else if (lclu > pr->lclu)
1853 node = node->rb_right;
1854 else if (lclu == pr->lclu)
1861 * __insert_pending - adds a pending cluster reservation to the set of
1862 * pending reservations
1864 * @inode - file containing the cluster
1865 * @lblk - logical block in the cluster to be added
1867 * Returns 0 on successful insertion and -ENOMEM on failure. If the
1868 * pending reservation is already in the set, returns successfully.
1870 static int __insert_pending(struct inode *inode, ext4_lblk_t lblk)
1872 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1873 struct ext4_pending_tree *tree = &EXT4_I(inode)->i_pending_tree;
1874 struct rb_node **p = &tree->root.rb_node;
1875 struct rb_node *parent = NULL;
1876 struct pending_reservation *pr;
1880 lclu = EXT4_B2C(sbi, lblk);
1881 /* search to find parent for insertion */
1884 pr = rb_entry(parent, struct pending_reservation, rb_node);
1886 if (lclu < pr->lclu) {
1888 } else if (lclu > pr->lclu) {
1889 p = &(*p)->rb_right;
1891 /* pending reservation already inserted */
1896 pr = kmem_cache_alloc(ext4_pending_cachep, GFP_ATOMIC);
1903 rb_link_node(&pr->rb_node, parent, p);
1904 rb_insert_color(&pr->rb_node, &tree->root);
1911 * __remove_pending - removes a pending cluster reservation from the set
1912 * of pending reservations
1914 * @inode - file containing the cluster
1915 * @lblk - logical block in the pending cluster reservation to be removed
1917 * Returns successfully if pending reservation is not a member of the set.
1919 static void __remove_pending(struct inode *inode, ext4_lblk_t lblk)
1921 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1922 struct pending_reservation *pr;
1923 struct ext4_pending_tree *tree;
1925 pr = __get_pending(inode, EXT4_B2C(sbi, lblk));
1927 tree = &EXT4_I(inode)->i_pending_tree;
1928 rb_erase(&pr->rb_node, &tree->root);
1929 kmem_cache_free(ext4_pending_cachep, pr);
1934 * ext4_remove_pending - removes a pending cluster reservation from the set
1935 * of pending reservations
1937 * @inode - file containing the cluster
1938 * @lblk - logical block in the pending cluster reservation to be removed
1940 * Locking for external use of __remove_pending.
1942 void ext4_remove_pending(struct inode *inode, ext4_lblk_t lblk)
1944 struct ext4_inode_info *ei = EXT4_I(inode);
1946 write_lock(&ei->i_es_lock);
1947 __remove_pending(inode, lblk);
1948 write_unlock(&ei->i_es_lock);
1952 * ext4_is_pending - determine whether a cluster has a pending reservation
1955 * @inode - file containing the cluster
1956 * @lblk - logical block in the cluster
1958 * Returns true if there's a pending reservation for the cluster in the
1959 * set of pending reservations, and false if not.
1961 bool ext4_is_pending(struct inode *inode, ext4_lblk_t lblk)
1963 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1964 struct ext4_inode_info *ei = EXT4_I(inode);
1967 read_lock(&ei->i_es_lock);
1968 ret = (bool)(__get_pending(inode, EXT4_B2C(sbi, lblk)) != NULL);
1969 read_unlock(&ei->i_es_lock);
1975 * ext4_es_insert_delayed_block - adds a delayed block to the extents status
1976 * tree, adding a pending reservation where
1979 * @inode - file containing the newly added block
1980 * @lblk - logical block to be added
1981 * @allocated - indicates whether a physical cluster has been allocated for
1982 * the logical cluster that contains the block
1984 * Returns 0 on success, negative error code on failure.
1986 int ext4_es_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk,
1989 struct extent_status newes;
1992 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
1995 es_debug("add [%u/1) delayed to extent status tree of inode %lu\n",
1996 lblk, inode->i_ino);
1998 newes.es_lblk = lblk;
2000 ext4_es_store_pblock_status(&newes, ~0, EXTENT_STATUS_DELAYED);
2001 trace_ext4_es_insert_delayed_block(inode, &newes, allocated);
2003 ext4_es_insert_extent_check(inode, &newes);
2005 write_lock(&EXT4_I(inode)->i_es_lock);
2007 err = __es_remove_extent(inode, lblk, lblk, NULL);
2011 err = __es_insert_extent(inode, &newes);
2012 if (err == -ENOMEM && __es_shrink(EXT4_SB(inode->i_sb),
2013 128, EXT4_I(inode)))
2019 __insert_pending(inode, lblk);
2022 write_unlock(&EXT4_I(inode)->i_es_lock);
2024 ext4_es_print_tree(inode);
2025 ext4_print_pending_tree(inode);
2031 * __es_delayed_clu - count number of clusters containing blocks that
2034 * @inode - file containing block range
2035 * @start - logical block defining start of range
2036 * @end - logical block defining end of range
2038 * Returns the number of clusters containing only delayed (not delayed
2039 * and unwritten) blocks in the range specified by @start and @end. Any
2040 * cluster or part of a cluster within the range and containing a delayed
2041 * and not unwritten block within the range is counted as a whole cluster.
2043 static unsigned int __es_delayed_clu(struct inode *inode, ext4_lblk_t start,
2046 struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
2047 struct extent_status *es;
2048 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2049 struct rb_node *node;
2050 ext4_lblk_t first_lclu, last_lclu;
2051 unsigned long long last_counted_lclu;
2054 /* guaranteed to be unequal to any ext4_lblk_t value */
2055 last_counted_lclu = ~0ULL;
2057 es = __es_tree_search(&tree->root, start);
2059 while (es && (es->es_lblk <= end)) {
2060 if (ext4_es_is_delonly(es)) {
2061 if (es->es_lblk <= start)
2062 first_lclu = EXT4_B2C(sbi, start);
2064 first_lclu = EXT4_B2C(sbi, es->es_lblk);
2066 if (ext4_es_end(es) >= end)
2067 last_lclu = EXT4_B2C(sbi, end);
2069 last_lclu = EXT4_B2C(sbi, ext4_es_end(es));
2071 if (first_lclu == last_counted_lclu)
2072 n += last_lclu - first_lclu;
2074 n += last_lclu - first_lclu + 1;
2075 last_counted_lclu = last_lclu;
2077 node = rb_next(&es->rb_node);
2080 es = rb_entry(node, struct extent_status, rb_node);
2087 * ext4_es_delayed_clu - count number of clusters containing blocks that
2088 * are both delayed and unwritten
2090 * @inode - file containing block range
2091 * @lblk - logical block defining start of range
2092 * @len - number of blocks in range
2094 * Locking for external use of __es_delayed_clu().
2096 unsigned int ext4_es_delayed_clu(struct inode *inode, ext4_lblk_t lblk,
2099 struct ext4_inode_info *ei = EXT4_I(inode);
2106 end = lblk + len - 1;
2107 WARN_ON(end < lblk);
2109 read_lock(&ei->i_es_lock);
2111 n = __es_delayed_clu(inode, lblk, end);
2113 read_unlock(&ei->i_es_lock);
2119 * __revise_pending - makes, cancels, or leaves unchanged pending cluster
2120 * reservations for a specified block range depending
2121 * upon the presence or absence of delayed blocks
2122 * outside the range within clusters at the ends of the
2125 * @inode - file containing the range
2126 * @lblk - logical block defining the start of range
2127 * @len - length of range in blocks
2129 * Used after a newly allocated extent is added to the extents status tree.
2130 * Requires that the extents in the range have either written or unwritten
2131 * status. Must be called while holding i_es_lock.
2133 static void __revise_pending(struct inode *inode, ext4_lblk_t lblk,
2136 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2137 ext4_lblk_t end = lblk + len - 1;
2138 ext4_lblk_t first, last;
2139 bool f_del = false, l_del = false;
2145 * Two cases - block range within single cluster and block range
2146 * spanning two or more clusters. Note that a cluster belonging
2147 * to a range starting and/or ending on a cluster boundary is treated
2148 * as if it does not contain a delayed extent. The new range may
2149 * have allocated space for previously delayed blocks out to the
2150 * cluster boundary, requiring that any pre-existing pending
2151 * reservation be canceled. Because this code only looks at blocks
2152 * outside the range, it should revise pending reservations
2153 * correctly even if the extent represented by the range can't be
2154 * inserted in the extents status tree due to ENOSPC.
2157 if (EXT4_B2C(sbi, lblk) == EXT4_B2C(sbi, end)) {
2158 first = EXT4_LBLK_CMASK(sbi, lblk);
2160 f_del = __es_scan_range(inode, &ext4_es_is_delonly,
2163 __insert_pending(inode, first);
2165 last = EXT4_LBLK_CMASK(sbi, end) +
2166 sbi->s_cluster_ratio - 1;
2168 l_del = __es_scan_range(inode,
2169 &ext4_es_is_delonly,
2172 __insert_pending(inode, last);
2174 __remove_pending(inode, last);
2177 first = EXT4_LBLK_CMASK(sbi, lblk);
2179 f_del = __es_scan_range(inode, &ext4_es_is_delonly,
2182 __insert_pending(inode, first);
2184 __remove_pending(inode, first);
2186 last = EXT4_LBLK_CMASK(sbi, end) + sbi->s_cluster_ratio - 1;
2188 l_del = __es_scan_range(inode, &ext4_es_is_delonly,
2191 __insert_pending(inode, last);
2193 __remove_pending(inode, last);