xref: /openbmc/linux/fs/ext4/extents_status.c (revision 7587eb18)
1 /*
2  *  fs/ext4/extents_status.c
3  *
4  * Written by Yongqiang Yang <xiaoqiangnk@gmail.com>
5  * Modified by
6  *	Allison Henderson <achender@linux.vnet.ibm.com>
7  *	Hugh Dickins <hughd@google.com>
8  *	Zheng Liu <wenqing.lz@taobao.com>
9  *
10  * Ext4 extents status tree core functions.
11  */
12 #include <linux/list_sort.h>
13 #include <linux/proc_fs.h>
14 #include <linux/seq_file.h>
15 #include "ext4.h"
16 
17 #include <trace/events/ext4.h>
18 
19 /*
20  * According to previous discussion in Ext4 Developer Workshop, we
21  * will introduce a new structure called io tree to track all extent
22  * status in order to solve some problems that we have met
23  * (e.g. Reservation space warning), and provide extent-level locking.
24  * Delay extent tree is the first step to achieve this goal.  It is
25  * original built by Yongqiang Yang.  At that time it is called delay
26  * extent tree, whose goal is only track delayed extents in memory to
27  * simplify the implementation of fiemap and bigalloc, and introduce
28  * lseek SEEK_DATA/SEEK_HOLE support.  That is why it is still called
29  * delay extent tree at the first commit.  But for better understand
30  * what it does, it has been rename to extent status tree.
31  *
32  * Step1:
33  * Currently the first step has been done.  All delayed extents are
34  * tracked in the tree.  It maintains the delayed extent when a delayed
35  * allocation is issued, and the delayed extent is written out or
36  * invalidated.  Therefore the implementation of fiemap and bigalloc
37  * are simplified, and SEEK_DATA/SEEK_HOLE are introduced.
38  *
39  * The following comment describes the implemenmtation of extent
40  * status tree and future works.
41  *
42  * Step2:
43  * In this step all extent status are tracked by extent status tree.
44  * Thus, we can first try to lookup a block mapping in this tree before
45  * finding it in extent tree.  Hence, single extent cache can be removed
46  * because extent status tree can do a better job.  Extents in status
47  * tree are loaded on-demand.  Therefore, the extent status tree may not
48  * contain all of the extents in a file.  Meanwhile we define a shrinker
49  * to reclaim memory from extent status tree because fragmented extent
50  * tree will make status tree cost too much memory.  written/unwritten/-
51  * hole extents in the tree will be reclaimed by this shrinker when we
52  * are under high memory pressure.  Delayed extents will not be
53  * reclimed because fiemap, bigalloc, and seek_data/hole need it.
54  */
55 
56 /*
57  * Extent status tree implementation for ext4.
58  *
59  *
60  * ==========================================================================
61  * Extent status tree tracks all extent status.
62  *
63  * 1. Why we need to implement extent status tree?
64  *
65  * Without extent status tree, ext4 identifies a delayed extent by looking
66  * up page cache, this has several deficiencies - complicated, buggy,
67  * and inefficient code.
68  *
69  * FIEMAP, SEEK_HOLE/DATA, bigalloc, and writeout all need to know if a
70  * block or a range of blocks are belonged to a delayed extent.
71  *
72  * Let us have a look at how they do without extent status tree.
73  *   --	FIEMAP
74  *	FIEMAP looks up page cache to identify delayed allocations from holes.
75  *
76  *   --	SEEK_HOLE/DATA
77  *	SEEK_HOLE/DATA has the same problem as FIEMAP.
78  *
79  *   --	bigalloc
80  *	bigalloc looks up page cache to figure out if a block is
81  *	already under delayed allocation or not to determine whether
82  *	quota reserving is needed for the cluster.
83  *
84  *   --	writeout
85  *	Writeout looks up whole page cache to see if a buffer is
86  *	mapped, If there are not very many delayed buffers, then it is
87  *	time comsuming.
88  *
89  * With extent status tree implementation, FIEMAP, SEEK_HOLE/DATA,
90  * bigalloc and writeout can figure out if a block or a range of
91  * blocks is under delayed allocation(belonged to a delayed extent) or
92  * not by searching the extent tree.
93  *
94  *
95  * ==========================================================================
96  * 2. Ext4 extent status tree impelmentation
97  *
98  *   --	extent
99  *	A extent is a range of blocks which are contiguous logically and
100  *	physically.  Unlike extent in extent tree, this extent in ext4 is
101  *	a in-memory struct, there is no corresponding on-disk data.  There
102  *	is no limit on length of extent, so an extent can contain as many
103  *	blocks as they are contiguous logically and physically.
104  *
105  *   --	extent status tree
106  *	Every inode has an extent status tree and all allocation blocks
107  *	are added to the tree with different status.  The extent in the
108  *	tree are ordered by logical block no.
109  *
110  *   --	operations on a extent status tree
111  *	There are three important operations on a delayed extent tree: find
112  *	next extent, adding a extent(a range of blocks) and removing a extent.
113  *
114  *   --	race on a extent status tree
115  *	Extent status tree is protected by inode->i_es_lock.
116  *
117  *   --	memory consumption
118  *      Fragmented extent tree will make extent status tree cost too much
119  *      memory.  Hence, we will reclaim written/unwritten/hole extents from
120  *      the tree under a heavy memory pressure.
121  *
122  *
123  * ==========================================================================
124  * 3. Performance analysis
125  *
126  *   --	overhead
127  *	1. There is a cache extent for write access, so if writes are
128  *	not very random, adding space operaions are in O(1) time.
129  *
130  *   --	gain
131  *	2. Code is much simpler, more readable, more maintainable and
132  *	more efficient.
133  *
134  *
135  * ==========================================================================
136  * 4. TODO list
137  *
138  *   -- Refactor delayed space reservation
139  *
140  *   -- Extent-level locking
141  */
142 
143 static struct kmem_cache *ext4_es_cachep;
144 
145 static int __es_insert_extent(struct inode *inode, struct extent_status *newes);
146 static int __es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
147 			      ext4_lblk_t end);
148 static int es_reclaim_extents(struct ext4_inode_info *ei, int *nr_to_scan);
149 static int __es_shrink(struct ext4_sb_info *sbi, int nr_to_scan,
150 		       struct ext4_inode_info *locked_ei);
151 
152 int __init ext4_init_es(void)
153 {
154 	ext4_es_cachep = kmem_cache_create("ext4_extent_status",
155 					   sizeof(struct extent_status),
156 					   0, (SLAB_RECLAIM_ACCOUNT), NULL);
157 	if (ext4_es_cachep == NULL)
158 		return -ENOMEM;
159 	return 0;
160 }
161 
162 void ext4_exit_es(void)
163 {
164 	if (ext4_es_cachep)
165 		kmem_cache_destroy(ext4_es_cachep);
166 }
167 
168 void ext4_es_init_tree(struct ext4_es_tree *tree)
169 {
170 	tree->root = RB_ROOT;
171 	tree->cache_es = NULL;
172 }
173 
174 #ifdef ES_DEBUG__
175 static void ext4_es_print_tree(struct inode *inode)
176 {
177 	struct ext4_es_tree *tree;
178 	struct rb_node *node;
179 
180 	printk(KERN_DEBUG "status extents for inode %lu:", inode->i_ino);
181 	tree = &EXT4_I(inode)->i_es_tree;
182 	node = rb_first(&tree->root);
183 	while (node) {
184 		struct extent_status *es;
185 		es = rb_entry(node, struct extent_status, rb_node);
186 		printk(KERN_DEBUG " [%u/%u) %llu %x",
187 		       es->es_lblk, es->es_len,
188 		       ext4_es_pblock(es), ext4_es_status(es));
189 		node = rb_next(node);
190 	}
191 	printk(KERN_DEBUG "\n");
192 }
193 #else
194 #define ext4_es_print_tree(inode)
195 #endif
196 
197 static inline ext4_lblk_t ext4_es_end(struct extent_status *es)
198 {
199 	BUG_ON(es->es_lblk + es->es_len < es->es_lblk);
200 	return es->es_lblk + es->es_len - 1;
201 }
202 
203 /*
204  * search through the tree for an delayed extent with a given offset.  If
205  * it can't be found, try to find next extent.
206  */
207 static struct extent_status *__es_tree_search(struct rb_root *root,
208 					      ext4_lblk_t lblk)
209 {
210 	struct rb_node *node = root->rb_node;
211 	struct extent_status *es = NULL;
212 
213 	while (node) {
214 		es = rb_entry(node, struct extent_status, rb_node);
215 		if (lblk < es->es_lblk)
216 			node = node->rb_left;
217 		else if (lblk > ext4_es_end(es))
218 			node = node->rb_right;
219 		else
220 			return es;
221 	}
222 
223 	if (es && lblk < es->es_lblk)
224 		return es;
225 
226 	if (es && lblk > ext4_es_end(es)) {
227 		node = rb_next(&es->rb_node);
228 		return node ? rb_entry(node, struct extent_status, rb_node) :
229 			      NULL;
230 	}
231 
232 	return NULL;
233 }
234 
235 /*
236  * ext4_es_find_delayed_extent_range: find the 1st delayed extent covering
237  * @es->lblk if it exists, otherwise, the next extent after @es->lblk.
238  *
239  * @inode: the inode which owns delayed extents
240  * @lblk: the offset where we start to search
241  * @end: the offset where we stop to search
242  * @es: delayed extent that we found
243  */
244 void ext4_es_find_delayed_extent_range(struct inode *inode,
245 				 ext4_lblk_t lblk, ext4_lblk_t end,
246 				 struct extent_status *es)
247 {
248 	struct ext4_es_tree *tree = NULL;
249 	struct extent_status *es1 = NULL;
250 	struct rb_node *node;
251 
252 	BUG_ON(es == NULL);
253 	BUG_ON(end < lblk);
254 	trace_ext4_es_find_delayed_extent_range_enter(inode, lblk);
255 
256 	read_lock(&EXT4_I(inode)->i_es_lock);
257 	tree = &EXT4_I(inode)->i_es_tree;
258 
259 	/* find extent in cache firstly */
260 	es->es_lblk = es->es_len = es->es_pblk = 0;
261 	if (tree->cache_es) {
262 		es1 = tree->cache_es;
263 		if (in_range(lblk, es1->es_lblk, es1->es_len)) {
264 			es_debug("%u cached by [%u/%u) %llu %x\n",
265 				 lblk, es1->es_lblk, es1->es_len,
266 				 ext4_es_pblock(es1), ext4_es_status(es1));
267 			goto out;
268 		}
269 	}
270 
271 	es1 = __es_tree_search(&tree->root, lblk);
272 
273 out:
274 	if (es1 && !ext4_es_is_delayed(es1)) {
275 		while ((node = rb_next(&es1->rb_node)) != NULL) {
276 			es1 = rb_entry(node, struct extent_status, rb_node);
277 			if (es1->es_lblk > end) {
278 				es1 = NULL;
279 				break;
280 			}
281 			if (ext4_es_is_delayed(es1))
282 				break;
283 		}
284 	}
285 
286 	if (es1 && ext4_es_is_delayed(es1)) {
287 		tree->cache_es = es1;
288 		es->es_lblk = es1->es_lblk;
289 		es->es_len = es1->es_len;
290 		es->es_pblk = es1->es_pblk;
291 	}
292 
293 	read_unlock(&EXT4_I(inode)->i_es_lock);
294 
295 	trace_ext4_es_find_delayed_extent_range_exit(inode, es);
296 }
297 
298 static void ext4_es_list_add(struct inode *inode)
299 {
300 	struct ext4_inode_info *ei = EXT4_I(inode);
301 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
302 
303 	if (!list_empty(&ei->i_es_list))
304 		return;
305 
306 	spin_lock(&sbi->s_es_lock);
307 	if (list_empty(&ei->i_es_list)) {
308 		list_add_tail(&ei->i_es_list, &sbi->s_es_list);
309 		sbi->s_es_nr_inode++;
310 	}
311 	spin_unlock(&sbi->s_es_lock);
312 }
313 
314 static void ext4_es_list_del(struct inode *inode)
315 {
316 	struct ext4_inode_info *ei = EXT4_I(inode);
317 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
318 
319 	spin_lock(&sbi->s_es_lock);
320 	if (!list_empty(&ei->i_es_list)) {
321 		list_del_init(&ei->i_es_list);
322 		sbi->s_es_nr_inode--;
323 		WARN_ON_ONCE(sbi->s_es_nr_inode < 0);
324 	}
325 	spin_unlock(&sbi->s_es_lock);
326 }
327 
328 static struct extent_status *
329 ext4_es_alloc_extent(struct inode *inode, ext4_lblk_t lblk, ext4_lblk_t len,
330 		     ext4_fsblk_t pblk)
331 {
332 	struct extent_status *es;
333 	es = kmem_cache_alloc(ext4_es_cachep, GFP_ATOMIC);
334 	if (es == NULL)
335 		return NULL;
336 	es->es_lblk = lblk;
337 	es->es_len = len;
338 	es->es_pblk = pblk;
339 
340 	/*
341 	 * We don't count delayed extent because we never try to reclaim them
342 	 */
343 	if (!ext4_es_is_delayed(es)) {
344 		if (!EXT4_I(inode)->i_es_shk_nr++)
345 			ext4_es_list_add(inode);
346 		percpu_counter_inc(&EXT4_SB(inode->i_sb)->
347 					s_es_stats.es_stats_shk_cnt);
348 	}
349 
350 	EXT4_I(inode)->i_es_all_nr++;
351 	percpu_counter_inc(&EXT4_SB(inode->i_sb)->s_es_stats.es_stats_all_cnt);
352 
353 	return es;
354 }
355 
356 static void ext4_es_free_extent(struct inode *inode, struct extent_status *es)
357 {
358 	EXT4_I(inode)->i_es_all_nr--;
359 	percpu_counter_dec(&EXT4_SB(inode->i_sb)->s_es_stats.es_stats_all_cnt);
360 
361 	/* Decrease the shrink counter when this es is not delayed */
362 	if (!ext4_es_is_delayed(es)) {
363 		BUG_ON(EXT4_I(inode)->i_es_shk_nr == 0);
364 		if (!--EXT4_I(inode)->i_es_shk_nr)
365 			ext4_es_list_del(inode);
366 		percpu_counter_dec(&EXT4_SB(inode->i_sb)->
367 					s_es_stats.es_stats_shk_cnt);
368 	}
369 
370 	kmem_cache_free(ext4_es_cachep, es);
371 }
372 
373 /*
374  * Check whether or not two extents can be merged
375  * Condition:
376  *  - logical block number is contiguous
377  *  - physical block number is contiguous
378  *  - status is equal
379  */
380 static int ext4_es_can_be_merged(struct extent_status *es1,
381 				 struct extent_status *es2)
382 {
383 	if (ext4_es_type(es1) != ext4_es_type(es2))
384 		return 0;
385 
386 	if (((__u64) es1->es_len) + es2->es_len > EXT_MAX_BLOCKS) {
387 		pr_warn("ES assertion failed when merging extents. "
388 			"The sum of lengths of es1 (%d) and es2 (%d) "
389 			"is bigger than allowed file size (%d)\n",
390 			es1->es_len, es2->es_len, EXT_MAX_BLOCKS);
391 		WARN_ON(1);
392 		return 0;
393 	}
394 
395 	if (((__u64) es1->es_lblk) + es1->es_len != es2->es_lblk)
396 		return 0;
397 
398 	if ((ext4_es_is_written(es1) || ext4_es_is_unwritten(es1)) &&
399 	    (ext4_es_pblock(es1) + es1->es_len == ext4_es_pblock(es2)))
400 		return 1;
401 
402 	if (ext4_es_is_hole(es1))
403 		return 1;
404 
405 	/* we need to check delayed extent is without unwritten status */
406 	if (ext4_es_is_delayed(es1) && !ext4_es_is_unwritten(es1))
407 		return 1;
408 
409 	return 0;
410 }
411 
412 static struct extent_status *
413 ext4_es_try_to_merge_left(struct inode *inode, struct extent_status *es)
414 {
415 	struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
416 	struct extent_status *es1;
417 	struct rb_node *node;
418 
419 	node = rb_prev(&es->rb_node);
420 	if (!node)
421 		return es;
422 
423 	es1 = rb_entry(node, struct extent_status, rb_node);
424 	if (ext4_es_can_be_merged(es1, es)) {
425 		es1->es_len += es->es_len;
426 		if (ext4_es_is_referenced(es))
427 			ext4_es_set_referenced(es1);
428 		rb_erase(&es->rb_node, &tree->root);
429 		ext4_es_free_extent(inode, es);
430 		es = es1;
431 	}
432 
433 	return es;
434 }
435 
436 static struct extent_status *
437 ext4_es_try_to_merge_right(struct inode *inode, struct extent_status *es)
438 {
439 	struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
440 	struct extent_status *es1;
441 	struct rb_node *node;
442 
443 	node = rb_next(&es->rb_node);
444 	if (!node)
445 		return es;
446 
447 	es1 = rb_entry(node, struct extent_status, rb_node);
448 	if (ext4_es_can_be_merged(es, es1)) {
449 		es->es_len += es1->es_len;
450 		if (ext4_es_is_referenced(es1))
451 			ext4_es_set_referenced(es);
452 		rb_erase(node, &tree->root);
453 		ext4_es_free_extent(inode, es1);
454 	}
455 
456 	return es;
457 }
458 
459 #ifdef ES_AGGRESSIVE_TEST
460 #include "ext4_extents.h"	/* Needed when ES_AGGRESSIVE_TEST is defined */
461 
462 static void ext4_es_insert_extent_ext_check(struct inode *inode,
463 					    struct extent_status *es)
464 {
465 	struct ext4_ext_path *path = NULL;
466 	struct ext4_extent *ex;
467 	ext4_lblk_t ee_block;
468 	ext4_fsblk_t ee_start;
469 	unsigned short ee_len;
470 	int depth, ee_status, es_status;
471 
472 	path = ext4_find_extent(inode, es->es_lblk, NULL, EXT4_EX_NOCACHE);
473 	if (IS_ERR(path))
474 		return;
475 
476 	depth = ext_depth(inode);
477 	ex = path[depth].p_ext;
478 
479 	if (ex) {
480 
481 		ee_block = le32_to_cpu(ex->ee_block);
482 		ee_start = ext4_ext_pblock(ex);
483 		ee_len = ext4_ext_get_actual_len(ex);
484 
485 		ee_status = ext4_ext_is_unwritten(ex) ? 1 : 0;
486 		es_status = ext4_es_is_unwritten(es) ? 1 : 0;
487 
488 		/*
489 		 * Make sure ex and es are not overlap when we try to insert
490 		 * a delayed/hole extent.
491 		 */
492 		if (!ext4_es_is_written(es) && !ext4_es_is_unwritten(es)) {
493 			if (in_range(es->es_lblk, ee_block, ee_len)) {
494 				pr_warn("ES insert assertion failed for "
495 					"inode: %lu we can find an extent "
496 					"at block [%d/%d/%llu/%c], but we "
497 					"want to add a delayed/hole extent "
498 					"[%d/%d/%llu/%x]\n",
499 					inode->i_ino, ee_block, ee_len,
500 					ee_start, ee_status ? 'u' : 'w',
501 					es->es_lblk, es->es_len,
502 					ext4_es_pblock(es), ext4_es_status(es));
503 			}
504 			goto out;
505 		}
506 
507 		/*
508 		 * We don't check ee_block == es->es_lblk, etc. because es
509 		 * might be a part of whole extent, vice versa.
510 		 */
511 		if (es->es_lblk < ee_block ||
512 		    ext4_es_pblock(es) != ee_start + es->es_lblk - ee_block) {
513 			pr_warn("ES insert assertion failed for inode: %lu "
514 				"ex_status [%d/%d/%llu/%c] != "
515 				"es_status [%d/%d/%llu/%c]\n", inode->i_ino,
516 				ee_block, ee_len, ee_start,
517 				ee_status ? 'u' : 'w', es->es_lblk, es->es_len,
518 				ext4_es_pblock(es), es_status ? 'u' : 'w');
519 			goto out;
520 		}
521 
522 		if (ee_status ^ es_status) {
523 			pr_warn("ES insert assertion failed for inode: %lu "
524 				"ex_status [%d/%d/%llu/%c] != "
525 				"es_status [%d/%d/%llu/%c]\n", inode->i_ino,
526 				ee_block, ee_len, ee_start,
527 				ee_status ? 'u' : 'w', es->es_lblk, es->es_len,
528 				ext4_es_pblock(es), es_status ? 'u' : 'w');
529 		}
530 	} else {
531 		/*
532 		 * We can't find an extent on disk.  So we need to make sure
533 		 * that we don't want to add an written/unwritten extent.
534 		 */
535 		if (!ext4_es_is_delayed(es) && !ext4_es_is_hole(es)) {
536 			pr_warn("ES insert assertion failed for inode: %lu "
537 				"can't find an extent at block %d but we want "
538 				"to add a written/unwritten extent "
539 				"[%d/%d/%llu/%x]\n", inode->i_ino,
540 				es->es_lblk, es->es_lblk, es->es_len,
541 				ext4_es_pblock(es), ext4_es_status(es));
542 		}
543 	}
544 out:
545 	ext4_ext_drop_refs(path);
546 	kfree(path);
547 }
548 
549 static void ext4_es_insert_extent_ind_check(struct inode *inode,
550 					    struct extent_status *es)
551 {
552 	struct ext4_map_blocks map;
553 	int retval;
554 
555 	/*
556 	 * Here we call ext4_ind_map_blocks to lookup a block mapping because
557 	 * 'Indirect' structure is defined in indirect.c.  So we couldn't
558 	 * access direct/indirect tree from outside.  It is too dirty to define
559 	 * this function in indirect.c file.
560 	 */
561 
562 	map.m_lblk = es->es_lblk;
563 	map.m_len = es->es_len;
564 
565 	retval = ext4_ind_map_blocks(NULL, inode, &map, 0);
566 	if (retval > 0) {
567 		if (ext4_es_is_delayed(es) || ext4_es_is_hole(es)) {
568 			/*
569 			 * We want to add a delayed/hole extent but this
570 			 * block has been allocated.
571 			 */
572 			pr_warn("ES insert assertion failed for inode: %lu "
573 				"We can find blocks but we want to add a "
574 				"delayed/hole extent [%d/%d/%llu/%x]\n",
575 				inode->i_ino, es->es_lblk, es->es_len,
576 				ext4_es_pblock(es), ext4_es_status(es));
577 			return;
578 		} else if (ext4_es_is_written(es)) {
579 			if (retval != es->es_len) {
580 				pr_warn("ES insert assertion failed for "
581 					"inode: %lu retval %d != es_len %d\n",
582 					inode->i_ino, retval, es->es_len);
583 				return;
584 			}
585 			if (map.m_pblk != ext4_es_pblock(es)) {
586 				pr_warn("ES insert assertion failed for "
587 					"inode: %lu m_pblk %llu != "
588 					"es_pblk %llu\n",
589 					inode->i_ino, map.m_pblk,
590 					ext4_es_pblock(es));
591 				return;
592 			}
593 		} else {
594 			/*
595 			 * We don't need to check unwritten extent because
596 			 * indirect-based file doesn't have it.
597 			 */
598 			BUG_ON(1);
599 		}
600 	} else if (retval == 0) {
601 		if (ext4_es_is_written(es)) {
602 			pr_warn("ES insert assertion failed for inode: %lu "
603 				"We can't find the block but we want to add "
604 				"a written extent [%d/%d/%llu/%x]\n",
605 				inode->i_ino, es->es_lblk, es->es_len,
606 				ext4_es_pblock(es), ext4_es_status(es));
607 			return;
608 		}
609 	}
610 }
611 
612 static inline void ext4_es_insert_extent_check(struct inode *inode,
613 					       struct extent_status *es)
614 {
615 	/*
616 	 * We don't need to worry about the race condition because
617 	 * caller takes i_data_sem locking.
618 	 */
619 	BUG_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
620 	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
621 		ext4_es_insert_extent_ext_check(inode, es);
622 	else
623 		ext4_es_insert_extent_ind_check(inode, es);
624 }
625 #else
626 static inline void ext4_es_insert_extent_check(struct inode *inode,
627 					       struct extent_status *es)
628 {
629 }
630 #endif
631 
632 static int __es_insert_extent(struct inode *inode, struct extent_status *newes)
633 {
634 	struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
635 	struct rb_node **p = &tree->root.rb_node;
636 	struct rb_node *parent = NULL;
637 	struct extent_status *es;
638 
639 	while (*p) {
640 		parent = *p;
641 		es = rb_entry(parent, struct extent_status, rb_node);
642 
643 		if (newes->es_lblk < es->es_lblk) {
644 			if (ext4_es_can_be_merged(newes, es)) {
645 				/*
646 				 * Here we can modify es_lblk directly
647 				 * because it isn't overlapped.
648 				 */
649 				es->es_lblk = newes->es_lblk;
650 				es->es_len += newes->es_len;
651 				if (ext4_es_is_written(es) ||
652 				    ext4_es_is_unwritten(es))
653 					ext4_es_store_pblock(es,
654 							     newes->es_pblk);
655 				es = ext4_es_try_to_merge_left(inode, es);
656 				goto out;
657 			}
658 			p = &(*p)->rb_left;
659 		} else if (newes->es_lblk > ext4_es_end(es)) {
660 			if (ext4_es_can_be_merged(es, newes)) {
661 				es->es_len += newes->es_len;
662 				es = ext4_es_try_to_merge_right(inode, es);
663 				goto out;
664 			}
665 			p = &(*p)->rb_right;
666 		} else {
667 			BUG_ON(1);
668 			return -EINVAL;
669 		}
670 	}
671 
672 	es = ext4_es_alloc_extent(inode, newes->es_lblk, newes->es_len,
673 				  newes->es_pblk);
674 	if (!es)
675 		return -ENOMEM;
676 	rb_link_node(&es->rb_node, parent, p);
677 	rb_insert_color(&es->rb_node, &tree->root);
678 
679 out:
680 	tree->cache_es = es;
681 	return 0;
682 }
683 
684 /*
685  * ext4_es_insert_extent() adds information to an inode's extent
686  * status tree.
687  *
688  * Return 0 on success, error code on failure.
689  */
690 int ext4_es_insert_extent(struct inode *inode, ext4_lblk_t lblk,
691 			  ext4_lblk_t len, ext4_fsblk_t pblk,
692 			  unsigned int status)
693 {
694 	struct extent_status newes;
695 	ext4_lblk_t end = lblk + len - 1;
696 	int err = 0;
697 
698 	es_debug("add [%u/%u) %llu %x to extent status tree of inode %lu\n",
699 		 lblk, len, pblk, status, inode->i_ino);
700 
701 	if (!len)
702 		return 0;
703 
704 	BUG_ON(end < lblk);
705 
706 	if ((status & EXTENT_STATUS_DELAYED) &&
707 	    (status & EXTENT_STATUS_WRITTEN)) {
708 		ext4_warning(inode->i_sb, "Inserting extent [%u/%u] as "
709 				" delayed and written which can potentially "
710 				" cause data loss.", lblk, len);
711 		WARN_ON(1);
712 	}
713 
714 	newes.es_lblk = lblk;
715 	newes.es_len = len;
716 	ext4_es_store_pblock_status(&newes, pblk, status);
717 	trace_ext4_es_insert_extent(inode, &newes);
718 
719 	ext4_es_insert_extent_check(inode, &newes);
720 
721 	write_lock(&EXT4_I(inode)->i_es_lock);
722 	err = __es_remove_extent(inode, lblk, end);
723 	if (err != 0)
724 		goto error;
725 retry:
726 	err = __es_insert_extent(inode, &newes);
727 	if (err == -ENOMEM && __es_shrink(EXT4_SB(inode->i_sb),
728 					  128, EXT4_I(inode)))
729 		goto retry;
730 	if (err == -ENOMEM && !ext4_es_is_delayed(&newes))
731 		err = 0;
732 
733 error:
734 	write_unlock(&EXT4_I(inode)->i_es_lock);
735 
736 	ext4_es_print_tree(inode);
737 
738 	return err;
739 }
740 
741 /*
742  * ext4_es_cache_extent() inserts information into the extent status
743  * tree if and only if there isn't information about the range in
744  * question already.
745  */
746 void ext4_es_cache_extent(struct inode *inode, ext4_lblk_t lblk,
747 			  ext4_lblk_t len, ext4_fsblk_t pblk,
748 			  unsigned int status)
749 {
750 	struct extent_status *es;
751 	struct extent_status newes;
752 	ext4_lblk_t end = lblk + len - 1;
753 
754 	newes.es_lblk = lblk;
755 	newes.es_len = len;
756 	ext4_es_store_pblock_status(&newes, pblk, status);
757 	trace_ext4_es_cache_extent(inode, &newes);
758 
759 	if (!len)
760 		return;
761 
762 	BUG_ON(end < lblk);
763 
764 	write_lock(&EXT4_I(inode)->i_es_lock);
765 
766 	es = __es_tree_search(&EXT4_I(inode)->i_es_tree.root, lblk);
767 	if (!es || es->es_lblk > end)
768 		__es_insert_extent(inode, &newes);
769 	write_unlock(&EXT4_I(inode)->i_es_lock);
770 }
771 
772 /*
773  * ext4_es_lookup_extent() looks up an extent in extent status tree.
774  *
775  * ext4_es_lookup_extent is called by ext4_map_blocks/ext4_da_map_blocks.
776  *
777  * Return: 1 on found, 0 on not
778  */
779 int ext4_es_lookup_extent(struct inode *inode, ext4_lblk_t lblk,
780 			  struct extent_status *es)
781 {
782 	struct ext4_es_tree *tree;
783 	struct ext4_es_stats *stats;
784 	struct extent_status *es1 = NULL;
785 	struct rb_node *node;
786 	int found = 0;
787 
788 	trace_ext4_es_lookup_extent_enter(inode, lblk);
789 	es_debug("lookup extent in block %u\n", lblk);
790 
791 	tree = &EXT4_I(inode)->i_es_tree;
792 	read_lock(&EXT4_I(inode)->i_es_lock);
793 
794 	/* find extent in cache firstly */
795 	es->es_lblk = es->es_len = es->es_pblk = 0;
796 	if (tree->cache_es) {
797 		es1 = tree->cache_es;
798 		if (in_range(lblk, es1->es_lblk, es1->es_len)) {
799 			es_debug("%u cached by [%u/%u)\n",
800 				 lblk, es1->es_lblk, es1->es_len);
801 			found = 1;
802 			goto out;
803 		}
804 	}
805 
806 	node = tree->root.rb_node;
807 	while (node) {
808 		es1 = rb_entry(node, struct extent_status, rb_node);
809 		if (lblk < es1->es_lblk)
810 			node = node->rb_left;
811 		else if (lblk > ext4_es_end(es1))
812 			node = node->rb_right;
813 		else {
814 			found = 1;
815 			break;
816 		}
817 	}
818 
819 out:
820 	stats = &EXT4_SB(inode->i_sb)->s_es_stats;
821 	if (found) {
822 		BUG_ON(!es1);
823 		es->es_lblk = es1->es_lblk;
824 		es->es_len = es1->es_len;
825 		es->es_pblk = es1->es_pblk;
826 		if (!ext4_es_is_referenced(es1))
827 			ext4_es_set_referenced(es1);
828 		stats->es_stats_cache_hits++;
829 	} else {
830 		stats->es_stats_cache_misses++;
831 	}
832 
833 	read_unlock(&EXT4_I(inode)->i_es_lock);
834 
835 	trace_ext4_es_lookup_extent_exit(inode, es, found);
836 	return found;
837 }
838 
839 static int __es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
840 			      ext4_lblk_t end)
841 {
842 	struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
843 	struct rb_node *node;
844 	struct extent_status *es;
845 	struct extent_status orig_es;
846 	ext4_lblk_t len1, len2;
847 	ext4_fsblk_t block;
848 	int err;
849 
850 retry:
851 	err = 0;
852 	es = __es_tree_search(&tree->root, lblk);
853 	if (!es)
854 		goto out;
855 	if (es->es_lblk > end)
856 		goto out;
857 
858 	/* Simply invalidate cache_es. */
859 	tree->cache_es = NULL;
860 
861 	orig_es.es_lblk = es->es_lblk;
862 	orig_es.es_len = es->es_len;
863 	orig_es.es_pblk = es->es_pblk;
864 
865 	len1 = lblk > es->es_lblk ? lblk - es->es_lblk : 0;
866 	len2 = ext4_es_end(es) > end ? ext4_es_end(es) - end : 0;
867 	if (len1 > 0)
868 		es->es_len = len1;
869 	if (len2 > 0) {
870 		if (len1 > 0) {
871 			struct extent_status newes;
872 
873 			newes.es_lblk = end + 1;
874 			newes.es_len = len2;
875 			block = 0x7FDEADBEEFULL;
876 			if (ext4_es_is_written(&orig_es) ||
877 			    ext4_es_is_unwritten(&orig_es))
878 				block = ext4_es_pblock(&orig_es) +
879 					orig_es.es_len - len2;
880 			ext4_es_store_pblock_status(&newes, block,
881 						    ext4_es_status(&orig_es));
882 			err = __es_insert_extent(inode, &newes);
883 			if (err) {
884 				es->es_lblk = orig_es.es_lblk;
885 				es->es_len = orig_es.es_len;
886 				if ((err == -ENOMEM) &&
887 				    __es_shrink(EXT4_SB(inode->i_sb),
888 							128, EXT4_I(inode)))
889 					goto retry;
890 				goto out;
891 			}
892 		} else {
893 			es->es_lblk = end + 1;
894 			es->es_len = len2;
895 			if (ext4_es_is_written(es) ||
896 			    ext4_es_is_unwritten(es)) {
897 				block = orig_es.es_pblk + orig_es.es_len - len2;
898 				ext4_es_store_pblock(es, block);
899 			}
900 		}
901 		goto out;
902 	}
903 
904 	if (len1 > 0) {
905 		node = rb_next(&es->rb_node);
906 		if (node)
907 			es = rb_entry(node, struct extent_status, rb_node);
908 		else
909 			es = NULL;
910 	}
911 
912 	while (es && ext4_es_end(es) <= end) {
913 		node = rb_next(&es->rb_node);
914 		rb_erase(&es->rb_node, &tree->root);
915 		ext4_es_free_extent(inode, es);
916 		if (!node) {
917 			es = NULL;
918 			break;
919 		}
920 		es = rb_entry(node, struct extent_status, rb_node);
921 	}
922 
923 	if (es && es->es_lblk < end + 1) {
924 		ext4_lblk_t orig_len = es->es_len;
925 
926 		len1 = ext4_es_end(es) - end;
927 		es->es_lblk = end + 1;
928 		es->es_len = len1;
929 		if (ext4_es_is_written(es) || ext4_es_is_unwritten(es)) {
930 			block = es->es_pblk + orig_len - len1;
931 			ext4_es_store_pblock(es, block);
932 		}
933 	}
934 
935 out:
936 	return err;
937 }
938 
939 /*
940  * ext4_es_remove_extent() removes a space from a extent status tree.
941  *
942  * Return 0 on success, error code on failure.
943  */
944 int ext4_es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
945 			  ext4_lblk_t len)
946 {
947 	ext4_lblk_t end;
948 	int err = 0;
949 
950 	trace_ext4_es_remove_extent(inode, lblk, len);
951 	es_debug("remove [%u/%u) from extent status tree of inode %lu\n",
952 		 lblk, len, inode->i_ino);
953 
954 	if (!len)
955 		return err;
956 
957 	end = lblk + len - 1;
958 	BUG_ON(end < lblk);
959 
960 	/*
961 	 * ext4_clear_inode() depends on us taking i_es_lock unconditionally
962 	 * so that we are sure __es_shrink() is done with the inode before it
963 	 * is reclaimed.
964 	 */
965 	write_lock(&EXT4_I(inode)->i_es_lock);
966 	err = __es_remove_extent(inode, lblk, end);
967 	write_unlock(&EXT4_I(inode)->i_es_lock);
968 	ext4_es_print_tree(inode);
969 	return err;
970 }
971 
972 static int __es_shrink(struct ext4_sb_info *sbi, int nr_to_scan,
973 		       struct ext4_inode_info *locked_ei)
974 {
975 	struct ext4_inode_info *ei;
976 	struct ext4_es_stats *es_stats;
977 	ktime_t start_time;
978 	u64 scan_time;
979 	int nr_to_walk;
980 	int nr_shrunk = 0;
981 	int retried = 0, nr_skipped = 0;
982 
983 	es_stats = &sbi->s_es_stats;
984 	start_time = ktime_get();
985 
986 retry:
987 	spin_lock(&sbi->s_es_lock);
988 	nr_to_walk = sbi->s_es_nr_inode;
989 	while (nr_to_walk-- > 0) {
990 		if (list_empty(&sbi->s_es_list)) {
991 			spin_unlock(&sbi->s_es_lock);
992 			goto out;
993 		}
994 		ei = list_first_entry(&sbi->s_es_list, struct ext4_inode_info,
995 				      i_es_list);
996 		/* Move the inode to the tail */
997 		list_move_tail(&ei->i_es_list, &sbi->s_es_list);
998 
999 		/*
1000 		 * Normally we try hard to avoid shrinking precached inodes,
1001 		 * but we will as a last resort.
1002 		 */
1003 		if (!retried && ext4_test_inode_state(&ei->vfs_inode,
1004 						EXT4_STATE_EXT_PRECACHED)) {
1005 			nr_skipped++;
1006 			continue;
1007 		}
1008 
1009 		if (ei == locked_ei || !write_trylock(&ei->i_es_lock)) {
1010 			nr_skipped++;
1011 			continue;
1012 		}
1013 		/*
1014 		 * Now we hold i_es_lock which protects us from inode reclaim
1015 		 * freeing inode under us
1016 		 */
1017 		spin_unlock(&sbi->s_es_lock);
1018 
1019 		nr_shrunk += es_reclaim_extents(ei, &nr_to_scan);
1020 		write_unlock(&ei->i_es_lock);
1021 
1022 		if (nr_to_scan <= 0)
1023 			goto out;
1024 		spin_lock(&sbi->s_es_lock);
1025 	}
1026 	spin_unlock(&sbi->s_es_lock);
1027 
1028 	/*
1029 	 * If we skipped any inodes, and we weren't able to make any
1030 	 * forward progress, try again to scan precached inodes.
1031 	 */
1032 	if ((nr_shrunk == 0) && nr_skipped && !retried) {
1033 		retried++;
1034 		goto retry;
1035 	}
1036 
1037 	if (locked_ei && nr_shrunk == 0)
1038 		nr_shrunk = es_reclaim_extents(locked_ei, &nr_to_scan);
1039 
1040 out:
1041 	scan_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1042 	if (likely(es_stats->es_stats_scan_time))
1043 		es_stats->es_stats_scan_time = (scan_time +
1044 				es_stats->es_stats_scan_time*3) / 4;
1045 	else
1046 		es_stats->es_stats_scan_time = scan_time;
1047 	if (scan_time > es_stats->es_stats_max_scan_time)
1048 		es_stats->es_stats_max_scan_time = scan_time;
1049 	if (likely(es_stats->es_stats_shrunk))
1050 		es_stats->es_stats_shrunk = (nr_shrunk +
1051 				es_stats->es_stats_shrunk*3) / 4;
1052 	else
1053 		es_stats->es_stats_shrunk = nr_shrunk;
1054 
1055 	trace_ext4_es_shrink(sbi->s_sb, nr_shrunk, scan_time,
1056 			     nr_skipped, retried);
1057 	return nr_shrunk;
1058 }
1059 
1060 static unsigned long ext4_es_count(struct shrinker *shrink,
1061 				   struct shrink_control *sc)
1062 {
1063 	unsigned long nr;
1064 	struct ext4_sb_info *sbi;
1065 
1066 	sbi = container_of(shrink, struct ext4_sb_info, s_es_shrinker);
1067 	nr = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt);
1068 	trace_ext4_es_shrink_count(sbi->s_sb, sc->nr_to_scan, nr);
1069 	return nr;
1070 }
1071 
1072 static unsigned long ext4_es_scan(struct shrinker *shrink,
1073 				  struct shrink_control *sc)
1074 {
1075 	struct ext4_sb_info *sbi = container_of(shrink,
1076 					struct ext4_sb_info, s_es_shrinker);
1077 	int nr_to_scan = sc->nr_to_scan;
1078 	int ret, nr_shrunk;
1079 
1080 	ret = percpu_counter_read_positive(&sbi->s_es_stats.es_stats_shk_cnt);
1081 	trace_ext4_es_shrink_scan_enter(sbi->s_sb, nr_to_scan, ret);
1082 
1083 	if (!nr_to_scan)
1084 		return ret;
1085 
1086 	nr_shrunk = __es_shrink(sbi, nr_to_scan, NULL);
1087 
1088 	trace_ext4_es_shrink_scan_exit(sbi->s_sb, nr_shrunk, ret);
1089 	return nr_shrunk;
1090 }
1091 
1092 int ext4_seq_es_shrinker_info_show(struct seq_file *seq, void *v)
1093 {
1094 	struct ext4_sb_info *sbi = EXT4_SB((struct super_block *) seq->private);
1095 	struct ext4_es_stats *es_stats = &sbi->s_es_stats;
1096 	struct ext4_inode_info *ei, *max = NULL;
1097 	unsigned int inode_cnt = 0;
1098 
1099 	if (v != SEQ_START_TOKEN)
1100 		return 0;
1101 
1102 	/* here we just find an inode that has the max nr. of objects */
1103 	spin_lock(&sbi->s_es_lock);
1104 	list_for_each_entry(ei, &sbi->s_es_list, i_es_list) {
1105 		inode_cnt++;
1106 		if (max && max->i_es_all_nr < ei->i_es_all_nr)
1107 			max = ei;
1108 		else if (!max)
1109 			max = ei;
1110 	}
1111 	spin_unlock(&sbi->s_es_lock);
1112 
1113 	seq_printf(seq, "stats:\n  %lld objects\n  %lld reclaimable objects\n",
1114 		   percpu_counter_sum_positive(&es_stats->es_stats_all_cnt),
1115 		   percpu_counter_sum_positive(&es_stats->es_stats_shk_cnt));
1116 	seq_printf(seq, "  %lu/%lu cache hits/misses\n",
1117 		   es_stats->es_stats_cache_hits,
1118 		   es_stats->es_stats_cache_misses);
1119 	if (inode_cnt)
1120 		seq_printf(seq, "  %d inodes on list\n", inode_cnt);
1121 
1122 	seq_printf(seq, "average:\n  %llu us scan time\n",
1123 	    div_u64(es_stats->es_stats_scan_time, 1000));
1124 	seq_printf(seq, "  %lu shrunk objects\n", es_stats->es_stats_shrunk);
1125 	if (inode_cnt)
1126 		seq_printf(seq,
1127 		    "maximum:\n  %lu inode (%u objects, %u reclaimable)\n"
1128 		    "  %llu us max scan time\n",
1129 		    max->vfs_inode.i_ino, max->i_es_all_nr, max->i_es_shk_nr,
1130 		    div_u64(es_stats->es_stats_max_scan_time, 1000));
1131 
1132 	return 0;
1133 }
1134 
1135 int ext4_es_register_shrinker(struct ext4_sb_info *sbi)
1136 {
1137 	int err;
1138 
1139 	/* Make sure we have enough bits for physical block number */
1140 	BUILD_BUG_ON(ES_SHIFT < 48);
1141 	INIT_LIST_HEAD(&sbi->s_es_list);
1142 	sbi->s_es_nr_inode = 0;
1143 	spin_lock_init(&sbi->s_es_lock);
1144 	sbi->s_es_stats.es_stats_shrunk = 0;
1145 	sbi->s_es_stats.es_stats_cache_hits = 0;
1146 	sbi->s_es_stats.es_stats_cache_misses = 0;
1147 	sbi->s_es_stats.es_stats_scan_time = 0;
1148 	sbi->s_es_stats.es_stats_max_scan_time = 0;
1149 	err = percpu_counter_init(&sbi->s_es_stats.es_stats_all_cnt, 0, GFP_KERNEL);
1150 	if (err)
1151 		return err;
1152 	err = percpu_counter_init(&sbi->s_es_stats.es_stats_shk_cnt, 0, GFP_KERNEL);
1153 	if (err)
1154 		goto err1;
1155 
1156 	sbi->s_es_shrinker.scan_objects = ext4_es_scan;
1157 	sbi->s_es_shrinker.count_objects = ext4_es_count;
1158 	sbi->s_es_shrinker.seeks = DEFAULT_SEEKS;
1159 	err = register_shrinker(&sbi->s_es_shrinker);
1160 	if (err)
1161 		goto err2;
1162 
1163 	return 0;
1164 
1165 err2:
1166 	percpu_counter_destroy(&sbi->s_es_stats.es_stats_shk_cnt);
1167 err1:
1168 	percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt);
1169 	return err;
1170 }
1171 
1172 void ext4_es_unregister_shrinker(struct ext4_sb_info *sbi)
1173 {
1174 	percpu_counter_destroy(&sbi->s_es_stats.es_stats_all_cnt);
1175 	percpu_counter_destroy(&sbi->s_es_stats.es_stats_shk_cnt);
1176 	unregister_shrinker(&sbi->s_es_shrinker);
1177 }
1178 
1179 /*
1180  * Shrink extents in given inode from ei->i_es_shrink_lblk till end. Scan at
1181  * most *nr_to_scan extents, update *nr_to_scan accordingly.
1182  *
1183  * Return 0 if we hit end of tree / interval, 1 if we exhausted nr_to_scan.
1184  * Increment *nr_shrunk by the number of reclaimed extents. Also update
1185  * ei->i_es_shrink_lblk to where we should continue scanning.
1186  */
1187 static int es_do_reclaim_extents(struct ext4_inode_info *ei, ext4_lblk_t end,
1188 				 int *nr_to_scan, int *nr_shrunk)
1189 {
1190 	struct inode *inode = &ei->vfs_inode;
1191 	struct ext4_es_tree *tree = &ei->i_es_tree;
1192 	struct extent_status *es;
1193 	struct rb_node *node;
1194 
1195 	es = __es_tree_search(&tree->root, ei->i_es_shrink_lblk);
1196 	if (!es)
1197 		goto out_wrap;
1198 	node = &es->rb_node;
1199 	while (*nr_to_scan > 0) {
1200 		if (es->es_lblk > end) {
1201 			ei->i_es_shrink_lblk = end + 1;
1202 			return 0;
1203 		}
1204 
1205 		(*nr_to_scan)--;
1206 		node = rb_next(&es->rb_node);
1207 		/*
1208 		 * We can't reclaim delayed extent from status tree because
1209 		 * fiemap, bigallic, and seek_data/hole need to use it.
1210 		 */
1211 		if (ext4_es_is_delayed(es))
1212 			goto next;
1213 		if (ext4_es_is_referenced(es)) {
1214 			ext4_es_clear_referenced(es);
1215 			goto next;
1216 		}
1217 
1218 		rb_erase(&es->rb_node, &tree->root);
1219 		ext4_es_free_extent(inode, es);
1220 		(*nr_shrunk)++;
1221 next:
1222 		if (!node)
1223 			goto out_wrap;
1224 		es = rb_entry(node, struct extent_status, rb_node);
1225 	}
1226 	ei->i_es_shrink_lblk = es->es_lblk;
1227 	return 1;
1228 out_wrap:
1229 	ei->i_es_shrink_lblk = 0;
1230 	return 0;
1231 }
1232 
1233 static int es_reclaim_extents(struct ext4_inode_info *ei, int *nr_to_scan)
1234 {
1235 	struct inode *inode = &ei->vfs_inode;
1236 	int nr_shrunk = 0;
1237 	ext4_lblk_t start = ei->i_es_shrink_lblk;
1238 	static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
1239 				      DEFAULT_RATELIMIT_BURST);
1240 
1241 	if (ei->i_es_shk_nr == 0)
1242 		return 0;
1243 
1244 	if (ext4_test_inode_state(inode, EXT4_STATE_EXT_PRECACHED) &&
1245 	    __ratelimit(&_rs))
1246 		ext4_warning(inode->i_sb, "forced shrink of precached extents");
1247 
1248 	if (!es_do_reclaim_extents(ei, EXT_MAX_BLOCKS, nr_to_scan, &nr_shrunk) &&
1249 	    start != 0)
1250 		es_do_reclaim_extents(ei, start - 1, nr_to_scan, &nr_shrunk);
1251 
1252 	ei->i_es_tree.cache_es = NULL;
1253 	return nr_shrunk;
1254 }
1255