Commit | Line | Data |
---|---|---|
654598be ZL |
1 | /* |
2 | * fs/ext4/extents_status.c | |
3 | * | |
4 | * Written by Yongqiang Yang <xiaoqiangnk@gmail.com> | |
5 | * Modified by | |
6 | * Allison Henderson <achender@linux.vnet.ibm.com> | |
7 | * Hugh Dickins <hughd@google.com> | |
8 | * Zheng Liu <wenqing.lz@taobao.com> | |
9 | * | |
10 | * Ext4 extents status tree core functions. | |
11 | */ | |
12 | #include <linux/rbtree.h> | |
13 | #include "ext4.h" | |
14 | #include "extents_status.h" | |
15 | #include "ext4_extents.h" | |
16 | ||
992e9fdd ZL |
17 | #include <trace/events/ext4.h> |
18 | ||
654598be ZL |
19 | /* |
20 | * According to previous discussion in Ext4 Developer Workshop, we | |
21 | * will introduce a new structure called io tree to track all extent | |
22 | * status in order to solve some problems that we have met | |
23 | * (e.g. Reservation space warning), and provide extent-level locking. | |
24 | * Delay extent tree is the first step to achieve this goal. It is | |
25 | * original built by Yongqiang Yang. At that time it is called delay | |
06b0c886 | 26 | * extent tree, whose goal is only track delayed extents in memory to |
654598be ZL |
27 | * simplify the implementation of fiemap and bigalloc, and introduce |
28 | * lseek SEEK_DATA/SEEK_HOLE support. That is why it is still called | |
06b0c886 ZL |
29 | * delay extent tree at the first commit. But for better understand |
30 | * what it does, it has been rename to extent status tree. | |
654598be | 31 | * |
06b0c886 ZL |
32 | * Step1: |
33 | * Currently the first step has been done. All delayed extents are | |
34 | * tracked in the tree. It maintains the delayed extent when a delayed | |
35 | * allocation is issued, and the delayed extent is written out or | |
654598be ZL |
36 | * invalidated. Therefore the implementation of fiemap and bigalloc |
37 | * are simplified, and SEEK_DATA/SEEK_HOLE are introduced. | |
38 | * | |
39 | * The following comment describes the implemenmtation of extent | |
40 | * status tree and future works. | |
06b0c886 ZL |
41 | * |
42 | * Step2: | |
43 | * In this step all extent status are tracked by extent status tree. | |
44 | * Thus, we can first try to lookup a block mapping in this tree before | |
45 | * finding it in extent tree. Hence, single extent cache can be removed | |
46 | * because extent status tree can do a better job. Extents in status | |
47 | * tree are loaded on-demand. Therefore, the extent status tree may not | |
48 | * contain all of the extents in a file. Meanwhile we define a shrinker | |
49 | * to reclaim memory from extent status tree because fragmented extent | |
50 | * tree will make status tree cost too much memory. written/unwritten/- | |
51 | * hole extents in the tree will be reclaimed by this shrinker when we | |
52 | * are under high memory pressure. Delayed extents will not be | |
53 | * reclimed because fiemap, bigalloc, and seek_data/hole need it. | |
654598be ZL |
54 | */ |
55 | ||
56 | /* | |
06b0c886 | 57 | * Extent status tree implementation for ext4. |
654598be ZL |
58 | * |
59 | * | |
60 | * ========================================================================== | |
06b0c886 | 61 | * Extent status tree tracks all extent status. |
654598be | 62 | * |
06b0c886 | 63 | * 1. Why we need to implement extent status tree? |
654598be | 64 | * |
06b0c886 | 65 | * Without extent status tree, ext4 identifies a delayed extent by looking |
654598be ZL |
66 | * up page cache, this has several deficiencies - complicated, buggy, |
67 | * and inefficient code. | |
68 | * | |
06b0c886 ZL |
69 | * FIEMAP, SEEK_HOLE/DATA, bigalloc, and writeout all need to know if a |
70 | * block or a range of blocks are belonged to a delayed extent. | |
654598be | 71 | * |
06b0c886 | 72 | * Let us have a look at how they do without extent status tree. |
654598be ZL |
73 | * -- FIEMAP |
74 | * FIEMAP looks up page cache to identify delayed allocations from holes. | |
75 | * | |
76 | * -- SEEK_HOLE/DATA | |
77 | * SEEK_HOLE/DATA has the same problem as FIEMAP. | |
78 | * | |
79 | * -- bigalloc | |
80 | * bigalloc looks up page cache to figure out if a block is | |
81 | * already under delayed allocation or not to determine whether | |
82 | * quota reserving is needed for the cluster. | |
83 | * | |
654598be ZL |
84 | * -- writeout |
85 | * Writeout looks up whole page cache to see if a buffer is | |
86 | * mapped, If there are not very many delayed buffers, then it is | |
87 | * time comsuming. | |
88 | * | |
06b0c886 | 89 | * With extent status tree implementation, FIEMAP, SEEK_HOLE/DATA, |
654598be ZL |
90 | * bigalloc and writeout can figure out if a block or a range of |
91 | * blocks is under delayed allocation(belonged to a delayed extent) or | |
06b0c886 | 92 | * not by searching the extent tree. |
654598be ZL |
93 | * |
94 | * | |
95 | * ========================================================================== | |
06b0c886 ZL |
96 | * 2. Ext4 extent status tree impelmentation |
97 | * | |
98 | * -- extent | |
99 | * A extent is a range of blocks which are contiguous logically and | |
100 | * physically. Unlike extent in extent tree, this extent in ext4 is | |
101 | * a in-memory struct, there is no corresponding on-disk data. There | |
102 | * is no limit on length of extent, so an extent can contain as many | |
103 | * blocks as they are contiguous logically and physically. | |
654598be | 104 | * |
06b0c886 ZL |
105 | * -- extent status tree |
106 | * Every inode has an extent status tree and all allocation blocks | |
107 | * are added to the tree with different status. The extent in the | |
108 | * tree are ordered by logical block no. | |
654598be | 109 | * |
06b0c886 ZL |
110 | * -- operations on a extent status tree |
111 | * There are three important operations on a delayed extent tree: find | |
112 | * next extent, adding a extent(a range of blocks) and removing a extent. | |
654598be | 113 | * |
06b0c886 ZL |
114 | * -- race on a extent status tree |
115 | * Extent status tree is protected by inode->i_es_lock. | |
654598be | 116 | * |
06b0c886 ZL |
117 | * -- memory consumption |
118 | * Fragmented extent tree will make extent status tree cost too much | |
119 | * memory. Hence, we will reclaim written/unwritten/hole extents from | |
120 | * the tree under a heavy memory pressure. | |
654598be ZL |
121 | * |
122 | * | |
123 | * ========================================================================== | |
06b0c886 ZL |
124 | * 3. Performance analysis |
125 | * | |
654598be ZL |
126 | * -- overhead |
127 | * 1. There is a cache extent for write access, so if writes are | |
128 | * not very random, adding space operaions are in O(1) time. | |
129 | * | |
130 | * -- gain | |
131 | * 2. Code is much simpler, more readable, more maintainable and | |
132 | * more efficient. | |
133 | * | |
134 | * | |
135 | * ========================================================================== | |
136 | * 4. TODO list | |
654598be | 137 | * |
06b0c886 | 138 | * -- Refactor delayed space reservation |
654598be ZL |
139 | * |
140 | * -- Extent-level locking | |
141 | */ | |
142 | ||
143 | static struct kmem_cache *ext4_es_cachep; | |
144 | ||
06b0c886 ZL |
145 | static int __es_insert_extent(struct ext4_es_tree *tree, |
146 | struct extent_status *newes); | |
147 | static int __es_remove_extent(struct ext4_es_tree *tree, ext4_lblk_t lblk, | |
148 | ext4_lblk_t end); | |
149 | ||
654598be ZL |
150 | int __init ext4_init_es(void) |
151 | { | |
152 | ext4_es_cachep = KMEM_CACHE(extent_status, SLAB_RECLAIM_ACCOUNT); | |
153 | if (ext4_es_cachep == NULL) | |
154 | return -ENOMEM; | |
155 | return 0; | |
156 | } | |
157 | ||
158 | void ext4_exit_es(void) | |
159 | { | |
160 | if (ext4_es_cachep) | |
161 | kmem_cache_destroy(ext4_es_cachep); | |
162 | } | |
163 | ||
164 | void ext4_es_init_tree(struct ext4_es_tree *tree) | |
165 | { | |
166 | tree->root = RB_ROOT; | |
167 | tree->cache_es = NULL; | |
168 | } | |
169 | ||
170 | #ifdef ES_DEBUG__ | |
171 | static void ext4_es_print_tree(struct inode *inode) | |
172 | { | |
173 | struct ext4_es_tree *tree; | |
174 | struct rb_node *node; | |
175 | ||
176 | printk(KERN_DEBUG "status extents for inode %lu:", inode->i_ino); | |
177 | tree = &EXT4_I(inode)->i_es_tree; | |
178 | node = rb_first(&tree->root); | |
179 | while (node) { | |
180 | struct extent_status *es; | |
181 | es = rb_entry(node, struct extent_status, rb_node); | |
fdc0212e ZL |
182 | printk(KERN_DEBUG " [%u/%u) %llu %llx", |
183 | es->es_lblk, es->es_len, | |
184 | ext4_es_pblock(es), ext4_es_status(es)); | |
654598be ZL |
185 | node = rb_next(node); |
186 | } | |
187 | printk(KERN_DEBUG "\n"); | |
188 | } | |
189 | #else | |
190 | #define ext4_es_print_tree(inode) | |
191 | #endif | |
192 | ||
06b0c886 | 193 | static inline ext4_lblk_t ext4_es_end(struct extent_status *es) |
654598be | 194 | { |
06b0c886 ZL |
195 | BUG_ON(es->es_lblk + es->es_len < es->es_lblk); |
196 | return es->es_lblk + es->es_len - 1; | |
654598be ZL |
197 | } |
198 | ||
199 | /* | |
200 | * search through the tree for an delayed extent with a given offset. If | |
201 | * it can't be found, try to find next extent. | |
202 | */ | |
203 | static struct extent_status *__es_tree_search(struct rb_root *root, | |
06b0c886 | 204 | ext4_lblk_t lblk) |
654598be ZL |
205 | { |
206 | struct rb_node *node = root->rb_node; | |
207 | struct extent_status *es = NULL; | |
208 | ||
209 | while (node) { | |
210 | es = rb_entry(node, struct extent_status, rb_node); | |
06b0c886 | 211 | if (lblk < es->es_lblk) |
654598be | 212 | node = node->rb_left; |
06b0c886 | 213 | else if (lblk > ext4_es_end(es)) |
654598be ZL |
214 | node = node->rb_right; |
215 | else | |
216 | return es; | |
217 | } | |
218 | ||
06b0c886 | 219 | if (es && lblk < es->es_lblk) |
654598be ZL |
220 | return es; |
221 | ||
06b0c886 | 222 | if (es && lblk > ext4_es_end(es)) { |
654598be ZL |
223 | node = rb_next(&es->rb_node); |
224 | return node ? rb_entry(node, struct extent_status, rb_node) : | |
225 | NULL; | |
226 | } | |
227 | ||
228 | return NULL; | |
229 | } | |
230 | ||
231 | /* | |
06b0c886 ZL |
232 | * ext4_es_find_extent: find the 1st delayed extent covering @es->lblk |
233 | * if it exists, otherwise, the next extent after @es->lblk. | |
654598be ZL |
234 | * |
235 | * @inode: the inode which owns delayed extents | |
236 | * @es: delayed extent that we found | |
237 | * | |
238 | * Returns the first block of the next extent after es, otherwise | |
fdc0212e | 239 | * EXT_MAX_BLOCKS if no extent is found. |
654598be ZL |
240 | * Delayed extent is returned via @es. |
241 | */ | |
242 | ext4_lblk_t ext4_es_find_extent(struct inode *inode, struct extent_status *es) | |
243 | { | |
244 | struct ext4_es_tree *tree = NULL; | |
245 | struct extent_status *es1 = NULL; | |
246 | struct rb_node *node; | |
247 | ext4_lblk_t ret = EXT_MAX_BLOCKS; | |
248 | ||
06b0c886 | 249 | trace_ext4_es_find_extent_enter(inode, es->es_lblk); |
992e9fdd | 250 | |
654598be ZL |
251 | read_lock(&EXT4_I(inode)->i_es_lock); |
252 | tree = &EXT4_I(inode)->i_es_tree; | |
253 | ||
fdc0212e ZL |
254 | /* find extent in cache firstly */ |
255 | es->es_len = es->es_pblk = 0; | |
654598be ZL |
256 | if (tree->cache_es) { |
257 | es1 = tree->cache_es; | |
06b0c886 | 258 | if (in_range(es->es_lblk, es1->es_lblk, es1->es_len)) { |
fdc0212e ZL |
259 | es_debug("%u cached by [%u/%u) %llu %llx\n", |
260 | es->es_lblk, es1->es_lblk, es1->es_len, | |
261 | ext4_es_pblock(es1), ext4_es_status(es1)); | |
654598be ZL |
262 | goto out; |
263 | } | |
264 | } | |
265 | ||
06b0c886 | 266 | es1 = __es_tree_search(&tree->root, es->es_lblk); |
654598be ZL |
267 | |
268 | out: | |
269 | if (es1) { | |
270 | tree->cache_es = es1; | |
06b0c886 ZL |
271 | es->es_lblk = es1->es_lblk; |
272 | es->es_len = es1->es_len; | |
fdc0212e | 273 | es->es_pblk = es1->es_pblk; |
654598be ZL |
274 | node = rb_next(&es1->rb_node); |
275 | if (node) { | |
276 | es1 = rb_entry(node, struct extent_status, rb_node); | |
06b0c886 | 277 | ret = es1->es_lblk; |
654598be ZL |
278 | } |
279 | } | |
280 | ||
281 | read_unlock(&EXT4_I(inode)->i_es_lock); | |
992e9fdd ZL |
282 | |
283 | trace_ext4_es_find_extent_exit(inode, es, ret); | |
654598be ZL |
284 | return ret; |
285 | } | |
286 | ||
287 | static struct extent_status * | |
fdc0212e | 288 | ext4_es_alloc_extent(ext4_lblk_t lblk, ext4_lblk_t len, ext4_fsblk_t pblk) |
654598be ZL |
289 | { |
290 | struct extent_status *es; | |
291 | es = kmem_cache_alloc(ext4_es_cachep, GFP_ATOMIC); | |
292 | if (es == NULL) | |
293 | return NULL; | |
06b0c886 ZL |
294 | es->es_lblk = lblk; |
295 | es->es_len = len; | |
fdc0212e | 296 | es->es_pblk = pblk; |
654598be ZL |
297 | return es; |
298 | } | |
299 | ||
300 | static void ext4_es_free_extent(struct extent_status *es) | |
301 | { | |
302 | kmem_cache_free(ext4_es_cachep, es); | |
303 | } | |
304 | ||
06b0c886 ZL |
305 | /* |
306 | * Check whether or not two extents can be merged | |
307 | * Condition: | |
308 | * - logical block number is contiguous | |
fdc0212e ZL |
309 | * - physical block number is contiguous |
310 | * - status is equal | |
06b0c886 ZL |
311 | */ |
312 | static int ext4_es_can_be_merged(struct extent_status *es1, | |
313 | struct extent_status *es2) | |
314 | { | |
315 | if (es1->es_lblk + es1->es_len != es2->es_lblk) | |
316 | return 0; | |
317 | ||
fdc0212e ZL |
318 | if (ext4_es_status(es1) != ext4_es_status(es2)) |
319 | return 0; | |
320 | ||
321 | if ((ext4_es_is_written(es1) || ext4_es_is_unwritten(es1)) && | |
322 | (ext4_es_pblock(es1) + es1->es_len != ext4_es_pblock(es2))) | |
323 | return 0; | |
324 | ||
06b0c886 ZL |
325 | return 1; |
326 | } | |
327 | ||
654598be ZL |
328 | static struct extent_status * |
329 | ext4_es_try_to_merge_left(struct ext4_es_tree *tree, struct extent_status *es) | |
330 | { | |
331 | struct extent_status *es1; | |
332 | struct rb_node *node; | |
333 | ||
334 | node = rb_prev(&es->rb_node); | |
335 | if (!node) | |
336 | return es; | |
337 | ||
338 | es1 = rb_entry(node, struct extent_status, rb_node); | |
06b0c886 ZL |
339 | if (ext4_es_can_be_merged(es1, es)) { |
340 | es1->es_len += es->es_len; | |
654598be ZL |
341 | rb_erase(&es->rb_node, &tree->root); |
342 | ext4_es_free_extent(es); | |
343 | es = es1; | |
344 | } | |
345 | ||
346 | return es; | |
347 | } | |
348 | ||
349 | static struct extent_status * | |
350 | ext4_es_try_to_merge_right(struct ext4_es_tree *tree, struct extent_status *es) | |
351 | { | |
352 | struct extent_status *es1; | |
353 | struct rb_node *node; | |
354 | ||
355 | node = rb_next(&es->rb_node); | |
356 | if (!node) | |
357 | return es; | |
358 | ||
359 | es1 = rb_entry(node, struct extent_status, rb_node); | |
06b0c886 ZL |
360 | if (ext4_es_can_be_merged(es, es1)) { |
361 | es->es_len += es1->es_len; | |
654598be ZL |
362 | rb_erase(node, &tree->root); |
363 | ext4_es_free_extent(es1); | |
364 | } | |
365 | ||
366 | return es; | |
367 | } | |
368 | ||
06b0c886 ZL |
369 | static int __es_insert_extent(struct ext4_es_tree *tree, |
370 | struct extent_status *newes) | |
654598be ZL |
371 | { |
372 | struct rb_node **p = &tree->root.rb_node; | |
373 | struct rb_node *parent = NULL; | |
374 | struct extent_status *es; | |
654598be ZL |
375 | |
376 | while (*p) { | |
377 | parent = *p; | |
378 | es = rb_entry(parent, struct extent_status, rb_node); | |
379 | ||
06b0c886 ZL |
380 | if (newes->es_lblk < es->es_lblk) { |
381 | if (ext4_es_can_be_merged(newes, es)) { | |
382 | /* | |
383 | * Here we can modify es_lblk directly | |
384 | * because it isn't overlapped. | |
385 | */ | |
386 | es->es_lblk = newes->es_lblk; | |
387 | es->es_len += newes->es_len; | |
fdc0212e ZL |
388 | if (ext4_es_is_written(es) || |
389 | ext4_es_is_unwritten(es)) | |
390 | ext4_es_store_pblock(es, | |
391 | newes->es_pblk); | |
654598be ZL |
392 | es = ext4_es_try_to_merge_left(tree, es); |
393 | goto out; | |
394 | } | |
395 | p = &(*p)->rb_left; | |
06b0c886 ZL |
396 | } else if (newes->es_lblk > ext4_es_end(es)) { |
397 | if (ext4_es_can_be_merged(es, newes)) { | |
398 | es->es_len += newes->es_len; | |
654598be ZL |
399 | es = ext4_es_try_to_merge_right(tree, es); |
400 | goto out; | |
401 | } | |
402 | p = &(*p)->rb_right; | |
403 | } else { | |
06b0c886 ZL |
404 | BUG_ON(1); |
405 | return -EINVAL; | |
654598be ZL |
406 | } |
407 | } | |
408 | ||
fdc0212e ZL |
409 | es = ext4_es_alloc_extent(newes->es_lblk, newes->es_len, |
410 | newes->es_pblk); | |
654598be ZL |
411 | if (!es) |
412 | return -ENOMEM; | |
413 | rb_link_node(&es->rb_node, parent, p); | |
414 | rb_insert_color(&es->rb_node, &tree->root); | |
415 | ||
416 | out: | |
417 | tree->cache_es = es; | |
418 | return 0; | |
419 | } | |
420 | ||
421 | /* | |
06b0c886 | 422 | * ext4_es_insert_extent() adds a space to a extent status tree. |
654598be ZL |
423 | * |
424 | * ext4_es_insert_extent is called by ext4_da_write_begin and | |
425 | * ext4_es_remove_extent. | |
426 | * | |
427 | * Return 0 on success, error code on failure. | |
428 | */ | |
06b0c886 | 429 | int ext4_es_insert_extent(struct inode *inode, ext4_lblk_t lblk, |
fdc0212e ZL |
430 | ext4_lblk_t len, ext4_fsblk_t pblk, |
431 | unsigned long long status) | |
654598be ZL |
432 | { |
433 | struct ext4_es_tree *tree; | |
06b0c886 ZL |
434 | struct extent_status newes; |
435 | ext4_lblk_t end = lblk + len - 1; | |
654598be ZL |
436 | int err = 0; |
437 | ||
fdc0212e ZL |
438 | es_debug("add [%u/%u) %llu %llx to extent status tree of inode %lu\n", |
439 | lblk, len, pblk, status, inode->i_ino); | |
06b0c886 ZL |
440 | |
441 | BUG_ON(end < lblk); | |
442 | ||
443 | newes.es_lblk = lblk; | |
444 | newes.es_len = len; | |
fdc0212e ZL |
445 | ext4_es_store_pblock(&newes, pblk); |
446 | ext4_es_store_status(&newes, status); | |
447 | trace_ext4_es_insert_extent(inode, &newes); | |
654598be ZL |
448 | |
449 | write_lock(&EXT4_I(inode)->i_es_lock); | |
450 | tree = &EXT4_I(inode)->i_es_tree; | |
06b0c886 ZL |
451 | err = __es_remove_extent(tree, lblk, end); |
452 | if (err != 0) | |
453 | goto error; | |
454 | err = __es_insert_extent(tree, &newes); | |
455 | ||
456 | error: | |
654598be ZL |
457 | write_unlock(&EXT4_I(inode)->i_es_lock); |
458 | ||
459 | ext4_es_print_tree(inode); | |
460 | ||
461 | return err; | |
462 | } | |
463 | ||
06b0c886 ZL |
464 | static int __es_remove_extent(struct ext4_es_tree *tree, ext4_lblk_t lblk, |
465 | ext4_lblk_t end) | |
654598be ZL |
466 | { |
467 | struct rb_node *node; | |
654598be ZL |
468 | struct extent_status *es; |
469 | struct extent_status orig_es; | |
06b0c886 | 470 | ext4_lblk_t len1, len2; |
fdc0212e | 471 | ext4_fsblk_t block; |
654598be ZL |
472 | int err = 0; |
473 | ||
06b0c886 | 474 | es = __es_tree_search(&tree->root, lblk); |
654598be ZL |
475 | if (!es) |
476 | goto out; | |
06b0c886 | 477 | if (es->es_lblk > end) |
654598be ZL |
478 | goto out; |
479 | ||
480 | /* Simply invalidate cache_es. */ | |
481 | tree->cache_es = NULL; | |
482 | ||
06b0c886 ZL |
483 | orig_es.es_lblk = es->es_lblk; |
484 | orig_es.es_len = es->es_len; | |
fdc0212e ZL |
485 | orig_es.es_pblk = es->es_pblk; |
486 | ||
06b0c886 ZL |
487 | len1 = lblk > es->es_lblk ? lblk - es->es_lblk : 0; |
488 | len2 = ext4_es_end(es) > end ? ext4_es_end(es) - end : 0; | |
654598be | 489 | if (len1 > 0) |
06b0c886 | 490 | es->es_len = len1; |
654598be ZL |
491 | if (len2 > 0) { |
492 | if (len1 > 0) { | |
06b0c886 ZL |
493 | struct extent_status newes; |
494 | ||
495 | newes.es_lblk = end + 1; | |
496 | newes.es_len = len2; | |
fdc0212e ZL |
497 | if (ext4_es_is_written(&orig_es) || |
498 | ext4_es_is_unwritten(&orig_es)) { | |
499 | block = ext4_es_pblock(&orig_es) + | |
500 | orig_es.es_len - len2; | |
501 | ext4_es_store_pblock(&newes, block); | |
502 | } | |
503 | ext4_es_store_status(&newes, ext4_es_status(&orig_es)); | |
06b0c886 | 504 | err = __es_insert_extent(tree, &newes); |
654598be | 505 | if (err) { |
06b0c886 ZL |
506 | es->es_lblk = orig_es.es_lblk; |
507 | es->es_len = orig_es.es_len; | |
654598be ZL |
508 | goto out; |
509 | } | |
510 | } else { | |
06b0c886 ZL |
511 | es->es_lblk = end + 1; |
512 | es->es_len = len2; | |
fdc0212e ZL |
513 | if (ext4_es_is_written(es) || |
514 | ext4_es_is_unwritten(es)) { | |
515 | block = orig_es.es_pblk + orig_es.es_len - len2; | |
516 | ext4_es_store_pblock(es, block); | |
517 | } | |
654598be ZL |
518 | } |
519 | goto out; | |
520 | } | |
521 | ||
522 | if (len1 > 0) { | |
523 | node = rb_next(&es->rb_node); | |
524 | if (node) | |
525 | es = rb_entry(node, struct extent_status, rb_node); | |
526 | else | |
527 | es = NULL; | |
528 | } | |
529 | ||
06b0c886 | 530 | while (es && ext4_es_end(es) <= end) { |
654598be ZL |
531 | node = rb_next(&es->rb_node); |
532 | rb_erase(&es->rb_node, &tree->root); | |
533 | ext4_es_free_extent(es); | |
534 | if (!node) { | |
535 | es = NULL; | |
536 | break; | |
537 | } | |
538 | es = rb_entry(node, struct extent_status, rb_node); | |
539 | } | |
540 | ||
06b0c886 | 541 | if (es && es->es_lblk < end + 1) { |
fdc0212e ZL |
542 | ext4_lblk_t orig_len = es->es_len; |
543 | ||
06b0c886 ZL |
544 | len1 = ext4_es_end(es) - end; |
545 | es->es_lblk = end + 1; | |
546 | es->es_len = len1; | |
fdc0212e ZL |
547 | if (ext4_es_is_written(es) || ext4_es_is_unwritten(es)) { |
548 | block = es->es_pblk + orig_len - len1; | |
549 | ext4_es_store_pblock(es, block); | |
550 | } | |
654598be ZL |
551 | } |
552 | ||
553 | out: | |
06b0c886 ZL |
554 | return err; |
555 | } | |
556 | ||
557 | /* | |
558 | * ext4_es_remove_extent() removes a space from a extent status tree. | |
559 | * | |
560 | * Return 0 on success, error code on failure. | |
561 | */ | |
562 | int ext4_es_remove_extent(struct inode *inode, ext4_lblk_t lblk, | |
563 | ext4_lblk_t len) | |
564 | { | |
565 | struct ext4_es_tree *tree; | |
566 | ext4_lblk_t end; | |
567 | int err = 0; | |
568 | ||
569 | trace_ext4_es_remove_extent(inode, lblk, len); | |
570 | es_debug("remove [%u/%u) from extent status tree of inode %lu\n", | |
571 | lblk, len, inode->i_ino); | |
572 | ||
573 | end = lblk + len - 1; | |
574 | BUG_ON(end < lblk); | |
575 | ||
576 | tree = &EXT4_I(inode)->i_es_tree; | |
577 | ||
578 | write_lock(&EXT4_I(inode)->i_es_lock); | |
579 | err = __es_remove_extent(tree, lblk, end); | |
654598be ZL |
580 | write_unlock(&EXT4_I(inode)->i_es_lock); |
581 | ext4_es_print_tree(inode); | |
582 | return err; | |
583 | } |