1 // SPDX-License-Identifier: GPL-2.0
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 */
13 #include <linux/list_sort.h>
14 #include <linux/proc_fs.h>
15 #include <linux/seq_file.h>
16 #include "ext4.h"
17
18 #include <trace/events/ext4.h>
19
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
27 * extent tree, whose goal is only track delayed extents in memory to
28 * simplify the implementation of fiemap and bigalloc, and introduce
29 * lseek SEEK_DATA/SEEK_HOLE support. That is why it is still called
30 * delay extent tree at the first commit. But for better understand
31 * what it does, it has been rename to extent status tree.
32 *
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
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.
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.
55 */
56
57 /*
58 * Extent status tree implementation for ext4.
59 *
60 *
61 * ==========================================================================
62 * Extent status tree tracks all extent status.
63 *
64 * 1. Why we need to implement extent status tree?
65 *
66 * Without extent status tree, ext4 identifies a delayed extent by looking
67 * up page cache, this has several deficiencies - complicated, buggy,
68 * and inefficient code.
69 *
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.
72 *
73 * Let us have a look at how they do without extent status tree.
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 *
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
88 * time consuming.
89 *
90 * With extent status tree implementation, FIEMAP, SEEK_HOLE/DATA,
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
93 * not by searching the extent tree.
94 *
95 *
96 * ==========================================================================
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.
105 *
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.
110 *
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.
114 *
115 * -- race on a extent status tree
116 * Extent status tree is protected by inode->i_es_lock.
117 *
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.
122 *
123 *
124 * ==========================================================================
125 * 3. Performance analysis
126 *
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
138 *
139 * -- Refactor delayed space reservation
140 *
141 * -- Extent-level locking
142 */
143
144 static struct kmem_cache *ext4_es_cachep;
145 static struct kmem_cache *ext4_pending_cachep;
146
147 static int __es_insert_extent(struct inode *inode, struct extent_status *newes,
148 struct extent_status *prealloc);
149 static int __es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
150 ext4_lblk_t end, int *reserved,
151 struct extent_status *prealloc);
152 static int es_reclaim_extents(struct ext4_inode_info *ei, int *nr_to_scan);
153 static int __es_shrink(struct ext4_sb_info *sbi, int nr_to_scan,
154 struct ext4_inode_info *locked_ei);
155 static int __revise_pending(struct inode *inode, ext4_lblk_t lblk,
156 ext4_lblk_t len,
157 struct pending_reservation **prealloc);
158
ext4_init_es(void)159 int __init ext4_init_es(void)
160 {
161 ext4_es_cachep = KMEM_CACHE(extent_status, SLAB_RECLAIM_ACCOUNT);
162 if (ext4_es_cachep == NULL)
163 return -ENOMEM;
164 return 0;
165 }
166
ext4_exit_es(void)167 void ext4_exit_es(void)
168 {
169 kmem_cache_destroy(ext4_es_cachep);
170 }
171
ext4_es_init_tree(struct ext4_es_tree * tree)172 void ext4_es_init_tree(struct ext4_es_tree *tree)
173 {
174 tree->root = RB_ROOT;
175 tree->cache_es = NULL;
176 }
177
178 #ifdef ES_DEBUG__
ext4_es_print_tree(struct inode * inode)179 static void ext4_es_print_tree(struct inode *inode)
180 {
181 struct ext4_es_tree *tree;
182 struct rb_node *node;
183
184 printk(KERN_DEBUG "status extents for inode %lu:", inode->i_ino);
185 tree = &EXT4_I(inode)->i_es_tree;
186 node = rb_first(&tree->root);
187 while (node) {
188 struct extent_status *es;
189 es = rb_entry(node, struct extent_status, rb_node);
190 printk(KERN_DEBUG " [%u/%u) %llu %x",
191 es->es_lblk, es->es_len,
192 ext4_es_pblock(es), ext4_es_status(es));
193 node = rb_next(node);
194 }
195 printk(KERN_DEBUG "\n");
196 }
197 #else
198 #define ext4_es_print_tree(inode)
199 #endif
200
ext4_es_end(struct extent_status * es)201 static inline ext4_lblk_t ext4_es_end(struct extent_status *es)
202 {
203 BUG_ON(es->es_lblk + es->es_len < es->es_lblk);
204 return es->es_lblk + es->es_len - 1;
205 }
206
207 /*
208 * search through the tree for an delayed extent with a given offset. If
209 * it can't be found, try to find next extent.
210 */
__es_tree_search(struct rb_root * root,ext4_lblk_t lblk)211 static struct extent_status *__es_tree_search(struct rb_root *root,
212 ext4_lblk_t lblk)
213 {
214 struct rb_node *node = root->rb_node;
215 struct extent_status *es = NULL;
216
217 while (node) {
218 es = rb_entry(node, struct extent_status, rb_node);
219 if (lblk < es->es_lblk)
220 node = node->rb_left;
221 else if (lblk > ext4_es_end(es))
222 node = node->rb_right;
223 else
224 return es;
225 }
226
227 if (es && lblk < es->es_lblk)
228 return es;
229
230 if (es && lblk > ext4_es_end(es)) {
231 node = rb_next(&es->rb_node);
232 return node ? rb_entry(node, struct extent_status, rb_node) :
233 NULL;
234 }
235
236 return NULL;
237 }
238
239 /*
240 * ext4_es_find_extent_range - find extent with specified status within block
241 * range or next extent following block range in
242 * extents status tree
243 *
244 * @inode - file containing the range
245 * @matching_fn - pointer to function that matches extents with desired status
246 * @lblk - logical block defining start of range
247 * @end - logical block defining end of range
248 * @es - extent found, if any
249 *
250 * Find the first extent within the block range specified by @lblk and @end
251 * in the extents status tree that satisfies @matching_fn. If a match
252 * is found, it's returned in @es. If not, and a matching extent is found
253 * beyond the block range, it's returned in @es. If no match is found, an
254 * extent is returned in @es whose es_lblk, es_len, and es_pblk components
255 * are 0.
256 */
__es_find_extent_range(struct inode * inode,int (* matching_fn)(struct extent_status * es),ext4_lblk_t lblk,ext4_lblk_t end,struct extent_status * es)257 static void __es_find_extent_range(struct inode *inode,
258 int (*matching_fn)(struct extent_status *es),
259 ext4_lblk_t lblk, ext4_lblk_t end,
260 struct extent_status *es)
261 {
262 struct ext4_es_tree *tree = NULL;
263 struct extent_status *es1 = NULL;
264 struct rb_node *node;
265
266 WARN_ON(es == NULL);
267 WARN_ON(end < lblk);
268
269 tree = &EXT4_I(inode)->i_es_tree;
270
271 /* see if the extent has been cached */
272 es->es_lblk = es->es_len = es->es_pblk = 0;
273 es1 = READ_ONCE(tree->cache_es);
274 if (es1 && in_range(lblk, es1->es_lblk, es1->es_len)) {
275 es_debug("%u cached by [%u/%u) %llu %x\n",
276 lblk, es1->es_lblk, es1->es_len,
277 ext4_es_pblock(es1), ext4_es_status(es1));
278 goto out;
279 }
280
281 es1 = __es_tree_search(&tree->root, lblk);
282
283 out:
284 if (es1 && !matching_fn(es1)) {
285 while ((node = rb_next(&es1->rb_node)) != NULL) {
286 es1 = rb_entry(node, struct extent_status, rb_node);
287 if (es1->es_lblk > end) {
288 es1 = NULL;
289 break;
290 }
291 if (matching_fn(es1))
292 break;
293 }
294 }
295
296 if (es1 && matching_fn(es1)) {
297 WRITE_ONCE(tree->cache_es, es1);
298 es->es_lblk = es1->es_lblk;
299 es->es_len = es1->es_len;
300 es->es_pblk = es1->es_pblk;
301 }
302
303 }
304
305 /*
306 * Locking for __es_find_extent_range() for external use
307 */
ext4_es_find_extent_range(struct inode * inode,int (* matching_fn)(struct extent_status * es),ext4_lblk_t lblk,ext4_lblk_t end,struct extent_status * es)308 void ext4_es_find_extent_range(struct inode *inode,
309 int (*matching_fn)(struct extent_status *es),
310 ext4_lblk_t lblk, ext4_lblk_t end,
311 struct extent_status *es)
312 {
313 es->es_lblk = es->es_len = es->es_pblk = 0;
314
315 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
316 return;
317
318 trace_ext4_es_find_extent_range_enter(inode, lblk);
319
320 read_lock(&EXT4_I(inode)->i_es_lock);
321 __es_find_extent_range(inode, matching_fn, lblk, end, es);
322 read_unlock(&EXT4_I(inode)->i_es_lock);
323
324 trace_ext4_es_find_extent_range_exit(inode, es);
325 }
326
327 /*
328 * __es_scan_range - search block range for block with specified status
329 * in extents status tree
330 *
331 * @inode - file containing the range
332 * @matching_fn - pointer to function that matches extents with desired status
333 * @lblk - logical block defining start of range
334 * @end - logical block defining end of range
335 *
336 * Returns true if at least one block in the specified block range satisfies
337 * the criterion specified by @matching_fn, and false if not. If at least
338 * one extent has the specified status, then there is at least one block
339 * in the cluster with that status. Should only be called by code that has
340 * taken i_es_lock.
341 */
__es_scan_range(struct inode * inode,int (* matching_fn)(struct extent_status * es),ext4_lblk_t start,ext4_lblk_t end)342 static bool __es_scan_range(struct inode *inode,
343 int (*matching_fn)(struct extent_status *es),
344 ext4_lblk_t start, ext4_lblk_t end)
345 {
346 struct extent_status es;
347
348 __es_find_extent_range(inode, matching_fn, start, end, &es);
349 if (es.es_len == 0)
350 return false; /* no matching extent in the tree */
351 else if (es.es_lblk <= start &&
352 start < es.es_lblk + es.es_len)
353 return true;
354 else if (start <= es.es_lblk && es.es_lblk <= end)
355 return true;
356 else
357 return false;
358 }
359 /*
360 * Locking for __es_scan_range() for external use
361 */
ext4_es_scan_range(struct inode * inode,int (* matching_fn)(struct extent_status * es),ext4_lblk_t lblk,ext4_lblk_t end)362 bool ext4_es_scan_range(struct inode *inode,
363 int (*matching_fn)(struct extent_status *es),
364 ext4_lblk_t lblk, ext4_lblk_t end)
365 {
366 bool ret;
367
368 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
369 return false;
370
371 read_lock(&EXT4_I(inode)->i_es_lock);
372 ret = __es_scan_range(inode, matching_fn, lblk, end);
373 read_unlock(&EXT4_I(inode)->i_es_lock);
374
375 return ret;
376 }
377
378 /*
379 * __es_scan_clu - search cluster for block with specified status in
380 * extents status tree
381 *
382 * @inode - file containing the cluster
383 * @matching_fn - pointer to function that matches extents with desired status
384 * @lblk - logical block in cluster to be searched
385 *
386 * Returns true if at least one extent in the cluster containing @lblk
387 * satisfies the criterion specified by @matching_fn, and false if not. If at
388 * least one extent has the specified status, then there is at least one block
389 * in the cluster with that status. Should only be called by code that has
390 * taken i_es_lock.
391 */
__es_scan_clu(struct inode * inode,int (* matching_fn)(struct extent_status * es),ext4_lblk_t lblk)392 static bool __es_scan_clu(struct inode *inode,
393 int (*matching_fn)(struct extent_status *es),
394 ext4_lblk_t lblk)
395 {
396 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
397 ext4_lblk_t lblk_start, lblk_end;
398
399 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
400 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
401
402 return __es_scan_range(inode, matching_fn, lblk_start, lblk_end);
403 }
404
405 /*
406 * Locking for __es_scan_clu() for external use
407 */
ext4_es_scan_clu(struct inode * inode,int (* matching_fn)(struct extent_status * es),ext4_lblk_t lblk)408 bool ext4_es_scan_clu(struct inode *inode,
409 int (*matching_fn)(struct extent_status *es),
410 ext4_lblk_t lblk)
411 {
412 bool ret;
413
414 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
415 return false;
416
417 read_lock(&EXT4_I(inode)->i_es_lock);
418 ret = __es_scan_clu(inode, matching_fn, lblk);
419 read_unlock(&EXT4_I(inode)->i_es_lock);
420
421 return ret;
422 }
423
ext4_es_list_add(struct inode * inode)424 static void ext4_es_list_add(struct inode *inode)
425 {
426 struct ext4_inode_info *ei = EXT4_I(inode);
427 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
428
429 if (!list_empty(&ei->i_es_list))
430 return;
431
432 spin_lock(&sbi->s_es_lock);
433 if (list_empty(&ei->i_es_list)) {
434 list_add_tail(&ei->i_es_list, &sbi->s_es_list);
435 sbi->s_es_nr_inode++;
436 }
437 spin_unlock(&sbi->s_es_lock);
438 }
439
ext4_es_list_del(struct inode * inode)440 static void ext4_es_list_del(struct inode *inode)
441 {
442 struct ext4_inode_info *ei = EXT4_I(inode);
443 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
444
445 spin_lock(&sbi->s_es_lock);
446 if (!list_empty(&ei->i_es_list)) {
447 list_del_init(&ei->i_es_list);
448 sbi->s_es_nr_inode--;
449 WARN_ON_ONCE(sbi->s_es_nr_inode < 0);
450 }
451 spin_unlock(&sbi->s_es_lock);
452 }
453
__alloc_pending(bool nofail)454 static inline struct pending_reservation *__alloc_pending(bool nofail)
455 {
456 if (!nofail)
457 return kmem_cache_alloc(ext4_pending_cachep, GFP_ATOMIC);
458
459 return kmem_cache_zalloc(ext4_pending_cachep, GFP_KERNEL | __GFP_NOFAIL);
460 }
461
__free_pending(struct pending_reservation * pr)462 static inline void __free_pending(struct pending_reservation *pr)
463 {
464 kmem_cache_free(ext4_pending_cachep, pr);
465 }
466
467 /*
468 * Returns true if we cannot fail to allocate memory for this extent_status
469 * entry and cannot reclaim it until its status changes.
470 */
ext4_es_must_keep(struct extent_status * es)471 static inline bool ext4_es_must_keep(struct extent_status *es)
472 {
473 /* fiemap, bigalloc, and seek_data/hole need to use it. */
474 if (ext4_es_is_delayed(es))
475 return true;
476
477 return false;
478 }
479
__es_alloc_extent(bool nofail)480 static inline struct extent_status *__es_alloc_extent(bool nofail)
481 {
482 if (!nofail)
483 return kmem_cache_alloc(ext4_es_cachep, GFP_ATOMIC);
484
485 return kmem_cache_zalloc(ext4_es_cachep, GFP_KERNEL | __GFP_NOFAIL);
486 }
487
ext4_es_init_extent(struct inode * inode,struct extent_status * es,ext4_lblk_t lblk,ext4_lblk_t len,ext4_fsblk_t pblk)488 static void ext4_es_init_extent(struct inode *inode, struct extent_status *es,
489 ext4_lblk_t lblk, ext4_lblk_t len, ext4_fsblk_t pblk)
490 {
491 es->es_lblk = lblk;
492 es->es_len = len;
493 es->es_pblk = pblk;
494
495 /* We never try to reclaim a must kept extent, so we don't count it. */
496 if (!ext4_es_must_keep(es)) {
497 if (!EXT4_I(inode)->i_es_shk_nr++)
498 ext4_es_list_add(inode);
499 percpu_counter_inc(&EXT4_SB(inode->i_sb)->
500 s_es_stats.es_stats_shk_cnt);
501 }
502
503 EXT4_I(inode)->i_es_all_nr++;
504 percpu_counter_inc(&EXT4_SB(inode->i_sb)->s_es_stats.es_stats_all_cnt);
505 }
506
__es_free_extent(struct extent_status * es)507 static inline void __es_free_extent(struct extent_status *es)
508 {
509 kmem_cache_free(ext4_es_cachep, es);
510 }
511
ext4_es_free_extent(struct inode * inode,struct extent_status * es)512 static void ext4_es_free_extent(struct inode *inode, struct extent_status *es)
513 {
514 EXT4_I(inode)->i_es_all_nr--;
515 percpu_counter_dec(&EXT4_SB(inode->i_sb)->s_es_stats.es_stats_all_cnt);
516
517 /* Decrease the shrink counter when we can reclaim the extent. */
518 if (!ext4_es_must_keep(es)) {
519 BUG_ON(EXT4_I(inode)->i_es_shk_nr == 0);
520 if (!--EXT4_I(inode)->i_es_shk_nr)
521 ext4_es_list_del(inode);
522 percpu_counter_dec(&EXT4_SB(inode->i_sb)->
523 s_es_stats.es_stats_shk_cnt);
524 }
525
526 __es_free_extent(es);
527 }
528
529 /*
530 * Check whether or not two extents can be merged
531 * Condition:
532 * - logical block number is contiguous
533 * - physical block number is contiguous
534 * - status is equal
535 */
ext4_es_can_be_merged(struct extent_status * es1,struct extent_status * es2)536 static int ext4_es_can_be_merged(struct extent_status *es1,
537 struct extent_status *es2)
538 {
539 if (ext4_es_type(es1) != ext4_es_type(es2))
540 return 0;
541
542 if (((__u64) es1->es_len) + es2->es_len > EXT_MAX_BLOCKS) {
543 pr_warn("ES assertion failed when merging extents. "
544 "The sum of lengths of es1 (%d) and es2 (%d) "
545 "is bigger than allowed file size (%d)\n",
546 es1->es_len, es2->es_len, EXT_MAX_BLOCKS);
547 WARN_ON(1);
548 return 0;
549 }
550
551 if (((__u64) es1->es_lblk) + es1->es_len != es2->es_lblk)
552 return 0;
553
554 if ((ext4_es_is_written(es1) || ext4_es_is_unwritten(es1)) &&
555 (ext4_es_pblock(es1) + es1->es_len == ext4_es_pblock(es2)))
556 return 1;
557
558 if (ext4_es_is_hole(es1))
559 return 1;
560
561 /* we need to check delayed extent */
562 if (ext4_es_is_delayed(es1))
563 return 1;
564
565 return 0;
566 }
567
568 static struct extent_status *
ext4_es_try_to_merge_left(struct inode * inode,struct extent_status * es)569 ext4_es_try_to_merge_left(struct inode *inode, struct extent_status *es)
570 {
571 struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
572 struct extent_status *es1;
573 struct rb_node *node;
574
575 node = rb_prev(&es->rb_node);
576 if (!node)
577 return es;
578
579 es1 = rb_entry(node, struct extent_status, rb_node);
580 if (ext4_es_can_be_merged(es1, es)) {
581 es1->es_len += es->es_len;
582 if (ext4_es_is_referenced(es))
583 ext4_es_set_referenced(es1);
584 rb_erase(&es->rb_node, &tree->root);
585 ext4_es_free_extent(inode, es);
586 es = es1;
587 }
588
589 return es;
590 }
591
592 static struct extent_status *
ext4_es_try_to_merge_right(struct inode * inode,struct extent_status * es)593 ext4_es_try_to_merge_right(struct inode *inode, struct extent_status *es)
594 {
595 struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
596 struct extent_status *es1;
597 struct rb_node *node;
598
599 node = rb_next(&es->rb_node);
600 if (!node)
601 return es;
602
603 es1 = rb_entry(node, struct extent_status, rb_node);
604 if (ext4_es_can_be_merged(es, es1)) {
605 es->es_len += es1->es_len;
606 if (ext4_es_is_referenced(es1))
607 ext4_es_set_referenced(es);
608 rb_erase(node, &tree->root);
609 ext4_es_free_extent(inode, es1);
610 }
611
612 return es;
613 }
614
615 #ifdef ES_AGGRESSIVE_TEST
616 #include "ext4_extents.h" /* Needed when ES_AGGRESSIVE_TEST is defined */
617
ext4_es_insert_extent_ext_check(struct inode * inode,struct extent_status * es)618 static void ext4_es_insert_extent_ext_check(struct inode *inode,
619 struct extent_status *es)
620 {
621 struct ext4_ext_path *path = NULL;
622 struct ext4_extent *ex;
623 ext4_lblk_t ee_block;
624 ext4_fsblk_t ee_start;
625 unsigned short ee_len;
626 int depth, ee_status, es_status;
627
628 path = ext4_find_extent(inode, es->es_lblk, NULL, EXT4_EX_NOCACHE);
629 if (IS_ERR(path))
630 return;
631
632 depth = ext_depth(inode);
633 ex = path[depth].p_ext;
634
635 if (ex) {
636
637 ee_block = le32_to_cpu(ex->ee_block);
638 ee_start = ext4_ext_pblock(ex);
639 ee_len = ext4_ext_get_actual_len(ex);
640
641 ee_status = ext4_ext_is_unwritten(ex) ? 1 : 0;
642 es_status = ext4_es_is_unwritten(es) ? 1 : 0;
643
644 /*
645 * Make sure ex and es are not overlap when we try to insert
646 * a delayed/hole extent.
647 */
648 if (!ext4_es_is_written(es) && !ext4_es_is_unwritten(es)) {
649 if (in_range(es->es_lblk, ee_block, ee_len)) {
650 pr_warn("ES insert assertion failed for "
651 "inode: %lu we can find an extent "
652 "at block [%d/%d/%llu/%c], but we "
653 "want to add a delayed/hole extent "
654 "[%d/%d/%llu/%x]\n",
655 inode->i_ino, ee_block, ee_len,
656 ee_start, ee_status ? 'u' : 'w',
657 es->es_lblk, es->es_len,
658 ext4_es_pblock(es), ext4_es_status(es));
659 }
660 goto out;
661 }
662
663 /*
664 * We don't check ee_block == es->es_lblk, etc. because es
665 * might be a part of whole extent, vice versa.
666 */
667 if (es->es_lblk < ee_block ||
668 ext4_es_pblock(es) != ee_start + es->es_lblk - ee_block) {
669 pr_warn("ES insert assertion failed for inode: %lu "
670 "ex_status [%d/%d/%llu/%c] != "
671 "es_status [%d/%d/%llu/%c]\n", inode->i_ino,
672 ee_block, ee_len, ee_start,
673 ee_status ? 'u' : 'w', es->es_lblk, es->es_len,
674 ext4_es_pblock(es), es_status ? 'u' : 'w');
675 goto out;
676 }
677
678 if (ee_status ^ es_status) {
679 pr_warn("ES insert assertion failed for inode: %lu "
680 "ex_status [%d/%d/%llu/%c] != "
681 "es_status [%d/%d/%llu/%c]\n", inode->i_ino,
682 ee_block, ee_len, ee_start,
683 ee_status ? 'u' : 'w', es->es_lblk, es->es_len,
684 ext4_es_pblock(es), es_status ? 'u' : 'w');
685 }
686 } else {
687 /*
688 * We can't find an extent on disk. So we need to make sure
689 * that we don't want to add an written/unwritten extent.
690 */
691 if (!ext4_es_is_delayed(es) && !ext4_es_is_hole(es)) {
692 pr_warn("ES insert assertion failed for inode: %lu "
693 "can't find an extent at block %d but we want "
694 "to add a written/unwritten extent "
695 "[%d/%d/%llu/%x]\n", inode->i_ino,
696 es->es_lblk, es->es_lblk, es->es_len,
697 ext4_es_pblock(es), ext4_es_status(es));
698 }
699 }
700 out:
701 ext4_free_ext_path(path);
702 }
703
ext4_es_insert_extent_ind_check(struct inode * inode,struct extent_status * es)704 static void ext4_es_insert_extent_ind_check(struct inode *inode,
705 struct extent_status *es)
706 {
707 struct ext4_map_blocks map;
708 int retval;
709
710 /*
711 * Here we call ext4_ind_map_blocks to lookup a block mapping because
712 * 'Indirect' structure is defined in indirect.c. So we couldn't
713 * access direct/indirect tree from outside. It is too dirty to define
714 * this function in indirect.c file.
715 */
716
717 map.m_lblk = es->es_lblk;
718 map.m_len = es->es_len;
719
720 retval = ext4_ind_map_blocks(NULL, inode, &map, 0);
721 if (retval > 0) {
722 if (ext4_es_is_delayed(es) || ext4_es_is_hole(es)) {
723 /*
724 * We want to add a delayed/hole extent but this
725 * block has been allocated.
726 */
727 pr_warn("ES insert assertion failed for inode: %lu "
728 "We can find blocks but we want to add a "
729 "delayed/hole extent [%d/%d/%llu/%x]\n",
730 inode->i_ino, es->es_lblk, es->es_len,
731 ext4_es_pblock(es), ext4_es_status(es));
732 return;
733 } else if (ext4_es_is_written(es)) {
734 if (retval != es->es_len) {
735 pr_warn("ES insert assertion failed for "
736 "inode: %lu retval %d != es_len %d\n",
737 inode->i_ino, retval, es->es_len);
738 return;
739 }
740 if (map.m_pblk != ext4_es_pblock(es)) {
741 pr_warn("ES insert assertion failed for "
742 "inode: %lu m_pblk %llu != "
743 "es_pblk %llu\n",
744 inode->i_ino, map.m_pblk,
745 ext4_es_pblock(es));
746 return;
747 }
748 } else {
749 /*
750 * We don't need to check unwritten extent because
751 * indirect-based file doesn't have it.
752 */
753 BUG();
754 }
755 } else if (retval == 0) {
756 if (ext4_es_is_written(es)) {
757 pr_warn("ES insert assertion failed for inode: %lu "
758 "We can't find the block but we want to add "
759 "a written extent [%d/%d/%llu/%x]\n",
760 inode->i_ino, es->es_lblk, es->es_len,
761 ext4_es_pblock(es), ext4_es_status(es));
762 return;
763 }
764 }
765 }
766
ext4_es_insert_extent_check(struct inode * inode,struct extent_status * es)767 static inline void ext4_es_insert_extent_check(struct inode *inode,
768 struct extent_status *es)
769 {
770 /*
771 * We don't need to worry about the race condition because
772 * caller takes i_data_sem locking.
773 */
774 BUG_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
775 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
776 ext4_es_insert_extent_ext_check(inode, es);
777 else
778 ext4_es_insert_extent_ind_check(inode, es);
779 }
780 #else
ext4_es_insert_extent_check(struct inode * inode,struct extent_status * es)781 static inline void ext4_es_insert_extent_check(struct inode *inode,
782 struct extent_status *es)
783 {
784 }
785 #endif
786
__es_insert_extent(struct inode * inode,struct extent_status * newes,struct extent_status * prealloc)787 static int __es_insert_extent(struct inode *inode, struct extent_status *newes,
788 struct extent_status *prealloc)
789 {
790 struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
791 struct rb_node **p = &tree->root.rb_node;
792 struct rb_node *parent = NULL;
793 struct extent_status *es;
794
795 while (*p) {
796 parent = *p;
797 es = rb_entry(parent, struct extent_status, rb_node);
798
799 if (newes->es_lblk < es->es_lblk) {
800 if (ext4_es_can_be_merged(newes, es)) {
801 /*
802 * Here we can modify es_lblk directly
803 * because it isn't overlapped.
804 */
805 es->es_lblk = newes->es_lblk;
806 es->es_len += newes->es_len;
807 if (ext4_es_is_written(es) ||
808 ext4_es_is_unwritten(es))
809 ext4_es_store_pblock(es,
810 newes->es_pblk);
811 es = ext4_es_try_to_merge_left(inode, es);
812 goto out;
813 }
814 p = &(*p)->rb_left;
815 } else if (newes->es_lblk > ext4_es_end(es)) {
816 if (ext4_es_can_be_merged(es, newes)) {
817 es->es_len += newes->es_len;
818 es = ext4_es_try_to_merge_right(inode, es);
819 goto out;
820 }
821 p = &(*p)->rb_right;
822 } else {
823 BUG();
824 return -EINVAL;
825 }
826 }
827
828 if (prealloc)
829 es = prealloc;
830 else
831 es = __es_alloc_extent(false);
832 if (!es)
833 return -ENOMEM;
834 ext4_es_init_extent(inode, es, newes->es_lblk, newes->es_len,
835 newes->es_pblk);
836
837 rb_link_node(&es->rb_node, parent, p);
838 rb_insert_color(&es->rb_node, &tree->root);
839
840 out:
841 tree->cache_es = es;
842 return 0;
843 }
844
845 /*
846 * ext4_es_insert_extent() adds information to an inode's extent
847 * status tree.
848 */
ext4_es_insert_extent(struct inode * inode,ext4_lblk_t lblk,ext4_lblk_t len,ext4_fsblk_t pblk,unsigned int status,bool delalloc_reserve_used)849 void ext4_es_insert_extent(struct inode *inode, ext4_lblk_t lblk,
850 ext4_lblk_t len, ext4_fsblk_t pblk,
851 unsigned int status, bool delalloc_reserve_used)
852 {
853 struct extent_status newes;
854 ext4_lblk_t end = lblk + len - 1;
855 int err1 = 0, err2 = 0, err3 = 0;
856 int resv_used = 0, pending = 0;
857 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
858 struct extent_status *es1 = NULL;
859 struct extent_status *es2 = NULL;
860 struct pending_reservation *pr = NULL;
861 bool revise_pending = false;
862
863 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
864 return;
865
866 es_debug("add [%u/%u) %llu %x %d to extent status tree of inode %lu\n",
867 lblk, len, pblk, status, delalloc_reserve_used, inode->i_ino);
868
869 if (!len)
870 return;
871
872 BUG_ON(end < lblk);
873 WARN_ON_ONCE(status & EXTENT_STATUS_DELAYED);
874
875 newes.es_lblk = lblk;
876 newes.es_len = len;
877 ext4_es_store_pblock_status(&newes, pblk, status);
878 trace_ext4_es_insert_extent(inode, &newes);
879
880 ext4_es_insert_extent_check(inode, &newes);
881
882 revise_pending = sbi->s_cluster_ratio > 1 &&
883 test_opt(inode->i_sb, DELALLOC) &&
884 (status & (EXTENT_STATUS_WRITTEN |
885 EXTENT_STATUS_UNWRITTEN));
886 retry:
887 if (err1 && !es1)
888 es1 = __es_alloc_extent(true);
889 if ((err1 || err2) && !es2)
890 es2 = __es_alloc_extent(true);
891 if ((err1 || err2 || err3 < 0) && revise_pending && !pr)
892 pr = __alloc_pending(true);
893 write_lock(&EXT4_I(inode)->i_es_lock);
894
895 err1 = __es_remove_extent(inode, lblk, end, &resv_used, es1);
896 if (err1 != 0)
897 goto error;
898 /* Free preallocated extent if it didn't get used. */
899 if (es1) {
900 if (!es1->es_len)
901 __es_free_extent(es1);
902 es1 = NULL;
903 }
904
905 err2 = __es_insert_extent(inode, &newes, es2);
906 if (err2 == -ENOMEM && !ext4_es_must_keep(&newes))
907 err2 = 0;
908 if (err2 != 0)
909 goto error;
910 /* Free preallocated extent if it didn't get used. */
911 if (es2) {
912 if (!es2->es_len)
913 __es_free_extent(es2);
914 es2 = NULL;
915 }
916
917 if (revise_pending) {
918 err3 = __revise_pending(inode, lblk, len, &pr);
919 if (err3 < 0)
920 goto error;
921 if (pr) {
922 __free_pending(pr);
923 pr = NULL;
924 }
925 pending = err3;
926 }
927 error:
928 write_unlock(&EXT4_I(inode)->i_es_lock);
929 /*
930 * Reduce the reserved cluster count to reflect successful deferred
931 * allocation of delayed allocated clusters or direct allocation of
932 * clusters discovered to be delayed allocated. Once allocated, a
933 * cluster is not included in the reserved count.
934 *
935 * When direct allocating (from fallocate, filemap, DIO, or clusters
936 * allocated when delalloc has been disabled by ext4_nonda_switch())
937 * an extent either 1) contains delayed blocks but start with
938 * non-delayed allocated blocks (e.g. hole) or 2) contains non-delayed
939 * allocated blocks which belong to delayed allocated clusters when
940 * bigalloc feature is enabled, quota has already been claimed by
941 * ext4_mb_new_blocks(), so release the quota reservations made for
942 * any previously delayed allocated clusters instead of claim them
943 * again.
944 */
945 resv_used += pending;
946 if (resv_used)
947 ext4_da_update_reserve_space(inode, resv_used,
948 delalloc_reserve_used);
949
950 if (err1 || err2 || err3 < 0)
951 goto retry;
952
953 ext4_es_print_tree(inode);
954 return;
955 }
956
957 /*
958 * ext4_es_cache_extent() inserts information into the extent status
959 * tree if and only if there isn't information about the range in
960 * question already.
961 */
ext4_es_cache_extent(struct inode * inode,ext4_lblk_t lblk,ext4_lblk_t len,ext4_fsblk_t pblk,unsigned int status)962 void ext4_es_cache_extent(struct inode *inode, ext4_lblk_t lblk,
963 ext4_lblk_t len, ext4_fsblk_t pblk,
964 unsigned int status)
965 {
966 struct extent_status *es;
967 struct extent_status newes;
968 ext4_lblk_t end = lblk + len - 1;
969
970 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
971 return;
972
973 newes.es_lblk = lblk;
974 newes.es_len = len;
975 ext4_es_store_pblock_status(&newes, pblk, status);
976 trace_ext4_es_cache_extent(inode, &newes);
977
978 if (!len)
979 return;
980
981 BUG_ON(end < lblk);
982
983 write_lock(&EXT4_I(inode)->i_es_lock);
984
985 es = __es_tree_search(&EXT4_I(inode)->i_es_tree.root, lblk);
986 if (!es || es->es_lblk > end)
987 __es_insert_extent(inode, &newes, NULL);
988 write_unlock(&EXT4_I(inode)->i_es_lock);
989 }
990
991 /*
992 * ext4_es_lookup_extent() looks up an extent in extent status tree.
993 *
994 * ext4_es_lookup_extent is called by ext4_map_blocks/ext4_da_map_blocks.
995 *
996 * Return: 1 on found, 0 on not
997 */
ext4_es_lookup_extent(struct inode * inode,ext4_lblk_t lblk,ext4_lblk_t * next_lblk,struct extent_status * es)998 int ext4_es_lookup_extent(struct inode *inode, ext4_lblk_t lblk,
999 ext4_lblk_t *next_lblk,
1000 struct extent_status *es)
1001 {
1002 struct ext4_es_tree *tree;
1003 struct ext4_es_stats *stats;
1004 struct extent_status *es1 = NULL;
1005 struct rb_node *node;
1006 int found = 0;
1007
1008 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
1009 return 0;
1010
1011 trace_ext4_es_lookup_extent_enter(inode, lblk);
1012 es_debug("lookup extent in block %u\n", lblk);
1013
1014 tree = &EXT4_I(inode)->i_es_tree;
1015 read_lock(&EXT4_I(inode)->i_es_lock);
1016
1017 /* find extent in cache firstly */
1018 es->es_lblk = es->es_len = es->es_pblk = 0;
1019 es1 = READ_ONCE(tree->cache_es);
1020 if (es1 && in_range(lblk, es1->es_lblk, es1->es_len)) {
1021 es_debug("%u cached by [%u/%u)\n",
1022 lblk, es1->es_lblk, es1->es_len);
1023 found = 1;
1024 goto out;
1025 }
1026
1027 node = tree->root.rb_node;
1028 while (node) {
1029 es1 = rb_entry(node, struct extent_status, rb_node);
1030 if (lblk < es1->es_lblk)
1031 node = node->rb_left;
1032 else if (lblk > ext4_es_end(es1))
1033 node = node->rb_right;
1034 else {
1035 found = 1;
1036 break;
1037 }
1038 }
1039
1040 out:
1041 stats = &EXT4_SB(inode->i_sb)->s_es_stats;
1042 if (found) {
1043 BUG_ON(!es1);
1044 es->es_lblk = es1->es_lblk;
1045 es->es_len = es1->es_len;
1046 es->es_pblk = es1->es_pblk;
1047 if (!ext4_es_is_referenced(es1))
1048 ext4_es_set_referenced(es1);
1049 percpu_counter_inc(&stats->es_stats_cache_hits);
1050 if (next_lblk) {
1051 node = rb_next(&es1->rb_node);
1052 if (node) {
1053 es1 = rb_entry(node, struct extent_status,
1054 rb_node);
1055 *next_lblk = es1->es_lblk;
1056 } else
1057 *next_lblk = 0;
1058 }
1059 } else {
1060 percpu_counter_inc(&stats->es_stats_cache_misses);
1061 }
1062
1063 read_unlock(&EXT4_I(inode)->i_es_lock);
1064
1065 trace_ext4_es_lookup_extent_exit(inode, es, found);
1066 return found;
1067 }
1068
1069 struct rsvd_count {
1070 int ndelayed;
1071 bool first_do_lblk_found;
1072 ext4_lblk_t first_do_lblk;
1073 ext4_lblk_t last_do_lblk;
1074 struct extent_status *left_es;
1075 bool partial;
1076 ext4_lblk_t lclu;
1077 };
1078
1079 /*
1080 * init_rsvd - initialize reserved count data before removing block range
1081 * in file from extent status tree
1082 *
1083 * @inode - file containing range
1084 * @lblk - first block in range
1085 * @es - pointer to first extent in range
1086 * @rc - pointer to reserved count data
1087 *
1088 * Assumes es is not NULL
1089 */
init_rsvd(struct inode * inode,ext4_lblk_t lblk,struct extent_status * es,struct rsvd_count * rc)1090 static void init_rsvd(struct inode *inode, ext4_lblk_t lblk,
1091 struct extent_status *es, struct rsvd_count *rc)
1092 {
1093 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1094 struct rb_node *node;
1095
1096 rc->ndelayed = 0;
1097
1098 /*
1099 * for bigalloc, note the first delayed block in the range has not
1100 * been found, record the extent containing the block to the left of
1101 * the region to be removed, if any, and note that there's no partial
1102 * cluster to track
1103 */
1104 if (sbi->s_cluster_ratio > 1) {
1105 rc->first_do_lblk_found = false;
1106 if (lblk > es->es_lblk) {
1107 rc->left_es = es;
1108 } else {
1109 node = rb_prev(&es->rb_node);
1110 rc->left_es = node ? rb_entry(node,
1111 struct extent_status,
1112 rb_node) : NULL;
1113 }
1114 rc->partial = false;
1115 }
1116 }
1117
1118 /*
1119 * count_rsvd - count the clusters containing delayed blocks in a range
1120 * within an extent and add to the running tally in rsvd_count
1121 *
1122 * @inode - file containing extent
1123 * @lblk - first block in range
1124 * @len - length of range in blocks
1125 * @es - pointer to extent containing clusters to be counted
1126 * @rc - pointer to reserved count data
1127 *
1128 * Tracks partial clusters found at the beginning and end of extents so
1129 * they aren't overcounted when they span adjacent extents
1130 */
count_rsvd(struct inode * inode,ext4_lblk_t lblk,long len,struct extent_status * es,struct rsvd_count * rc)1131 static void count_rsvd(struct inode *inode, ext4_lblk_t lblk, long len,
1132 struct extent_status *es, struct rsvd_count *rc)
1133 {
1134 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1135 ext4_lblk_t i, end, nclu;
1136
1137 if (!ext4_es_is_delayed(es))
1138 return;
1139
1140 WARN_ON(len <= 0);
1141
1142 if (sbi->s_cluster_ratio == 1) {
1143 rc->ndelayed += (int) len;
1144 return;
1145 }
1146
1147 /* bigalloc */
1148
1149 i = (lblk < es->es_lblk) ? es->es_lblk : lblk;
1150 end = lblk + (ext4_lblk_t) len - 1;
1151 end = (end > ext4_es_end(es)) ? ext4_es_end(es) : end;
1152
1153 /* record the first block of the first delayed extent seen */
1154 if (!rc->first_do_lblk_found) {
1155 rc->first_do_lblk = i;
1156 rc->first_do_lblk_found = true;
1157 }
1158
1159 /* update the last lblk in the region seen so far */
1160 rc->last_do_lblk = end;
1161
1162 /*
1163 * if we're tracking a partial cluster and the current extent
1164 * doesn't start with it, count it and stop tracking
1165 */
1166 if (rc->partial && (rc->lclu != EXT4_B2C(sbi, i))) {
1167 rc->ndelayed++;
1168 rc->partial = false;
1169 }
1170
1171 /*
1172 * if the first cluster doesn't start on a cluster boundary but
1173 * ends on one, count it
1174 */
1175 if (EXT4_LBLK_COFF(sbi, i) != 0) {
1176 if (end >= EXT4_LBLK_CFILL(sbi, i)) {
1177 rc->ndelayed++;
1178 rc->partial = false;
1179 i = EXT4_LBLK_CFILL(sbi, i) + 1;
1180 }
1181 }
1182
1183 /*
1184 * if the current cluster starts on a cluster boundary, count the
1185 * number of whole delayed clusters in the extent
1186 */
1187 if ((i + sbi->s_cluster_ratio - 1) <= end) {
1188 nclu = (end - i + 1) >> sbi->s_cluster_bits;
1189 rc->ndelayed += nclu;
1190 i += nclu << sbi->s_cluster_bits;
1191 }
1192
1193 /*
1194 * start tracking a partial cluster if there's a partial at the end
1195 * of the current extent and we're not already tracking one
1196 */
1197 if (!rc->partial && i <= end) {
1198 rc->partial = true;
1199 rc->lclu = EXT4_B2C(sbi, i);
1200 }
1201 }
1202
1203 /*
1204 * __pr_tree_search - search for a pending cluster reservation
1205 *
1206 * @root - root of pending reservation tree
1207 * @lclu - logical cluster to search for
1208 *
1209 * Returns the pending reservation for the cluster identified by @lclu
1210 * if found. If not, returns a reservation for the next cluster if any,
1211 * and if not, returns NULL.
1212 */
__pr_tree_search(struct rb_root * root,ext4_lblk_t lclu)1213 static struct pending_reservation *__pr_tree_search(struct rb_root *root,
1214 ext4_lblk_t lclu)
1215 {
1216 struct rb_node *node = root->rb_node;
1217 struct pending_reservation *pr = NULL;
1218
1219 while (node) {
1220 pr = rb_entry(node, struct pending_reservation, rb_node);
1221 if (lclu < pr->lclu)
1222 node = node->rb_left;
1223 else if (lclu > pr->lclu)
1224 node = node->rb_right;
1225 else
1226 return pr;
1227 }
1228 if (pr && lclu < pr->lclu)
1229 return pr;
1230 if (pr && lclu > pr->lclu) {
1231 node = rb_next(&pr->rb_node);
1232 return node ? rb_entry(node, struct pending_reservation,
1233 rb_node) : NULL;
1234 }
1235 return NULL;
1236 }
1237
1238 /*
1239 * get_rsvd - calculates and returns the number of cluster reservations to be
1240 * released when removing a block range from the extent status tree
1241 * and releases any pending reservations within the range
1242 *
1243 * @inode - file containing block range
1244 * @end - last block in range
1245 * @right_es - pointer to extent containing next block beyond end or NULL
1246 * @rc - pointer to reserved count data
1247 *
1248 * The number of reservations to be released is equal to the number of
1249 * clusters containing delayed blocks within the range, minus the number of
1250 * clusters still containing delayed blocks at the ends of the range, and
1251 * minus the number of pending reservations within the range.
1252 */
get_rsvd(struct inode * inode,ext4_lblk_t end,struct extent_status * right_es,struct rsvd_count * rc)1253 static unsigned int get_rsvd(struct inode *inode, ext4_lblk_t end,
1254 struct extent_status *right_es,
1255 struct rsvd_count *rc)
1256 {
1257 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1258 struct pending_reservation *pr;
1259 struct ext4_pending_tree *tree = &EXT4_I(inode)->i_pending_tree;
1260 struct rb_node *node;
1261 ext4_lblk_t first_lclu, last_lclu;
1262 bool left_delayed, right_delayed, count_pending;
1263 struct extent_status *es;
1264
1265 if (sbi->s_cluster_ratio > 1) {
1266 /* count any remaining partial cluster */
1267 if (rc->partial)
1268 rc->ndelayed++;
1269
1270 if (rc->ndelayed == 0)
1271 return 0;
1272
1273 first_lclu = EXT4_B2C(sbi, rc->first_do_lblk);
1274 last_lclu = EXT4_B2C(sbi, rc->last_do_lblk);
1275
1276 /*
1277 * decrease the delayed count by the number of clusters at the
1278 * ends of the range that still contain delayed blocks -
1279 * these clusters still need to be reserved
1280 */
1281 left_delayed = right_delayed = false;
1282
1283 es = rc->left_es;
1284 while (es && ext4_es_end(es) >=
1285 EXT4_LBLK_CMASK(sbi, rc->first_do_lblk)) {
1286 if (ext4_es_is_delayed(es)) {
1287 rc->ndelayed--;
1288 left_delayed = true;
1289 break;
1290 }
1291 node = rb_prev(&es->rb_node);
1292 if (!node)
1293 break;
1294 es = rb_entry(node, struct extent_status, rb_node);
1295 }
1296 if (right_es && (!left_delayed || first_lclu != last_lclu)) {
1297 if (end < ext4_es_end(right_es)) {
1298 es = right_es;
1299 } else {
1300 node = rb_next(&right_es->rb_node);
1301 es = node ? rb_entry(node, struct extent_status,
1302 rb_node) : NULL;
1303 }
1304 while (es && es->es_lblk <=
1305 EXT4_LBLK_CFILL(sbi, rc->last_do_lblk)) {
1306 if (ext4_es_is_delayed(es)) {
1307 rc->ndelayed--;
1308 right_delayed = true;
1309 break;
1310 }
1311 node = rb_next(&es->rb_node);
1312 if (!node)
1313 break;
1314 es = rb_entry(node, struct extent_status,
1315 rb_node);
1316 }
1317 }
1318
1319 /*
1320 * Determine the block range that should be searched for
1321 * pending reservations, if any. Clusters on the ends of the
1322 * original removed range containing delayed blocks are
1323 * excluded. They've already been accounted for and it's not
1324 * possible to determine if an associated pending reservation
1325 * should be released with the information available in the
1326 * extents status tree.
1327 */
1328 if (first_lclu == last_lclu) {
1329 if (left_delayed | right_delayed)
1330 count_pending = false;
1331 else
1332 count_pending = true;
1333 } else {
1334 if (left_delayed)
1335 first_lclu++;
1336 if (right_delayed)
1337 last_lclu--;
1338 if (first_lclu <= last_lclu)
1339 count_pending = true;
1340 else
1341 count_pending = false;
1342 }
1343
1344 /*
1345 * a pending reservation found between first_lclu and last_lclu
1346 * represents an allocated cluster that contained at least one
1347 * delayed block, so the delayed total must be reduced by one
1348 * for each pending reservation found and released
1349 */
1350 if (count_pending) {
1351 pr = __pr_tree_search(&tree->root, first_lclu);
1352 while (pr && pr->lclu <= last_lclu) {
1353 rc->ndelayed--;
1354 node = rb_next(&pr->rb_node);
1355 rb_erase(&pr->rb_node, &tree->root);
1356 __free_pending(pr);
1357 if (!node)
1358 break;
1359 pr = rb_entry(node, struct pending_reservation,
1360 rb_node);
1361 }
1362 }
1363 }
1364 return rc->ndelayed;
1365 }
1366
1367
1368 /*
1369 * __es_remove_extent - removes block range from extent status tree
1370 *
1371 * @inode - file containing range
1372 * @lblk - first block in range
1373 * @end - last block in range
1374 * @reserved - number of cluster reservations released
1375 * @prealloc - pre-allocated es to avoid memory allocation failures
1376 *
1377 * If @reserved is not NULL and delayed allocation is enabled, counts
1378 * block/cluster reservations freed by removing range and if bigalloc
1379 * enabled cancels pending reservations as needed. Returns 0 on success,
1380 * error code on failure.
1381 */
__es_remove_extent(struct inode * inode,ext4_lblk_t lblk,ext4_lblk_t end,int * reserved,struct extent_status * prealloc)1382 static int __es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
1383 ext4_lblk_t end, int *reserved,
1384 struct extent_status *prealloc)
1385 {
1386 struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
1387 struct rb_node *node;
1388 struct extent_status *es;
1389 struct extent_status orig_es;
1390 ext4_lblk_t len1, len2;
1391 ext4_fsblk_t block;
1392 int err = 0;
1393 bool count_reserved = true;
1394 struct rsvd_count rc;
1395
1396 if (reserved == NULL || !test_opt(inode->i_sb, DELALLOC))
1397 count_reserved = false;
1398
1399 es = __es_tree_search(&tree->root, lblk);
1400 if (!es)
1401 goto out;
1402 if (es->es_lblk > end)
1403 goto out;
1404
1405 /* Simply invalidate cache_es. */
1406 tree->cache_es = NULL;
1407 if (count_reserved)
1408 init_rsvd(inode, lblk, es, &rc);
1409
1410 orig_es.es_lblk = es->es_lblk;
1411 orig_es.es_len = es->es_len;
1412 orig_es.es_pblk = es->es_pblk;
1413
1414 len1 = lblk > es->es_lblk ? lblk - es->es_lblk : 0;
1415 len2 = ext4_es_end(es) > end ? ext4_es_end(es) - end : 0;
1416 if (len1 > 0)
1417 es->es_len = len1;
1418 if (len2 > 0) {
1419 if (len1 > 0) {
1420 struct extent_status newes;
1421
1422 newes.es_lblk = end + 1;
1423 newes.es_len = len2;
1424 block = 0x7FDEADBEEFULL;
1425 if (ext4_es_is_written(&orig_es) ||
1426 ext4_es_is_unwritten(&orig_es))
1427 block = ext4_es_pblock(&orig_es) +
1428 orig_es.es_len - len2;
1429 ext4_es_store_pblock_status(&newes, block,
1430 ext4_es_status(&orig_es));
1431 err = __es_insert_extent(inode, &newes, prealloc);
1432 if (err) {
1433 if (!ext4_es_must_keep(&newes))
1434 return 0;
1435
1436 es->es_lblk = orig_es.es_lblk;
1437 es->es_len = orig_es.es_len;
1438 goto out;
1439 }
1440 } else {
1441 es->es_lblk = end + 1;
1442 es->es_len = len2;
1443 if (ext4_es_is_written(es) ||
1444 ext4_es_is_unwritten(es)) {
1445 block = orig_es.es_pblk + orig_es.es_len - len2;
1446 ext4_es_store_pblock(es, block);
1447 }
1448 }
1449 if (count_reserved)
1450 count_rsvd(inode, orig_es.es_lblk + len1,
1451 orig_es.es_len - len1 - len2, &orig_es, &rc);
1452 goto out_get_reserved;
1453 }
1454
1455 if (len1 > 0) {
1456 if (count_reserved)
1457 count_rsvd(inode, lblk, orig_es.es_len - len1,
1458 &orig_es, &rc);
1459 node = rb_next(&es->rb_node);
1460 if (node)
1461 es = rb_entry(node, struct extent_status, rb_node);
1462 else
1463 es = NULL;
1464 }
1465
1466 while (es && ext4_es_end(es) <= end) {
1467 if (count_reserved)
1468 count_rsvd(inode, es->es_lblk, es->es_len, es, &rc);
1469 node = rb_next(&es->rb_node);
1470 rb_erase(&es->rb_node, &tree->root);
1471 ext4_es_free_extent(inode, es);
1472 if (!node) {
1473 es = NULL;
1474 break;
1475 }
1476 es = rb_entry(node, struct extent_status, rb_node);
1477 }
1478
1479 if (es && es->es_lblk < end + 1) {
1480 ext4_lblk_t orig_len = es->es_len;
1481
1482 len1 = ext4_es_end(es) - end;
1483 if (count_reserved)
1484 count_rsvd(inode, es->es_lblk, orig_len - len1,
1485 es, &rc);
1486 es->es_lblk = end + 1;
1487 es->es_len = len1;
1488 if (ext4_es_is_written(es) || ext4_es_is_unwritten(es)) {
1489 block = es->es_pblk + orig_len - len1;
1490 ext4_es_store_pblock(es, block);
1491 }
1492 }
1493
1494 out_get_reserved:
1495 if (count_reserved)
1496 *reserved = get_rsvd(inode, end, es, &rc);
1497 out:
1498 return err;
1499 }
1500
1501 /*
1502 * ext4_es_remove_extent - removes block range from extent status tree
1503 *
1504 * @inode - file containing range
1505 * @lblk - first block in range
1506 * @len - number of blocks to remove
1507 *
1508 * Reduces block/cluster reservation count and for bigalloc cancels pending
1509 * reservations as needed.
1510 */
ext4_es_remove_extent(struct inode * inode,ext4_lblk_t lblk,ext4_lblk_t len)1511 void ext4_es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
1512 ext4_lblk_t len)
1513 {
1514 ext4_lblk_t end;
1515 int err = 0;
1516 int reserved = 0;
1517 struct extent_status *es = NULL;
1518
1519 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
1520 return;
1521
1522 trace_ext4_es_remove_extent(inode, lblk, len);
1523 es_debug("remove [%u/%u) from extent status tree of inode %lu\n",
1524 lblk, len, inode->i_ino);
1525
1526 if (!len)
1527 return;
1528
1529 end = lblk + len - 1;
1530 BUG_ON(end < lblk);
1531
1532 retry:
1533 if (err && !es)
1534 es = __es_alloc_extent(true);
1535 /*
1536 * ext4_clear_inode() depends on us taking i_es_lock unconditionally
1537 * so that we are sure __es_shrink() is done with the inode before it
1538 * is reclaimed.
1539 */
1540 write_lock(&EXT4_I(inode)->i_es_lock);
1541 err = __es_remove_extent(inode, lblk, end, &reserved, es);
1542 /* Free preallocated extent if it didn't get used. */
1543 if (es) {
1544 if (!es->es_len)
1545 __es_free_extent(es);
1546 es = NULL;
1547 }
1548 write_unlock(&EXT4_I(inode)->i_es_lock);
1549 if (err)
1550 goto retry;
1551
1552 ext4_es_print_tree(inode);
1553 ext4_da_release_space(inode, reserved);
1554 return;
1555 }
1556
__es_shrink(struct ext4_sb_info * sbi,int nr_to_scan,struct ext4_inode_info * locked_ei)1557 static int __es_shrink(struct ext4_sb_info *sbi, int nr_to_scan,
1558 struct ext4_inode_info *locked_ei)
1559 {
1560 struct ext4_inode_info *ei;
1561 struct ext4_es_stats *es_stats;
1562 ktime_t start_time;
1563 u64 scan_time;
1564 int nr_to_walk;
1565 int nr_shrunk = 0;
1566 int retried = 0, nr_skipped = 0;
1567
1568 es_stats = &sbi->s_es_stats;
1569 start_time = ktime_get();
1570
1571 retry:
1572 spin_lock(&sbi->s_es_lock);
1573 nr_to_walk = sbi->s_es_nr_inode;
1574 while (nr_to_walk-- > 0) {
1575 if (list_empty(&sbi->s_es_list)) {
1576 spin_unlock(&sbi->s_es_lock);
1577 goto out;
1578 }
1579 ei = list_first_entry(&sbi->s_es_list, struct ext4_inode_info,
1580 i_es_list);
1581 /* Move the inode to the tail */
1582 list_move_tail(&ei->i_es_list, &sbi->s_es_list);
1583
1584 /*
1585 * Normally we try hard to avoid shrinking precached inodes,
1586 * but we will as a last resort.
1587 */
1588 if (!retried && ext4_test_inode_state(&ei->vfs_inode,
1589 EXT4_STATE_EXT_PRECACHED)) {
1590 nr_skipped++;
1591 continue;
1592 }
1593
1594 if (ei == locked_ei || !write_trylock(&ei->i_es_lock)) {
1595 nr_skipped++;
1596 continue;
1597 }
1598 /*
1599 * Now we hold i_es_lock which protects us from inode reclaim
1600 * freeing inode under us
1601 */
1602 spin_unlock(&sbi->s_es_lock);
1603
1604 nr_shrunk += es_reclaim_extents(ei, &nr_to_scan);
1605 write_unlock(&ei->i_es_lock);
1606
1607 if (nr_to_scan <= 0)
1608 goto out;
1609 spin_lock(&sbi->s_es_lock);
1610 }
1611 spin_unlock(&sbi->s_es_lock);
1612
1613 /*
1614 * If we skipped any inodes, and we weren't able to make any
1615 * forward progress, try again to scan precached inodes.
1616 */
1617 if ((nr_shrunk == 0) && nr_skipped && !retried) {
1618 retried++;
1619 goto retry;
1620 }
1621
1622 if (locked_ei && nr_shrunk == 0)
1623 nr_shrunk = es_reclaim_extents(locked_ei, &nr_to_scan);
1624
1625 out:
1626 scan_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1627 if (likely(es_stats->es_stats_scan_time))
1628 es_stats->es_stats_scan_time = (scan_time +
1629 es_stats->es_stats_scan_time*3) / 4;
1630 else
1631 es_stats->es_stats_scan_time = scan_time;
1632 if (scan_time > es_stats->es_stats_max_scan_time)
1633 es_stats->es_stats_max_scan_time = scan_time;
1634 if (likely(es_stats->es_stats_shrunk))
1635 es_stats->es_stats_shrunk = (nr_shrunk +
1636 es_stats->es_stats_shrunk*3) / 4;
1637 else
1638 es_stats->es_stats_shrunk = nr_shrunk;
1639
1640 trace_ext4_es_shrink(sbi->s_sb, nr_shrunk, scan_time,
1641 nr_skipped, retried);
1642 return nr_shrunk;
1643 }
1644
ext4_es_count(struct shrinker * shrink,struct shrink_control * sc)1645 static unsigned long ext4_es_count(struct shrinker *shrink,
1646 struct shrink_control *sc)
1647 {
1648 unsigned long nr;
1649 struct ext4_sb_info *sbi;
1650
1651 sbi = shrink->private_data;
1652 nr = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt);
1653 trace_ext4_es_shrink_count(sbi->s_sb, sc->nr_to_scan, nr);
1654 return nr;
1655 }
1656
ext4_es_scan(struct shrinker * shrink,struct shrink_control * sc)1657 static unsigned long ext4_es_scan(struct shrinker *shrink,
1658 struct shrink_control *sc)
1659 {
1660 struct ext4_sb_info *sbi = shrink->private_data;
1661 int nr_to_scan = sc->nr_to_scan;
1662 int ret, nr_shrunk;
1663
1664 ret = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt);
1665 trace_ext4_es_shrink_scan_enter(sbi->s_sb, nr_to_scan, ret);
1666
1667 nr_shrunk = __es_shrink(sbi, nr_to_scan, NULL);
1668
1669 ret = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt);
1670 trace_ext4_es_shrink_scan_exit(sbi->s_sb, nr_shrunk, ret);
1671 return nr_shrunk;
1672 }
1673
ext4_seq_es_shrinker_info_show(struct seq_file * seq,void * v)1674 int ext4_seq_es_shrinker_info_show(struct seq_file *seq, void *v)
1675 {
1676 struct ext4_sb_info *sbi = EXT4_SB((struct super_block *) seq->private);
1677 struct ext4_es_stats *es_stats = &sbi->s_es_stats;
1678 struct ext4_inode_info *ei, *max = NULL;
1679 unsigned int inode_cnt = 0;
1680
1681 if (v != SEQ_START_TOKEN)
1682 return 0;
1683
1684 /* here we just find an inode that has the max nr. of objects */
1685 spin_lock(&sbi->s_es_lock);
1686 list_for_each_entry(ei, &sbi->s_es_list, i_es_list) {
1687 inode_cnt++;
1688 if (max && max->i_es_all_nr < ei->i_es_all_nr)
1689 max = ei;
1690 else if (!max)
1691 max = ei;
1692 }
1693 spin_unlock(&sbi->s_es_lock);
1694
1695 seq_printf(seq, "stats:\n %lld objects\n %lld reclaimable objects\n",
1696 percpu_counter_sum_positive(&es_stats->es_stats_all_cnt),
1697 percpu_counter_sum_positive(&es_stats->es_stats_shk_cnt));
1698 seq_printf(seq, " %lld/%lld cache hits/misses\n",
1699 percpu_counter_sum_positive(&es_stats->es_stats_cache_hits),
1700 percpu_counter_sum_positive(&es_stats->es_stats_cache_misses));
1701 if (inode_cnt)
1702 seq_printf(seq, " %d inodes on list\n", inode_cnt);
1703
1704 seq_printf(seq, "average:\n %llu us scan time\n",
1705 div_u64(es_stats->es_stats_scan_time, 1000));
1706 seq_printf(seq, " %lu shrunk objects\n", es_stats->es_stats_shrunk);
1707 if (inode_cnt)
1708 seq_printf(seq,
1709 "maximum:\n %lu inode (%u objects, %u reclaimable)\n"
1710 " %llu us max scan time\n",
1711 max->vfs_inode.i_ino, max->i_es_all_nr, max->i_es_shk_nr,
1712 div_u64(es_stats->es_stats_max_scan_time, 1000));
1713
1714 return 0;
1715 }
1716
ext4_es_register_shrinker(struct ext4_sb_info * sbi)1717 int ext4_es_register_shrinker(struct ext4_sb_info *sbi)
1718 {
1719 int err;
1720
1721 /* Make sure we have enough bits for physical block number */
1722 BUILD_BUG_ON(ES_SHIFT < 48);
1723 INIT_LIST_HEAD(&sbi->s_es_list);
1724 sbi->s_es_nr_inode = 0;
1725 spin_lock_init(&sbi->s_es_lock);
1726 sbi->s_es_stats.es_stats_shrunk = 0;
1727 err = percpu_counter_init(&sbi->s_es_stats.es_stats_cache_hits, 0,
1728 GFP_KERNEL);
1729 if (err)
1730 return err;
1731 err = percpu_counter_init(&sbi->s_es_stats.es_stats_cache_misses, 0,
1732 GFP_KERNEL);
1733 if (err)
1734 goto err1;
1735 sbi->s_es_stats.es_stats_scan_time = 0;
1736 sbi->s_es_stats.es_stats_max_scan_time = 0;
1737 err = percpu_counter_init(&sbi->s_es_stats.es_stats_all_cnt, 0, GFP_KERNEL);
1738 if (err)
1739 goto err2;
1740 err = percpu_counter_init(&sbi->s_es_stats.es_stats_shk_cnt, 0, GFP_KERNEL);
1741 if (err)
1742 goto err3;
1743
1744 sbi->s_es_shrinker = shrinker_alloc(0, "ext4-es:%s", sbi->s_sb->s_id);
1745 if (!sbi->s_es_shrinker) {
1746 err = -ENOMEM;
1747 goto err4;
1748 }
1749
1750 sbi->s_es_shrinker->scan_objects = ext4_es_scan;
1751 sbi->s_es_shrinker->count_objects = ext4_es_count;
1752 sbi->s_es_shrinker->private_data = sbi;
1753
1754 shrinker_register(sbi->s_es_shrinker);
1755
1756 return 0;
1757 err4:
1758 percpu_counter_destroy(&sbi->s_es_stats.es_stats_shk_cnt);
1759 err3:
1760 percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt);
1761 err2:
1762 percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_misses);
1763 err1:
1764 percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_hits);
1765 return err;
1766 }
1767
ext4_es_unregister_shrinker(struct ext4_sb_info * sbi)1768 void ext4_es_unregister_shrinker(struct ext4_sb_info *sbi)
1769 {
1770 percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_hits);
1771 percpu_counter_destroy(&sbi->s_es_stats.es_stats_cache_misses);
1772 percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt);
1773 percpu_counter_destroy(&sbi->s_es_stats.es_stats_shk_cnt);
1774 shrinker_free(sbi->s_es_shrinker);
1775 }
1776
1777 /*
1778 * Shrink extents in given inode from ei->i_es_shrink_lblk till end. Scan at
1779 * most *nr_to_scan extents, update *nr_to_scan accordingly.
1780 *
1781 * Return 0 if we hit end of tree / interval, 1 if we exhausted nr_to_scan.
1782 * Increment *nr_shrunk by the number of reclaimed extents. Also update
1783 * ei->i_es_shrink_lblk to where we should continue scanning.
1784 */
es_do_reclaim_extents(struct ext4_inode_info * ei,ext4_lblk_t end,int * nr_to_scan,int * nr_shrunk)1785 static int es_do_reclaim_extents(struct ext4_inode_info *ei, ext4_lblk_t end,
1786 int *nr_to_scan, int *nr_shrunk)
1787 {
1788 struct inode *inode = &ei->vfs_inode;
1789 struct ext4_es_tree *tree = &ei->i_es_tree;
1790 struct extent_status *es;
1791 struct rb_node *node;
1792
1793 es = __es_tree_search(&tree->root, ei->i_es_shrink_lblk);
1794 if (!es)
1795 goto out_wrap;
1796
1797 while (*nr_to_scan > 0) {
1798 if (es->es_lblk > end) {
1799 ei->i_es_shrink_lblk = end + 1;
1800 return 0;
1801 }
1802
1803 (*nr_to_scan)--;
1804 node = rb_next(&es->rb_node);
1805
1806 if (ext4_es_must_keep(es))
1807 goto next;
1808 if (ext4_es_is_referenced(es)) {
1809 ext4_es_clear_referenced(es);
1810 goto next;
1811 }
1812
1813 rb_erase(&es->rb_node, &tree->root);
1814 ext4_es_free_extent(inode, es);
1815 (*nr_shrunk)++;
1816 next:
1817 if (!node)
1818 goto out_wrap;
1819 es = rb_entry(node, struct extent_status, rb_node);
1820 }
1821 ei->i_es_shrink_lblk = es->es_lblk;
1822 return 1;
1823 out_wrap:
1824 ei->i_es_shrink_lblk = 0;
1825 return 0;
1826 }
1827
es_reclaim_extents(struct ext4_inode_info * ei,int * nr_to_scan)1828 static int es_reclaim_extents(struct ext4_inode_info *ei, int *nr_to_scan)
1829 {
1830 struct inode *inode = &ei->vfs_inode;
1831 int nr_shrunk = 0;
1832 ext4_lblk_t start = ei->i_es_shrink_lblk;
1833 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
1834 DEFAULT_RATELIMIT_BURST);
1835
1836 if (ei->i_es_shk_nr == 0)
1837 return 0;
1838
1839 if (ext4_test_inode_state(inode, EXT4_STATE_EXT_PRECACHED) &&
1840 __ratelimit(&_rs))
1841 ext4_warning(inode->i_sb, "forced shrink of precached extents");
1842
1843 if (!es_do_reclaim_extents(ei, EXT_MAX_BLOCKS, nr_to_scan, &nr_shrunk) &&
1844 start != 0)
1845 es_do_reclaim_extents(ei, start - 1, nr_to_scan, &nr_shrunk);
1846
1847 ei->i_es_tree.cache_es = NULL;
1848 return nr_shrunk;
1849 }
1850
1851 /*
1852 * Called to support EXT4_IOC_CLEAR_ES_CACHE. We can only remove
1853 * discretionary entries from the extent status cache. (Some entries
1854 * must be present for proper operations.)
1855 */
ext4_clear_inode_es(struct inode * inode)1856 void ext4_clear_inode_es(struct inode *inode)
1857 {
1858 struct ext4_inode_info *ei = EXT4_I(inode);
1859 struct extent_status *es;
1860 struct ext4_es_tree *tree;
1861 struct rb_node *node;
1862
1863 write_lock(&ei->i_es_lock);
1864 tree = &EXT4_I(inode)->i_es_tree;
1865 tree->cache_es = NULL;
1866 node = rb_first(&tree->root);
1867 while (node) {
1868 es = rb_entry(node, struct extent_status, rb_node);
1869 node = rb_next(node);
1870 if (!ext4_es_must_keep(es)) {
1871 rb_erase(&es->rb_node, &tree->root);
1872 ext4_es_free_extent(inode, es);
1873 }
1874 }
1875 ext4_clear_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
1876 write_unlock(&ei->i_es_lock);
1877 }
1878
1879 #ifdef ES_DEBUG__
ext4_print_pending_tree(struct inode * inode)1880 static void ext4_print_pending_tree(struct inode *inode)
1881 {
1882 struct ext4_pending_tree *tree;
1883 struct rb_node *node;
1884 struct pending_reservation *pr;
1885
1886 printk(KERN_DEBUG "pending reservations for inode %lu:", inode->i_ino);
1887 tree = &EXT4_I(inode)->i_pending_tree;
1888 node = rb_first(&tree->root);
1889 while (node) {
1890 pr = rb_entry(node, struct pending_reservation, rb_node);
1891 printk(KERN_DEBUG " %u", pr->lclu);
1892 node = rb_next(node);
1893 }
1894 printk(KERN_DEBUG "\n");
1895 }
1896 #else
1897 #define ext4_print_pending_tree(inode)
1898 #endif
1899
ext4_init_pending(void)1900 int __init ext4_init_pending(void)
1901 {
1902 ext4_pending_cachep = KMEM_CACHE(pending_reservation, SLAB_RECLAIM_ACCOUNT);
1903 if (ext4_pending_cachep == NULL)
1904 return -ENOMEM;
1905 return 0;
1906 }
1907
ext4_exit_pending(void)1908 void ext4_exit_pending(void)
1909 {
1910 kmem_cache_destroy(ext4_pending_cachep);
1911 }
1912
ext4_init_pending_tree(struct ext4_pending_tree * tree)1913 void ext4_init_pending_tree(struct ext4_pending_tree *tree)
1914 {
1915 tree->root = RB_ROOT;
1916 }
1917
1918 /*
1919 * __get_pending - retrieve a pointer to a pending reservation
1920 *
1921 * @inode - file containing the pending cluster reservation
1922 * @lclu - logical cluster of interest
1923 *
1924 * Returns a pointer to a pending reservation if it's a member of
1925 * the set, and NULL if not. Must be called holding i_es_lock.
1926 */
__get_pending(struct inode * inode,ext4_lblk_t lclu)1927 static struct pending_reservation *__get_pending(struct inode *inode,
1928 ext4_lblk_t lclu)
1929 {
1930 struct ext4_pending_tree *tree;
1931 struct rb_node *node;
1932 struct pending_reservation *pr = NULL;
1933
1934 tree = &EXT4_I(inode)->i_pending_tree;
1935 node = (&tree->root)->rb_node;
1936
1937 while (node) {
1938 pr = rb_entry(node, struct pending_reservation, rb_node);
1939 if (lclu < pr->lclu)
1940 node = node->rb_left;
1941 else if (lclu > pr->lclu)
1942 node = node->rb_right;
1943 else if (lclu == pr->lclu)
1944 return pr;
1945 }
1946 return NULL;
1947 }
1948
1949 /*
1950 * __insert_pending - adds a pending cluster reservation to the set of
1951 * pending reservations
1952 *
1953 * @inode - file containing the cluster
1954 * @lblk - logical block in the cluster to be added
1955 * @prealloc - preallocated pending entry
1956 *
1957 * Returns 1 on successful insertion and -ENOMEM on failure. If the
1958 * pending reservation is already in the set, returns successfully.
1959 */
__insert_pending(struct inode * inode,ext4_lblk_t lblk,struct pending_reservation ** prealloc)1960 static int __insert_pending(struct inode *inode, ext4_lblk_t lblk,
1961 struct pending_reservation **prealloc)
1962 {
1963 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1964 struct ext4_pending_tree *tree = &EXT4_I(inode)->i_pending_tree;
1965 struct rb_node **p = &tree->root.rb_node;
1966 struct rb_node *parent = NULL;
1967 struct pending_reservation *pr;
1968 ext4_lblk_t lclu;
1969 int ret = 0;
1970
1971 lclu = EXT4_B2C(sbi, lblk);
1972 /* search to find parent for insertion */
1973 while (*p) {
1974 parent = *p;
1975 pr = rb_entry(parent, struct pending_reservation, rb_node);
1976
1977 if (lclu < pr->lclu) {
1978 p = &(*p)->rb_left;
1979 } else if (lclu > pr->lclu) {
1980 p = &(*p)->rb_right;
1981 } else {
1982 /* pending reservation already inserted */
1983 goto out;
1984 }
1985 }
1986
1987 if (likely(*prealloc == NULL)) {
1988 pr = __alloc_pending(false);
1989 if (!pr) {
1990 ret = -ENOMEM;
1991 goto out;
1992 }
1993 } else {
1994 pr = *prealloc;
1995 *prealloc = NULL;
1996 }
1997 pr->lclu = lclu;
1998
1999 rb_link_node(&pr->rb_node, parent, p);
2000 rb_insert_color(&pr->rb_node, &tree->root);
2001 ret = 1;
2002
2003 out:
2004 return ret;
2005 }
2006
2007 /*
2008 * __remove_pending - removes a pending cluster reservation from the set
2009 * of pending reservations
2010 *
2011 * @inode - file containing the cluster
2012 * @lblk - logical block in the pending cluster reservation to be removed
2013 *
2014 * Returns successfully if pending reservation is not a member of the set.
2015 */
__remove_pending(struct inode * inode,ext4_lblk_t lblk)2016 static void __remove_pending(struct inode *inode, ext4_lblk_t lblk)
2017 {
2018 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2019 struct pending_reservation *pr;
2020 struct ext4_pending_tree *tree;
2021
2022 pr = __get_pending(inode, EXT4_B2C(sbi, lblk));
2023 if (pr != NULL) {
2024 tree = &EXT4_I(inode)->i_pending_tree;
2025 rb_erase(&pr->rb_node, &tree->root);
2026 __free_pending(pr);
2027 }
2028 }
2029
2030 /*
2031 * ext4_remove_pending - removes a pending cluster reservation from the set
2032 * of pending reservations
2033 *
2034 * @inode - file containing the cluster
2035 * @lblk - logical block in the pending cluster reservation to be removed
2036 *
2037 * Locking for external use of __remove_pending.
2038 */
ext4_remove_pending(struct inode * inode,ext4_lblk_t lblk)2039 void ext4_remove_pending(struct inode *inode, ext4_lblk_t lblk)
2040 {
2041 struct ext4_inode_info *ei = EXT4_I(inode);
2042
2043 write_lock(&ei->i_es_lock);
2044 __remove_pending(inode, lblk);
2045 write_unlock(&ei->i_es_lock);
2046 }
2047
2048 /*
2049 * ext4_is_pending - determine whether a cluster has a pending reservation
2050 * on it
2051 *
2052 * @inode - file containing the cluster
2053 * @lblk - logical block in the cluster
2054 *
2055 * Returns true if there's a pending reservation for the cluster in the
2056 * set of pending reservations, and false if not.
2057 */
ext4_is_pending(struct inode * inode,ext4_lblk_t lblk)2058 bool ext4_is_pending(struct inode *inode, ext4_lblk_t lblk)
2059 {
2060 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2061 struct ext4_inode_info *ei = EXT4_I(inode);
2062 bool ret;
2063
2064 read_lock(&ei->i_es_lock);
2065 ret = (bool)(__get_pending(inode, EXT4_B2C(sbi, lblk)) != NULL);
2066 read_unlock(&ei->i_es_lock);
2067
2068 return ret;
2069 }
2070
2071 /*
2072 * ext4_es_insert_delayed_extent - adds some delayed blocks to the extents
2073 * status tree, adding a pending reservation
2074 * where needed
2075 *
2076 * @inode - file containing the newly added block
2077 * @lblk - start logical block to be added
2078 * @len - length of blocks to be added
2079 * @lclu_allocated/end_allocated - indicates whether a physical cluster has
2080 * been allocated for the logical cluster
2081 * that contains the start/end block. Note that
2082 * end_allocated should always be set to false
2083 * if the start and the end block are in the
2084 * same cluster
2085 */
ext4_es_insert_delayed_extent(struct inode * inode,ext4_lblk_t lblk,ext4_lblk_t len,bool lclu_allocated,bool end_allocated)2086 void ext4_es_insert_delayed_extent(struct inode *inode, ext4_lblk_t lblk,
2087 ext4_lblk_t len, bool lclu_allocated,
2088 bool end_allocated)
2089 {
2090 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2091 struct extent_status newes;
2092 ext4_lblk_t end = lblk + len - 1;
2093 int err1 = 0, err2 = 0, err3 = 0;
2094 struct extent_status *es1 = NULL;
2095 struct extent_status *es2 = NULL;
2096 struct pending_reservation *pr1 = NULL;
2097 struct pending_reservation *pr2 = NULL;
2098
2099 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
2100 return;
2101
2102 es_debug("add [%u/%u) delayed to extent status tree of inode %lu\n",
2103 lblk, len, inode->i_ino);
2104 if (!len)
2105 return;
2106
2107 WARN_ON_ONCE((EXT4_B2C(sbi, lblk) == EXT4_B2C(sbi, end)) &&
2108 end_allocated);
2109
2110 newes.es_lblk = lblk;
2111 newes.es_len = len;
2112 ext4_es_store_pblock_status(&newes, ~0, EXTENT_STATUS_DELAYED);
2113 trace_ext4_es_insert_delayed_extent(inode, &newes, lclu_allocated,
2114 end_allocated);
2115
2116 ext4_es_insert_extent_check(inode, &newes);
2117
2118 retry:
2119 if (err1 && !es1)
2120 es1 = __es_alloc_extent(true);
2121 if ((err1 || err2) && !es2)
2122 es2 = __es_alloc_extent(true);
2123 if (err1 || err2 || err3 < 0) {
2124 if (lclu_allocated && !pr1)
2125 pr1 = __alloc_pending(true);
2126 if (end_allocated && !pr2)
2127 pr2 = __alloc_pending(true);
2128 }
2129 write_lock(&EXT4_I(inode)->i_es_lock);
2130
2131 err1 = __es_remove_extent(inode, lblk, end, NULL, es1);
2132 if (err1 != 0)
2133 goto error;
2134 /* Free preallocated extent if it didn't get used. */
2135 if (es1) {
2136 if (!es1->es_len)
2137 __es_free_extent(es1);
2138 es1 = NULL;
2139 }
2140
2141 err2 = __es_insert_extent(inode, &newes, es2);
2142 if (err2 != 0)
2143 goto error;
2144 /* Free preallocated extent if it didn't get used. */
2145 if (es2) {
2146 if (!es2->es_len)
2147 __es_free_extent(es2);
2148 es2 = NULL;
2149 }
2150
2151 if (lclu_allocated) {
2152 err3 = __insert_pending(inode, lblk, &pr1);
2153 if (err3 < 0)
2154 goto error;
2155 if (pr1) {
2156 __free_pending(pr1);
2157 pr1 = NULL;
2158 }
2159 }
2160 if (end_allocated) {
2161 err3 = __insert_pending(inode, end, &pr2);
2162 if (err3 < 0)
2163 goto error;
2164 if (pr2) {
2165 __free_pending(pr2);
2166 pr2 = NULL;
2167 }
2168 }
2169 error:
2170 write_unlock(&EXT4_I(inode)->i_es_lock);
2171 if (err1 || err2 || err3 < 0)
2172 goto retry;
2173
2174 ext4_es_print_tree(inode);
2175 ext4_print_pending_tree(inode);
2176 return;
2177 }
2178
2179 /*
2180 * __revise_pending - makes, cancels, or leaves unchanged pending cluster
2181 * reservations for a specified block range depending
2182 * upon the presence or absence of delayed blocks
2183 * outside the range within clusters at the ends of the
2184 * range
2185 *
2186 * @inode - file containing the range
2187 * @lblk - logical block defining the start of range
2188 * @len - length of range in blocks
2189 * @prealloc - preallocated pending entry
2190 *
2191 * Used after a newly allocated extent is added to the extents status tree.
2192 * Requires that the extents in the range have either written or unwritten
2193 * status. Must be called while holding i_es_lock. Returns number of new
2194 * inserts pending cluster on insert pendings, returns 0 on remove pendings,
2195 * return -ENOMEM on failure.
2196 */
__revise_pending(struct inode * inode,ext4_lblk_t lblk,ext4_lblk_t len,struct pending_reservation ** prealloc)2197 static int __revise_pending(struct inode *inode, ext4_lblk_t lblk,
2198 ext4_lblk_t len,
2199 struct pending_reservation **prealloc)
2200 {
2201 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2202 ext4_lblk_t end = lblk + len - 1;
2203 ext4_lblk_t first, last;
2204 bool f_del = false, l_del = false;
2205 int pendings = 0;
2206 int ret = 0;
2207
2208 if (len == 0)
2209 return 0;
2210
2211 /*
2212 * Two cases - block range within single cluster and block range
2213 * spanning two or more clusters. Note that a cluster belonging
2214 * to a range starting and/or ending on a cluster boundary is treated
2215 * as if it does not contain a delayed extent. The new range may
2216 * have allocated space for previously delayed blocks out to the
2217 * cluster boundary, requiring that any pre-existing pending
2218 * reservation be canceled. Because this code only looks at blocks
2219 * outside the range, it should revise pending reservations
2220 * correctly even if the extent represented by the range can't be
2221 * inserted in the extents status tree due to ENOSPC.
2222 */
2223
2224 if (EXT4_B2C(sbi, lblk) == EXT4_B2C(sbi, end)) {
2225 first = EXT4_LBLK_CMASK(sbi, lblk);
2226 if (first != lblk)
2227 f_del = __es_scan_range(inode, &ext4_es_is_delayed,
2228 first, lblk - 1);
2229 if (f_del) {
2230 ret = __insert_pending(inode, first, prealloc);
2231 if (ret < 0)
2232 goto out;
2233 pendings += ret;
2234 } else {
2235 last = EXT4_LBLK_CMASK(sbi, end) +
2236 sbi->s_cluster_ratio - 1;
2237 if (last != end)
2238 l_del = __es_scan_range(inode,
2239 &ext4_es_is_delayed,
2240 end + 1, last);
2241 if (l_del) {
2242 ret = __insert_pending(inode, last, prealloc);
2243 if (ret < 0)
2244 goto out;
2245 pendings += ret;
2246 } else
2247 __remove_pending(inode, last);
2248 }
2249 } else {
2250 first = EXT4_LBLK_CMASK(sbi, lblk);
2251 if (first != lblk)
2252 f_del = __es_scan_range(inode, &ext4_es_is_delayed,
2253 first, lblk - 1);
2254 if (f_del) {
2255 ret = __insert_pending(inode, first, prealloc);
2256 if (ret < 0)
2257 goto out;
2258 pendings += ret;
2259 } else
2260 __remove_pending(inode, first);
2261
2262 last = EXT4_LBLK_CMASK(sbi, end) + sbi->s_cluster_ratio - 1;
2263 if (last != end)
2264 l_del = __es_scan_range(inode, &ext4_es_is_delayed,
2265 end + 1, last);
2266 if (l_del) {
2267 ret = __insert_pending(inode, last, prealloc);
2268 if (ret < 0)
2269 goto out;
2270 pendings += ret;
2271 } else
2272 __remove_pending(inode, last);
2273 }
2274 out:
2275 return (ret < 0) ? ret : pendings;
2276 }
2277