]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
654598be ZL |
2 | /* |
3 | * fs/ext4/extents_status.c | |
4 | * | |
5 | * Written by Yongqiang Yang <xiaoqiangnk@gmail.com> | |
6 | * Modified by | |
7 | * Allison Henderson <achender@linux.vnet.ibm.com> | |
8 | * Hugh Dickins <hughd@google.com> | |
9 | * Zheng Liu <wenqing.lz@taobao.com> | |
10 | * | |
11 | * Ext4 extents status tree core functions. | |
12 | */ | |
d3922a77 | 13 | #include <linux/list_sort.h> |
eb68d0e2 ZL |
14 | #include <linux/proc_fs.h> |
15 | #include <linux/seq_file.h> | |
654598be | 16 | #include "ext4.h" |
654598be | 17 | |
992e9fdd ZL |
18 | #include <trace/events/ext4.h> |
19 | ||
654598be ZL |
20 | /* |
21 | * According to previous discussion in Ext4 Developer Workshop, we | |
22 | * will introduce a new structure called io tree to track all extent | |
23 | * status in order to solve some problems that we have met | |
24 | * (e.g. Reservation space warning), and provide extent-level locking. | |
25 | * Delay extent tree is the first step to achieve this goal. It is | |
26 | * original built by Yongqiang Yang. At that time it is called delay | |
06b0c886 | 27 | * extent tree, whose goal is only track delayed extents in memory to |
654598be ZL |
28 | * simplify the implementation of fiemap and bigalloc, and introduce |
29 | * lseek SEEK_DATA/SEEK_HOLE support. That is why it is still called | |
06b0c886 ZL |
30 | * delay extent tree at the first commit. But for better understand |
31 | * what it does, it has been rename to extent status tree. | |
654598be | 32 | * |
06b0c886 ZL |
33 | * Step1: |
34 | * Currently the first step has been done. All delayed extents are | |
35 | * tracked in the tree. It maintains the delayed extent when a delayed | |
36 | * allocation is issued, and the delayed extent is written out or | |
654598be ZL |
37 | * invalidated. Therefore the implementation of fiemap and bigalloc |
38 | * are simplified, and SEEK_DATA/SEEK_HOLE are introduced. | |
39 | * | |
40 | * The following comment describes the implemenmtation of extent | |
41 | * status tree and future works. | |
06b0c886 ZL |
42 | * |
43 | * Step2: | |
44 | * In this step all extent status are tracked by extent status tree. | |
45 | * Thus, we can first try to lookup a block mapping in this tree before | |
46 | * finding it in extent tree. Hence, single extent cache can be removed | |
47 | * because extent status tree can do a better job. Extents in status | |
48 | * tree are loaded on-demand. Therefore, the extent status tree may not | |
49 | * contain all of the extents in a file. Meanwhile we define a shrinker | |
50 | * to reclaim memory from extent status tree because fragmented extent | |
51 | * tree will make status tree cost too much memory. written/unwritten/- | |
52 | * hole extents in the tree will be reclaimed by this shrinker when we | |
53 | * are under high memory pressure. Delayed extents will not be | |
54 | * reclimed because fiemap, bigalloc, and seek_data/hole need it. | |
654598be ZL |
55 | */ |
56 | ||
57 | /* | |
06b0c886 | 58 | * Extent status tree implementation for ext4. |
654598be ZL |
59 | * |
60 | * | |
61 | * ========================================================================== | |
06b0c886 | 62 | * Extent status tree tracks all extent status. |
654598be | 63 | * |
06b0c886 | 64 | * 1. Why we need to implement extent status tree? |
654598be | 65 | * |
06b0c886 | 66 | * Without extent status tree, ext4 identifies a delayed extent by looking |
654598be ZL |
67 | * up page cache, this has several deficiencies - complicated, buggy, |
68 | * and inefficient code. | |
69 | * | |
06b0c886 ZL |
70 | * FIEMAP, SEEK_HOLE/DATA, bigalloc, and writeout all need to know if a |
71 | * block or a range of blocks are belonged to a delayed extent. | |
654598be | 72 | * |
06b0c886 | 73 | * Let us have a look at how they do without extent status tree. |
654598be ZL |
74 | * -- FIEMAP |
75 | * FIEMAP looks up page cache to identify delayed allocations from holes. | |
76 | * | |
77 | * -- SEEK_HOLE/DATA | |
78 | * SEEK_HOLE/DATA has the same problem as FIEMAP. | |
79 | * | |
80 | * -- bigalloc | |
81 | * bigalloc looks up page cache to figure out if a block is | |
82 | * already under delayed allocation or not to determine whether | |
83 | * quota reserving is needed for the cluster. | |
84 | * | |
654598be ZL |
85 | * -- writeout |
86 | * Writeout looks up whole page cache to see if a buffer is | |
87 | * mapped, If there are not very many delayed buffers, then it is | |
3f8b6fb7 | 88 | * time consuming. |
654598be | 89 | * |
06b0c886 | 90 | * With extent status tree implementation, FIEMAP, SEEK_HOLE/DATA, |
654598be ZL |
91 | * bigalloc and writeout can figure out if a block or a range of |
92 | * blocks is under delayed allocation(belonged to a delayed extent) or | |
06b0c886 | 93 | * not by searching the extent tree. |
654598be ZL |
94 | * |
95 | * | |
96 | * ========================================================================== | |
06b0c886 ZL |
97 | * 2. Ext4 extent status tree impelmentation |
98 | * | |
99 | * -- extent | |
100 | * A extent is a range of blocks which are contiguous logically and | |
101 | * physically. Unlike extent in extent tree, this extent in ext4 is | |
102 | * a in-memory struct, there is no corresponding on-disk data. There | |
103 | * is no limit on length of extent, so an extent can contain as many | |
104 | * blocks as they are contiguous logically and physically. | |
654598be | 105 | * |
06b0c886 ZL |
106 | * -- extent status tree |
107 | * Every inode has an extent status tree and all allocation blocks | |
108 | * are added to the tree with different status. The extent in the | |
109 | * tree are ordered by logical block no. | |
654598be | 110 | * |
06b0c886 ZL |
111 | * -- operations on a extent status tree |
112 | * There are three important operations on a delayed extent tree: find | |
113 | * next extent, adding a extent(a range of blocks) and removing a extent. | |
654598be | 114 | * |
06b0c886 ZL |
115 | * -- race on a extent status tree |
116 | * Extent status tree is protected by inode->i_es_lock. | |
654598be | 117 | * |
06b0c886 ZL |
118 | * -- memory consumption |
119 | * Fragmented extent tree will make extent status tree cost too much | |
120 | * memory. Hence, we will reclaim written/unwritten/hole extents from | |
121 | * the tree under a heavy memory pressure. | |
654598be ZL |
122 | * |
123 | * | |
124 | * ========================================================================== | |
06b0c886 ZL |
125 | * 3. Performance analysis |
126 | * | |
654598be ZL |
127 | * -- overhead |
128 | * 1. There is a cache extent for write access, so if writes are | |
129 | * not very random, adding space operaions are in O(1) time. | |
130 | * | |
131 | * -- gain | |
132 | * 2. Code is much simpler, more readable, more maintainable and | |
133 | * more efficient. | |
134 | * | |
135 | * | |
136 | * ========================================================================== | |
137 | * 4. TODO list | |
654598be | 138 | * |
06b0c886 | 139 | * -- Refactor delayed space reservation |
654598be ZL |
140 | * |
141 | * -- Extent-level locking | |
142 | */ | |
143 | ||
144 | static struct kmem_cache *ext4_es_cachep; | |
1dc0aa46 | 145 | static struct kmem_cache *ext4_pending_cachep; |
654598be | 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, | |
8fcc3a58 | 149 | ext4_lblk_t end, int *reserved); |
dd475925 | 150 | static int es_reclaim_extents(struct ext4_inode_info *ei, int *nr_to_scan); |
edaa53ca ZL |
151 | static int __es_shrink(struct ext4_sb_info *sbi, int nr_to_scan, |
152 | struct ext4_inode_info *locked_ei); | |
b6bf9171 EW |
153 | static void __revise_pending(struct inode *inode, ext4_lblk_t lblk, |
154 | ext4_lblk_t len); | |
06b0c886 | 155 | |
654598be ZL |
156 | int __init ext4_init_es(void) |
157 | { | |
24630774 TT |
158 | ext4_es_cachep = kmem_cache_create("ext4_extent_status", |
159 | sizeof(struct extent_status), | |
160 | 0, (SLAB_RECLAIM_ACCOUNT), NULL); | |
654598be ZL |
161 | if (ext4_es_cachep == NULL) |
162 | return -ENOMEM; | |
163 | return 0; | |
164 | } | |
165 | ||
166 | void ext4_exit_es(void) | |
167 | { | |
21c580d8 | 168 | kmem_cache_destroy(ext4_es_cachep); |
654598be ZL |
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 | /* | |
ad431025 EW |
239 | * ext4_es_find_extent_range - find extent with specified status within block |
240 | * range or next extent following block range in | |
241 | * extents status tree | |
654598be | 242 | * |
ad431025 EW |
243 | * @inode - file containing the range |
244 | * @matching_fn - pointer to function that matches extents with desired status | |
245 | * @lblk - logical block defining start of range | |
246 | * @end - logical block defining end of range | |
247 | * @es - extent found, if any | |
248 | * | |
249 | * Find the first extent within the block range specified by @lblk and @end | |
250 | * in the extents status tree that satisfies @matching_fn. If a match | |
251 | * is found, it's returned in @es. If not, and a matching extent is found | |
252 | * beyond the block range, it's returned in @es. If no match is found, an | |
253 | * extent is returned in @es whose es_lblk, es_len, and es_pblk components | |
254 | * are 0. | |
654598be | 255 | */ |
ad431025 EW |
256 | static void __es_find_extent_range(struct inode *inode, |
257 | int (*matching_fn)(struct extent_status *es), | |
258 | ext4_lblk_t lblk, ext4_lblk_t end, | |
259 | struct extent_status *es) | |
654598be ZL |
260 | { |
261 | struct ext4_es_tree *tree = NULL; | |
262 | struct extent_status *es1 = NULL; | |
263 | struct rb_node *node; | |
654598be | 264 | |
ad431025 EW |
265 | WARN_ON(es == NULL); |
266 | WARN_ON(end < lblk); | |
992e9fdd | 267 | |
654598be ZL |
268 | tree = &EXT4_I(inode)->i_es_tree; |
269 | ||
ad431025 | 270 | /* see if the extent has been cached */ |
be401363 | 271 | es->es_lblk = es->es_len = es->es_pblk = 0; |
654598be ZL |
272 | if (tree->cache_es) { |
273 | es1 = tree->cache_es; | |
be401363 | 274 | if (in_range(lblk, es1->es_lblk, es1->es_len)) { |
3be78c73 | 275 | es_debug("%u cached by [%u/%u) %llu %x\n", |
be401363 | 276 | lblk, es1->es_lblk, es1->es_len, |
fdc0212e | 277 | ext4_es_pblock(es1), ext4_es_status(es1)); |
654598be ZL |
278 | goto out; |
279 | } | |
280 | } | |
281 | ||
be401363 | 282 | es1 = __es_tree_search(&tree->root, lblk); |
654598be ZL |
283 | |
284 | out: | |
ad431025 | 285 | if (es1 && !matching_fn(es1)) { |
be401363 ZL |
286 | while ((node = rb_next(&es1->rb_node)) != NULL) { |
287 | es1 = rb_entry(node, struct extent_status, rb_node); | |
e30b5dca YZ |
288 | if (es1->es_lblk > end) { |
289 | es1 = NULL; | |
290 | break; | |
291 | } | |
ad431025 | 292 | if (matching_fn(es1)) |
be401363 ZL |
293 | break; |
294 | } | |
295 | } | |
296 | ||
ad431025 | 297 | if (es1 && matching_fn(es1)) { |
654598be | 298 | tree->cache_es = es1; |
06b0c886 ZL |
299 | es->es_lblk = es1->es_lblk; |
300 | es->es_len = es1->es_len; | |
fdc0212e | 301 | es->es_pblk = es1->es_pblk; |
654598be ZL |
302 | } |
303 | ||
ad431025 EW |
304 | } |
305 | ||
306 | /* | |
307 | * Locking for __es_find_extent_range() for external use | |
308 | */ | |
309 | void ext4_es_find_extent_range(struct inode *inode, | |
310 | int (*matching_fn)(struct extent_status *es), | |
311 | ext4_lblk_t lblk, ext4_lblk_t end, | |
312 | struct extent_status *es) | |
313 | { | |
314 | trace_ext4_es_find_extent_range_enter(inode, lblk); | |
315 | ||
316 | read_lock(&EXT4_I(inode)->i_es_lock); | |
317 | __es_find_extent_range(inode, matching_fn, lblk, end, es); | |
318 | read_unlock(&EXT4_I(inode)->i_es_lock); | |
319 | ||
320 | trace_ext4_es_find_extent_range_exit(inode, es); | |
321 | } | |
322 | ||
323 | /* | |
324 | * __es_scan_range - search block range for block with specified status | |
325 | * in extents status tree | |
326 | * | |
327 | * @inode - file containing the range | |
328 | * @matching_fn - pointer to function that matches extents with desired status | |
329 | * @lblk - logical block defining start of range | |
330 | * @end - logical block defining end of range | |
331 | * | |
332 | * Returns true if at least one block in the specified block range satisfies | |
333 | * the criterion specified by @matching_fn, and false if not. If at least | |
334 | * one extent has the specified status, then there is at least one block | |
335 | * in the cluster with that status. Should only be called by code that has | |
336 | * taken i_es_lock. | |
337 | */ | |
338 | static bool __es_scan_range(struct inode *inode, | |
339 | int (*matching_fn)(struct extent_status *es), | |
340 | ext4_lblk_t start, ext4_lblk_t end) | |
341 | { | |
342 | struct extent_status es; | |
343 | ||
344 | __es_find_extent_range(inode, matching_fn, start, end, &es); | |
345 | if (es.es_len == 0) | |
346 | return false; /* no matching extent in the tree */ | |
347 | else if (es.es_lblk <= start && | |
348 | start < es.es_lblk + es.es_len) | |
349 | return true; | |
350 | else if (start <= es.es_lblk && es.es_lblk <= end) | |
351 | return true; | |
352 | else | |
353 | return false; | |
354 | } | |
355 | /* | |
356 | * Locking for __es_scan_range() for external use | |
357 | */ | |
358 | bool ext4_es_scan_range(struct inode *inode, | |
359 | int (*matching_fn)(struct extent_status *es), | |
360 | ext4_lblk_t lblk, ext4_lblk_t end) | |
361 | { | |
362 | bool ret; | |
363 | ||
364 | read_lock(&EXT4_I(inode)->i_es_lock); | |
365 | ret = __es_scan_range(inode, matching_fn, lblk, end); | |
366 | read_unlock(&EXT4_I(inode)->i_es_lock); | |
367 | ||
368 | return ret; | |
369 | } | |
370 | ||
371 | /* | |
372 | * __es_scan_clu - search cluster for block with specified status in | |
373 | * extents status tree | |
374 | * | |
375 | * @inode - file containing the cluster | |
376 | * @matching_fn - pointer to function that matches extents with desired status | |
377 | * @lblk - logical block in cluster to be searched | |
378 | * | |
379 | * Returns true if at least one extent in the cluster containing @lblk | |
380 | * satisfies the criterion specified by @matching_fn, and false if not. If at | |
381 | * least one extent has the specified status, then there is at least one block | |
382 | * in the cluster with that status. Should only be called by code that has | |
383 | * taken i_es_lock. | |
384 | */ | |
385 | static bool __es_scan_clu(struct inode *inode, | |
386 | int (*matching_fn)(struct extent_status *es), | |
387 | ext4_lblk_t lblk) | |
388 | { | |
389 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
390 | ext4_lblk_t lblk_start, lblk_end; | |
391 | ||
392 | lblk_start = EXT4_LBLK_CMASK(sbi, lblk); | |
393 | lblk_end = lblk_start + sbi->s_cluster_ratio - 1; | |
394 | ||
395 | return __es_scan_range(inode, matching_fn, lblk_start, lblk_end); | |
396 | } | |
397 | ||
398 | /* | |
399 | * Locking for __es_scan_clu() for external use | |
400 | */ | |
401 | bool ext4_es_scan_clu(struct inode *inode, | |
402 | int (*matching_fn)(struct extent_status *es), | |
403 | ext4_lblk_t lblk) | |
404 | { | |
405 | bool ret; | |
406 | ||
407 | read_lock(&EXT4_I(inode)->i_es_lock); | |
408 | ret = __es_scan_clu(inode, matching_fn, lblk); | |
654598be | 409 | read_unlock(&EXT4_I(inode)->i_es_lock); |
992e9fdd | 410 | |
ad431025 | 411 | return ret; |
654598be ZL |
412 | } |
413 | ||
b0dea4c1 | 414 | static void ext4_es_list_add(struct inode *inode) |
edaa53ca ZL |
415 | { |
416 | struct ext4_inode_info *ei = EXT4_I(inode); | |
417 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
418 | ||
419 | if (!list_empty(&ei->i_es_list)) | |
420 | return; | |
421 | ||
422 | spin_lock(&sbi->s_es_lock); | |
423 | if (list_empty(&ei->i_es_list)) { | |
424 | list_add_tail(&ei->i_es_list, &sbi->s_es_list); | |
425 | sbi->s_es_nr_inode++; | |
426 | } | |
427 | spin_unlock(&sbi->s_es_lock); | |
428 | } | |
429 | ||
b0dea4c1 | 430 | static void ext4_es_list_del(struct inode *inode) |
edaa53ca ZL |
431 | { |
432 | struct ext4_inode_info *ei = EXT4_I(inode); | |
433 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
434 | ||
435 | spin_lock(&sbi->s_es_lock); | |
436 | if (!list_empty(&ei->i_es_list)) { | |
437 | list_del_init(&ei->i_es_list); | |
438 | sbi->s_es_nr_inode--; | |
439 | WARN_ON_ONCE(sbi->s_es_nr_inode < 0); | |
440 | } | |
441 | spin_unlock(&sbi->s_es_lock); | |
442 | } | |
443 | ||
654598be | 444 | static struct extent_status * |
bdedbb7b ZL |
445 | ext4_es_alloc_extent(struct inode *inode, ext4_lblk_t lblk, ext4_lblk_t len, |
446 | ext4_fsblk_t pblk) | |
654598be ZL |
447 | { |
448 | struct extent_status *es; | |
449 | es = kmem_cache_alloc(ext4_es_cachep, GFP_ATOMIC); | |
450 | if (es == NULL) | |
451 | return NULL; | |
06b0c886 ZL |
452 | es->es_lblk = lblk; |
453 | es->es_len = len; | |
fdc0212e | 454 | es->es_pblk = pblk; |
74cd15cd ZL |
455 | |
456 | /* | |
457 | * We don't count delayed extent because we never try to reclaim them | |
458 | */ | |
24630774 | 459 | if (!ext4_es_is_delayed(es)) { |
b0dea4c1 JK |
460 | if (!EXT4_I(inode)->i_es_shk_nr++) |
461 | ext4_es_list_add(inode); | |
eb68d0e2 | 462 | percpu_counter_inc(&EXT4_SB(inode->i_sb)-> |
edaa53ca | 463 | s_es_stats.es_stats_shk_cnt); |
24630774 | 464 | } |
74cd15cd | 465 | |
eb68d0e2 ZL |
466 | EXT4_I(inode)->i_es_all_nr++; |
467 | percpu_counter_inc(&EXT4_SB(inode->i_sb)->s_es_stats.es_stats_all_cnt); | |
468 | ||
654598be ZL |
469 | return es; |
470 | } | |
471 | ||
bdedbb7b | 472 | static void ext4_es_free_extent(struct inode *inode, struct extent_status *es) |
654598be | 473 | { |
eb68d0e2 ZL |
474 | EXT4_I(inode)->i_es_all_nr--; |
475 | percpu_counter_dec(&EXT4_SB(inode->i_sb)->s_es_stats.es_stats_all_cnt); | |
476 | ||
edaa53ca | 477 | /* Decrease the shrink counter when this es is not delayed */ |
74cd15cd | 478 | if (!ext4_es_is_delayed(es)) { |
edaa53ca | 479 | BUG_ON(EXT4_I(inode)->i_es_shk_nr == 0); |
b0dea4c1 JK |
480 | if (!--EXT4_I(inode)->i_es_shk_nr) |
481 | ext4_es_list_del(inode); | |
eb68d0e2 | 482 | percpu_counter_dec(&EXT4_SB(inode->i_sb)-> |
edaa53ca | 483 | s_es_stats.es_stats_shk_cnt); |
74cd15cd ZL |
484 | } |
485 | ||
654598be ZL |
486 | kmem_cache_free(ext4_es_cachep, es); |
487 | } | |
488 | ||
06b0c886 ZL |
489 | /* |
490 | * Check whether or not two extents can be merged | |
491 | * Condition: | |
492 | * - logical block number is contiguous | |
fdc0212e ZL |
493 | * - physical block number is contiguous |
494 | * - status is equal | |
06b0c886 ZL |
495 | */ |
496 | static int ext4_es_can_be_merged(struct extent_status *es1, | |
497 | struct extent_status *es2) | |
498 | { | |
2be12de9 | 499 | if (ext4_es_type(es1) != ext4_es_type(es2)) |
06b0c886 ZL |
500 | return 0; |
501 | ||
0baaea64 LC |
502 | if (((__u64) es1->es_len) + es2->es_len > EXT_MAX_BLOCKS) { |
503 | pr_warn("ES assertion failed when merging extents. " | |
504 | "The sum of lengths of es1 (%d) and es2 (%d) " | |
505 | "is bigger than allowed file size (%d)\n", | |
506 | es1->es_len, es2->es_len, EXT_MAX_BLOCKS); | |
507 | WARN_ON(1); | |
fdc0212e | 508 | return 0; |
0baaea64 | 509 | } |
fdc0212e | 510 | |
bd384364 | 511 | if (((__u64) es1->es_lblk) + es1->es_len != es2->es_lblk) |
fdc0212e ZL |
512 | return 0; |
513 | ||
bd384364 ZL |
514 | if ((ext4_es_is_written(es1) || ext4_es_is_unwritten(es1)) && |
515 | (ext4_es_pblock(es1) + es1->es_len == ext4_es_pblock(es2))) | |
516 | return 1; | |
517 | ||
518 | if (ext4_es_is_hole(es1)) | |
519 | return 1; | |
520 | ||
521 | /* we need to check delayed extent is without unwritten status */ | |
522 | if (ext4_es_is_delayed(es1) && !ext4_es_is_unwritten(es1)) | |
523 | return 1; | |
524 | ||
525 | return 0; | |
06b0c886 ZL |
526 | } |
527 | ||
654598be | 528 | static struct extent_status * |
bdedbb7b | 529 | ext4_es_try_to_merge_left(struct inode *inode, struct extent_status *es) |
654598be | 530 | { |
bdedbb7b | 531 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be ZL |
532 | struct extent_status *es1; |
533 | struct rb_node *node; | |
534 | ||
535 | node = rb_prev(&es->rb_node); | |
536 | if (!node) | |
537 | return es; | |
538 | ||
539 | es1 = rb_entry(node, struct extent_status, rb_node); | |
06b0c886 ZL |
540 | if (ext4_es_can_be_merged(es1, es)) { |
541 | es1->es_len += es->es_len; | |
2be12de9 JK |
542 | if (ext4_es_is_referenced(es)) |
543 | ext4_es_set_referenced(es1); | |
654598be | 544 | rb_erase(&es->rb_node, &tree->root); |
bdedbb7b | 545 | ext4_es_free_extent(inode, es); |
654598be ZL |
546 | es = es1; |
547 | } | |
548 | ||
549 | return es; | |
550 | } | |
551 | ||
552 | static struct extent_status * | |
bdedbb7b | 553 | ext4_es_try_to_merge_right(struct inode *inode, struct extent_status *es) |
654598be | 554 | { |
bdedbb7b | 555 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be ZL |
556 | struct extent_status *es1; |
557 | struct rb_node *node; | |
558 | ||
559 | node = rb_next(&es->rb_node); | |
560 | if (!node) | |
561 | return es; | |
562 | ||
563 | es1 = rb_entry(node, struct extent_status, rb_node); | |
06b0c886 ZL |
564 | if (ext4_es_can_be_merged(es, es1)) { |
565 | es->es_len += es1->es_len; | |
2be12de9 JK |
566 | if (ext4_es_is_referenced(es1)) |
567 | ext4_es_set_referenced(es); | |
654598be | 568 | rb_erase(node, &tree->root); |
bdedbb7b | 569 | ext4_es_free_extent(inode, es1); |
654598be ZL |
570 | } |
571 | ||
572 | return es; | |
573 | } | |
574 | ||
921f266b | 575 | #ifdef ES_AGGRESSIVE_TEST |
d7b2a00c ZL |
576 | #include "ext4_extents.h" /* Needed when ES_AGGRESSIVE_TEST is defined */ |
577 | ||
921f266b DM |
578 | static void ext4_es_insert_extent_ext_check(struct inode *inode, |
579 | struct extent_status *es) | |
580 | { | |
581 | struct ext4_ext_path *path = NULL; | |
582 | struct ext4_extent *ex; | |
583 | ext4_lblk_t ee_block; | |
584 | ext4_fsblk_t ee_start; | |
585 | unsigned short ee_len; | |
586 | int depth, ee_status, es_status; | |
587 | ||
ed8a1a76 | 588 | path = ext4_find_extent(inode, es->es_lblk, NULL, EXT4_EX_NOCACHE); |
921f266b DM |
589 | if (IS_ERR(path)) |
590 | return; | |
591 | ||
592 | depth = ext_depth(inode); | |
593 | ex = path[depth].p_ext; | |
594 | ||
595 | if (ex) { | |
596 | ||
597 | ee_block = le32_to_cpu(ex->ee_block); | |
598 | ee_start = ext4_ext_pblock(ex); | |
599 | ee_len = ext4_ext_get_actual_len(ex); | |
600 | ||
556615dc | 601 | ee_status = ext4_ext_is_unwritten(ex) ? 1 : 0; |
921f266b DM |
602 | es_status = ext4_es_is_unwritten(es) ? 1 : 0; |
603 | ||
604 | /* | |
605 | * Make sure ex and es are not overlap when we try to insert | |
606 | * a delayed/hole extent. | |
607 | */ | |
608 | if (!ext4_es_is_written(es) && !ext4_es_is_unwritten(es)) { | |
609 | if (in_range(es->es_lblk, ee_block, ee_len)) { | |
bdafe42a | 610 | pr_warn("ES insert assertion failed for " |
921f266b DM |
611 | "inode: %lu we can find an extent " |
612 | "at block [%d/%d/%llu/%c], but we " | |
ce140cdd EW |
613 | "want to add a delayed/hole extent " |
614 | "[%d/%d/%llu/%x]\n", | |
921f266b DM |
615 | inode->i_ino, ee_block, ee_len, |
616 | ee_start, ee_status ? 'u' : 'w', | |
617 | es->es_lblk, es->es_len, | |
618 | ext4_es_pblock(es), ext4_es_status(es)); | |
619 | } | |
620 | goto out; | |
621 | } | |
622 | ||
623 | /* | |
624 | * We don't check ee_block == es->es_lblk, etc. because es | |
625 | * might be a part of whole extent, vice versa. | |
626 | */ | |
627 | if (es->es_lblk < ee_block || | |
628 | ext4_es_pblock(es) != ee_start + es->es_lblk - ee_block) { | |
bdafe42a | 629 | pr_warn("ES insert assertion failed for inode: %lu " |
921f266b DM |
630 | "ex_status [%d/%d/%llu/%c] != " |
631 | "es_status [%d/%d/%llu/%c]\n", inode->i_ino, | |
632 | ee_block, ee_len, ee_start, | |
633 | ee_status ? 'u' : 'w', es->es_lblk, es->es_len, | |
634 | ext4_es_pblock(es), es_status ? 'u' : 'w'); | |
635 | goto out; | |
636 | } | |
637 | ||
638 | if (ee_status ^ es_status) { | |
bdafe42a | 639 | pr_warn("ES insert assertion failed for inode: %lu " |
921f266b DM |
640 | "ex_status [%d/%d/%llu/%c] != " |
641 | "es_status [%d/%d/%llu/%c]\n", inode->i_ino, | |
642 | ee_block, ee_len, ee_start, | |
643 | ee_status ? 'u' : 'w', es->es_lblk, es->es_len, | |
644 | ext4_es_pblock(es), es_status ? 'u' : 'w'); | |
645 | } | |
646 | } else { | |
647 | /* | |
648 | * We can't find an extent on disk. So we need to make sure | |
649 | * that we don't want to add an written/unwritten extent. | |
650 | */ | |
651 | if (!ext4_es_is_delayed(es) && !ext4_es_is_hole(es)) { | |
bdafe42a | 652 | pr_warn("ES insert assertion failed for inode: %lu " |
921f266b | 653 | "can't find an extent at block %d but we want " |
ce140cdd EW |
654 | "to add a written/unwritten extent " |
655 | "[%d/%d/%llu/%x]\n", inode->i_ino, | |
921f266b DM |
656 | es->es_lblk, es->es_lblk, es->es_len, |
657 | ext4_es_pblock(es), ext4_es_status(es)); | |
658 | } | |
659 | } | |
660 | out: | |
b7ea89ad TT |
661 | ext4_ext_drop_refs(path); |
662 | kfree(path); | |
921f266b DM |
663 | } |
664 | ||
665 | static void ext4_es_insert_extent_ind_check(struct inode *inode, | |
666 | struct extent_status *es) | |
667 | { | |
668 | struct ext4_map_blocks map; | |
669 | int retval; | |
670 | ||
671 | /* | |
672 | * Here we call ext4_ind_map_blocks to lookup a block mapping because | |
673 | * 'Indirect' structure is defined in indirect.c. So we couldn't | |
674 | * access direct/indirect tree from outside. It is too dirty to define | |
675 | * this function in indirect.c file. | |
676 | */ | |
677 | ||
678 | map.m_lblk = es->es_lblk; | |
679 | map.m_len = es->es_len; | |
680 | ||
681 | retval = ext4_ind_map_blocks(NULL, inode, &map, 0); | |
682 | if (retval > 0) { | |
683 | if (ext4_es_is_delayed(es) || ext4_es_is_hole(es)) { | |
684 | /* | |
685 | * We want to add a delayed/hole extent but this | |
686 | * block has been allocated. | |
687 | */ | |
bdafe42a | 688 | pr_warn("ES insert assertion failed for inode: %lu " |
921f266b | 689 | "We can find blocks but we want to add a " |
ce140cdd | 690 | "delayed/hole extent [%d/%d/%llu/%x]\n", |
921f266b DM |
691 | inode->i_ino, es->es_lblk, es->es_len, |
692 | ext4_es_pblock(es), ext4_es_status(es)); | |
693 | return; | |
694 | } else if (ext4_es_is_written(es)) { | |
695 | if (retval != es->es_len) { | |
bdafe42a | 696 | pr_warn("ES insert assertion failed for " |
921f266b DM |
697 | "inode: %lu retval %d != es_len %d\n", |
698 | inode->i_ino, retval, es->es_len); | |
699 | return; | |
700 | } | |
701 | if (map.m_pblk != ext4_es_pblock(es)) { | |
bdafe42a | 702 | pr_warn("ES insert assertion failed for " |
921f266b DM |
703 | "inode: %lu m_pblk %llu != " |
704 | "es_pblk %llu\n", | |
705 | inode->i_ino, map.m_pblk, | |
706 | ext4_es_pblock(es)); | |
707 | return; | |
708 | } | |
709 | } else { | |
710 | /* | |
711 | * We don't need to check unwritten extent because | |
712 | * indirect-based file doesn't have it. | |
713 | */ | |
1e83bc81 | 714 | BUG(); |
921f266b DM |
715 | } |
716 | } else if (retval == 0) { | |
717 | if (ext4_es_is_written(es)) { | |
bdafe42a | 718 | pr_warn("ES insert assertion failed for inode: %lu " |
921f266b | 719 | "We can't find the block but we want to add " |
ce140cdd | 720 | "a written extent [%d/%d/%llu/%x]\n", |
921f266b DM |
721 | inode->i_ino, es->es_lblk, es->es_len, |
722 | ext4_es_pblock(es), ext4_es_status(es)); | |
723 | return; | |
724 | } | |
725 | } | |
726 | } | |
727 | ||
728 | static inline void ext4_es_insert_extent_check(struct inode *inode, | |
729 | struct extent_status *es) | |
730 | { | |
731 | /* | |
732 | * We don't need to worry about the race condition because | |
733 | * caller takes i_data_sem locking. | |
734 | */ | |
735 | BUG_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem)); | |
736 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
737 | ext4_es_insert_extent_ext_check(inode, es); | |
738 | else | |
739 | ext4_es_insert_extent_ind_check(inode, es); | |
740 | } | |
741 | #else | |
742 | static inline void ext4_es_insert_extent_check(struct inode *inode, | |
743 | struct extent_status *es) | |
744 | { | |
745 | } | |
746 | #endif | |
747 | ||
bdedbb7b | 748 | static int __es_insert_extent(struct inode *inode, struct extent_status *newes) |
654598be | 749 | { |
bdedbb7b | 750 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be ZL |
751 | struct rb_node **p = &tree->root.rb_node; |
752 | struct rb_node *parent = NULL; | |
753 | struct extent_status *es; | |
654598be ZL |
754 | |
755 | while (*p) { | |
756 | parent = *p; | |
757 | es = rb_entry(parent, struct extent_status, rb_node); | |
758 | ||
06b0c886 ZL |
759 | if (newes->es_lblk < es->es_lblk) { |
760 | if (ext4_es_can_be_merged(newes, es)) { | |
761 | /* | |
762 | * Here we can modify es_lblk directly | |
763 | * because it isn't overlapped. | |
764 | */ | |
765 | es->es_lblk = newes->es_lblk; | |
766 | es->es_len += newes->es_len; | |
fdc0212e ZL |
767 | if (ext4_es_is_written(es) || |
768 | ext4_es_is_unwritten(es)) | |
769 | ext4_es_store_pblock(es, | |
770 | newes->es_pblk); | |
bdedbb7b | 771 | es = ext4_es_try_to_merge_left(inode, es); |
654598be ZL |
772 | goto out; |
773 | } | |
774 | p = &(*p)->rb_left; | |
06b0c886 ZL |
775 | } else if (newes->es_lblk > ext4_es_end(es)) { |
776 | if (ext4_es_can_be_merged(es, newes)) { | |
777 | es->es_len += newes->es_len; | |
bdedbb7b | 778 | es = ext4_es_try_to_merge_right(inode, es); |
654598be ZL |
779 | goto out; |
780 | } | |
781 | p = &(*p)->rb_right; | |
782 | } else { | |
1e83bc81 | 783 | BUG(); |
06b0c886 | 784 | return -EINVAL; |
654598be ZL |
785 | } |
786 | } | |
787 | ||
bdedbb7b | 788 | es = ext4_es_alloc_extent(inode, newes->es_lblk, newes->es_len, |
fdc0212e | 789 | newes->es_pblk); |
654598be ZL |
790 | if (!es) |
791 | return -ENOMEM; | |
792 | rb_link_node(&es->rb_node, parent, p); | |
793 | rb_insert_color(&es->rb_node, &tree->root); | |
794 | ||
795 | out: | |
796 | tree->cache_es = es; | |
797 | return 0; | |
798 | } | |
799 | ||
800 | /* | |
bdafe42a TT |
801 | * ext4_es_insert_extent() adds information to an inode's extent |
802 | * status tree. | |
654598be ZL |
803 | * |
804 | * Return 0 on success, error code on failure. | |
805 | */ | |
06b0c886 | 806 | int ext4_es_insert_extent(struct inode *inode, ext4_lblk_t lblk, |
fdc0212e | 807 | ext4_lblk_t len, ext4_fsblk_t pblk, |
3be78c73 | 808 | unsigned int status) |
654598be | 809 | { |
06b0c886 ZL |
810 | struct extent_status newes; |
811 | ext4_lblk_t end = lblk + len - 1; | |
654598be | 812 | int err = 0; |
b6bf9171 | 813 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
654598be | 814 | |
3be78c73 | 815 | es_debug("add [%u/%u) %llu %x to extent status tree of inode %lu\n", |
fdc0212e | 816 | lblk, len, pblk, status, inode->i_ino); |
06b0c886 | 817 | |
d4381472 EG |
818 | if (!len) |
819 | return 0; | |
820 | ||
06b0c886 ZL |
821 | BUG_ON(end < lblk); |
822 | ||
d2dc317d LC |
823 | if ((status & EXTENT_STATUS_DELAYED) && |
824 | (status & EXTENT_STATUS_WRITTEN)) { | |
825 | ext4_warning(inode->i_sb, "Inserting extent [%u/%u] as " | |
826 | " delayed and written which can potentially " | |
8d2ae1cb | 827 | " cause data loss.", lblk, len); |
d2dc317d LC |
828 | WARN_ON(1); |
829 | } | |
830 | ||
06b0c886 ZL |
831 | newes.es_lblk = lblk; |
832 | newes.es_len = len; | |
9a6633b1 | 833 | ext4_es_store_pblock_status(&newes, pblk, status); |
fdc0212e | 834 | trace_ext4_es_insert_extent(inode, &newes); |
654598be | 835 | |
921f266b DM |
836 | ext4_es_insert_extent_check(inode, &newes); |
837 | ||
654598be | 838 | write_lock(&EXT4_I(inode)->i_es_lock); |
8fcc3a58 | 839 | err = __es_remove_extent(inode, lblk, end, NULL); |
06b0c886 ZL |
840 | if (err != 0) |
841 | goto error; | |
e15f742c | 842 | retry: |
bdedbb7b | 843 | err = __es_insert_extent(inode, &newes); |
edaa53ca | 844 | if (err == -ENOMEM && __es_shrink(EXT4_SB(inode->i_sb), |
dd475925 | 845 | 128, EXT4_I(inode))) |
e15f742c TT |
846 | goto retry; |
847 | if (err == -ENOMEM && !ext4_es_is_delayed(&newes)) | |
848 | err = 0; | |
06b0c886 | 849 | |
b6bf9171 EW |
850 | if (sbi->s_cluster_ratio > 1 && test_opt(inode->i_sb, DELALLOC) && |
851 | (status & EXTENT_STATUS_WRITTEN || | |
852 | status & EXTENT_STATUS_UNWRITTEN)) | |
853 | __revise_pending(inode, lblk, len); | |
854 | ||
06b0c886 | 855 | error: |
654598be ZL |
856 | write_unlock(&EXT4_I(inode)->i_es_lock); |
857 | ||
858 | ext4_es_print_tree(inode); | |
859 | ||
860 | return err; | |
861 | } | |
862 | ||
107a7bd3 TT |
863 | /* |
864 | * ext4_es_cache_extent() inserts information into the extent status | |
865 | * tree if and only if there isn't information about the range in | |
866 | * question already. | |
867 | */ | |
868 | void ext4_es_cache_extent(struct inode *inode, ext4_lblk_t lblk, | |
869 | ext4_lblk_t len, ext4_fsblk_t pblk, | |
870 | unsigned int status) | |
871 | { | |
872 | struct extent_status *es; | |
873 | struct extent_status newes; | |
874 | ext4_lblk_t end = lblk + len - 1; | |
875 | ||
876 | newes.es_lblk = lblk; | |
877 | newes.es_len = len; | |
9a6633b1 | 878 | ext4_es_store_pblock_status(&newes, pblk, status); |
107a7bd3 TT |
879 | trace_ext4_es_cache_extent(inode, &newes); |
880 | ||
881 | if (!len) | |
882 | return; | |
883 | ||
884 | BUG_ON(end < lblk); | |
885 | ||
886 | write_lock(&EXT4_I(inode)->i_es_lock); | |
887 | ||
888 | es = __es_tree_search(&EXT4_I(inode)->i_es_tree.root, lblk); | |
7869a4a6 TT |
889 | if (!es || es->es_lblk > end) |
890 | __es_insert_extent(inode, &newes); | |
107a7bd3 TT |
891 | write_unlock(&EXT4_I(inode)->i_es_lock); |
892 | } | |
893 | ||
d100eef2 ZL |
894 | /* |
895 | * ext4_es_lookup_extent() looks up an extent in extent status tree. | |
896 | * | |
897 | * ext4_es_lookup_extent is called by ext4_map_blocks/ext4_da_map_blocks. | |
898 | * | |
899 | * Return: 1 on found, 0 on not | |
900 | */ | |
901 | int ext4_es_lookup_extent(struct inode *inode, ext4_lblk_t lblk, | |
bb5835ed | 902 | ext4_lblk_t *next_lblk, |
d100eef2 ZL |
903 | struct extent_status *es) |
904 | { | |
905 | struct ext4_es_tree *tree; | |
eb68d0e2 | 906 | struct ext4_es_stats *stats; |
d100eef2 ZL |
907 | struct extent_status *es1 = NULL; |
908 | struct rb_node *node; | |
909 | int found = 0; | |
910 | ||
911 | trace_ext4_es_lookup_extent_enter(inode, lblk); | |
912 | es_debug("lookup extent in block %u\n", lblk); | |
913 | ||
914 | tree = &EXT4_I(inode)->i_es_tree; | |
915 | read_lock(&EXT4_I(inode)->i_es_lock); | |
916 | ||
917 | /* find extent in cache firstly */ | |
918 | es->es_lblk = es->es_len = es->es_pblk = 0; | |
919 | if (tree->cache_es) { | |
920 | es1 = tree->cache_es; | |
921 | if (in_range(lblk, es1->es_lblk, es1->es_len)) { | |
922 | es_debug("%u cached by [%u/%u)\n", | |
923 | lblk, es1->es_lblk, es1->es_len); | |
924 | found = 1; | |
925 | goto out; | |
926 | } | |
927 | } | |
928 | ||
929 | node = tree->root.rb_node; | |
930 | while (node) { | |
931 | es1 = rb_entry(node, struct extent_status, rb_node); | |
932 | if (lblk < es1->es_lblk) | |
933 | node = node->rb_left; | |
934 | else if (lblk > ext4_es_end(es1)) | |
935 | node = node->rb_right; | |
936 | else { | |
937 | found = 1; | |
938 | break; | |
939 | } | |
940 | } | |
941 | ||
942 | out: | |
eb68d0e2 | 943 | stats = &EXT4_SB(inode->i_sb)->s_es_stats; |
d100eef2 ZL |
944 | if (found) { |
945 | BUG_ON(!es1); | |
946 | es->es_lblk = es1->es_lblk; | |
947 | es->es_len = es1->es_len; | |
948 | es->es_pblk = es1->es_pblk; | |
87d8a74b JK |
949 | if (!ext4_es_is_referenced(es1)) |
950 | ext4_es_set_referenced(es1); | |
520f897a | 951 | percpu_counter_inc(&stats->es_stats_cache_hits); |
bb5835ed TT |
952 | if (next_lblk) { |
953 | node = rb_next(&es1->rb_node); | |
954 | if (node) { | |
955 | es1 = rb_entry(node, struct extent_status, | |
956 | rb_node); | |
957 | *next_lblk = es1->es_lblk; | |
958 | } else | |
959 | *next_lblk = 0; | |
960 | } | |
eb68d0e2 | 961 | } else { |
520f897a | 962 | percpu_counter_inc(&stats->es_stats_cache_misses); |
d100eef2 ZL |
963 | } |
964 | ||
965 | read_unlock(&EXT4_I(inode)->i_es_lock); | |
966 | ||
967 | trace_ext4_es_lookup_extent_exit(inode, es, found); | |
968 | return found; | |
969 | } | |
970 | ||
8fcc3a58 EW |
971 | struct rsvd_count { |
972 | int ndelonly; | |
973 | bool first_do_lblk_found; | |
974 | ext4_lblk_t first_do_lblk; | |
975 | ext4_lblk_t last_do_lblk; | |
976 | struct extent_status *left_es; | |
977 | bool partial; | |
978 | ext4_lblk_t lclu; | |
979 | }; | |
980 | ||
981 | /* | |
982 | * init_rsvd - initialize reserved count data before removing block range | |
983 | * in file from extent status tree | |
984 | * | |
985 | * @inode - file containing range | |
986 | * @lblk - first block in range | |
987 | * @es - pointer to first extent in range | |
988 | * @rc - pointer to reserved count data | |
989 | * | |
990 | * Assumes es is not NULL | |
991 | */ | |
992 | static void init_rsvd(struct inode *inode, ext4_lblk_t lblk, | |
993 | struct extent_status *es, struct rsvd_count *rc) | |
994 | { | |
995 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
996 | struct rb_node *node; | |
997 | ||
998 | rc->ndelonly = 0; | |
999 | ||
1000 | /* | |
1001 | * for bigalloc, note the first delonly block in the range has not | |
1002 | * been found, record the extent containing the block to the left of | |
1003 | * the region to be removed, if any, and note that there's no partial | |
1004 | * cluster to track | |
1005 | */ | |
1006 | if (sbi->s_cluster_ratio > 1) { | |
1007 | rc->first_do_lblk_found = false; | |
1008 | if (lblk > es->es_lblk) { | |
1009 | rc->left_es = es; | |
1010 | } else { | |
1011 | node = rb_prev(&es->rb_node); | |
1012 | rc->left_es = node ? rb_entry(node, | |
1013 | struct extent_status, | |
1014 | rb_node) : NULL; | |
1015 | } | |
1016 | rc->partial = false; | |
1017 | } | |
1018 | } | |
1019 | ||
1020 | /* | |
1021 | * count_rsvd - count the clusters containing delayed and not unwritten | |
1022 | * (delonly) blocks in a range within an extent and add to | |
1023 | * the running tally in rsvd_count | |
1024 | * | |
1025 | * @inode - file containing extent | |
1026 | * @lblk - first block in range | |
1027 | * @len - length of range in blocks | |
1028 | * @es - pointer to extent containing clusters to be counted | |
1029 | * @rc - pointer to reserved count data | |
1030 | * | |
1031 | * Tracks partial clusters found at the beginning and end of extents so | |
1032 | * they aren't overcounted when they span adjacent extents | |
1033 | */ | |
1034 | static void count_rsvd(struct inode *inode, ext4_lblk_t lblk, long len, | |
1035 | struct extent_status *es, struct rsvd_count *rc) | |
1036 | { | |
1037 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1038 | ext4_lblk_t i, end, nclu; | |
1039 | ||
1040 | if (!ext4_es_is_delonly(es)) | |
1041 | return; | |
1042 | ||
1043 | WARN_ON(len <= 0); | |
1044 | ||
1045 | if (sbi->s_cluster_ratio == 1) { | |
1046 | rc->ndelonly += (int) len; | |
1047 | return; | |
1048 | } | |
1049 | ||
1050 | /* bigalloc */ | |
1051 | ||
1052 | i = (lblk < es->es_lblk) ? es->es_lblk : lblk; | |
1053 | end = lblk + (ext4_lblk_t) len - 1; | |
1054 | end = (end > ext4_es_end(es)) ? ext4_es_end(es) : end; | |
1055 | ||
1056 | /* record the first block of the first delonly extent seen */ | |
39c0ae16 | 1057 | if (!rc->first_do_lblk_found) { |
8fcc3a58 EW |
1058 | rc->first_do_lblk = i; |
1059 | rc->first_do_lblk_found = true; | |
1060 | } | |
1061 | ||
1062 | /* update the last lblk in the region seen so far */ | |
1063 | rc->last_do_lblk = end; | |
1064 | ||
1065 | /* | |
1066 | * if we're tracking a partial cluster and the current extent | |
1067 | * doesn't start with it, count it and stop tracking | |
1068 | */ | |
1069 | if (rc->partial && (rc->lclu != EXT4_B2C(sbi, i))) { | |
1070 | rc->ndelonly++; | |
1071 | rc->partial = false; | |
1072 | } | |
1073 | ||
1074 | /* | |
1075 | * if the first cluster doesn't start on a cluster boundary but | |
1076 | * ends on one, count it | |
1077 | */ | |
1078 | if (EXT4_LBLK_COFF(sbi, i) != 0) { | |
1079 | if (end >= EXT4_LBLK_CFILL(sbi, i)) { | |
1080 | rc->ndelonly++; | |
1081 | rc->partial = false; | |
1082 | i = EXT4_LBLK_CFILL(sbi, i) + 1; | |
1083 | } | |
1084 | } | |
1085 | ||
1086 | /* | |
1087 | * if the current cluster starts on a cluster boundary, count the | |
1088 | * number of whole delonly clusters in the extent | |
1089 | */ | |
1090 | if ((i + sbi->s_cluster_ratio - 1) <= end) { | |
1091 | nclu = (end - i + 1) >> sbi->s_cluster_bits; | |
1092 | rc->ndelonly += nclu; | |
1093 | i += nclu << sbi->s_cluster_bits; | |
1094 | } | |
1095 | ||
1096 | /* | |
1097 | * start tracking a partial cluster if there's a partial at the end | |
1098 | * of the current extent and we're not already tracking one | |
1099 | */ | |
1100 | if (!rc->partial && i <= end) { | |
1101 | rc->partial = true; | |
1102 | rc->lclu = EXT4_B2C(sbi, i); | |
1103 | } | |
1104 | } | |
1105 | ||
1106 | /* | |
1107 | * __pr_tree_search - search for a pending cluster reservation | |
1108 | * | |
1109 | * @root - root of pending reservation tree | |
1110 | * @lclu - logical cluster to search for | |
1111 | * | |
1112 | * Returns the pending reservation for the cluster identified by @lclu | |
1113 | * if found. If not, returns a reservation for the next cluster if any, | |
1114 | * and if not, returns NULL. | |
1115 | */ | |
1116 | static struct pending_reservation *__pr_tree_search(struct rb_root *root, | |
1117 | ext4_lblk_t lclu) | |
1118 | { | |
1119 | struct rb_node *node = root->rb_node; | |
1120 | struct pending_reservation *pr = NULL; | |
1121 | ||
1122 | while (node) { | |
1123 | pr = rb_entry(node, struct pending_reservation, rb_node); | |
1124 | if (lclu < pr->lclu) | |
1125 | node = node->rb_left; | |
1126 | else if (lclu > pr->lclu) | |
1127 | node = node->rb_right; | |
1128 | else | |
1129 | return pr; | |
1130 | } | |
1131 | if (pr && lclu < pr->lclu) | |
1132 | return pr; | |
1133 | if (pr && lclu > pr->lclu) { | |
1134 | node = rb_next(&pr->rb_node); | |
1135 | return node ? rb_entry(node, struct pending_reservation, | |
1136 | rb_node) : NULL; | |
1137 | } | |
1138 | return NULL; | |
1139 | } | |
1140 | ||
1141 | /* | |
1142 | * get_rsvd - calculates and returns the number of cluster reservations to be | |
1143 | * released when removing a block range from the extent status tree | |
1144 | * and releases any pending reservations within the range | |
1145 | * | |
1146 | * @inode - file containing block range | |
1147 | * @end - last block in range | |
1148 | * @right_es - pointer to extent containing next block beyond end or NULL | |
1149 | * @rc - pointer to reserved count data | |
1150 | * | |
1151 | * The number of reservations to be released is equal to the number of | |
1152 | * clusters containing delayed and not unwritten (delonly) blocks within | |
1153 | * the range, minus the number of clusters still containing delonly blocks | |
1154 | * at the ends of the range, and minus the number of pending reservations | |
1155 | * within the range. | |
1156 | */ | |
1157 | static unsigned int get_rsvd(struct inode *inode, ext4_lblk_t end, | |
1158 | struct extent_status *right_es, | |
1159 | struct rsvd_count *rc) | |
1160 | { | |
1161 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1162 | struct pending_reservation *pr; | |
1163 | struct ext4_pending_tree *tree = &EXT4_I(inode)->i_pending_tree; | |
1164 | struct rb_node *node; | |
1165 | ext4_lblk_t first_lclu, last_lclu; | |
1166 | bool left_delonly, right_delonly, count_pending; | |
1167 | struct extent_status *es; | |
1168 | ||
1169 | if (sbi->s_cluster_ratio > 1) { | |
1170 | /* count any remaining partial cluster */ | |
1171 | if (rc->partial) | |
1172 | rc->ndelonly++; | |
1173 | ||
1174 | if (rc->ndelonly == 0) | |
1175 | return 0; | |
1176 | ||
1177 | first_lclu = EXT4_B2C(sbi, rc->first_do_lblk); | |
1178 | last_lclu = EXT4_B2C(sbi, rc->last_do_lblk); | |
1179 | ||
1180 | /* | |
1181 | * decrease the delonly count by the number of clusters at the | |
1182 | * ends of the range that still contain delonly blocks - | |
1183 | * these clusters still need to be reserved | |
1184 | */ | |
1185 | left_delonly = right_delonly = false; | |
1186 | ||
1187 | es = rc->left_es; | |
1188 | while (es && ext4_es_end(es) >= | |
1189 | EXT4_LBLK_CMASK(sbi, rc->first_do_lblk)) { | |
1190 | if (ext4_es_is_delonly(es)) { | |
1191 | rc->ndelonly--; | |
1192 | left_delonly = true; | |
1193 | break; | |
1194 | } | |
1195 | node = rb_prev(&es->rb_node); | |
1196 | if (!node) | |
1197 | break; | |
1198 | es = rb_entry(node, struct extent_status, rb_node); | |
1199 | } | |
1200 | if (right_es && (!left_delonly || first_lclu != last_lclu)) { | |
1201 | if (end < ext4_es_end(right_es)) { | |
1202 | es = right_es; | |
1203 | } else { | |
1204 | node = rb_next(&right_es->rb_node); | |
1205 | es = node ? rb_entry(node, struct extent_status, | |
1206 | rb_node) : NULL; | |
1207 | } | |
1208 | while (es && es->es_lblk <= | |
1209 | EXT4_LBLK_CFILL(sbi, rc->last_do_lblk)) { | |
1210 | if (ext4_es_is_delonly(es)) { | |
1211 | rc->ndelonly--; | |
1212 | right_delonly = true; | |
1213 | break; | |
1214 | } | |
1215 | node = rb_next(&es->rb_node); | |
1216 | if (!node) | |
1217 | break; | |
1218 | es = rb_entry(node, struct extent_status, | |
1219 | rb_node); | |
1220 | } | |
1221 | } | |
1222 | ||
1223 | /* | |
1224 | * Determine the block range that should be searched for | |
1225 | * pending reservations, if any. Clusters on the ends of the | |
1226 | * original removed range containing delonly blocks are | |
1227 | * excluded. They've already been accounted for and it's not | |
1228 | * possible to determine if an associated pending reservation | |
1229 | * should be released with the information available in the | |
1230 | * extents status tree. | |
1231 | */ | |
1232 | if (first_lclu == last_lclu) { | |
1233 | if (left_delonly | right_delonly) | |
1234 | count_pending = false; | |
1235 | else | |
1236 | count_pending = true; | |
1237 | } else { | |
1238 | if (left_delonly) | |
1239 | first_lclu++; | |
1240 | if (right_delonly) | |
1241 | last_lclu--; | |
1242 | if (first_lclu <= last_lclu) | |
1243 | count_pending = true; | |
1244 | else | |
1245 | count_pending = false; | |
1246 | } | |
1247 | ||
1248 | /* | |
1249 | * a pending reservation found between first_lclu and last_lclu | |
1250 | * represents an allocated cluster that contained at least one | |
1251 | * delonly block, so the delonly total must be reduced by one | |
1252 | * for each pending reservation found and released | |
1253 | */ | |
1254 | if (count_pending) { | |
1255 | pr = __pr_tree_search(&tree->root, first_lclu); | |
1256 | while (pr && pr->lclu <= last_lclu) { | |
1257 | rc->ndelonly--; | |
1258 | node = rb_next(&pr->rb_node); | |
1259 | rb_erase(&pr->rb_node, &tree->root); | |
1260 | kmem_cache_free(ext4_pending_cachep, pr); | |
1261 | if (!node) | |
1262 | break; | |
1263 | pr = rb_entry(node, struct pending_reservation, | |
1264 | rb_node); | |
1265 | } | |
1266 | } | |
1267 | } | |
1268 | return rc->ndelonly; | |
1269 | } | |
1270 | ||
1271 | ||
1272 | /* | |
1273 | * __es_remove_extent - removes block range from extent status tree | |
1274 | * | |
1275 | * @inode - file containing range | |
1276 | * @lblk - first block in range | |
1277 | * @end - last block in range | |
1278 | * @reserved - number of cluster reservations released | |
1279 | * | |
1280 | * If @reserved is not NULL and delayed allocation is enabled, counts | |
1281 | * block/cluster reservations freed by removing range and if bigalloc | |
1282 | * enabled cancels pending reservations as needed. Returns 0 on success, | |
1283 | * error code on failure. | |
1284 | */ | |
bdedbb7b | 1285 | static int __es_remove_extent(struct inode *inode, ext4_lblk_t lblk, |
8fcc3a58 | 1286 | ext4_lblk_t end, int *reserved) |
654598be | 1287 | { |
bdedbb7b | 1288 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be | 1289 | struct rb_node *node; |
654598be ZL |
1290 | struct extent_status *es; |
1291 | struct extent_status orig_es; | |
06b0c886 | 1292 | ext4_lblk_t len1, len2; |
fdc0212e | 1293 | ext4_fsblk_t block; |
e15f742c | 1294 | int err; |
8fcc3a58 EW |
1295 | bool count_reserved = true; |
1296 | struct rsvd_count rc; | |
654598be | 1297 | |
8fcc3a58 EW |
1298 | if (reserved == NULL || !test_opt(inode->i_sb, DELALLOC)) |
1299 | count_reserved = false; | |
e15f742c TT |
1300 | retry: |
1301 | err = 0; | |
8fcc3a58 | 1302 | |
06b0c886 | 1303 | es = __es_tree_search(&tree->root, lblk); |
654598be ZL |
1304 | if (!es) |
1305 | goto out; | |
06b0c886 | 1306 | if (es->es_lblk > end) |
654598be ZL |
1307 | goto out; |
1308 | ||
1309 | /* Simply invalidate cache_es. */ | |
1310 | tree->cache_es = NULL; | |
8fcc3a58 EW |
1311 | if (count_reserved) |
1312 | init_rsvd(inode, lblk, es, &rc); | |
654598be | 1313 | |
06b0c886 ZL |
1314 | orig_es.es_lblk = es->es_lblk; |
1315 | orig_es.es_len = es->es_len; | |
fdc0212e ZL |
1316 | orig_es.es_pblk = es->es_pblk; |
1317 | ||
06b0c886 ZL |
1318 | len1 = lblk > es->es_lblk ? lblk - es->es_lblk : 0; |
1319 | len2 = ext4_es_end(es) > end ? ext4_es_end(es) - end : 0; | |
654598be | 1320 | if (len1 > 0) |
06b0c886 | 1321 | es->es_len = len1; |
654598be ZL |
1322 | if (len2 > 0) { |
1323 | if (len1 > 0) { | |
06b0c886 ZL |
1324 | struct extent_status newes; |
1325 | ||
1326 | newes.es_lblk = end + 1; | |
1327 | newes.es_len = len2; | |
666525df | 1328 | block = 0x7FDEADBEEFULL; |
fdc0212e | 1329 | if (ext4_es_is_written(&orig_es) || |
9a6633b1 | 1330 | ext4_es_is_unwritten(&orig_es)) |
fdc0212e ZL |
1331 | block = ext4_es_pblock(&orig_es) + |
1332 | orig_es.es_len - len2; | |
9a6633b1 TT |
1333 | ext4_es_store_pblock_status(&newes, block, |
1334 | ext4_es_status(&orig_es)); | |
bdedbb7b | 1335 | err = __es_insert_extent(inode, &newes); |
654598be | 1336 | if (err) { |
06b0c886 ZL |
1337 | es->es_lblk = orig_es.es_lblk; |
1338 | es->es_len = orig_es.es_len; | |
e15f742c | 1339 | if ((err == -ENOMEM) && |
edaa53ca | 1340 | __es_shrink(EXT4_SB(inode->i_sb), |
dd475925 | 1341 | 128, EXT4_I(inode))) |
e15f742c | 1342 | goto retry; |
654598be ZL |
1343 | goto out; |
1344 | } | |
1345 | } else { | |
06b0c886 ZL |
1346 | es->es_lblk = end + 1; |
1347 | es->es_len = len2; | |
fdc0212e ZL |
1348 | if (ext4_es_is_written(es) || |
1349 | ext4_es_is_unwritten(es)) { | |
1350 | block = orig_es.es_pblk + orig_es.es_len - len2; | |
1351 | ext4_es_store_pblock(es, block); | |
1352 | } | |
654598be | 1353 | } |
8fcc3a58 EW |
1354 | if (count_reserved) |
1355 | count_rsvd(inode, lblk, orig_es.es_len - len1 - len2, | |
1356 | &orig_es, &rc); | |
654598be ZL |
1357 | goto out; |
1358 | } | |
1359 | ||
1360 | if (len1 > 0) { | |
8fcc3a58 EW |
1361 | if (count_reserved) |
1362 | count_rsvd(inode, lblk, orig_es.es_len - len1, | |
1363 | &orig_es, &rc); | |
654598be ZL |
1364 | node = rb_next(&es->rb_node); |
1365 | if (node) | |
1366 | es = rb_entry(node, struct extent_status, rb_node); | |
1367 | else | |
1368 | es = NULL; | |
1369 | } | |
1370 | ||
06b0c886 | 1371 | while (es && ext4_es_end(es) <= end) { |
8fcc3a58 EW |
1372 | if (count_reserved) |
1373 | count_rsvd(inode, es->es_lblk, es->es_len, es, &rc); | |
654598be ZL |
1374 | node = rb_next(&es->rb_node); |
1375 | rb_erase(&es->rb_node, &tree->root); | |
bdedbb7b | 1376 | ext4_es_free_extent(inode, es); |
654598be ZL |
1377 | if (!node) { |
1378 | es = NULL; | |
1379 | break; | |
1380 | } | |
1381 | es = rb_entry(node, struct extent_status, rb_node); | |
1382 | } | |
1383 | ||
06b0c886 | 1384 | if (es && es->es_lblk < end + 1) { |
fdc0212e ZL |
1385 | ext4_lblk_t orig_len = es->es_len; |
1386 | ||
06b0c886 | 1387 | len1 = ext4_es_end(es) - end; |
8fcc3a58 EW |
1388 | if (count_reserved) |
1389 | count_rsvd(inode, es->es_lblk, orig_len - len1, | |
1390 | es, &rc); | |
06b0c886 ZL |
1391 | es->es_lblk = end + 1; |
1392 | es->es_len = len1; | |
fdc0212e ZL |
1393 | if (ext4_es_is_written(es) || ext4_es_is_unwritten(es)) { |
1394 | block = es->es_pblk + orig_len - len1; | |
1395 | ext4_es_store_pblock(es, block); | |
1396 | } | |
654598be ZL |
1397 | } |
1398 | ||
8fcc3a58 EW |
1399 | if (count_reserved) |
1400 | *reserved = get_rsvd(inode, end, es, &rc); | |
654598be | 1401 | out: |
06b0c886 ZL |
1402 | return err; |
1403 | } | |
1404 | ||
1405 | /* | |
8fcc3a58 | 1406 | * ext4_es_remove_extent - removes block range from extent status tree |
06b0c886 | 1407 | * |
8fcc3a58 EW |
1408 | * @inode - file containing range |
1409 | * @lblk - first block in range | |
1410 | * @len - number of blocks to remove | |
1411 | * | |
1412 | * Reduces block/cluster reservation count and for bigalloc cancels pending | |
1413 | * reservations as needed. Returns 0 on success, error code on failure. | |
06b0c886 ZL |
1414 | */ |
1415 | int ext4_es_remove_extent(struct inode *inode, ext4_lblk_t lblk, | |
1416 | ext4_lblk_t len) | |
1417 | { | |
06b0c886 ZL |
1418 | ext4_lblk_t end; |
1419 | int err = 0; | |
8fcc3a58 | 1420 | int reserved = 0; |
06b0c886 ZL |
1421 | |
1422 | trace_ext4_es_remove_extent(inode, lblk, len); | |
1423 | es_debug("remove [%u/%u) from extent status tree of inode %lu\n", | |
1424 | lblk, len, inode->i_ino); | |
1425 | ||
d4381472 EG |
1426 | if (!len) |
1427 | return err; | |
1428 | ||
06b0c886 ZL |
1429 | end = lblk + len - 1; |
1430 | BUG_ON(end < lblk); | |
1431 | ||
edaa53ca ZL |
1432 | /* |
1433 | * ext4_clear_inode() depends on us taking i_es_lock unconditionally | |
1434 | * so that we are sure __es_shrink() is done with the inode before it | |
1435 | * is reclaimed. | |
1436 | */ | |
06b0c886 | 1437 | write_lock(&EXT4_I(inode)->i_es_lock); |
8fcc3a58 | 1438 | err = __es_remove_extent(inode, lblk, end, &reserved); |
654598be ZL |
1439 | write_unlock(&EXT4_I(inode)->i_es_lock); |
1440 | ext4_es_print_tree(inode); | |
8fcc3a58 | 1441 | ext4_da_release_space(inode, reserved); |
654598be ZL |
1442 | return err; |
1443 | } | |
74cd15cd | 1444 | |
edaa53ca ZL |
1445 | static int __es_shrink(struct ext4_sb_info *sbi, int nr_to_scan, |
1446 | struct ext4_inode_info *locked_ei) | |
74cd15cd | 1447 | { |
74cd15cd | 1448 | struct ext4_inode_info *ei; |
eb68d0e2 | 1449 | struct ext4_es_stats *es_stats; |
eb68d0e2 ZL |
1450 | ktime_t start_time; |
1451 | u64 scan_time; | |
edaa53ca | 1452 | int nr_to_walk; |
1ab6c499 | 1453 | int nr_shrunk = 0; |
edaa53ca | 1454 | int retried = 0, nr_skipped = 0; |
74cd15cd | 1455 | |
eb68d0e2 ZL |
1456 | es_stats = &sbi->s_es_stats; |
1457 | start_time = ktime_get(); | |
d3922a77 | 1458 | |
7869a4a6 | 1459 | retry: |
edaa53ca ZL |
1460 | spin_lock(&sbi->s_es_lock); |
1461 | nr_to_walk = sbi->s_es_nr_inode; | |
1462 | while (nr_to_walk-- > 0) { | |
edaa53ca ZL |
1463 | if (list_empty(&sbi->s_es_list)) { |
1464 | spin_unlock(&sbi->s_es_lock); | |
1465 | goto out; | |
1466 | } | |
1467 | ei = list_first_entry(&sbi->s_es_list, struct ext4_inode_info, | |
1468 | i_es_list); | |
1469 | /* Move the inode to the tail */ | |
dd475925 | 1470 | list_move_tail(&ei->i_es_list, &sbi->s_es_list); |
74cd15cd | 1471 | |
7869a4a6 | 1472 | /* |
edaa53ca ZL |
1473 | * Normally we try hard to avoid shrinking precached inodes, |
1474 | * but we will as a last resort. | |
7869a4a6 | 1475 | */ |
edaa53ca ZL |
1476 | if (!retried && ext4_test_inode_state(&ei->vfs_inode, |
1477 | EXT4_STATE_EXT_PRECACHED)) { | |
7869a4a6 | 1478 | nr_skipped++; |
74cd15cd ZL |
1479 | continue; |
1480 | } | |
d3922a77 | 1481 | |
edaa53ca ZL |
1482 | if (ei == locked_ei || !write_trylock(&ei->i_es_lock)) { |
1483 | nr_skipped++; | |
d3922a77 | 1484 | continue; |
edaa53ca ZL |
1485 | } |
1486 | /* | |
1487 | * Now we hold i_es_lock which protects us from inode reclaim | |
1488 | * freeing inode under us | |
1489 | */ | |
1490 | spin_unlock(&sbi->s_es_lock); | |
74cd15cd | 1491 | |
dd475925 | 1492 | nr_shrunk += es_reclaim_extents(ei, &nr_to_scan); |
74cd15cd ZL |
1493 | write_unlock(&ei->i_es_lock); |
1494 | ||
dd475925 | 1495 | if (nr_to_scan <= 0) |
edaa53ca ZL |
1496 | goto out; |
1497 | spin_lock(&sbi->s_es_lock); | |
74cd15cd | 1498 | } |
edaa53ca | 1499 | spin_unlock(&sbi->s_es_lock); |
7869a4a6 TT |
1500 | |
1501 | /* | |
1502 | * If we skipped any inodes, and we weren't able to make any | |
edaa53ca | 1503 | * forward progress, try again to scan precached inodes. |
7869a4a6 TT |
1504 | */ |
1505 | if ((nr_shrunk == 0) && nr_skipped && !retried) { | |
1506 | retried++; | |
7869a4a6 TT |
1507 | goto retry; |
1508 | } | |
1509 | ||
e15f742c | 1510 | if (locked_ei && nr_shrunk == 0) |
dd475925 | 1511 | nr_shrunk = es_reclaim_extents(locked_ei, &nr_to_scan); |
e15f742c | 1512 | |
edaa53ca | 1513 | out: |
eb68d0e2 ZL |
1514 | scan_time = ktime_to_ns(ktime_sub(ktime_get(), start_time)); |
1515 | if (likely(es_stats->es_stats_scan_time)) | |
1516 | es_stats->es_stats_scan_time = (scan_time + | |
1517 | es_stats->es_stats_scan_time*3) / 4; | |
1518 | else | |
1519 | es_stats->es_stats_scan_time = scan_time; | |
1520 | if (scan_time > es_stats->es_stats_max_scan_time) | |
1521 | es_stats->es_stats_max_scan_time = scan_time; | |
1522 | if (likely(es_stats->es_stats_shrunk)) | |
1523 | es_stats->es_stats_shrunk = (nr_shrunk + | |
1524 | es_stats->es_stats_shrunk*3) / 4; | |
1525 | else | |
1526 | es_stats->es_stats_shrunk = nr_shrunk; | |
1527 | ||
edaa53ca | 1528 | trace_ext4_es_shrink(sbi->s_sb, nr_shrunk, scan_time, |
eb68d0e2 | 1529 | nr_skipped, retried); |
e15f742c TT |
1530 | return nr_shrunk; |
1531 | } | |
1532 | ||
1ab6c499 DC |
1533 | static unsigned long ext4_es_count(struct shrinker *shrink, |
1534 | struct shrink_control *sc) | |
1535 | { | |
1536 | unsigned long nr; | |
1537 | struct ext4_sb_info *sbi; | |
1538 | ||
1539 | sbi = container_of(shrink, struct ext4_sb_info, s_es_shrinker); | |
edaa53ca | 1540 | nr = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt); |
e963bb1d | 1541 | trace_ext4_es_shrink_count(sbi->s_sb, sc->nr_to_scan, nr); |
1ab6c499 DC |
1542 | return nr; |
1543 | } | |
1544 | ||
1545 | static unsigned long ext4_es_scan(struct shrinker *shrink, | |
1546 | struct shrink_control *sc) | |
e15f742c TT |
1547 | { |
1548 | struct ext4_sb_info *sbi = container_of(shrink, | |
1549 | struct ext4_sb_info, s_es_shrinker); | |
1550 | int nr_to_scan = sc->nr_to_scan; | |
1551 | int ret, nr_shrunk; | |
1552 | ||
edaa53ca | 1553 | ret = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt); |
e963bb1d | 1554 | trace_ext4_es_shrink_scan_enter(sbi->s_sb, nr_to_scan, ret); |
e15f742c TT |
1555 | |
1556 | if (!nr_to_scan) | |
1557 | return ret; | |
1558 | ||
edaa53ca | 1559 | nr_shrunk = __es_shrink(sbi, nr_to_scan, NULL); |
e15f742c | 1560 | |
e963bb1d | 1561 | trace_ext4_es_shrink_scan_exit(sbi->s_sb, nr_shrunk, ret); |
1ab6c499 | 1562 | return nr_shrunk; |
74cd15cd ZL |
1563 | } |
1564 | ||
ebd173be | 1565 | int ext4_seq_es_shrinker_info_show(struct seq_file *seq, void *v) |
74cd15cd | 1566 | { |
ebd173be | 1567 | struct ext4_sb_info *sbi = EXT4_SB((struct super_block *) seq->private); |
eb68d0e2 ZL |
1568 | struct ext4_es_stats *es_stats = &sbi->s_es_stats; |
1569 | struct ext4_inode_info *ei, *max = NULL; | |
1570 | unsigned int inode_cnt = 0; | |
1571 | ||
1572 | if (v != SEQ_START_TOKEN) | |
1573 | return 0; | |
1574 | ||
1575 | /* here we just find an inode that has the max nr. of objects */ | |
edaa53ca ZL |
1576 | spin_lock(&sbi->s_es_lock); |
1577 | list_for_each_entry(ei, &sbi->s_es_list, i_es_list) { | |
eb68d0e2 ZL |
1578 | inode_cnt++; |
1579 | if (max && max->i_es_all_nr < ei->i_es_all_nr) | |
1580 | max = ei; | |
1581 | else if (!max) | |
1582 | max = ei; | |
1583 | } | |
edaa53ca | 1584 | spin_unlock(&sbi->s_es_lock); |
eb68d0e2 ZL |
1585 | |
1586 | seq_printf(seq, "stats:\n %lld objects\n %lld reclaimable objects\n", | |
1587 | percpu_counter_sum_positive(&es_stats->es_stats_all_cnt), | |
edaa53ca | 1588 | percpu_counter_sum_positive(&es_stats->es_stats_shk_cnt)); |
520f897a YG |
1589 | seq_printf(seq, " %lld/%lld cache hits/misses\n", |
1590 | percpu_counter_sum_positive(&es_stats->es_stats_cache_hits), | |
1591 | percpu_counter_sum_positive(&es_stats->es_stats_cache_misses)); | |
eb68d0e2 | 1592 | if (inode_cnt) |
edaa53ca | 1593 | seq_printf(seq, " %d inodes on list\n", inode_cnt); |
eb68d0e2 ZL |
1594 | |
1595 | seq_printf(seq, "average:\n %llu us scan time\n", | |
1596 | div_u64(es_stats->es_stats_scan_time, 1000)); | |
1597 | seq_printf(seq, " %lu shrunk objects\n", es_stats->es_stats_shrunk); | |
1598 | if (inode_cnt) | |
1599 | seq_printf(seq, | |
1600 | "maximum:\n %lu inode (%u objects, %u reclaimable)\n" | |
1601 | " %llu us max scan time\n", | |
edaa53ca | 1602 | max->vfs_inode.i_ino, max->i_es_all_nr, max->i_es_shk_nr, |
eb68d0e2 ZL |
1603 | div_u64(es_stats->es_stats_max_scan_time, 1000)); |
1604 | ||
1605 | return 0; | |
1606 | } | |
1607 | ||
eb68d0e2 | 1608 | int ext4_es_register_shrinker(struct ext4_sb_info *sbi) |
74cd15cd | 1609 | { |
eb68d0e2 ZL |
1610 | int err; |
1611 | ||
624d0f1d JK |
1612 | /* Make sure we have enough bits for physical block number */ |
1613 | BUILD_BUG_ON(ES_SHIFT < 48); | |
edaa53ca ZL |
1614 | INIT_LIST_HEAD(&sbi->s_es_list); |
1615 | sbi->s_es_nr_inode = 0; | |
1616 | spin_lock_init(&sbi->s_es_lock); | |
eb68d0e2 | 1617 | sbi->s_es_stats.es_stats_shrunk = 0; |
520f897a YG |
1618 | err = percpu_counter_init(&sbi->s_es_stats.es_stats_cache_hits, 0, |
1619 | GFP_KERNEL); | |
1620 | if (err) | |
1621 | return err; | |
1622 | err = percpu_counter_init(&sbi->s_es_stats.es_stats_cache_misses, 0, | |
1623 | GFP_KERNEL); | |
1624 | if (err) | |
1625 | goto err1; | |
eb68d0e2 ZL |
1626 | sbi->s_es_stats.es_stats_scan_time = 0; |
1627 | sbi->s_es_stats.es_stats_max_scan_time = 0; | |
c2661b80 | 1628 | err = percpu_counter_init(&sbi->s_es_stats.es_stats_all_cnt, 0, GFP_KERNEL); |
eb68d0e2 | 1629 | if (err) |
520f897a | 1630 | goto err2; |
edaa53ca | 1631 | err = percpu_counter_init(&sbi->s_es_stats.es_stats_shk_cnt, 0, GFP_KERNEL); |
eb68d0e2 | 1632 | if (err) |
520f897a | 1633 | goto err3; |
eb68d0e2 | 1634 | |
1ab6c499 DC |
1635 | sbi->s_es_shrinker.scan_objects = ext4_es_scan; |
1636 | sbi->s_es_shrinker.count_objects = ext4_es_count; | |
74cd15cd | 1637 | sbi->s_es_shrinker.seeks = DEFAULT_SEEKS; |
eb68d0e2 ZL |
1638 | err = register_shrinker(&sbi->s_es_shrinker); |
1639 | if (err) | |
520f897a | 1640 | goto err4; |
eb68d0e2 | 1641 | |
eb68d0e2 | 1642 | return 0; |
520f897a | 1643 | err4: |
edaa53ca | 1644 | percpu_counter_destroy(&sbi->s_es_stats.es_stats_shk_cnt); |
520f897a | 1645 | err3: |
eb68d0e2 | 1646 | percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt); |
520f897a YG |
1647 | err2: |
1648 | percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_misses); | |
1649 | err1: | |
1650 | percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_hits); | |
eb68d0e2 | 1651 | return err; |
74cd15cd ZL |
1652 | } |
1653 | ||
d3922a77 | 1654 | void ext4_es_unregister_shrinker(struct ext4_sb_info *sbi) |
74cd15cd | 1655 | { |
520f897a YG |
1656 | percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_hits); |
1657 | percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_misses); | |
eb68d0e2 | 1658 | percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt); |
edaa53ca | 1659 | percpu_counter_destroy(&sbi->s_es_stats.es_stats_shk_cnt); |
d3922a77 | 1660 | unregister_shrinker(&sbi->s_es_shrinker); |
74cd15cd ZL |
1661 | } |
1662 | ||
dd475925 JK |
1663 | /* |
1664 | * Shrink extents in given inode from ei->i_es_shrink_lblk till end. Scan at | |
1665 | * most *nr_to_scan extents, update *nr_to_scan accordingly. | |
1666 | * | |
1667 | * Return 0 if we hit end of tree / interval, 1 if we exhausted nr_to_scan. | |
1668 | * Increment *nr_shrunk by the number of reclaimed extents. Also update | |
1669 | * ei->i_es_shrink_lblk to where we should continue scanning. | |
1670 | */ | |
1671 | static int es_do_reclaim_extents(struct ext4_inode_info *ei, ext4_lblk_t end, | |
1672 | int *nr_to_scan, int *nr_shrunk) | |
74cd15cd ZL |
1673 | { |
1674 | struct inode *inode = &ei->vfs_inode; | |
1675 | struct ext4_es_tree *tree = &ei->i_es_tree; | |
74cd15cd | 1676 | struct extent_status *es; |
dd475925 | 1677 | struct rb_node *node; |
74cd15cd | 1678 | |
dd475925 JK |
1679 | es = __es_tree_search(&tree->root, ei->i_es_shrink_lblk); |
1680 | if (!es) | |
1681 | goto out_wrap; | |
8fcc3a58 | 1682 | |
dd475925 JK |
1683 | while (*nr_to_scan > 0) { |
1684 | if (es->es_lblk > end) { | |
1685 | ei->i_es_shrink_lblk = end + 1; | |
1686 | return 0; | |
1687 | } | |
7869a4a6 | 1688 | |
dd475925 | 1689 | (*nr_to_scan)--; |
74cd15cd ZL |
1690 | node = rb_next(&es->rb_node); |
1691 | /* | |
1692 | * We can't reclaim delayed extent from status tree because | |
1693 | * fiemap, bigallic, and seek_data/hole need to use it. | |
1694 | */ | |
2be12de9 JK |
1695 | if (ext4_es_is_delayed(es)) |
1696 | goto next; | |
1697 | if (ext4_es_is_referenced(es)) { | |
1698 | ext4_es_clear_referenced(es); | |
1699 | goto next; | |
74cd15cd | 1700 | } |
2be12de9 JK |
1701 | |
1702 | rb_erase(&es->rb_node, &tree->root); | |
1703 | ext4_es_free_extent(inode, es); | |
1704 | (*nr_shrunk)++; | |
1705 | next: | |
dd475925 JK |
1706 | if (!node) |
1707 | goto out_wrap; | |
1708 | es = rb_entry(node, struct extent_status, rb_node); | |
74cd15cd | 1709 | } |
dd475925 JK |
1710 | ei->i_es_shrink_lblk = es->es_lblk; |
1711 | return 1; | |
1712 | out_wrap: | |
1713 | ei->i_es_shrink_lblk = 0; | |
1714 | return 0; | |
1715 | } | |
1716 | ||
1717 | static int es_reclaim_extents(struct ext4_inode_info *ei, int *nr_to_scan) | |
1718 | { | |
1719 | struct inode *inode = &ei->vfs_inode; | |
1720 | int nr_shrunk = 0; | |
1721 | ext4_lblk_t start = ei->i_es_shrink_lblk; | |
1722 | static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL, | |
1723 | DEFAULT_RATELIMIT_BURST); | |
1724 | ||
1725 | if (ei->i_es_shk_nr == 0) | |
1726 | return 0; | |
1727 | ||
1728 | if (ext4_test_inode_state(inode, EXT4_STATE_EXT_PRECACHED) && | |
1729 | __ratelimit(&_rs)) | |
1730 | ext4_warning(inode->i_sb, "forced shrink of precached extents"); | |
1731 | ||
1732 | if (!es_do_reclaim_extents(ei, EXT_MAX_BLOCKS, nr_to_scan, &nr_shrunk) && | |
1733 | start != 0) | |
1734 | es_do_reclaim_extents(ei, start - 1, nr_to_scan, &nr_shrunk); | |
1735 | ||
1736 | ei->i_es_tree.cache_es = NULL; | |
74cd15cd ZL |
1737 | return nr_shrunk; |
1738 | } | |
1dc0aa46 | 1739 | |
b0c013e2 TT |
1740 | /* |
1741 | * Called to support EXT4_IOC_CLEAR_ES_CACHE. We can only remove | |
1742 | * discretionary entries from the extent status cache. (Some entries | |
1743 | * must be present for proper operations.) | |
1744 | */ | |
1745 | void ext4_clear_inode_es(struct inode *inode) | |
1746 | { | |
1747 | struct ext4_inode_info *ei = EXT4_I(inode); | |
1748 | struct extent_status *es; | |
1749 | struct ext4_es_tree *tree; | |
1750 | struct rb_node *node; | |
1751 | ||
1752 | write_lock(&ei->i_es_lock); | |
1753 | tree = &EXT4_I(inode)->i_es_tree; | |
1754 | tree->cache_es = NULL; | |
1755 | node = rb_first(&tree->root); | |
1756 | while (node) { | |
1757 | es = rb_entry(node, struct extent_status, rb_node); | |
1758 | node = rb_next(node); | |
1759 | if (!ext4_es_is_delayed(es)) { | |
1760 | rb_erase(&es->rb_node, &tree->root); | |
1761 | ext4_es_free_extent(inode, es); | |
1762 | } | |
1763 | } | |
1764 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_PRECACHED); | |
1765 | write_unlock(&ei->i_es_lock); | |
1766 | } | |
1767 | ||
1dc0aa46 EW |
1768 | #ifdef ES_DEBUG__ |
1769 | static void ext4_print_pending_tree(struct inode *inode) | |
1770 | { | |
1771 | struct ext4_pending_tree *tree; | |
1772 | struct rb_node *node; | |
1773 | struct pending_reservation *pr; | |
1774 | ||
1775 | printk(KERN_DEBUG "pending reservations for inode %lu:", inode->i_ino); | |
1776 | tree = &EXT4_I(inode)->i_pending_tree; | |
1777 | node = rb_first(&tree->root); | |
1778 | while (node) { | |
1779 | pr = rb_entry(node, struct pending_reservation, rb_node); | |
1780 | printk(KERN_DEBUG " %u", pr->lclu); | |
1781 | node = rb_next(node); | |
1782 | } | |
1783 | printk(KERN_DEBUG "\n"); | |
1784 | } | |
1785 | #else | |
1786 | #define ext4_print_pending_tree(inode) | |
1787 | #endif | |
1788 | ||
1789 | int __init ext4_init_pending(void) | |
1790 | { | |
1791 | ext4_pending_cachep = kmem_cache_create("ext4_pending_reservation", | |
1792 | sizeof(struct pending_reservation), | |
1793 | 0, (SLAB_RECLAIM_ACCOUNT), NULL); | |
1794 | if (ext4_pending_cachep == NULL) | |
1795 | return -ENOMEM; | |
1796 | return 0; | |
1797 | } | |
1798 | ||
1799 | void ext4_exit_pending(void) | |
1800 | { | |
1801 | kmem_cache_destroy(ext4_pending_cachep); | |
1802 | } | |
1803 | ||
1804 | void ext4_init_pending_tree(struct ext4_pending_tree *tree) | |
1805 | { | |
1806 | tree->root = RB_ROOT; | |
1807 | } | |
1808 | ||
1809 | /* | |
1810 | * __get_pending - retrieve a pointer to a pending reservation | |
1811 | * | |
1812 | * @inode - file containing the pending cluster reservation | |
1813 | * @lclu - logical cluster of interest | |
1814 | * | |
1815 | * Returns a pointer to a pending reservation if it's a member of | |
1816 | * the set, and NULL if not. Must be called holding i_es_lock. | |
1817 | */ | |
1818 | static struct pending_reservation *__get_pending(struct inode *inode, | |
1819 | ext4_lblk_t lclu) | |
1820 | { | |
1821 | struct ext4_pending_tree *tree; | |
1822 | struct rb_node *node; | |
1823 | struct pending_reservation *pr = NULL; | |
1824 | ||
1825 | tree = &EXT4_I(inode)->i_pending_tree; | |
1826 | node = (&tree->root)->rb_node; | |
1827 | ||
1828 | while (node) { | |
1829 | pr = rb_entry(node, struct pending_reservation, rb_node); | |
1830 | if (lclu < pr->lclu) | |
1831 | node = node->rb_left; | |
1832 | else if (lclu > pr->lclu) | |
1833 | node = node->rb_right; | |
1834 | else if (lclu == pr->lclu) | |
1835 | return pr; | |
1836 | } | |
1837 | return NULL; | |
1838 | } | |
1839 | ||
1840 | /* | |
1841 | * __insert_pending - adds a pending cluster reservation to the set of | |
1842 | * pending reservations | |
1843 | * | |
1844 | * @inode - file containing the cluster | |
1845 | * @lblk - logical block in the cluster to be added | |
1846 | * | |
1847 | * Returns 0 on successful insertion and -ENOMEM on failure. If the | |
1848 | * pending reservation is already in the set, returns successfully. | |
1849 | */ | |
1850 | static int __insert_pending(struct inode *inode, ext4_lblk_t lblk) | |
1851 | { | |
1852 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1853 | struct ext4_pending_tree *tree = &EXT4_I(inode)->i_pending_tree; | |
1854 | struct rb_node **p = &tree->root.rb_node; | |
1855 | struct rb_node *parent = NULL; | |
1856 | struct pending_reservation *pr; | |
1857 | ext4_lblk_t lclu; | |
1858 | int ret = 0; | |
1859 | ||
1860 | lclu = EXT4_B2C(sbi, lblk); | |
1861 | /* search to find parent for insertion */ | |
1862 | while (*p) { | |
1863 | parent = *p; | |
1864 | pr = rb_entry(parent, struct pending_reservation, rb_node); | |
1865 | ||
1866 | if (lclu < pr->lclu) { | |
1867 | p = &(*p)->rb_left; | |
1868 | } else if (lclu > pr->lclu) { | |
1869 | p = &(*p)->rb_right; | |
1870 | } else { | |
1871 | /* pending reservation already inserted */ | |
1872 | goto out; | |
1873 | } | |
1874 | } | |
1875 | ||
1876 | pr = kmem_cache_alloc(ext4_pending_cachep, GFP_ATOMIC); | |
1877 | if (pr == NULL) { | |
1878 | ret = -ENOMEM; | |
1879 | goto out; | |
1880 | } | |
1881 | pr->lclu = lclu; | |
1882 | ||
1883 | rb_link_node(&pr->rb_node, parent, p); | |
1884 | rb_insert_color(&pr->rb_node, &tree->root); | |
1885 | ||
1886 | out: | |
1887 | return ret; | |
1888 | } | |
1889 | ||
1890 | /* | |
1891 | * __remove_pending - removes a pending cluster reservation from the set | |
1892 | * of pending reservations | |
1893 | * | |
1894 | * @inode - file containing the cluster | |
1895 | * @lblk - logical block in the pending cluster reservation to be removed | |
1896 | * | |
1897 | * Returns successfully if pending reservation is not a member of the set. | |
1898 | */ | |
1899 | static void __remove_pending(struct inode *inode, ext4_lblk_t lblk) | |
1900 | { | |
1901 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1902 | struct pending_reservation *pr; | |
1903 | struct ext4_pending_tree *tree; | |
1904 | ||
1905 | pr = __get_pending(inode, EXT4_B2C(sbi, lblk)); | |
1906 | if (pr != NULL) { | |
1907 | tree = &EXT4_I(inode)->i_pending_tree; | |
1908 | rb_erase(&pr->rb_node, &tree->root); | |
1909 | kmem_cache_free(ext4_pending_cachep, pr); | |
1910 | } | |
1911 | } | |
1912 | ||
1913 | /* | |
1914 | * ext4_remove_pending - removes a pending cluster reservation from the set | |
1915 | * of pending reservations | |
1916 | * | |
1917 | * @inode - file containing the cluster | |
1918 | * @lblk - logical block in the pending cluster reservation to be removed | |
1919 | * | |
1920 | * Locking for external use of __remove_pending. | |
1921 | */ | |
1922 | void ext4_remove_pending(struct inode *inode, ext4_lblk_t lblk) | |
1923 | { | |
1924 | struct ext4_inode_info *ei = EXT4_I(inode); | |
1925 | ||
1926 | write_lock(&ei->i_es_lock); | |
1927 | __remove_pending(inode, lblk); | |
1928 | write_unlock(&ei->i_es_lock); | |
1929 | } | |
1930 | ||
1931 | /* | |
1932 | * ext4_is_pending - determine whether a cluster has a pending reservation | |
1933 | * on it | |
1934 | * | |
1935 | * @inode - file containing the cluster | |
1936 | * @lblk - logical block in the cluster | |
1937 | * | |
1938 | * Returns true if there's a pending reservation for the cluster in the | |
1939 | * set of pending reservations, and false if not. | |
1940 | */ | |
1941 | bool ext4_is_pending(struct inode *inode, ext4_lblk_t lblk) | |
1942 | { | |
1943 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1944 | struct ext4_inode_info *ei = EXT4_I(inode); | |
1945 | bool ret; | |
1946 | ||
1947 | read_lock(&ei->i_es_lock); | |
1948 | ret = (bool)(__get_pending(inode, EXT4_B2C(sbi, lblk)) != NULL); | |
1949 | read_unlock(&ei->i_es_lock); | |
1950 | ||
1951 | return ret; | |
1952 | } | |
0b02f4c0 EW |
1953 | |
1954 | /* | |
1955 | * ext4_es_insert_delayed_block - adds a delayed block to the extents status | |
1956 | * tree, adding a pending reservation where | |
1957 | * needed | |
1958 | * | |
1959 | * @inode - file containing the newly added block | |
1960 | * @lblk - logical block to be added | |
1961 | * @allocated - indicates whether a physical cluster has been allocated for | |
1962 | * the logical cluster that contains the block | |
1963 | * | |
1964 | * Returns 0 on success, negative error code on failure. | |
1965 | */ | |
1966 | int ext4_es_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk, | |
1967 | bool allocated) | |
1968 | { | |
1969 | struct extent_status newes; | |
1970 | int err = 0; | |
1971 | ||
1972 | es_debug("add [%u/1) delayed to extent status tree of inode %lu\n", | |
1973 | lblk, inode->i_ino); | |
1974 | ||
1975 | newes.es_lblk = lblk; | |
1976 | newes.es_len = 1; | |
1977 | ext4_es_store_pblock_status(&newes, ~0, EXTENT_STATUS_DELAYED); | |
1978 | trace_ext4_es_insert_delayed_block(inode, &newes, allocated); | |
1979 | ||
1980 | ext4_es_insert_extent_check(inode, &newes); | |
1981 | ||
1982 | write_lock(&EXT4_I(inode)->i_es_lock); | |
1983 | ||
8fcc3a58 | 1984 | err = __es_remove_extent(inode, lblk, lblk, NULL); |
0b02f4c0 EW |
1985 | if (err != 0) |
1986 | goto error; | |
1987 | retry: | |
1988 | err = __es_insert_extent(inode, &newes); | |
1989 | if (err == -ENOMEM && __es_shrink(EXT4_SB(inode->i_sb), | |
1990 | 128, EXT4_I(inode))) | |
1991 | goto retry; | |
1992 | if (err != 0) | |
1993 | goto error; | |
1994 | ||
1995 | if (allocated) | |
1996 | __insert_pending(inode, lblk); | |
1997 | ||
1998 | error: | |
1999 | write_unlock(&EXT4_I(inode)->i_es_lock); | |
2000 | ||
2001 | ext4_es_print_tree(inode); | |
2002 | ext4_print_pending_tree(inode); | |
2003 | ||
2004 | return err; | |
2005 | } | |
b6bf9171 EW |
2006 | |
2007 | /* | |
2008 | * __es_delayed_clu - count number of clusters containing blocks that | |
2009 | * are delayed only | |
2010 | * | |
2011 | * @inode - file containing block range | |
2012 | * @start - logical block defining start of range | |
2013 | * @end - logical block defining end of range | |
2014 | * | |
2015 | * Returns the number of clusters containing only delayed (not delayed | |
2016 | * and unwritten) blocks in the range specified by @start and @end. Any | |
2017 | * cluster or part of a cluster within the range and containing a delayed | |
2018 | * and not unwritten block within the range is counted as a whole cluster. | |
2019 | */ | |
2020 | static unsigned int __es_delayed_clu(struct inode *inode, ext4_lblk_t start, | |
2021 | ext4_lblk_t end) | |
2022 | { | |
2023 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; | |
2024 | struct extent_status *es; | |
2025 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
2026 | struct rb_node *node; | |
2027 | ext4_lblk_t first_lclu, last_lclu; | |
2028 | unsigned long long last_counted_lclu; | |
2029 | unsigned int n = 0; | |
2030 | ||
2031 | /* guaranteed to be unequal to any ext4_lblk_t value */ | |
2032 | last_counted_lclu = ~0ULL; | |
2033 | ||
2034 | es = __es_tree_search(&tree->root, start); | |
2035 | ||
2036 | while (es && (es->es_lblk <= end)) { | |
2037 | if (ext4_es_is_delonly(es)) { | |
2038 | if (es->es_lblk <= start) | |
2039 | first_lclu = EXT4_B2C(sbi, start); | |
2040 | else | |
2041 | first_lclu = EXT4_B2C(sbi, es->es_lblk); | |
2042 | ||
2043 | if (ext4_es_end(es) >= end) | |
2044 | last_lclu = EXT4_B2C(sbi, end); | |
2045 | else | |
2046 | last_lclu = EXT4_B2C(sbi, ext4_es_end(es)); | |
2047 | ||
2048 | if (first_lclu == last_counted_lclu) | |
2049 | n += last_lclu - first_lclu; | |
2050 | else | |
2051 | n += last_lclu - first_lclu + 1; | |
2052 | last_counted_lclu = last_lclu; | |
2053 | } | |
2054 | node = rb_next(&es->rb_node); | |
2055 | if (!node) | |
2056 | break; | |
2057 | es = rb_entry(node, struct extent_status, rb_node); | |
2058 | } | |
2059 | ||
2060 | return n; | |
2061 | } | |
2062 | ||
2063 | /* | |
2064 | * ext4_es_delayed_clu - count number of clusters containing blocks that | |
2065 | * are both delayed and unwritten | |
2066 | * | |
2067 | * @inode - file containing block range | |
2068 | * @lblk - logical block defining start of range | |
2069 | * @len - number of blocks in range | |
2070 | * | |
2071 | * Locking for external use of __es_delayed_clu(). | |
2072 | */ | |
2073 | unsigned int ext4_es_delayed_clu(struct inode *inode, ext4_lblk_t lblk, | |
2074 | ext4_lblk_t len) | |
2075 | { | |
2076 | struct ext4_inode_info *ei = EXT4_I(inode); | |
2077 | ext4_lblk_t end; | |
2078 | unsigned int n; | |
2079 | ||
2080 | if (len == 0) | |
2081 | return 0; | |
2082 | ||
2083 | end = lblk + len - 1; | |
2084 | WARN_ON(end < lblk); | |
2085 | ||
2086 | read_lock(&ei->i_es_lock); | |
2087 | ||
2088 | n = __es_delayed_clu(inode, lblk, end); | |
2089 | ||
2090 | read_unlock(&ei->i_es_lock); | |
2091 | ||
2092 | return n; | |
2093 | } | |
2094 | ||
2095 | /* | |
2096 | * __revise_pending - makes, cancels, or leaves unchanged pending cluster | |
2097 | * reservations for a specified block range depending | |
2098 | * upon the presence or absence of delayed blocks | |
2099 | * outside the range within clusters at the ends of the | |
2100 | * range | |
2101 | * | |
2102 | * @inode - file containing the range | |
2103 | * @lblk - logical block defining the start of range | |
2104 | * @len - length of range in blocks | |
2105 | * | |
2106 | * Used after a newly allocated extent is added to the extents status tree. | |
2107 | * Requires that the extents in the range have either written or unwritten | |
2108 | * status. Must be called while holding i_es_lock. | |
2109 | */ | |
2110 | static void __revise_pending(struct inode *inode, ext4_lblk_t lblk, | |
2111 | ext4_lblk_t len) | |
2112 | { | |
2113 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
2114 | ext4_lblk_t end = lblk + len - 1; | |
2115 | ext4_lblk_t first, last; | |
2116 | bool f_del = false, l_del = false; | |
2117 | ||
2118 | if (len == 0) | |
2119 | return; | |
2120 | ||
2121 | /* | |
2122 | * Two cases - block range within single cluster and block range | |
2123 | * spanning two or more clusters. Note that a cluster belonging | |
2124 | * to a range starting and/or ending on a cluster boundary is treated | |
2125 | * as if it does not contain a delayed extent. The new range may | |
2126 | * have allocated space for previously delayed blocks out to the | |
2127 | * cluster boundary, requiring that any pre-existing pending | |
2128 | * reservation be canceled. Because this code only looks at blocks | |
2129 | * outside the range, it should revise pending reservations | |
2130 | * correctly even if the extent represented by the range can't be | |
2131 | * inserted in the extents status tree due to ENOSPC. | |
2132 | */ | |
2133 | ||
2134 | if (EXT4_B2C(sbi, lblk) == EXT4_B2C(sbi, end)) { | |
2135 | first = EXT4_LBLK_CMASK(sbi, lblk); | |
2136 | if (first != lblk) | |
2137 | f_del = __es_scan_range(inode, &ext4_es_is_delonly, | |
2138 | first, lblk - 1); | |
2139 | if (f_del) { | |
2140 | __insert_pending(inode, first); | |
2141 | } else { | |
2142 | last = EXT4_LBLK_CMASK(sbi, end) + | |
2143 | sbi->s_cluster_ratio - 1; | |
2144 | if (last != end) | |
2145 | l_del = __es_scan_range(inode, | |
2146 | &ext4_es_is_delonly, | |
2147 | end + 1, last); | |
2148 | if (l_del) | |
2149 | __insert_pending(inode, last); | |
2150 | else | |
2151 | __remove_pending(inode, last); | |
2152 | } | |
2153 | } else { | |
2154 | first = EXT4_LBLK_CMASK(sbi, lblk); | |
2155 | if (first != lblk) | |
2156 | f_del = __es_scan_range(inode, &ext4_es_is_delonly, | |
2157 | first, lblk - 1); | |
2158 | if (f_del) | |
2159 | __insert_pending(inode, first); | |
2160 | else | |
2161 | __remove_pending(inode, first); | |
2162 | ||
2163 | last = EXT4_LBLK_CMASK(sbi, end) + sbi->s_cluster_ratio - 1; | |
2164 | if (last != end) | |
2165 | l_del = __es_scan_range(inode, &ext4_es_is_delonly, | |
2166 | end + 1, last); | |
2167 | if (l_del) | |
2168 | __insert_pending(inode, last); | |
2169 | else | |
2170 | __remove_pending(inode, last); | |
2171 | } | |
2172 | } |