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> | |
d3922a77 | 13 | #include <linux/list_sort.h> |
654598be ZL |
14 | #include "ext4.h" |
15 | #include "extents_status.h" | |
654598be | 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); | |
e15f742c TT |
150 | static int __ext4_es_shrink(struct ext4_sb_info *sbi, int nr_to_scan, |
151 | struct ext4_inode_info *locked_ei); | |
06b0c886 | 152 | |
654598be ZL |
153 | int __init ext4_init_es(void) |
154 | { | |
24630774 TT |
155 | ext4_es_cachep = kmem_cache_create("ext4_extent_status", |
156 | sizeof(struct extent_status), | |
157 | 0, (SLAB_RECLAIM_ACCOUNT), NULL); | |
654598be ZL |
158 | if (ext4_es_cachep == NULL) |
159 | return -ENOMEM; | |
160 | return 0; | |
161 | } | |
162 | ||
163 | void ext4_exit_es(void) | |
164 | { | |
165 | if (ext4_es_cachep) | |
166 | kmem_cache_destroy(ext4_es_cachep); | |
167 | } | |
168 | ||
169 | void ext4_es_init_tree(struct ext4_es_tree *tree) | |
170 | { | |
171 | tree->root = RB_ROOT; | |
172 | tree->cache_es = NULL; | |
173 | } | |
174 | ||
175 | #ifdef ES_DEBUG__ | |
176 | static void ext4_es_print_tree(struct inode *inode) | |
177 | { | |
178 | struct ext4_es_tree *tree; | |
179 | struct rb_node *node; | |
180 | ||
181 | printk(KERN_DEBUG "status extents for inode %lu:", inode->i_ino); | |
182 | tree = &EXT4_I(inode)->i_es_tree; | |
183 | node = rb_first(&tree->root); | |
184 | while (node) { | |
185 | struct extent_status *es; | |
186 | es = rb_entry(node, struct extent_status, rb_node); | |
ce140cdd | 187 | printk(KERN_DEBUG " [%u/%u) %llu %x", |
fdc0212e ZL |
188 | es->es_lblk, es->es_len, |
189 | ext4_es_pblock(es), ext4_es_status(es)); | |
654598be ZL |
190 | node = rb_next(node); |
191 | } | |
192 | printk(KERN_DEBUG "\n"); | |
193 | } | |
194 | #else | |
195 | #define ext4_es_print_tree(inode) | |
196 | #endif | |
197 | ||
06b0c886 | 198 | static inline ext4_lblk_t ext4_es_end(struct extent_status *es) |
654598be | 199 | { |
06b0c886 ZL |
200 | BUG_ON(es->es_lblk + es->es_len < es->es_lblk); |
201 | return es->es_lblk + es->es_len - 1; | |
654598be ZL |
202 | } |
203 | ||
204 | /* | |
205 | * search through the tree for an delayed extent with a given offset. If | |
206 | * it can't be found, try to find next extent. | |
207 | */ | |
208 | static struct extent_status *__es_tree_search(struct rb_root *root, | |
06b0c886 | 209 | ext4_lblk_t lblk) |
654598be ZL |
210 | { |
211 | struct rb_node *node = root->rb_node; | |
212 | struct extent_status *es = NULL; | |
213 | ||
214 | while (node) { | |
215 | es = rb_entry(node, struct extent_status, rb_node); | |
06b0c886 | 216 | if (lblk < es->es_lblk) |
654598be | 217 | node = node->rb_left; |
06b0c886 | 218 | else if (lblk > ext4_es_end(es)) |
654598be ZL |
219 | node = node->rb_right; |
220 | else | |
221 | return es; | |
222 | } | |
223 | ||
06b0c886 | 224 | if (es && lblk < es->es_lblk) |
654598be ZL |
225 | return es; |
226 | ||
06b0c886 | 227 | if (es && lblk > ext4_es_end(es)) { |
654598be ZL |
228 | node = rb_next(&es->rb_node); |
229 | return node ? rb_entry(node, struct extent_status, rb_node) : | |
230 | NULL; | |
231 | } | |
232 | ||
233 | return NULL; | |
234 | } | |
235 | ||
236 | /* | |
e30b5dca YZ |
237 | * ext4_es_find_delayed_extent_range: find the 1st delayed extent covering |
238 | * @es->lblk if it exists, otherwise, the next extent after @es->lblk. | |
654598be ZL |
239 | * |
240 | * @inode: the inode which owns delayed extents | |
be401363 | 241 | * @lblk: the offset where we start to search |
e30b5dca | 242 | * @end: the offset where we stop to search |
654598be | 243 | * @es: delayed extent that we found |
654598be | 244 | */ |
e30b5dca YZ |
245 | void ext4_es_find_delayed_extent_range(struct inode *inode, |
246 | ext4_lblk_t lblk, ext4_lblk_t end, | |
be401363 | 247 | struct extent_status *es) |
654598be ZL |
248 | { |
249 | struct ext4_es_tree *tree = NULL; | |
250 | struct extent_status *es1 = NULL; | |
251 | struct rb_node *node; | |
654598be | 252 | |
be401363 | 253 | BUG_ON(es == NULL); |
e30b5dca YZ |
254 | BUG_ON(end < lblk); |
255 | trace_ext4_es_find_delayed_extent_range_enter(inode, lblk); | |
992e9fdd | 256 | |
654598be ZL |
257 | read_lock(&EXT4_I(inode)->i_es_lock); |
258 | tree = &EXT4_I(inode)->i_es_tree; | |
259 | ||
fdc0212e | 260 | /* find extent in cache firstly */ |
be401363 | 261 | es->es_lblk = es->es_len = es->es_pblk = 0; |
654598be ZL |
262 | if (tree->cache_es) { |
263 | es1 = tree->cache_es; | |
be401363 | 264 | if (in_range(lblk, es1->es_lblk, es1->es_len)) { |
3be78c73 | 265 | es_debug("%u cached by [%u/%u) %llu %x\n", |
be401363 | 266 | lblk, es1->es_lblk, es1->es_len, |
fdc0212e | 267 | ext4_es_pblock(es1), ext4_es_status(es1)); |
654598be ZL |
268 | goto out; |
269 | } | |
270 | } | |
271 | ||
be401363 | 272 | es1 = __es_tree_search(&tree->root, lblk); |
654598be ZL |
273 | |
274 | out: | |
be401363 ZL |
275 | if (es1 && !ext4_es_is_delayed(es1)) { |
276 | while ((node = rb_next(&es1->rb_node)) != NULL) { | |
277 | es1 = rb_entry(node, struct extent_status, rb_node); | |
e30b5dca YZ |
278 | if (es1->es_lblk > end) { |
279 | es1 = NULL; | |
280 | break; | |
281 | } | |
be401363 ZL |
282 | if (ext4_es_is_delayed(es1)) |
283 | break; | |
284 | } | |
285 | } | |
286 | ||
287 | if (es1 && ext4_es_is_delayed(es1)) { | |
654598be | 288 | tree->cache_es = es1; |
06b0c886 ZL |
289 | es->es_lblk = es1->es_lblk; |
290 | es->es_len = es1->es_len; | |
fdc0212e | 291 | es->es_pblk = es1->es_pblk; |
654598be ZL |
292 | } |
293 | ||
294 | read_unlock(&EXT4_I(inode)->i_es_lock); | |
992e9fdd | 295 | |
e30b5dca | 296 | trace_ext4_es_find_delayed_extent_range_exit(inode, es); |
654598be ZL |
297 | } |
298 | ||
299 | static struct extent_status * | |
bdedbb7b ZL |
300 | ext4_es_alloc_extent(struct inode *inode, ext4_lblk_t lblk, ext4_lblk_t len, |
301 | ext4_fsblk_t pblk) | |
654598be ZL |
302 | { |
303 | struct extent_status *es; | |
304 | es = kmem_cache_alloc(ext4_es_cachep, GFP_ATOMIC); | |
305 | if (es == NULL) | |
306 | return NULL; | |
06b0c886 ZL |
307 | es->es_lblk = lblk; |
308 | es->es_len = len; | |
fdc0212e | 309 | es->es_pblk = pblk; |
74cd15cd ZL |
310 | |
311 | /* | |
312 | * We don't count delayed extent because we never try to reclaim them | |
313 | */ | |
24630774 | 314 | if (!ext4_es_is_delayed(es)) { |
74cd15cd | 315 | EXT4_I(inode)->i_es_lru_nr++; |
1ac6466f | 316 | percpu_counter_inc(&EXT4_SB(inode->i_sb)->s_extent_cache_cnt); |
24630774 | 317 | } |
74cd15cd | 318 | |
654598be ZL |
319 | return es; |
320 | } | |
321 | ||
bdedbb7b | 322 | static void ext4_es_free_extent(struct inode *inode, struct extent_status *es) |
654598be | 323 | { |
74cd15cd ZL |
324 | /* Decrease the lru counter when this es is not delayed */ |
325 | if (!ext4_es_is_delayed(es)) { | |
326 | BUG_ON(EXT4_I(inode)->i_es_lru_nr == 0); | |
327 | EXT4_I(inode)->i_es_lru_nr--; | |
1ac6466f | 328 | percpu_counter_dec(&EXT4_SB(inode->i_sb)->s_extent_cache_cnt); |
74cd15cd ZL |
329 | } |
330 | ||
654598be ZL |
331 | kmem_cache_free(ext4_es_cachep, es); |
332 | } | |
333 | ||
06b0c886 ZL |
334 | /* |
335 | * Check whether or not two extents can be merged | |
336 | * Condition: | |
337 | * - logical block number is contiguous | |
fdc0212e ZL |
338 | * - physical block number is contiguous |
339 | * - status is equal | |
06b0c886 ZL |
340 | */ |
341 | static int ext4_es_can_be_merged(struct extent_status *es1, | |
342 | struct extent_status *es2) | |
343 | { | |
bd384364 | 344 | if (ext4_es_status(es1) != ext4_es_status(es2)) |
06b0c886 ZL |
345 | return 0; |
346 | ||
0baaea64 LC |
347 | if (((__u64) es1->es_len) + es2->es_len > EXT_MAX_BLOCKS) { |
348 | pr_warn("ES assertion failed when merging extents. " | |
349 | "The sum of lengths of es1 (%d) and es2 (%d) " | |
350 | "is bigger than allowed file size (%d)\n", | |
351 | es1->es_len, es2->es_len, EXT_MAX_BLOCKS); | |
352 | WARN_ON(1); | |
fdc0212e | 353 | return 0; |
0baaea64 | 354 | } |
fdc0212e | 355 | |
bd384364 | 356 | if (((__u64) es1->es_lblk) + es1->es_len != es2->es_lblk) |
fdc0212e ZL |
357 | return 0; |
358 | ||
bd384364 ZL |
359 | if ((ext4_es_is_written(es1) || ext4_es_is_unwritten(es1)) && |
360 | (ext4_es_pblock(es1) + es1->es_len == ext4_es_pblock(es2))) | |
361 | return 1; | |
362 | ||
363 | if (ext4_es_is_hole(es1)) | |
364 | return 1; | |
365 | ||
366 | /* we need to check delayed extent is without unwritten status */ | |
367 | if (ext4_es_is_delayed(es1) && !ext4_es_is_unwritten(es1)) | |
368 | return 1; | |
369 | ||
370 | return 0; | |
06b0c886 ZL |
371 | } |
372 | ||
654598be | 373 | static struct extent_status * |
bdedbb7b | 374 | ext4_es_try_to_merge_left(struct inode *inode, struct extent_status *es) |
654598be | 375 | { |
bdedbb7b | 376 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be ZL |
377 | struct extent_status *es1; |
378 | struct rb_node *node; | |
379 | ||
380 | node = rb_prev(&es->rb_node); | |
381 | if (!node) | |
382 | return es; | |
383 | ||
384 | es1 = rb_entry(node, struct extent_status, rb_node); | |
06b0c886 ZL |
385 | if (ext4_es_can_be_merged(es1, es)) { |
386 | es1->es_len += es->es_len; | |
654598be | 387 | rb_erase(&es->rb_node, &tree->root); |
bdedbb7b | 388 | ext4_es_free_extent(inode, es); |
654598be ZL |
389 | es = es1; |
390 | } | |
391 | ||
392 | return es; | |
393 | } | |
394 | ||
395 | static struct extent_status * | |
bdedbb7b | 396 | ext4_es_try_to_merge_right(struct inode *inode, struct extent_status *es) |
654598be | 397 | { |
bdedbb7b | 398 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be ZL |
399 | struct extent_status *es1; |
400 | struct rb_node *node; | |
401 | ||
402 | node = rb_next(&es->rb_node); | |
403 | if (!node) | |
404 | return es; | |
405 | ||
406 | es1 = rb_entry(node, struct extent_status, rb_node); | |
06b0c886 ZL |
407 | if (ext4_es_can_be_merged(es, es1)) { |
408 | es->es_len += es1->es_len; | |
654598be | 409 | rb_erase(node, &tree->root); |
bdedbb7b | 410 | ext4_es_free_extent(inode, es1); |
654598be ZL |
411 | } |
412 | ||
413 | return es; | |
414 | } | |
415 | ||
921f266b | 416 | #ifdef ES_AGGRESSIVE_TEST |
d7b2a00c ZL |
417 | #include "ext4_extents.h" /* Needed when ES_AGGRESSIVE_TEST is defined */ |
418 | ||
921f266b DM |
419 | static void ext4_es_insert_extent_ext_check(struct inode *inode, |
420 | struct extent_status *es) | |
421 | { | |
422 | struct ext4_ext_path *path = NULL; | |
423 | struct ext4_extent *ex; | |
424 | ext4_lblk_t ee_block; | |
425 | ext4_fsblk_t ee_start; | |
426 | unsigned short ee_len; | |
427 | int depth, ee_status, es_status; | |
428 | ||
107a7bd3 | 429 | path = ext4_ext_find_extent(inode, es->es_lblk, NULL, EXT4_EX_NOCACHE); |
921f266b DM |
430 | if (IS_ERR(path)) |
431 | return; | |
432 | ||
433 | depth = ext_depth(inode); | |
434 | ex = path[depth].p_ext; | |
435 | ||
436 | if (ex) { | |
437 | ||
438 | ee_block = le32_to_cpu(ex->ee_block); | |
439 | ee_start = ext4_ext_pblock(ex); | |
440 | ee_len = ext4_ext_get_actual_len(ex); | |
441 | ||
556615dc | 442 | ee_status = ext4_ext_is_unwritten(ex) ? 1 : 0; |
921f266b DM |
443 | es_status = ext4_es_is_unwritten(es) ? 1 : 0; |
444 | ||
445 | /* | |
446 | * Make sure ex and es are not overlap when we try to insert | |
447 | * a delayed/hole extent. | |
448 | */ | |
449 | if (!ext4_es_is_written(es) && !ext4_es_is_unwritten(es)) { | |
450 | if (in_range(es->es_lblk, ee_block, ee_len)) { | |
bdafe42a | 451 | pr_warn("ES insert assertion failed for " |
921f266b DM |
452 | "inode: %lu we can find an extent " |
453 | "at block [%d/%d/%llu/%c], but we " | |
ce140cdd EW |
454 | "want to add a delayed/hole extent " |
455 | "[%d/%d/%llu/%x]\n", | |
921f266b DM |
456 | inode->i_ino, ee_block, ee_len, |
457 | ee_start, ee_status ? 'u' : 'w', | |
458 | es->es_lblk, es->es_len, | |
459 | ext4_es_pblock(es), ext4_es_status(es)); | |
460 | } | |
461 | goto out; | |
462 | } | |
463 | ||
464 | /* | |
465 | * We don't check ee_block == es->es_lblk, etc. because es | |
466 | * might be a part of whole extent, vice versa. | |
467 | */ | |
468 | if (es->es_lblk < ee_block || | |
469 | ext4_es_pblock(es) != ee_start + es->es_lblk - ee_block) { | |
bdafe42a | 470 | pr_warn("ES insert assertion failed for inode: %lu " |
921f266b DM |
471 | "ex_status [%d/%d/%llu/%c] != " |
472 | "es_status [%d/%d/%llu/%c]\n", inode->i_ino, | |
473 | ee_block, ee_len, ee_start, | |
474 | ee_status ? 'u' : 'w', es->es_lblk, es->es_len, | |
475 | ext4_es_pblock(es), es_status ? 'u' : 'w'); | |
476 | goto out; | |
477 | } | |
478 | ||
479 | if (ee_status ^ es_status) { | |
bdafe42a | 480 | pr_warn("ES insert assertion failed for inode: %lu " |
921f266b DM |
481 | "ex_status [%d/%d/%llu/%c] != " |
482 | "es_status [%d/%d/%llu/%c]\n", inode->i_ino, | |
483 | ee_block, ee_len, ee_start, | |
484 | ee_status ? 'u' : 'w', es->es_lblk, es->es_len, | |
485 | ext4_es_pblock(es), es_status ? 'u' : 'w'); | |
486 | } | |
487 | } else { | |
488 | /* | |
489 | * We can't find an extent on disk. So we need to make sure | |
490 | * that we don't want to add an written/unwritten extent. | |
491 | */ | |
492 | if (!ext4_es_is_delayed(es) && !ext4_es_is_hole(es)) { | |
bdafe42a | 493 | pr_warn("ES insert assertion failed for inode: %lu " |
921f266b | 494 | "can't find an extent at block %d but we want " |
ce140cdd EW |
495 | "to add a written/unwritten extent " |
496 | "[%d/%d/%llu/%x]\n", inode->i_ino, | |
921f266b DM |
497 | es->es_lblk, es->es_lblk, es->es_len, |
498 | ext4_es_pblock(es), ext4_es_status(es)); | |
499 | } | |
500 | } | |
501 | out: | |
502 | if (path) { | |
503 | ext4_ext_drop_refs(path); | |
504 | kfree(path); | |
505 | } | |
506 | } | |
507 | ||
508 | static void ext4_es_insert_extent_ind_check(struct inode *inode, | |
509 | struct extent_status *es) | |
510 | { | |
511 | struct ext4_map_blocks map; | |
512 | int retval; | |
513 | ||
514 | /* | |
515 | * Here we call ext4_ind_map_blocks to lookup a block mapping because | |
516 | * 'Indirect' structure is defined in indirect.c. So we couldn't | |
517 | * access direct/indirect tree from outside. It is too dirty to define | |
518 | * this function in indirect.c file. | |
519 | */ | |
520 | ||
521 | map.m_lblk = es->es_lblk; | |
522 | map.m_len = es->es_len; | |
523 | ||
524 | retval = ext4_ind_map_blocks(NULL, inode, &map, 0); | |
525 | if (retval > 0) { | |
526 | if (ext4_es_is_delayed(es) || ext4_es_is_hole(es)) { | |
527 | /* | |
528 | * We want to add a delayed/hole extent but this | |
529 | * block has been allocated. | |
530 | */ | |
bdafe42a | 531 | pr_warn("ES insert assertion failed for inode: %lu " |
921f266b | 532 | "We can find blocks but we want to add a " |
ce140cdd | 533 | "delayed/hole extent [%d/%d/%llu/%x]\n", |
921f266b DM |
534 | inode->i_ino, es->es_lblk, es->es_len, |
535 | ext4_es_pblock(es), ext4_es_status(es)); | |
536 | return; | |
537 | } else if (ext4_es_is_written(es)) { | |
538 | if (retval != es->es_len) { | |
bdafe42a | 539 | pr_warn("ES insert assertion failed for " |
921f266b DM |
540 | "inode: %lu retval %d != es_len %d\n", |
541 | inode->i_ino, retval, es->es_len); | |
542 | return; | |
543 | } | |
544 | if (map.m_pblk != ext4_es_pblock(es)) { | |
bdafe42a | 545 | pr_warn("ES insert assertion failed for " |
921f266b DM |
546 | "inode: %lu m_pblk %llu != " |
547 | "es_pblk %llu\n", | |
548 | inode->i_ino, map.m_pblk, | |
549 | ext4_es_pblock(es)); | |
550 | return; | |
551 | } | |
552 | } else { | |
553 | /* | |
554 | * We don't need to check unwritten extent because | |
555 | * indirect-based file doesn't have it. | |
556 | */ | |
557 | BUG_ON(1); | |
558 | } | |
559 | } else if (retval == 0) { | |
560 | if (ext4_es_is_written(es)) { | |
bdafe42a | 561 | pr_warn("ES insert assertion failed for inode: %lu " |
921f266b | 562 | "We can't find the block but we want to add " |
ce140cdd | 563 | "a written extent [%d/%d/%llu/%x]\n", |
921f266b DM |
564 | inode->i_ino, es->es_lblk, es->es_len, |
565 | ext4_es_pblock(es), ext4_es_status(es)); | |
566 | return; | |
567 | } | |
568 | } | |
569 | } | |
570 | ||
571 | static inline void ext4_es_insert_extent_check(struct inode *inode, | |
572 | struct extent_status *es) | |
573 | { | |
574 | /* | |
575 | * We don't need to worry about the race condition because | |
576 | * caller takes i_data_sem locking. | |
577 | */ | |
578 | BUG_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem)); | |
579 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
580 | ext4_es_insert_extent_ext_check(inode, es); | |
581 | else | |
582 | ext4_es_insert_extent_ind_check(inode, es); | |
583 | } | |
584 | #else | |
585 | static inline void ext4_es_insert_extent_check(struct inode *inode, | |
586 | struct extent_status *es) | |
587 | { | |
588 | } | |
589 | #endif | |
590 | ||
bdedbb7b | 591 | static int __es_insert_extent(struct inode *inode, struct extent_status *newes) |
654598be | 592 | { |
bdedbb7b | 593 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be ZL |
594 | struct rb_node **p = &tree->root.rb_node; |
595 | struct rb_node *parent = NULL; | |
596 | struct extent_status *es; | |
654598be ZL |
597 | |
598 | while (*p) { | |
599 | parent = *p; | |
600 | es = rb_entry(parent, struct extent_status, rb_node); | |
601 | ||
06b0c886 ZL |
602 | if (newes->es_lblk < es->es_lblk) { |
603 | if (ext4_es_can_be_merged(newes, es)) { | |
604 | /* | |
605 | * Here we can modify es_lblk directly | |
606 | * because it isn't overlapped. | |
607 | */ | |
608 | es->es_lblk = newes->es_lblk; | |
609 | es->es_len += newes->es_len; | |
fdc0212e ZL |
610 | if (ext4_es_is_written(es) || |
611 | ext4_es_is_unwritten(es)) | |
612 | ext4_es_store_pblock(es, | |
613 | newes->es_pblk); | |
bdedbb7b | 614 | es = ext4_es_try_to_merge_left(inode, es); |
654598be ZL |
615 | goto out; |
616 | } | |
617 | p = &(*p)->rb_left; | |
06b0c886 ZL |
618 | } else if (newes->es_lblk > ext4_es_end(es)) { |
619 | if (ext4_es_can_be_merged(es, newes)) { | |
620 | es->es_len += newes->es_len; | |
bdedbb7b | 621 | es = ext4_es_try_to_merge_right(inode, es); |
654598be ZL |
622 | goto out; |
623 | } | |
624 | p = &(*p)->rb_right; | |
625 | } else { | |
06b0c886 ZL |
626 | BUG_ON(1); |
627 | return -EINVAL; | |
654598be ZL |
628 | } |
629 | } | |
630 | ||
bdedbb7b | 631 | es = ext4_es_alloc_extent(inode, newes->es_lblk, newes->es_len, |
fdc0212e | 632 | newes->es_pblk); |
654598be ZL |
633 | if (!es) |
634 | return -ENOMEM; | |
635 | rb_link_node(&es->rb_node, parent, p); | |
636 | rb_insert_color(&es->rb_node, &tree->root); | |
637 | ||
638 | out: | |
639 | tree->cache_es = es; | |
640 | return 0; | |
641 | } | |
642 | ||
643 | /* | |
bdafe42a TT |
644 | * ext4_es_insert_extent() adds information to an inode's extent |
645 | * status tree. | |
654598be ZL |
646 | * |
647 | * Return 0 on success, error code on failure. | |
648 | */ | |
06b0c886 | 649 | int ext4_es_insert_extent(struct inode *inode, ext4_lblk_t lblk, |
fdc0212e | 650 | ext4_lblk_t len, ext4_fsblk_t pblk, |
3be78c73 | 651 | unsigned int status) |
654598be | 652 | { |
06b0c886 ZL |
653 | struct extent_status newes; |
654 | ext4_lblk_t end = lblk + len - 1; | |
654598be ZL |
655 | int err = 0; |
656 | ||
3be78c73 | 657 | es_debug("add [%u/%u) %llu %x to extent status tree of inode %lu\n", |
fdc0212e | 658 | lblk, len, pblk, status, inode->i_ino); |
06b0c886 | 659 | |
d4381472 EG |
660 | if (!len) |
661 | return 0; | |
662 | ||
06b0c886 ZL |
663 | BUG_ON(end < lblk); |
664 | ||
665 | newes.es_lblk = lblk; | |
666 | newes.es_len = len; | |
9a6633b1 | 667 | ext4_es_store_pblock_status(&newes, pblk, status); |
fdc0212e | 668 | trace_ext4_es_insert_extent(inode, &newes); |
654598be | 669 | |
921f266b DM |
670 | ext4_es_insert_extent_check(inode, &newes); |
671 | ||
654598be | 672 | write_lock(&EXT4_I(inode)->i_es_lock); |
bdedbb7b | 673 | err = __es_remove_extent(inode, lblk, end); |
06b0c886 ZL |
674 | if (err != 0) |
675 | goto error; | |
e15f742c | 676 | retry: |
bdedbb7b | 677 | err = __es_insert_extent(inode, &newes); |
e15f742c TT |
678 | if (err == -ENOMEM && __ext4_es_shrink(EXT4_SB(inode->i_sb), 1, |
679 | EXT4_I(inode))) | |
680 | goto retry; | |
681 | if (err == -ENOMEM && !ext4_es_is_delayed(&newes)) | |
682 | err = 0; | |
06b0c886 ZL |
683 | |
684 | error: | |
654598be ZL |
685 | write_unlock(&EXT4_I(inode)->i_es_lock); |
686 | ||
687 | ext4_es_print_tree(inode); | |
688 | ||
689 | return err; | |
690 | } | |
691 | ||
107a7bd3 TT |
692 | /* |
693 | * ext4_es_cache_extent() inserts information into the extent status | |
694 | * tree if and only if there isn't information about the range in | |
695 | * question already. | |
696 | */ | |
697 | void ext4_es_cache_extent(struct inode *inode, ext4_lblk_t lblk, | |
698 | ext4_lblk_t len, ext4_fsblk_t pblk, | |
699 | unsigned int status) | |
700 | { | |
701 | struct extent_status *es; | |
702 | struct extent_status newes; | |
703 | ext4_lblk_t end = lblk + len - 1; | |
704 | ||
705 | newes.es_lblk = lblk; | |
706 | newes.es_len = len; | |
9a6633b1 | 707 | ext4_es_store_pblock_status(&newes, pblk, status); |
107a7bd3 TT |
708 | trace_ext4_es_cache_extent(inode, &newes); |
709 | ||
710 | if (!len) | |
711 | return; | |
712 | ||
713 | BUG_ON(end < lblk); | |
714 | ||
715 | write_lock(&EXT4_I(inode)->i_es_lock); | |
716 | ||
717 | es = __es_tree_search(&EXT4_I(inode)->i_es_tree.root, lblk); | |
7869a4a6 TT |
718 | if (!es || es->es_lblk > end) |
719 | __es_insert_extent(inode, &newes); | |
107a7bd3 TT |
720 | write_unlock(&EXT4_I(inode)->i_es_lock); |
721 | } | |
722 | ||
d100eef2 ZL |
723 | /* |
724 | * ext4_es_lookup_extent() looks up an extent in extent status tree. | |
725 | * | |
726 | * ext4_es_lookup_extent is called by ext4_map_blocks/ext4_da_map_blocks. | |
727 | * | |
728 | * Return: 1 on found, 0 on not | |
729 | */ | |
730 | int ext4_es_lookup_extent(struct inode *inode, ext4_lblk_t lblk, | |
731 | struct extent_status *es) | |
732 | { | |
733 | struct ext4_es_tree *tree; | |
734 | struct extent_status *es1 = NULL; | |
735 | struct rb_node *node; | |
736 | int found = 0; | |
737 | ||
738 | trace_ext4_es_lookup_extent_enter(inode, lblk); | |
739 | es_debug("lookup extent in block %u\n", lblk); | |
740 | ||
741 | tree = &EXT4_I(inode)->i_es_tree; | |
742 | read_lock(&EXT4_I(inode)->i_es_lock); | |
743 | ||
744 | /* find extent in cache firstly */ | |
745 | es->es_lblk = es->es_len = es->es_pblk = 0; | |
746 | if (tree->cache_es) { | |
747 | es1 = tree->cache_es; | |
748 | if (in_range(lblk, es1->es_lblk, es1->es_len)) { | |
749 | es_debug("%u cached by [%u/%u)\n", | |
750 | lblk, es1->es_lblk, es1->es_len); | |
751 | found = 1; | |
752 | goto out; | |
753 | } | |
754 | } | |
755 | ||
756 | node = tree->root.rb_node; | |
757 | while (node) { | |
758 | es1 = rb_entry(node, struct extent_status, rb_node); | |
759 | if (lblk < es1->es_lblk) | |
760 | node = node->rb_left; | |
761 | else if (lblk > ext4_es_end(es1)) | |
762 | node = node->rb_right; | |
763 | else { | |
764 | found = 1; | |
765 | break; | |
766 | } | |
767 | } | |
768 | ||
769 | out: | |
770 | if (found) { | |
771 | BUG_ON(!es1); | |
772 | es->es_lblk = es1->es_lblk; | |
773 | es->es_len = es1->es_len; | |
774 | es->es_pblk = es1->es_pblk; | |
775 | } | |
776 | ||
777 | read_unlock(&EXT4_I(inode)->i_es_lock); | |
778 | ||
779 | trace_ext4_es_lookup_extent_exit(inode, es, found); | |
780 | return found; | |
781 | } | |
782 | ||
bdedbb7b ZL |
783 | static int __es_remove_extent(struct inode *inode, ext4_lblk_t lblk, |
784 | ext4_lblk_t end) | |
654598be | 785 | { |
bdedbb7b | 786 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be | 787 | struct rb_node *node; |
654598be ZL |
788 | struct extent_status *es; |
789 | struct extent_status orig_es; | |
06b0c886 | 790 | ext4_lblk_t len1, len2; |
fdc0212e | 791 | ext4_fsblk_t block; |
e15f742c | 792 | int err; |
654598be | 793 | |
e15f742c TT |
794 | retry: |
795 | err = 0; | |
06b0c886 | 796 | es = __es_tree_search(&tree->root, lblk); |
654598be ZL |
797 | if (!es) |
798 | goto out; | |
06b0c886 | 799 | if (es->es_lblk > end) |
654598be ZL |
800 | goto out; |
801 | ||
802 | /* Simply invalidate cache_es. */ | |
803 | tree->cache_es = NULL; | |
804 | ||
06b0c886 ZL |
805 | orig_es.es_lblk = es->es_lblk; |
806 | orig_es.es_len = es->es_len; | |
fdc0212e ZL |
807 | orig_es.es_pblk = es->es_pblk; |
808 | ||
06b0c886 ZL |
809 | len1 = lblk > es->es_lblk ? lblk - es->es_lblk : 0; |
810 | len2 = ext4_es_end(es) > end ? ext4_es_end(es) - end : 0; | |
654598be | 811 | if (len1 > 0) |
06b0c886 | 812 | es->es_len = len1; |
654598be ZL |
813 | if (len2 > 0) { |
814 | if (len1 > 0) { | |
06b0c886 ZL |
815 | struct extent_status newes; |
816 | ||
817 | newes.es_lblk = end + 1; | |
818 | newes.es_len = len2; | |
666525df | 819 | block = 0x7FDEADBEEFULL; |
fdc0212e | 820 | if (ext4_es_is_written(&orig_es) || |
9a6633b1 | 821 | ext4_es_is_unwritten(&orig_es)) |
fdc0212e ZL |
822 | block = ext4_es_pblock(&orig_es) + |
823 | orig_es.es_len - len2; | |
9a6633b1 TT |
824 | ext4_es_store_pblock_status(&newes, block, |
825 | ext4_es_status(&orig_es)); | |
bdedbb7b | 826 | err = __es_insert_extent(inode, &newes); |
654598be | 827 | if (err) { |
06b0c886 ZL |
828 | es->es_lblk = orig_es.es_lblk; |
829 | es->es_len = orig_es.es_len; | |
e15f742c TT |
830 | if ((err == -ENOMEM) && |
831 | __ext4_es_shrink(EXT4_SB(inode->i_sb), 1, | |
832 | EXT4_I(inode))) | |
833 | goto retry; | |
654598be ZL |
834 | goto out; |
835 | } | |
836 | } else { | |
06b0c886 ZL |
837 | es->es_lblk = end + 1; |
838 | es->es_len = len2; | |
fdc0212e ZL |
839 | if (ext4_es_is_written(es) || |
840 | ext4_es_is_unwritten(es)) { | |
841 | block = orig_es.es_pblk + orig_es.es_len - len2; | |
842 | ext4_es_store_pblock(es, block); | |
843 | } | |
654598be ZL |
844 | } |
845 | goto out; | |
846 | } | |
847 | ||
848 | if (len1 > 0) { | |
849 | node = rb_next(&es->rb_node); | |
850 | if (node) | |
851 | es = rb_entry(node, struct extent_status, rb_node); | |
852 | else | |
853 | es = NULL; | |
854 | } | |
855 | ||
06b0c886 | 856 | while (es && ext4_es_end(es) <= end) { |
654598be ZL |
857 | node = rb_next(&es->rb_node); |
858 | rb_erase(&es->rb_node, &tree->root); | |
bdedbb7b | 859 | ext4_es_free_extent(inode, es); |
654598be ZL |
860 | if (!node) { |
861 | es = NULL; | |
862 | break; | |
863 | } | |
864 | es = rb_entry(node, struct extent_status, rb_node); | |
865 | } | |
866 | ||
06b0c886 | 867 | if (es && es->es_lblk < end + 1) { |
fdc0212e ZL |
868 | ext4_lblk_t orig_len = es->es_len; |
869 | ||
06b0c886 ZL |
870 | len1 = ext4_es_end(es) - end; |
871 | es->es_lblk = end + 1; | |
872 | es->es_len = len1; | |
fdc0212e ZL |
873 | if (ext4_es_is_written(es) || ext4_es_is_unwritten(es)) { |
874 | block = es->es_pblk + orig_len - len1; | |
875 | ext4_es_store_pblock(es, block); | |
876 | } | |
654598be ZL |
877 | } |
878 | ||
879 | out: | |
06b0c886 ZL |
880 | return err; |
881 | } | |
882 | ||
883 | /* | |
884 | * ext4_es_remove_extent() removes a space from a extent status tree. | |
885 | * | |
886 | * Return 0 on success, error code on failure. | |
887 | */ | |
888 | int ext4_es_remove_extent(struct inode *inode, ext4_lblk_t lblk, | |
889 | ext4_lblk_t len) | |
890 | { | |
06b0c886 ZL |
891 | ext4_lblk_t end; |
892 | int err = 0; | |
893 | ||
894 | trace_ext4_es_remove_extent(inode, lblk, len); | |
895 | es_debug("remove [%u/%u) from extent status tree of inode %lu\n", | |
896 | lblk, len, inode->i_ino); | |
897 | ||
d4381472 EG |
898 | if (!len) |
899 | return err; | |
900 | ||
06b0c886 ZL |
901 | end = lblk + len - 1; |
902 | BUG_ON(end < lblk); | |
903 | ||
06b0c886 | 904 | write_lock(&EXT4_I(inode)->i_es_lock); |
bdedbb7b | 905 | err = __es_remove_extent(inode, lblk, end); |
654598be ZL |
906 | write_unlock(&EXT4_I(inode)->i_es_lock); |
907 | ext4_es_print_tree(inode); | |
908 | return err; | |
909 | } | |
74cd15cd | 910 | |
d3922a77 ZL |
911 | static int ext4_inode_touch_time_cmp(void *priv, struct list_head *a, |
912 | struct list_head *b) | |
913 | { | |
914 | struct ext4_inode_info *eia, *eib; | |
915 | eia = list_entry(a, struct ext4_inode_info, i_es_lru); | |
916 | eib = list_entry(b, struct ext4_inode_info, i_es_lru); | |
917 | ||
7869a4a6 TT |
918 | if (ext4_test_inode_state(&eia->vfs_inode, EXT4_STATE_EXT_PRECACHED) && |
919 | !ext4_test_inode_state(&eib->vfs_inode, EXT4_STATE_EXT_PRECACHED)) | |
920 | return 1; | |
921 | if (!ext4_test_inode_state(&eia->vfs_inode, EXT4_STATE_EXT_PRECACHED) && | |
922 | ext4_test_inode_state(&eib->vfs_inode, EXT4_STATE_EXT_PRECACHED)) | |
923 | return -1; | |
d3922a77 ZL |
924 | if (eia->i_touch_when == eib->i_touch_when) |
925 | return 0; | |
926 | if (time_after(eia->i_touch_when, eib->i_touch_when)) | |
927 | return 1; | |
928 | else | |
929 | return -1; | |
930 | } | |
931 | ||
e15f742c TT |
932 | static int __ext4_es_shrink(struct ext4_sb_info *sbi, int nr_to_scan, |
933 | struct ext4_inode_info *locked_ei) | |
74cd15cd | 934 | { |
74cd15cd | 935 | struct ext4_inode_info *ei; |
d3922a77 | 936 | struct list_head *cur, *tmp; |
7869a4a6 | 937 | LIST_HEAD(skipped); |
1ab6c499 | 938 | int nr_shrunk = 0; |
7869a4a6 | 939 | int retried = 0, skip_precached = 1, nr_skipped = 0; |
74cd15cd | 940 | |
74cd15cd | 941 | spin_lock(&sbi->s_es_lru_lock); |
d3922a77 | 942 | |
7869a4a6 | 943 | retry: |
74cd15cd | 944 | list_for_each_safe(cur, tmp, &sbi->s_es_lru) { |
1ab6c499 DC |
945 | int shrunk; |
946 | ||
d3922a77 ZL |
947 | /* |
948 | * If we have already reclaimed all extents from extent | |
949 | * status tree, just stop the loop immediately. | |
950 | */ | |
951 | if (percpu_counter_read_positive(&sbi->s_extent_cache_cnt) == 0) | |
952 | break; | |
74cd15cd ZL |
953 | |
954 | ei = list_entry(cur, struct ext4_inode_info, i_es_lru); | |
955 | ||
7869a4a6 TT |
956 | /* |
957 | * Skip the inode that is newer than the last_sorted | |
958 | * time. Normally we try hard to avoid shrinking | |
959 | * precached inodes, but we will as a last resort. | |
960 | */ | |
961 | if ((sbi->s_es_last_sorted < ei->i_touch_when) || | |
962 | (skip_precached && ext4_test_inode_state(&ei->vfs_inode, | |
963 | EXT4_STATE_EXT_PRECACHED))) { | |
964 | nr_skipped++; | |
965 | list_move_tail(cur, &skipped); | |
74cd15cd ZL |
966 | continue; |
967 | } | |
d3922a77 | 968 | |
3f1f9b85 TT |
969 | if (ei->i_es_lru_nr == 0 || ei == locked_ei || |
970 | !write_trylock(&ei->i_es_lock)) | |
d3922a77 | 971 | continue; |
74cd15cd | 972 | |
1ab6c499 | 973 | shrunk = __es_try_to_reclaim_extents(ei, nr_to_scan); |
d3922a77 ZL |
974 | if (ei->i_es_lru_nr == 0) |
975 | list_del_init(&ei->i_es_lru); | |
74cd15cd ZL |
976 | write_unlock(&ei->i_es_lock); |
977 | ||
1ab6c499 DC |
978 | nr_shrunk += shrunk; |
979 | nr_to_scan -= shrunk; | |
74cd15cd ZL |
980 | if (nr_to_scan == 0) |
981 | break; | |
982 | } | |
d3922a77 ZL |
983 | |
984 | /* Move the newer inodes into the tail of the LRU list. */ | |
7869a4a6 TT |
985 | list_splice_tail(&skipped, &sbi->s_es_lru); |
986 | INIT_LIST_HEAD(&skipped); | |
987 | ||
988 | /* | |
989 | * If we skipped any inodes, and we weren't able to make any | |
990 | * forward progress, sort the list and try again. | |
991 | */ | |
992 | if ((nr_shrunk == 0) && nr_skipped && !retried) { | |
993 | retried++; | |
994 | list_sort(NULL, &sbi->s_es_lru, ext4_inode_touch_time_cmp); | |
995 | sbi->s_es_last_sorted = jiffies; | |
996 | ei = list_first_entry(&sbi->s_es_lru, struct ext4_inode_info, | |
997 | i_es_lru); | |
998 | /* | |
999 | * If there are no non-precached inodes left on the | |
1000 | * list, start releasing precached extents. | |
1001 | */ | |
1002 | if (ext4_test_inode_state(&ei->vfs_inode, | |
1003 | EXT4_STATE_EXT_PRECACHED)) | |
1004 | skip_precached = 0; | |
1005 | goto retry; | |
1006 | } | |
1007 | ||
74cd15cd | 1008 | spin_unlock(&sbi->s_es_lru_lock); |
74cd15cd | 1009 | |
e15f742c | 1010 | if (locked_ei && nr_shrunk == 0) |
7869a4a6 | 1011 | nr_shrunk = __es_try_to_reclaim_extents(locked_ei, nr_to_scan); |
e15f742c TT |
1012 | |
1013 | return nr_shrunk; | |
1014 | } | |
1015 | ||
1ab6c499 DC |
1016 | static unsigned long ext4_es_count(struct shrinker *shrink, |
1017 | struct shrink_control *sc) | |
1018 | { | |
1019 | unsigned long nr; | |
1020 | struct ext4_sb_info *sbi; | |
1021 | ||
1022 | sbi = container_of(shrink, struct ext4_sb_info, s_es_shrinker); | |
1023 | nr = percpu_counter_read_positive(&sbi->s_extent_cache_cnt); | |
1024 | trace_ext4_es_shrink_enter(sbi->s_sb, sc->nr_to_scan, nr); | |
1025 | return nr; | |
1026 | } | |
1027 | ||
1028 | static unsigned long ext4_es_scan(struct shrinker *shrink, | |
1029 | struct shrink_control *sc) | |
e15f742c TT |
1030 | { |
1031 | struct ext4_sb_info *sbi = container_of(shrink, | |
1032 | struct ext4_sb_info, s_es_shrinker); | |
1033 | int nr_to_scan = sc->nr_to_scan; | |
1034 | int ret, nr_shrunk; | |
1035 | ||
1036 | ret = percpu_counter_read_positive(&sbi->s_extent_cache_cnt); | |
1037 | trace_ext4_es_shrink_enter(sbi->s_sb, nr_to_scan, ret); | |
1038 | ||
1039 | if (!nr_to_scan) | |
1040 | return ret; | |
1041 | ||
1042 | nr_shrunk = __ext4_es_shrink(sbi, nr_to_scan, NULL); | |
1043 | ||
24630774 | 1044 | trace_ext4_es_shrink_exit(sbi->s_sb, nr_shrunk, ret); |
1ab6c499 | 1045 | return nr_shrunk; |
74cd15cd ZL |
1046 | } |
1047 | ||
d3922a77 | 1048 | void ext4_es_register_shrinker(struct ext4_sb_info *sbi) |
74cd15cd | 1049 | { |
74cd15cd ZL |
1050 | INIT_LIST_HEAD(&sbi->s_es_lru); |
1051 | spin_lock_init(&sbi->s_es_lru_lock); | |
d3922a77 | 1052 | sbi->s_es_last_sorted = 0; |
1ab6c499 DC |
1053 | sbi->s_es_shrinker.scan_objects = ext4_es_scan; |
1054 | sbi->s_es_shrinker.count_objects = ext4_es_count; | |
74cd15cd ZL |
1055 | sbi->s_es_shrinker.seeks = DEFAULT_SEEKS; |
1056 | register_shrinker(&sbi->s_es_shrinker); | |
1057 | } | |
1058 | ||
d3922a77 | 1059 | void ext4_es_unregister_shrinker(struct ext4_sb_info *sbi) |
74cd15cd | 1060 | { |
d3922a77 | 1061 | unregister_shrinker(&sbi->s_es_shrinker); |
74cd15cd ZL |
1062 | } |
1063 | ||
1064 | void ext4_es_lru_add(struct inode *inode) | |
1065 | { | |
1066 | struct ext4_inode_info *ei = EXT4_I(inode); | |
1067 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1068 | ||
d3922a77 ZL |
1069 | ei->i_touch_when = jiffies; |
1070 | ||
1071 | if (!list_empty(&ei->i_es_lru)) | |
1072 | return; | |
1073 | ||
74cd15cd ZL |
1074 | spin_lock(&sbi->s_es_lru_lock); |
1075 | if (list_empty(&ei->i_es_lru)) | |
1076 | list_add_tail(&ei->i_es_lru, &sbi->s_es_lru); | |
74cd15cd ZL |
1077 | spin_unlock(&sbi->s_es_lru_lock); |
1078 | } | |
1079 | ||
1080 | void ext4_es_lru_del(struct inode *inode) | |
1081 | { | |
1082 | struct ext4_inode_info *ei = EXT4_I(inode); | |
1083 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1084 | ||
1085 | spin_lock(&sbi->s_es_lru_lock); | |
1086 | if (!list_empty(&ei->i_es_lru)) | |
1087 | list_del_init(&ei->i_es_lru); | |
1088 | spin_unlock(&sbi->s_es_lru_lock); | |
1089 | } | |
1090 | ||
74cd15cd ZL |
1091 | static int __es_try_to_reclaim_extents(struct ext4_inode_info *ei, |
1092 | int nr_to_scan) | |
1093 | { | |
1094 | struct inode *inode = &ei->vfs_inode; | |
1095 | struct ext4_es_tree *tree = &ei->i_es_tree; | |
1096 | struct rb_node *node; | |
1097 | struct extent_status *es; | |
1ab6c499 | 1098 | unsigned long nr_shrunk = 0; |
7869a4a6 TT |
1099 | static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL, |
1100 | DEFAULT_RATELIMIT_BURST); | |
74cd15cd ZL |
1101 | |
1102 | if (ei->i_es_lru_nr == 0) | |
1103 | return 0; | |
1104 | ||
7869a4a6 TT |
1105 | if (ext4_test_inode_state(inode, EXT4_STATE_EXT_PRECACHED) && |
1106 | __ratelimit(&_rs)) | |
1107 | ext4_warning(inode->i_sb, "forced shrink of precached extents"); | |
1108 | ||
74cd15cd ZL |
1109 | node = rb_first(&tree->root); |
1110 | while (node != NULL) { | |
1111 | es = rb_entry(node, struct extent_status, rb_node); | |
1112 | node = rb_next(&es->rb_node); | |
1113 | /* | |
1114 | * We can't reclaim delayed extent from status tree because | |
1115 | * fiemap, bigallic, and seek_data/hole need to use it. | |
1116 | */ | |
1117 | if (!ext4_es_is_delayed(es)) { | |
1118 | rb_erase(&es->rb_node, &tree->root); | |
1119 | ext4_es_free_extent(inode, es); | |
1120 | nr_shrunk++; | |
1121 | if (--nr_to_scan == 0) | |
1122 | break; | |
1123 | } | |
1124 | } | |
1125 | tree->cache_es = NULL; | |
1126 | return nr_shrunk; | |
1127 | } |