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, | |
06b0c886 | 149 | ext4_lblk_t end); |
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 | */ | |
714 | BUG_ON(1); | |
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 { | |
06b0c886 ZL |
783 | BUG_ON(1); |
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); |
bdedbb7b | 839 | err = __es_remove_extent(inode, lblk, end); |
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, | |
902 | struct extent_status *es) | |
903 | { | |
904 | struct ext4_es_tree *tree; | |
eb68d0e2 | 905 | struct ext4_es_stats *stats; |
d100eef2 ZL |
906 | struct extent_status *es1 = NULL; |
907 | struct rb_node *node; | |
908 | int found = 0; | |
909 | ||
910 | trace_ext4_es_lookup_extent_enter(inode, lblk); | |
911 | es_debug("lookup extent in block %u\n", lblk); | |
912 | ||
913 | tree = &EXT4_I(inode)->i_es_tree; | |
914 | read_lock(&EXT4_I(inode)->i_es_lock); | |
915 | ||
916 | /* find extent in cache firstly */ | |
917 | es->es_lblk = es->es_len = es->es_pblk = 0; | |
918 | if (tree->cache_es) { | |
919 | es1 = tree->cache_es; | |
920 | if (in_range(lblk, es1->es_lblk, es1->es_len)) { | |
921 | es_debug("%u cached by [%u/%u)\n", | |
922 | lblk, es1->es_lblk, es1->es_len); | |
923 | found = 1; | |
924 | goto out; | |
925 | } | |
926 | } | |
927 | ||
928 | node = tree->root.rb_node; | |
929 | while (node) { | |
930 | es1 = rb_entry(node, struct extent_status, rb_node); | |
931 | if (lblk < es1->es_lblk) | |
932 | node = node->rb_left; | |
933 | else if (lblk > ext4_es_end(es1)) | |
934 | node = node->rb_right; | |
935 | else { | |
936 | found = 1; | |
937 | break; | |
938 | } | |
939 | } | |
940 | ||
941 | out: | |
eb68d0e2 | 942 | stats = &EXT4_SB(inode->i_sb)->s_es_stats; |
d100eef2 ZL |
943 | if (found) { |
944 | BUG_ON(!es1); | |
945 | es->es_lblk = es1->es_lblk; | |
946 | es->es_len = es1->es_len; | |
947 | es->es_pblk = es1->es_pblk; | |
87d8a74b JK |
948 | if (!ext4_es_is_referenced(es1)) |
949 | ext4_es_set_referenced(es1); | |
eb68d0e2 ZL |
950 | stats->es_stats_cache_hits++; |
951 | } else { | |
952 | stats->es_stats_cache_misses++; | |
d100eef2 ZL |
953 | } |
954 | ||
955 | read_unlock(&EXT4_I(inode)->i_es_lock); | |
956 | ||
957 | trace_ext4_es_lookup_extent_exit(inode, es, found); | |
958 | return found; | |
959 | } | |
960 | ||
bdedbb7b ZL |
961 | static int __es_remove_extent(struct inode *inode, ext4_lblk_t lblk, |
962 | ext4_lblk_t end) | |
654598be | 963 | { |
bdedbb7b | 964 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; |
654598be | 965 | struct rb_node *node; |
654598be ZL |
966 | struct extent_status *es; |
967 | struct extent_status orig_es; | |
06b0c886 | 968 | ext4_lblk_t len1, len2; |
fdc0212e | 969 | ext4_fsblk_t block; |
e15f742c | 970 | int err; |
654598be | 971 | |
e15f742c TT |
972 | retry: |
973 | err = 0; | |
06b0c886 | 974 | es = __es_tree_search(&tree->root, lblk); |
654598be ZL |
975 | if (!es) |
976 | goto out; | |
06b0c886 | 977 | if (es->es_lblk > end) |
654598be ZL |
978 | goto out; |
979 | ||
980 | /* Simply invalidate cache_es. */ | |
981 | tree->cache_es = NULL; | |
982 | ||
06b0c886 ZL |
983 | orig_es.es_lblk = es->es_lblk; |
984 | orig_es.es_len = es->es_len; | |
fdc0212e ZL |
985 | orig_es.es_pblk = es->es_pblk; |
986 | ||
06b0c886 ZL |
987 | len1 = lblk > es->es_lblk ? lblk - es->es_lblk : 0; |
988 | len2 = ext4_es_end(es) > end ? ext4_es_end(es) - end : 0; | |
654598be | 989 | if (len1 > 0) |
06b0c886 | 990 | es->es_len = len1; |
654598be ZL |
991 | if (len2 > 0) { |
992 | if (len1 > 0) { | |
06b0c886 ZL |
993 | struct extent_status newes; |
994 | ||
995 | newes.es_lblk = end + 1; | |
996 | newes.es_len = len2; | |
666525df | 997 | block = 0x7FDEADBEEFULL; |
fdc0212e | 998 | if (ext4_es_is_written(&orig_es) || |
9a6633b1 | 999 | ext4_es_is_unwritten(&orig_es)) |
fdc0212e ZL |
1000 | block = ext4_es_pblock(&orig_es) + |
1001 | orig_es.es_len - len2; | |
9a6633b1 TT |
1002 | ext4_es_store_pblock_status(&newes, block, |
1003 | ext4_es_status(&orig_es)); | |
bdedbb7b | 1004 | err = __es_insert_extent(inode, &newes); |
654598be | 1005 | if (err) { |
06b0c886 ZL |
1006 | es->es_lblk = orig_es.es_lblk; |
1007 | es->es_len = orig_es.es_len; | |
e15f742c | 1008 | if ((err == -ENOMEM) && |
edaa53ca | 1009 | __es_shrink(EXT4_SB(inode->i_sb), |
dd475925 | 1010 | 128, EXT4_I(inode))) |
e15f742c | 1011 | goto retry; |
654598be ZL |
1012 | goto out; |
1013 | } | |
1014 | } else { | |
06b0c886 ZL |
1015 | es->es_lblk = end + 1; |
1016 | es->es_len = len2; | |
fdc0212e ZL |
1017 | if (ext4_es_is_written(es) || |
1018 | ext4_es_is_unwritten(es)) { | |
1019 | block = orig_es.es_pblk + orig_es.es_len - len2; | |
1020 | ext4_es_store_pblock(es, block); | |
1021 | } | |
654598be ZL |
1022 | } |
1023 | goto out; | |
1024 | } | |
1025 | ||
1026 | if (len1 > 0) { | |
1027 | node = rb_next(&es->rb_node); | |
1028 | if (node) | |
1029 | es = rb_entry(node, struct extent_status, rb_node); | |
1030 | else | |
1031 | es = NULL; | |
1032 | } | |
1033 | ||
06b0c886 | 1034 | while (es && ext4_es_end(es) <= end) { |
654598be ZL |
1035 | node = rb_next(&es->rb_node); |
1036 | rb_erase(&es->rb_node, &tree->root); | |
bdedbb7b | 1037 | ext4_es_free_extent(inode, es); |
654598be ZL |
1038 | if (!node) { |
1039 | es = NULL; | |
1040 | break; | |
1041 | } | |
1042 | es = rb_entry(node, struct extent_status, rb_node); | |
1043 | } | |
1044 | ||
06b0c886 | 1045 | if (es && es->es_lblk < end + 1) { |
fdc0212e ZL |
1046 | ext4_lblk_t orig_len = es->es_len; |
1047 | ||
06b0c886 ZL |
1048 | len1 = ext4_es_end(es) - end; |
1049 | es->es_lblk = end + 1; | |
1050 | es->es_len = len1; | |
fdc0212e ZL |
1051 | if (ext4_es_is_written(es) || ext4_es_is_unwritten(es)) { |
1052 | block = es->es_pblk + orig_len - len1; | |
1053 | ext4_es_store_pblock(es, block); | |
1054 | } | |
654598be ZL |
1055 | } |
1056 | ||
1057 | out: | |
06b0c886 ZL |
1058 | return err; |
1059 | } | |
1060 | ||
1061 | /* | |
1062 | * ext4_es_remove_extent() removes a space from a extent status tree. | |
1063 | * | |
1064 | * Return 0 on success, error code on failure. | |
1065 | */ | |
1066 | int ext4_es_remove_extent(struct inode *inode, ext4_lblk_t lblk, | |
1067 | ext4_lblk_t len) | |
1068 | { | |
06b0c886 ZL |
1069 | ext4_lblk_t end; |
1070 | int err = 0; | |
1071 | ||
1072 | trace_ext4_es_remove_extent(inode, lblk, len); | |
1073 | es_debug("remove [%u/%u) from extent status tree of inode %lu\n", | |
1074 | lblk, len, inode->i_ino); | |
1075 | ||
d4381472 EG |
1076 | if (!len) |
1077 | return err; | |
1078 | ||
06b0c886 ZL |
1079 | end = lblk + len - 1; |
1080 | BUG_ON(end < lblk); | |
1081 | ||
edaa53ca ZL |
1082 | /* |
1083 | * ext4_clear_inode() depends on us taking i_es_lock unconditionally | |
1084 | * so that we are sure __es_shrink() is done with the inode before it | |
1085 | * is reclaimed. | |
1086 | */ | |
06b0c886 | 1087 | write_lock(&EXT4_I(inode)->i_es_lock); |
bdedbb7b | 1088 | err = __es_remove_extent(inode, lblk, end); |
654598be ZL |
1089 | write_unlock(&EXT4_I(inode)->i_es_lock); |
1090 | ext4_es_print_tree(inode); | |
1091 | return err; | |
1092 | } | |
74cd15cd | 1093 | |
edaa53ca ZL |
1094 | static int __es_shrink(struct ext4_sb_info *sbi, int nr_to_scan, |
1095 | struct ext4_inode_info *locked_ei) | |
74cd15cd | 1096 | { |
74cd15cd | 1097 | struct ext4_inode_info *ei; |
eb68d0e2 | 1098 | struct ext4_es_stats *es_stats; |
eb68d0e2 ZL |
1099 | ktime_t start_time; |
1100 | u64 scan_time; | |
edaa53ca | 1101 | int nr_to_walk; |
1ab6c499 | 1102 | int nr_shrunk = 0; |
edaa53ca | 1103 | int retried = 0, nr_skipped = 0; |
74cd15cd | 1104 | |
eb68d0e2 ZL |
1105 | es_stats = &sbi->s_es_stats; |
1106 | start_time = ktime_get(); | |
d3922a77 | 1107 | |
7869a4a6 | 1108 | retry: |
edaa53ca ZL |
1109 | spin_lock(&sbi->s_es_lock); |
1110 | nr_to_walk = sbi->s_es_nr_inode; | |
1111 | while (nr_to_walk-- > 0) { | |
edaa53ca ZL |
1112 | if (list_empty(&sbi->s_es_list)) { |
1113 | spin_unlock(&sbi->s_es_lock); | |
1114 | goto out; | |
1115 | } | |
1116 | ei = list_first_entry(&sbi->s_es_list, struct ext4_inode_info, | |
1117 | i_es_list); | |
1118 | /* Move the inode to the tail */ | |
dd475925 | 1119 | list_move_tail(&ei->i_es_list, &sbi->s_es_list); |
74cd15cd | 1120 | |
7869a4a6 | 1121 | /* |
edaa53ca ZL |
1122 | * Normally we try hard to avoid shrinking precached inodes, |
1123 | * but we will as a last resort. | |
7869a4a6 | 1124 | */ |
edaa53ca ZL |
1125 | if (!retried && ext4_test_inode_state(&ei->vfs_inode, |
1126 | EXT4_STATE_EXT_PRECACHED)) { | |
7869a4a6 | 1127 | nr_skipped++; |
74cd15cd ZL |
1128 | continue; |
1129 | } | |
d3922a77 | 1130 | |
edaa53ca ZL |
1131 | if (ei == locked_ei || !write_trylock(&ei->i_es_lock)) { |
1132 | nr_skipped++; | |
d3922a77 | 1133 | continue; |
edaa53ca ZL |
1134 | } |
1135 | /* | |
1136 | * Now we hold i_es_lock which protects us from inode reclaim | |
1137 | * freeing inode under us | |
1138 | */ | |
1139 | spin_unlock(&sbi->s_es_lock); | |
74cd15cd | 1140 | |
dd475925 | 1141 | nr_shrunk += es_reclaim_extents(ei, &nr_to_scan); |
74cd15cd ZL |
1142 | write_unlock(&ei->i_es_lock); |
1143 | ||
dd475925 | 1144 | if (nr_to_scan <= 0) |
edaa53ca ZL |
1145 | goto out; |
1146 | spin_lock(&sbi->s_es_lock); | |
74cd15cd | 1147 | } |
edaa53ca | 1148 | spin_unlock(&sbi->s_es_lock); |
7869a4a6 TT |
1149 | |
1150 | /* | |
1151 | * If we skipped any inodes, and we weren't able to make any | |
edaa53ca | 1152 | * forward progress, try again to scan precached inodes. |
7869a4a6 TT |
1153 | */ |
1154 | if ((nr_shrunk == 0) && nr_skipped && !retried) { | |
1155 | retried++; | |
7869a4a6 TT |
1156 | goto retry; |
1157 | } | |
1158 | ||
e15f742c | 1159 | if (locked_ei && nr_shrunk == 0) |
dd475925 | 1160 | nr_shrunk = es_reclaim_extents(locked_ei, &nr_to_scan); |
e15f742c | 1161 | |
edaa53ca | 1162 | out: |
eb68d0e2 ZL |
1163 | scan_time = ktime_to_ns(ktime_sub(ktime_get(), start_time)); |
1164 | if (likely(es_stats->es_stats_scan_time)) | |
1165 | es_stats->es_stats_scan_time = (scan_time + | |
1166 | es_stats->es_stats_scan_time*3) / 4; | |
1167 | else | |
1168 | es_stats->es_stats_scan_time = scan_time; | |
1169 | if (scan_time > es_stats->es_stats_max_scan_time) | |
1170 | es_stats->es_stats_max_scan_time = scan_time; | |
1171 | if (likely(es_stats->es_stats_shrunk)) | |
1172 | es_stats->es_stats_shrunk = (nr_shrunk + | |
1173 | es_stats->es_stats_shrunk*3) / 4; | |
1174 | else | |
1175 | es_stats->es_stats_shrunk = nr_shrunk; | |
1176 | ||
edaa53ca | 1177 | trace_ext4_es_shrink(sbi->s_sb, nr_shrunk, scan_time, |
eb68d0e2 | 1178 | nr_skipped, retried); |
e15f742c TT |
1179 | return nr_shrunk; |
1180 | } | |
1181 | ||
1ab6c499 DC |
1182 | static unsigned long ext4_es_count(struct shrinker *shrink, |
1183 | struct shrink_control *sc) | |
1184 | { | |
1185 | unsigned long nr; | |
1186 | struct ext4_sb_info *sbi; | |
1187 | ||
1188 | sbi = container_of(shrink, struct ext4_sb_info, s_es_shrinker); | |
edaa53ca | 1189 | nr = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt); |
e963bb1d | 1190 | trace_ext4_es_shrink_count(sbi->s_sb, sc->nr_to_scan, nr); |
1ab6c499 DC |
1191 | return nr; |
1192 | } | |
1193 | ||
1194 | static unsigned long ext4_es_scan(struct shrinker *shrink, | |
1195 | struct shrink_control *sc) | |
e15f742c TT |
1196 | { |
1197 | struct ext4_sb_info *sbi = container_of(shrink, | |
1198 | struct ext4_sb_info, s_es_shrinker); | |
1199 | int nr_to_scan = sc->nr_to_scan; | |
1200 | int ret, nr_shrunk; | |
1201 | ||
edaa53ca | 1202 | ret = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt); |
e963bb1d | 1203 | trace_ext4_es_shrink_scan_enter(sbi->s_sb, nr_to_scan, ret); |
e15f742c TT |
1204 | |
1205 | if (!nr_to_scan) | |
1206 | return ret; | |
1207 | ||
edaa53ca | 1208 | nr_shrunk = __es_shrink(sbi, nr_to_scan, NULL); |
e15f742c | 1209 | |
e963bb1d | 1210 | trace_ext4_es_shrink_scan_exit(sbi->s_sb, nr_shrunk, ret); |
1ab6c499 | 1211 | return nr_shrunk; |
74cd15cd ZL |
1212 | } |
1213 | ||
ebd173be | 1214 | int ext4_seq_es_shrinker_info_show(struct seq_file *seq, void *v) |
74cd15cd | 1215 | { |
ebd173be | 1216 | struct ext4_sb_info *sbi = EXT4_SB((struct super_block *) seq->private); |
eb68d0e2 ZL |
1217 | struct ext4_es_stats *es_stats = &sbi->s_es_stats; |
1218 | struct ext4_inode_info *ei, *max = NULL; | |
1219 | unsigned int inode_cnt = 0; | |
1220 | ||
1221 | if (v != SEQ_START_TOKEN) | |
1222 | return 0; | |
1223 | ||
1224 | /* here we just find an inode that has the max nr. of objects */ | |
edaa53ca ZL |
1225 | spin_lock(&sbi->s_es_lock); |
1226 | list_for_each_entry(ei, &sbi->s_es_list, i_es_list) { | |
eb68d0e2 ZL |
1227 | inode_cnt++; |
1228 | if (max && max->i_es_all_nr < ei->i_es_all_nr) | |
1229 | max = ei; | |
1230 | else if (!max) | |
1231 | max = ei; | |
1232 | } | |
edaa53ca | 1233 | spin_unlock(&sbi->s_es_lock); |
eb68d0e2 ZL |
1234 | |
1235 | seq_printf(seq, "stats:\n %lld objects\n %lld reclaimable objects\n", | |
1236 | percpu_counter_sum_positive(&es_stats->es_stats_all_cnt), | |
edaa53ca | 1237 | percpu_counter_sum_positive(&es_stats->es_stats_shk_cnt)); |
eb68d0e2 ZL |
1238 | seq_printf(seq, " %lu/%lu cache hits/misses\n", |
1239 | es_stats->es_stats_cache_hits, | |
1240 | es_stats->es_stats_cache_misses); | |
eb68d0e2 | 1241 | if (inode_cnt) |
edaa53ca | 1242 | seq_printf(seq, " %d inodes on list\n", inode_cnt); |
eb68d0e2 ZL |
1243 | |
1244 | seq_printf(seq, "average:\n %llu us scan time\n", | |
1245 | div_u64(es_stats->es_stats_scan_time, 1000)); | |
1246 | seq_printf(seq, " %lu shrunk objects\n", es_stats->es_stats_shrunk); | |
1247 | if (inode_cnt) | |
1248 | seq_printf(seq, | |
1249 | "maximum:\n %lu inode (%u objects, %u reclaimable)\n" | |
1250 | " %llu us max scan time\n", | |
edaa53ca | 1251 | max->vfs_inode.i_ino, max->i_es_all_nr, max->i_es_shk_nr, |
eb68d0e2 ZL |
1252 | div_u64(es_stats->es_stats_max_scan_time, 1000)); |
1253 | ||
1254 | return 0; | |
1255 | } | |
1256 | ||
eb68d0e2 | 1257 | int ext4_es_register_shrinker(struct ext4_sb_info *sbi) |
74cd15cd | 1258 | { |
eb68d0e2 ZL |
1259 | int err; |
1260 | ||
624d0f1d JK |
1261 | /* Make sure we have enough bits for physical block number */ |
1262 | BUILD_BUG_ON(ES_SHIFT < 48); | |
edaa53ca ZL |
1263 | INIT_LIST_HEAD(&sbi->s_es_list); |
1264 | sbi->s_es_nr_inode = 0; | |
1265 | spin_lock_init(&sbi->s_es_lock); | |
eb68d0e2 ZL |
1266 | sbi->s_es_stats.es_stats_shrunk = 0; |
1267 | sbi->s_es_stats.es_stats_cache_hits = 0; | |
1268 | sbi->s_es_stats.es_stats_cache_misses = 0; | |
1269 | sbi->s_es_stats.es_stats_scan_time = 0; | |
1270 | sbi->s_es_stats.es_stats_max_scan_time = 0; | |
c2661b80 | 1271 | err = percpu_counter_init(&sbi->s_es_stats.es_stats_all_cnt, 0, GFP_KERNEL); |
eb68d0e2 ZL |
1272 | if (err) |
1273 | return err; | |
edaa53ca | 1274 | err = percpu_counter_init(&sbi->s_es_stats.es_stats_shk_cnt, 0, GFP_KERNEL); |
eb68d0e2 ZL |
1275 | if (err) |
1276 | goto err1; | |
1277 | ||
1ab6c499 DC |
1278 | sbi->s_es_shrinker.scan_objects = ext4_es_scan; |
1279 | sbi->s_es_shrinker.count_objects = ext4_es_count; | |
74cd15cd | 1280 | sbi->s_es_shrinker.seeks = DEFAULT_SEEKS; |
eb68d0e2 ZL |
1281 | err = register_shrinker(&sbi->s_es_shrinker); |
1282 | if (err) | |
1283 | goto err2; | |
1284 | ||
eb68d0e2 ZL |
1285 | return 0; |
1286 | ||
1287 | err2: | |
edaa53ca | 1288 | percpu_counter_destroy(&sbi->s_es_stats.es_stats_shk_cnt); |
eb68d0e2 ZL |
1289 | err1: |
1290 | percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt); | |
1291 | return err; | |
74cd15cd ZL |
1292 | } |
1293 | ||
d3922a77 | 1294 | void ext4_es_unregister_shrinker(struct ext4_sb_info *sbi) |
74cd15cd | 1295 | { |
eb68d0e2 | 1296 | percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt); |
edaa53ca | 1297 | percpu_counter_destroy(&sbi->s_es_stats.es_stats_shk_cnt); |
d3922a77 | 1298 | unregister_shrinker(&sbi->s_es_shrinker); |
74cd15cd ZL |
1299 | } |
1300 | ||
dd475925 JK |
1301 | /* |
1302 | * Shrink extents in given inode from ei->i_es_shrink_lblk till end. Scan at | |
1303 | * most *nr_to_scan extents, update *nr_to_scan accordingly. | |
1304 | * | |
1305 | * Return 0 if we hit end of tree / interval, 1 if we exhausted nr_to_scan. | |
1306 | * Increment *nr_shrunk by the number of reclaimed extents. Also update | |
1307 | * ei->i_es_shrink_lblk to where we should continue scanning. | |
1308 | */ | |
1309 | static int es_do_reclaim_extents(struct ext4_inode_info *ei, ext4_lblk_t end, | |
1310 | int *nr_to_scan, int *nr_shrunk) | |
74cd15cd ZL |
1311 | { |
1312 | struct inode *inode = &ei->vfs_inode; | |
1313 | struct ext4_es_tree *tree = &ei->i_es_tree; | |
74cd15cd | 1314 | struct extent_status *es; |
dd475925 | 1315 | struct rb_node *node; |
74cd15cd | 1316 | |
dd475925 JK |
1317 | es = __es_tree_search(&tree->root, ei->i_es_shrink_lblk); |
1318 | if (!es) | |
1319 | goto out_wrap; | |
1320 | node = &es->rb_node; | |
1321 | while (*nr_to_scan > 0) { | |
1322 | if (es->es_lblk > end) { | |
1323 | ei->i_es_shrink_lblk = end + 1; | |
1324 | return 0; | |
1325 | } | |
7869a4a6 | 1326 | |
dd475925 | 1327 | (*nr_to_scan)--; |
74cd15cd ZL |
1328 | node = rb_next(&es->rb_node); |
1329 | /* | |
1330 | * We can't reclaim delayed extent from status tree because | |
1331 | * fiemap, bigallic, and seek_data/hole need to use it. | |
1332 | */ | |
2be12de9 JK |
1333 | if (ext4_es_is_delayed(es)) |
1334 | goto next; | |
1335 | if (ext4_es_is_referenced(es)) { | |
1336 | ext4_es_clear_referenced(es); | |
1337 | goto next; | |
74cd15cd | 1338 | } |
2be12de9 JK |
1339 | |
1340 | rb_erase(&es->rb_node, &tree->root); | |
1341 | ext4_es_free_extent(inode, es); | |
1342 | (*nr_shrunk)++; | |
1343 | next: | |
dd475925 JK |
1344 | if (!node) |
1345 | goto out_wrap; | |
1346 | es = rb_entry(node, struct extent_status, rb_node); | |
74cd15cd | 1347 | } |
dd475925 JK |
1348 | ei->i_es_shrink_lblk = es->es_lblk; |
1349 | return 1; | |
1350 | out_wrap: | |
1351 | ei->i_es_shrink_lblk = 0; | |
1352 | return 0; | |
1353 | } | |
1354 | ||
1355 | static int es_reclaim_extents(struct ext4_inode_info *ei, int *nr_to_scan) | |
1356 | { | |
1357 | struct inode *inode = &ei->vfs_inode; | |
1358 | int nr_shrunk = 0; | |
1359 | ext4_lblk_t start = ei->i_es_shrink_lblk; | |
1360 | static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL, | |
1361 | DEFAULT_RATELIMIT_BURST); | |
1362 | ||
1363 | if (ei->i_es_shk_nr == 0) | |
1364 | return 0; | |
1365 | ||
1366 | if (ext4_test_inode_state(inode, EXT4_STATE_EXT_PRECACHED) && | |
1367 | __ratelimit(&_rs)) | |
1368 | ext4_warning(inode->i_sb, "forced shrink of precached extents"); | |
1369 | ||
1370 | if (!es_do_reclaim_extents(ei, EXT_MAX_BLOCKS, nr_to_scan, &nr_shrunk) && | |
1371 | start != 0) | |
1372 | es_do_reclaim_extents(ei, start - 1, nr_to_scan, &nr_shrunk); | |
1373 | ||
1374 | ei->i_es_tree.cache_es = NULL; | |
74cd15cd ZL |
1375 | return nr_shrunk; |
1376 | } | |
1dc0aa46 EW |
1377 | |
1378 | #ifdef ES_DEBUG__ | |
1379 | static void ext4_print_pending_tree(struct inode *inode) | |
1380 | { | |
1381 | struct ext4_pending_tree *tree; | |
1382 | struct rb_node *node; | |
1383 | struct pending_reservation *pr; | |
1384 | ||
1385 | printk(KERN_DEBUG "pending reservations for inode %lu:", inode->i_ino); | |
1386 | tree = &EXT4_I(inode)->i_pending_tree; | |
1387 | node = rb_first(&tree->root); | |
1388 | while (node) { | |
1389 | pr = rb_entry(node, struct pending_reservation, rb_node); | |
1390 | printk(KERN_DEBUG " %u", pr->lclu); | |
1391 | node = rb_next(node); | |
1392 | } | |
1393 | printk(KERN_DEBUG "\n"); | |
1394 | } | |
1395 | #else | |
1396 | #define ext4_print_pending_tree(inode) | |
1397 | #endif | |
1398 | ||
1399 | int __init ext4_init_pending(void) | |
1400 | { | |
1401 | ext4_pending_cachep = kmem_cache_create("ext4_pending_reservation", | |
1402 | sizeof(struct pending_reservation), | |
1403 | 0, (SLAB_RECLAIM_ACCOUNT), NULL); | |
1404 | if (ext4_pending_cachep == NULL) | |
1405 | return -ENOMEM; | |
1406 | return 0; | |
1407 | } | |
1408 | ||
1409 | void ext4_exit_pending(void) | |
1410 | { | |
1411 | kmem_cache_destroy(ext4_pending_cachep); | |
1412 | } | |
1413 | ||
1414 | void ext4_init_pending_tree(struct ext4_pending_tree *tree) | |
1415 | { | |
1416 | tree->root = RB_ROOT; | |
1417 | } | |
1418 | ||
1419 | /* | |
1420 | * __get_pending - retrieve a pointer to a pending reservation | |
1421 | * | |
1422 | * @inode - file containing the pending cluster reservation | |
1423 | * @lclu - logical cluster of interest | |
1424 | * | |
1425 | * Returns a pointer to a pending reservation if it's a member of | |
1426 | * the set, and NULL if not. Must be called holding i_es_lock. | |
1427 | */ | |
1428 | static struct pending_reservation *__get_pending(struct inode *inode, | |
1429 | ext4_lblk_t lclu) | |
1430 | { | |
1431 | struct ext4_pending_tree *tree; | |
1432 | struct rb_node *node; | |
1433 | struct pending_reservation *pr = NULL; | |
1434 | ||
1435 | tree = &EXT4_I(inode)->i_pending_tree; | |
1436 | node = (&tree->root)->rb_node; | |
1437 | ||
1438 | while (node) { | |
1439 | pr = rb_entry(node, struct pending_reservation, rb_node); | |
1440 | if (lclu < pr->lclu) | |
1441 | node = node->rb_left; | |
1442 | else if (lclu > pr->lclu) | |
1443 | node = node->rb_right; | |
1444 | else if (lclu == pr->lclu) | |
1445 | return pr; | |
1446 | } | |
1447 | return NULL; | |
1448 | } | |
1449 | ||
1450 | /* | |
1451 | * __insert_pending - adds a pending cluster reservation to the set of | |
1452 | * pending reservations | |
1453 | * | |
1454 | * @inode - file containing the cluster | |
1455 | * @lblk - logical block in the cluster to be added | |
1456 | * | |
1457 | * Returns 0 on successful insertion and -ENOMEM on failure. If the | |
1458 | * pending reservation is already in the set, returns successfully. | |
1459 | */ | |
1460 | static int __insert_pending(struct inode *inode, ext4_lblk_t lblk) | |
1461 | { | |
1462 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1463 | struct ext4_pending_tree *tree = &EXT4_I(inode)->i_pending_tree; | |
1464 | struct rb_node **p = &tree->root.rb_node; | |
1465 | struct rb_node *parent = NULL; | |
1466 | struct pending_reservation *pr; | |
1467 | ext4_lblk_t lclu; | |
1468 | int ret = 0; | |
1469 | ||
1470 | lclu = EXT4_B2C(sbi, lblk); | |
1471 | /* search to find parent for insertion */ | |
1472 | while (*p) { | |
1473 | parent = *p; | |
1474 | pr = rb_entry(parent, struct pending_reservation, rb_node); | |
1475 | ||
1476 | if (lclu < pr->lclu) { | |
1477 | p = &(*p)->rb_left; | |
1478 | } else if (lclu > pr->lclu) { | |
1479 | p = &(*p)->rb_right; | |
1480 | } else { | |
1481 | /* pending reservation already inserted */ | |
1482 | goto out; | |
1483 | } | |
1484 | } | |
1485 | ||
1486 | pr = kmem_cache_alloc(ext4_pending_cachep, GFP_ATOMIC); | |
1487 | if (pr == NULL) { | |
1488 | ret = -ENOMEM; | |
1489 | goto out; | |
1490 | } | |
1491 | pr->lclu = lclu; | |
1492 | ||
1493 | rb_link_node(&pr->rb_node, parent, p); | |
1494 | rb_insert_color(&pr->rb_node, &tree->root); | |
1495 | ||
1496 | out: | |
1497 | return ret; | |
1498 | } | |
1499 | ||
1500 | /* | |
1501 | * __remove_pending - removes a pending cluster reservation from the set | |
1502 | * of pending reservations | |
1503 | * | |
1504 | * @inode - file containing the cluster | |
1505 | * @lblk - logical block in the pending cluster reservation to be removed | |
1506 | * | |
1507 | * Returns successfully if pending reservation is not a member of the set. | |
1508 | */ | |
1509 | static void __remove_pending(struct inode *inode, ext4_lblk_t lblk) | |
1510 | { | |
1511 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1512 | struct pending_reservation *pr; | |
1513 | struct ext4_pending_tree *tree; | |
1514 | ||
1515 | pr = __get_pending(inode, EXT4_B2C(sbi, lblk)); | |
1516 | if (pr != NULL) { | |
1517 | tree = &EXT4_I(inode)->i_pending_tree; | |
1518 | rb_erase(&pr->rb_node, &tree->root); | |
1519 | kmem_cache_free(ext4_pending_cachep, pr); | |
1520 | } | |
1521 | } | |
1522 | ||
1523 | /* | |
1524 | * ext4_remove_pending - removes a pending cluster reservation from the set | |
1525 | * of pending reservations | |
1526 | * | |
1527 | * @inode - file containing the cluster | |
1528 | * @lblk - logical block in the pending cluster reservation to be removed | |
1529 | * | |
1530 | * Locking for external use of __remove_pending. | |
1531 | */ | |
1532 | void ext4_remove_pending(struct inode *inode, ext4_lblk_t lblk) | |
1533 | { | |
1534 | struct ext4_inode_info *ei = EXT4_I(inode); | |
1535 | ||
1536 | write_lock(&ei->i_es_lock); | |
1537 | __remove_pending(inode, lblk); | |
1538 | write_unlock(&ei->i_es_lock); | |
1539 | } | |
1540 | ||
1541 | /* | |
1542 | * ext4_is_pending - determine whether a cluster has a pending reservation | |
1543 | * on it | |
1544 | * | |
1545 | * @inode - file containing the cluster | |
1546 | * @lblk - logical block in the cluster | |
1547 | * | |
1548 | * Returns true if there's a pending reservation for the cluster in the | |
1549 | * set of pending reservations, and false if not. | |
1550 | */ | |
1551 | bool ext4_is_pending(struct inode *inode, ext4_lblk_t lblk) | |
1552 | { | |
1553 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1554 | struct ext4_inode_info *ei = EXT4_I(inode); | |
1555 | bool ret; | |
1556 | ||
1557 | read_lock(&ei->i_es_lock); | |
1558 | ret = (bool)(__get_pending(inode, EXT4_B2C(sbi, lblk)) != NULL); | |
1559 | read_unlock(&ei->i_es_lock); | |
1560 | ||
1561 | return ret; | |
1562 | } | |
0b02f4c0 EW |
1563 | |
1564 | /* | |
1565 | * ext4_es_insert_delayed_block - adds a delayed block to the extents status | |
1566 | * tree, adding a pending reservation where | |
1567 | * needed | |
1568 | * | |
1569 | * @inode - file containing the newly added block | |
1570 | * @lblk - logical block to be added | |
1571 | * @allocated - indicates whether a physical cluster has been allocated for | |
1572 | * the logical cluster that contains the block | |
1573 | * | |
1574 | * Returns 0 on success, negative error code on failure. | |
1575 | */ | |
1576 | int ext4_es_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk, | |
1577 | bool allocated) | |
1578 | { | |
1579 | struct extent_status newes; | |
1580 | int err = 0; | |
1581 | ||
1582 | es_debug("add [%u/1) delayed to extent status tree of inode %lu\n", | |
1583 | lblk, inode->i_ino); | |
1584 | ||
1585 | newes.es_lblk = lblk; | |
1586 | newes.es_len = 1; | |
1587 | ext4_es_store_pblock_status(&newes, ~0, EXTENT_STATUS_DELAYED); | |
1588 | trace_ext4_es_insert_delayed_block(inode, &newes, allocated); | |
1589 | ||
1590 | ext4_es_insert_extent_check(inode, &newes); | |
1591 | ||
1592 | write_lock(&EXT4_I(inode)->i_es_lock); | |
1593 | ||
1594 | err = __es_remove_extent(inode, lblk, lblk); | |
1595 | if (err != 0) | |
1596 | goto error; | |
1597 | retry: | |
1598 | err = __es_insert_extent(inode, &newes); | |
1599 | if (err == -ENOMEM && __es_shrink(EXT4_SB(inode->i_sb), | |
1600 | 128, EXT4_I(inode))) | |
1601 | goto retry; | |
1602 | if (err != 0) | |
1603 | goto error; | |
1604 | ||
1605 | if (allocated) | |
1606 | __insert_pending(inode, lblk); | |
1607 | ||
1608 | error: | |
1609 | write_unlock(&EXT4_I(inode)->i_es_lock); | |
1610 | ||
1611 | ext4_es_print_tree(inode); | |
1612 | ext4_print_pending_tree(inode); | |
1613 | ||
1614 | return err; | |
1615 | } | |
b6bf9171 EW |
1616 | |
1617 | /* | |
1618 | * __es_delayed_clu - count number of clusters containing blocks that | |
1619 | * are delayed only | |
1620 | * | |
1621 | * @inode - file containing block range | |
1622 | * @start - logical block defining start of range | |
1623 | * @end - logical block defining end of range | |
1624 | * | |
1625 | * Returns the number of clusters containing only delayed (not delayed | |
1626 | * and unwritten) blocks in the range specified by @start and @end. Any | |
1627 | * cluster or part of a cluster within the range and containing a delayed | |
1628 | * and not unwritten block within the range is counted as a whole cluster. | |
1629 | */ | |
1630 | static unsigned int __es_delayed_clu(struct inode *inode, ext4_lblk_t start, | |
1631 | ext4_lblk_t end) | |
1632 | { | |
1633 | struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree; | |
1634 | struct extent_status *es; | |
1635 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1636 | struct rb_node *node; | |
1637 | ext4_lblk_t first_lclu, last_lclu; | |
1638 | unsigned long long last_counted_lclu; | |
1639 | unsigned int n = 0; | |
1640 | ||
1641 | /* guaranteed to be unequal to any ext4_lblk_t value */ | |
1642 | last_counted_lclu = ~0ULL; | |
1643 | ||
1644 | es = __es_tree_search(&tree->root, start); | |
1645 | ||
1646 | while (es && (es->es_lblk <= end)) { | |
1647 | if (ext4_es_is_delonly(es)) { | |
1648 | if (es->es_lblk <= start) | |
1649 | first_lclu = EXT4_B2C(sbi, start); | |
1650 | else | |
1651 | first_lclu = EXT4_B2C(sbi, es->es_lblk); | |
1652 | ||
1653 | if (ext4_es_end(es) >= end) | |
1654 | last_lclu = EXT4_B2C(sbi, end); | |
1655 | else | |
1656 | last_lclu = EXT4_B2C(sbi, ext4_es_end(es)); | |
1657 | ||
1658 | if (first_lclu == last_counted_lclu) | |
1659 | n += last_lclu - first_lclu; | |
1660 | else | |
1661 | n += last_lclu - first_lclu + 1; | |
1662 | last_counted_lclu = last_lclu; | |
1663 | } | |
1664 | node = rb_next(&es->rb_node); | |
1665 | if (!node) | |
1666 | break; | |
1667 | es = rb_entry(node, struct extent_status, rb_node); | |
1668 | } | |
1669 | ||
1670 | return n; | |
1671 | } | |
1672 | ||
1673 | /* | |
1674 | * ext4_es_delayed_clu - count number of clusters containing blocks that | |
1675 | * are both delayed and unwritten | |
1676 | * | |
1677 | * @inode - file containing block range | |
1678 | * @lblk - logical block defining start of range | |
1679 | * @len - number of blocks in range | |
1680 | * | |
1681 | * Locking for external use of __es_delayed_clu(). | |
1682 | */ | |
1683 | unsigned int ext4_es_delayed_clu(struct inode *inode, ext4_lblk_t lblk, | |
1684 | ext4_lblk_t len) | |
1685 | { | |
1686 | struct ext4_inode_info *ei = EXT4_I(inode); | |
1687 | ext4_lblk_t end; | |
1688 | unsigned int n; | |
1689 | ||
1690 | if (len == 0) | |
1691 | return 0; | |
1692 | ||
1693 | end = lblk + len - 1; | |
1694 | WARN_ON(end < lblk); | |
1695 | ||
1696 | read_lock(&ei->i_es_lock); | |
1697 | ||
1698 | n = __es_delayed_clu(inode, lblk, end); | |
1699 | ||
1700 | read_unlock(&ei->i_es_lock); | |
1701 | ||
1702 | return n; | |
1703 | } | |
1704 | ||
1705 | /* | |
1706 | * __revise_pending - makes, cancels, or leaves unchanged pending cluster | |
1707 | * reservations for a specified block range depending | |
1708 | * upon the presence or absence of delayed blocks | |
1709 | * outside the range within clusters at the ends of the | |
1710 | * range | |
1711 | * | |
1712 | * @inode - file containing the range | |
1713 | * @lblk - logical block defining the start of range | |
1714 | * @len - length of range in blocks | |
1715 | * | |
1716 | * Used after a newly allocated extent is added to the extents status tree. | |
1717 | * Requires that the extents in the range have either written or unwritten | |
1718 | * status. Must be called while holding i_es_lock. | |
1719 | */ | |
1720 | static void __revise_pending(struct inode *inode, ext4_lblk_t lblk, | |
1721 | ext4_lblk_t len) | |
1722 | { | |
1723 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1724 | ext4_lblk_t end = lblk + len - 1; | |
1725 | ext4_lblk_t first, last; | |
1726 | bool f_del = false, l_del = false; | |
1727 | ||
1728 | if (len == 0) | |
1729 | return; | |
1730 | ||
1731 | /* | |
1732 | * Two cases - block range within single cluster and block range | |
1733 | * spanning two or more clusters. Note that a cluster belonging | |
1734 | * to a range starting and/or ending on a cluster boundary is treated | |
1735 | * as if it does not contain a delayed extent. The new range may | |
1736 | * have allocated space for previously delayed blocks out to the | |
1737 | * cluster boundary, requiring that any pre-existing pending | |
1738 | * reservation be canceled. Because this code only looks at blocks | |
1739 | * outside the range, it should revise pending reservations | |
1740 | * correctly even if the extent represented by the range can't be | |
1741 | * inserted in the extents status tree due to ENOSPC. | |
1742 | */ | |
1743 | ||
1744 | if (EXT4_B2C(sbi, lblk) == EXT4_B2C(sbi, end)) { | |
1745 | first = EXT4_LBLK_CMASK(sbi, lblk); | |
1746 | if (first != lblk) | |
1747 | f_del = __es_scan_range(inode, &ext4_es_is_delonly, | |
1748 | first, lblk - 1); | |
1749 | if (f_del) { | |
1750 | __insert_pending(inode, first); | |
1751 | } else { | |
1752 | last = EXT4_LBLK_CMASK(sbi, end) + | |
1753 | sbi->s_cluster_ratio - 1; | |
1754 | if (last != end) | |
1755 | l_del = __es_scan_range(inode, | |
1756 | &ext4_es_is_delonly, | |
1757 | end + 1, last); | |
1758 | if (l_del) | |
1759 | __insert_pending(inode, last); | |
1760 | else | |
1761 | __remove_pending(inode, last); | |
1762 | } | |
1763 | } else { | |
1764 | first = EXT4_LBLK_CMASK(sbi, lblk); | |
1765 | if (first != lblk) | |
1766 | f_del = __es_scan_range(inode, &ext4_es_is_delonly, | |
1767 | first, lblk - 1); | |
1768 | if (f_del) | |
1769 | __insert_pending(inode, first); | |
1770 | else | |
1771 | __remove_pending(inode, first); | |
1772 | ||
1773 | last = EXT4_LBLK_CMASK(sbi, end) + sbi->s_cluster_ratio - 1; | |
1774 | if (last != end) | |
1775 | l_del = __es_scan_range(inode, &ext4_es_is_delonly, | |
1776 | end + 1, last); | |
1777 | if (l_del) | |
1778 | __insert_pending(inode, last); | |
1779 | else | |
1780 | __remove_pending(inode, last); | |
1781 | } | |
1782 | } | |
f456767d EW |
1783 | |
1784 | /* | |
1785 | * ext4_es_remove_blks - remove block range from extents status tree and | |
1786 | * reduce reservation count or cancel pending | |
1787 | * reservation as needed | |
1788 | * | |
1789 | * @inode - file containing range | |
1790 | * @lblk - first block in range | |
1791 | * @len - number of blocks to remove | |
1792 | * | |
1793 | */ | |
1794 | void ext4_es_remove_blks(struct inode *inode, ext4_lblk_t lblk, | |
1795 | ext4_lblk_t len) | |
1796 | { | |
1797 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1798 | unsigned int clu_size, reserved = 0; | |
1799 | ext4_lblk_t last_lclu, first, length, remainder, last; | |
1800 | bool delonly; | |
1801 | int err = 0; | |
1802 | struct pending_reservation *pr; | |
1803 | struct ext4_pending_tree *tree; | |
1804 | ||
1805 | /* | |
1806 | * Process cluster by cluster for bigalloc - there may be up to | |
1807 | * two clusters in a 4k page with a 1k block size and two blocks | |
1808 | * per cluster. Also necessary for systems with larger page sizes | |
1809 | * and potentially larger block sizes. | |
1810 | */ | |
1811 | clu_size = sbi->s_cluster_ratio; | |
1812 | last_lclu = EXT4_B2C(sbi, lblk + len - 1); | |
1813 | ||
1814 | write_lock(&EXT4_I(inode)->i_es_lock); | |
1815 | ||
1816 | for (first = lblk, remainder = len; | |
1817 | remainder > 0; | |
1818 | first += length, remainder -= length) { | |
1819 | ||
1820 | if (EXT4_B2C(sbi, first) == last_lclu) | |
1821 | length = remainder; | |
1822 | else | |
1823 | length = clu_size - EXT4_LBLK_COFF(sbi, first); | |
1824 | ||
1825 | /* | |
1826 | * The BH_Delay flag, which triggers calls to this function, | |
1827 | * and the contents of the extents status tree can be | |
1828 | * inconsistent due to writepages activity. So, note whether | |
1829 | * the blocks to be removed actually belong to an extent with | |
1830 | * delayed only status. | |
1831 | */ | |
1832 | delonly = __es_scan_clu(inode, &ext4_es_is_delonly, first); | |
1833 | ||
1834 | /* | |
1835 | * because of the writepages effect, written and unwritten | |
1836 | * blocks could be removed here | |
1837 | */ | |
1838 | last = first + length - 1; | |
1839 | err = __es_remove_extent(inode, first, last); | |
1840 | if (err) | |
1841 | ext4_warning(inode->i_sb, | |
1842 | "%s: couldn't remove page (err = %d)", | |
1843 | __func__, err); | |
1844 | ||
1845 | /* non-bigalloc case: simply count the cluster for release */ | |
1846 | if (sbi->s_cluster_ratio == 1 && delonly) { | |
1847 | reserved++; | |
1848 | continue; | |
1849 | } | |
1850 | ||
1851 | /* | |
1852 | * bigalloc case: if all delayed allocated only blocks have | |
1853 | * just been removed from a cluster, either cancel a pending | |
1854 | * reservation if it exists or count a cluster for release | |
1855 | */ | |
1856 | if (delonly && | |
1857 | !__es_scan_clu(inode, &ext4_es_is_delonly, first)) { | |
1858 | pr = __get_pending(inode, EXT4_B2C(sbi, first)); | |
1859 | if (pr != NULL) { | |
1860 | tree = &EXT4_I(inode)->i_pending_tree; | |
1861 | rb_erase(&pr->rb_node, &tree->root); | |
1862 | kmem_cache_free(ext4_pending_cachep, pr); | |
1863 | } else { | |
1864 | reserved++; | |
1865 | } | |
1866 | } | |
1867 | } | |
1868 | ||
1869 | write_unlock(&EXT4_I(inode)->i_es_lock); | |
1870 | ||
1871 | ext4_da_release_space(inode, reserved); | |
1872 | } |