xref: /openbmc/linux/fs/btrfs/defrag.c (revision 0bb020dc)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5 
6 #include <linux/sched.h>
7 #include "ctree.h"
8 #include "disk-io.h"
9 #include "print-tree.h"
10 #include "transaction.h"
11 #include "locking.h"
12 #include "accessors.h"
13 #include "messages.h"
14 #include "delalloc-space.h"
15 #include "subpage.h"
16 #include "defrag.h"
17 #include "file-item.h"
18 #include "super.h"
19 
20 static struct kmem_cache *btrfs_inode_defrag_cachep;
21 
22 /*
23  * When auto defrag is enabled we queue up these defrag structs to remember
24  * which inodes need defragging passes.
25  */
26 struct inode_defrag {
27 	struct rb_node rb_node;
28 	/* Inode number */
29 	u64 ino;
30 	/*
31 	 * Transid where the defrag was added, we search for extents newer than
32 	 * this.
33 	 */
34 	u64 transid;
35 
36 	/* Root objectid */
37 	u64 root;
38 
39 	/*
40 	 * The extent size threshold for autodefrag.
41 	 *
42 	 * This value is different for compressed/non-compressed extents, thus
43 	 * needs to be passed from higher layer.
44 	 * (aka, inode_should_defrag())
45 	 */
46 	u32 extent_thresh;
47 };
48 
__compare_inode_defrag(struct inode_defrag * defrag1,struct inode_defrag * defrag2)49 static int __compare_inode_defrag(struct inode_defrag *defrag1,
50 				  struct inode_defrag *defrag2)
51 {
52 	if (defrag1->root > defrag2->root)
53 		return 1;
54 	else if (defrag1->root < defrag2->root)
55 		return -1;
56 	else if (defrag1->ino > defrag2->ino)
57 		return 1;
58 	else if (defrag1->ino < defrag2->ino)
59 		return -1;
60 	else
61 		return 0;
62 }
63 
64 /*
65  * Pop a record for an inode into the defrag tree.  The lock must be held
66  * already.
67  *
68  * If you're inserting a record for an older transid than an existing record,
69  * the transid already in the tree is lowered.
70  *
71  * If an existing record is found the defrag item you pass in is freed.
72  */
__btrfs_add_inode_defrag(struct btrfs_inode * inode,struct inode_defrag * defrag)73 static int __btrfs_add_inode_defrag(struct btrfs_inode *inode,
74 				    struct inode_defrag *defrag)
75 {
76 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
77 	struct inode_defrag *entry;
78 	struct rb_node **p;
79 	struct rb_node *parent = NULL;
80 	int ret;
81 
82 	p = &fs_info->defrag_inodes.rb_node;
83 	while (*p) {
84 		parent = *p;
85 		entry = rb_entry(parent, struct inode_defrag, rb_node);
86 
87 		ret = __compare_inode_defrag(defrag, entry);
88 		if (ret < 0)
89 			p = &parent->rb_left;
90 		else if (ret > 0)
91 			p = &parent->rb_right;
92 		else {
93 			/*
94 			 * If we're reinserting an entry for an old defrag run,
95 			 * make sure to lower the transid of our existing
96 			 * record.
97 			 */
98 			if (defrag->transid < entry->transid)
99 				entry->transid = defrag->transid;
100 			entry->extent_thresh = min(defrag->extent_thresh,
101 						   entry->extent_thresh);
102 			return -EEXIST;
103 		}
104 	}
105 	set_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags);
106 	rb_link_node(&defrag->rb_node, parent, p);
107 	rb_insert_color(&defrag->rb_node, &fs_info->defrag_inodes);
108 	return 0;
109 }
110 
__need_auto_defrag(struct btrfs_fs_info * fs_info)111 static inline int __need_auto_defrag(struct btrfs_fs_info *fs_info)
112 {
113 	if (!btrfs_test_opt(fs_info, AUTO_DEFRAG))
114 		return 0;
115 
116 	if (btrfs_fs_closing(fs_info))
117 		return 0;
118 
119 	return 1;
120 }
121 
122 /*
123  * Insert a defrag record for this inode if auto defrag is enabled.
124  */
btrfs_add_inode_defrag(struct btrfs_trans_handle * trans,struct btrfs_inode * inode,u32 extent_thresh)125 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
126 			   struct btrfs_inode *inode, u32 extent_thresh)
127 {
128 	struct btrfs_root *root = inode->root;
129 	struct btrfs_fs_info *fs_info = root->fs_info;
130 	struct inode_defrag *defrag;
131 	u64 transid;
132 	int ret;
133 
134 	if (!__need_auto_defrag(fs_info))
135 		return 0;
136 
137 	if (test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags))
138 		return 0;
139 
140 	if (trans)
141 		transid = trans->transid;
142 	else
143 		transid = inode->root->last_trans;
144 
145 	defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
146 	if (!defrag)
147 		return -ENOMEM;
148 
149 	defrag->ino = btrfs_ino(inode);
150 	defrag->transid = transid;
151 	defrag->root = root->root_key.objectid;
152 	defrag->extent_thresh = extent_thresh;
153 
154 	spin_lock(&fs_info->defrag_inodes_lock);
155 	if (!test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags)) {
156 		/*
157 		 * If we set IN_DEFRAG flag and evict the inode from memory,
158 		 * and then re-read this inode, this new inode doesn't have
159 		 * IN_DEFRAG flag. At the case, we may find the existed defrag.
160 		 */
161 		ret = __btrfs_add_inode_defrag(inode, defrag);
162 		if (ret)
163 			kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
164 	} else {
165 		kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
166 	}
167 	spin_unlock(&fs_info->defrag_inodes_lock);
168 	return 0;
169 }
170 
171 /*
172  * Pick the defragable inode that we want, if it doesn't exist, we will get the
173  * next one.
174  */
btrfs_pick_defrag_inode(struct btrfs_fs_info * fs_info,u64 root,u64 ino)175 static struct inode_defrag *btrfs_pick_defrag_inode(
176 			struct btrfs_fs_info *fs_info, u64 root, u64 ino)
177 {
178 	struct inode_defrag *entry = NULL;
179 	struct inode_defrag tmp;
180 	struct rb_node *p;
181 	struct rb_node *parent = NULL;
182 	int ret;
183 
184 	tmp.ino = ino;
185 	tmp.root = root;
186 
187 	spin_lock(&fs_info->defrag_inodes_lock);
188 	p = fs_info->defrag_inodes.rb_node;
189 	while (p) {
190 		parent = p;
191 		entry = rb_entry(parent, struct inode_defrag, rb_node);
192 
193 		ret = __compare_inode_defrag(&tmp, entry);
194 		if (ret < 0)
195 			p = parent->rb_left;
196 		else if (ret > 0)
197 			p = parent->rb_right;
198 		else
199 			goto out;
200 	}
201 
202 	if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
203 		parent = rb_next(parent);
204 		if (parent)
205 			entry = rb_entry(parent, struct inode_defrag, rb_node);
206 		else
207 			entry = NULL;
208 	}
209 out:
210 	if (entry)
211 		rb_erase(parent, &fs_info->defrag_inodes);
212 	spin_unlock(&fs_info->defrag_inodes_lock);
213 	return entry;
214 }
215 
btrfs_cleanup_defrag_inodes(struct btrfs_fs_info * fs_info)216 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
217 {
218 	struct inode_defrag *defrag;
219 	struct rb_node *node;
220 
221 	spin_lock(&fs_info->defrag_inodes_lock);
222 	node = rb_first(&fs_info->defrag_inodes);
223 	while (node) {
224 		rb_erase(node, &fs_info->defrag_inodes);
225 		defrag = rb_entry(node, struct inode_defrag, rb_node);
226 		kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
227 
228 		cond_resched_lock(&fs_info->defrag_inodes_lock);
229 
230 		node = rb_first(&fs_info->defrag_inodes);
231 	}
232 	spin_unlock(&fs_info->defrag_inodes_lock);
233 }
234 
235 #define BTRFS_DEFRAG_BATCH	1024
236 
__btrfs_run_defrag_inode(struct btrfs_fs_info * fs_info,struct inode_defrag * defrag)237 static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
238 				    struct inode_defrag *defrag)
239 {
240 	struct btrfs_root *inode_root;
241 	struct inode *inode;
242 	struct btrfs_ioctl_defrag_range_args range;
243 	int ret = 0;
244 	u64 cur = 0;
245 
246 again:
247 	if (test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state))
248 		goto cleanup;
249 	if (!__need_auto_defrag(fs_info))
250 		goto cleanup;
251 
252 	/* Get the inode */
253 	inode_root = btrfs_get_fs_root(fs_info, defrag->root, true);
254 	if (IS_ERR(inode_root)) {
255 		ret = PTR_ERR(inode_root);
256 		goto cleanup;
257 	}
258 
259 	inode = btrfs_iget(fs_info->sb, defrag->ino, inode_root);
260 	btrfs_put_root(inode_root);
261 	if (IS_ERR(inode)) {
262 		ret = PTR_ERR(inode);
263 		goto cleanup;
264 	}
265 
266 	if (cur >= i_size_read(inode)) {
267 		iput(inode);
268 		goto cleanup;
269 	}
270 
271 	/* Do a chunk of defrag */
272 	clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
273 	memset(&range, 0, sizeof(range));
274 	range.len = (u64)-1;
275 	range.start = cur;
276 	range.extent_thresh = defrag->extent_thresh;
277 
278 	sb_start_write(fs_info->sb);
279 	ret = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
280 				       BTRFS_DEFRAG_BATCH);
281 	sb_end_write(fs_info->sb);
282 	iput(inode);
283 
284 	if (ret < 0)
285 		goto cleanup;
286 
287 	cur = max(cur + fs_info->sectorsize, range.start);
288 	goto again;
289 
290 cleanup:
291 	kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
292 	return ret;
293 }
294 
295 /*
296  * Run through the list of inodes in the FS that need defragging.
297  */
btrfs_run_defrag_inodes(struct btrfs_fs_info * fs_info)298 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
299 {
300 	struct inode_defrag *defrag;
301 	u64 first_ino = 0;
302 	u64 root_objectid = 0;
303 
304 	atomic_inc(&fs_info->defrag_running);
305 	while (1) {
306 		/* Pause the auto defragger. */
307 		if (test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state))
308 			break;
309 
310 		if (!__need_auto_defrag(fs_info))
311 			break;
312 
313 		/* find an inode to defrag */
314 		defrag = btrfs_pick_defrag_inode(fs_info, root_objectid, first_ino);
315 		if (!defrag) {
316 			if (root_objectid || first_ino) {
317 				root_objectid = 0;
318 				first_ino = 0;
319 				continue;
320 			} else {
321 				break;
322 			}
323 		}
324 
325 		first_ino = defrag->ino + 1;
326 		root_objectid = defrag->root;
327 
328 		__btrfs_run_defrag_inode(fs_info, defrag);
329 	}
330 	atomic_dec(&fs_info->defrag_running);
331 
332 	/*
333 	 * During unmount, we use the transaction_wait queue to wait for the
334 	 * defragger to stop.
335 	 */
336 	wake_up(&fs_info->transaction_wait);
337 	return 0;
338 }
339 
340 /*
341  * Defrag all the leaves in a given btree.
342  * Read all the leaves and try to get key order to
343  * better reflect disk order
344  */
345 
btrfs_defrag_leaves(struct btrfs_trans_handle * trans,struct btrfs_root * root)346 int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
347 			struct btrfs_root *root)
348 {
349 	struct btrfs_path *path = NULL;
350 	struct btrfs_key key;
351 	int ret = 0;
352 	int wret;
353 	int level;
354 	int next_key_ret = 0;
355 	u64 last_ret = 0;
356 
357 	if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
358 		goto out;
359 
360 	path = btrfs_alloc_path();
361 	if (!path) {
362 		ret = -ENOMEM;
363 		goto out;
364 	}
365 
366 	level = btrfs_header_level(root->node);
367 
368 	if (level == 0)
369 		goto out;
370 
371 	if (root->defrag_progress.objectid == 0) {
372 		struct extent_buffer *root_node;
373 		u32 nritems;
374 
375 		root_node = btrfs_lock_root_node(root);
376 		nritems = btrfs_header_nritems(root_node);
377 		root->defrag_max.objectid = 0;
378 		/* from above we know this is not a leaf */
379 		btrfs_node_key_to_cpu(root_node, &root->defrag_max,
380 				      nritems - 1);
381 		btrfs_tree_unlock(root_node);
382 		free_extent_buffer(root_node);
383 		memset(&key, 0, sizeof(key));
384 	} else {
385 		memcpy(&key, &root->defrag_progress, sizeof(key));
386 	}
387 
388 	path->keep_locks = 1;
389 
390 	ret = btrfs_search_forward(root, &key, path, BTRFS_OLDEST_GENERATION);
391 	if (ret < 0)
392 		goto out;
393 	if (ret > 0) {
394 		ret = 0;
395 		goto out;
396 	}
397 	btrfs_release_path(path);
398 	/*
399 	 * We don't need a lock on a leaf. btrfs_realloc_node() will lock all
400 	 * leafs from path->nodes[1], so set lowest_level to 1 to avoid later
401 	 * a deadlock (attempting to write lock an already write locked leaf).
402 	 */
403 	path->lowest_level = 1;
404 	wret = btrfs_search_slot(trans, root, &key, path, 0, 1);
405 
406 	if (wret < 0) {
407 		ret = wret;
408 		goto out;
409 	}
410 	if (!path->nodes[1]) {
411 		ret = 0;
412 		goto out;
413 	}
414 	/*
415 	 * The node at level 1 must always be locked when our path has
416 	 * keep_locks set and lowest_level is 1, regardless of the value of
417 	 * path->slots[1].
418 	 */
419 	BUG_ON(path->locks[1] == 0);
420 	ret = btrfs_realloc_node(trans, root,
421 				 path->nodes[1], 0,
422 				 &last_ret,
423 				 &root->defrag_progress);
424 	if (ret) {
425 		WARN_ON(ret == -EAGAIN);
426 		goto out;
427 	}
428 	/*
429 	 * Now that we reallocated the node we can find the next key. Note that
430 	 * btrfs_find_next_key() can release our path and do another search
431 	 * without COWing, this is because even with path->keep_locks = 1,
432 	 * btrfs_search_slot() / ctree.c:unlock_up() does not keeps a lock on a
433 	 * node when path->slots[node_level - 1] does not point to the last
434 	 * item or a slot beyond the last item (ctree.c:unlock_up()). Therefore
435 	 * we search for the next key after reallocating our node.
436 	 */
437 	path->slots[1] = btrfs_header_nritems(path->nodes[1]);
438 	next_key_ret = btrfs_find_next_key(root, path, &key, 1,
439 					   BTRFS_OLDEST_GENERATION);
440 	if (next_key_ret == 0) {
441 		memcpy(&root->defrag_progress, &key, sizeof(key));
442 		ret = -EAGAIN;
443 	}
444 out:
445 	btrfs_free_path(path);
446 	if (ret == -EAGAIN) {
447 		if (root->defrag_max.objectid > root->defrag_progress.objectid)
448 			goto done;
449 		if (root->defrag_max.type > root->defrag_progress.type)
450 			goto done;
451 		if (root->defrag_max.offset > root->defrag_progress.offset)
452 			goto done;
453 		ret = 0;
454 	}
455 done:
456 	if (ret != -EAGAIN)
457 		memset(&root->defrag_progress, 0,
458 		       sizeof(root->defrag_progress));
459 
460 	return ret;
461 }
462 
463 /*
464  * Defrag specific helper to get an extent map.
465  *
466  * Differences between this and btrfs_get_extent() are:
467  *
468  * - No extent_map will be added to inode->extent_tree
469  *   To reduce memory usage in the long run.
470  *
471  * - Extra optimization to skip file extents older than @newer_than
472  *   By using btrfs_search_forward() we can skip entire file ranges that
473  *   have extents created in past transactions, because btrfs_search_forward()
474  *   will not visit leaves and nodes with a generation smaller than given
475  *   minimal generation threshold (@newer_than).
476  *
477  * Return valid em if we find a file extent matching the requirement.
478  * Return NULL if we can not find a file extent matching the requirement.
479  *
480  * Return ERR_PTR() for error.
481  */
defrag_get_extent(struct btrfs_inode * inode,u64 start,u64 newer_than)482 static struct extent_map *defrag_get_extent(struct btrfs_inode *inode,
483 					    u64 start, u64 newer_than)
484 {
485 	struct btrfs_root *root = inode->root;
486 	struct btrfs_file_extent_item *fi;
487 	struct btrfs_path path = { 0 };
488 	struct extent_map *em;
489 	struct btrfs_key key;
490 	u64 ino = btrfs_ino(inode);
491 	int ret;
492 
493 	em = alloc_extent_map();
494 	if (!em) {
495 		ret = -ENOMEM;
496 		goto err;
497 	}
498 
499 	key.objectid = ino;
500 	key.type = BTRFS_EXTENT_DATA_KEY;
501 	key.offset = start;
502 
503 	if (newer_than) {
504 		ret = btrfs_search_forward(root, &key, &path, newer_than);
505 		if (ret < 0)
506 			goto err;
507 		/* Can't find anything newer */
508 		if (ret > 0)
509 			goto not_found;
510 	} else {
511 		ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
512 		if (ret < 0)
513 			goto err;
514 	}
515 	if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
516 		/*
517 		 * If btrfs_search_slot() makes path to point beyond nritems,
518 		 * we should not have an empty leaf, as this inode must at
519 		 * least have its INODE_ITEM.
520 		 */
521 		ASSERT(btrfs_header_nritems(path.nodes[0]));
522 		path.slots[0] = btrfs_header_nritems(path.nodes[0]) - 1;
523 	}
524 	btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
525 	/* Perfect match, no need to go one slot back */
526 	if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY &&
527 	    key.offset == start)
528 		goto iterate;
529 
530 	/* We didn't find a perfect match, needs to go one slot back */
531 	if (path.slots[0] > 0) {
532 		btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
533 		if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
534 			path.slots[0]--;
535 	}
536 
537 iterate:
538 	/* Iterate through the path to find a file extent covering @start */
539 	while (true) {
540 		u64 extent_end;
541 
542 		if (path.slots[0] >= btrfs_header_nritems(path.nodes[0]))
543 			goto next;
544 
545 		btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
546 
547 		/*
548 		 * We may go one slot back to INODE_REF/XATTR item, then
549 		 * need to go forward until we reach an EXTENT_DATA.
550 		 * But we should still has the correct ino as key.objectid.
551 		 */
552 		if (WARN_ON(key.objectid < ino) || key.type < BTRFS_EXTENT_DATA_KEY)
553 			goto next;
554 
555 		/* It's beyond our target range, definitely not extent found */
556 		if (key.objectid > ino || key.type > BTRFS_EXTENT_DATA_KEY)
557 			goto not_found;
558 
559 		/*
560 		 *	|	|<- File extent ->|
561 		 *	\- start
562 		 *
563 		 * This means there is a hole between start and key.offset.
564 		 */
565 		if (key.offset > start) {
566 			em->start = start;
567 			em->orig_start = start;
568 			em->block_start = EXTENT_MAP_HOLE;
569 			em->len = key.offset - start;
570 			break;
571 		}
572 
573 		fi = btrfs_item_ptr(path.nodes[0], path.slots[0],
574 				    struct btrfs_file_extent_item);
575 		extent_end = btrfs_file_extent_end(&path);
576 
577 		/*
578 		 *	|<- file extent ->|	|
579 		 *				\- start
580 		 *
581 		 * We haven't reached start, search next slot.
582 		 */
583 		if (extent_end <= start)
584 			goto next;
585 
586 		/* Now this extent covers @start, convert it to em */
587 		btrfs_extent_item_to_extent_map(inode, &path, fi, em);
588 		break;
589 next:
590 		ret = btrfs_next_item(root, &path);
591 		if (ret < 0)
592 			goto err;
593 		if (ret > 0)
594 			goto not_found;
595 	}
596 	btrfs_release_path(&path);
597 	return em;
598 
599 not_found:
600 	btrfs_release_path(&path);
601 	free_extent_map(em);
602 	return NULL;
603 
604 err:
605 	btrfs_release_path(&path);
606 	free_extent_map(em);
607 	return ERR_PTR(ret);
608 }
609 
defrag_lookup_extent(struct inode * inode,u64 start,u64 newer_than,bool locked)610 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start,
611 					       u64 newer_than, bool locked)
612 {
613 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
614 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
615 	struct extent_map *em;
616 	const u32 sectorsize = BTRFS_I(inode)->root->fs_info->sectorsize;
617 
618 	/*
619 	 * Hopefully we have this extent in the tree already, try without the
620 	 * full extent lock.
621 	 */
622 	read_lock(&em_tree->lock);
623 	em = lookup_extent_mapping(em_tree, start, sectorsize);
624 	read_unlock(&em_tree->lock);
625 
626 	/*
627 	 * We can get a merged extent, in that case, we need to re-search
628 	 * tree to get the original em for defrag.
629 	 *
630 	 * If @newer_than is 0 or em::generation < newer_than, we can trust
631 	 * this em, as either we don't care about the generation, or the
632 	 * merged extent map will be rejected anyway.
633 	 */
634 	if (em && test_bit(EXTENT_FLAG_MERGED, &em->flags) &&
635 	    newer_than && em->generation >= newer_than) {
636 		free_extent_map(em);
637 		em = NULL;
638 	}
639 
640 	if (!em) {
641 		struct extent_state *cached = NULL;
642 		u64 end = start + sectorsize - 1;
643 
644 		/* Get the big lock and read metadata off disk. */
645 		if (!locked)
646 			lock_extent(io_tree, start, end, &cached);
647 		em = defrag_get_extent(BTRFS_I(inode), start, newer_than);
648 		if (!locked)
649 			unlock_extent(io_tree, start, end, &cached);
650 
651 		if (IS_ERR(em))
652 			return NULL;
653 	}
654 
655 	return em;
656 }
657 
get_extent_max_capacity(const struct btrfs_fs_info * fs_info,const struct extent_map * em)658 static u32 get_extent_max_capacity(const struct btrfs_fs_info *fs_info,
659 				   const struct extent_map *em)
660 {
661 	if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
662 		return BTRFS_MAX_COMPRESSED;
663 	return fs_info->max_extent_size;
664 }
665 
defrag_check_next_extent(struct inode * inode,struct extent_map * em,u32 extent_thresh,u64 newer_than,bool locked)666 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em,
667 				     u32 extent_thresh, u64 newer_than, bool locked)
668 {
669 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
670 	struct extent_map *next;
671 	bool ret = false;
672 
673 	/* This is the last extent */
674 	if (em->start + em->len >= i_size_read(inode))
675 		return false;
676 
677 	/*
678 	 * Here we need to pass @newer_then when checking the next extent, or
679 	 * we will hit a case we mark current extent for defrag, but the next
680 	 * one will not be a target.
681 	 * This will just cause extra IO without really reducing the fragments.
682 	 */
683 	next = defrag_lookup_extent(inode, em->start + em->len, newer_than, locked);
684 	/* No more em or hole */
685 	if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
686 		goto out;
687 	if (test_bit(EXTENT_FLAG_PREALLOC, &next->flags))
688 		goto out;
689 	/*
690 	 * If the next extent is at its max capacity, defragging current extent
691 	 * makes no sense, as the total number of extents won't change.
692 	 */
693 	if (next->len >= get_extent_max_capacity(fs_info, em))
694 		goto out;
695 	/* Skip older extent */
696 	if (next->generation < newer_than)
697 		goto out;
698 	/* Also check extent size */
699 	if (next->len >= extent_thresh)
700 		goto out;
701 
702 	ret = true;
703 out:
704 	free_extent_map(next);
705 	return ret;
706 }
707 
708 /*
709  * Prepare one page to be defragged.
710  *
711  * This will ensure:
712  *
713  * - Returned page is locked and has been set up properly.
714  * - No ordered extent exists in the page.
715  * - The page is uptodate.
716  *
717  * NOTE: Caller should also wait for page writeback after the cluster is
718  * prepared, here we don't do writeback wait for each page.
719  */
defrag_prepare_one_page(struct btrfs_inode * inode,pgoff_t index)720 static struct page *defrag_prepare_one_page(struct btrfs_inode *inode, pgoff_t index)
721 {
722 	struct address_space *mapping = inode->vfs_inode.i_mapping;
723 	gfp_t mask = btrfs_alloc_write_mask(mapping);
724 	u64 page_start = (u64)index << PAGE_SHIFT;
725 	u64 page_end = page_start + PAGE_SIZE - 1;
726 	struct extent_state *cached_state = NULL;
727 	struct page *page;
728 	int ret;
729 
730 again:
731 	page = find_or_create_page(mapping, index, mask);
732 	if (!page)
733 		return ERR_PTR(-ENOMEM);
734 
735 	/*
736 	 * Since we can defragment files opened read-only, we can encounter
737 	 * transparent huge pages here (see CONFIG_READ_ONLY_THP_FOR_FS). We
738 	 * can't do I/O using huge pages yet, so return an error for now.
739 	 * Filesystem transparent huge pages are typically only used for
740 	 * executables that explicitly enable them, so this isn't very
741 	 * restrictive.
742 	 */
743 	if (PageCompound(page)) {
744 		unlock_page(page);
745 		put_page(page);
746 		return ERR_PTR(-ETXTBSY);
747 	}
748 
749 	ret = set_page_extent_mapped(page);
750 	if (ret < 0) {
751 		unlock_page(page);
752 		put_page(page);
753 		return ERR_PTR(ret);
754 	}
755 
756 	/* Wait for any existing ordered extent in the range */
757 	while (1) {
758 		struct btrfs_ordered_extent *ordered;
759 
760 		lock_extent(&inode->io_tree, page_start, page_end, &cached_state);
761 		ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
762 		unlock_extent(&inode->io_tree, page_start, page_end,
763 			      &cached_state);
764 		if (!ordered)
765 			break;
766 
767 		unlock_page(page);
768 		btrfs_start_ordered_extent(ordered);
769 		btrfs_put_ordered_extent(ordered);
770 		lock_page(page);
771 		/*
772 		 * We unlocked the page above, so we need check if it was
773 		 * released or not.
774 		 */
775 		if (page->mapping != mapping || !PagePrivate(page)) {
776 			unlock_page(page);
777 			put_page(page);
778 			goto again;
779 		}
780 	}
781 
782 	/*
783 	 * Now the page range has no ordered extent any more.  Read the page to
784 	 * make it uptodate.
785 	 */
786 	if (!PageUptodate(page)) {
787 		btrfs_read_folio(NULL, page_folio(page));
788 		lock_page(page);
789 		if (page->mapping != mapping || !PagePrivate(page)) {
790 			unlock_page(page);
791 			put_page(page);
792 			goto again;
793 		}
794 		if (!PageUptodate(page)) {
795 			unlock_page(page);
796 			put_page(page);
797 			return ERR_PTR(-EIO);
798 		}
799 	}
800 	return page;
801 }
802 
803 struct defrag_target_range {
804 	struct list_head list;
805 	u64 start;
806 	u64 len;
807 };
808 
809 /*
810  * Collect all valid target extents.
811  *
812  * @start:	   file offset to lookup
813  * @len:	   length to lookup
814  * @extent_thresh: file extent size threshold, any extent size >= this value
815  *		   will be ignored
816  * @newer_than:    only defrag extents newer than this value
817  * @do_compress:   whether the defrag is doing compression
818  *		   if true, @extent_thresh will be ignored and all regular
819  *		   file extents meeting @newer_than will be targets.
820  * @locked:	   if the range has already held extent lock
821  * @target_list:   list of targets file extents
822  */
defrag_collect_targets(struct btrfs_inode * inode,u64 start,u64 len,u32 extent_thresh,u64 newer_than,bool do_compress,bool locked,struct list_head * target_list,u64 * last_scanned_ret)823 static int defrag_collect_targets(struct btrfs_inode *inode,
824 				  u64 start, u64 len, u32 extent_thresh,
825 				  u64 newer_than, bool do_compress,
826 				  bool locked, struct list_head *target_list,
827 				  u64 *last_scanned_ret)
828 {
829 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
830 	bool last_is_target = false;
831 	u64 cur = start;
832 	int ret = 0;
833 
834 	while (cur < start + len) {
835 		struct extent_map *em;
836 		struct defrag_target_range *new;
837 		bool next_mergeable = true;
838 		u64 range_len;
839 
840 		last_is_target = false;
841 		em = defrag_lookup_extent(&inode->vfs_inode, cur, newer_than, locked);
842 		if (!em)
843 			break;
844 
845 		/*
846 		 * If the file extent is an inlined one, we may still want to
847 		 * defrag it (fallthrough) if it will cause a regular extent.
848 		 * This is for users who want to convert inline extents to
849 		 * regular ones through max_inline= mount option.
850 		 */
851 		if (em->block_start == EXTENT_MAP_INLINE &&
852 		    em->len <= inode->root->fs_info->max_inline)
853 			goto next;
854 
855 		/* Skip hole/delalloc/preallocated extents */
856 		if (em->block_start == EXTENT_MAP_HOLE ||
857 		    em->block_start == EXTENT_MAP_DELALLOC ||
858 		    test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
859 			goto next;
860 
861 		/* Skip older extent */
862 		if (em->generation < newer_than)
863 			goto next;
864 
865 		/* This em is under writeback, no need to defrag */
866 		if (em->generation == (u64)-1)
867 			goto next;
868 
869 		/*
870 		 * Our start offset might be in the middle of an existing extent
871 		 * map, so take that into account.
872 		 */
873 		range_len = em->len - (cur - em->start);
874 		/*
875 		 * If this range of the extent map is already flagged for delalloc,
876 		 * skip it, because:
877 		 *
878 		 * 1) We could deadlock later, when trying to reserve space for
879 		 *    delalloc, because in case we can't immediately reserve space
880 		 *    the flusher can start delalloc and wait for the respective
881 		 *    ordered extents to complete. The deadlock would happen
882 		 *    because we do the space reservation while holding the range
883 		 *    locked, and starting writeback, or finishing an ordered
884 		 *    extent, requires locking the range;
885 		 *
886 		 * 2) If there's delalloc there, it means there's dirty pages for
887 		 *    which writeback has not started yet (we clean the delalloc
888 		 *    flag when starting writeback and after creating an ordered
889 		 *    extent). If we mark pages in an adjacent range for defrag,
890 		 *    then we will have a larger contiguous range for delalloc,
891 		 *    very likely resulting in a larger extent after writeback is
892 		 *    triggered (except in a case of free space fragmentation).
893 		 */
894 		if (test_range_bit(&inode->io_tree, cur, cur + range_len - 1,
895 				   EXTENT_DELALLOC, 0, NULL))
896 			goto next;
897 
898 		/*
899 		 * For do_compress case, we want to compress all valid file
900 		 * extents, thus no @extent_thresh or mergeable check.
901 		 */
902 		if (do_compress)
903 			goto add;
904 
905 		/* Skip too large extent */
906 		if (em->len >= extent_thresh)
907 			goto next;
908 
909 		/*
910 		 * Skip extents already at its max capacity, this is mostly for
911 		 * compressed extents, which max cap is only 128K.
912 		 */
913 		if (em->len >= get_extent_max_capacity(fs_info, em))
914 			goto next;
915 
916 		/*
917 		 * Normally there are no more extents after an inline one, thus
918 		 * @next_mergeable will normally be false and not defragged.
919 		 * So if an inline extent passed all above checks, just add it
920 		 * for defrag, and be converted to regular extents.
921 		 */
922 		if (em->block_start == EXTENT_MAP_INLINE)
923 			goto add;
924 
925 		next_mergeable = defrag_check_next_extent(&inode->vfs_inode, em,
926 						extent_thresh, newer_than, locked);
927 		if (!next_mergeable) {
928 			struct defrag_target_range *last;
929 
930 			/* Empty target list, no way to merge with last entry */
931 			if (list_empty(target_list))
932 				goto next;
933 			last = list_entry(target_list->prev,
934 					  struct defrag_target_range, list);
935 			/* Not mergeable with last entry */
936 			if (last->start + last->len != cur)
937 				goto next;
938 
939 			/* Mergeable, fall through to add it to @target_list. */
940 		}
941 
942 add:
943 		last_is_target = true;
944 		range_len = min(extent_map_end(em), start + len) - cur;
945 		/*
946 		 * This one is a good target, check if it can be merged into
947 		 * last range of the target list.
948 		 */
949 		if (!list_empty(target_list)) {
950 			struct defrag_target_range *last;
951 
952 			last = list_entry(target_list->prev,
953 					  struct defrag_target_range, list);
954 			ASSERT(last->start + last->len <= cur);
955 			if (last->start + last->len == cur) {
956 				/* Mergeable, enlarge the last entry */
957 				last->len += range_len;
958 				goto next;
959 			}
960 			/* Fall through to allocate a new entry */
961 		}
962 
963 		/* Allocate new defrag_target_range */
964 		new = kmalloc(sizeof(*new), GFP_NOFS);
965 		if (!new) {
966 			free_extent_map(em);
967 			ret = -ENOMEM;
968 			break;
969 		}
970 		new->start = cur;
971 		new->len = range_len;
972 		list_add_tail(&new->list, target_list);
973 
974 next:
975 		cur = extent_map_end(em);
976 		free_extent_map(em);
977 	}
978 	if (ret < 0) {
979 		struct defrag_target_range *entry;
980 		struct defrag_target_range *tmp;
981 
982 		list_for_each_entry_safe(entry, tmp, target_list, list) {
983 			list_del_init(&entry->list);
984 			kfree(entry);
985 		}
986 	}
987 	if (!ret && last_scanned_ret) {
988 		/*
989 		 * If the last extent is not a target, the caller can skip to
990 		 * the end of that extent.
991 		 * Otherwise, we can only go the end of the specified range.
992 		 */
993 		if (!last_is_target)
994 			*last_scanned_ret = max(cur, *last_scanned_ret);
995 		else
996 			*last_scanned_ret = max(start + len, *last_scanned_ret);
997 	}
998 	return ret;
999 }
1000 
1001 #define CLUSTER_SIZE	(SZ_256K)
1002 static_assert(PAGE_ALIGNED(CLUSTER_SIZE));
1003 
1004 /*
1005  * Defrag one contiguous target range.
1006  *
1007  * @inode:	target inode
1008  * @target:	target range to defrag
1009  * @pages:	locked pages covering the defrag range
1010  * @nr_pages:	number of locked pages
1011  *
1012  * Caller should ensure:
1013  *
1014  * - Pages are prepared
1015  *   Pages should be locked, no ordered extent in the pages range,
1016  *   no writeback.
1017  *
1018  * - Extent bits are locked
1019  */
defrag_one_locked_target(struct btrfs_inode * inode,struct defrag_target_range * target,struct page ** pages,int nr_pages,struct extent_state ** cached_state)1020 static int defrag_one_locked_target(struct btrfs_inode *inode,
1021 				    struct defrag_target_range *target,
1022 				    struct page **pages, int nr_pages,
1023 				    struct extent_state **cached_state)
1024 {
1025 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1026 	struct extent_changeset *data_reserved = NULL;
1027 	const u64 start = target->start;
1028 	const u64 len = target->len;
1029 	unsigned long last_index = (start + len - 1) >> PAGE_SHIFT;
1030 	unsigned long start_index = start >> PAGE_SHIFT;
1031 	unsigned long first_index = page_index(pages[0]);
1032 	int ret = 0;
1033 	int i;
1034 
1035 	ASSERT(last_index - first_index + 1 <= nr_pages);
1036 
1037 	ret = btrfs_delalloc_reserve_space(inode, &data_reserved, start, len);
1038 	if (ret < 0)
1039 		return ret;
1040 	clear_extent_bit(&inode->io_tree, start, start + len - 1,
1041 			 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1042 			 EXTENT_DEFRAG, cached_state);
1043 	set_extent_bit(&inode->io_tree, start, start + len - 1,
1044 		       EXTENT_DELALLOC | EXTENT_DEFRAG, cached_state);
1045 
1046 	/* Update the page status */
1047 	for (i = start_index - first_index; i <= last_index - first_index; i++) {
1048 		ClearPageChecked(pages[i]);
1049 		btrfs_page_clamp_set_dirty(fs_info, pages[i], start, len);
1050 	}
1051 	btrfs_delalloc_release_extents(inode, len);
1052 	extent_changeset_free(data_reserved);
1053 
1054 	return ret;
1055 }
1056 
defrag_one_range(struct btrfs_inode * inode,u64 start,u32 len,u32 extent_thresh,u64 newer_than,bool do_compress,u64 * last_scanned_ret)1057 static int defrag_one_range(struct btrfs_inode *inode, u64 start, u32 len,
1058 			    u32 extent_thresh, u64 newer_than, bool do_compress,
1059 			    u64 *last_scanned_ret)
1060 {
1061 	struct extent_state *cached_state = NULL;
1062 	struct defrag_target_range *entry;
1063 	struct defrag_target_range *tmp;
1064 	LIST_HEAD(target_list);
1065 	struct page **pages;
1066 	const u32 sectorsize = inode->root->fs_info->sectorsize;
1067 	u64 last_index = (start + len - 1) >> PAGE_SHIFT;
1068 	u64 start_index = start >> PAGE_SHIFT;
1069 	unsigned int nr_pages = last_index - start_index + 1;
1070 	int ret = 0;
1071 	int i;
1072 
1073 	ASSERT(nr_pages <= CLUSTER_SIZE / PAGE_SIZE);
1074 	ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(len, sectorsize));
1075 
1076 	pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
1077 	if (!pages)
1078 		return -ENOMEM;
1079 
1080 	/* Prepare all pages */
1081 	for (i = 0; i < nr_pages; i++) {
1082 		pages[i] = defrag_prepare_one_page(inode, start_index + i);
1083 		if (IS_ERR(pages[i])) {
1084 			ret = PTR_ERR(pages[i]);
1085 			pages[i] = NULL;
1086 			goto free_pages;
1087 		}
1088 	}
1089 	for (i = 0; i < nr_pages; i++)
1090 		wait_on_page_writeback(pages[i]);
1091 
1092 	/* Lock the pages range */
1093 	lock_extent(&inode->io_tree, start_index << PAGE_SHIFT,
1094 		    (last_index << PAGE_SHIFT) + PAGE_SIZE - 1,
1095 		    &cached_state);
1096 	/*
1097 	 * Now we have a consistent view about the extent map, re-check
1098 	 * which range really needs to be defragged.
1099 	 *
1100 	 * And this time we have extent locked already, pass @locked = true
1101 	 * so that we won't relock the extent range and cause deadlock.
1102 	 */
1103 	ret = defrag_collect_targets(inode, start, len, extent_thresh,
1104 				     newer_than, do_compress, true,
1105 				     &target_list, last_scanned_ret);
1106 	if (ret < 0)
1107 		goto unlock_extent;
1108 
1109 	list_for_each_entry(entry, &target_list, list) {
1110 		ret = defrag_one_locked_target(inode, entry, pages, nr_pages,
1111 					       &cached_state);
1112 		if (ret < 0)
1113 			break;
1114 	}
1115 
1116 	list_for_each_entry_safe(entry, tmp, &target_list, list) {
1117 		list_del_init(&entry->list);
1118 		kfree(entry);
1119 	}
1120 unlock_extent:
1121 	unlock_extent(&inode->io_tree, start_index << PAGE_SHIFT,
1122 		      (last_index << PAGE_SHIFT) + PAGE_SIZE - 1,
1123 		      &cached_state);
1124 free_pages:
1125 	for (i = 0; i < nr_pages; i++) {
1126 		if (pages[i]) {
1127 			unlock_page(pages[i]);
1128 			put_page(pages[i]);
1129 		}
1130 	}
1131 	kfree(pages);
1132 	return ret;
1133 }
1134 
defrag_one_cluster(struct btrfs_inode * inode,struct file_ra_state * ra,u64 start,u32 len,u32 extent_thresh,u64 newer_than,bool do_compress,unsigned long * sectors_defragged,unsigned long max_sectors,u64 * last_scanned_ret)1135 static int defrag_one_cluster(struct btrfs_inode *inode,
1136 			      struct file_ra_state *ra,
1137 			      u64 start, u32 len, u32 extent_thresh,
1138 			      u64 newer_than, bool do_compress,
1139 			      unsigned long *sectors_defragged,
1140 			      unsigned long max_sectors,
1141 			      u64 *last_scanned_ret)
1142 {
1143 	const u32 sectorsize = inode->root->fs_info->sectorsize;
1144 	struct defrag_target_range *entry;
1145 	struct defrag_target_range *tmp;
1146 	LIST_HEAD(target_list);
1147 	int ret;
1148 
1149 	ret = defrag_collect_targets(inode, start, len, extent_thresh,
1150 				     newer_than, do_compress, false,
1151 				     &target_list, NULL);
1152 	if (ret < 0)
1153 		goto out;
1154 
1155 	list_for_each_entry(entry, &target_list, list) {
1156 		u32 range_len = entry->len;
1157 
1158 		/* Reached or beyond the limit */
1159 		if (max_sectors && *sectors_defragged >= max_sectors) {
1160 			ret = 1;
1161 			break;
1162 		}
1163 
1164 		if (max_sectors)
1165 			range_len = min_t(u32, range_len,
1166 				(max_sectors - *sectors_defragged) * sectorsize);
1167 
1168 		/*
1169 		 * If defrag_one_range() has updated last_scanned_ret,
1170 		 * our range may already be invalid (e.g. hole punched).
1171 		 * Skip if our range is before last_scanned_ret, as there is
1172 		 * no need to defrag the range anymore.
1173 		 */
1174 		if (entry->start + range_len <= *last_scanned_ret)
1175 			continue;
1176 
1177 		if (ra)
1178 			page_cache_sync_readahead(inode->vfs_inode.i_mapping,
1179 				ra, NULL, entry->start >> PAGE_SHIFT,
1180 				((entry->start + range_len - 1) >> PAGE_SHIFT) -
1181 				(entry->start >> PAGE_SHIFT) + 1);
1182 		/*
1183 		 * Here we may not defrag any range if holes are punched before
1184 		 * we locked the pages.
1185 		 * But that's fine, it only affects the @sectors_defragged
1186 		 * accounting.
1187 		 */
1188 		ret = defrag_one_range(inode, entry->start, range_len,
1189 				       extent_thresh, newer_than, do_compress,
1190 				       last_scanned_ret);
1191 		if (ret < 0)
1192 			break;
1193 		*sectors_defragged += range_len >>
1194 				      inode->root->fs_info->sectorsize_bits;
1195 	}
1196 out:
1197 	list_for_each_entry_safe(entry, tmp, &target_list, list) {
1198 		list_del_init(&entry->list);
1199 		kfree(entry);
1200 	}
1201 	if (ret >= 0)
1202 		*last_scanned_ret = max(*last_scanned_ret, start + len);
1203 	return ret;
1204 }
1205 
1206 /*
1207  * Entry point to file defragmentation.
1208  *
1209  * @inode:	   inode to be defragged
1210  * @ra:		   readahead state (can be NUL)
1211  * @range:	   defrag options including range and flags
1212  * @newer_than:	   minimum transid to defrag
1213  * @max_to_defrag: max number of sectors to be defragged, if 0, the whole inode
1214  *		   will be defragged.
1215  *
1216  * Return <0 for error.
1217  * Return >=0 for the number of sectors defragged, and range->start will be updated
1218  * to indicate the file offset where next defrag should be started at.
1219  * (Mostly for autodefrag, which sets @max_to_defrag thus we may exit early without
1220  *  defragging all the range).
1221  */
btrfs_defrag_file(struct inode * inode,struct file_ra_state * ra,struct btrfs_ioctl_defrag_range_args * range,u64 newer_than,unsigned long max_to_defrag)1222 int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra,
1223 		      struct btrfs_ioctl_defrag_range_args *range,
1224 		      u64 newer_than, unsigned long max_to_defrag)
1225 {
1226 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1227 	unsigned long sectors_defragged = 0;
1228 	u64 isize = i_size_read(inode);
1229 	u64 cur;
1230 	u64 last_byte;
1231 	bool do_compress = (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS);
1232 	bool ra_allocated = false;
1233 	int compress_type = BTRFS_COMPRESS_ZLIB;
1234 	int ret = 0;
1235 	u32 extent_thresh = range->extent_thresh;
1236 	pgoff_t start_index;
1237 
1238 	if (isize == 0)
1239 		return 0;
1240 
1241 	if (range->start >= isize)
1242 		return -EINVAL;
1243 
1244 	if (do_compress) {
1245 		if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1246 			return -EINVAL;
1247 		if (range->compress_type)
1248 			compress_type = range->compress_type;
1249 	}
1250 
1251 	if (extent_thresh == 0)
1252 		extent_thresh = SZ_256K;
1253 
1254 	if (range->start + range->len > range->start) {
1255 		/* Got a specific range */
1256 		last_byte = min(isize, range->start + range->len);
1257 	} else {
1258 		/* Defrag until file end */
1259 		last_byte = isize;
1260 	}
1261 
1262 	/* Align the range */
1263 	cur = round_down(range->start, fs_info->sectorsize);
1264 	last_byte = round_up(last_byte, fs_info->sectorsize) - 1;
1265 
1266 	/*
1267 	 * If we were not given a ra, allocate a readahead context. As
1268 	 * readahead is just an optimization, defrag will work without it so
1269 	 * we don't error out.
1270 	 */
1271 	if (!ra) {
1272 		ra_allocated = true;
1273 		ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1274 		if (ra)
1275 			file_ra_state_init(ra, inode->i_mapping);
1276 	}
1277 
1278 	/*
1279 	 * Make writeback start from the beginning of the range, so that the
1280 	 * defrag range can be written sequentially.
1281 	 */
1282 	start_index = cur >> PAGE_SHIFT;
1283 	if (start_index < inode->i_mapping->writeback_index)
1284 		inode->i_mapping->writeback_index = start_index;
1285 
1286 	while (cur < last_byte) {
1287 		const unsigned long prev_sectors_defragged = sectors_defragged;
1288 		u64 last_scanned = cur;
1289 		u64 cluster_end;
1290 
1291 		if (btrfs_defrag_cancelled(fs_info)) {
1292 			ret = -EAGAIN;
1293 			break;
1294 		}
1295 
1296 		/* We want the cluster end at page boundary when possible */
1297 		cluster_end = (((cur >> PAGE_SHIFT) +
1298 			       (SZ_256K >> PAGE_SHIFT)) << PAGE_SHIFT) - 1;
1299 		cluster_end = min(cluster_end, last_byte);
1300 
1301 		btrfs_inode_lock(BTRFS_I(inode), 0);
1302 		if (IS_SWAPFILE(inode)) {
1303 			ret = -ETXTBSY;
1304 			btrfs_inode_unlock(BTRFS_I(inode), 0);
1305 			break;
1306 		}
1307 		if (!(inode->i_sb->s_flags & SB_ACTIVE)) {
1308 			btrfs_inode_unlock(BTRFS_I(inode), 0);
1309 			break;
1310 		}
1311 		if (do_compress)
1312 			BTRFS_I(inode)->defrag_compress = compress_type;
1313 		ret = defrag_one_cluster(BTRFS_I(inode), ra, cur,
1314 				cluster_end + 1 - cur, extent_thresh,
1315 				newer_than, do_compress, &sectors_defragged,
1316 				max_to_defrag, &last_scanned);
1317 
1318 		if (sectors_defragged > prev_sectors_defragged)
1319 			balance_dirty_pages_ratelimited(inode->i_mapping);
1320 
1321 		btrfs_inode_unlock(BTRFS_I(inode), 0);
1322 		if (ret < 0)
1323 			break;
1324 		cur = max(cluster_end + 1, last_scanned);
1325 		if (ret > 0) {
1326 			ret = 0;
1327 			break;
1328 		}
1329 		cond_resched();
1330 	}
1331 
1332 	if (ra_allocated)
1333 		kfree(ra);
1334 	/*
1335 	 * Update range.start for autodefrag, this will indicate where to start
1336 	 * in next run.
1337 	 */
1338 	range->start = cur;
1339 	if (sectors_defragged) {
1340 		/*
1341 		 * We have defragged some sectors, for compression case they
1342 		 * need to be written back immediately.
1343 		 */
1344 		if (range->flags & BTRFS_DEFRAG_RANGE_START_IO) {
1345 			filemap_flush(inode->i_mapping);
1346 			if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1347 				     &BTRFS_I(inode)->runtime_flags))
1348 				filemap_flush(inode->i_mapping);
1349 		}
1350 		if (range->compress_type == BTRFS_COMPRESS_LZO)
1351 			btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1352 		else if (range->compress_type == BTRFS_COMPRESS_ZSTD)
1353 			btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1354 		ret = sectors_defragged;
1355 	}
1356 	if (do_compress) {
1357 		btrfs_inode_lock(BTRFS_I(inode), 0);
1358 		BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1359 		btrfs_inode_unlock(BTRFS_I(inode), 0);
1360 	}
1361 	return ret;
1362 }
1363 
btrfs_auto_defrag_exit(void)1364 void __cold btrfs_auto_defrag_exit(void)
1365 {
1366 	kmem_cache_destroy(btrfs_inode_defrag_cachep);
1367 }
1368 
btrfs_auto_defrag_init(void)1369 int __init btrfs_auto_defrag_init(void)
1370 {
1371 	btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
1372 					sizeof(struct inode_defrag), 0,
1373 					SLAB_MEM_SPREAD,
1374 					NULL);
1375 	if (!btrfs_inode_defrag_cachep)
1376 		return -ENOMEM;
1377 
1378 	return 0;
1379 }
1380