1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2011 Fujitsu.  All rights reserved.
4  * Written by Miao Xie <miaox@cn.fujitsu.com>
5  */
6 
7 #include <linux/slab.h>
8 #include <linux/iversion.h>
9 #include "misc.h"
10 #include "delayed-inode.h"
11 #include "disk-io.h"
12 #include "transaction.h"
13 #include "ctree.h"
14 #include "qgroup.h"
15 #include "locking.h"
16 #include "inode-item.h"
17 
18 #define BTRFS_DELAYED_WRITEBACK		512
19 #define BTRFS_DELAYED_BACKGROUND	128
20 #define BTRFS_DELAYED_BATCH		16
21 
22 static struct kmem_cache *delayed_node_cache;
23 
btrfs_delayed_inode_init(void)24 int __init btrfs_delayed_inode_init(void)
25 {
26 	delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
27 					sizeof(struct btrfs_delayed_node),
28 					0,
29 					SLAB_MEM_SPREAD,
30 					NULL);
31 	if (!delayed_node_cache)
32 		return -ENOMEM;
33 	return 0;
34 }
35 
btrfs_delayed_inode_exit(void)36 void __cold btrfs_delayed_inode_exit(void)
37 {
38 	kmem_cache_destroy(delayed_node_cache);
39 }
40 
btrfs_init_delayed_node(struct btrfs_delayed_node * delayed_node,struct btrfs_root * root,u64 inode_id)41 static inline void btrfs_init_delayed_node(
42 				struct btrfs_delayed_node *delayed_node,
43 				struct btrfs_root *root, u64 inode_id)
44 {
45 	delayed_node->root = root;
46 	delayed_node->inode_id = inode_id;
47 	refcount_set(&delayed_node->refs, 0);
48 	delayed_node->ins_root = RB_ROOT_CACHED;
49 	delayed_node->del_root = RB_ROOT_CACHED;
50 	mutex_init(&delayed_node->mutex);
51 	INIT_LIST_HEAD(&delayed_node->n_list);
52 	INIT_LIST_HEAD(&delayed_node->p_list);
53 }
54 
btrfs_get_delayed_node(struct btrfs_inode * btrfs_inode)55 static struct btrfs_delayed_node *btrfs_get_delayed_node(
56 		struct btrfs_inode *btrfs_inode)
57 {
58 	struct btrfs_root *root = btrfs_inode->root;
59 	u64 ino = btrfs_ino(btrfs_inode);
60 	struct btrfs_delayed_node *node;
61 
62 	node = READ_ONCE(btrfs_inode->delayed_node);
63 	if (node) {
64 		refcount_inc(&node->refs);
65 		return node;
66 	}
67 
68 	spin_lock(&root->inode_lock);
69 	node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
70 
71 	if (node) {
72 		if (btrfs_inode->delayed_node) {
73 			refcount_inc(&node->refs);	/* can be accessed */
74 			BUG_ON(btrfs_inode->delayed_node != node);
75 			spin_unlock(&root->inode_lock);
76 			return node;
77 		}
78 
79 		/*
80 		 * It's possible that we're racing into the middle of removing
81 		 * this node from the radix tree.  In this case, the refcount
82 		 * was zero and it should never go back to one.  Just return
83 		 * NULL like it was never in the radix at all; our release
84 		 * function is in the process of removing it.
85 		 *
86 		 * Some implementations of refcount_inc refuse to bump the
87 		 * refcount once it has hit zero.  If we don't do this dance
88 		 * here, refcount_inc() may decide to just WARN_ONCE() instead
89 		 * of actually bumping the refcount.
90 		 *
91 		 * If this node is properly in the radix, we want to bump the
92 		 * refcount twice, once for the inode and once for this get
93 		 * operation.
94 		 */
95 		if (refcount_inc_not_zero(&node->refs)) {
96 			refcount_inc(&node->refs);
97 			btrfs_inode->delayed_node = node;
98 		} else {
99 			node = NULL;
100 		}
101 
102 		spin_unlock(&root->inode_lock);
103 		return node;
104 	}
105 	spin_unlock(&root->inode_lock);
106 
107 	return NULL;
108 }
109 
110 /* Will return either the node or PTR_ERR(-ENOMEM) */
btrfs_get_or_create_delayed_node(struct btrfs_inode * btrfs_inode)111 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
112 		struct btrfs_inode *btrfs_inode)
113 {
114 	struct btrfs_delayed_node *node;
115 	struct btrfs_root *root = btrfs_inode->root;
116 	u64 ino = btrfs_ino(btrfs_inode);
117 	int ret;
118 
119 again:
120 	node = btrfs_get_delayed_node(btrfs_inode);
121 	if (node)
122 		return node;
123 
124 	node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
125 	if (!node)
126 		return ERR_PTR(-ENOMEM);
127 	btrfs_init_delayed_node(node, root, ino);
128 
129 	/* cached in the btrfs inode and can be accessed */
130 	refcount_set(&node->refs, 2);
131 
132 	ret = radix_tree_preload(GFP_NOFS);
133 	if (ret) {
134 		kmem_cache_free(delayed_node_cache, node);
135 		return ERR_PTR(ret);
136 	}
137 
138 	spin_lock(&root->inode_lock);
139 	ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
140 	if (ret == -EEXIST) {
141 		spin_unlock(&root->inode_lock);
142 		kmem_cache_free(delayed_node_cache, node);
143 		radix_tree_preload_end();
144 		goto again;
145 	}
146 	btrfs_inode->delayed_node = node;
147 	spin_unlock(&root->inode_lock);
148 	radix_tree_preload_end();
149 
150 	return node;
151 }
152 
153 /*
154  * Call it when holding delayed_node->mutex
155  *
156  * If mod = 1, add this node into the prepared list.
157  */
btrfs_queue_delayed_node(struct btrfs_delayed_root * root,struct btrfs_delayed_node * node,int mod)158 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
159 				     struct btrfs_delayed_node *node,
160 				     int mod)
161 {
162 	spin_lock(&root->lock);
163 	if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
164 		if (!list_empty(&node->p_list))
165 			list_move_tail(&node->p_list, &root->prepare_list);
166 		else if (mod)
167 			list_add_tail(&node->p_list, &root->prepare_list);
168 	} else {
169 		list_add_tail(&node->n_list, &root->node_list);
170 		list_add_tail(&node->p_list, &root->prepare_list);
171 		refcount_inc(&node->refs);	/* inserted into list */
172 		root->nodes++;
173 		set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
174 	}
175 	spin_unlock(&root->lock);
176 }
177 
178 /* Call it when holding delayed_node->mutex */
btrfs_dequeue_delayed_node(struct btrfs_delayed_root * root,struct btrfs_delayed_node * node)179 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
180 				       struct btrfs_delayed_node *node)
181 {
182 	spin_lock(&root->lock);
183 	if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
184 		root->nodes--;
185 		refcount_dec(&node->refs);	/* not in the list */
186 		list_del_init(&node->n_list);
187 		if (!list_empty(&node->p_list))
188 			list_del_init(&node->p_list);
189 		clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
190 	}
191 	spin_unlock(&root->lock);
192 }
193 
btrfs_first_delayed_node(struct btrfs_delayed_root * delayed_root)194 static struct btrfs_delayed_node *btrfs_first_delayed_node(
195 			struct btrfs_delayed_root *delayed_root)
196 {
197 	struct list_head *p;
198 	struct btrfs_delayed_node *node = NULL;
199 
200 	spin_lock(&delayed_root->lock);
201 	if (list_empty(&delayed_root->node_list))
202 		goto out;
203 
204 	p = delayed_root->node_list.next;
205 	node = list_entry(p, struct btrfs_delayed_node, n_list);
206 	refcount_inc(&node->refs);
207 out:
208 	spin_unlock(&delayed_root->lock);
209 
210 	return node;
211 }
212 
btrfs_next_delayed_node(struct btrfs_delayed_node * node)213 static struct btrfs_delayed_node *btrfs_next_delayed_node(
214 						struct btrfs_delayed_node *node)
215 {
216 	struct btrfs_delayed_root *delayed_root;
217 	struct list_head *p;
218 	struct btrfs_delayed_node *next = NULL;
219 
220 	delayed_root = node->root->fs_info->delayed_root;
221 	spin_lock(&delayed_root->lock);
222 	if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
223 		/* not in the list */
224 		if (list_empty(&delayed_root->node_list))
225 			goto out;
226 		p = delayed_root->node_list.next;
227 	} else if (list_is_last(&node->n_list, &delayed_root->node_list))
228 		goto out;
229 	else
230 		p = node->n_list.next;
231 
232 	next = list_entry(p, struct btrfs_delayed_node, n_list);
233 	refcount_inc(&next->refs);
234 out:
235 	spin_unlock(&delayed_root->lock);
236 
237 	return next;
238 }
239 
__btrfs_release_delayed_node(struct btrfs_delayed_node * delayed_node,int mod)240 static void __btrfs_release_delayed_node(
241 				struct btrfs_delayed_node *delayed_node,
242 				int mod)
243 {
244 	struct btrfs_delayed_root *delayed_root;
245 
246 	if (!delayed_node)
247 		return;
248 
249 	delayed_root = delayed_node->root->fs_info->delayed_root;
250 
251 	mutex_lock(&delayed_node->mutex);
252 	if (delayed_node->count)
253 		btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
254 	else
255 		btrfs_dequeue_delayed_node(delayed_root, delayed_node);
256 	mutex_unlock(&delayed_node->mutex);
257 
258 	if (refcount_dec_and_test(&delayed_node->refs)) {
259 		struct btrfs_root *root = delayed_node->root;
260 
261 		spin_lock(&root->inode_lock);
262 		/*
263 		 * Once our refcount goes to zero, nobody is allowed to bump it
264 		 * back up.  We can delete it now.
265 		 */
266 		ASSERT(refcount_read(&delayed_node->refs) == 0);
267 		radix_tree_delete(&root->delayed_nodes_tree,
268 				  delayed_node->inode_id);
269 		spin_unlock(&root->inode_lock);
270 		kmem_cache_free(delayed_node_cache, delayed_node);
271 	}
272 }
273 
btrfs_release_delayed_node(struct btrfs_delayed_node * node)274 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
275 {
276 	__btrfs_release_delayed_node(node, 0);
277 }
278 
btrfs_first_prepared_delayed_node(struct btrfs_delayed_root * delayed_root)279 static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
280 					struct btrfs_delayed_root *delayed_root)
281 {
282 	struct list_head *p;
283 	struct btrfs_delayed_node *node = NULL;
284 
285 	spin_lock(&delayed_root->lock);
286 	if (list_empty(&delayed_root->prepare_list))
287 		goto out;
288 
289 	p = delayed_root->prepare_list.next;
290 	list_del_init(p);
291 	node = list_entry(p, struct btrfs_delayed_node, p_list);
292 	refcount_inc(&node->refs);
293 out:
294 	spin_unlock(&delayed_root->lock);
295 
296 	return node;
297 }
298 
btrfs_release_prepared_delayed_node(struct btrfs_delayed_node * node)299 static inline void btrfs_release_prepared_delayed_node(
300 					struct btrfs_delayed_node *node)
301 {
302 	__btrfs_release_delayed_node(node, 1);
303 }
304 
btrfs_alloc_delayed_item(u16 data_len,struct btrfs_delayed_node * node,enum btrfs_delayed_item_type type)305 static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u16 data_len,
306 					   struct btrfs_delayed_node *node,
307 					   enum btrfs_delayed_item_type type)
308 {
309 	struct btrfs_delayed_item *item;
310 
311 	item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
312 	if (item) {
313 		item->data_len = data_len;
314 		item->type = type;
315 		item->bytes_reserved = 0;
316 		item->delayed_node = node;
317 		RB_CLEAR_NODE(&item->rb_node);
318 		INIT_LIST_HEAD(&item->log_list);
319 		item->logged = false;
320 		refcount_set(&item->refs, 1);
321 	}
322 	return item;
323 }
324 
325 /*
326  * __btrfs_lookup_delayed_item - look up the delayed item by key
327  * @delayed_node: pointer to the delayed node
328  * @index:	  the dir index value to lookup (offset of a dir index key)
329  *
330  * Note: if we don't find the right item, we will return the prev item and
331  * the next item.
332  */
__btrfs_lookup_delayed_item(struct rb_root * root,u64 index)333 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
334 				struct rb_root *root,
335 				u64 index)
336 {
337 	struct rb_node *node = root->rb_node;
338 	struct btrfs_delayed_item *delayed_item = NULL;
339 
340 	while (node) {
341 		delayed_item = rb_entry(node, struct btrfs_delayed_item,
342 					rb_node);
343 		if (delayed_item->index < index)
344 			node = node->rb_right;
345 		else if (delayed_item->index > index)
346 			node = node->rb_left;
347 		else
348 			return delayed_item;
349 	}
350 
351 	return NULL;
352 }
353 
__btrfs_add_delayed_item(struct btrfs_delayed_node * delayed_node,struct btrfs_delayed_item * ins)354 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
355 				    struct btrfs_delayed_item *ins)
356 {
357 	struct rb_node **p, *node;
358 	struct rb_node *parent_node = NULL;
359 	struct rb_root_cached *root;
360 	struct btrfs_delayed_item *item;
361 	bool leftmost = true;
362 
363 	if (ins->type == BTRFS_DELAYED_INSERTION_ITEM)
364 		root = &delayed_node->ins_root;
365 	else
366 		root = &delayed_node->del_root;
367 
368 	p = &root->rb_root.rb_node;
369 	node = &ins->rb_node;
370 
371 	while (*p) {
372 		parent_node = *p;
373 		item = rb_entry(parent_node, struct btrfs_delayed_item,
374 				 rb_node);
375 
376 		if (item->index < ins->index) {
377 			p = &(*p)->rb_right;
378 			leftmost = false;
379 		} else if (item->index > ins->index) {
380 			p = &(*p)->rb_left;
381 		} else {
382 			return -EEXIST;
383 		}
384 	}
385 
386 	rb_link_node(node, parent_node, p);
387 	rb_insert_color_cached(node, root, leftmost);
388 
389 	if (ins->type == BTRFS_DELAYED_INSERTION_ITEM &&
390 	    ins->index >= delayed_node->index_cnt)
391 		delayed_node->index_cnt = ins->index + 1;
392 
393 	delayed_node->count++;
394 	atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
395 	return 0;
396 }
397 
finish_one_item(struct btrfs_delayed_root * delayed_root)398 static void finish_one_item(struct btrfs_delayed_root *delayed_root)
399 {
400 	int seq = atomic_inc_return(&delayed_root->items_seq);
401 
402 	/* atomic_dec_return implies a barrier */
403 	if ((atomic_dec_return(&delayed_root->items) <
404 	    BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0))
405 		cond_wake_up_nomb(&delayed_root->wait);
406 }
407 
__btrfs_remove_delayed_item(struct btrfs_delayed_item * delayed_item)408 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
409 {
410 	struct rb_root_cached *root;
411 	struct btrfs_delayed_root *delayed_root;
412 
413 	/* Not inserted, ignore it. */
414 	if (RB_EMPTY_NODE(&delayed_item->rb_node))
415 		return;
416 
417 	delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
418 
419 	BUG_ON(!delayed_root);
420 
421 	if (delayed_item->type == BTRFS_DELAYED_INSERTION_ITEM)
422 		root = &delayed_item->delayed_node->ins_root;
423 	else
424 		root = &delayed_item->delayed_node->del_root;
425 
426 	rb_erase_cached(&delayed_item->rb_node, root);
427 	RB_CLEAR_NODE(&delayed_item->rb_node);
428 	delayed_item->delayed_node->count--;
429 
430 	finish_one_item(delayed_root);
431 }
432 
btrfs_release_delayed_item(struct btrfs_delayed_item * item)433 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
434 {
435 	if (item) {
436 		__btrfs_remove_delayed_item(item);
437 		if (refcount_dec_and_test(&item->refs))
438 			kfree(item);
439 	}
440 }
441 
__btrfs_first_delayed_insertion_item(struct btrfs_delayed_node * delayed_node)442 static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
443 					struct btrfs_delayed_node *delayed_node)
444 {
445 	struct rb_node *p;
446 	struct btrfs_delayed_item *item = NULL;
447 
448 	p = rb_first_cached(&delayed_node->ins_root);
449 	if (p)
450 		item = rb_entry(p, struct btrfs_delayed_item, rb_node);
451 
452 	return item;
453 }
454 
__btrfs_first_delayed_deletion_item(struct btrfs_delayed_node * delayed_node)455 static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
456 					struct btrfs_delayed_node *delayed_node)
457 {
458 	struct rb_node *p;
459 	struct btrfs_delayed_item *item = NULL;
460 
461 	p = rb_first_cached(&delayed_node->del_root);
462 	if (p)
463 		item = rb_entry(p, struct btrfs_delayed_item, rb_node);
464 
465 	return item;
466 }
467 
__btrfs_next_delayed_item(struct btrfs_delayed_item * item)468 static struct btrfs_delayed_item *__btrfs_next_delayed_item(
469 						struct btrfs_delayed_item *item)
470 {
471 	struct rb_node *p;
472 	struct btrfs_delayed_item *next = NULL;
473 
474 	p = rb_next(&item->rb_node);
475 	if (p)
476 		next = rb_entry(p, struct btrfs_delayed_item, rb_node);
477 
478 	return next;
479 }
480 
btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle * trans,struct btrfs_delayed_item * item)481 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
482 					       struct btrfs_delayed_item *item)
483 {
484 	struct btrfs_block_rsv *src_rsv;
485 	struct btrfs_block_rsv *dst_rsv;
486 	struct btrfs_fs_info *fs_info = trans->fs_info;
487 	u64 num_bytes;
488 	int ret;
489 
490 	if (!trans->bytes_reserved)
491 		return 0;
492 
493 	src_rsv = trans->block_rsv;
494 	dst_rsv = &fs_info->delayed_block_rsv;
495 
496 	num_bytes = btrfs_calc_insert_metadata_size(fs_info, 1);
497 
498 	/*
499 	 * Here we migrate space rsv from transaction rsv, since have already
500 	 * reserved space when starting a transaction.  So no need to reserve
501 	 * qgroup space here.
502 	 */
503 	ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
504 	if (!ret) {
505 		trace_btrfs_space_reservation(fs_info, "delayed_item",
506 					      item->delayed_node->inode_id,
507 					      num_bytes, 1);
508 		/*
509 		 * For insertions we track reserved metadata space by accounting
510 		 * for the number of leaves that will be used, based on the delayed
511 		 * node's index_items_size field.
512 		 */
513 		if (item->type == BTRFS_DELAYED_DELETION_ITEM)
514 			item->bytes_reserved = num_bytes;
515 	}
516 
517 	return ret;
518 }
519 
btrfs_delayed_item_release_metadata(struct btrfs_root * root,struct btrfs_delayed_item * item)520 static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
521 						struct btrfs_delayed_item *item)
522 {
523 	struct btrfs_block_rsv *rsv;
524 	struct btrfs_fs_info *fs_info = root->fs_info;
525 
526 	if (!item->bytes_reserved)
527 		return;
528 
529 	rsv = &fs_info->delayed_block_rsv;
530 	/*
531 	 * Check btrfs_delayed_item_reserve_metadata() to see why we don't need
532 	 * to release/reserve qgroup space.
533 	 */
534 	trace_btrfs_space_reservation(fs_info, "delayed_item",
535 				      item->delayed_node->inode_id,
536 				      item->bytes_reserved, 0);
537 	btrfs_block_rsv_release(fs_info, rsv, item->bytes_reserved, NULL);
538 }
539 
btrfs_delayed_item_release_leaves(struct btrfs_delayed_node * node,unsigned int num_leaves)540 static void btrfs_delayed_item_release_leaves(struct btrfs_delayed_node *node,
541 					      unsigned int num_leaves)
542 {
543 	struct btrfs_fs_info *fs_info = node->root->fs_info;
544 	const u64 bytes = btrfs_calc_insert_metadata_size(fs_info, num_leaves);
545 
546 	/* There are no space reservations during log replay, bail out. */
547 	if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
548 		return;
549 
550 	trace_btrfs_space_reservation(fs_info, "delayed_item", node->inode_id,
551 				      bytes, 0);
552 	btrfs_block_rsv_release(fs_info, &fs_info->delayed_block_rsv, bytes, NULL);
553 }
554 
btrfs_delayed_inode_reserve_metadata(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_delayed_node * node)555 static int btrfs_delayed_inode_reserve_metadata(
556 					struct btrfs_trans_handle *trans,
557 					struct btrfs_root *root,
558 					struct btrfs_delayed_node *node)
559 {
560 	struct btrfs_fs_info *fs_info = root->fs_info;
561 	struct btrfs_block_rsv *src_rsv;
562 	struct btrfs_block_rsv *dst_rsv;
563 	u64 num_bytes;
564 	int ret;
565 
566 	src_rsv = trans->block_rsv;
567 	dst_rsv = &fs_info->delayed_block_rsv;
568 
569 	num_bytes = btrfs_calc_metadata_size(fs_info, 1);
570 
571 	/*
572 	 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
573 	 * which doesn't reserve space for speed.  This is a problem since we
574 	 * still need to reserve space for this update, so try to reserve the
575 	 * space.
576 	 *
577 	 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
578 	 * we always reserve enough to update the inode item.
579 	 */
580 	if (!src_rsv || (!trans->bytes_reserved &&
581 			 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
582 		ret = btrfs_qgroup_reserve_meta(root, num_bytes,
583 					  BTRFS_QGROUP_RSV_META_PREALLOC, true);
584 		if (ret < 0)
585 			return ret;
586 		ret = btrfs_block_rsv_add(fs_info, dst_rsv, num_bytes,
587 					  BTRFS_RESERVE_NO_FLUSH);
588 		/* NO_FLUSH could only fail with -ENOSPC */
589 		ASSERT(ret == 0 || ret == -ENOSPC);
590 		if (ret)
591 			btrfs_qgroup_free_meta_prealloc(root, num_bytes);
592 	} else {
593 		ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
594 	}
595 
596 	if (!ret) {
597 		trace_btrfs_space_reservation(fs_info, "delayed_inode",
598 					      node->inode_id, num_bytes, 1);
599 		node->bytes_reserved = num_bytes;
600 	}
601 
602 	return ret;
603 }
604 
btrfs_delayed_inode_release_metadata(struct btrfs_fs_info * fs_info,struct btrfs_delayed_node * node,bool qgroup_free)605 static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
606 						struct btrfs_delayed_node *node,
607 						bool qgroup_free)
608 {
609 	struct btrfs_block_rsv *rsv;
610 
611 	if (!node->bytes_reserved)
612 		return;
613 
614 	rsv = &fs_info->delayed_block_rsv;
615 	trace_btrfs_space_reservation(fs_info, "delayed_inode",
616 				      node->inode_id, node->bytes_reserved, 0);
617 	btrfs_block_rsv_release(fs_info, rsv, node->bytes_reserved, NULL);
618 	if (qgroup_free)
619 		btrfs_qgroup_free_meta_prealloc(node->root,
620 				node->bytes_reserved);
621 	else
622 		btrfs_qgroup_convert_reserved_meta(node->root,
623 				node->bytes_reserved);
624 	node->bytes_reserved = 0;
625 }
626 
627 /*
628  * Insert a single delayed item or a batch of delayed items, as many as possible
629  * that fit in a leaf. The delayed items (dir index keys) are sorted by their key
630  * in the rbtree, and if there's a gap between two consecutive dir index items,
631  * then it means at some point we had delayed dir indexes to add but they got
632  * removed (by btrfs_delete_delayed_dir_index()) before we attempted to flush them
633  * into the subvolume tree. Dir index keys also have their offsets coming from a
634  * monotonically increasing counter, so we can't get new keys with an offset that
635  * fits within a gap between delayed dir index items.
636  */
btrfs_insert_delayed_item(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct btrfs_delayed_item * first_item)637 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
638 				     struct btrfs_root *root,
639 				     struct btrfs_path *path,
640 				     struct btrfs_delayed_item *first_item)
641 {
642 	struct btrfs_fs_info *fs_info = root->fs_info;
643 	struct btrfs_delayed_node *node = first_item->delayed_node;
644 	LIST_HEAD(item_list);
645 	struct btrfs_delayed_item *curr;
646 	struct btrfs_delayed_item *next;
647 	const int max_size = BTRFS_LEAF_DATA_SIZE(fs_info);
648 	struct btrfs_item_batch batch;
649 	struct btrfs_key first_key;
650 	const u32 first_data_size = first_item->data_len;
651 	int total_size;
652 	char *ins_data = NULL;
653 	int ret;
654 	bool continuous_keys_only = false;
655 
656 	lockdep_assert_held(&node->mutex);
657 
658 	/*
659 	 * During normal operation the delayed index offset is continuously
660 	 * increasing, so we can batch insert all items as there will not be any
661 	 * overlapping keys in the tree.
662 	 *
663 	 * The exception to this is log replay, where we may have interleaved
664 	 * offsets in the tree, so our batch needs to be continuous keys only in
665 	 * order to ensure we do not end up with out of order items in our leaf.
666 	 */
667 	if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
668 		continuous_keys_only = true;
669 
670 	/*
671 	 * For delayed items to insert, we track reserved metadata bytes based
672 	 * on the number of leaves that we will use.
673 	 * See btrfs_insert_delayed_dir_index() and
674 	 * btrfs_delayed_item_reserve_metadata()).
675 	 */
676 	ASSERT(first_item->bytes_reserved == 0);
677 
678 	list_add_tail(&first_item->tree_list, &item_list);
679 	batch.total_data_size = first_data_size;
680 	batch.nr = 1;
681 	total_size = first_data_size + sizeof(struct btrfs_item);
682 	curr = first_item;
683 
684 	while (true) {
685 		int next_size;
686 
687 		next = __btrfs_next_delayed_item(curr);
688 		if (!next)
689 			break;
690 
691 		/*
692 		 * We cannot allow gaps in the key space if we're doing log
693 		 * replay.
694 		 */
695 		if (continuous_keys_only && (next->index != curr->index + 1))
696 			break;
697 
698 		ASSERT(next->bytes_reserved == 0);
699 
700 		next_size = next->data_len + sizeof(struct btrfs_item);
701 		if (total_size + next_size > max_size)
702 			break;
703 
704 		list_add_tail(&next->tree_list, &item_list);
705 		batch.nr++;
706 		total_size += next_size;
707 		batch.total_data_size += next->data_len;
708 		curr = next;
709 	}
710 
711 	if (batch.nr == 1) {
712 		first_key.objectid = node->inode_id;
713 		first_key.type = BTRFS_DIR_INDEX_KEY;
714 		first_key.offset = first_item->index;
715 		batch.keys = &first_key;
716 		batch.data_sizes = &first_data_size;
717 	} else {
718 		struct btrfs_key *ins_keys;
719 		u32 *ins_sizes;
720 		int i = 0;
721 
722 		ins_data = kmalloc(batch.nr * sizeof(u32) +
723 				   batch.nr * sizeof(struct btrfs_key), GFP_NOFS);
724 		if (!ins_data) {
725 			ret = -ENOMEM;
726 			goto out;
727 		}
728 		ins_sizes = (u32 *)ins_data;
729 		ins_keys = (struct btrfs_key *)(ins_data + batch.nr * sizeof(u32));
730 		batch.keys = ins_keys;
731 		batch.data_sizes = ins_sizes;
732 		list_for_each_entry(curr, &item_list, tree_list) {
733 			ins_keys[i].objectid = node->inode_id;
734 			ins_keys[i].type = BTRFS_DIR_INDEX_KEY;
735 			ins_keys[i].offset = curr->index;
736 			ins_sizes[i] = curr->data_len;
737 			i++;
738 		}
739 	}
740 
741 	ret = btrfs_insert_empty_items(trans, root, path, &batch);
742 	if (ret)
743 		goto out;
744 
745 	list_for_each_entry(curr, &item_list, tree_list) {
746 		char *data_ptr;
747 
748 		data_ptr = btrfs_item_ptr(path->nodes[0], path->slots[0], char);
749 		write_extent_buffer(path->nodes[0], &curr->data,
750 				    (unsigned long)data_ptr, curr->data_len);
751 		path->slots[0]++;
752 	}
753 
754 	/*
755 	 * Now release our path before releasing the delayed items and their
756 	 * metadata reservations, so that we don't block other tasks for more
757 	 * time than needed.
758 	 */
759 	btrfs_release_path(path);
760 
761 	ASSERT(node->index_item_leaves > 0);
762 
763 	/*
764 	 * For normal operations we will batch an entire leaf's worth of delayed
765 	 * items, so if there are more items to process we can decrement
766 	 * index_item_leaves by 1 as we inserted 1 leaf's worth of items.
767 	 *
768 	 * However for log replay we may not have inserted an entire leaf's
769 	 * worth of items, we may have not had continuous items, so decrementing
770 	 * here would mess up the index_item_leaves accounting.  For this case
771 	 * only clean up the accounting when there are no items left.
772 	 */
773 	if (next && !continuous_keys_only) {
774 		/*
775 		 * We inserted one batch of items into a leaf a there are more
776 		 * items to flush in a future batch, now release one unit of
777 		 * metadata space from the delayed block reserve, corresponding
778 		 * the leaf we just flushed to.
779 		 */
780 		btrfs_delayed_item_release_leaves(node, 1);
781 		node->index_item_leaves--;
782 	} else if (!next) {
783 		/*
784 		 * There are no more items to insert. We can have a number of
785 		 * reserved leaves > 1 here - this happens when many dir index
786 		 * items are added and then removed before they are flushed (file
787 		 * names with a very short life, never span a transaction). So
788 		 * release all remaining leaves.
789 		 */
790 		btrfs_delayed_item_release_leaves(node, node->index_item_leaves);
791 		node->index_item_leaves = 0;
792 	}
793 
794 	list_for_each_entry_safe(curr, next, &item_list, tree_list) {
795 		list_del(&curr->tree_list);
796 		btrfs_release_delayed_item(curr);
797 	}
798 out:
799 	kfree(ins_data);
800 	return ret;
801 }
802 
btrfs_insert_delayed_items(struct btrfs_trans_handle * trans,struct btrfs_path * path,struct btrfs_root * root,struct btrfs_delayed_node * node)803 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
804 				      struct btrfs_path *path,
805 				      struct btrfs_root *root,
806 				      struct btrfs_delayed_node *node)
807 {
808 	int ret = 0;
809 
810 	while (ret == 0) {
811 		struct btrfs_delayed_item *curr;
812 
813 		mutex_lock(&node->mutex);
814 		curr = __btrfs_first_delayed_insertion_item(node);
815 		if (!curr) {
816 			mutex_unlock(&node->mutex);
817 			break;
818 		}
819 		ret = btrfs_insert_delayed_item(trans, root, path, curr);
820 		mutex_unlock(&node->mutex);
821 	}
822 
823 	return ret;
824 }
825 
btrfs_batch_delete_items(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct btrfs_delayed_item * item)826 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
827 				    struct btrfs_root *root,
828 				    struct btrfs_path *path,
829 				    struct btrfs_delayed_item *item)
830 {
831 	const u64 ino = item->delayed_node->inode_id;
832 	struct btrfs_fs_info *fs_info = root->fs_info;
833 	struct btrfs_delayed_item *curr, *next;
834 	struct extent_buffer *leaf = path->nodes[0];
835 	LIST_HEAD(batch_list);
836 	int nitems, slot, last_slot;
837 	int ret;
838 	u64 total_reserved_size = item->bytes_reserved;
839 
840 	ASSERT(leaf != NULL);
841 
842 	slot = path->slots[0];
843 	last_slot = btrfs_header_nritems(leaf) - 1;
844 	/*
845 	 * Our caller always gives us a path pointing to an existing item, so
846 	 * this can not happen.
847 	 */
848 	ASSERT(slot <= last_slot);
849 	if (WARN_ON(slot > last_slot))
850 		return -ENOENT;
851 
852 	nitems = 1;
853 	curr = item;
854 	list_add_tail(&curr->tree_list, &batch_list);
855 
856 	/*
857 	 * Keep checking if the next delayed item matches the next item in the
858 	 * leaf - if so, we can add it to the batch of items to delete from the
859 	 * leaf.
860 	 */
861 	while (slot < last_slot) {
862 		struct btrfs_key key;
863 
864 		next = __btrfs_next_delayed_item(curr);
865 		if (!next)
866 			break;
867 
868 		slot++;
869 		btrfs_item_key_to_cpu(leaf, &key, slot);
870 		if (key.objectid != ino ||
871 		    key.type != BTRFS_DIR_INDEX_KEY ||
872 		    key.offset != next->index)
873 			break;
874 		nitems++;
875 		curr = next;
876 		list_add_tail(&curr->tree_list, &batch_list);
877 		total_reserved_size += curr->bytes_reserved;
878 	}
879 
880 	ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
881 	if (ret)
882 		return ret;
883 
884 	/* In case of BTRFS_FS_LOG_RECOVERING items won't have reserved space */
885 	if (total_reserved_size > 0) {
886 		/*
887 		 * Check btrfs_delayed_item_reserve_metadata() to see why we
888 		 * don't need to release/reserve qgroup space.
889 		 */
890 		trace_btrfs_space_reservation(fs_info, "delayed_item", ino,
891 					      total_reserved_size, 0);
892 		btrfs_block_rsv_release(fs_info, &fs_info->delayed_block_rsv,
893 					total_reserved_size, NULL);
894 	}
895 
896 	list_for_each_entry_safe(curr, next, &batch_list, tree_list) {
897 		list_del(&curr->tree_list);
898 		btrfs_release_delayed_item(curr);
899 	}
900 
901 	return 0;
902 }
903 
btrfs_delete_delayed_items(struct btrfs_trans_handle * trans,struct btrfs_path * path,struct btrfs_root * root,struct btrfs_delayed_node * node)904 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
905 				      struct btrfs_path *path,
906 				      struct btrfs_root *root,
907 				      struct btrfs_delayed_node *node)
908 {
909 	struct btrfs_key key;
910 	int ret = 0;
911 
912 	key.objectid = node->inode_id;
913 	key.type = BTRFS_DIR_INDEX_KEY;
914 
915 	while (ret == 0) {
916 		struct btrfs_delayed_item *item;
917 
918 		mutex_lock(&node->mutex);
919 		item = __btrfs_first_delayed_deletion_item(node);
920 		if (!item) {
921 			mutex_unlock(&node->mutex);
922 			break;
923 		}
924 
925 		key.offset = item->index;
926 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
927 		if (ret > 0) {
928 			/*
929 			 * There's no matching item in the leaf. This means we
930 			 * have already deleted this item in a past run of the
931 			 * delayed items. We ignore errors when running delayed
932 			 * items from an async context, through a work queue job
933 			 * running btrfs_async_run_delayed_root(), and don't
934 			 * release delayed items that failed to complete. This
935 			 * is because we will retry later, and at transaction
936 			 * commit time we always run delayed items and will
937 			 * then deal with errors if they fail to run again.
938 			 *
939 			 * So just release delayed items for which we can't find
940 			 * an item in the tree, and move to the next item.
941 			 */
942 			btrfs_release_path(path);
943 			btrfs_release_delayed_item(item);
944 			ret = 0;
945 		} else if (ret == 0) {
946 			ret = btrfs_batch_delete_items(trans, root, path, item);
947 			btrfs_release_path(path);
948 		}
949 
950 		/*
951 		 * We unlock and relock on each iteration, this is to prevent
952 		 * blocking other tasks for too long while we are being run from
953 		 * the async context (work queue job). Those tasks are typically
954 		 * running system calls like creat/mkdir/rename/unlink/etc which
955 		 * need to add delayed items to this delayed node.
956 		 */
957 		mutex_unlock(&node->mutex);
958 	}
959 
960 	return ret;
961 }
962 
btrfs_release_delayed_inode(struct btrfs_delayed_node * delayed_node)963 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
964 {
965 	struct btrfs_delayed_root *delayed_root;
966 
967 	if (delayed_node &&
968 	    test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
969 		BUG_ON(!delayed_node->root);
970 		clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
971 		delayed_node->count--;
972 
973 		delayed_root = delayed_node->root->fs_info->delayed_root;
974 		finish_one_item(delayed_root);
975 	}
976 }
977 
btrfs_release_delayed_iref(struct btrfs_delayed_node * delayed_node)978 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
979 {
980 
981 	if (test_and_clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags)) {
982 		struct btrfs_delayed_root *delayed_root;
983 
984 		ASSERT(delayed_node->root);
985 		delayed_node->count--;
986 
987 		delayed_root = delayed_node->root->fs_info->delayed_root;
988 		finish_one_item(delayed_root);
989 	}
990 }
991 
__btrfs_update_delayed_inode(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct btrfs_delayed_node * node)992 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
993 					struct btrfs_root *root,
994 					struct btrfs_path *path,
995 					struct btrfs_delayed_node *node)
996 {
997 	struct btrfs_fs_info *fs_info = root->fs_info;
998 	struct btrfs_key key;
999 	struct btrfs_inode_item *inode_item;
1000 	struct extent_buffer *leaf;
1001 	int mod;
1002 	int ret;
1003 
1004 	key.objectid = node->inode_id;
1005 	key.type = BTRFS_INODE_ITEM_KEY;
1006 	key.offset = 0;
1007 
1008 	if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1009 		mod = -1;
1010 	else
1011 		mod = 1;
1012 
1013 	ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1014 	if (ret > 0)
1015 		ret = -ENOENT;
1016 	if (ret < 0)
1017 		goto out;
1018 
1019 	leaf = path->nodes[0];
1020 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
1021 				    struct btrfs_inode_item);
1022 	write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1023 			    sizeof(struct btrfs_inode_item));
1024 	btrfs_mark_buffer_dirty(leaf);
1025 
1026 	if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1027 		goto out;
1028 
1029 	path->slots[0]++;
1030 	if (path->slots[0] >= btrfs_header_nritems(leaf))
1031 		goto search;
1032 again:
1033 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1034 	if (key.objectid != node->inode_id)
1035 		goto out;
1036 
1037 	if (key.type != BTRFS_INODE_REF_KEY &&
1038 	    key.type != BTRFS_INODE_EXTREF_KEY)
1039 		goto out;
1040 
1041 	/*
1042 	 * Delayed iref deletion is for the inode who has only one link,
1043 	 * so there is only one iref. The case that several irefs are
1044 	 * in the same item doesn't exist.
1045 	 */
1046 	btrfs_del_item(trans, root, path);
1047 out:
1048 	btrfs_release_delayed_iref(node);
1049 	btrfs_release_path(path);
1050 err_out:
1051 	btrfs_delayed_inode_release_metadata(fs_info, node, (ret < 0));
1052 	btrfs_release_delayed_inode(node);
1053 
1054 	/*
1055 	 * If we fail to update the delayed inode we need to abort the
1056 	 * transaction, because we could leave the inode with the improper
1057 	 * counts behind.
1058 	 */
1059 	if (ret && ret != -ENOENT)
1060 		btrfs_abort_transaction(trans, ret);
1061 
1062 	return ret;
1063 
1064 search:
1065 	btrfs_release_path(path);
1066 
1067 	key.type = BTRFS_INODE_EXTREF_KEY;
1068 	key.offset = -1;
1069 
1070 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1071 	if (ret < 0)
1072 		goto err_out;
1073 	ASSERT(ret);
1074 
1075 	ret = 0;
1076 	leaf = path->nodes[0];
1077 	path->slots[0]--;
1078 	goto again;
1079 }
1080 
btrfs_update_delayed_inode(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct btrfs_delayed_node * node)1081 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1082 					     struct btrfs_root *root,
1083 					     struct btrfs_path *path,
1084 					     struct btrfs_delayed_node *node)
1085 {
1086 	int ret;
1087 
1088 	mutex_lock(&node->mutex);
1089 	if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1090 		mutex_unlock(&node->mutex);
1091 		return 0;
1092 	}
1093 
1094 	ret = __btrfs_update_delayed_inode(trans, root, path, node);
1095 	mutex_unlock(&node->mutex);
1096 	return ret;
1097 }
1098 
1099 static inline int
__btrfs_commit_inode_delayed_items(struct btrfs_trans_handle * trans,struct btrfs_path * path,struct btrfs_delayed_node * node)1100 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1101 				   struct btrfs_path *path,
1102 				   struct btrfs_delayed_node *node)
1103 {
1104 	int ret;
1105 
1106 	ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1107 	if (ret)
1108 		return ret;
1109 
1110 	ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1111 	if (ret)
1112 		return ret;
1113 
1114 	ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1115 	return ret;
1116 }
1117 
1118 /*
1119  * Called when committing the transaction.
1120  * Returns 0 on success.
1121  * Returns < 0 on error and returns with an aborted transaction with any
1122  * outstanding delayed items cleaned up.
1123  */
__btrfs_run_delayed_items(struct btrfs_trans_handle * trans,int nr)1124 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, int nr)
1125 {
1126 	struct btrfs_fs_info *fs_info = trans->fs_info;
1127 	struct btrfs_delayed_root *delayed_root;
1128 	struct btrfs_delayed_node *curr_node, *prev_node;
1129 	struct btrfs_path *path;
1130 	struct btrfs_block_rsv *block_rsv;
1131 	int ret = 0;
1132 	bool count = (nr > 0);
1133 
1134 	if (TRANS_ABORTED(trans))
1135 		return -EIO;
1136 
1137 	path = btrfs_alloc_path();
1138 	if (!path)
1139 		return -ENOMEM;
1140 
1141 	block_rsv = trans->block_rsv;
1142 	trans->block_rsv = &fs_info->delayed_block_rsv;
1143 
1144 	delayed_root = fs_info->delayed_root;
1145 
1146 	curr_node = btrfs_first_delayed_node(delayed_root);
1147 	while (curr_node && (!count || nr--)) {
1148 		ret = __btrfs_commit_inode_delayed_items(trans, path,
1149 							 curr_node);
1150 		if (ret) {
1151 			btrfs_release_delayed_node(curr_node);
1152 			curr_node = NULL;
1153 			btrfs_abort_transaction(trans, ret);
1154 			break;
1155 		}
1156 
1157 		prev_node = curr_node;
1158 		curr_node = btrfs_next_delayed_node(curr_node);
1159 		btrfs_release_delayed_node(prev_node);
1160 	}
1161 
1162 	if (curr_node)
1163 		btrfs_release_delayed_node(curr_node);
1164 	btrfs_free_path(path);
1165 	trans->block_rsv = block_rsv;
1166 
1167 	return ret;
1168 }
1169 
btrfs_run_delayed_items(struct btrfs_trans_handle * trans)1170 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans)
1171 {
1172 	return __btrfs_run_delayed_items(trans, -1);
1173 }
1174 
btrfs_run_delayed_items_nr(struct btrfs_trans_handle * trans,int nr)1175 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, int nr)
1176 {
1177 	return __btrfs_run_delayed_items(trans, nr);
1178 }
1179 
btrfs_commit_inode_delayed_items(struct btrfs_trans_handle * trans,struct btrfs_inode * inode)1180 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1181 				     struct btrfs_inode *inode)
1182 {
1183 	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1184 	struct btrfs_path *path;
1185 	struct btrfs_block_rsv *block_rsv;
1186 	int ret;
1187 
1188 	if (!delayed_node)
1189 		return 0;
1190 
1191 	mutex_lock(&delayed_node->mutex);
1192 	if (!delayed_node->count) {
1193 		mutex_unlock(&delayed_node->mutex);
1194 		btrfs_release_delayed_node(delayed_node);
1195 		return 0;
1196 	}
1197 	mutex_unlock(&delayed_node->mutex);
1198 
1199 	path = btrfs_alloc_path();
1200 	if (!path) {
1201 		btrfs_release_delayed_node(delayed_node);
1202 		return -ENOMEM;
1203 	}
1204 
1205 	block_rsv = trans->block_rsv;
1206 	trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1207 
1208 	ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1209 
1210 	btrfs_release_delayed_node(delayed_node);
1211 	btrfs_free_path(path);
1212 	trans->block_rsv = block_rsv;
1213 
1214 	return ret;
1215 }
1216 
btrfs_commit_inode_delayed_inode(struct btrfs_inode * inode)1217 int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode)
1218 {
1219 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1220 	struct btrfs_trans_handle *trans;
1221 	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1222 	struct btrfs_path *path;
1223 	struct btrfs_block_rsv *block_rsv;
1224 	int ret;
1225 
1226 	if (!delayed_node)
1227 		return 0;
1228 
1229 	mutex_lock(&delayed_node->mutex);
1230 	if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1231 		mutex_unlock(&delayed_node->mutex);
1232 		btrfs_release_delayed_node(delayed_node);
1233 		return 0;
1234 	}
1235 	mutex_unlock(&delayed_node->mutex);
1236 
1237 	trans = btrfs_join_transaction(delayed_node->root);
1238 	if (IS_ERR(trans)) {
1239 		ret = PTR_ERR(trans);
1240 		goto out;
1241 	}
1242 
1243 	path = btrfs_alloc_path();
1244 	if (!path) {
1245 		ret = -ENOMEM;
1246 		goto trans_out;
1247 	}
1248 
1249 	block_rsv = trans->block_rsv;
1250 	trans->block_rsv = &fs_info->delayed_block_rsv;
1251 
1252 	mutex_lock(&delayed_node->mutex);
1253 	if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1254 		ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1255 						   path, delayed_node);
1256 	else
1257 		ret = 0;
1258 	mutex_unlock(&delayed_node->mutex);
1259 
1260 	btrfs_free_path(path);
1261 	trans->block_rsv = block_rsv;
1262 trans_out:
1263 	btrfs_end_transaction(trans);
1264 	btrfs_btree_balance_dirty(fs_info);
1265 out:
1266 	btrfs_release_delayed_node(delayed_node);
1267 
1268 	return ret;
1269 }
1270 
btrfs_remove_delayed_node(struct btrfs_inode * inode)1271 void btrfs_remove_delayed_node(struct btrfs_inode *inode)
1272 {
1273 	struct btrfs_delayed_node *delayed_node;
1274 
1275 	delayed_node = READ_ONCE(inode->delayed_node);
1276 	if (!delayed_node)
1277 		return;
1278 
1279 	inode->delayed_node = NULL;
1280 	btrfs_release_delayed_node(delayed_node);
1281 }
1282 
1283 struct btrfs_async_delayed_work {
1284 	struct btrfs_delayed_root *delayed_root;
1285 	int nr;
1286 	struct btrfs_work work;
1287 };
1288 
btrfs_async_run_delayed_root(struct btrfs_work * work)1289 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1290 {
1291 	struct btrfs_async_delayed_work *async_work;
1292 	struct btrfs_delayed_root *delayed_root;
1293 	struct btrfs_trans_handle *trans;
1294 	struct btrfs_path *path;
1295 	struct btrfs_delayed_node *delayed_node = NULL;
1296 	struct btrfs_root *root;
1297 	struct btrfs_block_rsv *block_rsv;
1298 	int total_done = 0;
1299 
1300 	async_work = container_of(work, struct btrfs_async_delayed_work, work);
1301 	delayed_root = async_work->delayed_root;
1302 
1303 	path = btrfs_alloc_path();
1304 	if (!path)
1305 		goto out;
1306 
1307 	do {
1308 		if (atomic_read(&delayed_root->items) <
1309 		    BTRFS_DELAYED_BACKGROUND / 2)
1310 			break;
1311 
1312 		delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1313 		if (!delayed_node)
1314 			break;
1315 
1316 		root = delayed_node->root;
1317 
1318 		trans = btrfs_join_transaction(root);
1319 		if (IS_ERR(trans)) {
1320 			btrfs_release_path(path);
1321 			btrfs_release_prepared_delayed_node(delayed_node);
1322 			total_done++;
1323 			continue;
1324 		}
1325 
1326 		block_rsv = trans->block_rsv;
1327 		trans->block_rsv = &root->fs_info->delayed_block_rsv;
1328 
1329 		__btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1330 
1331 		trans->block_rsv = block_rsv;
1332 		btrfs_end_transaction(trans);
1333 		btrfs_btree_balance_dirty_nodelay(root->fs_info);
1334 
1335 		btrfs_release_path(path);
1336 		btrfs_release_prepared_delayed_node(delayed_node);
1337 		total_done++;
1338 
1339 	} while ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK)
1340 		 || total_done < async_work->nr);
1341 
1342 	btrfs_free_path(path);
1343 out:
1344 	wake_up(&delayed_root->wait);
1345 	kfree(async_work);
1346 }
1347 
1348 
btrfs_wq_run_delayed_node(struct btrfs_delayed_root * delayed_root,struct btrfs_fs_info * fs_info,int nr)1349 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1350 				     struct btrfs_fs_info *fs_info, int nr)
1351 {
1352 	struct btrfs_async_delayed_work *async_work;
1353 
1354 	async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1355 	if (!async_work)
1356 		return -ENOMEM;
1357 
1358 	async_work->delayed_root = delayed_root;
1359 	btrfs_init_work(&async_work->work, btrfs_async_run_delayed_root, NULL,
1360 			NULL);
1361 	async_work->nr = nr;
1362 
1363 	btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1364 	return 0;
1365 }
1366 
btrfs_assert_delayed_root_empty(struct btrfs_fs_info * fs_info)1367 void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1368 {
1369 	WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
1370 }
1371 
could_end_wait(struct btrfs_delayed_root * delayed_root,int seq)1372 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1373 {
1374 	int val = atomic_read(&delayed_root->items_seq);
1375 
1376 	if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1377 		return 1;
1378 
1379 	if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1380 		return 1;
1381 
1382 	return 0;
1383 }
1384 
btrfs_balance_delayed_items(struct btrfs_fs_info * fs_info)1385 void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1386 {
1387 	struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1388 
1389 	if ((atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) ||
1390 		btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1391 		return;
1392 
1393 	if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1394 		int seq;
1395 		int ret;
1396 
1397 		seq = atomic_read(&delayed_root->items_seq);
1398 
1399 		ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1400 		if (ret)
1401 			return;
1402 
1403 		wait_event_interruptible(delayed_root->wait,
1404 					 could_end_wait(delayed_root, seq));
1405 		return;
1406 	}
1407 
1408 	btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1409 }
1410 
1411 /* Will return 0 or -ENOMEM */
btrfs_insert_delayed_dir_index(struct btrfs_trans_handle * trans,const char * name,int name_len,struct btrfs_inode * dir,struct btrfs_disk_key * disk_key,u8 type,u64 index)1412 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1413 				   const char *name, int name_len,
1414 				   struct btrfs_inode *dir,
1415 				   struct btrfs_disk_key *disk_key, u8 type,
1416 				   u64 index)
1417 {
1418 	struct btrfs_fs_info *fs_info = trans->fs_info;
1419 	const unsigned int leaf_data_size = BTRFS_LEAF_DATA_SIZE(fs_info);
1420 	struct btrfs_delayed_node *delayed_node;
1421 	struct btrfs_delayed_item *delayed_item;
1422 	struct btrfs_dir_item *dir_item;
1423 	bool reserve_leaf_space;
1424 	u32 data_len;
1425 	int ret;
1426 
1427 	delayed_node = btrfs_get_or_create_delayed_node(dir);
1428 	if (IS_ERR(delayed_node))
1429 		return PTR_ERR(delayed_node);
1430 
1431 	delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len,
1432 						delayed_node,
1433 						BTRFS_DELAYED_INSERTION_ITEM);
1434 	if (!delayed_item) {
1435 		ret = -ENOMEM;
1436 		goto release_node;
1437 	}
1438 
1439 	delayed_item->index = index;
1440 
1441 	dir_item = (struct btrfs_dir_item *)delayed_item->data;
1442 	dir_item->location = *disk_key;
1443 	btrfs_set_stack_dir_transid(dir_item, trans->transid);
1444 	btrfs_set_stack_dir_data_len(dir_item, 0);
1445 	btrfs_set_stack_dir_name_len(dir_item, name_len);
1446 	btrfs_set_stack_dir_type(dir_item, type);
1447 	memcpy((char *)(dir_item + 1), name, name_len);
1448 
1449 	data_len = delayed_item->data_len + sizeof(struct btrfs_item);
1450 
1451 	mutex_lock(&delayed_node->mutex);
1452 
1453 	if (delayed_node->index_item_leaves == 0 ||
1454 	    delayed_node->curr_index_batch_size + data_len > leaf_data_size) {
1455 		delayed_node->curr_index_batch_size = data_len;
1456 		reserve_leaf_space = true;
1457 	} else {
1458 		delayed_node->curr_index_batch_size += data_len;
1459 		reserve_leaf_space = false;
1460 	}
1461 
1462 	if (reserve_leaf_space) {
1463 		ret = btrfs_delayed_item_reserve_metadata(trans, delayed_item);
1464 		/*
1465 		 * Space was reserved for a dir index item insertion when we
1466 		 * started the transaction, so getting a failure here should be
1467 		 * impossible.
1468 		 */
1469 		if (WARN_ON(ret)) {
1470 			mutex_unlock(&delayed_node->mutex);
1471 			btrfs_release_delayed_item(delayed_item);
1472 			goto release_node;
1473 		}
1474 
1475 		delayed_node->index_item_leaves++;
1476 	} else if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
1477 		const u64 bytes = btrfs_calc_insert_metadata_size(fs_info, 1);
1478 
1479 		/*
1480 		 * Adding the new dir index item does not require touching another
1481 		 * leaf, so we can release 1 unit of metadata that was previously
1482 		 * reserved when starting the transaction. This applies only to
1483 		 * the case where we had a transaction start and excludes the
1484 		 * transaction join case (when replaying log trees).
1485 		 */
1486 		trace_btrfs_space_reservation(fs_info, "transaction",
1487 					      trans->transid, bytes, 0);
1488 		btrfs_block_rsv_release(fs_info, trans->block_rsv, bytes, NULL);
1489 		ASSERT(trans->bytes_reserved >= bytes);
1490 		trans->bytes_reserved -= bytes;
1491 	}
1492 
1493 	ret = __btrfs_add_delayed_item(delayed_node, delayed_item);
1494 	if (unlikely(ret)) {
1495 		btrfs_err(trans->fs_info,
1496 			  "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1497 			  name_len, name, delayed_node->root->root_key.objectid,
1498 			  delayed_node->inode_id, ret);
1499 		BUG();
1500 	}
1501 	mutex_unlock(&delayed_node->mutex);
1502 
1503 release_node:
1504 	btrfs_release_delayed_node(delayed_node);
1505 	return ret;
1506 }
1507 
btrfs_delete_delayed_insertion_item(struct btrfs_fs_info * fs_info,struct btrfs_delayed_node * node,u64 index)1508 static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
1509 					       struct btrfs_delayed_node *node,
1510 					       u64 index)
1511 {
1512 	struct btrfs_delayed_item *item;
1513 
1514 	mutex_lock(&node->mutex);
1515 	item = __btrfs_lookup_delayed_item(&node->ins_root.rb_root, index);
1516 	if (!item) {
1517 		mutex_unlock(&node->mutex);
1518 		return 1;
1519 	}
1520 
1521 	/*
1522 	 * For delayed items to insert, we track reserved metadata bytes based
1523 	 * on the number of leaves that we will use.
1524 	 * See btrfs_insert_delayed_dir_index() and
1525 	 * btrfs_delayed_item_reserve_metadata()).
1526 	 */
1527 	ASSERT(item->bytes_reserved == 0);
1528 	ASSERT(node->index_item_leaves > 0);
1529 
1530 	/*
1531 	 * If there's only one leaf reserved, we can decrement this item from the
1532 	 * current batch, otherwise we can not because we don't know which leaf
1533 	 * it belongs to. With the current limit on delayed items, we rarely
1534 	 * accumulate enough dir index items to fill more than one leaf (even
1535 	 * when using a leaf size of 4K).
1536 	 */
1537 	if (node->index_item_leaves == 1) {
1538 		const u32 data_len = item->data_len + sizeof(struct btrfs_item);
1539 
1540 		ASSERT(node->curr_index_batch_size >= data_len);
1541 		node->curr_index_batch_size -= data_len;
1542 	}
1543 
1544 	btrfs_release_delayed_item(item);
1545 
1546 	/* If we now have no more dir index items, we can release all leaves. */
1547 	if (RB_EMPTY_ROOT(&node->ins_root.rb_root)) {
1548 		btrfs_delayed_item_release_leaves(node, node->index_item_leaves);
1549 		node->index_item_leaves = 0;
1550 	}
1551 
1552 	mutex_unlock(&node->mutex);
1553 	return 0;
1554 }
1555 
btrfs_delete_delayed_dir_index(struct btrfs_trans_handle * trans,struct btrfs_inode * dir,u64 index)1556 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1557 				   struct btrfs_inode *dir, u64 index)
1558 {
1559 	struct btrfs_delayed_node *node;
1560 	struct btrfs_delayed_item *item;
1561 	int ret;
1562 
1563 	node = btrfs_get_or_create_delayed_node(dir);
1564 	if (IS_ERR(node))
1565 		return PTR_ERR(node);
1566 
1567 	ret = btrfs_delete_delayed_insertion_item(trans->fs_info, node, index);
1568 	if (!ret)
1569 		goto end;
1570 
1571 	item = btrfs_alloc_delayed_item(0, node, BTRFS_DELAYED_DELETION_ITEM);
1572 	if (!item) {
1573 		ret = -ENOMEM;
1574 		goto end;
1575 	}
1576 
1577 	item->index = index;
1578 
1579 	ret = btrfs_delayed_item_reserve_metadata(trans, item);
1580 	/*
1581 	 * we have reserved enough space when we start a new transaction,
1582 	 * so reserving metadata failure is impossible.
1583 	 */
1584 	if (ret < 0) {
1585 		btrfs_err(trans->fs_info,
1586 "metadata reservation failed for delayed dir item deltiona, should have been reserved");
1587 		btrfs_release_delayed_item(item);
1588 		goto end;
1589 	}
1590 
1591 	mutex_lock(&node->mutex);
1592 	ret = __btrfs_add_delayed_item(node, item);
1593 	if (unlikely(ret)) {
1594 		btrfs_err(trans->fs_info,
1595 			  "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1596 			  index, node->root->root_key.objectid,
1597 			  node->inode_id, ret);
1598 		btrfs_delayed_item_release_metadata(dir->root, item);
1599 		btrfs_release_delayed_item(item);
1600 	}
1601 	mutex_unlock(&node->mutex);
1602 end:
1603 	btrfs_release_delayed_node(node);
1604 	return ret;
1605 }
1606 
btrfs_inode_delayed_dir_index_count(struct btrfs_inode * inode)1607 int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode)
1608 {
1609 	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1610 
1611 	if (!delayed_node)
1612 		return -ENOENT;
1613 
1614 	/*
1615 	 * Since we have held i_mutex of this directory, it is impossible that
1616 	 * a new directory index is added into the delayed node and index_cnt
1617 	 * is updated now. So we needn't lock the delayed node.
1618 	 */
1619 	if (!delayed_node->index_cnt) {
1620 		btrfs_release_delayed_node(delayed_node);
1621 		return -EINVAL;
1622 	}
1623 
1624 	inode->index_cnt = delayed_node->index_cnt;
1625 	btrfs_release_delayed_node(delayed_node);
1626 	return 0;
1627 }
1628 
btrfs_readdir_get_delayed_items(struct inode * inode,struct list_head * ins_list,struct list_head * del_list)1629 bool btrfs_readdir_get_delayed_items(struct inode *inode,
1630 				     struct list_head *ins_list,
1631 				     struct list_head *del_list)
1632 {
1633 	struct btrfs_delayed_node *delayed_node;
1634 	struct btrfs_delayed_item *item;
1635 
1636 	delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1637 	if (!delayed_node)
1638 		return false;
1639 
1640 	/*
1641 	 * We can only do one readdir with delayed items at a time because of
1642 	 * item->readdir_list.
1643 	 */
1644 	btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
1645 	btrfs_inode_lock(inode, 0);
1646 
1647 	mutex_lock(&delayed_node->mutex);
1648 	item = __btrfs_first_delayed_insertion_item(delayed_node);
1649 	while (item) {
1650 		refcount_inc(&item->refs);
1651 		list_add_tail(&item->readdir_list, ins_list);
1652 		item = __btrfs_next_delayed_item(item);
1653 	}
1654 
1655 	item = __btrfs_first_delayed_deletion_item(delayed_node);
1656 	while (item) {
1657 		refcount_inc(&item->refs);
1658 		list_add_tail(&item->readdir_list, del_list);
1659 		item = __btrfs_next_delayed_item(item);
1660 	}
1661 	mutex_unlock(&delayed_node->mutex);
1662 	/*
1663 	 * This delayed node is still cached in the btrfs inode, so refs
1664 	 * must be > 1 now, and we needn't check it is going to be freed
1665 	 * or not.
1666 	 *
1667 	 * Besides that, this function is used to read dir, we do not
1668 	 * insert/delete delayed items in this period. So we also needn't
1669 	 * requeue or dequeue this delayed node.
1670 	 */
1671 	refcount_dec(&delayed_node->refs);
1672 
1673 	return true;
1674 }
1675 
btrfs_readdir_put_delayed_items(struct inode * inode,struct list_head * ins_list,struct list_head * del_list)1676 void btrfs_readdir_put_delayed_items(struct inode *inode,
1677 				     struct list_head *ins_list,
1678 				     struct list_head *del_list)
1679 {
1680 	struct btrfs_delayed_item *curr, *next;
1681 
1682 	list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1683 		list_del(&curr->readdir_list);
1684 		if (refcount_dec_and_test(&curr->refs))
1685 			kfree(curr);
1686 	}
1687 
1688 	list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1689 		list_del(&curr->readdir_list);
1690 		if (refcount_dec_and_test(&curr->refs))
1691 			kfree(curr);
1692 	}
1693 
1694 	/*
1695 	 * The VFS is going to do up_read(), so we need to downgrade back to a
1696 	 * read lock.
1697 	 */
1698 	downgrade_write(&inode->i_rwsem);
1699 }
1700 
btrfs_should_delete_dir_index(struct list_head * del_list,u64 index)1701 int btrfs_should_delete_dir_index(struct list_head *del_list,
1702 				  u64 index)
1703 {
1704 	struct btrfs_delayed_item *curr;
1705 	int ret = 0;
1706 
1707 	list_for_each_entry(curr, del_list, readdir_list) {
1708 		if (curr->index > index)
1709 			break;
1710 		if (curr->index == index) {
1711 			ret = 1;
1712 			break;
1713 		}
1714 	}
1715 	return ret;
1716 }
1717 
1718 /*
1719  * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1720  *
1721  */
btrfs_readdir_delayed_dir_index(struct dir_context * ctx,struct list_head * ins_list)1722 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1723 				    struct list_head *ins_list)
1724 {
1725 	struct btrfs_dir_item *di;
1726 	struct btrfs_delayed_item *curr, *next;
1727 	struct btrfs_key location;
1728 	char *name;
1729 	int name_len;
1730 	int over = 0;
1731 	unsigned char d_type;
1732 
1733 	if (list_empty(ins_list))
1734 		return 0;
1735 
1736 	/*
1737 	 * Changing the data of the delayed item is impossible. So
1738 	 * we needn't lock them. And we have held i_mutex of the
1739 	 * directory, nobody can delete any directory indexes now.
1740 	 */
1741 	list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1742 		list_del(&curr->readdir_list);
1743 
1744 		if (curr->index < ctx->pos) {
1745 			if (refcount_dec_and_test(&curr->refs))
1746 				kfree(curr);
1747 			continue;
1748 		}
1749 
1750 		ctx->pos = curr->index;
1751 
1752 		di = (struct btrfs_dir_item *)curr->data;
1753 		name = (char *)(di + 1);
1754 		name_len = btrfs_stack_dir_name_len(di);
1755 
1756 		d_type = fs_ftype_to_dtype(di->type);
1757 		btrfs_disk_key_to_cpu(&location, &di->location);
1758 
1759 		over = !dir_emit(ctx, name, name_len,
1760 			       location.objectid, d_type);
1761 
1762 		if (refcount_dec_and_test(&curr->refs))
1763 			kfree(curr);
1764 
1765 		if (over)
1766 			return 1;
1767 		ctx->pos++;
1768 	}
1769 	return 0;
1770 }
1771 
fill_stack_inode_item(struct btrfs_trans_handle * trans,struct btrfs_inode_item * inode_item,struct inode * inode)1772 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1773 				  struct btrfs_inode_item *inode_item,
1774 				  struct inode *inode)
1775 {
1776 	u64 flags;
1777 
1778 	btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1779 	btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1780 	btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1781 	btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1782 	btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1783 	btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1784 	btrfs_set_stack_inode_generation(inode_item,
1785 					 BTRFS_I(inode)->generation);
1786 	btrfs_set_stack_inode_sequence(inode_item,
1787 				       inode_peek_iversion(inode));
1788 	btrfs_set_stack_inode_transid(inode_item, trans->transid);
1789 	btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1790 	flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
1791 					  BTRFS_I(inode)->ro_flags);
1792 	btrfs_set_stack_inode_flags(inode_item, flags);
1793 	btrfs_set_stack_inode_block_group(inode_item, 0);
1794 
1795 	btrfs_set_stack_timespec_sec(&inode_item->atime,
1796 				     inode->i_atime.tv_sec);
1797 	btrfs_set_stack_timespec_nsec(&inode_item->atime,
1798 				      inode->i_atime.tv_nsec);
1799 
1800 	btrfs_set_stack_timespec_sec(&inode_item->mtime,
1801 				     inode->i_mtime.tv_sec);
1802 	btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1803 				      inode->i_mtime.tv_nsec);
1804 
1805 	btrfs_set_stack_timespec_sec(&inode_item->ctime,
1806 				     inode->i_ctime.tv_sec);
1807 	btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1808 				      inode->i_ctime.tv_nsec);
1809 
1810 	btrfs_set_stack_timespec_sec(&inode_item->otime,
1811 				     BTRFS_I(inode)->i_otime.tv_sec);
1812 	btrfs_set_stack_timespec_nsec(&inode_item->otime,
1813 				     BTRFS_I(inode)->i_otime.tv_nsec);
1814 }
1815 
btrfs_fill_inode(struct inode * inode,u32 * rdev)1816 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1817 {
1818 	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1819 	struct btrfs_delayed_node *delayed_node;
1820 	struct btrfs_inode_item *inode_item;
1821 
1822 	delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1823 	if (!delayed_node)
1824 		return -ENOENT;
1825 
1826 	mutex_lock(&delayed_node->mutex);
1827 	if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1828 		mutex_unlock(&delayed_node->mutex);
1829 		btrfs_release_delayed_node(delayed_node);
1830 		return -ENOENT;
1831 	}
1832 
1833 	inode_item = &delayed_node->inode_item;
1834 
1835 	i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1836 	i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1837 	btrfs_i_size_write(BTRFS_I(inode), btrfs_stack_inode_size(inode_item));
1838 	btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
1839 			round_up(i_size_read(inode), fs_info->sectorsize));
1840 	inode->i_mode = btrfs_stack_inode_mode(inode_item);
1841 	set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1842 	inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1843 	BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1844         BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1845 
1846 	inode_set_iversion_queried(inode,
1847 				   btrfs_stack_inode_sequence(inode_item));
1848 	inode->i_rdev = 0;
1849 	*rdev = btrfs_stack_inode_rdev(inode_item);
1850 	btrfs_inode_split_flags(btrfs_stack_inode_flags(inode_item),
1851 				&BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
1852 
1853 	inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1854 	inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1855 
1856 	inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1857 	inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1858 
1859 	inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1860 	inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1861 
1862 	BTRFS_I(inode)->i_otime.tv_sec =
1863 		btrfs_stack_timespec_sec(&inode_item->otime);
1864 	BTRFS_I(inode)->i_otime.tv_nsec =
1865 		btrfs_stack_timespec_nsec(&inode_item->otime);
1866 
1867 	inode->i_generation = BTRFS_I(inode)->generation;
1868 	BTRFS_I(inode)->index_cnt = (u64)-1;
1869 
1870 	mutex_unlock(&delayed_node->mutex);
1871 	btrfs_release_delayed_node(delayed_node);
1872 	return 0;
1873 }
1874 
btrfs_delayed_update_inode(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * inode)1875 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1876 			       struct btrfs_root *root,
1877 			       struct btrfs_inode *inode)
1878 {
1879 	struct btrfs_delayed_node *delayed_node;
1880 	int ret = 0;
1881 
1882 	delayed_node = btrfs_get_or_create_delayed_node(inode);
1883 	if (IS_ERR(delayed_node))
1884 		return PTR_ERR(delayed_node);
1885 
1886 	mutex_lock(&delayed_node->mutex);
1887 	if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1888 		fill_stack_inode_item(trans, &delayed_node->inode_item,
1889 				      &inode->vfs_inode);
1890 		goto release_node;
1891 	}
1892 
1893 	ret = btrfs_delayed_inode_reserve_metadata(trans, root, delayed_node);
1894 	if (ret)
1895 		goto release_node;
1896 
1897 	fill_stack_inode_item(trans, &delayed_node->inode_item, &inode->vfs_inode);
1898 	set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1899 	delayed_node->count++;
1900 	atomic_inc(&root->fs_info->delayed_root->items);
1901 release_node:
1902 	mutex_unlock(&delayed_node->mutex);
1903 	btrfs_release_delayed_node(delayed_node);
1904 	return ret;
1905 }
1906 
btrfs_delayed_delete_inode_ref(struct btrfs_inode * inode)1907 int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
1908 {
1909 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1910 	struct btrfs_delayed_node *delayed_node;
1911 
1912 	/*
1913 	 * we don't do delayed inode updates during log recovery because it
1914 	 * leads to enospc problems.  This means we also can't do
1915 	 * delayed inode refs
1916 	 */
1917 	if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1918 		return -EAGAIN;
1919 
1920 	delayed_node = btrfs_get_or_create_delayed_node(inode);
1921 	if (IS_ERR(delayed_node))
1922 		return PTR_ERR(delayed_node);
1923 
1924 	/*
1925 	 * We don't reserve space for inode ref deletion is because:
1926 	 * - We ONLY do async inode ref deletion for the inode who has only
1927 	 *   one link(i_nlink == 1), it means there is only one inode ref.
1928 	 *   And in most case, the inode ref and the inode item are in the
1929 	 *   same leaf, and we will deal with them at the same time.
1930 	 *   Since we are sure we will reserve the space for the inode item,
1931 	 *   it is unnecessary to reserve space for inode ref deletion.
1932 	 * - If the inode ref and the inode item are not in the same leaf,
1933 	 *   We also needn't worry about enospc problem, because we reserve
1934 	 *   much more space for the inode update than it needs.
1935 	 * - At the worst, we can steal some space from the global reservation.
1936 	 *   It is very rare.
1937 	 */
1938 	mutex_lock(&delayed_node->mutex);
1939 	if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1940 		goto release_node;
1941 
1942 	set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1943 	delayed_node->count++;
1944 	atomic_inc(&fs_info->delayed_root->items);
1945 release_node:
1946 	mutex_unlock(&delayed_node->mutex);
1947 	btrfs_release_delayed_node(delayed_node);
1948 	return 0;
1949 }
1950 
__btrfs_kill_delayed_node(struct btrfs_delayed_node * delayed_node)1951 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1952 {
1953 	struct btrfs_root *root = delayed_node->root;
1954 	struct btrfs_fs_info *fs_info = root->fs_info;
1955 	struct btrfs_delayed_item *curr_item, *prev_item;
1956 
1957 	mutex_lock(&delayed_node->mutex);
1958 	curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1959 	while (curr_item) {
1960 		prev_item = curr_item;
1961 		curr_item = __btrfs_next_delayed_item(prev_item);
1962 		btrfs_release_delayed_item(prev_item);
1963 	}
1964 
1965 	if (delayed_node->index_item_leaves > 0) {
1966 		btrfs_delayed_item_release_leaves(delayed_node,
1967 					  delayed_node->index_item_leaves);
1968 		delayed_node->index_item_leaves = 0;
1969 	}
1970 
1971 	curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1972 	while (curr_item) {
1973 		btrfs_delayed_item_release_metadata(root, curr_item);
1974 		prev_item = curr_item;
1975 		curr_item = __btrfs_next_delayed_item(prev_item);
1976 		btrfs_release_delayed_item(prev_item);
1977 	}
1978 
1979 	btrfs_release_delayed_iref(delayed_node);
1980 
1981 	if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1982 		btrfs_delayed_inode_release_metadata(fs_info, delayed_node, false);
1983 		btrfs_release_delayed_inode(delayed_node);
1984 	}
1985 	mutex_unlock(&delayed_node->mutex);
1986 }
1987 
btrfs_kill_delayed_inode_items(struct btrfs_inode * inode)1988 void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
1989 {
1990 	struct btrfs_delayed_node *delayed_node;
1991 
1992 	delayed_node = btrfs_get_delayed_node(inode);
1993 	if (!delayed_node)
1994 		return;
1995 
1996 	__btrfs_kill_delayed_node(delayed_node);
1997 	btrfs_release_delayed_node(delayed_node);
1998 }
1999 
btrfs_kill_all_delayed_nodes(struct btrfs_root * root)2000 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
2001 {
2002 	u64 inode_id = 0;
2003 	struct btrfs_delayed_node *delayed_nodes[8];
2004 	int i, n;
2005 
2006 	while (1) {
2007 		spin_lock(&root->inode_lock);
2008 		n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
2009 					   (void **)delayed_nodes, inode_id,
2010 					   ARRAY_SIZE(delayed_nodes));
2011 		if (!n) {
2012 			spin_unlock(&root->inode_lock);
2013 			break;
2014 		}
2015 
2016 		inode_id = delayed_nodes[n - 1]->inode_id + 1;
2017 		for (i = 0; i < n; i++) {
2018 			/*
2019 			 * Don't increase refs in case the node is dead and
2020 			 * about to be removed from the tree in the loop below
2021 			 */
2022 			if (!refcount_inc_not_zero(&delayed_nodes[i]->refs))
2023 				delayed_nodes[i] = NULL;
2024 		}
2025 		spin_unlock(&root->inode_lock);
2026 
2027 		for (i = 0; i < n; i++) {
2028 			if (!delayed_nodes[i])
2029 				continue;
2030 			__btrfs_kill_delayed_node(delayed_nodes[i]);
2031 			btrfs_release_delayed_node(delayed_nodes[i]);
2032 		}
2033 	}
2034 }
2035 
btrfs_destroy_delayed_inodes(struct btrfs_fs_info * fs_info)2036 void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
2037 {
2038 	struct btrfs_delayed_node *curr_node, *prev_node;
2039 
2040 	curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
2041 	while (curr_node) {
2042 		__btrfs_kill_delayed_node(curr_node);
2043 
2044 		prev_node = curr_node;
2045 		curr_node = btrfs_next_delayed_node(curr_node);
2046 		btrfs_release_delayed_node(prev_node);
2047 	}
2048 }
2049 
btrfs_log_get_delayed_items(struct btrfs_inode * inode,struct list_head * ins_list,struct list_head * del_list)2050 void btrfs_log_get_delayed_items(struct btrfs_inode *inode,
2051 				 struct list_head *ins_list,
2052 				 struct list_head *del_list)
2053 {
2054 	struct btrfs_delayed_node *node;
2055 	struct btrfs_delayed_item *item;
2056 
2057 	node = btrfs_get_delayed_node(inode);
2058 	if (!node)
2059 		return;
2060 
2061 	mutex_lock(&node->mutex);
2062 	item = __btrfs_first_delayed_insertion_item(node);
2063 	while (item) {
2064 		/*
2065 		 * It's possible that the item is already in a log list. This
2066 		 * can happen in case two tasks are trying to log the same
2067 		 * directory. For example if we have tasks A and task B:
2068 		 *
2069 		 * Task A collected the delayed items into a log list while
2070 		 * under the inode's log_mutex (at btrfs_log_inode()), but it
2071 		 * only releases the items after logging the inodes they point
2072 		 * to (if they are new inodes), which happens after unlocking
2073 		 * the log mutex;
2074 		 *
2075 		 * Task B enters btrfs_log_inode() and acquires the log_mutex
2076 		 * of the same directory inode, before task B releases the
2077 		 * delayed items. This can happen for example when logging some
2078 		 * inode we need to trigger logging of its parent directory, so
2079 		 * logging two files that have the same parent directory can
2080 		 * lead to this.
2081 		 *
2082 		 * If this happens, just ignore delayed items already in a log
2083 		 * list. All the tasks logging the directory are under a log
2084 		 * transaction and whichever finishes first can not sync the log
2085 		 * before the other completes and leaves the log transaction.
2086 		 */
2087 		if (!item->logged && list_empty(&item->log_list)) {
2088 			refcount_inc(&item->refs);
2089 			list_add_tail(&item->log_list, ins_list);
2090 		}
2091 		item = __btrfs_next_delayed_item(item);
2092 	}
2093 
2094 	item = __btrfs_first_delayed_deletion_item(node);
2095 	while (item) {
2096 		/* It may be non-empty, for the same reason mentioned above. */
2097 		if (!item->logged && list_empty(&item->log_list)) {
2098 			refcount_inc(&item->refs);
2099 			list_add_tail(&item->log_list, del_list);
2100 		}
2101 		item = __btrfs_next_delayed_item(item);
2102 	}
2103 	mutex_unlock(&node->mutex);
2104 
2105 	/*
2106 	 * We are called during inode logging, which means the inode is in use
2107 	 * and can not be evicted before we finish logging the inode. So we never
2108 	 * have the last reference on the delayed inode.
2109 	 * Also, we don't use btrfs_release_delayed_node() because that would
2110 	 * requeue the delayed inode (change its order in the list of prepared
2111 	 * nodes) and we don't want to do such change because we don't create or
2112 	 * delete delayed items.
2113 	 */
2114 	ASSERT(refcount_read(&node->refs) > 1);
2115 	refcount_dec(&node->refs);
2116 }
2117 
btrfs_log_put_delayed_items(struct btrfs_inode * inode,struct list_head * ins_list,struct list_head * del_list)2118 void btrfs_log_put_delayed_items(struct btrfs_inode *inode,
2119 				 struct list_head *ins_list,
2120 				 struct list_head *del_list)
2121 {
2122 	struct btrfs_delayed_node *node;
2123 	struct btrfs_delayed_item *item;
2124 	struct btrfs_delayed_item *next;
2125 
2126 	node = btrfs_get_delayed_node(inode);
2127 	if (!node)
2128 		return;
2129 
2130 	mutex_lock(&node->mutex);
2131 
2132 	list_for_each_entry_safe(item, next, ins_list, log_list) {
2133 		item->logged = true;
2134 		list_del_init(&item->log_list);
2135 		if (refcount_dec_and_test(&item->refs))
2136 			kfree(item);
2137 	}
2138 
2139 	list_for_each_entry_safe(item, next, del_list, log_list) {
2140 		item->logged = true;
2141 		list_del_init(&item->log_list);
2142 		if (refcount_dec_and_test(&item->refs))
2143 			kfree(item);
2144 	}
2145 
2146 	mutex_unlock(&node->mutex);
2147 
2148 	/*
2149 	 * We are called during inode logging, which means the inode is in use
2150 	 * and can not be evicted before we finish logging the inode. So we never
2151 	 * have the last reference on the delayed inode.
2152 	 * Also, we don't use btrfs_release_delayed_node() because that would
2153 	 * requeue the delayed inode (change its order in the list of prepared
2154 	 * nodes) and we don't want to do such change because we don't create or
2155 	 * delete delayed items.
2156 	 */
2157 	ASSERT(refcount_read(&node->refs) > 1);
2158 	refcount_dec(&node->refs);
2159 }
2160