2 * fs/ext4/extents_status.c
4 * Written by Yongqiang Yang <xiaoqiangnk@gmail.com>
6 * Allison Henderson <achender@linux.vnet.ibm.com>
7 * Hugh Dickins <hughd@google.com>
8 * Zheng Liu <wenqing.lz@taobao.com>
10 * Ext4 extents status tree core functions.
12 #include <linux/rbtree.h>
14 #include "extents_status.h"
15 #include "ext4_extents.h"
18 * According to previous discussion in Ext4 Developer Workshop, we
19 * will introduce a new structure called io tree to track all extent
20 * status in order to solve some problems that we have met
21 * (e.g. Reservation space warning), and provide extent-level locking.
22 * Delay extent tree is the first step to achieve this goal. It is
23 * original built by Yongqiang Yang. At that time it is called delay
24 * extent tree, whose goal is only track delay extent in memory to
25 * simplify the implementation of fiemap and bigalloc, and introduce
26 * lseek SEEK_DATA/SEEK_HOLE support. That is why it is still called
27 * delay extent tree at the following comment. But for better
28 * understand what it does, it has been rename to extent status tree.
30 * Currently the first step has been done. All delay extents are
31 * tracked in the tree. It maintains the delay extent when a delay
32 * allocation is issued, and the delay extent is written out or
33 * invalidated. Therefore the implementation of fiemap and bigalloc
34 * are simplified, and SEEK_DATA/SEEK_HOLE are introduced.
36 * The following comment describes the implemenmtation of extent
37 * status tree and future works.
41 * extents status tree implementation for ext4.
44 * ==========================================================================
45 * Extents status encompass delayed extents and extent locks
47 * 1. Why delayed extent implementation ?
49 * Without delayed extent, ext4 identifies a delayed extent by looking
50 * up page cache, this has several deficiencies - complicated, buggy,
51 * and inefficient code.
53 * FIEMAP, SEEK_HOLE/DATA, bigalloc, punch hole and writeout all need
54 * to know if a block or a range of blocks are belonged to a delayed
57 * Let us have a look at how they do without delayed extents implementation.
59 * FIEMAP looks up page cache to identify delayed allocations from holes.
62 * SEEK_HOLE/DATA has the same problem as FIEMAP.
65 * bigalloc looks up page cache to figure out if a block is
66 * already under delayed allocation or not to determine whether
67 * quota reserving is needed for the cluster.
70 * punch hole looks up page cache to identify a delayed extent.
73 * Writeout looks up whole page cache to see if a buffer is
74 * mapped, If there are not very many delayed buffers, then it is
77 * With delayed extents implementation, FIEMAP, SEEK_HOLE/DATA,
78 * bigalloc and writeout can figure out if a block or a range of
79 * blocks is under delayed allocation(belonged to a delayed extent) or
80 * not by searching the delayed extent tree.
83 * ==========================================================================
84 * 2. ext4 delayed extents impelmentation
87 * A delayed extent is a range of blocks which are contiguous
88 * logically and under delayed allocation. Unlike extent in
89 * ext4, delayed extent in ext4 is a in-memory struct, there is
90 * no corresponding on-disk data. There is no limit on length of
91 * delayed extent, so a delayed extent can contain as many blocks
92 * as they are contiguous logically.
94 * -- delayed extent tree
95 * Every inode has a delayed extent tree and all under delayed
96 * allocation blocks are added to the tree as delayed extents.
97 * Delayed extents in the tree are ordered by logical block no.
99 * -- operations on a delayed extent tree
100 * There are three operations on a delayed extent tree: find next
101 * delayed extent, adding a space(a range of blocks) and removing
104 * -- race on a delayed extent tree
105 * Delayed extent tree is protected inode->i_es_lock.
108 * ==========================================================================
109 * 3. performance analysis
111 * 1. There is a cache extent for write access, so if writes are
112 * not very random, adding space operaions are in O(1) time.
115 * 2. Code is much simpler, more readable, more maintainable and
119 * ==========================================================================
121 * -- Track all extent status
123 * -- Improve get block process
125 * -- Extent-level locking
128 static struct kmem_cache *ext4_es_cachep;
130 int __init ext4_init_es(void)
132 ext4_es_cachep = KMEM_CACHE(extent_status, SLAB_RECLAIM_ACCOUNT);
133 if (ext4_es_cachep == NULL)
138 void ext4_exit_es(void)
141 kmem_cache_destroy(ext4_es_cachep);
144 void ext4_es_init_tree(struct ext4_es_tree *tree)
146 tree->root = RB_ROOT;
147 tree->cache_es = NULL;
151 static void ext4_es_print_tree(struct inode *inode)
153 struct ext4_es_tree *tree;
154 struct rb_node *node;
156 printk(KERN_DEBUG "status extents for inode %lu:", inode->i_ino);
157 tree = &EXT4_I(inode)->i_es_tree;
158 node = rb_first(&tree->root);
160 struct extent_status *es;
161 es = rb_entry(node, struct extent_status, rb_node);
162 printk(KERN_DEBUG " [%u/%u)", es->start, es->len);
163 node = rb_next(node);
165 printk(KERN_DEBUG "\n");
168 #define ext4_es_print_tree(inode)
171 static inline ext4_lblk_t extent_status_end(struct extent_status *es)
173 BUG_ON(es->start + es->len < es->start);
174 return es->start + es->len - 1;
178 * search through the tree for an delayed extent with a given offset. If
179 * it can't be found, try to find next extent.
181 static struct extent_status *__es_tree_search(struct rb_root *root,
184 struct rb_node *node = root->rb_node;
185 struct extent_status *es = NULL;
188 es = rb_entry(node, struct extent_status, rb_node);
189 if (offset < es->start)
190 node = node->rb_left;
191 else if (offset > extent_status_end(es))
192 node = node->rb_right;
197 if (es && offset < es->start)
200 if (es && offset > extent_status_end(es)) {
201 node = rb_next(&es->rb_node);
202 return node ? rb_entry(node, struct extent_status, rb_node) :
210 * ext4_es_find_extent: find the 1st delayed extent covering @es->start
211 * if it exists, otherwise, the next extent after @es->start.
213 * @inode: the inode which owns delayed extents
214 * @es: delayed extent that we found
216 * Returns the first block of the next extent after es, otherwise
217 * EXT_MAX_BLOCKS if no delay extent is found.
218 * Delayed extent is returned via @es.
220 ext4_lblk_t ext4_es_find_extent(struct inode *inode, struct extent_status *es)
222 struct ext4_es_tree *tree = NULL;
223 struct extent_status *es1 = NULL;
224 struct rb_node *node;
225 ext4_lblk_t ret = EXT_MAX_BLOCKS;
227 read_lock(&EXT4_I(inode)->i_es_lock);
228 tree = &EXT4_I(inode)->i_es_tree;
230 /* find delay extent in cache firstly */
231 if (tree->cache_es) {
232 es1 = tree->cache_es;
233 if (in_range(es->start, es1->start, es1->len)) {
234 es_debug("%u cached by [%u/%u)\n",
235 es->start, es1->start, es1->len);
241 es1 = __es_tree_search(&tree->root, es->start);
245 tree->cache_es = es1;
246 es->start = es1->start;
248 node = rb_next(&es1->rb_node);
250 es1 = rb_entry(node, struct extent_status, rb_node);
255 read_unlock(&EXT4_I(inode)->i_es_lock);
259 static struct extent_status *
260 ext4_es_alloc_extent(ext4_lblk_t start, ext4_lblk_t len)
262 struct extent_status *es;
263 es = kmem_cache_alloc(ext4_es_cachep, GFP_ATOMIC);
271 static void ext4_es_free_extent(struct extent_status *es)
273 kmem_cache_free(ext4_es_cachep, es);
276 static struct extent_status *
277 ext4_es_try_to_merge_left(struct ext4_es_tree *tree, struct extent_status *es)
279 struct extent_status *es1;
280 struct rb_node *node;
282 node = rb_prev(&es->rb_node);
286 es1 = rb_entry(node, struct extent_status, rb_node);
287 if (es->start == extent_status_end(es1) + 1) {
289 rb_erase(&es->rb_node, &tree->root);
290 ext4_es_free_extent(es);
297 static struct extent_status *
298 ext4_es_try_to_merge_right(struct ext4_es_tree *tree, struct extent_status *es)
300 struct extent_status *es1;
301 struct rb_node *node;
303 node = rb_next(&es->rb_node);
307 es1 = rb_entry(node, struct extent_status, rb_node);
308 if (es1->start == extent_status_end(es) + 1) {
310 rb_erase(node, &tree->root);
311 ext4_es_free_extent(es1);
317 static int __es_insert_extent(struct ext4_es_tree *tree, ext4_lblk_t offset,
320 struct rb_node **p = &tree->root.rb_node;
321 struct rb_node *parent = NULL;
322 struct extent_status *es;
323 ext4_lblk_t end = offset + len - 1;
325 BUG_ON(end < offset);
327 if (es && offset == (extent_status_end(es) + 1)) {
328 es_debug("cached by [%u/%u)\n", es->start, es->len);
330 es = ext4_es_try_to_merge_right(tree, es);
332 } else if (es && es->start == end + 1) {
333 es_debug("cached by [%u/%u)\n", es->start, es->len);
336 es = ext4_es_try_to_merge_left(tree, es);
338 } else if (es && es->start <= offset &&
339 end <= extent_status_end(es)) {
340 es_debug("cached by [%u/%u)\n", es->start, es->len);
346 es = rb_entry(parent, struct extent_status, rb_node);
348 if (offset < es->start) {
349 if (es->start == end + 1) {
352 es = ext4_es_try_to_merge_left(tree, es);
356 } else if (offset > extent_status_end(es)) {
357 if (offset == extent_status_end(es) + 1) {
359 es = ext4_es_try_to_merge_right(tree, es);
364 if (extent_status_end(es) <= end)
365 es->len = offset - es->start + len;
370 es = ext4_es_alloc_extent(offset, len);
373 rb_link_node(&es->rb_node, parent, p);
374 rb_insert_color(&es->rb_node, &tree->root);
382 * ext4_es_insert_extent() adds a space to a delayed extent tree.
383 * Caller holds inode->i_es_lock.
385 * ext4_es_insert_extent is called by ext4_da_write_begin and
386 * ext4_es_remove_extent.
388 * Return 0 on success, error code on failure.
390 int ext4_es_insert_extent(struct inode *inode, ext4_lblk_t offset,
393 struct ext4_es_tree *tree;
396 es_debug("add [%u/%u) to extent status tree of inode %lu\n",
397 offset, len, inode->i_ino);
399 write_lock(&EXT4_I(inode)->i_es_lock);
400 tree = &EXT4_I(inode)->i_es_tree;
401 err = __es_insert_extent(tree, offset, len);
402 write_unlock(&EXT4_I(inode)->i_es_lock);
404 ext4_es_print_tree(inode);
410 * ext4_es_remove_extent() removes a space from a delayed extent tree.
411 * Caller holds inode->i_es_lock.
413 * Return 0 on success, error code on failure.
415 int ext4_es_remove_extent(struct inode *inode, ext4_lblk_t offset,
418 struct rb_node *node;
419 struct ext4_es_tree *tree;
420 struct extent_status *es;
421 struct extent_status orig_es;
422 ext4_lblk_t len1, len2, end;
425 es_debug("remove [%u/%u) from extent status tree of inode %lu\n",
426 offset, len, inode->i_ino);
428 end = offset + len - 1;
429 BUG_ON(end < offset);
430 write_lock(&EXT4_I(inode)->i_es_lock);
431 tree = &EXT4_I(inode)->i_es_tree;
432 es = __es_tree_search(&tree->root, offset);
438 /* Simply invalidate cache_es. */
439 tree->cache_es = NULL;
441 orig_es.start = es->start;
442 orig_es.len = es->len;
443 len1 = offset > es->start ? offset - es->start : 0;
444 len2 = extent_status_end(es) > end ?
445 extent_status_end(es) - end : 0;
450 err = __es_insert_extent(tree, end + 1, len2);
452 es->start = orig_es.start;
453 es->len = orig_es.len;
464 node = rb_next(&es->rb_node);
466 es = rb_entry(node, struct extent_status, rb_node);
471 while (es && extent_status_end(es) <= end) {
472 node = rb_next(&es->rb_node);
473 rb_erase(&es->rb_node, &tree->root);
474 ext4_es_free_extent(es);
479 es = rb_entry(node, struct extent_status, rb_node);
482 if (es && es->start < end + 1) {
483 len1 = extent_status_end(es) - end;
489 write_unlock(&EXT4_I(inode)->i_es_lock);
490 ext4_es_print_tree(inode);