]>
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> |
eb68d0e2 ZL |
14 | #include <linux/proc_fs.h> |
15 | #include <linux/seq_file.h> | |
654598be ZL |
16 | #include "ext4.h" |
17 | #include "extents_status.h" | |
654598be | 18 | |
992e9fdd ZL |
19 | #include <trace/events/ext4.h> |
20 | ||
654598be ZL |
21 | /* |
22 | * According to previous discussion in Ext4 Developer Workshop, we | |
23 | * will introduce a new structure called io tree to track all extent | |
24 | * status in order to solve some problems that we have met | |
25 | * (e.g. Reservation space warning), and provide extent-level locking. | |
26 | * Delay extent tree is the first step to achieve this goal. It is | |
27 | * original built by Yongqiang Yang. At that time it is called delay | |
06b0c886 | 28 | * extent tree, whose goal is only track delayed extents in memory to |
654598be ZL |
29 | * simplify the implementation of fiemap and bigalloc, and introduce |
30 | * lseek SEEK_DATA/SEEK_HOLE support. That is why it is still called | |
06b0c886 ZL |
31 | * delay extent tree at the first commit. But for better understand |
32 | * what it does, it has been rename to extent status tree. | |
654598be | 33 | * |
06b0c886 ZL |
34 | * Step1: |
35 | * Currently the first step has been done. All delayed extents are | |
36 | * tracked in the tree. It maintains the delayed extent when a delayed | |
37 | * allocation is issued, and the delayed extent is written out or | |
654598be ZL |
38 | * invalidated. Therefore the implementation of fiemap and bigalloc |
39 | * are simplified, and SEEK_DATA/SEEK_HOLE are introduced. | |
40 | * | |
41 | * The following comment describes the implemenmtation of extent | |
42 | * status tree and future works. | |
06b0c886 ZL |
43 | * |
44 | * Step2: | |
45 | * In this step all extent status are tracked by extent status tree. | |
46 | * Thus, we can first try to lookup a block mapping in this tree before | |
47 | * finding it in extent tree. Hence, single extent cache can be removed | |
48 | * because extent status tree can do a better job. Extents in status | |
49 | * tree are loaded on-demand. Therefore, the extent status tree may not | |
50 | * contain all of the extents in a file. Meanwhile we define a shrinker | |
51 | * to reclaim memory from extent status tree because fragmented extent | |
52 | * tree will make status tree cost too much memory. written/unwritten/- | |
53 | * hole extents in the tree will be reclaimed by this shrinker when we | |
54 | * are under high memory pressure. Delayed extents will not be | |
55 | * reclimed because fiemap, bigalloc, and seek_data/hole need it. | |
654598be ZL |
56 | */ |
57 | ||
58 | /* | |
06b0c886 | 59 | * Extent status tree implementation for ext4. |
654598be ZL |
60 | * |
61 | * | |
62 | * ========================================================================== | |
06b0c886 | 63 | * Extent status tree tracks all extent status. |
654598be | 64 | * |
06b0c886 | 65 | * 1. Why we need to implement extent status tree? |
654598be | 66 | * |
06b0c886 | 67 | * Without extent status tree, ext4 identifies a delayed extent by looking |
654598be ZL |
68 | * up page cache, this has several deficiencies - complicated, buggy, |
69 | * and inefficient code. | |
70 | * | |
06b0c886 ZL |
71 | * FIEMAP, SEEK_HOLE/DATA, bigalloc, and writeout all need to know if a |
72 | * block or a range of blocks are belonged to a delayed extent. | |
654598be | 73 | * |
06b0c886 | 74 | * Let us have a look at how they do without extent status tree. |
654598be ZL |
75 | * -- FIEMAP |
76 | * FIEMAP looks up page cache to identify delayed allocations from holes. | |
77 | * | |
78 | * -- SEEK_HOLE/DATA | |
79 | * SEEK_HOLE/DATA has the same problem as FIEMAP. | |
80 | * | |
81 | * -- bigalloc | |
82 | * bigalloc looks up page cache to figure out if a block is | |
83 | * already under delayed allocation or not to determine whether | |
84 | * quota reserving is needed for the cluster. | |
85 | * | |
654598be ZL |
86 | * -- writeout |
87 | * Writeout looks up whole page cache to see if a buffer is | |
88 | * mapped, If there are not very many delayed buffers, then it is | |
89 | * time comsuming. | |
90 | * | |
06b0c886 | 91 | * With extent status tree implementation, FIEMAP, SEEK_HOLE/DATA, |
654598be ZL |
92 | * bigalloc and writeout can figure out if a block or a range of |
93 | * blocks is under delayed allocation(belonged to a delayed extent) or | |
06b0c886 | 94 | * not by searching the extent tree. |
654598be ZL |
95 | * |
96 | * | |
97 | * ========================================================================== | |
06b0c886 ZL |
98 | * 2. Ext4 extent status tree impelmentation |
99 | * | |
100 | * -- extent | |
101 | * A extent is a range of blocks which are contiguous logically and | |
102 | * physically. Unlike extent in extent tree, this extent in ext4 is | |
103 | * a in-memory struct, there is no corresponding on-disk data. There | |
104 | * is no limit on length of extent, so an extent can contain as many | |
105 | * blocks as they are contiguous logically and physically. | |
654598be | 106 | * |
06b0c886 ZL |
107 | * -- extent status tree |
108 | * Every inode has an extent status tree and all allocation blocks | |
109 | * are added to the tree with different status. The extent in the | |
110 | * tree are ordered by logical block no. | |
654598be | 111 | * |
06b0c886 ZL |
112 | * -- operations on a extent status tree |
113 | * There are three important operations on a delayed extent tree: find | |
114 | * next extent, adding a extent(a range of blocks) and removing a extent. | |
654598be | 115 | * |
06b0c886 ZL |
116 | * -- race on a extent status tree |
117 | * Extent status tree is protected by inode->i_es_lock. | |
654598be | 118 | * |
06b0c886 ZL |
119 | * -- memory consumption |
120 | * Fragmented extent tree will make extent status tree cost too much | |
121 | * memory. Hence, we will reclaim written/unwritten/hole extents from | |
122 | * the tree under a heavy memory pressure. | |
654598be ZL |
123 | * |
124 | * | |
125 | * ========================================================================== | |
06b0c886 ZL |
126 | * 3. Performance analysis |
127 | * | |
654598be ZL |
128 | * -- overhead |
129 | * 1. There is a cache extent for write access, so if writes are | |
130 | * not very random, adding space operaions are in O(1) time. | |
131 | * | |
132 | * -- gain | |
133 | * 2. Code is much simpler, more readable, more maintainable and | |
134 | * more efficient. | |
135 | * | |
136 | * | |
137 | * ========================================================================== | |
138 | * 4. TODO list | |
654598be | 139 | * |
06b0c886 | 140 | * -- Refactor delayed space reservation |
654598be ZL |
141 | * |
142 | * -- Extent-level locking | |
143 | */ | |
144 | ||
145 | static struct kmem_cache *ext4_es_cachep; | |
146 | ||
bdedbb7b ZL |
147 | static int __es_insert_extent(struct inode *inode, struct extent_status *newes); |
148 | static int __es_remove_extent(struct inode *inode, ext4_lblk_t lblk, | |
06b0c886 | 149 | ext4_lblk_t end); |
74cd15cd ZL |
150 | static int __es_try_to_reclaim_extents(struct ext4_inode_info *ei, |
151 | int nr_to_scan); | |
e15f742c TT |
152 | static int __ext4_es_shrink(struct ext4_sb_info *sbi, int nr_to_scan, |
153 | struct ext4_inode_info *locked_ei); | |
06b0c886 | 154 | |
654598be ZL |
155 | int __init ext4_init_es(void) |
156 | { | |
24630774 TT |
157 | ext4_es_cachep = kmem_cache_create("ext4_extent_status", |
158 | sizeof(struct extent_status), | |
159 | 0, (SLAB_RECLAIM_ACCOUNT), NULL); | |
654598be ZL |
160 | if (ext4_es_cachep == NULL) |
161 | return -ENOMEM; | |
162 | return 0; | |
163 | } | |
164 | ||
165 | void ext4_exit_es(void) | |
166 | { | |
167 | if (ext4_es_cachep) | |
168 | kmem_cache_destroy(ext4_es_cachep); | |
169 | } | |
170 | ||
171 | void ext4_es_init_tree(struct ext4_es_tree *tree) | |
172 | { | |
173 | tree->root = RB_ROOT; | |
174 | tree->cache_es = NULL; | |
175 | } | |
176 | ||
177 | #ifdef ES_DEBUG__ | |
178 | static void ext4_es_print_tree(struct inode *inode) | |
179 | { | |
180 | struct ext4_es_tree *tree; | |
181 | struct rb_node *node; | |
182 | ||
183 | printk(KERN_DEBUG "status extents for inode %lu:", inode->i_ino); | |
184 | tree = &EXT4_I(inode)->i_es_tree; | |
185 | node = rb_first(&tree->root); | |
186 | while (node) { | |
187 | struct extent_status *es; | |
188 | es = rb_entry(node, struct extent_status, rb_node); | |
ce140cdd | 189 | printk(KERN_DEBUG " [%u/%u) %llu %x", |
fdc0212e ZL |
190 | es->es_lblk, es->es_len, |
191 | ext4_es_pblock(es), ext4_es_status(es)); | |
654598be ZL |
192 | node = rb_next(node); |
193 | } | |
194 | printk(KERN_DEBUG "\n"); | |
195 | } | |
196 | #else | |
197 | #define ext4_es_print_tree(inode) | |
198 | #endif | |
199 | ||
06b0c886 | 200 | static inline ext4_lblk_t ext4_es_end(struct extent_status *es) |
654598be | 201 | { |
06b0c886 ZL |
202 | BUG_ON(es->es_lblk + es->es_len < es->es_lblk); |
203 | return es->es_lblk + es->es_len - 1; | |
654598be ZL |
204 | } |
205 | ||
206 | /* | |
207 | * search through the tree for an delayed extent with a given offset. If | |
208 | * it can't be found, try to find next extent. | |
209 | */ | |
210 | static struct extent_status *__es_tree_search(struct rb_root *root, | |
06b0c886 | 211 | ext4_lblk_t lblk) |
654598be ZL |
212 | { |
213 | struct rb_node *node = root->rb_node; | |
214 | struct extent_status *es = NULL; | |
215 | ||
216 | while (node) { | |
217 | es = rb_entry(node, struct extent_status, rb_node); | |
06b0c886 | 218 | if (lblk < es->es_lblk) |
654598be | 219 | node = node->rb_left; |
06b0c886 | 220 | else if (lblk > ext4_es_end(es)) |
654598be ZL |
221 | node = node->rb_right; |
222 | else | |
223 | return es; | |
224 | } | |
225 | ||
06b0c886 | 226 | if (es && lblk < es->es_lblk) |
654598be ZL |
227 | return es; |
228 | ||
06b0c886 | 229 | if (es && lblk > ext4_es_end(es)) { |
654598be ZL |
230 | node = rb_next(&es->rb_node); |
231 | return node ? rb_entry(node, struct extent_status, rb_node) : | |
232 | NULL; | |
233 | } | |
234 | ||
235 | return NULL; | |
236 | } | |
237 | ||
238 | /* | |
e30b5dca YZ |
239 | * ext4_es_find_delayed_extent_range: find the 1st delayed extent covering |
240 | * @es->lblk if it exists, otherwise, the next extent after @es->lblk. | |
654598be ZL |
241 | * |
242 | * @inode: the inode which owns delayed extents | |
be401363 | 243 | * @lblk: the offset where we start to search |
e30b5dca | 244 | * @end: the offset where we stop to search |
654598be | 245 | * @es: delayed extent that we found |
654598be | 246 | */ |
e30b5dca YZ |
247 | void ext4_es_find_delayed_extent_range(struct inode *inode, |
248 | ext4_lblk_t lblk, ext4_lblk_t end, | |
be401363 | 249 | struct extent_status *es) |
654598be ZL |
250 | { |
251 | struct ext4_es_tree *tree = NULL; | |
252 | struct extent_status *es1 = NULL; | |
253 | struct rb_node *node; | |
654598be | 254 | |
be401363 | 255 | BUG_ON(es == NULL); |
e30b5dca YZ |
256 | BUG_ON(end < lblk); |
257 | trace_ext4_es_find_delayed_extent_range_enter(inode, lblk); | |
992e9fdd | 258 | |
654598be ZL |
259 | read_lock(&EXT4_I(inode)->i_es_lock); |
260 | tree = &EXT4_I(inode)->i_es_tree; | |
261 | ||
fdc0212e | 262 | /* find extent in cache firstly */ |
be401363 | 263 | es->es_lblk = es->es_len = es->es_pblk = 0; |
654598be ZL |
264 | if (tree->cache_es) { |
265 | es1 = tree->cache_es; | |
be401363 | 266 | if (in_range(lblk, es1->es_lblk, es1->es_len)) { |
3be78c73 | 267 | es_debug("%u cached by [%u/%u) %llu %x\n", |
be401363 | 268 | lblk, es1->es_lblk, es1->es_len, |
fdc0212e | 269 | ext4_es_pblock(es1), ext4_es_status(es1)); |
654598be ZL |
270 | goto out; |
271 | } | |
272 | } | |
273 | ||
be401363 | 274 | es1 = __es_tree_search(&tree->root, lblk); |
654598be ZL |
275 | |
276 | out: | |
be401363 ZL |
277 | if (es1 && !ext4_es_is_delayed(es1)) { |
278 | while ((node = rb_next(&es1->rb_node)) != NULL) { | |
279 | es1 = rb_entry(node, struct extent_status, rb_node); | |
e30b5dca YZ |
280 | if (es1->es_lblk > end) { |
281 | es1 = NULL; | |
282 | break; | |
283 | } | |
be401363 ZL |
284 | if (ext4_es_is_delayed(es1)) |
285 | break; | |
286 | } | |
287 | } | |
288 | ||
289 | if (es1 && ext4_es_is_delayed(es1)) { | |
654598be | 290 | tree->cache_es = es1; |
06b0c886 ZL |
291 | es->es_lblk = es1->es_lblk; |
292 | es->es_len = es1->es_len; | |
fdc0212e | 293 | es->es_pblk = es1->es_pblk; |
654598be ZL |
294 | } |
295 | ||
296 | read_unlock(&EXT4_I(inode)->i_es_lock); | |
992e9fdd | 297 | |
e30b5dca | 298 | trace_ext4_es_find_delayed_extent_range_exit(inode, es); |
654598be ZL |
299 | } |
300 | ||
301 | static struct extent_status * | |
bdedbb7b ZL |
302 | ext4_es_alloc_extent(struct inode *inode, ext4_lblk_t lblk, ext4_lblk_t len, |
303 | ext4_fsblk_t pblk) | |
654598be ZL |
304 | { |
305 | struct extent_status *es; | |
306 | es = kmem_cache_alloc(ext4_es_cachep, GFP_ATOMIC); | |
307 | if (es == NULL) | |
308 | return NULL; | |
06b0c886 ZL |
309 | es->es_lblk = lblk; |
310 | es->es_len = len; | |
fdc0212e | 311 | es->es_pblk = pblk; |
74cd15cd ZL |
312 | |
313 | /* | |
314 | * We don't count delayed extent because we never try to reclaim them | |
315 | */ | |
24630774 | 316 | if (!ext4_es_is_delayed(es)) { |
74cd15cd | 317 | EXT4_I(inode)->i_es_lru_nr++; |
eb68d0e2 ZL |
318 | percpu_counter_inc(&EXT4_SB(inode->i_sb)-> |
319 | s_es_stats.es_stats_lru_cnt); | |
24630774 | 320 | } |
74cd15cd | 321 | |
eb68d0e2 ZL |
322 | EXT4_I(inode)->i_es_all_nr++; |
323 | percpu_counter_inc(&EXT4_SB(inode->i_sb)->s_es_stats.es_stats_all_cnt); | |
324 | ||
654598be ZL |
325 | return es; |
326 | } | |
327 | ||
bdedbb7b | 328 | static void ext4_es_free_extent(struct inode *inode, struct extent_status *es) |
654598be | 329 | { |
eb68d0e2 ZL |
330 | EXT4_I(inode)->i_es_all_nr--; |
331 | percpu_counter_dec(&EXT4_SB(inode->i_sb)->s_es_stats.es_stats_all_cnt); | |
332 | ||
74cd15cd ZL |
333 | /* Decrease the lru counter when this es is not delayed */ |
334 | if (!ext4_es_is_delayed(es)) { | |
335 | BUG_ON(EXT4_I(inode)->i_es_lru_nr == 0); | |
336 | EXT4_I(inode)->i_es_lru_nr--; | |
eb68d0e2 ZL |
337 | percpu_counter_dec(&EXT4_SB(inode->i_sb)-> |
338 | s_es_stats.es_stats_lru_cnt); | |
74cd15cd ZL |
339 | } |
340 | ||
654598be ZL |
341 | kmem_cache_free(ext4_es_cachep, es); |
342 | } | |
343 | ||
06b0c886 ZL |
344 | /* |
345 | * Check whether or not two extents can be merged | |
346 | * Condition: | |
347 | * - logical block number is contiguous | |
fdc0212e ZL |
348 | * - physical block number is contiguous |
349 | * - status is equal | |
06b0c886 ZL |
350 | */ |
351 | static int ext4_es_can_be_merged(struct extent_status *es1, | |
352 | struct extent_status *es2) | |
353 | { | |
bd384364 | 354 | if (ext4_es_status(es1) != ext4_es_status(es2)) |
06b0c886 ZL |
355 | return 0; |
356 | ||
0baaea64 LC |
357 | if (((__u64) es1->es_len) + es2->es_len > EXT_MAX_BLOCKS) { |
358 | pr_warn("ES assertion failed when merging extents. " | |
359 | "The sum of lengths of es1 (%d) and es2 (%d) " | |
360 | "is bigger than allowed file size (%d)\n", | |
361 | es1->es_len, es2->es_len, EXT_MAX_BLOCKS); | |
362 | WARN_ON(1); | |
fdc0212e | 363 | return 0; |
0baaea64 | 364 | } |
fdc0212e | 365 | |
bd384364 | 366 | if (((__u64) es1->es_lblk) + es1->es_len != es2->es_lblk) |
fdc0212e ZL |
367 | return 0; |
368 | ||
bd384364 ZL |
369 | if ((ext4_es_is_written(es1) || ext4_es_is_unwritten(es1)) && |
370 | (ext4_es_pblock(es1) + es1->es_len == ext4_es_pblock(es2))) | |
371 | return 1; | |
372 | ||
373 | if (ext4_es_is_hole(es1)) | |
374 | return 1; | |
375 | ||
376 | /* we need to check delayed extent is without unwritten status */ | |
377 | if (ext4_es_is_delayed(es1) && !ext4_es_is_unwritten(es1)) | |
378 | return 1; | |
379 | ||
380 | return 0; | |
06b0c886 ZL |
381 | } |
382 | ||
654598be | 383 | static struct extent_status * |
bdedbb7b | 384 | ext4_es_try_to_merge_left(struct inode *inode, struct extent_status *es) |
654598be | 385 | { |
bdedbb7b | 386 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be ZL |
387 | struct extent_status *es1; |
388 | struct rb_node *node; | |
389 | ||
390 | node = rb_prev(&es->rb_node); | |
391 | if (!node) | |
392 | return es; | |
393 | ||
394 | es1 = rb_entry(node, struct extent_status, rb_node); | |
06b0c886 ZL |
395 | if (ext4_es_can_be_merged(es1, es)) { |
396 | es1->es_len += es->es_len; | |
654598be | 397 | rb_erase(&es->rb_node, &tree->root); |
bdedbb7b | 398 | ext4_es_free_extent(inode, es); |
654598be ZL |
399 | es = es1; |
400 | } | |
401 | ||
402 | return es; | |
403 | } | |
404 | ||
405 | static struct extent_status * | |
bdedbb7b | 406 | ext4_es_try_to_merge_right(struct inode *inode, struct extent_status *es) |
654598be | 407 | { |
bdedbb7b | 408 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be ZL |
409 | struct extent_status *es1; |
410 | struct rb_node *node; | |
411 | ||
412 | node = rb_next(&es->rb_node); | |
413 | if (!node) | |
414 | return es; | |
415 | ||
416 | es1 = rb_entry(node, struct extent_status, rb_node); | |
06b0c886 ZL |
417 | if (ext4_es_can_be_merged(es, es1)) { |
418 | es->es_len += es1->es_len; | |
654598be | 419 | rb_erase(node, &tree->root); |
bdedbb7b | 420 | ext4_es_free_extent(inode, es1); |
654598be ZL |
421 | } |
422 | ||
423 | return es; | |
424 | } | |
425 | ||
921f266b | 426 | #ifdef ES_AGGRESSIVE_TEST |
d7b2a00c ZL |
427 | #include "ext4_extents.h" /* Needed when ES_AGGRESSIVE_TEST is defined */ |
428 | ||
921f266b DM |
429 | static void ext4_es_insert_extent_ext_check(struct inode *inode, |
430 | struct extent_status *es) | |
431 | { | |
432 | struct ext4_ext_path *path = NULL; | |
433 | struct ext4_extent *ex; | |
434 | ext4_lblk_t ee_block; | |
435 | ext4_fsblk_t ee_start; | |
436 | unsigned short ee_len; | |
437 | int depth, ee_status, es_status; | |
438 | ||
ed8a1a76 | 439 | path = ext4_find_extent(inode, es->es_lblk, NULL, EXT4_EX_NOCACHE); |
921f266b DM |
440 | if (IS_ERR(path)) |
441 | return; | |
442 | ||
443 | depth = ext_depth(inode); | |
444 | ex = path[depth].p_ext; | |
445 | ||
446 | if (ex) { | |
447 | ||
448 | ee_block = le32_to_cpu(ex->ee_block); | |
449 | ee_start = ext4_ext_pblock(ex); | |
450 | ee_len = ext4_ext_get_actual_len(ex); | |
451 | ||
556615dc | 452 | ee_status = ext4_ext_is_unwritten(ex) ? 1 : 0; |
921f266b DM |
453 | es_status = ext4_es_is_unwritten(es) ? 1 : 0; |
454 | ||
455 | /* | |
456 | * Make sure ex and es are not overlap when we try to insert | |
457 | * a delayed/hole extent. | |
458 | */ | |
459 | if (!ext4_es_is_written(es) && !ext4_es_is_unwritten(es)) { | |
460 | if (in_range(es->es_lblk, ee_block, ee_len)) { | |
bdafe42a | 461 | pr_warn("ES insert assertion failed for " |
921f266b DM |
462 | "inode: %lu we can find an extent " |
463 | "at block [%d/%d/%llu/%c], but we " | |
ce140cdd EW |
464 | "want to add a delayed/hole extent " |
465 | "[%d/%d/%llu/%x]\n", | |
921f266b DM |
466 | inode->i_ino, ee_block, ee_len, |
467 | ee_start, ee_status ? 'u' : 'w', | |
468 | es->es_lblk, es->es_len, | |
469 | ext4_es_pblock(es), ext4_es_status(es)); | |
470 | } | |
471 | goto out; | |
472 | } | |
473 | ||
474 | /* | |
475 | * We don't check ee_block == es->es_lblk, etc. because es | |
476 | * might be a part of whole extent, vice versa. | |
477 | */ | |
478 | if (es->es_lblk < ee_block || | |
479 | ext4_es_pblock(es) != ee_start + es->es_lblk - ee_block) { | |
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 | goto out; | |
487 | } | |
488 | ||
489 | if (ee_status ^ es_status) { | |
bdafe42a | 490 | pr_warn("ES insert assertion failed for inode: %lu " |
921f266b DM |
491 | "ex_status [%d/%d/%llu/%c] != " |
492 | "es_status [%d/%d/%llu/%c]\n", inode->i_ino, | |
493 | ee_block, ee_len, ee_start, | |
494 | ee_status ? 'u' : 'w', es->es_lblk, es->es_len, | |
495 | ext4_es_pblock(es), es_status ? 'u' : 'w'); | |
496 | } | |
497 | } else { | |
498 | /* | |
499 | * We can't find an extent on disk. So we need to make sure | |
500 | * that we don't want to add an written/unwritten extent. | |
501 | */ | |
502 | if (!ext4_es_is_delayed(es) && !ext4_es_is_hole(es)) { | |
bdafe42a | 503 | pr_warn("ES insert assertion failed for inode: %lu " |
921f266b | 504 | "can't find an extent at block %d but we want " |
ce140cdd EW |
505 | "to add a written/unwritten extent " |
506 | "[%d/%d/%llu/%x]\n", inode->i_ino, | |
921f266b DM |
507 | es->es_lblk, es->es_lblk, es->es_len, |
508 | ext4_es_pblock(es), ext4_es_status(es)); | |
509 | } | |
510 | } | |
511 | out: | |
b7ea89ad TT |
512 | ext4_ext_drop_refs(path); |
513 | kfree(path); | |
921f266b DM |
514 | } |
515 | ||
516 | static void ext4_es_insert_extent_ind_check(struct inode *inode, | |
517 | struct extent_status *es) | |
518 | { | |
519 | struct ext4_map_blocks map; | |
520 | int retval; | |
521 | ||
522 | /* | |
523 | * Here we call ext4_ind_map_blocks to lookup a block mapping because | |
524 | * 'Indirect' structure is defined in indirect.c. So we couldn't | |
525 | * access direct/indirect tree from outside. It is too dirty to define | |
526 | * this function in indirect.c file. | |
527 | */ | |
528 | ||
529 | map.m_lblk = es->es_lblk; | |
530 | map.m_len = es->es_len; | |
531 | ||
532 | retval = ext4_ind_map_blocks(NULL, inode, &map, 0); | |
533 | if (retval > 0) { | |
534 | if (ext4_es_is_delayed(es) || ext4_es_is_hole(es)) { | |
535 | /* | |
536 | * We want to add a delayed/hole extent but this | |
537 | * block has been allocated. | |
538 | */ | |
bdafe42a | 539 | pr_warn("ES insert assertion failed for inode: %lu " |
921f266b | 540 | "We can find blocks but we want to add a " |
ce140cdd | 541 | "delayed/hole extent [%d/%d/%llu/%x]\n", |
921f266b DM |
542 | inode->i_ino, es->es_lblk, es->es_len, |
543 | ext4_es_pblock(es), ext4_es_status(es)); | |
544 | return; | |
545 | } else if (ext4_es_is_written(es)) { | |
546 | if (retval != es->es_len) { | |
bdafe42a | 547 | pr_warn("ES insert assertion failed for " |
921f266b DM |
548 | "inode: %lu retval %d != es_len %d\n", |
549 | inode->i_ino, retval, es->es_len); | |
550 | return; | |
551 | } | |
552 | if (map.m_pblk != ext4_es_pblock(es)) { | |
bdafe42a | 553 | pr_warn("ES insert assertion failed for " |
921f266b DM |
554 | "inode: %lu m_pblk %llu != " |
555 | "es_pblk %llu\n", | |
556 | inode->i_ino, map.m_pblk, | |
557 | ext4_es_pblock(es)); | |
558 | return; | |
559 | } | |
560 | } else { | |
561 | /* | |
562 | * We don't need to check unwritten extent because | |
563 | * indirect-based file doesn't have it. | |
564 | */ | |
565 | BUG_ON(1); | |
566 | } | |
567 | } else if (retval == 0) { | |
568 | if (ext4_es_is_written(es)) { | |
bdafe42a | 569 | pr_warn("ES insert assertion failed for inode: %lu " |
921f266b | 570 | "We can't find the block but we want to add " |
ce140cdd | 571 | "a written extent [%d/%d/%llu/%x]\n", |
921f266b DM |
572 | inode->i_ino, es->es_lblk, es->es_len, |
573 | ext4_es_pblock(es), ext4_es_status(es)); | |
574 | return; | |
575 | } | |
576 | } | |
577 | } | |
578 | ||
579 | static inline void ext4_es_insert_extent_check(struct inode *inode, | |
580 | struct extent_status *es) | |
581 | { | |
582 | /* | |
583 | * We don't need to worry about the race condition because | |
584 | * caller takes i_data_sem locking. | |
585 | */ | |
586 | BUG_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem)); | |
587 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
588 | ext4_es_insert_extent_ext_check(inode, es); | |
589 | else | |
590 | ext4_es_insert_extent_ind_check(inode, es); | |
591 | } | |
592 | #else | |
593 | static inline void ext4_es_insert_extent_check(struct inode *inode, | |
594 | struct extent_status *es) | |
595 | { | |
596 | } | |
597 | #endif | |
598 | ||
bdedbb7b | 599 | static int __es_insert_extent(struct inode *inode, struct extent_status *newes) |
654598be | 600 | { |
bdedbb7b | 601 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be ZL |
602 | struct rb_node **p = &tree->root.rb_node; |
603 | struct rb_node *parent = NULL; | |
604 | struct extent_status *es; | |
654598be ZL |
605 | |
606 | while (*p) { | |
607 | parent = *p; | |
608 | es = rb_entry(parent, struct extent_status, rb_node); | |
609 | ||
06b0c886 ZL |
610 | if (newes->es_lblk < es->es_lblk) { |
611 | if (ext4_es_can_be_merged(newes, es)) { | |
612 | /* | |
613 | * Here we can modify es_lblk directly | |
614 | * because it isn't overlapped. | |
615 | */ | |
616 | es->es_lblk = newes->es_lblk; | |
617 | es->es_len += newes->es_len; | |
fdc0212e ZL |
618 | if (ext4_es_is_written(es) || |
619 | ext4_es_is_unwritten(es)) | |
620 | ext4_es_store_pblock(es, | |
621 | newes->es_pblk); | |
bdedbb7b | 622 | es = ext4_es_try_to_merge_left(inode, es); |
654598be ZL |
623 | goto out; |
624 | } | |
625 | p = &(*p)->rb_left; | |
06b0c886 ZL |
626 | } else if (newes->es_lblk > ext4_es_end(es)) { |
627 | if (ext4_es_can_be_merged(es, newes)) { | |
628 | es->es_len += newes->es_len; | |
bdedbb7b | 629 | es = ext4_es_try_to_merge_right(inode, es); |
654598be ZL |
630 | goto out; |
631 | } | |
632 | p = &(*p)->rb_right; | |
633 | } else { | |
06b0c886 ZL |
634 | BUG_ON(1); |
635 | return -EINVAL; | |
654598be ZL |
636 | } |
637 | } | |
638 | ||
bdedbb7b | 639 | es = ext4_es_alloc_extent(inode, newes->es_lblk, newes->es_len, |
fdc0212e | 640 | newes->es_pblk); |
654598be ZL |
641 | if (!es) |
642 | return -ENOMEM; | |
643 | rb_link_node(&es->rb_node, parent, p); | |
644 | rb_insert_color(&es->rb_node, &tree->root); | |
645 | ||
646 | out: | |
647 | tree->cache_es = es; | |
648 | return 0; | |
649 | } | |
650 | ||
651 | /* | |
bdafe42a TT |
652 | * ext4_es_insert_extent() adds information to an inode's extent |
653 | * status tree. | |
654598be ZL |
654 | * |
655 | * Return 0 on success, error code on failure. | |
656 | */ | |
06b0c886 | 657 | int ext4_es_insert_extent(struct inode *inode, ext4_lblk_t lblk, |
fdc0212e | 658 | ext4_lblk_t len, ext4_fsblk_t pblk, |
3be78c73 | 659 | unsigned int status) |
654598be | 660 | { |
06b0c886 ZL |
661 | struct extent_status newes; |
662 | ext4_lblk_t end = lblk + len - 1; | |
654598be ZL |
663 | int err = 0; |
664 | ||
3be78c73 | 665 | es_debug("add [%u/%u) %llu %x to extent status tree of inode %lu\n", |
fdc0212e | 666 | lblk, len, pblk, status, inode->i_ino); |
06b0c886 | 667 | |
d4381472 EG |
668 | if (!len) |
669 | return 0; | |
670 | ||
06b0c886 ZL |
671 | BUG_ON(end < lblk); |
672 | ||
673 | newes.es_lblk = lblk; | |
674 | newes.es_len = len; | |
9a6633b1 | 675 | ext4_es_store_pblock_status(&newes, pblk, status); |
fdc0212e | 676 | trace_ext4_es_insert_extent(inode, &newes); |
654598be | 677 | |
921f266b DM |
678 | ext4_es_insert_extent_check(inode, &newes); |
679 | ||
654598be | 680 | write_lock(&EXT4_I(inode)->i_es_lock); |
bdedbb7b | 681 | err = __es_remove_extent(inode, lblk, end); |
06b0c886 ZL |
682 | if (err != 0) |
683 | goto error; | |
e15f742c | 684 | retry: |
bdedbb7b | 685 | err = __es_insert_extent(inode, &newes); |
e15f742c TT |
686 | if (err == -ENOMEM && __ext4_es_shrink(EXT4_SB(inode->i_sb), 1, |
687 | EXT4_I(inode))) | |
688 | goto retry; | |
689 | if (err == -ENOMEM && !ext4_es_is_delayed(&newes)) | |
690 | err = 0; | |
06b0c886 ZL |
691 | |
692 | error: | |
654598be ZL |
693 | write_unlock(&EXT4_I(inode)->i_es_lock); |
694 | ||
695 | ext4_es_print_tree(inode); | |
696 | ||
697 | return err; | |
698 | } | |
699 | ||
107a7bd3 TT |
700 | /* |
701 | * ext4_es_cache_extent() inserts information into the extent status | |
702 | * tree if and only if there isn't information about the range in | |
703 | * question already. | |
704 | */ | |
705 | void ext4_es_cache_extent(struct inode *inode, ext4_lblk_t lblk, | |
706 | ext4_lblk_t len, ext4_fsblk_t pblk, | |
707 | unsigned int status) | |
708 | { | |
709 | struct extent_status *es; | |
710 | struct extent_status newes; | |
711 | ext4_lblk_t end = lblk + len - 1; | |
712 | ||
713 | newes.es_lblk = lblk; | |
714 | newes.es_len = len; | |
9a6633b1 | 715 | ext4_es_store_pblock_status(&newes, pblk, status); |
107a7bd3 TT |
716 | trace_ext4_es_cache_extent(inode, &newes); |
717 | ||
718 | if (!len) | |
719 | return; | |
720 | ||
721 | BUG_ON(end < lblk); | |
722 | ||
723 | write_lock(&EXT4_I(inode)->i_es_lock); | |
724 | ||
725 | es = __es_tree_search(&EXT4_I(inode)->i_es_tree.root, lblk); | |
7869a4a6 TT |
726 | if (!es || es->es_lblk > end) |
727 | __es_insert_extent(inode, &newes); | |
107a7bd3 TT |
728 | write_unlock(&EXT4_I(inode)->i_es_lock); |
729 | } | |
730 | ||
d100eef2 ZL |
731 | /* |
732 | * ext4_es_lookup_extent() looks up an extent in extent status tree. | |
733 | * | |
734 | * ext4_es_lookup_extent is called by ext4_map_blocks/ext4_da_map_blocks. | |
735 | * | |
736 | * Return: 1 on found, 0 on not | |
737 | */ | |
738 | int ext4_es_lookup_extent(struct inode *inode, ext4_lblk_t lblk, | |
739 | struct extent_status *es) | |
740 | { | |
741 | struct ext4_es_tree *tree; | |
eb68d0e2 | 742 | struct ext4_es_stats *stats; |
d100eef2 ZL |
743 | struct extent_status *es1 = NULL; |
744 | struct rb_node *node; | |
745 | int found = 0; | |
746 | ||
747 | trace_ext4_es_lookup_extent_enter(inode, lblk); | |
748 | es_debug("lookup extent in block %u\n", lblk); | |
749 | ||
750 | tree = &EXT4_I(inode)->i_es_tree; | |
751 | read_lock(&EXT4_I(inode)->i_es_lock); | |
752 | ||
753 | /* find extent in cache firstly */ | |
754 | es->es_lblk = es->es_len = es->es_pblk = 0; | |
755 | if (tree->cache_es) { | |
756 | es1 = tree->cache_es; | |
757 | if (in_range(lblk, es1->es_lblk, es1->es_len)) { | |
758 | es_debug("%u cached by [%u/%u)\n", | |
759 | lblk, es1->es_lblk, es1->es_len); | |
760 | found = 1; | |
761 | goto out; | |
762 | } | |
763 | } | |
764 | ||
765 | node = tree->root.rb_node; | |
766 | while (node) { | |
767 | es1 = rb_entry(node, struct extent_status, rb_node); | |
768 | if (lblk < es1->es_lblk) | |
769 | node = node->rb_left; | |
770 | else if (lblk > ext4_es_end(es1)) | |
771 | node = node->rb_right; | |
772 | else { | |
773 | found = 1; | |
774 | break; | |
775 | } | |
776 | } | |
777 | ||
778 | out: | |
eb68d0e2 | 779 | stats = &EXT4_SB(inode->i_sb)->s_es_stats; |
d100eef2 ZL |
780 | if (found) { |
781 | BUG_ON(!es1); | |
782 | es->es_lblk = es1->es_lblk; | |
783 | es->es_len = es1->es_len; | |
784 | es->es_pblk = es1->es_pblk; | |
eb68d0e2 ZL |
785 | stats->es_stats_cache_hits++; |
786 | } else { | |
787 | stats->es_stats_cache_misses++; | |
d100eef2 ZL |
788 | } |
789 | ||
790 | read_unlock(&EXT4_I(inode)->i_es_lock); | |
791 | ||
792 | trace_ext4_es_lookup_extent_exit(inode, es, found); | |
793 | return found; | |
794 | } | |
795 | ||
bdedbb7b ZL |
796 | static int __es_remove_extent(struct inode *inode, ext4_lblk_t lblk, |
797 | ext4_lblk_t end) | |
654598be | 798 | { |
bdedbb7b | 799 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be | 800 | struct rb_node *node; |
654598be ZL |
801 | struct extent_status *es; |
802 | struct extent_status orig_es; | |
06b0c886 | 803 | ext4_lblk_t len1, len2; |
fdc0212e | 804 | ext4_fsblk_t block; |
e15f742c | 805 | int err; |
654598be | 806 | |
e15f742c TT |
807 | retry: |
808 | err = 0; | |
06b0c886 | 809 | es = __es_tree_search(&tree->root, lblk); |
654598be ZL |
810 | if (!es) |
811 | goto out; | |
06b0c886 | 812 | if (es->es_lblk > end) |
654598be ZL |
813 | goto out; |
814 | ||
815 | /* Simply invalidate cache_es. */ | |
816 | tree->cache_es = NULL; | |
817 | ||
06b0c886 ZL |
818 | orig_es.es_lblk = es->es_lblk; |
819 | orig_es.es_len = es->es_len; | |
fdc0212e ZL |
820 | orig_es.es_pblk = es->es_pblk; |
821 | ||
06b0c886 ZL |
822 | len1 = lblk > es->es_lblk ? lblk - es->es_lblk : 0; |
823 | len2 = ext4_es_end(es) > end ? ext4_es_end(es) - end : 0; | |
654598be | 824 | if (len1 > 0) |
06b0c886 | 825 | es->es_len = len1; |
654598be ZL |
826 | if (len2 > 0) { |
827 | if (len1 > 0) { | |
06b0c886 ZL |
828 | struct extent_status newes; |
829 | ||
830 | newes.es_lblk = end + 1; | |
831 | newes.es_len = len2; | |
666525df | 832 | block = 0x7FDEADBEEFULL; |
fdc0212e | 833 | if (ext4_es_is_written(&orig_es) || |
9a6633b1 | 834 | ext4_es_is_unwritten(&orig_es)) |
fdc0212e ZL |
835 | block = ext4_es_pblock(&orig_es) + |
836 | orig_es.es_len - len2; | |
9a6633b1 TT |
837 | ext4_es_store_pblock_status(&newes, block, |
838 | ext4_es_status(&orig_es)); | |
bdedbb7b | 839 | err = __es_insert_extent(inode, &newes); |
654598be | 840 | if (err) { |
06b0c886 ZL |
841 | es->es_lblk = orig_es.es_lblk; |
842 | es->es_len = orig_es.es_len; | |
e15f742c TT |
843 | if ((err == -ENOMEM) && |
844 | __ext4_es_shrink(EXT4_SB(inode->i_sb), 1, | |
845 | EXT4_I(inode))) | |
846 | goto retry; | |
654598be ZL |
847 | goto out; |
848 | } | |
849 | } else { | |
06b0c886 ZL |
850 | es->es_lblk = end + 1; |
851 | es->es_len = len2; | |
fdc0212e ZL |
852 | if (ext4_es_is_written(es) || |
853 | ext4_es_is_unwritten(es)) { | |
854 | block = orig_es.es_pblk + orig_es.es_len - len2; | |
855 | ext4_es_store_pblock(es, block); | |
856 | } | |
654598be ZL |
857 | } |
858 | goto out; | |
859 | } | |
860 | ||
861 | if (len1 > 0) { | |
862 | node = rb_next(&es->rb_node); | |
863 | if (node) | |
864 | es = rb_entry(node, struct extent_status, rb_node); | |
865 | else | |
866 | es = NULL; | |
867 | } | |
868 | ||
06b0c886 | 869 | while (es && ext4_es_end(es) <= end) { |
654598be ZL |
870 | node = rb_next(&es->rb_node); |
871 | rb_erase(&es->rb_node, &tree->root); | |
bdedbb7b | 872 | ext4_es_free_extent(inode, es); |
654598be ZL |
873 | if (!node) { |
874 | es = NULL; | |
875 | break; | |
876 | } | |
877 | es = rb_entry(node, struct extent_status, rb_node); | |
878 | } | |
879 | ||
06b0c886 | 880 | if (es && es->es_lblk < end + 1) { |
fdc0212e ZL |
881 | ext4_lblk_t orig_len = es->es_len; |
882 | ||
06b0c886 ZL |
883 | len1 = ext4_es_end(es) - end; |
884 | es->es_lblk = end + 1; | |
885 | es->es_len = len1; | |
fdc0212e ZL |
886 | if (ext4_es_is_written(es) || ext4_es_is_unwritten(es)) { |
887 | block = es->es_pblk + orig_len - len1; | |
888 | ext4_es_store_pblock(es, block); | |
889 | } | |
654598be ZL |
890 | } |
891 | ||
892 | out: | |
06b0c886 ZL |
893 | return err; |
894 | } | |
895 | ||
896 | /* | |
897 | * ext4_es_remove_extent() removes a space from a extent status tree. | |
898 | * | |
899 | * Return 0 on success, error code on failure. | |
900 | */ | |
901 | int ext4_es_remove_extent(struct inode *inode, ext4_lblk_t lblk, | |
902 | ext4_lblk_t len) | |
903 | { | |
06b0c886 ZL |
904 | ext4_lblk_t end; |
905 | int err = 0; | |
906 | ||
907 | trace_ext4_es_remove_extent(inode, lblk, len); | |
908 | es_debug("remove [%u/%u) from extent status tree of inode %lu\n", | |
909 | lblk, len, inode->i_ino); | |
910 | ||
d4381472 EG |
911 | if (!len) |
912 | return err; | |
913 | ||
06b0c886 ZL |
914 | end = lblk + len - 1; |
915 | BUG_ON(end < lblk); | |
916 | ||
06b0c886 | 917 | write_lock(&EXT4_I(inode)->i_es_lock); |
bdedbb7b | 918 | err = __es_remove_extent(inode, lblk, end); |
654598be ZL |
919 | write_unlock(&EXT4_I(inode)->i_es_lock); |
920 | ext4_es_print_tree(inode); | |
921 | return err; | |
922 | } | |
74cd15cd | 923 | |
d3922a77 ZL |
924 | static int ext4_inode_touch_time_cmp(void *priv, struct list_head *a, |
925 | struct list_head *b) | |
926 | { | |
927 | struct ext4_inode_info *eia, *eib; | |
928 | eia = list_entry(a, struct ext4_inode_info, i_es_lru); | |
929 | eib = list_entry(b, struct ext4_inode_info, i_es_lru); | |
930 | ||
7869a4a6 TT |
931 | if (ext4_test_inode_state(&eia->vfs_inode, EXT4_STATE_EXT_PRECACHED) && |
932 | !ext4_test_inode_state(&eib->vfs_inode, EXT4_STATE_EXT_PRECACHED)) | |
933 | return 1; | |
934 | if (!ext4_test_inode_state(&eia->vfs_inode, EXT4_STATE_EXT_PRECACHED) && | |
935 | ext4_test_inode_state(&eib->vfs_inode, EXT4_STATE_EXT_PRECACHED)) | |
936 | return -1; | |
d3922a77 ZL |
937 | if (eia->i_touch_when == eib->i_touch_when) |
938 | return 0; | |
939 | if (time_after(eia->i_touch_when, eib->i_touch_when)) | |
940 | return 1; | |
941 | else | |
942 | return -1; | |
943 | } | |
944 | ||
e15f742c TT |
945 | static int __ext4_es_shrink(struct ext4_sb_info *sbi, int nr_to_scan, |
946 | struct ext4_inode_info *locked_ei) | |
74cd15cd | 947 | { |
74cd15cd | 948 | struct ext4_inode_info *ei; |
eb68d0e2 | 949 | struct ext4_es_stats *es_stats; |
d3922a77 | 950 | struct list_head *cur, *tmp; |
7869a4a6 | 951 | LIST_HEAD(skipped); |
eb68d0e2 ZL |
952 | ktime_t start_time; |
953 | u64 scan_time; | |
1ab6c499 | 954 | int nr_shrunk = 0; |
7869a4a6 | 955 | int retried = 0, skip_precached = 1, nr_skipped = 0; |
74cd15cd | 956 | |
eb68d0e2 ZL |
957 | es_stats = &sbi->s_es_stats; |
958 | start_time = ktime_get(); | |
74cd15cd | 959 | spin_lock(&sbi->s_es_lru_lock); |
d3922a77 | 960 | |
7869a4a6 | 961 | retry: |
74cd15cd | 962 | list_for_each_safe(cur, tmp, &sbi->s_es_lru) { |
1ab6c499 DC |
963 | int shrunk; |
964 | ||
d3922a77 ZL |
965 | /* |
966 | * If we have already reclaimed all extents from extent | |
967 | * status tree, just stop the loop immediately. | |
968 | */ | |
eb68d0e2 ZL |
969 | if (percpu_counter_read_positive( |
970 | &es_stats->es_stats_lru_cnt) == 0) | |
d3922a77 | 971 | break; |
74cd15cd ZL |
972 | |
973 | ei = list_entry(cur, struct ext4_inode_info, i_es_lru); | |
974 | ||
7869a4a6 TT |
975 | /* |
976 | * Skip the inode that is newer than the last_sorted | |
977 | * time. Normally we try hard to avoid shrinking | |
978 | * precached inodes, but we will as a last resort. | |
979 | */ | |
eb68d0e2 | 980 | if ((es_stats->es_stats_last_sorted < ei->i_touch_when) || |
7869a4a6 TT |
981 | (skip_precached && ext4_test_inode_state(&ei->vfs_inode, |
982 | EXT4_STATE_EXT_PRECACHED))) { | |
983 | nr_skipped++; | |
984 | list_move_tail(cur, &skipped); | |
74cd15cd ZL |
985 | continue; |
986 | } | |
d3922a77 | 987 | |
3f1f9b85 TT |
988 | if (ei->i_es_lru_nr == 0 || ei == locked_ei || |
989 | !write_trylock(&ei->i_es_lock)) | |
d3922a77 | 990 | continue; |
74cd15cd | 991 | |
1ab6c499 | 992 | shrunk = __es_try_to_reclaim_extents(ei, nr_to_scan); |
d3922a77 ZL |
993 | if (ei->i_es_lru_nr == 0) |
994 | list_del_init(&ei->i_es_lru); | |
74cd15cd ZL |
995 | write_unlock(&ei->i_es_lock); |
996 | ||
1ab6c499 DC |
997 | nr_shrunk += shrunk; |
998 | nr_to_scan -= shrunk; | |
74cd15cd ZL |
999 | if (nr_to_scan == 0) |
1000 | break; | |
1001 | } | |
d3922a77 ZL |
1002 | |
1003 | /* Move the newer inodes into the tail of the LRU list. */ | |
7869a4a6 TT |
1004 | list_splice_tail(&skipped, &sbi->s_es_lru); |
1005 | INIT_LIST_HEAD(&skipped); | |
1006 | ||
1007 | /* | |
1008 | * If we skipped any inodes, and we weren't able to make any | |
1009 | * forward progress, sort the list and try again. | |
1010 | */ | |
1011 | if ((nr_shrunk == 0) && nr_skipped && !retried) { | |
1012 | retried++; | |
1013 | list_sort(NULL, &sbi->s_es_lru, ext4_inode_touch_time_cmp); | |
eb68d0e2 | 1014 | es_stats->es_stats_last_sorted = jiffies; |
7869a4a6 TT |
1015 | ei = list_first_entry(&sbi->s_es_lru, struct ext4_inode_info, |
1016 | i_es_lru); | |
1017 | /* | |
1018 | * If there are no non-precached inodes left on the | |
1019 | * list, start releasing precached extents. | |
1020 | */ | |
1021 | if (ext4_test_inode_state(&ei->vfs_inode, | |
1022 | EXT4_STATE_EXT_PRECACHED)) | |
1023 | skip_precached = 0; | |
1024 | goto retry; | |
1025 | } | |
1026 | ||
74cd15cd | 1027 | spin_unlock(&sbi->s_es_lru_lock); |
74cd15cd | 1028 | |
e15f742c | 1029 | if (locked_ei && nr_shrunk == 0) |
7869a4a6 | 1030 | nr_shrunk = __es_try_to_reclaim_extents(locked_ei, nr_to_scan); |
e15f742c | 1031 | |
eb68d0e2 ZL |
1032 | scan_time = ktime_to_ns(ktime_sub(ktime_get(), start_time)); |
1033 | if (likely(es_stats->es_stats_scan_time)) | |
1034 | es_stats->es_stats_scan_time = (scan_time + | |
1035 | es_stats->es_stats_scan_time*3) / 4; | |
1036 | else | |
1037 | es_stats->es_stats_scan_time = scan_time; | |
1038 | if (scan_time > es_stats->es_stats_max_scan_time) | |
1039 | es_stats->es_stats_max_scan_time = scan_time; | |
1040 | if (likely(es_stats->es_stats_shrunk)) | |
1041 | es_stats->es_stats_shrunk = (nr_shrunk + | |
1042 | es_stats->es_stats_shrunk*3) / 4; | |
1043 | else | |
1044 | es_stats->es_stats_shrunk = nr_shrunk; | |
1045 | ||
1046 | trace_ext4_es_shrink(sbi->s_sb, nr_shrunk, scan_time, skip_precached, | |
1047 | nr_skipped, retried); | |
e15f742c TT |
1048 | return nr_shrunk; |
1049 | } | |
1050 | ||
1ab6c499 DC |
1051 | static unsigned long ext4_es_count(struct shrinker *shrink, |
1052 | struct shrink_control *sc) | |
1053 | { | |
1054 | unsigned long nr; | |
1055 | struct ext4_sb_info *sbi; | |
1056 | ||
1057 | sbi = container_of(shrink, struct ext4_sb_info, s_es_shrinker); | |
eb68d0e2 | 1058 | nr = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_lru_cnt); |
e963bb1d | 1059 | trace_ext4_es_shrink_count(sbi->s_sb, sc->nr_to_scan, nr); |
1ab6c499 DC |
1060 | return nr; |
1061 | } | |
1062 | ||
1063 | static unsigned long ext4_es_scan(struct shrinker *shrink, | |
1064 | struct shrink_control *sc) | |
e15f742c TT |
1065 | { |
1066 | struct ext4_sb_info *sbi = container_of(shrink, | |
1067 | struct ext4_sb_info, s_es_shrinker); | |
1068 | int nr_to_scan = sc->nr_to_scan; | |
1069 | int ret, nr_shrunk; | |
1070 | ||
eb68d0e2 | 1071 | ret = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_lru_cnt); |
e963bb1d | 1072 | trace_ext4_es_shrink_scan_enter(sbi->s_sb, nr_to_scan, ret); |
e15f742c TT |
1073 | |
1074 | if (!nr_to_scan) | |
1075 | return ret; | |
1076 | ||
1077 | nr_shrunk = __ext4_es_shrink(sbi, nr_to_scan, NULL); | |
1078 | ||
e963bb1d | 1079 | trace_ext4_es_shrink_scan_exit(sbi->s_sb, nr_shrunk, ret); |
1ab6c499 | 1080 | return nr_shrunk; |
74cd15cd ZL |
1081 | } |
1082 | ||
eb68d0e2 | 1083 | static void *ext4_es_seq_shrinker_info_start(struct seq_file *seq, loff_t *pos) |
74cd15cd | 1084 | { |
eb68d0e2 ZL |
1085 | return *pos ? NULL : SEQ_START_TOKEN; |
1086 | } | |
1087 | ||
1088 | static void * | |
1089 | ext4_es_seq_shrinker_info_next(struct seq_file *seq, void *v, loff_t *pos) | |
1090 | { | |
1091 | return NULL; | |
1092 | } | |
1093 | ||
1094 | static int ext4_es_seq_shrinker_info_show(struct seq_file *seq, void *v) | |
1095 | { | |
1096 | struct ext4_sb_info *sbi = seq->private; | |
1097 | struct ext4_es_stats *es_stats = &sbi->s_es_stats; | |
1098 | struct ext4_inode_info *ei, *max = NULL; | |
1099 | unsigned int inode_cnt = 0; | |
1100 | ||
1101 | if (v != SEQ_START_TOKEN) | |
1102 | return 0; | |
1103 | ||
1104 | /* here we just find an inode that has the max nr. of objects */ | |
1105 | spin_lock(&sbi->s_es_lru_lock); | |
1106 | list_for_each_entry(ei, &sbi->s_es_lru, i_es_lru) { | |
1107 | inode_cnt++; | |
1108 | if (max && max->i_es_all_nr < ei->i_es_all_nr) | |
1109 | max = ei; | |
1110 | else if (!max) | |
1111 | max = ei; | |
1112 | } | |
1113 | spin_unlock(&sbi->s_es_lru_lock); | |
1114 | ||
1115 | seq_printf(seq, "stats:\n %lld objects\n %lld reclaimable objects\n", | |
1116 | percpu_counter_sum_positive(&es_stats->es_stats_all_cnt), | |
1117 | percpu_counter_sum_positive(&es_stats->es_stats_lru_cnt)); | |
1118 | seq_printf(seq, " %lu/%lu cache hits/misses\n", | |
1119 | es_stats->es_stats_cache_hits, | |
1120 | es_stats->es_stats_cache_misses); | |
1121 | if (es_stats->es_stats_last_sorted != 0) | |
1122 | seq_printf(seq, " %u ms last sorted interval\n", | |
1123 | jiffies_to_msecs(jiffies - | |
1124 | es_stats->es_stats_last_sorted)); | |
1125 | if (inode_cnt) | |
1126 | seq_printf(seq, " %d inodes on lru list\n", inode_cnt); | |
1127 | ||
1128 | seq_printf(seq, "average:\n %llu us scan time\n", | |
1129 | div_u64(es_stats->es_stats_scan_time, 1000)); | |
1130 | seq_printf(seq, " %lu shrunk objects\n", es_stats->es_stats_shrunk); | |
1131 | if (inode_cnt) | |
1132 | seq_printf(seq, | |
1133 | "maximum:\n %lu inode (%u objects, %u reclaimable)\n" | |
1134 | " %llu us max scan time\n", | |
1135 | max->vfs_inode.i_ino, max->i_es_all_nr, max->i_es_lru_nr, | |
1136 | div_u64(es_stats->es_stats_max_scan_time, 1000)); | |
1137 | ||
1138 | return 0; | |
1139 | } | |
1140 | ||
1141 | static void ext4_es_seq_shrinker_info_stop(struct seq_file *seq, void *v) | |
1142 | { | |
1143 | } | |
1144 | ||
1145 | static const struct seq_operations ext4_es_seq_shrinker_info_ops = { | |
1146 | .start = ext4_es_seq_shrinker_info_start, | |
1147 | .next = ext4_es_seq_shrinker_info_next, | |
1148 | .stop = ext4_es_seq_shrinker_info_stop, | |
1149 | .show = ext4_es_seq_shrinker_info_show, | |
1150 | }; | |
1151 | ||
1152 | static int | |
1153 | ext4_es_seq_shrinker_info_open(struct inode *inode, struct file *file) | |
1154 | { | |
1155 | int ret; | |
1156 | ||
1157 | ret = seq_open(file, &ext4_es_seq_shrinker_info_ops); | |
1158 | if (!ret) { | |
1159 | struct seq_file *m = file->private_data; | |
1160 | m->private = PDE_DATA(inode); | |
1161 | } | |
1162 | ||
1163 | return ret; | |
1164 | } | |
1165 | ||
1166 | static int | |
1167 | ext4_es_seq_shrinker_info_release(struct inode *inode, struct file *file) | |
1168 | { | |
1169 | return seq_release(inode, file); | |
1170 | } | |
1171 | ||
1172 | static const struct file_operations ext4_es_seq_shrinker_info_fops = { | |
1173 | .owner = THIS_MODULE, | |
1174 | .open = ext4_es_seq_shrinker_info_open, | |
1175 | .read = seq_read, | |
1176 | .llseek = seq_lseek, | |
1177 | .release = ext4_es_seq_shrinker_info_release, | |
1178 | }; | |
1179 | ||
1180 | int ext4_es_register_shrinker(struct ext4_sb_info *sbi) | |
74cd15cd | 1181 | { |
eb68d0e2 ZL |
1182 | int err; |
1183 | ||
74cd15cd ZL |
1184 | INIT_LIST_HEAD(&sbi->s_es_lru); |
1185 | spin_lock_init(&sbi->s_es_lru_lock); | |
eb68d0e2 ZL |
1186 | sbi->s_es_stats.es_stats_last_sorted = 0; |
1187 | sbi->s_es_stats.es_stats_shrunk = 0; | |
1188 | sbi->s_es_stats.es_stats_cache_hits = 0; | |
1189 | sbi->s_es_stats.es_stats_cache_misses = 0; | |
1190 | sbi->s_es_stats.es_stats_scan_time = 0; | |
1191 | sbi->s_es_stats.es_stats_max_scan_time = 0; | |
c2661b80 | 1192 | err = percpu_counter_init(&sbi->s_es_stats.es_stats_all_cnt, 0, GFP_KERNEL); |
eb68d0e2 ZL |
1193 | if (err) |
1194 | return err; | |
c2661b80 | 1195 | err = percpu_counter_init(&sbi->s_es_stats.es_stats_lru_cnt, 0, GFP_KERNEL); |
eb68d0e2 ZL |
1196 | if (err) |
1197 | goto err1; | |
1198 | ||
1ab6c499 DC |
1199 | sbi->s_es_shrinker.scan_objects = ext4_es_scan; |
1200 | sbi->s_es_shrinker.count_objects = ext4_es_count; | |
74cd15cd | 1201 | sbi->s_es_shrinker.seeks = DEFAULT_SEEKS; |
eb68d0e2 ZL |
1202 | err = register_shrinker(&sbi->s_es_shrinker); |
1203 | if (err) | |
1204 | goto err2; | |
1205 | ||
1206 | if (sbi->s_proc) | |
1207 | proc_create_data("es_shrinker_info", S_IRUGO, sbi->s_proc, | |
1208 | &ext4_es_seq_shrinker_info_fops, sbi); | |
1209 | ||
1210 | return 0; | |
1211 | ||
1212 | err2: | |
1213 | percpu_counter_destroy(&sbi->s_es_stats.es_stats_lru_cnt); | |
1214 | err1: | |
1215 | percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt); | |
1216 | return err; | |
74cd15cd ZL |
1217 | } |
1218 | ||
d3922a77 | 1219 | void ext4_es_unregister_shrinker(struct ext4_sb_info *sbi) |
74cd15cd | 1220 | { |
eb68d0e2 ZL |
1221 | if (sbi->s_proc) |
1222 | remove_proc_entry("es_shrinker_info", sbi->s_proc); | |
1223 | percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt); | |
1224 | percpu_counter_destroy(&sbi->s_es_stats.es_stats_lru_cnt); | |
d3922a77 | 1225 | unregister_shrinker(&sbi->s_es_shrinker); |
74cd15cd ZL |
1226 | } |
1227 | ||
1228 | void ext4_es_lru_add(struct inode *inode) | |
1229 | { | |
1230 | struct ext4_inode_info *ei = EXT4_I(inode); | |
1231 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1232 | ||
d3922a77 ZL |
1233 | ei->i_touch_when = jiffies; |
1234 | ||
1235 | if (!list_empty(&ei->i_es_lru)) | |
1236 | return; | |
1237 | ||
74cd15cd ZL |
1238 | spin_lock(&sbi->s_es_lru_lock); |
1239 | if (list_empty(&ei->i_es_lru)) | |
1240 | list_add_tail(&ei->i_es_lru, &sbi->s_es_lru); | |
74cd15cd ZL |
1241 | spin_unlock(&sbi->s_es_lru_lock); |
1242 | } | |
1243 | ||
1244 | void ext4_es_lru_del(struct inode *inode) | |
1245 | { | |
1246 | struct ext4_inode_info *ei = EXT4_I(inode); | |
1247 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1248 | ||
1249 | spin_lock(&sbi->s_es_lru_lock); | |
1250 | if (!list_empty(&ei->i_es_lru)) | |
1251 | list_del_init(&ei->i_es_lru); | |
1252 | spin_unlock(&sbi->s_es_lru_lock); | |
1253 | } | |
1254 | ||
74cd15cd ZL |
1255 | static int __es_try_to_reclaim_extents(struct ext4_inode_info *ei, |
1256 | int nr_to_scan) | |
1257 | { | |
1258 | struct inode *inode = &ei->vfs_inode; | |
1259 | struct ext4_es_tree *tree = &ei->i_es_tree; | |
1260 | struct rb_node *node; | |
1261 | struct extent_status *es; | |
1ab6c499 | 1262 | unsigned long nr_shrunk = 0; |
7869a4a6 TT |
1263 | static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL, |
1264 | DEFAULT_RATELIMIT_BURST); | |
74cd15cd ZL |
1265 | |
1266 | if (ei->i_es_lru_nr == 0) | |
1267 | return 0; | |
1268 | ||
7869a4a6 TT |
1269 | if (ext4_test_inode_state(inode, EXT4_STATE_EXT_PRECACHED) && |
1270 | __ratelimit(&_rs)) | |
1271 | ext4_warning(inode->i_sb, "forced shrink of precached extents"); | |
1272 | ||
74cd15cd ZL |
1273 | node = rb_first(&tree->root); |
1274 | while (node != NULL) { | |
1275 | es = rb_entry(node, struct extent_status, rb_node); | |
1276 | node = rb_next(&es->rb_node); | |
1277 | /* | |
1278 | * We can't reclaim delayed extent from status tree because | |
1279 | * fiemap, bigallic, and seek_data/hole need to use it. | |
1280 | */ | |
1281 | if (!ext4_es_is_delayed(es)) { | |
1282 | rb_erase(&es->rb_node, &tree->root); | |
1283 | ext4_es_free_extent(inode, es); | |
1284 | nr_shrunk++; | |
1285 | if (--nr_to_scan == 0) | |
1286 | break; | |
1287 | } | |
1288 | } | |
1289 | tree->cache_es = NULL; | |
1290 | return nr_shrunk; | |
1291 | } |