cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

ordered-data.c (31108B)


      1// SPDX-License-Identifier: GPL-2.0
      2/*
      3 * Copyright (C) 2007 Oracle.  All rights reserved.
      4 */
      5
      6#include <linux/slab.h>
      7#include <linux/blkdev.h>
      8#include <linux/writeback.h>
      9#include <linux/sched/mm.h>
     10#include "misc.h"
     11#include "ctree.h"
     12#include "transaction.h"
     13#include "btrfs_inode.h"
     14#include "extent_io.h"
     15#include "disk-io.h"
     16#include "compression.h"
     17#include "delalloc-space.h"
     18#include "qgroup.h"
     19#include "subpage.h"
     20
     21static struct kmem_cache *btrfs_ordered_extent_cache;
     22
     23static u64 entry_end(struct btrfs_ordered_extent *entry)
     24{
     25	if (entry->file_offset + entry->num_bytes < entry->file_offset)
     26		return (u64)-1;
     27	return entry->file_offset + entry->num_bytes;
     28}
     29
     30/* returns NULL if the insertion worked, or it returns the node it did find
     31 * in the tree
     32 */
     33static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
     34				   struct rb_node *node)
     35{
     36	struct rb_node **p = &root->rb_node;
     37	struct rb_node *parent = NULL;
     38	struct btrfs_ordered_extent *entry;
     39
     40	while (*p) {
     41		parent = *p;
     42		entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
     43
     44		if (file_offset < entry->file_offset)
     45			p = &(*p)->rb_left;
     46		else if (file_offset >= entry_end(entry))
     47			p = &(*p)->rb_right;
     48		else
     49			return parent;
     50	}
     51
     52	rb_link_node(node, parent, p);
     53	rb_insert_color(node, root);
     54	return NULL;
     55}
     56
     57/*
     58 * look for a given offset in the tree, and if it can't be found return the
     59 * first lesser offset
     60 */
     61static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
     62				     struct rb_node **prev_ret)
     63{
     64	struct rb_node *n = root->rb_node;
     65	struct rb_node *prev = NULL;
     66	struct rb_node *test;
     67	struct btrfs_ordered_extent *entry;
     68	struct btrfs_ordered_extent *prev_entry = NULL;
     69
     70	while (n) {
     71		entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
     72		prev = n;
     73		prev_entry = entry;
     74
     75		if (file_offset < entry->file_offset)
     76			n = n->rb_left;
     77		else if (file_offset >= entry_end(entry))
     78			n = n->rb_right;
     79		else
     80			return n;
     81	}
     82	if (!prev_ret)
     83		return NULL;
     84
     85	while (prev && file_offset >= entry_end(prev_entry)) {
     86		test = rb_next(prev);
     87		if (!test)
     88			break;
     89		prev_entry = rb_entry(test, struct btrfs_ordered_extent,
     90				      rb_node);
     91		if (file_offset < entry_end(prev_entry))
     92			break;
     93
     94		prev = test;
     95	}
     96	if (prev)
     97		prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
     98				      rb_node);
     99	while (prev && file_offset < entry_end(prev_entry)) {
    100		test = rb_prev(prev);
    101		if (!test)
    102			break;
    103		prev_entry = rb_entry(test, struct btrfs_ordered_extent,
    104				      rb_node);
    105		prev = test;
    106	}
    107	*prev_ret = prev;
    108	return NULL;
    109}
    110
    111static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
    112			  u64 len)
    113{
    114	if (file_offset + len <= entry->file_offset ||
    115	    entry->file_offset + entry->num_bytes <= file_offset)
    116		return 0;
    117	return 1;
    118}
    119
    120/*
    121 * look find the first ordered struct that has this offset, otherwise
    122 * the first one less than this offset
    123 */
    124static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
    125					  u64 file_offset)
    126{
    127	struct rb_root *root = &tree->tree;
    128	struct rb_node *prev = NULL;
    129	struct rb_node *ret;
    130	struct btrfs_ordered_extent *entry;
    131
    132	if (tree->last) {
    133		entry = rb_entry(tree->last, struct btrfs_ordered_extent,
    134				 rb_node);
    135		if (in_range(file_offset, entry->file_offset, entry->num_bytes))
    136			return tree->last;
    137	}
    138	ret = __tree_search(root, file_offset, &prev);
    139	if (!ret)
    140		ret = prev;
    141	if (ret)
    142		tree->last = ret;
    143	return ret;
    144}
    145
    146/**
    147 * Add an ordered extent to the per-inode tree.
    148 *
    149 * @inode:           Inode that this extent is for.
    150 * @file_offset:     Logical offset in file where the extent starts.
    151 * @num_bytes:       Logical length of extent in file.
    152 * @ram_bytes:       Full length of unencoded data.
    153 * @disk_bytenr:     Offset of extent on disk.
    154 * @disk_num_bytes:  Size of extent on disk.
    155 * @offset:          Offset into unencoded data where file data starts.
    156 * @flags:           Flags specifying type of extent (1 << BTRFS_ORDERED_*).
    157 * @compress_type:   Compression algorithm used for data.
    158 *
    159 * Most of these parameters correspond to &struct btrfs_file_extent_item. The
    160 * tree is given a single reference on the ordered extent that was inserted.
    161 *
    162 * Return: 0 or -ENOMEM.
    163 */
    164int btrfs_add_ordered_extent(struct btrfs_inode *inode, u64 file_offset,
    165			     u64 num_bytes, u64 ram_bytes, u64 disk_bytenr,
    166			     u64 disk_num_bytes, u64 offset, unsigned flags,
    167			     int compress_type)
    168{
    169	struct btrfs_root *root = inode->root;
    170	struct btrfs_fs_info *fs_info = root->fs_info;
    171	struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
    172	struct rb_node *node;
    173	struct btrfs_ordered_extent *entry;
    174	int ret;
    175
    176	if (flags &
    177	    ((1 << BTRFS_ORDERED_NOCOW) | (1 << BTRFS_ORDERED_PREALLOC))) {
    178		/* For nocow write, we can release the qgroup rsv right now */
    179		ret = btrfs_qgroup_free_data(inode, NULL, file_offset, num_bytes);
    180		if (ret < 0)
    181			return ret;
    182		ret = 0;
    183	} else {
    184		/*
    185		 * The ordered extent has reserved qgroup space, release now
    186		 * and pass the reserved number for qgroup_record to free.
    187		 */
    188		ret = btrfs_qgroup_release_data(inode, file_offset, num_bytes);
    189		if (ret < 0)
    190			return ret;
    191	}
    192	entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
    193	if (!entry)
    194		return -ENOMEM;
    195
    196	entry->file_offset = file_offset;
    197	entry->num_bytes = num_bytes;
    198	entry->ram_bytes = ram_bytes;
    199	entry->disk_bytenr = disk_bytenr;
    200	entry->disk_num_bytes = disk_num_bytes;
    201	entry->offset = offset;
    202	entry->bytes_left = num_bytes;
    203	entry->inode = igrab(&inode->vfs_inode);
    204	entry->compress_type = compress_type;
    205	entry->truncated_len = (u64)-1;
    206	entry->qgroup_rsv = ret;
    207	entry->physical = (u64)-1;
    208
    209	ASSERT((flags & ~BTRFS_ORDERED_TYPE_FLAGS) == 0);
    210	entry->flags = flags;
    211
    212	percpu_counter_add_batch(&fs_info->ordered_bytes, num_bytes,
    213				 fs_info->delalloc_batch);
    214
    215	/* one ref for the tree */
    216	refcount_set(&entry->refs, 1);
    217	init_waitqueue_head(&entry->wait);
    218	INIT_LIST_HEAD(&entry->list);
    219	INIT_LIST_HEAD(&entry->log_list);
    220	INIT_LIST_HEAD(&entry->root_extent_list);
    221	INIT_LIST_HEAD(&entry->work_list);
    222	init_completion(&entry->completion);
    223
    224	trace_btrfs_ordered_extent_add(inode, entry);
    225
    226	spin_lock_irq(&tree->lock);
    227	node = tree_insert(&tree->tree, file_offset,
    228			   &entry->rb_node);
    229	if (node)
    230		btrfs_panic(fs_info, -EEXIST,
    231				"inconsistency in ordered tree at offset %llu",
    232				file_offset);
    233	spin_unlock_irq(&tree->lock);
    234
    235	spin_lock(&root->ordered_extent_lock);
    236	list_add_tail(&entry->root_extent_list,
    237		      &root->ordered_extents);
    238	root->nr_ordered_extents++;
    239	if (root->nr_ordered_extents == 1) {
    240		spin_lock(&fs_info->ordered_root_lock);
    241		BUG_ON(!list_empty(&root->ordered_root));
    242		list_add_tail(&root->ordered_root, &fs_info->ordered_roots);
    243		spin_unlock(&fs_info->ordered_root_lock);
    244	}
    245	spin_unlock(&root->ordered_extent_lock);
    246
    247	/*
    248	 * We don't need the count_max_extents here, we can assume that all of
    249	 * that work has been done at higher layers, so this is truly the
    250	 * smallest the extent is going to get.
    251	 */
    252	spin_lock(&inode->lock);
    253	btrfs_mod_outstanding_extents(inode, 1);
    254	spin_unlock(&inode->lock);
    255
    256	return 0;
    257}
    258
    259/*
    260 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
    261 * when an ordered extent is finished.  If the list covers more than one
    262 * ordered extent, it is split across multiples.
    263 */
    264void btrfs_add_ordered_sum(struct btrfs_ordered_extent *entry,
    265			   struct btrfs_ordered_sum *sum)
    266{
    267	struct btrfs_ordered_inode_tree *tree;
    268
    269	tree = &BTRFS_I(entry->inode)->ordered_tree;
    270	spin_lock_irq(&tree->lock);
    271	list_add_tail(&sum->list, &entry->list);
    272	spin_unlock_irq(&tree->lock);
    273}
    274
    275/*
    276 * Mark all ordered extents io inside the specified range finished.
    277 *
    278 * @page:	 The invovled page for the opeartion.
    279 *		 For uncompressed buffered IO, the page status also needs to be
    280 *		 updated to indicate whether the pending ordered io is finished.
    281 *		 Can be NULL for direct IO and compressed write.
    282 *		 For these cases, callers are ensured they won't execute the
    283 *		 endio function twice.
    284 * @finish_func: The function to be executed when all the IO of an ordered
    285 *		 extent are finished.
    286 *
    287 * This function is called for endio, thus the range must have ordered
    288 * extent(s) coveri it.
    289 */
    290void btrfs_mark_ordered_io_finished(struct btrfs_inode *inode,
    291				struct page *page, u64 file_offset,
    292				u64 num_bytes, btrfs_func_t finish_func,
    293				bool uptodate)
    294{
    295	struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
    296	struct btrfs_fs_info *fs_info = inode->root->fs_info;
    297	struct btrfs_workqueue *wq;
    298	struct rb_node *node;
    299	struct btrfs_ordered_extent *entry = NULL;
    300	unsigned long flags;
    301	u64 cur = file_offset;
    302
    303	if (btrfs_is_free_space_inode(inode))
    304		wq = fs_info->endio_freespace_worker;
    305	else
    306		wq = fs_info->endio_write_workers;
    307
    308	if (page)
    309		ASSERT(page->mapping && page_offset(page) <= file_offset &&
    310		       file_offset + num_bytes <= page_offset(page) + PAGE_SIZE);
    311
    312	spin_lock_irqsave(&tree->lock, flags);
    313	while (cur < file_offset + num_bytes) {
    314		u64 entry_end;
    315		u64 end;
    316		u32 len;
    317
    318		node = tree_search(tree, cur);
    319		/* No ordered extents at all */
    320		if (!node)
    321			break;
    322
    323		entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
    324		entry_end = entry->file_offset + entry->num_bytes;
    325		/*
    326		 * |<-- OE --->|  |
    327		 *		  cur
    328		 * Go to next OE.
    329		 */
    330		if (cur >= entry_end) {
    331			node = rb_next(node);
    332			/* No more ordered extents, exit */
    333			if (!node)
    334				break;
    335			entry = rb_entry(node, struct btrfs_ordered_extent,
    336					 rb_node);
    337
    338			/* Go to next ordered extent and continue */
    339			cur = entry->file_offset;
    340			continue;
    341		}
    342		/*
    343		 * |	|<--- OE --->|
    344		 * cur
    345		 * Go to the start of OE.
    346		 */
    347		if (cur < entry->file_offset) {
    348			cur = entry->file_offset;
    349			continue;
    350		}
    351
    352		/*
    353		 * Now we are definitely inside one ordered extent.
    354		 *
    355		 * |<--- OE --->|
    356		 *	|
    357		 *	cur
    358		 */
    359		end = min(entry->file_offset + entry->num_bytes,
    360			  file_offset + num_bytes) - 1;
    361		ASSERT(end + 1 - cur < U32_MAX);
    362		len = end + 1 - cur;
    363
    364		if (page) {
    365			/*
    366			 * Ordered (Private2) bit indicates whether we still
    367			 * have pending io unfinished for the ordered extent.
    368			 *
    369			 * If there's no such bit, we need to skip to next range.
    370			 */
    371			if (!btrfs_page_test_ordered(fs_info, page, cur, len)) {
    372				cur += len;
    373				continue;
    374			}
    375			btrfs_page_clear_ordered(fs_info, page, cur, len);
    376		}
    377
    378		/* Now we're fine to update the accounting */
    379		if (unlikely(len > entry->bytes_left)) {
    380			WARN_ON(1);
    381			btrfs_crit(fs_info,
    382"bad ordered extent accounting, root=%llu ino=%llu OE offset=%llu OE len=%llu to_dec=%u left=%llu",
    383				   inode->root->root_key.objectid,
    384				   btrfs_ino(inode),
    385				   entry->file_offset,
    386				   entry->num_bytes,
    387				   len, entry->bytes_left);
    388			entry->bytes_left = 0;
    389		} else {
    390			entry->bytes_left -= len;
    391		}
    392
    393		if (!uptodate)
    394			set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
    395
    396		/*
    397		 * All the IO of the ordered extent is finished, we need to queue
    398		 * the finish_func to be executed.
    399		 */
    400		if (entry->bytes_left == 0) {
    401			set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
    402			cond_wake_up(&entry->wait);
    403			refcount_inc(&entry->refs);
    404			spin_unlock_irqrestore(&tree->lock, flags);
    405			btrfs_init_work(&entry->work, finish_func, NULL, NULL);
    406			btrfs_queue_work(wq, &entry->work);
    407			spin_lock_irqsave(&tree->lock, flags);
    408		}
    409		cur += len;
    410	}
    411	spin_unlock_irqrestore(&tree->lock, flags);
    412}
    413
    414/*
    415 * Finish IO for one ordered extent across a given range.  The range can only
    416 * contain one ordered extent.
    417 *
    418 * @cached:	 The cached ordered extent. If not NULL, we can skip the tree
    419 *               search and use the ordered extent directly.
    420 * 		 Will be also used to store the finished ordered extent.
    421 * @file_offset: File offset for the finished IO
    422 * @io_size:	 Length of the finish IO range
    423 *
    424 * Return true if the ordered extent is finished in the range, and update
    425 * @cached.
    426 * Return false otherwise.
    427 *
    428 * NOTE: The range can NOT cross multiple ordered extents.
    429 * Thus caller should ensure the range doesn't cross ordered extents.
    430 */
    431bool btrfs_dec_test_ordered_pending(struct btrfs_inode *inode,
    432				    struct btrfs_ordered_extent **cached,
    433				    u64 file_offset, u64 io_size)
    434{
    435	struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
    436	struct rb_node *node;
    437	struct btrfs_ordered_extent *entry = NULL;
    438	unsigned long flags;
    439	bool finished = false;
    440
    441	spin_lock_irqsave(&tree->lock, flags);
    442	if (cached && *cached) {
    443		entry = *cached;
    444		goto have_entry;
    445	}
    446
    447	node = tree_search(tree, file_offset);
    448	if (!node)
    449		goto out;
    450
    451	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
    452have_entry:
    453	if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
    454		goto out;
    455
    456	if (io_size > entry->bytes_left)
    457		btrfs_crit(inode->root->fs_info,
    458			   "bad ordered accounting left %llu size %llu",
    459		       entry->bytes_left, io_size);
    460
    461	entry->bytes_left -= io_size;
    462
    463	if (entry->bytes_left == 0) {
    464		/*
    465		 * Ensure only one caller can set the flag and finished_ret
    466		 * accordingly
    467		 */
    468		finished = !test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
    469		/* test_and_set_bit implies a barrier */
    470		cond_wake_up_nomb(&entry->wait);
    471	}
    472out:
    473	if (finished && cached && entry) {
    474		*cached = entry;
    475		refcount_inc(&entry->refs);
    476	}
    477	spin_unlock_irqrestore(&tree->lock, flags);
    478	return finished;
    479}
    480
    481/*
    482 * used to drop a reference on an ordered extent.  This will free
    483 * the extent if the last reference is dropped
    484 */
    485void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
    486{
    487	struct list_head *cur;
    488	struct btrfs_ordered_sum *sum;
    489
    490	trace_btrfs_ordered_extent_put(BTRFS_I(entry->inode), entry);
    491
    492	if (refcount_dec_and_test(&entry->refs)) {
    493		ASSERT(list_empty(&entry->root_extent_list));
    494		ASSERT(list_empty(&entry->log_list));
    495		ASSERT(RB_EMPTY_NODE(&entry->rb_node));
    496		if (entry->inode)
    497			btrfs_add_delayed_iput(entry->inode);
    498		while (!list_empty(&entry->list)) {
    499			cur = entry->list.next;
    500			sum = list_entry(cur, struct btrfs_ordered_sum, list);
    501			list_del(&sum->list);
    502			kvfree(sum);
    503		}
    504		kmem_cache_free(btrfs_ordered_extent_cache, entry);
    505	}
    506}
    507
    508/*
    509 * remove an ordered extent from the tree.  No references are dropped
    510 * and waiters are woken up.
    511 */
    512void btrfs_remove_ordered_extent(struct btrfs_inode *btrfs_inode,
    513				 struct btrfs_ordered_extent *entry)
    514{
    515	struct btrfs_ordered_inode_tree *tree;
    516	struct btrfs_root *root = btrfs_inode->root;
    517	struct btrfs_fs_info *fs_info = root->fs_info;
    518	struct rb_node *node;
    519	bool pending;
    520
    521	/* This is paired with btrfs_add_ordered_extent. */
    522	spin_lock(&btrfs_inode->lock);
    523	btrfs_mod_outstanding_extents(btrfs_inode, -1);
    524	spin_unlock(&btrfs_inode->lock);
    525	if (root != fs_info->tree_root) {
    526		u64 release;
    527
    528		if (test_bit(BTRFS_ORDERED_ENCODED, &entry->flags))
    529			release = entry->disk_num_bytes;
    530		else
    531			release = entry->num_bytes;
    532		btrfs_delalloc_release_metadata(btrfs_inode, release, false);
    533	}
    534
    535	percpu_counter_add_batch(&fs_info->ordered_bytes, -entry->num_bytes,
    536				 fs_info->delalloc_batch);
    537
    538	tree = &btrfs_inode->ordered_tree;
    539	spin_lock_irq(&tree->lock);
    540	node = &entry->rb_node;
    541	rb_erase(node, &tree->tree);
    542	RB_CLEAR_NODE(node);
    543	if (tree->last == node)
    544		tree->last = NULL;
    545	set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
    546	pending = test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags);
    547	spin_unlock_irq(&tree->lock);
    548
    549	/*
    550	 * The current running transaction is waiting on us, we need to let it
    551	 * know that we're complete and wake it up.
    552	 */
    553	if (pending) {
    554		struct btrfs_transaction *trans;
    555
    556		/*
    557		 * The checks for trans are just a formality, it should be set,
    558		 * but if it isn't we don't want to deref/assert under the spin
    559		 * lock, so be nice and check if trans is set, but ASSERT() so
    560		 * if it isn't set a developer will notice.
    561		 */
    562		spin_lock(&fs_info->trans_lock);
    563		trans = fs_info->running_transaction;
    564		if (trans)
    565			refcount_inc(&trans->use_count);
    566		spin_unlock(&fs_info->trans_lock);
    567
    568		ASSERT(trans);
    569		if (trans) {
    570			if (atomic_dec_and_test(&trans->pending_ordered))
    571				wake_up(&trans->pending_wait);
    572			btrfs_put_transaction(trans);
    573		}
    574	}
    575
    576	spin_lock(&root->ordered_extent_lock);
    577	list_del_init(&entry->root_extent_list);
    578	root->nr_ordered_extents--;
    579
    580	trace_btrfs_ordered_extent_remove(btrfs_inode, entry);
    581
    582	if (!root->nr_ordered_extents) {
    583		spin_lock(&fs_info->ordered_root_lock);
    584		BUG_ON(list_empty(&root->ordered_root));
    585		list_del_init(&root->ordered_root);
    586		spin_unlock(&fs_info->ordered_root_lock);
    587	}
    588	spin_unlock(&root->ordered_extent_lock);
    589	wake_up(&entry->wait);
    590}
    591
    592static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
    593{
    594	struct btrfs_ordered_extent *ordered;
    595
    596	ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
    597	btrfs_start_ordered_extent(ordered, 1);
    598	complete(&ordered->completion);
    599}
    600
    601/*
    602 * wait for all the ordered extents in a root.  This is done when balancing
    603 * space between drives.
    604 */
    605u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr,
    606			       const u64 range_start, const u64 range_len)
    607{
    608	struct btrfs_fs_info *fs_info = root->fs_info;
    609	LIST_HEAD(splice);
    610	LIST_HEAD(skipped);
    611	LIST_HEAD(works);
    612	struct btrfs_ordered_extent *ordered, *next;
    613	u64 count = 0;
    614	const u64 range_end = range_start + range_len;
    615
    616	mutex_lock(&root->ordered_extent_mutex);
    617	spin_lock(&root->ordered_extent_lock);
    618	list_splice_init(&root->ordered_extents, &splice);
    619	while (!list_empty(&splice) && nr) {
    620		ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
    621					   root_extent_list);
    622
    623		if (range_end <= ordered->disk_bytenr ||
    624		    ordered->disk_bytenr + ordered->disk_num_bytes <= range_start) {
    625			list_move_tail(&ordered->root_extent_list, &skipped);
    626			cond_resched_lock(&root->ordered_extent_lock);
    627			continue;
    628		}
    629
    630		list_move_tail(&ordered->root_extent_list,
    631			       &root->ordered_extents);
    632		refcount_inc(&ordered->refs);
    633		spin_unlock(&root->ordered_extent_lock);
    634
    635		btrfs_init_work(&ordered->flush_work,
    636				btrfs_run_ordered_extent_work, NULL, NULL);
    637		list_add_tail(&ordered->work_list, &works);
    638		btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work);
    639
    640		cond_resched();
    641		spin_lock(&root->ordered_extent_lock);
    642		if (nr != U64_MAX)
    643			nr--;
    644		count++;
    645	}
    646	list_splice_tail(&skipped, &root->ordered_extents);
    647	list_splice_tail(&splice, &root->ordered_extents);
    648	spin_unlock(&root->ordered_extent_lock);
    649
    650	list_for_each_entry_safe(ordered, next, &works, work_list) {
    651		list_del_init(&ordered->work_list);
    652		wait_for_completion(&ordered->completion);
    653		btrfs_put_ordered_extent(ordered);
    654		cond_resched();
    655	}
    656	mutex_unlock(&root->ordered_extent_mutex);
    657
    658	return count;
    659}
    660
    661void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr,
    662			     const u64 range_start, const u64 range_len)
    663{
    664	struct btrfs_root *root;
    665	struct list_head splice;
    666	u64 done;
    667
    668	INIT_LIST_HEAD(&splice);
    669
    670	mutex_lock(&fs_info->ordered_operations_mutex);
    671	spin_lock(&fs_info->ordered_root_lock);
    672	list_splice_init(&fs_info->ordered_roots, &splice);
    673	while (!list_empty(&splice) && nr) {
    674		root = list_first_entry(&splice, struct btrfs_root,
    675					ordered_root);
    676		root = btrfs_grab_root(root);
    677		BUG_ON(!root);
    678		list_move_tail(&root->ordered_root,
    679			       &fs_info->ordered_roots);
    680		spin_unlock(&fs_info->ordered_root_lock);
    681
    682		done = btrfs_wait_ordered_extents(root, nr,
    683						  range_start, range_len);
    684		btrfs_put_root(root);
    685
    686		spin_lock(&fs_info->ordered_root_lock);
    687		if (nr != U64_MAX) {
    688			nr -= done;
    689		}
    690	}
    691	list_splice_tail(&splice, &fs_info->ordered_roots);
    692	spin_unlock(&fs_info->ordered_root_lock);
    693	mutex_unlock(&fs_info->ordered_operations_mutex);
    694}
    695
    696/*
    697 * Used to start IO or wait for a given ordered extent to finish.
    698 *
    699 * If wait is one, this effectively waits on page writeback for all the pages
    700 * in the extent, and it waits on the io completion code to insert
    701 * metadata into the btree corresponding to the extent
    702 */
    703void btrfs_start_ordered_extent(struct btrfs_ordered_extent *entry, int wait)
    704{
    705	u64 start = entry->file_offset;
    706	u64 end = start + entry->num_bytes - 1;
    707	struct btrfs_inode *inode = BTRFS_I(entry->inode);
    708
    709	trace_btrfs_ordered_extent_start(inode, entry);
    710
    711	/*
    712	 * pages in the range can be dirty, clean or writeback.  We
    713	 * start IO on any dirty ones so the wait doesn't stall waiting
    714	 * for the flusher thread to find them
    715	 */
    716	if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
    717		filemap_fdatawrite_range(inode->vfs_inode.i_mapping, start, end);
    718	if (wait) {
    719		wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
    720						 &entry->flags));
    721	}
    722}
    723
    724/*
    725 * Used to wait on ordered extents across a large range of bytes.
    726 */
    727int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
    728{
    729	int ret = 0;
    730	int ret_wb = 0;
    731	u64 end;
    732	u64 orig_end;
    733	struct btrfs_ordered_extent *ordered;
    734
    735	if (start + len < start) {
    736		orig_end = INT_LIMIT(loff_t);
    737	} else {
    738		orig_end = start + len - 1;
    739		if (orig_end > INT_LIMIT(loff_t))
    740			orig_end = INT_LIMIT(loff_t);
    741	}
    742
    743	/* start IO across the range first to instantiate any delalloc
    744	 * extents
    745	 */
    746	ret = btrfs_fdatawrite_range(inode, start, orig_end);
    747	if (ret)
    748		return ret;
    749
    750	/*
    751	 * If we have a writeback error don't return immediately. Wait first
    752	 * for any ordered extents that haven't completed yet. This is to make
    753	 * sure no one can dirty the same page ranges and call writepages()
    754	 * before the ordered extents complete - to avoid failures (-EEXIST)
    755	 * when adding the new ordered extents to the ordered tree.
    756	 */
    757	ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
    758
    759	end = orig_end;
    760	while (1) {
    761		ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode), end);
    762		if (!ordered)
    763			break;
    764		if (ordered->file_offset > orig_end) {
    765			btrfs_put_ordered_extent(ordered);
    766			break;
    767		}
    768		if (ordered->file_offset + ordered->num_bytes <= start) {
    769			btrfs_put_ordered_extent(ordered);
    770			break;
    771		}
    772		btrfs_start_ordered_extent(ordered, 1);
    773		end = ordered->file_offset;
    774		/*
    775		 * If the ordered extent had an error save the error but don't
    776		 * exit without waiting first for all other ordered extents in
    777		 * the range to complete.
    778		 */
    779		if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
    780			ret = -EIO;
    781		btrfs_put_ordered_extent(ordered);
    782		if (end == 0 || end == start)
    783			break;
    784		end--;
    785	}
    786	return ret_wb ? ret_wb : ret;
    787}
    788
    789/*
    790 * find an ordered extent corresponding to file_offset.  return NULL if
    791 * nothing is found, otherwise take a reference on the extent and return it
    792 */
    793struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct btrfs_inode *inode,
    794							 u64 file_offset)
    795{
    796	struct btrfs_ordered_inode_tree *tree;
    797	struct rb_node *node;
    798	struct btrfs_ordered_extent *entry = NULL;
    799	unsigned long flags;
    800
    801	tree = &inode->ordered_tree;
    802	spin_lock_irqsave(&tree->lock, flags);
    803	node = tree_search(tree, file_offset);
    804	if (!node)
    805		goto out;
    806
    807	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
    808	if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
    809		entry = NULL;
    810	if (entry)
    811		refcount_inc(&entry->refs);
    812out:
    813	spin_unlock_irqrestore(&tree->lock, flags);
    814	return entry;
    815}
    816
    817/* Since the DIO code tries to lock a wide area we need to look for any ordered
    818 * extents that exist in the range, rather than just the start of the range.
    819 */
    820struct btrfs_ordered_extent *btrfs_lookup_ordered_range(
    821		struct btrfs_inode *inode, u64 file_offset, u64 len)
    822{
    823	struct btrfs_ordered_inode_tree *tree;
    824	struct rb_node *node;
    825	struct btrfs_ordered_extent *entry = NULL;
    826
    827	tree = &inode->ordered_tree;
    828	spin_lock_irq(&tree->lock);
    829	node = tree_search(tree, file_offset);
    830	if (!node) {
    831		node = tree_search(tree, file_offset + len);
    832		if (!node)
    833			goto out;
    834	}
    835
    836	while (1) {
    837		entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
    838		if (range_overlaps(entry, file_offset, len))
    839			break;
    840
    841		if (entry->file_offset >= file_offset + len) {
    842			entry = NULL;
    843			break;
    844		}
    845		entry = NULL;
    846		node = rb_next(node);
    847		if (!node)
    848			break;
    849	}
    850out:
    851	if (entry)
    852		refcount_inc(&entry->refs);
    853	spin_unlock_irq(&tree->lock);
    854	return entry;
    855}
    856
    857/*
    858 * Adds all ordered extents to the given list. The list ends up sorted by the
    859 * file_offset of the ordered extents.
    860 */
    861void btrfs_get_ordered_extents_for_logging(struct btrfs_inode *inode,
    862					   struct list_head *list)
    863{
    864	struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
    865	struct rb_node *n;
    866
    867	ASSERT(inode_is_locked(&inode->vfs_inode));
    868
    869	spin_lock_irq(&tree->lock);
    870	for (n = rb_first(&tree->tree); n; n = rb_next(n)) {
    871		struct btrfs_ordered_extent *ordered;
    872
    873		ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
    874
    875		if (test_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
    876			continue;
    877
    878		ASSERT(list_empty(&ordered->log_list));
    879		list_add_tail(&ordered->log_list, list);
    880		refcount_inc(&ordered->refs);
    881	}
    882	spin_unlock_irq(&tree->lock);
    883}
    884
    885/*
    886 * lookup and return any extent before 'file_offset'.  NULL is returned
    887 * if none is found
    888 */
    889struct btrfs_ordered_extent *
    890btrfs_lookup_first_ordered_extent(struct btrfs_inode *inode, u64 file_offset)
    891{
    892	struct btrfs_ordered_inode_tree *tree;
    893	struct rb_node *node;
    894	struct btrfs_ordered_extent *entry = NULL;
    895
    896	tree = &inode->ordered_tree;
    897	spin_lock_irq(&tree->lock);
    898	node = tree_search(tree, file_offset);
    899	if (!node)
    900		goto out;
    901
    902	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
    903	refcount_inc(&entry->refs);
    904out:
    905	spin_unlock_irq(&tree->lock);
    906	return entry;
    907}
    908
    909/*
    910 * Lookup the first ordered extent that overlaps the range
    911 * [@file_offset, @file_offset + @len).
    912 *
    913 * The difference between this and btrfs_lookup_first_ordered_extent() is
    914 * that this one won't return any ordered extent that does not overlap the range.
    915 * And the difference against btrfs_lookup_ordered_extent() is, this function
    916 * ensures the first ordered extent gets returned.
    917 */
    918struct btrfs_ordered_extent *btrfs_lookup_first_ordered_range(
    919			struct btrfs_inode *inode, u64 file_offset, u64 len)
    920{
    921	struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
    922	struct rb_node *node;
    923	struct rb_node *cur;
    924	struct rb_node *prev;
    925	struct rb_node *next;
    926	struct btrfs_ordered_extent *entry = NULL;
    927
    928	spin_lock_irq(&tree->lock);
    929	node = tree->tree.rb_node;
    930	/*
    931	 * Here we don't want to use tree_search() which will use tree->last
    932	 * and screw up the search order.
    933	 * And __tree_search() can't return the adjacent ordered extents
    934	 * either, thus here we do our own search.
    935	 */
    936	while (node) {
    937		entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
    938
    939		if (file_offset < entry->file_offset) {
    940			node = node->rb_left;
    941		} else if (file_offset >= entry_end(entry)) {
    942			node = node->rb_right;
    943		} else {
    944			/*
    945			 * Direct hit, got an ordered extent that starts at
    946			 * @file_offset
    947			 */
    948			goto out;
    949		}
    950	}
    951	if (!entry) {
    952		/* Empty tree */
    953		goto out;
    954	}
    955
    956	cur = &entry->rb_node;
    957	/* We got an entry around @file_offset, check adjacent entries */
    958	if (entry->file_offset < file_offset) {
    959		prev = cur;
    960		next = rb_next(cur);
    961	} else {
    962		prev = rb_prev(cur);
    963		next = cur;
    964	}
    965	if (prev) {
    966		entry = rb_entry(prev, struct btrfs_ordered_extent, rb_node);
    967		if (range_overlaps(entry, file_offset, len))
    968			goto out;
    969	}
    970	if (next) {
    971		entry = rb_entry(next, struct btrfs_ordered_extent, rb_node);
    972		if (range_overlaps(entry, file_offset, len))
    973			goto out;
    974	}
    975	/* No ordered extent in the range */
    976	entry = NULL;
    977out:
    978	if (entry)
    979		refcount_inc(&entry->refs);
    980	spin_unlock_irq(&tree->lock);
    981	return entry;
    982}
    983
    984/*
    985 * btrfs_flush_ordered_range - Lock the passed range and ensures all pending
    986 * ordered extents in it are run to completion.
    987 *
    988 * @inode:        Inode whose ordered tree is to be searched
    989 * @start:        Beginning of range to flush
    990 * @end:          Last byte of range to lock
    991 * @cached_state: If passed, will return the extent state responsible for the
    992 * locked range. It's the caller's responsibility to free the cached state.
    993 *
    994 * This function always returns with the given range locked, ensuring after it's
    995 * called no order extent can be pending.
    996 */
    997void btrfs_lock_and_flush_ordered_range(struct btrfs_inode *inode, u64 start,
    998					u64 end,
    999					struct extent_state **cached_state)
   1000{
   1001	struct btrfs_ordered_extent *ordered;
   1002	struct extent_state *cache = NULL;
   1003	struct extent_state **cachedp = &cache;
   1004
   1005	if (cached_state)
   1006		cachedp = cached_state;
   1007
   1008	while (1) {
   1009		lock_extent_bits(&inode->io_tree, start, end, cachedp);
   1010		ordered = btrfs_lookup_ordered_range(inode, start,
   1011						     end - start + 1);
   1012		if (!ordered) {
   1013			/*
   1014			 * If no external cached_state has been passed then
   1015			 * decrement the extra ref taken for cachedp since we
   1016			 * aren't exposing it outside of this function
   1017			 */
   1018			if (!cached_state)
   1019				refcount_dec(&cache->refs);
   1020			break;
   1021		}
   1022		unlock_extent_cached(&inode->io_tree, start, end, cachedp);
   1023		btrfs_start_ordered_extent(ordered, 1);
   1024		btrfs_put_ordered_extent(ordered);
   1025	}
   1026}
   1027
   1028static int clone_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pos,
   1029				u64 len)
   1030{
   1031	struct inode *inode = ordered->inode;
   1032	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
   1033	u64 file_offset = ordered->file_offset + pos;
   1034	u64 disk_bytenr = ordered->disk_bytenr + pos;
   1035	unsigned long flags = ordered->flags & BTRFS_ORDERED_TYPE_FLAGS;
   1036
   1037	/*
   1038	 * The splitting extent is already counted and will be added again in
   1039	 * btrfs_add_ordered_extent_*(). Subtract len to avoid double counting.
   1040	 */
   1041	percpu_counter_add_batch(&fs_info->ordered_bytes, -len,
   1042				 fs_info->delalloc_batch);
   1043	WARN_ON_ONCE(flags & (1 << BTRFS_ORDERED_COMPRESSED));
   1044	return btrfs_add_ordered_extent(BTRFS_I(inode), file_offset, len, len,
   1045					disk_bytenr, len, 0, flags,
   1046					ordered->compress_type);
   1047}
   1048
   1049int btrfs_split_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pre,
   1050				u64 post)
   1051{
   1052	struct inode *inode = ordered->inode;
   1053	struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
   1054	struct rb_node *node;
   1055	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
   1056	int ret = 0;
   1057
   1058	spin_lock_irq(&tree->lock);
   1059	/* Remove from tree once */
   1060	node = &ordered->rb_node;
   1061	rb_erase(node, &tree->tree);
   1062	RB_CLEAR_NODE(node);
   1063	if (tree->last == node)
   1064		tree->last = NULL;
   1065
   1066	ordered->file_offset += pre;
   1067	ordered->disk_bytenr += pre;
   1068	ordered->num_bytes -= (pre + post);
   1069	ordered->disk_num_bytes -= (pre + post);
   1070	ordered->bytes_left -= (pre + post);
   1071
   1072	/* Re-insert the node */
   1073	node = tree_insert(&tree->tree, ordered->file_offset, &ordered->rb_node);
   1074	if (node)
   1075		btrfs_panic(fs_info, -EEXIST,
   1076			"zoned: inconsistency in ordered tree at offset %llu",
   1077			    ordered->file_offset);
   1078
   1079	spin_unlock_irq(&tree->lock);
   1080
   1081	if (pre)
   1082		ret = clone_ordered_extent(ordered, 0, pre);
   1083	if (ret == 0 && post)
   1084		ret = clone_ordered_extent(ordered, pre + ordered->disk_num_bytes,
   1085					   post);
   1086
   1087	return ret;
   1088}
   1089
   1090int __init ordered_data_init(void)
   1091{
   1092	btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
   1093				     sizeof(struct btrfs_ordered_extent), 0,
   1094				     SLAB_MEM_SPREAD,
   1095				     NULL);
   1096	if (!btrfs_ordered_extent_cache)
   1097		return -ENOMEM;
   1098
   1099	return 0;
   1100}
   1101
   1102void __cold ordered_data_exit(void)
   1103{
   1104	kmem_cache_destroy(btrfs_ordered_extent_cache);
   1105}