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 | /* | |
e30b5dca YZ |
235 | * ext4_es_find_delayed_extent_range: find the 1st delayed extent covering |
236 | * @es->lblk 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 |
e30b5dca | 240 | * @end: the offset where we stop to search |
654598be | 241 | * @es: delayed extent that we found |
654598be | 242 | */ |
e30b5dca YZ |
243 | void ext4_es_find_delayed_extent_range(struct inode *inode, |
244 | ext4_lblk_t lblk, ext4_lblk_t end, | |
be401363 | 245 | struct extent_status *es) |
654598be ZL |
246 | { |
247 | struct ext4_es_tree *tree = NULL; | |
248 | struct extent_status *es1 = NULL; | |
249 | struct rb_node *node; | |
654598be | 250 | |
be401363 | 251 | BUG_ON(es == NULL); |
e30b5dca YZ |
252 | BUG_ON(end < lblk); |
253 | trace_ext4_es_find_delayed_extent_range_enter(inode, lblk); | |
992e9fdd | 254 | |
654598be ZL |
255 | read_lock(&EXT4_I(inode)->i_es_lock); |
256 | tree = &EXT4_I(inode)->i_es_tree; | |
257 | ||
fdc0212e | 258 | /* find extent in cache firstly */ |
be401363 | 259 | es->es_lblk = es->es_len = es->es_pblk = 0; |
654598be ZL |
260 | if (tree->cache_es) { |
261 | es1 = tree->cache_es; | |
be401363 | 262 | if (in_range(lblk, es1->es_lblk, es1->es_len)) { |
fdc0212e | 263 | es_debug("%u cached by [%u/%u) %llu %llx\n", |
be401363 | 264 | lblk, es1->es_lblk, es1->es_len, |
fdc0212e | 265 | ext4_es_pblock(es1), ext4_es_status(es1)); |
654598be ZL |
266 | goto out; |
267 | } | |
268 | } | |
269 | ||
be401363 | 270 | es1 = __es_tree_search(&tree->root, lblk); |
654598be ZL |
271 | |
272 | out: | |
be401363 ZL |
273 | if (es1 && !ext4_es_is_delayed(es1)) { |
274 | while ((node = rb_next(&es1->rb_node)) != NULL) { | |
275 | es1 = rb_entry(node, struct extent_status, rb_node); | |
e30b5dca YZ |
276 | if (es1->es_lblk > end) { |
277 | es1 = NULL; | |
278 | break; | |
279 | } | |
be401363 ZL |
280 | if (ext4_es_is_delayed(es1)) |
281 | break; | |
282 | } | |
283 | } | |
284 | ||
285 | if (es1 && ext4_es_is_delayed(es1)) { | |
654598be | 286 | tree->cache_es = es1; |
06b0c886 ZL |
287 | es->es_lblk = es1->es_lblk; |
288 | es->es_len = es1->es_len; | |
fdc0212e | 289 | es->es_pblk = es1->es_pblk; |
654598be ZL |
290 | } |
291 | ||
292 | read_unlock(&EXT4_I(inode)->i_es_lock); | |
992e9fdd | 293 | |
74cd15cd | 294 | ext4_es_lru_add(inode); |
e30b5dca | 295 | trace_ext4_es_find_delayed_extent_range_exit(inode, es); |
654598be ZL |
296 | } |
297 | ||
298 | static struct extent_status * | |
bdedbb7b ZL |
299 | ext4_es_alloc_extent(struct inode *inode, ext4_lblk_t lblk, ext4_lblk_t len, |
300 | ext4_fsblk_t pblk) | |
654598be ZL |
301 | { |
302 | struct extent_status *es; | |
303 | es = kmem_cache_alloc(ext4_es_cachep, GFP_ATOMIC); | |
304 | if (es == NULL) | |
305 | return NULL; | |
06b0c886 ZL |
306 | es->es_lblk = lblk; |
307 | es->es_len = len; | |
fdc0212e | 308 | es->es_pblk = pblk; |
74cd15cd ZL |
309 | |
310 | /* | |
311 | * We don't count delayed extent because we never try to reclaim them | |
312 | */ | |
24630774 | 313 | if (!ext4_es_is_delayed(es)) { |
74cd15cd | 314 | EXT4_I(inode)->i_es_lru_nr++; |
1ac6466f | 315 | percpu_counter_inc(&EXT4_SB(inode->i_sb)->s_extent_cache_cnt); |
24630774 | 316 | } |
74cd15cd | 317 | |
654598be ZL |
318 | return es; |
319 | } | |
320 | ||
bdedbb7b | 321 | static void ext4_es_free_extent(struct inode *inode, struct extent_status *es) |
654598be | 322 | { |
74cd15cd ZL |
323 | /* Decrease the lru counter when this es is not delayed */ |
324 | if (!ext4_es_is_delayed(es)) { | |
325 | BUG_ON(EXT4_I(inode)->i_es_lru_nr == 0); | |
326 | EXT4_I(inode)->i_es_lru_nr--; | |
1ac6466f | 327 | percpu_counter_dec(&EXT4_SB(inode->i_sb)->s_extent_cache_cnt); |
74cd15cd ZL |
328 | } |
329 | ||
654598be ZL |
330 | kmem_cache_free(ext4_es_cachep, es); |
331 | } | |
332 | ||
06b0c886 ZL |
333 | /* |
334 | * Check whether or not two extents can be merged | |
335 | * Condition: | |
336 | * - logical block number is contiguous | |
fdc0212e ZL |
337 | * - physical block number is contiguous |
338 | * - status is equal | |
06b0c886 ZL |
339 | */ |
340 | static int ext4_es_can_be_merged(struct extent_status *es1, | |
341 | struct extent_status *es2) | |
342 | { | |
bd384364 | 343 | if (ext4_es_status(es1) != ext4_es_status(es2)) |
06b0c886 ZL |
344 | return 0; |
345 | ||
bd384364 | 346 | if (((__u64) es1->es_len) + es2->es_len > 0xFFFFFFFFULL) |
fdc0212e ZL |
347 | return 0; |
348 | ||
bd384364 | 349 | if (((__u64) es1->es_lblk) + es1->es_len != es2->es_lblk) |
fdc0212e ZL |
350 | return 0; |
351 | ||
bd384364 ZL |
352 | if ((ext4_es_is_written(es1) || ext4_es_is_unwritten(es1)) && |
353 | (ext4_es_pblock(es1) + es1->es_len == ext4_es_pblock(es2))) | |
354 | return 1; | |
355 | ||
356 | if (ext4_es_is_hole(es1)) | |
357 | return 1; | |
358 | ||
359 | /* we need to check delayed extent is without unwritten status */ | |
360 | if (ext4_es_is_delayed(es1) && !ext4_es_is_unwritten(es1)) | |
361 | return 1; | |
362 | ||
363 | return 0; | |
06b0c886 ZL |
364 | } |
365 | ||
654598be | 366 | static struct extent_status * |
bdedbb7b | 367 | ext4_es_try_to_merge_left(struct inode *inode, struct extent_status *es) |
654598be | 368 | { |
bdedbb7b | 369 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be ZL |
370 | struct extent_status *es1; |
371 | struct rb_node *node; | |
372 | ||
373 | node = rb_prev(&es->rb_node); | |
374 | if (!node) | |
375 | return es; | |
376 | ||
377 | es1 = rb_entry(node, struct extent_status, rb_node); | |
06b0c886 ZL |
378 | if (ext4_es_can_be_merged(es1, es)) { |
379 | es1->es_len += es->es_len; | |
654598be | 380 | rb_erase(&es->rb_node, &tree->root); |
bdedbb7b | 381 | ext4_es_free_extent(inode, es); |
654598be ZL |
382 | es = es1; |
383 | } | |
384 | ||
385 | return es; | |
386 | } | |
387 | ||
388 | static struct extent_status * | |
bdedbb7b | 389 | ext4_es_try_to_merge_right(struct inode *inode, struct extent_status *es) |
654598be | 390 | { |
bdedbb7b | 391 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be ZL |
392 | struct extent_status *es1; |
393 | struct rb_node *node; | |
394 | ||
395 | node = rb_next(&es->rb_node); | |
396 | if (!node) | |
397 | return es; | |
398 | ||
399 | es1 = rb_entry(node, struct extent_status, rb_node); | |
06b0c886 ZL |
400 | if (ext4_es_can_be_merged(es, es1)) { |
401 | es->es_len += es1->es_len; | |
654598be | 402 | rb_erase(node, &tree->root); |
bdedbb7b | 403 | ext4_es_free_extent(inode, es1); |
654598be ZL |
404 | } |
405 | ||
406 | return es; | |
407 | } | |
408 | ||
921f266b DM |
409 | #ifdef ES_AGGRESSIVE_TEST |
410 | static void ext4_es_insert_extent_ext_check(struct inode *inode, | |
411 | struct extent_status *es) | |
412 | { | |
413 | struct ext4_ext_path *path = NULL; | |
414 | struct ext4_extent *ex; | |
415 | ext4_lblk_t ee_block; | |
416 | ext4_fsblk_t ee_start; | |
417 | unsigned short ee_len; | |
418 | int depth, ee_status, es_status; | |
419 | ||
420 | path = ext4_ext_find_extent(inode, es->es_lblk, NULL); | |
421 | if (IS_ERR(path)) | |
422 | return; | |
423 | ||
424 | depth = ext_depth(inode); | |
425 | ex = path[depth].p_ext; | |
426 | ||
427 | if (ex) { | |
428 | ||
429 | ee_block = le32_to_cpu(ex->ee_block); | |
430 | ee_start = ext4_ext_pblock(ex); | |
431 | ee_len = ext4_ext_get_actual_len(ex); | |
432 | ||
433 | ee_status = ext4_ext_is_uninitialized(ex) ? 1 : 0; | |
434 | es_status = ext4_es_is_unwritten(es) ? 1 : 0; | |
435 | ||
436 | /* | |
437 | * Make sure ex and es are not overlap when we try to insert | |
438 | * a delayed/hole extent. | |
439 | */ | |
440 | if (!ext4_es_is_written(es) && !ext4_es_is_unwritten(es)) { | |
441 | if (in_range(es->es_lblk, ee_block, ee_len)) { | |
442 | pr_warn("ES insert assertation failed for " | |
443 | "inode: %lu we can find an extent " | |
444 | "at block [%d/%d/%llu/%c], but we " | |
445 | "want to add an delayed/hole extent " | |
446 | "[%d/%d/%llu/%llx]\n", | |
447 | inode->i_ino, ee_block, ee_len, | |
448 | ee_start, ee_status ? 'u' : 'w', | |
449 | es->es_lblk, es->es_len, | |
450 | ext4_es_pblock(es), ext4_es_status(es)); | |
451 | } | |
452 | goto out; | |
453 | } | |
454 | ||
455 | /* | |
456 | * We don't check ee_block == es->es_lblk, etc. because es | |
457 | * might be a part of whole extent, vice versa. | |
458 | */ | |
459 | if (es->es_lblk < ee_block || | |
460 | ext4_es_pblock(es) != ee_start + es->es_lblk - ee_block) { | |
461 | pr_warn("ES insert assertation failed for inode: %lu " | |
462 | "ex_status [%d/%d/%llu/%c] != " | |
463 | "es_status [%d/%d/%llu/%c]\n", inode->i_ino, | |
464 | ee_block, ee_len, ee_start, | |
465 | ee_status ? 'u' : 'w', es->es_lblk, es->es_len, | |
466 | ext4_es_pblock(es), es_status ? 'u' : 'w'); | |
467 | goto out; | |
468 | } | |
469 | ||
470 | if (ee_status ^ es_status) { | |
471 | pr_warn("ES insert assertation failed for inode: %lu " | |
472 | "ex_status [%d/%d/%llu/%c] != " | |
473 | "es_status [%d/%d/%llu/%c]\n", inode->i_ino, | |
474 | ee_block, ee_len, ee_start, | |
475 | ee_status ? 'u' : 'w', es->es_lblk, es->es_len, | |
476 | ext4_es_pblock(es), es_status ? 'u' : 'w'); | |
477 | } | |
478 | } else { | |
479 | /* | |
480 | * We can't find an extent on disk. So we need to make sure | |
481 | * that we don't want to add an written/unwritten extent. | |
482 | */ | |
483 | if (!ext4_es_is_delayed(es) && !ext4_es_is_hole(es)) { | |
484 | pr_warn("ES insert assertation failed for inode: %lu " | |
485 | "can't find an extent at block %d but we want " | |
486 | "to add an written/unwritten extent " | |
487 | "[%d/%d/%llu/%llx]\n", inode->i_ino, | |
488 | es->es_lblk, es->es_lblk, es->es_len, | |
489 | ext4_es_pblock(es), ext4_es_status(es)); | |
490 | } | |
491 | } | |
492 | out: | |
493 | if (path) { | |
494 | ext4_ext_drop_refs(path); | |
495 | kfree(path); | |
496 | } | |
497 | } | |
498 | ||
499 | static void ext4_es_insert_extent_ind_check(struct inode *inode, | |
500 | struct extent_status *es) | |
501 | { | |
502 | struct ext4_map_blocks map; | |
503 | int retval; | |
504 | ||
505 | /* | |
506 | * Here we call ext4_ind_map_blocks to lookup a block mapping because | |
507 | * 'Indirect' structure is defined in indirect.c. So we couldn't | |
508 | * access direct/indirect tree from outside. It is too dirty to define | |
509 | * this function in indirect.c file. | |
510 | */ | |
511 | ||
512 | map.m_lblk = es->es_lblk; | |
513 | map.m_len = es->es_len; | |
514 | ||
515 | retval = ext4_ind_map_blocks(NULL, inode, &map, 0); | |
516 | if (retval > 0) { | |
517 | if (ext4_es_is_delayed(es) || ext4_es_is_hole(es)) { | |
518 | /* | |
519 | * We want to add a delayed/hole extent but this | |
520 | * block has been allocated. | |
521 | */ | |
522 | pr_warn("ES insert assertation failed for inode: %lu " | |
523 | "We can find blocks but we want to add a " | |
524 | "delayed/hole extent [%d/%d/%llu/%llx]\n", | |
525 | inode->i_ino, es->es_lblk, es->es_len, | |
526 | ext4_es_pblock(es), ext4_es_status(es)); | |
527 | return; | |
528 | } else if (ext4_es_is_written(es)) { | |
529 | if (retval != es->es_len) { | |
530 | pr_warn("ES insert assertation failed for " | |
531 | "inode: %lu retval %d != es_len %d\n", | |
532 | inode->i_ino, retval, es->es_len); | |
533 | return; | |
534 | } | |
535 | if (map.m_pblk != ext4_es_pblock(es)) { | |
536 | pr_warn("ES insert assertation failed for " | |
537 | "inode: %lu m_pblk %llu != " | |
538 | "es_pblk %llu\n", | |
539 | inode->i_ino, map.m_pblk, | |
540 | ext4_es_pblock(es)); | |
541 | return; | |
542 | } | |
543 | } else { | |
544 | /* | |
545 | * We don't need to check unwritten extent because | |
546 | * indirect-based file doesn't have it. | |
547 | */ | |
548 | BUG_ON(1); | |
549 | } | |
550 | } else if (retval == 0) { | |
551 | if (ext4_es_is_written(es)) { | |
552 | pr_warn("ES insert assertation failed for inode: %lu " | |
553 | "We can't find the block but we want to add " | |
554 | "an written extent [%d/%d/%llu/%llx]\n", | |
555 | inode->i_ino, es->es_lblk, es->es_len, | |
556 | ext4_es_pblock(es), ext4_es_status(es)); | |
557 | return; | |
558 | } | |
559 | } | |
560 | } | |
561 | ||
562 | static inline void ext4_es_insert_extent_check(struct inode *inode, | |
563 | struct extent_status *es) | |
564 | { | |
565 | /* | |
566 | * We don't need to worry about the race condition because | |
567 | * caller takes i_data_sem locking. | |
568 | */ | |
569 | BUG_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem)); | |
570 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
571 | ext4_es_insert_extent_ext_check(inode, es); | |
572 | else | |
573 | ext4_es_insert_extent_ind_check(inode, es); | |
574 | } | |
575 | #else | |
576 | static inline void ext4_es_insert_extent_check(struct inode *inode, | |
577 | struct extent_status *es) | |
578 | { | |
579 | } | |
580 | #endif | |
581 | ||
bdedbb7b | 582 | static int __es_insert_extent(struct inode *inode, struct extent_status *newes) |
654598be | 583 | { |
bdedbb7b | 584 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be ZL |
585 | struct rb_node **p = &tree->root.rb_node; |
586 | struct rb_node *parent = NULL; | |
587 | struct extent_status *es; | |
654598be ZL |
588 | |
589 | while (*p) { | |
590 | parent = *p; | |
591 | es = rb_entry(parent, struct extent_status, rb_node); | |
592 | ||
06b0c886 ZL |
593 | if (newes->es_lblk < es->es_lblk) { |
594 | if (ext4_es_can_be_merged(newes, es)) { | |
595 | /* | |
596 | * Here we can modify es_lblk directly | |
597 | * because it isn't overlapped. | |
598 | */ | |
599 | es->es_lblk = newes->es_lblk; | |
600 | es->es_len += newes->es_len; | |
fdc0212e ZL |
601 | if (ext4_es_is_written(es) || |
602 | ext4_es_is_unwritten(es)) | |
603 | ext4_es_store_pblock(es, | |
604 | newes->es_pblk); | |
bdedbb7b | 605 | es = ext4_es_try_to_merge_left(inode, es); |
654598be ZL |
606 | goto out; |
607 | } | |
608 | p = &(*p)->rb_left; | |
06b0c886 ZL |
609 | } else if (newes->es_lblk > ext4_es_end(es)) { |
610 | if (ext4_es_can_be_merged(es, newes)) { | |
611 | es->es_len += newes->es_len; | |
bdedbb7b | 612 | es = ext4_es_try_to_merge_right(inode, es); |
654598be ZL |
613 | goto out; |
614 | } | |
615 | p = &(*p)->rb_right; | |
616 | } else { | |
06b0c886 ZL |
617 | BUG_ON(1); |
618 | return -EINVAL; | |
654598be ZL |
619 | } |
620 | } | |
621 | ||
bdedbb7b | 622 | es = ext4_es_alloc_extent(inode, newes->es_lblk, newes->es_len, |
fdc0212e | 623 | newes->es_pblk); |
654598be ZL |
624 | if (!es) |
625 | return -ENOMEM; | |
626 | rb_link_node(&es->rb_node, parent, p); | |
627 | rb_insert_color(&es->rb_node, &tree->root); | |
628 | ||
629 | out: | |
630 | tree->cache_es = es; | |
631 | return 0; | |
632 | } | |
633 | ||
634 | /* | |
06b0c886 | 635 | * ext4_es_insert_extent() adds a space to a extent status tree. |
654598be ZL |
636 | * |
637 | * ext4_es_insert_extent is called by ext4_da_write_begin and | |
638 | * ext4_es_remove_extent. | |
639 | * | |
640 | * Return 0 on success, error code on failure. | |
641 | */ | |
06b0c886 | 642 | int ext4_es_insert_extent(struct inode *inode, ext4_lblk_t lblk, |
fdc0212e ZL |
643 | ext4_lblk_t len, ext4_fsblk_t pblk, |
644 | unsigned long long status) | |
654598be | 645 | { |
06b0c886 ZL |
646 | struct extent_status newes; |
647 | ext4_lblk_t end = lblk + len - 1; | |
654598be ZL |
648 | int err = 0; |
649 | ||
fdc0212e ZL |
650 | es_debug("add [%u/%u) %llu %llx to extent status tree of inode %lu\n", |
651 | lblk, len, pblk, status, inode->i_ino); | |
06b0c886 | 652 | |
d4381472 EG |
653 | if (!len) |
654 | return 0; | |
655 | ||
06b0c886 ZL |
656 | BUG_ON(end < lblk); |
657 | ||
658 | newes.es_lblk = lblk; | |
659 | newes.es_len = len; | |
fdc0212e ZL |
660 | ext4_es_store_pblock(&newes, pblk); |
661 | ext4_es_store_status(&newes, status); | |
662 | trace_ext4_es_insert_extent(inode, &newes); | |
654598be | 663 | |
921f266b DM |
664 | ext4_es_insert_extent_check(inode, &newes); |
665 | ||
654598be | 666 | write_lock(&EXT4_I(inode)->i_es_lock); |
bdedbb7b | 667 | err = __es_remove_extent(inode, lblk, end); |
06b0c886 ZL |
668 | if (err != 0) |
669 | goto error; | |
bdedbb7b | 670 | err = __es_insert_extent(inode, &newes); |
06b0c886 ZL |
671 | |
672 | error: | |
654598be ZL |
673 | write_unlock(&EXT4_I(inode)->i_es_lock); |
674 | ||
74cd15cd | 675 | ext4_es_lru_add(inode); |
654598be ZL |
676 | ext4_es_print_tree(inode); |
677 | ||
678 | return err; | |
679 | } | |
680 | ||
d100eef2 ZL |
681 | /* |
682 | * ext4_es_lookup_extent() looks up an extent in extent status tree. | |
683 | * | |
684 | * ext4_es_lookup_extent is called by ext4_map_blocks/ext4_da_map_blocks. | |
685 | * | |
686 | * Return: 1 on found, 0 on not | |
687 | */ | |
688 | int ext4_es_lookup_extent(struct inode *inode, ext4_lblk_t lblk, | |
689 | struct extent_status *es) | |
690 | { | |
691 | struct ext4_es_tree *tree; | |
692 | struct extent_status *es1 = NULL; | |
693 | struct rb_node *node; | |
694 | int found = 0; | |
695 | ||
696 | trace_ext4_es_lookup_extent_enter(inode, lblk); | |
697 | es_debug("lookup extent in block %u\n", lblk); | |
698 | ||
699 | tree = &EXT4_I(inode)->i_es_tree; | |
700 | read_lock(&EXT4_I(inode)->i_es_lock); | |
701 | ||
702 | /* find extent in cache firstly */ | |
703 | es->es_lblk = es->es_len = es->es_pblk = 0; | |
704 | if (tree->cache_es) { | |
705 | es1 = tree->cache_es; | |
706 | if (in_range(lblk, es1->es_lblk, es1->es_len)) { | |
707 | es_debug("%u cached by [%u/%u)\n", | |
708 | lblk, es1->es_lblk, es1->es_len); | |
709 | found = 1; | |
710 | goto out; | |
711 | } | |
712 | } | |
713 | ||
714 | node = tree->root.rb_node; | |
715 | while (node) { | |
716 | es1 = rb_entry(node, struct extent_status, rb_node); | |
717 | if (lblk < es1->es_lblk) | |
718 | node = node->rb_left; | |
719 | else if (lblk > ext4_es_end(es1)) | |
720 | node = node->rb_right; | |
721 | else { | |
722 | found = 1; | |
723 | break; | |
724 | } | |
725 | } | |
726 | ||
727 | out: | |
728 | if (found) { | |
729 | BUG_ON(!es1); | |
730 | es->es_lblk = es1->es_lblk; | |
731 | es->es_len = es1->es_len; | |
732 | es->es_pblk = es1->es_pblk; | |
733 | } | |
734 | ||
735 | read_unlock(&EXT4_I(inode)->i_es_lock); | |
736 | ||
74cd15cd | 737 | ext4_es_lru_add(inode); |
d100eef2 ZL |
738 | trace_ext4_es_lookup_extent_exit(inode, es, found); |
739 | return found; | |
740 | } | |
741 | ||
bdedbb7b ZL |
742 | static int __es_remove_extent(struct inode *inode, ext4_lblk_t lblk, |
743 | ext4_lblk_t end) | |
654598be | 744 | { |
bdedbb7b | 745 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be | 746 | struct rb_node *node; |
654598be ZL |
747 | struct extent_status *es; |
748 | struct extent_status orig_es; | |
06b0c886 | 749 | ext4_lblk_t len1, len2; |
fdc0212e | 750 | ext4_fsblk_t block; |
654598be ZL |
751 | int err = 0; |
752 | ||
06b0c886 | 753 | es = __es_tree_search(&tree->root, lblk); |
654598be ZL |
754 | if (!es) |
755 | goto out; | |
06b0c886 | 756 | if (es->es_lblk > end) |
654598be ZL |
757 | goto out; |
758 | ||
759 | /* Simply invalidate cache_es. */ | |
760 | tree->cache_es = NULL; | |
761 | ||
06b0c886 ZL |
762 | orig_es.es_lblk = es->es_lblk; |
763 | orig_es.es_len = es->es_len; | |
fdc0212e ZL |
764 | orig_es.es_pblk = es->es_pblk; |
765 | ||
06b0c886 ZL |
766 | len1 = lblk > es->es_lblk ? lblk - es->es_lblk : 0; |
767 | len2 = ext4_es_end(es) > end ? ext4_es_end(es) - end : 0; | |
654598be | 768 | if (len1 > 0) |
06b0c886 | 769 | es->es_len = len1; |
654598be ZL |
770 | if (len2 > 0) { |
771 | if (len1 > 0) { | |
06b0c886 ZL |
772 | struct extent_status newes; |
773 | ||
774 | newes.es_lblk = end + 1; | |
775 | newes.es_len = len2; | |
fdc0212e ZL |
776 | if (ext4_es_is_written(&orig_es) || |
777 | ext4_es_is_unwritten(&orig_es)) { | |
778 | block = ext4_es_pblock(&orig_es) + | |
779 | orig_es.es_len - len2; | |
780 | ext4_es_store_pblock(&newes, block); | |
781 | } | |
782 | ext4_es_store_status(&newes, ext4_es_status(&orig_es)); | |
bdedbb7b | 783 | err = __es_insert_extent(inode, &newes); |
654598be | 784 | if (err) { |
06b0c886 ZL |
785 | es->es_lblk = orig_es.es_lblk; |
786 | es->es_len = orig_es.es_len; | |
654598be ZL |
787 | goto out; |
788 | } | |
789 | } else { | |
06b0c886 ZL |
790 | es->es_lblk = end + 1; |
791 | es->es_len = len2; | |
fdc0212e ZL |
792 | if (ext4_es_is_written(es) || |
793 | ext4_es_is_unwritten(es)) { | |
794 | block = orig_es.es_pblk + orig_es.es_len - len2; | |
795 | ext4_es_store_pblock(es, block); | |
796 | } | |
654598be ZL |
797 | } |
798 | goto out; | |
799 | } | |
800 | ||
801 | if (len1 > 0) { | |
802 | node = rb_next(&es->rb_node); | |
803 | if (node) | |
804 | es = rb_entry(node, struct extent_status, rb_node); | |
805 | else | |
806 | es = NULL; | |
807 | } | |
808 | ||
06b0c886 | 809 | while (es && ext4_es_end(es) <= end) { |
654598be ZL |
810 | node = rb_next(&es->rb_node); |
811 | rb_erase(&es->rb_node, &tree->root); | |
bdedbb7b | 812 | ext4_es_free_extent(inode, es); |
654598be ZL |
813 | if (!node) { |
814 | es = NULL; | |
815 | break; | |
816 | } | |
817 | es = rb_entry(node, struct extent_status, rb_node); | |
818 | } | |
819 | ||
06b0c886 | 820 | if (es && es->es_lblk < end + 1) { |
fdc0212e ZL |
821 | ext4_lblk_t orig_len = es->es_len; |
822 | ||
06b0c886 ZL |
823 | len1 = ext4_es_end(es) - end; |
824 | es->es_lblk = end + 1; | |
825 | es->es_len = len1; | |
fdc0212e ZL |
826 | if (ext4_es_is_written(es) || ext4_es_is_unwritten(es)) { |
827 | block = es->es_pblk + orig_len - len1; | |
828 | ext4_es_store_pblock(es, block); | |
829 | } | |
654598be ZL |
830 | } |
831 | ||
832 | out: | |
06b0c886 ZL |
833 | return err; |
834 | } | |
835 | ||
836 | /* | |
837 | * ext4_es_remove_extent() removes a space from a extent status tree. | |
838 | * | |
839 | * Return 0 on success, error code on failure. | |
840 | */ | |
841 | int ext4_es_remove_extent(struct inode *inode, ext4_lblk_t lblk, | |
842 | ext4_lblk_t len) | |
843 | { | |
06b0c886 ZL |
844 | ext4_lblk_t end; |
845 | int err = 0; | |
846 | ||
847 | trace_ext4_es_remove_extent(inode, lblk, len); | |
848 | es_debug("remove [%u/%u) from extent status tree of inode %lu\n", | |
849 | lblk, len, inode->i_ino); | |
850 | ||
d4381472 EG |
851 | if (!len) |
852 | return err; | |
853 | ||
06b0c886 ZL |
854 | end = lblk + len - 1; |
855 | BUG_ON(end < lblk); | |
856 | ||
06b0c886 | 857 | write_lock(&EXT4_I(inode)->i_es_lock); |
bdedbb7b | 858 | err = __es_remove_extent(inode, lblk, end); |
654598be ZL |
859 | write_unlock(&EXT4_I(inode)->i_es_lock); |
860 | ext4_es_print_tree(inode); | |
861 | return err; | |
862 | } | |
74cd15cd | 863 | |
adb23551 ZL |
864 | int ext4_es_zeroout(struct inode *inode, struct ext4_extent *ex) |
865 | { | |
866 | ext4_lblk_t ee_block; | |
867 | ext4_fsblk_t ee_pblock; | |
868 | unsigned int ee_len; | |
869 | ||
870 | ee_block = le32_to_cpu(ex->ee_block); | |
871 | ee_len = ext4_ext_get_actual_len(ex); | |
872 | ee_pblock = ext4_ext_pblock(ex); | |
873 | ||
874 | if (ee_len == 0) | |
875 | return 0; | |
876 | ||
877 | return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock, | |
878 | EXTENT_STATUS_WRITTEN); | |
879 | } | |
880 | ||
74cd15cd ZL |
881 | static int ext4_es_shrink(struct shrinker *shrink, struct shrink_control *sc) |
882 | { | |
883 | struct ext4_sb_info *sbi = container_of(shrink, | |
884 | struct ext4_sb_info, s_es_shrinker); | |
885 | struct ext4_inode_info *ei; | |
886 | struct list_head *cur, *tmp, scanned; | |
887 | int nr_to_scan = sc->nr_to_scan; | |
888 | int ret, nr_shrunk = 0; | |
889 | ||
1ac6466f | 890 | ret = percpu_counter_read_positive(&sbi->s_extent_cache_cnt); |
24630774 | 891 | trace_ext4_es_shrink_enter(sbi->s_sb, nr_to_scan, ret); |
74cd15cd ZL |
892 | |
893 | if (!nr_to_scan) | |
24630774 | 894 | return ret; |
74cd15cd ZL |
895 | |
896 | INIT_LIST_HEAD(&scanned); | |
897 | ||
898 | spin_lock(&sbi->s_es_lru_lock); | |
899 | list_for_each_safe(cur, tmp, &sbi->s_es_lru) { | |
900 | list_move_tail(cur, &scanned); | |
901 | ||
902 | ei = list_entry(cur, struct ext4_inode_info, i_es_lru); | |
903 | ||
904 | read_lock(&ei->i_es_lock); | |
905 | if (ei->i_es_lru_nr == 0) { | |
906 | read_unlock(&ei->i_es_lock); | |
907 | continue; | |
908 | } | |
909 | read_unlock(&ei->i_es_lock); | |
910 | ||
911 | write_lock(&ei->i_es_lock); | |
912 | ret = __es_try_to_reclaim_extents(ei, nr_to_scan); | |
913 | write_unlock(&ei->i_es_lock); | |
914 | ||
915 | nr_shrunk += ret; | |
916 | nr_to_scan -= ret; | |
917 | if (nr_to_scan == 0) | |
918 | break; | |
919 | } | |
920 | list_splice_tail(&scanned, &sbi->s_es_lru); | |
921 | spin_unlock(&sbi->s_es_lru_lock); | |
74cd15cd | 922 | |
1ac6466f | 923 | ret = percpu_counter_read_positive(&sbi->s_extent_cache_cnt); |
24630774 TT |
924 | trace_ext4_es_shrink_exit(sbi->s_sb, nr_shrunk, ret); |
925 | return ret; | |
74cd15cd ZL |
926 | } |
927 | ||
928 | void ext4_es_register_shrinker(struct super_block *sb) | |
929 | { | |
930 | struct ext4_sb_info *sbi; | |
931 | ||
932 | sbi = EXT4_SB(sb); | |
933 | INIT_LIST_HEAD(&sbi->s_es_lru); | |
934 | spin_lock_init(&sbi->s_es_lru_lock); | |
935 | sbi->s_es_shrinker.shrink = ext4_es_shrink; | |
936 | sbi->s_es_shrinker.seeks = DEFAULT_SEEKS; | |
937 | register_shrinker(&sbi->s_es_shrinker); | |
938 | } | |
939 | ||
940 | void ext4_es_unregister_shrinker(struct super_block *sb) | |
941 | { | |
942 | unregister_shrinker(&EXT4_SB(sb)->s_es_shrinker); | |
943 | } | |
944 | ||
945 | void ext4_es_lru_add(struct inode *inode) | |
946 | { | |
947 | struct ext4_inode_info *ei = EXT4_I(inode); | |
948 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
949 | ||
950 | spin_lock(&sbi->s_es_lru_lock); | |
951 | if (list_empty(&ei->i_es_lru)) | |
952 | list_add_tail(&ei->i_es_lru, &sbi->s_es_lru); | |
953 | else | |
954 | list_move_tail(&ei->i_es_lru, &sbi->s_es_lru); | |
955 | spin_unlock(&sbi->s_es_lru_lock); | |
956 | } | |
957 | ||
958 | void ext4_es_lru_del(struct inode *inode) | |
959 | { | |
960 | struct ext4_inode_info *ei = EXT4_I(inode); | |
961 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
962 | ||
963 | spin_lock(&sbi->s_es_lru_lock); | |
964 | if (!list_empty(&ei->i_es_lru)) | |
965 | list_del_init(&ei->i_es_lru); | |
966 | spin_unlock(&sbi->s_es_lru_lock); | |
967 | } | |
968 | ||
74cd15cd ZL |
969 | static int __es_try_to_reclaim_extents(struct ext4_inode_info *ei, |
970 | int nr_to_scan) | |
971 | { | |
972 | struct inode *inode = &ei->vfs_inode; | |
973 | struct ext4_es_tree *tree = &ei->i_es_tree; | |
974 | struct rb_node *node; | |
975 | struct extent_status *es; | |
976 | int nr_shrunk = 0; | |
977 | ||
978 | if (ei->i_es_lru_nr == 0) | |
979 | return 0; | |
980 | ||
981 | node = rb_first(&tree->root); | |
982 | while (node != NULL) { | |
983 | es = rb_entry(node, struct extent_status, rb_node); | |
984 | node = rb_next(&es->rb_node); | |
985 | /* | |
986 | * We can't reclaim delayed extent from status tree because | |
987 | * fiemap, bigallic, and seek_data/hole need to use it. | |
988 | */ | |
989 | if (!ext4_es_is_delayed(es)) { | |
990 | rb_erase(&es->rb_node, &tree->root); | |
991 | ext4_es_free_extent(inode, es); | |
992 | nr_shrunk++; | |
993 | if (--nr_to_scan == 0) | |
994 | break; | |
995 | } | |
996 | } | |
997 | tree->cache_es = NULL; | |
998 | return nr_shrunk; | |
999 | } |