]>
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 | ||
bdedbb7b ZL |
145 | static int __es_insert_extent(struct inode *inode, struct extent_status *newes); |
146 | static int __es_remove_extent(struct inode *inode, ext4_lblk_t lblk, | |
06b0c886 | 147 | ext4_lblk_t end); |
74cd15cd ZL |
148 | static int __es_try_to_reclaim_extents(struct ext4_inode_info *ei, |
149 | int nr_to_scan); | |
06b0c886 | 150 | |
654598be ZL |
151 | int __init ext4_init_es(void) |
152 | { | |
24630774 TT |
153 | ext4_es_cachep = kmem_cache_create("ext4_extent_status", |
154 | sizeof(struct extent_status), | |
155 | 0, (SLAB_RECLAIM_ACCOUNT), NULL); | |
654598be ZL |
156 | if (ext4_es_cachep == NULL) |
157 | return -ENOMEM; | |
158 | return 0; | |
159 | } | |
160 | ||
161 | void ext4_exit_es(void) | |
162 | { | |
163 | if (ext4_es_cachep) | |
164 | kmem_cache_destroy(ext4_es_cachep); | |
165 | } | |
166 | ||
167 | void ext4_es_init_tree(struct ext4_es_tree *tree) | |
168 | { | |
169 | tree->root = RB_ROOT; | |
170 | tree->cache_es = NULL; | |
171 | } | |
172 | ||
173 | #ifdef ES_DEBUG__ | |
174 | static void ext4_es_print_tree(struct inode *inode) | |
175 | { | |
176 | struct ext4_es_tree *tree; | |
177 | struct rb_node *node; | |
178 | ||
179 | printk(KERN_DEBUG "status extents for inode %lu:", inode->i_ino); | |
180 | tree = &EXT4_I(inode)->i_es_tree; | |
181 | node = rb_first(&tree->root); | |
182 | while (node) { | |
183 | struct extent_status *es; | |
184 | es = rb_entry(node, struct extent_status, rb_node); | |
fdc0212e ZL |
185 | printk(KERN_DEBUG " [%u/%u) %llu %llx", |
186 | es->es_lblk, es->es_len, | |
187 | ext4_es_pblock(es), ext4_es_status(es)); | |
654598be ZL |
188 | node = rb_next(node); |
189 | } | |
190 | printk(KERN_DEBUG "\n"); | |
191 | } | |
192 | #else | |
193 | #define ext4_es_print_tree(inode) | |
194 | #endif | |
195 | ||
06b0c886 | 196 | static inline ext4_lblk_t ext4_es_end(struct extent_status *es) |
654598be | 197 | { |
06b0c886 ZL |
198 | BUG_ON(es->es_lblk + es->es_len < es->es_lblk); |
199 | return es->es_lblk + es->es_len - 1; | |
654598be ZL |
200 | } |
201 | ||
202 | /* | |
203 | * search through the tree for an delayed extent with a given offset. If | |
204 | * it can't be found, try to find next extent. | |
205 | */ | |
206 | static struct extent_status *__es_tree_search(struct rb_root *root, | |
06b0c886 | 207 | ext4_lblk_t lblk) |
654598be ZL |
208 | { |
209 | struct rb_node *node = root->rb_node; | |
210 | struct extent_status *es = NULL; | |
211 | ||
212 | while (node) { | |
213 | es = rb_entry(node, struct extent_status, rb_node); | |
06b0c886 | 214 | if (lblk < es->es_lblk) |
654598be | 215 | node = node->rb_left; |
06b0c886 | 216 | else if (lblk > ext4_es_end(es)) |
654598be ZL |
217 | node = node->rb_right; |
218 | else | |
219 | return es; | |
220 | } | |
221 | ||
06b0c886 | 222 | if (es && lblk < es->es_lblk) |
654598be ZL |
223 | return es; |
224 | ||
06b0c886 | 225 | if (es && lblk > ext4_es_end(es)) { |
654598be ZL |
226 | node = rb_next(&es->rb_node); |
227 | return node ? rb_entry(node, struct extent_status, rb_node) : | |
228 | NULL; | |
229 | } | |
230 | ||
231 | return NULL; | |
232 | } | |
233 | ||
234 | /* | |
be401363 | 235 | * ext4_es_find_delayed_extent: find the 1st delayed extent covering @es->lblk |
06b0c886 | 236 | * if it exists, otherwise, the next extent after @es->lblk. |
654598be ZL |
237 | * |
238 | * @inode: the inode which owns delayed extents | |
be401363 | 239 | * @lblk: the offset where we start to search |
654598be | 240 | * @es: delayed extent that we found |
654598be | 241 | */ |
be401363 ZL |
242 | void ext4_es_find_delayed_extent(struct inode *inode, ext4_lblk_t lblk, |
243 | struct extent_status *es) | |
654598be ZL |
244 | { |
245 | struct ext4_es_tree *tree = NULL; | |
246 | struct extent_status *es1 = NULL; | |
247 | struct rb_node *node; | |
654598be | 248 | |
be401363 ZL |
249 | BUG_ON(es == NULL); |
250 | trace_ext4_es_find_delayed_extent_enter(inode, lblk); | |
992e9fdd | 251 | |
654598be ZL |
252 | read_lock(&EXT4_I(inode)->i_es_lock); |
253 | tree = &EXT4_I(inode)->i_es_tree; | |
254 | ||
fdc0212e | 255 | /* find extent in cache firstly */ |
be401363 | 256 | es->es_lblk = es->es_len = es->es_pblk = 0; |
654598be ZL |
257 | if (tree->cache_es) { |
258 | es1 = tree->cache_es; | |
be401363 | 259 | if (in_range(lblk, es1->es_lblk, es1->es_len)) { |
fdc0212e | 260 | es_debug("%u cached by [%u/%u) %llu %llx\n", |
be401363 | 261 | lblk, es1->es_lblk, es1->es_len, |
fdc0212e | 262 | ext4_es_pblock(es1), ext4_es_status(es1)); |
654598be ZL |
263 | goto out; |
264 | } | |
265 | } | |
266 | ||
be401363 | 267 | es1 = __es_tree_search(&tree->root, lblk); |
654598be ZL |
268 | |
269 | out: | |
be401363 ZL |
270 | if (es1 && !ext4_es_is_delayed(es1)) { |
271 | while ((node = rb_next(&es1->rb_node)) != NULL) { | |
272 | es1 = rb_entry(node, struct extent_status, rb_node); | |
273 | if (ext4_es_is_delayed(es1)) | |
274 | break; | |
275 | } | |
276 | } | |
277 | ||
278 | if (es1 && ext4_es_is_delayed(es1)) { | |
654598be | 279 | tree->cache_es = es1; |
06b0c886 ZL |
280 | es->es_lblk = es1->es_lblk; |
281 | es->es_len = es1->es_len; | |
fdc0212e | 282 | es->es_pblk = es1->es_pblk; |
654598be ZL |
283 | } |
284 | ||
285 | read_unlock(&EXT4_I(inode)->i_es_lock); | |
992e9fdd | 286 | |
74cd15cd | 287 | ext4_es_lru_add(inode); |
be401363 | 288 | trace_ext4_es_find_delayed_extent_exit(inode, es); |
654598be ZL |
289 | } |
290 | ||
291 | static struct extent_status * | |
bdedbb7b ZL |
292 | ext4_es_alloc_extent(struct inode *inode, ext4_lblk_t lblk, ext4_lblk_t len, |
293 | ext4_fsblk_t pblk) | |
654598be ZL |
294 | { |
295 | struct extent_status *es; | |
296 | es = kmem_cache_alloc(ext4_es_cachep, GFP_ATOMIC); | |
297 | if (es == NULL) | |
298 | return NULL; | |
06b0c886 ZL |
299 | es->es_lblk = lblk; |
300 | es->es_len = len; | |
fdc0212e | 301 | es->es_pblk = pblk; |
74cd15cd ZL |
302 | |
303 | /* | |
304 | * We don't count delayed extent because we never try to reclaim them | |
305 | */ | |
24630774 | 306 | if (!ext4_es_is_delayed(es)) { |
74cd15cd | 307 | EXT4_I(inode)->i_es_lru_nr++; |
1ac6466f | 308 | percpu_counter_inc(&EXT4_SB(inode->i_sb)->s_extent_cache_cnt); |
24630774 | 309 | } |
74cd15cd | 310 | |
654598be ZL |
311 | return es; |
312 | } | |
313 | ||
bdedbb7b | 314 | static void ext4_es_free_extent(struct inode *inode, struct extent_status *es) |
654598be | 315 | { |
74cd15cd ZL |
316 | /* Decrease the lru counter when this es is not delayed */ |
317 | if (!ext4_es_is_delayed(es)) { | |
318 | BUG_ON(EXT4_I(inode)->i_es_lru_nr == 0); | |
319 | EXT4_I(inode)->i_es_lru_nr--; | |
1ac6466f | 320 | percpu_counter_dec(&EXT4_SB(inode->i_sb)->s_extent_cache_cnt); |
74cd15cd ZL |
321 | } |
322 | ||
654598be ZL |
323 | kmem_cache_free(ext4_es_cachep, es); |
324 | } | |
325 | ||
06b0c886 ZL |
326 | /* |
327 | * Check whether or not two extents can be merged | |
328 | * Condition: | |
329 | * - logical block number is contiguous | |
fdc0212e ZL |
330 | * - physical block number is contiguous |
331 | * - status is equal | |
06b0c886 ZL |
332 | */ |
333 | static int ext4_es_can_be_merged(struct extent_status *es1, | |
334 | struct extent_status *es2) | |
335 | { | |
bd384364 | 336 | if (ext4_es_status(es1) != ext4_es_status(es2)) |
06b0c886 ZL |
337 | return 0; |
338 | ||
bd384364 | 339 | if (((__u64) es1->es_len) + es2->es_len > 0xFFFFFFFFULL) |
fdc0212e ZL |
340 | return 0; |
341 | ||
bd384364 | 342 | if (((__u64) es1->es_lblk) + es1->es_len != es2->es_lblk) |
fdc0212e ZL |
343 | return 0; |
344 | ||
bd384364 ZL |
345 | if ((ext4_es_is_written(es1) || ext4_es_is_unwritten(es1)) && |
346 | (ext4_es_pblock(es1) + es1->es_len == ext4_es_pblock(es2))) | |
347 | return 1; | |
348 | ||
349 | if (ext4_es_is_hole(es1)) | |
350 | return 1; | |
351 | ||
352 | /* we need to check delayed extent is without unwritten status */ | |
353 | if (ext4_es_is_delayed(es1) && !ext4_es_is_unwritten(es1)) | |
354 | return 1; | |
355 | ||
356 | return 0; | |
06b0c886 ZL |
357 | } |
358 | ||
654598be | 359 | static struct extent_status * |
bdedbb7b | 360 | ext4_es_try_to_merge_left(struct inode *inode, struct extent_status *es) |
654598be | 361 | { |
bdedbb7b | 362 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be ZL |
363 | struct extent_status *es1; |
364 | struct rb_node *node; | |
365 | ||
366 | node = rb_prev(&es->rb_node); | |
367 | if (!node) | |
368 | return es; | |
369 | ||
370 | es1 = rb_entry(node, struct extent_status, rb_node); | |
06b0c886 ZL |
371 | if (ext4_es_can_be_merged(es1, es)) { |
372 | es1->es_len += es->es_len; | |
654598be | 373 | rb_erase(&es->rb_node, &tree->root); |
bdedbb7b | 374 | ext4_es_free_extent(inode, es); |
654598be ZL |
375 | es = es1; |
376 | } | |
377 | ||
378 | return es; | |
379 | } | |
380 | ||
381 | static struct extent_status * | |
bdedbb7b | 382 | ext4_es_try_to_merge_right(struct inode *inode, struct extent_status *es) |
654598be | 383 | { |
bdedbb7b | 384 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be ZL |
385 | struct extent_status *es1; |
386 | struct rb_node *node; | |
387 | ||
388 | node = rb_next(&es->rb_node); | |
389 | if (!node) | |
390 | return es; | |
391 | ||
392 | es1 = rb_entry(node, struct extent_status, rb_node); | |
06b0c886 ZL |
393 | if (ext4_es_can_be_merged(es, es1)) { |
394 | es->es_len += es1->es_len; | |
654598be | 395 | rb_erase(node, &tree->root); |
bdedbb7b | 396 | ext4_es_free_extent(inode, es1); |
654598be ZL |
397 | } |
398 | ||
399 | return es; | |
400 | } | |
401 | ||
921f266b DM |
402 | #ifdef ES_AGGRESSIVE_TEST |
403 | static void ext4_es_insert_extent_ext_check(struct inode *inode, | |
404 | struct extent_status *es) | |
405 | { | |
406 | struct ext4_ext_path *path = NULL; | |
407 | struct ext4_extent *ex; | |
408 | ext4_lblk_t ee_block; | |
409 | ext4_fsblk_t ee_start; | |
410 | unsigned short ee_len; | |
411 | int depth, ee_status, es_status; | |
412 | ||
413 | path = ext4_ext_find_extent(inode, es->es_lblk, NULL); | |
414 | if (IS_ERR(path)) | |
415 | return; | |
416 | ||
417 | depth = ext_depth(inode); | |
418 | ex = path[depth].p_ext; | |
419 | ||
420 | if (ex) { | |
421 | ||
422 | ee_block = le32_to_cpu(ex->ee_block); | |
423 | ee_start = ext4_ext_pblock(ex); | |
424 | ee_len = ext4_ext_get_actual_len(ex); | |
425 | ||
426 | ee_status = ext4_ext_is_uninitialized(ex) ? 1 : 0; | |
427 | es_status = ext4_es_is_unwritten(es) ? 1 : 0; | |
428 | ||
429 | /* | |
430 | * Make sure ex and es are not overlap when we try to insert | |
431 | * a delayed/hole extent. | |
432 | */ | |
433 | if (!ext4_es_is_written(es) && !ext4_es_is_unwritten(es)) { | |
434 | if (in_range(es->es_lblk, ee_block, ee_len)) { | |
435 | pr_warn("ES insert assertation failed for " | |
436 | "inode: %lu we can find an extent " | |
437 | "at block [%d/%d/%llu/%c], but we " | |
438 | "want to add an delayed/hole extent " | |
439 | "[%d/%d/%llu/%llx]\n", | |
440 | inode->i_ino, ee_block, ee_len, | |
441 | ee_start, ee_status ? 'u' : 'w', | |
442 | es->es_lblk, es->es_len, | |
443 | ext4_es_pblock(es), ext4_es_status(es)); | |
444 | } | |
445 | goto out; | |
446 | } | |
447 | ||
448 | /* | |
449 | * We don't check ee_block == es->es_lblk, etc. because es | |
450 | * might be a part of whole extent, vice versa. | |
451 | */ | |
452 | if (es->es_lblk < ee_block || | |
453 | ext4_es_pblock(es) != ee_start + es->es_lblk - ee_block) { | |
454 | pr_warn("ES insert assertation failed for inode: %lu " | |
455 | "ex_status [%d/%d/%llu/%c] != " | |
456 | "es_status [%d/%d/%llu/%c]\n", inode->i_ino, | |
457 | ee_block, ee_len, ee_start, | |
458 | ee_status ? 'u' : 'w', es->es_lblk, es->es_len, | |
459 | ext4_es_pblock(es), es_status ? 'u' : 'w'); | |
460 | goto out; | |
461 | } | |
462 | ||
463 | if (ee_status ^ es_status) { | |
464 | pr_warn("ES insert assertation failed for inode: %lu " | |
465 | "ex_status [%d/%d/%llu/%c] != " | |
466 | "es_status [%d/%d/%llu/%c]\n", inode->i_ino, | |
467 | ee_block, ee_len, ee_start, | |
468 | ee_status ? 'u' : 'w', es->es_lblk, es->es_len, | |
469 | ext4_es_pblock(es), es_status ? 'u' : 'w'); | |
470 | } | |
471 | } else { | |
472 | /* | |
473 | * We can't find an extent on disk. So we need to make sure | |
474 | * that we don't want to add an written/unwritten extent. | |
475 | */ | |
476 | if (!ext4_es_is_delayed(es) && !ext4_es_is_hole(es)) { | |
477 | pr_warn("ES insert assertation failed for inode: %lu " | |
478 | "can't find an extent at block %d but we want " | |
479 | "to add an written/unwritten extent " | |
480 | "[%d/%d/%llu/%llx]\n", inode->i_ino, | |
481 | es->es_lblk, es->es_lblk, es->es_len, | |
482 | ext4_es_pblock(es), ext4_es_status(es)); | |
483 | } | |
484 | } | |
485 | out: | |
486 | if (path) { | |
487 | ext4_ext_drop_refs(path); | |
488 | kfree(path); | |
489 | } | |
490 | } | |
491 | ||
492 | static void ext4_es_insert_extent_ind_check(struct inode *inode, | |
493 | struct extent_status *es) | |
494 | { | |
495 | struct ext4_map_blocks map; | |
496 | int retval; | |
497 | ||
498 | /* | |
499 | * Here we call ext4_ind_map_blocks to lookup a block mapping because | |
500 | * 'Indirect' structure is defined in indirect.c. So we couldn't | |
501 | * access direct/indirect tree from outside. It is too dirty to define | |
502 | * this function in indirect.c file. | |
503 | */ | |
504 | ||
505 | map.m_lblk = es->es_lblk; | |
506 | map.m_len = es->es_len; | |
507 | ||
508 | retval = ext4_ind_map_blocks(NULL, inode, &map, 0); | |
509 | if (retval > 0) { | |
510 | if (ext4_es_is_delayed(es) || ext4_es_is_hole(es)) { | |
511 | /* | |
512 | * We want to add a delayed/hole extent but this | |
513 | * block has been allocated. | |
514 | */ | |
515 | pr_warn("ES insert assertation failed for inode: %lu " | |
516 | "We can find blocks but we want to add a " | |
517 | "delayed/hole extent [%d/%d/%llu/%llx]\n", | |
518 | inode->i_ino, es->es_lblk, es->es_len, | |
519 | ext4_es_pblock(es), ext4_es_status(es)); | |
520 | return; | |
521 | } else if (ext4_es_is_written(es)) { | |
522 | if (retval != es->es_len) { | |
523 | pr_warn("ES insert assertation failed for " | |
524 | "inode: %lu retval %d != es_len %d\n", | |
525 | inode->i_ino, retval, es->es_len); | |
526 | return; | |
527 | } | |
528 | if (map.m_pblk != ext4_es_pblock(es)) { | |
529 | pr_warn("ES insert assertation failed for " | |
530 | "inode: %lu m_pblk %llu != " | |
531 | "es_pblk %llu\n", | |
532 | inode->i_ino, map.m_pblk, | |
533 | ext4_es_pblock(es)); | |
534 | return; | |
535 | } | |
536 | } else { | |
537 | /* | |
538 | * We don't need to check unwritten extent because | |
539 | * indirect-based file doesn't have it. | |
540 | */ | |
541 | BUG_ON(1); | |
542 | } | |
543 | } else if (retval == 0) { | |
544 | if (ext4_es_is_written(es)) { | |
545 | pr_warn("ES insert assertation failed for inode: %lu " | |
546 | "We can't find the block but we want to add " | |
547 | "an written extent [%d/%d/%llu/%llx]\n", | |
548 | inode->i_ino, es->es_lblk, es->es_len, | |
549 | ext4_es_pblock(es), ext4_es_status(es)); | |
550 | return; | |
551 | } | |
552 | } | |
553 | } | |
554 | ||
555 | static inline void ext4_es_insert_extent_check(struct inode *inode, | |
556 | struct extent_status *es) | |
557 | { | |
558 | /* | |
559 | * We don't need to worry about the race condition because | |
560 | * caller takes i_data_sem locking. | |
561 | */ | |
562 | BUG_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem)); | |
563 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
564 | ext4_es_insert_extent_ext_check(inode, es); | |
565 | else | |
566 | ext4_es_insert_extent_ind_check(inode, es); | |
567 | } | |
568 | #else | |
569 | static inline void ext4_es_insert_extent_check(struct inode *inode, | |
570 | struct extent_status *es) | |
571 | { | |
572 | } | |
573 | #endif | |
574 | ||
bdedbb7b | 575 | static int __es_insert_extent(struct inode *inode, struct extent_status *newes) |
654598be | 576 | { |
bdedbb7b | 577 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be ZL |
578 | struct rb_node **p = &tree->root.rb_node; |
579 | struct rb_node *parent = NULL; | |
580 | struct extent_status *es; | |
654598be ZL |
581 | |
582 | while (*p) { | |
583 | parent = *p; | |
584 | es = rb_entry(parent, struct extent_status, rb_node); | |
585 | ||
06b0c886 ZL |
586 | if (newes->es_lblk < es->es_lblk) { |
587 | if (ext4_es_can_be_merged(newes, es)) { | |
588 | /* | |
589 | * Here we can modify es_lblk directly | |
590 | * because it isn't overlapped. | |
591 | */ | |
592 | es->es_lblk = newes->es_lblk; | |
593 | es->es_len += newes->es_len; | |
fdc0212e ZL |
594 | if (ext4_es_is_written(es) || |
595 | ext4_es_is_unwritten(es)) | |
596 | ext4_es_store_pblock(es, | |
597 | newes->es_pblk); | |
bdedbb7b | 598 | es = ext4_es_try_to_merge_left(inode, es); |
654598be ZL |
599 | goto out; |
600 | } | |
601 | p = &(*p)->rb_left; | |
06b0c886 ZL |
602 | } else if (newes->es_lblk > ext4_es_end(es)) { |
603 | if (ext4_es_can_be_merged(es, newes)) { | |
604 | es->es_len += newes->es_len; | |
bdedbb7b | 605 | es = ext4_es_try_to_merge_right(inode, es); |
654598be ZL |
606 | goto out; |
607 | } | |
608 | p = &(*p)->rb_right; | |
609 | } else { | |
06b0c886 ZL |
610 | BUG_ON(1); |
611 | return -EINVAL; | |
654598be ZL |
612 | } |
613 | } | |
614 | ||
bdedbb7b | 615 | es = ext4_es_alloc_extent(inode, newes->es_lblk, newes->es_len, |
fdc0212e | 616 | newes->es_pblk); |
654598be ZL |
617 | if (!es) |
618 | return -ENOMEM; | |
619 | rb_link_node(&es->rb_node, parent, p); | |
620 | rb_insert_color(&es->rb_node, &tree->root); | |
621 | ||
622 | out: | |
623 | tree->cache_es = es; | |
624 | return 0; | |
625 | } | |
626 | ||
627 | /* | |
06b0c886 | 628 | * ext4_es_insert_extent() adds a space to a extent status tree. |
654598be ZL |
629 | * |
630 | * ext4_es_insert_extent is called by ext4_da_write_begin and | |
631 | * ext4_es_remove_extent. | |
632 | * | |
633 | * Return 0 on success, error code on failure. | |
634 | */ | |
06b0c886 | 635 | int ext4_es_insert_extent(struct inode *inode, ext4_lblk_t lblk, |
fdc0212e ZL |
636 | ext4_lblk_t len, ext4_fsblk_t pblk, |
637 | unsigned long long status) | |
654598be | 638 | { |
06b0c886 ZL |
639 | struct extent_status newes; |
640 | ext4_lblk_t end = lblk + len - 1; | |
654598be ZL |
641 | int err = 0; |
642 | ||
fdc0212e ZL |
643 | es_debug("add [%u/%u) %llu %llx to extent status tree of inode %lu\n", |
644 | lblk, len, pblk, status, inode->i_ino); | |
06b0c886 | 645 | |
d4381472 EG |
646 | if (!len) |
647 | return 0; | |
648 | ||
06b0c886 ZL |
649 | BUG_ON(end < lblk); |
650 | ||
651 | newes.es_lblk = lblk; | |
652 | newes.es_len = len; | |
fdc0212e ZL |
653 | ext4_es_store_pblock(&newes, pblk); |
654 | ext4_es_store_status(&newes, status); | |
655 | trace_ext4_es_insert_extent(inode, &newes); | |
654598be | 656 | |
921f266b DM |
657 | ext4_es_insert_extent_check(inode, &newes); |
658 | ||
654598be | 659 | write_lock(&EXT4_I(inode)->i_es_lock); |
bdedbb7b | 660 | err = __es_remove_extent(inode, lblk, end); |
06b0c886 ZL |
661 | if (err != 0) |
662 | goto error; | |
bdedbb7b | 663 | err = __es_insert_extent(inode, &newes); |
06b0c886 ZL |
664 | |
665 | error: | |
654598be ZL |
666 | write_unlock(&EXT4_I(inode)->i_es_lock); |
667 | ||
74cd15cd | 668 | ext4_es_lru_add(inode); |
654598be ZL |
669 | ext4_es_print_tree(inode); |
670 | ||
671 | return err; | |
672 | } | |
673 | ||
d100eef2 ZL |
674 | /* |
675 | * ext4_es_lookup_extent() looks up an extent in extent status tree. | |
676 | * | |
677 | * ext4_es_lookup_extent is called by ext4_map_blocks/ext4_da_map_blocks. | |
678 | * | |
679 | * Return: 1 on found, 0 on not | |
680 | */ | |
681 | int ext4_es_lookup_extent(struct inode *inode, ext4_lblk_t lblk, | |
682 | struct extent_status *es) | |
683 | { | |
684 | struct ext4_es_tree *tree; | |
685 | struct extent_status *es1 = NULL; | |
686 | struct rb_node *node; | |
687 | int found = 0; | |
688 | ||
689 | trace_ext4_es_lookup_extent_enter(inode, lblk); | |
690 | es_debug("lookup extent in block %u\n", lblk); | |
691 | ||
692 | tree = &EXT4_I(inode)->i_es_tree; | |
693 | read_lock(&EXT4_I(inode)->i_es_lock); | |
694 | ||
695 | /* find extent in cache firstly */ | |
696 | es->es_lblk = es->es_len = es->es_pblk = 0; | |
697 | if (tree->cache_es) { | |
698 | es1 = tree->cache_es; | |
699 | if (in_range(lblk, es1->es_lblk, es1->es_len)) { | |
700 | es_debug("%u cached by [%u/%u)\n", | |
701 | lblk, es1->es_lblk, es1->es_len); | |
702 | found = 1; | |
703 | goto out; | |
704 | } | |
705 | } | |
706 | ||
707 | node = tree->root.rb_node; | |
708 | while (node) { | |
709 | es1 = rb_entry(node, struct extent_status, rb_node); | |
710 | if (lblk < es1->es_lblk) | |
711 | node = node->rb_left; | |
712 | else if (lblk > ext4_es_end(es1)) | |
713 | node = node->rb_right; | |
714 | else { | |
715 | found = 1; | |
716 | break; | |
717 | } | |
718 | } | |
719 | ||
720 | out: | |
721 | if (found) { | |
722 | BUG_ON(!es1); | |
723 | es->es_lblk = es1->es_lblk; | |
724 | es->es_len = es1->es_len; | |
725 | es->es_pblk = es1->es_pblk; | |
726 | } | |
727 | ||
728 | read_unlock(&EXT4_I(inode)->i_es_lock); | |
729 | ||
74cd15cd | 730 | ext4_es_lru_add(inode); |
d100eef2 ZL |
731 | trace_ext4_es_lookup_extent_exit(inode, es, found); |
732 | return found; | |
733 | } | |
734 | ||
bdedbb7b ZL |
735 | static int __es_remove_extent(struct inode *inode, ext4_lblk_t lblk, |
736 | ext4_lblk_t end) | |
654598be | 737 | { |
bdedbb7b | 738 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be | 739 | struct rb_node *node; |
654598be ZL |
740 | struct extent_status *es; |
741 | struct extent_status orig_es; | |
06b0c886 | 742 | ext4_lblk_t len1, len2; |
fdc0212e | 743 | ext4_fsblk_t block; |
654598be ZL |
744 | int err = 0; |
745 | ||
06b0c886 | 746 | es = __es_tree_search(&tree->root, lblk); |
654598be ZL |
747 | if (!es) |
748 | goto out; | |
06b0c886 | 749 | if (es->es_lblk > end) |
654598be ZL |
750 | goto out; |
751 | ||
752 | /* Simply invalidate cache_es. */ | |
753 | tree->cache_es = NULL; | |
754 | ||
06b0c886 ZL |
755 | orig_es.es_lblk = es->es_lblk; |
756 | orig_es.es_len = es->es_len; | |
fdc0212e ZL |
757 | orig_es.es_pblk = es->es_pblk; |
758 | ||
06b0c886 ZL |
759 | len1 = lblk > es->es_lblk ? lblk - es->es_lblk : 0; |
760 | len2 = ext4_es_end(es) > end ? ext4_es_end(es) - end : 0; | |
654598be | 761 | if (len1 > 0) |
06b0c886 | 762 | es->es_len = len1; |
654598be ZL |
763 | if (len2 > 0) { |
764 | if (len1 > 0) { | |
06b0c886 ZL |
765 | struct extent_status newes; |
766 | ||
767 | newes.es_lblk = end + 1; | |
768 | newes.es_len = len2; | |
fdc0212e ZL |
769 | if (ext4_es_is_written(&orig_es) || |
770 | ext4_es_is_unwritten(&orig_es)) { | |
771 | block = ext4_es_pblock(&orig_es) + | |
772 | orig_es.es_len - len2; | |
773 | ext4_es_store_pblock(&newes, block); | |
774 | } | |
775 | ext4_es_store_status(&newes, ext4_es_status(&orig_es)); | |
bdedbb7b | 776 | err = __es_insert_extent(inode, &newes); |
654598be | 777 | if (err) { |
06b0c886 ZL |
778 | es->es_lblk = orig_es.es_lblk; |
779 | es->es_len = orig_es.es_len; | |
654598be ZL |
780 | goto out; |
781 | } | |
782 | } else { | |
06b0c886 ZL |
783 | es->es_lblk = end + 1; |
784 | es->es_len = len2; | |
fdc0212e ZL |
785 | if (ext4_es_is_written(es) || |
786 | ext4_es_is_unwritten(es)) { | |
787 | block = orig_es.es_pblk + orig_es.es_len - len2; | |
788 | ext4_es_store_pblock(es, block); | |
789 | } | |
654598be ZL |
790 | } |
791 | goto out; | |
792 | } | |
793 | ||
794 | if (len1 > 0) { | |
795 | node = rb_next(&es->rb_node); | |
796 | if (node) | |
797 | es = rb_entry(node, struct extent_status, rb_node); | |
798 | else | |
799 | es = NULL; | |
800 | } | |
801 | ||
06b0c886 | 802 | while (es && ext4_es_end(es) <= end) { |
654598be ZL |
803 | node = rb_next(&es->rb_node); |
804 | rb_erase(&es->rb_node, &tree->root); | |
bdedbb7b | 805 | ext4_es_free_extent(inode, es); |
654598be ZL |
806 | if (!node) { |
807 | es = NULL; | |
808 | break; | |
809 | } | |
810 | es = rb_entry(node, struct extent_status, rb_node); | |
811 | } | |
812 | ||
06b0c886 | 813 | if (es && es->es_lblk < end + 1) { |
fdc0212e ZL |
814 | ext4_lblk_t orig_len = es->es_len; |
815 | ||
06b0c886 ZL |
816 | len1 = ext4_es_end(es) - end; |
817 | es->es_lblk = end + 1; | |
818 | es->es_len = len1; | |
fdc0212e ZL |
819 | if (ext4_es_is_written(es) || ext4_es_is_unwritten(es)) { |
820 | block = es->es_pblk + orig_len - len1; | |
821 | ext4_es_store_pblock(es, block); | |
822 | } | |
654598be ZL |
823 | } |
824 | ||
825 | out: | |
06b0c886 ZL |
826 | return err; |
827 | } | |
828 | ||
829 | /* | |
830 | * ext4_es_remove_extent() removes a space from a extent status tree. | |
831 | * | |
832 | * Return 0 on success, error code on failure. | |
833 | */ | |
834 | int ext4_es_remove_extent(struct inode *inode, ext4_lblk_t lblk, | |
835 | ext4_lblk_t len) | |
836 | { | |
06b0c886 ZL |
837 | ext4_lblk_t end; |
838 | int err = 0; | |
839 | ||
840 | trace_ext4_es_remove_extent(inode, lblk, len); | |
841 | es_debug("remove [%u/%u) from extent status tree of inode %lu\n", | |
842 | lblk, len, inode->i_ino); | |
843 | ||
d4381472 EG |
844 | if (!len) |
845 | return err; | |
846 | ||
06b0c886 ZL |
847 | end = lblk + len - 1; |
848 | BUG_ON(end < lblk); | |
849 | ||
06b0c886 | 850 | write_lock(&EXT4_I(inode)->i_es_lock); |
bdedbb7b | 851 | err = __es_remove_extent(inode, lblk, end); |
654598be ZL |
852 | write_unlock(&EXT4_I(inode)->i_es_lock); |
853 | ext4_es_print_tree(inode); | |
854 | return err; | |
855 | } | |
74cd15cd | 856 | |
adb23551 ZL |
857 | int ext4_es_zeroout(struct inode *inode, struct ext4_extent *ex) |
858 | { | |
859 | ext4_lblk_t ee_block; | |
860 | ext4_fsblk_t ee_pblock; | |
861 | unsigned int ee_len; | |
862 | ||
863 | ee_block = le32_to_cpu(ex->ee_block); | |
864 | ee_len = ext4_ext_get_actual_len(ex); | |
865 | ee_pblock = ext4_ext_pblock(ex); | |
866 | ||
867 | if (ee_len == 0) | |
868 | return 0; | |
869 | ||
870 | return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock, | |
871 | EXTENT_STATUS_WRITTEN); | |
872 | } | |
873 | ||
74cd15cd ZL |
874 | static int ext4_es_shrink(struct shrinker *shrink, struct shrink_control *sc) |
875 | { | |
876 | struct ext4_sb_info *sbi = container_of(shrink, | |
877 | struct ext4_sb_info, s_es_shrinker); | |
878 | struct ext4_inode_info *ei; | |
879 | struct list_head *cur, *tmp, scanned; | |
880 | int nr_to_scan = sc->nr_to_scan; | |
881 | int ret, nr_shrunk = 0; | |
882 | ||
1ac6466f | 883 | ret = percpu_counter_read_positive(&sbi->s_extent_cache_cnt); |
24630774 | 884 | trace_ext4_es_shrink_enter(sbi->s_sb, nr_to_scan, ret); |
74cd15cd ZL |
885 | |
886 | if (!nr_to_scan) | |
24630774 | 887 | return ret; |
74cd15cd ZL |
888 | |
889 | INIT_LIST_HEAD(&scanned); | |
890 | ||
891 | spin_lock(&sbi->s_es_lru_lock); | |
892 | list_for_each_safe(cur, tmp, &sbi->s_es_lru) { | |
893 | list_move_tail(cur, &scanned); | |
894 | ||
895 | ei = list_entry(cur, struct ext4_inode_info, i_es_lru); | |
896 | ||
897 | read_lock(&ei->i_es_lock); | |
898 | if (ei->i_es_lru_nr == 0) { | |
899 | read_unlock(&ei->i_es_lock); | |
900 | continue; | |
901 | } | |
902 | read_unlock(&ei->i_es_lock); | |
903 | ||
904 | write_lock(&ei->i_es_lock); | |
905 | ret = __es_try_to_reclaim_extents(ei, nr_to_scan); | |
906 | write_unlock(&ei->i_es_lock); | |
907 | ||
908 | nr_shrunk += ret; | |
909 | nr_to_scan -= ret; | |
910 | if (nr_to_scan == 0) | |
911 | break; | |
912 | } | |
913 | list_splice_tail(&scanned, &sbi->s_es_lru); | |
914 | spin_unlock(&sbi->s_es_lru_lock); | |
74cd15cd | 915 | |
1ac6466f | 916 | ret = percpu_counter_read_positive(&sbi->s_extent_cache_cnt); |
24630774 TT |
917 | trace_ext4_es_shrink_exit(sbi->s_sb, nr_shrunk, ret); |
918 | return ret; | |
74cd15cd ZL |
919 | } |
920 | ||
921 | void ext4_es_register_shrinker(struct super_block *sb) | |
922 | { | |
923 | struct ext4_sb_info *sbi; | |
924 | ||
925 | sbi = EXT4_SB(sb); | |
926 | INIT_LIST_HEAD(&sbi->s_es_lru); | |
927 | spin_lock_init(&sbi->s_es_lru_lock); | |
928 | sbi->s_es_shrinker.shrink = ext4_es_shrink; | |
929 | sbi->s_es_shrinker.seeks = DEFAULT_SEEKS; | |
930 | register_shrinker(&sbi->s_es_shrinker); | |
931 | } | |
932 | ||
933 | void ext4_es_unregister_shrinker(struct super_block *sb) | |
934 | { | |
935 | unregister_shrinker(&EXT4_SB(sb)->s_es_shrinker); | |
936 | } | |
937 | ||
938 | void ext4_es_lru_add(struct inode *inode) | |
939 | { | |
940 | struct ext4_inode_info *ei = EXT4_I(inode); | |
941 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
942 | ||
943 | spin_lock(&sbi->s_es_lru_lock); | |
944 | if (list_empty(&ei->i_es_lru)) | |
945 | list_add_tail(&ei->i_es_lru, &sbi->s_es_lru); | |
946 | else | |
947 | list_move_tail(&ei->i_es_lru, &sbi->s_es_lru); | |
948 | spin_unlock(&sbi->s_es_lru_lock); | |
949 | } | |
950 | ||
951 | void ext4_es_lru_del(struct inode *inode) | |
952 | { | |
953 | struct ext4_inode_info *ei = EXT4_I(inode); | |
954 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
955 | ||
956 | spin_lock(&sbi->s_es_lru_lock); | |
957 | if (!list_empty(&ei->i_es_lru)) | |
958 | list_del_init(&ei->i_es_lru); | |
959 | spin_unlock(&sbi->s_es_lru_lock); | |
960 | } | |
961 | ||
74cd15cd ZL |
962 | static int __es_try_to_reclaim_extents(struct ext4_inode_info *ei, |
963 | int nr_to_scan) | |
964 | { | |
965 | struct inode *inode = &ei->vfs_inode; | |
966 | struct ext4_es_tree *tree = &ei->i_es_tree; | |
967 | struct rb_node *node; | |
968 | struct extent_status *es; | |
969 | int nr_shrunk = 0; | |
970 | ||
971 | if (ei->i_es_lru_nr == 0) | |
972 | return 0; | |
973 | ||
974 | node = rb_first(&tree->root); | |
975 | while (node != NULL) { | |
976 | es = rb_entry(node, struct extent_status, rb_node); | |
977 | node = rb_next(&es->rb_node); | |
978 | /* | |
979 | * We can't reclaim delayed extent from status tree because | |
980 | * fiemap, bigallic, and seek_data/hole need to use it. | |
981 | */ | |
982 | if (!ext4_es_is_delayed(es)) { | |
983 | rb_erase(&es->rb_node, &tree->root); | |
984 | ext4_es_free_extent(inode, es); | |
985 | nr_shrunk++; | |
986 | if (--nr_to_scan == 0) | |
987 | break; | |
988 | } | |
989 | } | |
990 | tree->cache_es = NULL; | |
991 | return nr_shrunk; | |
992 | } |