btrfs: make fast fsyncs wait only for writeback

Currently regardless of a full or a fast fsync we always wait for ordered
extents to complete, and then start logging the inode after that. However
for fast fsyncs we can just wait for the writeback to complete, we don't
need to wait for the ordered extents to complete since we use the list of
modified extents maps to figure out which extents we must log and we can
get their checksums directly from the ordered extents that are still in
flight, otherwise look them up from the checksums tree.

Until commit b5e6c3e170b770 ("btrfs: always wait on ordered extents at
fsync time"), for fast fsyncs, we used to start logging without even
waiting for the writeback to complete first, we would wait for it to
complete after logging, while holding a transaction open, which lead to
performance issues when using cgroups and probably for other cases too,
as wait for IO while holding a transaction handle should be avoided as
much as possible. After that, for fast fsyncs, we started to wait for
ordered extents to complete before starting to log, which adds some
latency to fsyncs and we even got at least one report about a performance
drop which bisected to that particular change:

https://lore.kernel.org/linux-btrfs/20181109215148.GF23260@techsingularity.net/

This change makes fast fsyncs only wait for writeback to finish before
starting to log the inode, instead of waiting for both the writeback to
finish and for the ordered extents to complete. This brings back part of
the logic we had that extracts checksums from in flight ordered extents,
which are not yet in the checksums tree, and making sure transaction
commits wait for the completion of ordered extents previously logged
(by far most of the time they have already completed by the time a
transaction commit starts, resulting in no wait at all), to avoid any
data loss if an ordered extent completes after the transaction used to
log an inode is committed, followed by a power failure.

When there are no other tasks accessing the checksums and the subvolume
btrees, the ordered extent completion is pretty fast, typically taking
100 to 200 microseconds only in my observations. However when there are
other tasks accessing these btrees, ordered extent completion can take a
lot more time due to lock contention on nodes and leaves of these btrees.
I've seen cases over 2 milliseconds, which starts to be significant. In
particular when we do have concurrent fsyncs against different files there
is a lot of contention on the checksums btree, since we have many tasks
writing the checksums into the btree and other tasks that already started
the logging phase are doing lookups for checksums in the btree.

This change also turns all ranged fsyncs into full ranged fsyncs, which
is something we already did when not using the NO_HOLES features or when
doing a full fsync. This is to guarantee we never miss checksums due to
writeback having been triggered only for a part of an extent, and we end
up logging the full extent but only checksums for the written range, which
results in missing checksums after log replay. Allowing ranged fsyncs to
operate again only in the original range, when using the NO_HOLES feature
and doing a fast fsync is doable but requires some non trivial changes to
the writeback path, which can always be worked on later if needed, but I
don't think they are a very common use case.

Several tests were performed using fio for different numbers of concurrent
jobs, each writing and fsyncing its own file, for both sequential and
random file writes. The tests were run on bare metal, no virtualization,
on a box with 12 cores (Intel i7-8700), 64Gb of RAM and a NVMe device,
with a kernel configuration that is the default of typical distributions
(debian in this case), without debug options enabled (kasan, kmemleak,
slub debug, debug of page allocations, lock debugging, etc).

The following script that calls fio was used:

  $ cat test-fsync.sh
  #!/bin/bash

  DEV=/dev/nvme0n1
  MNT=/mnt/btrfs
  MOUNT_OPTIONS="-o ssd -o space_cache=v2"
  MKFS_OPTIONS="-d single -m single"

  if [ $# -ne 5 ]; then
    echo "Use $0 NUM_JOBS FILE_SIZE FSYNC_FREQ BLOCK_SIZE [write|randwrite]"
    exit 1
  fi

  NUM_JOBS=$1
  FILE_SIZE=$2
  FSYNC_FREQ=$3
  BLOCK_SIZE=$4
  WRITE_MODE=$5

  if [ "$WRITE_MODE" != "write" ] && [ "$WRITE_MODE" != "randwrite" ]; then
    echo "Invalid WRITE_MODE, must be 'write' or 'randwrite'"
    exit 1
  fi

  cat <<EOF > /tmp/fio-job.ini
  [writers]
  rw=$WRITE_MODE
  fsync=$FSYNC_FREQ
  fallocate=none
  group_reporting=1
  direct=0
  bs=$BLOCK_SIZE
  ioengine=sync
  size=$FILE_SIZE
  directory=$MNT
  numjobs=$NUM_JOBS
  EOF

  echo "performance" | tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor

  echo
  echo "Using config:"
  echo
  cat /tmp/fio-job.ini
  echo

  umount $MNT &> /dev/null
  mkfs.btrfs -f $MKFS_OPTIONS $DEV
  mount $MOUNT_OPTIONS $DEV $MNT
  fio /tmp/fio-job.ini
  umount $MNT

The results were the following:

*************************
*** sequential writes ***
*************************

==== 1 job, 8GiB file, fsync frequency 1, block size 64KiB ====

Before patch:

WRITE: bw=36.6MiB/s (38.4MB/s), 36.6MiB/s-36.6MiB/s (38.4MB/s-38.4MB/s), io=8192MiB (8590MB), run=223689-223689msec

After patch:

WRITE: bw=40.2MiB/s (42.1MB/s), 40.2MiB/s-40.2MiB/s (42.1MB/s-42.1MB/s), io=8192MiB (8590MB), run=203980-203980msec
(+9.8%, -8.8% runtime)

==== 2 jobs, 4GiB files, fsync frequency 1, block size 64KiB ====

Before patch:

WRITE: bw=35.8MiB/s (37.5MB/s), 35.8MiB/s-35.8MiB/s (37.5MB/s-37.5MB/s), io=8192MiB (8590MB), run=228950-228950msec

After patch:

WRITE: bw=43.5MiB/s (45.6MB/s), 43.5MiB/s-43.5MiB/s (45.6MB/s-45.6MB/s), io=8192MiB (8590MB), run=188272-188272msec
(+21.5% throughput, -17.8% runtime)

==== 4 jobs, 2GiB files, fsync frequency 1, block size 64KiB ====

Before patch:

WRITE: bw=50.1MiB/s (52.6MB/s), 50.1MiB/s-50.1MiB/s (52.6MB/s-52.6MB/s), io=8192MiB (8590MB), run=163446-163446msec

After patch:

WRITE: bw=64.5MiB/s (67.6MB/s), 64.5MiB/s-64.5MiB/s (67.6MB/s-67.6MB/s), io=8192MiB (8590MB), run=126987-126987msec
(+28.7% throughput, -22.3% runtime)

==== 8 jobs, 1GiB files, fsync frequency 1, block size 64KiB ====

Before patch:

WRITE: bw=64.0MiB/s (68.1MB/s), 64.0MiB/s-64.0MiB/s (68.1MB/s-68.1MB/s), io=8192MiB (8590MB), run=126075-126075msec

After patch:

WRITE: bw=86.8MiB/s (91.0MB/s), 86.8MiB/s-86.8MiB/s (91.0MB/s-91.0MB/s), io=8192MiB (8590MB), run=94358-94358msec
(+35.6% throughput, -25.2% runtime)

==== 16 jobs, 512MiB files, fsync frequency 1, block size 64KiB ====

Before patch:

WRITE: bw=79.8MiB/s (83.6MB/s), 79.8MiB/s-79.8MiB/s (83.6MB/s-83.6MB/s), io=8192MiB (8590MB), run=102694-102694msec

After patch:

WRITE: bw=107MiB/s (112MB/s), 107MiB/s-107MiB/s (112MB/s-112MB/s), io=8192MiB (8590MB), run=76446-76446msec
(+34.1% throughput, -25.6% runtime)

==== 32 jobs, 512MiB files, fsync frequency 1, block size 64KiB ====

Before patch:

WRITE: bw=93.2MiB/s (97.7MB/s), 93.2MiB/s-93.2MiB/s (97.7MB/s-97.7MB/s), io=16.0GiB (17.2GB), run=175836-175836msec

After patch:

WRITE: bw=111MiB/s (117MB/s), 111MiB/s-111MiB/s (117MB/s-117MB/s), io=16.0GiB (17.2GB), run=147001-147001msec
(+19.1% throughput, -16.4% runtime)

==== 64 jobs, 512MiB files, fsync frequency 1, block size 64KiB ====

Before patch:

WRITE: bw=108MiB/s (114MB/s), 108MiB/s-108MiB/s (114MB/s-114MB/s), io=32.0GiB (34.4GB), run=302656-302656msec

After patch:

WRITE: bw=133MiB/s (140MB/s), 133MiB/s-133MiB/s (140MB/s-140MB/s), io=32.0GiB (34.4GB), run=246003-246003msec
(+23.1% throughput, -18.7% runtime)

************************
***   random writes  ***
************************

==== 1 job, 8GiB file, fsync frequency 16, block size 4KiB ====

Before patch:

WRITE: bw=11.5MiB/s (12.0MB/s), 11.5MiB/s-11.5MiB/s (12.0MB/s-12.0MB/s), io=8192MiB (8590MB), run=714281-714281msec

After patch:

WRITE: bw=11.6MiB/s (12.2MB/s), 11.6MiB/s-11.6MiB/s (12.2MB/s-12.2MB/s), io=8192MiB (8590MB), run=705959-705959msec
(+0.9% throughput, -1.7% runtime)

==== 2 jobs, 4GiB files, fsync frequency 16, block size 4KiB ====

Before patch:

WRITE: bw=12.8MiB/s (13.5MB/s), 12.8MiB/s-12.8MiB/s (13.5MB/s-13.5MB/s), io=8192MiB (8590MB), run=638101-638101msec

After patch:

WRITE: bw=13.1MiB/s (13.7MB/s), 13.1MiB/s-13.1MiB/s (13.7MB/s-13.7MB/s), io=8192MiB (8590MB), run=625374-625374msec
(+2.3% throughput, -2.0% runtime)

==== 4 jobs, 2GiB files, fsync frequency 16, block size 4KiB ====

Before patch:

WRITE: bw=15.4MiB/s (16.2MB/s), 15.4MiB/s-15.4MiB/s (16.2MB/s-16.2MB/s), io=8192MiB (8590MB), run=531146-531146msec

After patch:

WRITE: bw=17.8MiB/s (18.7MB/s), 17.8MiB/s-17.8MiB/s (18.7MB/s-18.7MB/s), io=8192MiB (8590MB), run=460431-460431msec
(+15.6% throughput, -13.3% runtime)

==== 8 jobs, 1GiB files, fsync frequency 16, block size 4KiB ====

Before patch:

WRITE: bw=19.9MiB/s (20.8MB/s), 19.9MiB/s-19.9MiB/s (20.8MB/s-20.8MB/s), io=8192MiB (8590MB), run=412664-412664msec

After patch:

WRITE: bw=22.2MiB/s (23.3MB/s), 22.2MiB/s-22.2MiB/s (23.3MB/s-23.3MB/s), io=8192MiB (8590MB), run=368589-368589msec
(+11.6% throughput, -10.7% runtime)

==== 16 jobs, 512MiB files, fsync frequency 16, block size 4KiB ====

Before patch:

WRITE: bw=29.3MiB/s (30.7MB/s), 29.3MiB/s-29.3MiB/s (30.7MB/s-30.7MB/s), io=8192MiB (8590MB), run=279924-279924msec

After patch:

WRITE: bw=30.4MiB/s (31.9MB/s), 30.4MiB/s-30.4MiB/s (31.9MB/s-31.9MB/s), io=8192MiB (8590MB), run=269258-269258msec
(+3.8% throughput, -3.8% runtime)

==== 32 jobs, 512MiB files, fsync frequency 16, block size 4KiB ====

Before patch:

WRITE: bw=36.9MiB/s (38.7MB/s), 36.9MiB/s-36.9MiB/s (38.7MB/s-38.7MB/s), io=16.0GiB (17.2GB), run=443581-443581msec

After patch:

WRITE: bw=41.6MiB/s (43.6MB/s), 41.6MiB/s-41.6MiB/s (43.6MB/s-43.6MB/s), io=16.0GiB (17.2GB), run=394114-394114msec
(+12.7% throughput, -11.2% runtime)

==== 64 jobs, 512MiB files, fsync frequency 16, block size 4KiB ====

Before patch:

WRITE: bw=45.9MiB/s (48.1MB/s), 45.9MiB/s-45.9MiB/s (48.1MB/s-48.1MB/s), io=32.0GiB (34.4GB), run=714614-714614msec

After patch:

WRITE: bw=48.8MiB/s (51.1MB/s), 48.8MiB/s-48.8MiB/s (51.1MB/s-51.1MB/s), io=32.0GiB (34.4GB), run=672087-672087msec
(+6.3% throughput, -6.0% runtime)

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

8 files changed, 277 insertions, 106 deletions

diff --git a/fs/btrfs/btrfs_inode.h b/fs/btrfs/btrfs_inode.h
index c47b6c6fea9f..00f7831d0902 100644
--- a/fs/btrfs/btrfs_inode.h
+++ b/fs/btrfs/btrfs_inode.h
@@ -25,6 +25,11 @@ enum {
 	BTRFS_INODE_DUMMY,
 	BTRFS_INODE_IN_DEFRAG,
 	BTRFS_INODE_HAS_ASYNC_EXTENT,
+	 /*
+	  * Always set under the VFS' inode lock, otherwise it can cause races
+	  * during fsync (we start as a fast fsync and then end up in a full
+	  * fsync racing with ordered extent completion).
+	  */
 	BTRFS_INODE_NEEDS_FULL_SYNC,
 	BTRFS_INODE_COPY_EVERYTHING,
 	BTRFS_INODE_IN_DELALLOC_LIST,
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index de14ed402390..b859d5ae7f80 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -2114,20 +2114,24 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
 	struct btrfs_trans_handle *trans;
 	struct btrfs_log_ctx ctx;
 	int ret = 0, err;
+	u64 len;
+	bool full_sync;
 
 	trace_btrfs_sync_file(file, datasync);
 
 	btrfs_init_log_ctx(&ctx, inode);
 
 	/*
-	 * Set the range to full if the NO_HOLES feature is not enabled.
-	 * This is to avoid missing file extent items representing holes after
-	 * replaying the log.
+	 * Always set the range to a full range, otherwise we can get into
+	 * several problems, from missing file extent items to represent holes
+	 * when not using the NO_HOLES feature, to log tree corruption due to
+	 * races between hole detection during logging and completion of ordered
+	 * extents outside the range, to missing checksums due to ordered extents
+	 * for which we flushed only a subset of their pages.
 	 */
-	if (!btrfs_fs_incompat(fs_info, NO_HOLES)) {
-		start = 0;
-		end = LLONG_MAX;
-	}
+	start = 0;
+	end = LLONG_MAX;
+	len = (u64)LLONG_MAX + 1;
 
 	/*
 	 * We write the dirty pages in the range and wait until they complete
@@ -2151,19 +2155,12 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
 	atomic_inc(&root->log_batch);
 
 	/*
-	 * If the inode needs a full sync, make sure we use a full range to
-	 * avoid log tree corruption, due to hole detection racing with ordered
-	 * extent completion for adjacent ranges and races between logging and
-	 * completion of ordered extents for adjancent ranges - both races
-	 * could lead to file extent items in the log with overlapping ranges.
-	 * Do this while holding the inode lock, to avoid races with other
-	 * tasks.
+	 * Always check for the full sync flag while holding the inode's lock,
+	 * to avoid races with other tasks. The flag must be either set all the
+	 * time during logging or always off all the time while logging.
 	 */
-	if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
-		     &BTRFS_I(inode)->runtime_flags)) {
-		start = 0;
-		end = LLONG_MAX;
-	}
+	full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
+			     &BTRFS_I(inode)->runtime_flags);
 
 	/*
 	 * Before we acquired the inode's lock, someone may have dirtied more
@@ -2194,20 +2191,42 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
 	 * We have to do this here to avoid the priority inversion of waiting on
 	 * IO of a lower priority task while holding a transaction open.
 	 *
-	 * Also, the range length can be represented by u64, we have to do the
-	 * typecasts to avoid signed overflow if it's [0, LLONG_MAX].
+	 * For a full fsync we wait for the ordered extents to complete while
+	 * for a fast fsync we wait just for writeback to complete, and then
+	 * attach the ordered extents to the transaction so that a transaction
+	 * commit waits for their completion, to avoid data loss if we fsync,
+	 * the current transaction commits before the ordered extents complete
+	 * and a power failure happens right after that.
 	 */
-	ret = btrfs_wait_ordered_range(inode, start, (u64)end - (u64)start + 1);
-	if (ret) {
-		up_write(&BTRFS_I(inode)->dio_sem);
-		inode_unlock(inode);
-		goto out;
+	if (full_sync) {
+		ret = btrfs_wait_ordered_range(inode, start, len);
+	} else {
+		/*
+		 * Get our ordered extents as soon as possible to avoid doing
+		 * checksum lookups in the csum tree, and use instead the
+		 * checksums attached to the ordered extents.
+		 */
+		btrfs_get_ordered_extents_for_logging(BTRFS_I(inode),
+						      &ctx.ordered_extents);
+		ret = filemap_fdatawait_range(inode->i_mapping, start, end);
 	}
+
+	if (ret)
+		goto out_release_extents;
+
 	atomic_inc(&root->log_batch);
 
+	/*
+	 * If we are doing a fast fsync we can not bail out if the inode's
+	 * last_trans is <= then the last committed transaction, because we only
+	 * update the last_trans of the inode during ordered extent completion,
+	 * and for a fast fsync we don't wait for that, we only wait for the
+	 * writeback to complete.
+	 */
 	smp_mb();
 	if (btrfs_inode_in_log(BTRFS_I(inode), fs_info->generation) ||
-	    BTRFS_I(inode)->last_trans <= fs_info->last_trans_committed) {
+	    (BTRFS_I(inode)->last_trans <= fs_info->last_trans_committed &&
+	     (full_sync || list_empty(&ctx.ordered_extents)))) {
 		/*
 		 * We've had everything committed since the last time we were
 		 * modified so clear this flag in case it was set for whatever
@@ -2223,9 +2242,7 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
 		 * checked called fsync.
 		 */
 		ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err);
-		up_write(&BTRFS_I(inode)->dio_sem);
-		inode_unlock(inode);
-		goto out;
+		goto out_release_extents;
 	}
 
 	/*
@@ -2242,12 +2259,11 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
 	trans = btrfs_start_transaction(root, 0);
 	if (IS_ERR(trans)) {
 		ret = PTR_ERR(trans);
-		up_write(&BTRFS_I(inode)->dio_sem);
-		inode_unlock(inode);
-		goto out;
+		goto out_release_extents;
 	}
 
-	ret = btrfs_log_dentry_safe(trans, dentry, start, end, &ctx);
+	ret = btrfs_log_dentry_safe(trans, dentry, &ctx);
+	btrfs_release_log_ctx_extents(&ctx);
 	if (ret < 0) {
 		/* Fallthrough and commit/free transaction. */
 		ret = 1;
@@ -2274,6 +2290,13 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
 				goto out;
 			}
 		}
+		if (!full_sync) {
+			ret = btrfs_wait_ordered_range(inode, start, len);
+			if (ret) {
+				btrfs_end_transaction(trans);
+				goto out;
+			}
+		}
 		ret = btrfs_commit_transaction(trans);
 	} else {
 		ret = btrfs_end_transaction(trans);
@@ -2284,6 +2307,12 @@ out:
 	if (!ret)
 		ret = err;
 	return ret > 0 ? -EIO : ret;
+
+out_release_extents:
+	btrfs_release_log_ctx_extents(&ctx);
+	up_write(&BTRFS_I(inode)->dio_sem);
+	inode_unlock(inode);
+	goto out;
 }
 
 static const struct vm_operations_struct btrfs_file_vm_ops = {
diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c
index ebac13389e7e..4732c5b89460 100644
--- a/fs/btrfs/ordered-data.c
+++ b/fs/btrfs/ordered-data.c
@@ -212,6 +212,7 @@ static int __btrfs_add_ordered_extent(struct btrfs_inode *inode, u64 file_offset
 	refcount_set(&entry->refs, 1);
 	init_waitqueue_head(&entry->wait);
 	INIT_LIST_HEAD(&entry->list);
+	INIT_LIST_HEAD(&entry->log_list);
 	INIT_LIST_HEAD(&entry->root_extent_list);
 	INIT_LIST_HEAD(&entry->work_list);
 	init_completion(&entry->completion);
@@ -445,6 +446,7 @@ void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
 
 	if (refcount_dec_and_test(&entry->refs)) {
 		ASSERT(list_empty(&entry->root_extent_list));
+		ASSERT(list_empty(&entry->log_list));
 		ASSERT(RB_EMPTY_NODE(&entry->rb_node));
 		if (entry->inode)
 			btrfs_add_delayed_iput(entry->inode);
@@ -470,6 +472,7 @@ void btrfs_remove_ordered_extent(struct inode *inode,
 	struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
 	struct btrfs_root *root = btrfs_inode->root;
 	struct rb_node *node;
+	bool pending;
 
 	/* This is paired with btrfs_add_ordered_extent. */
 	spin_lock(&btrfs_inode->lock);
@@ -491,8 +494,36 @@ void btrfs_remove_ordered_extent(struct inode *inode,
 	if (tree->last == node)
 		tree->last = NULL;
 	set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
+	pending = test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags);
 	spin_unlock_irq(&tree->lock);
 
+	/*
+	 * The current running transaction is waiting on us, we need to let it
+	 * know that we're complete and wake it up.
+	 */
+	if (pending) {
+		struct btrfs_transaction *trans;
+
+		/*
+		 * The checks for trans are just a formality, it should be set,
+		 * but if it isn't we don't want to deref/assert under the spin
+		 * lock, so be nice and check if trans is set, but ASSERT() so
+		 * if it isn't set a developer will notice.
+		 */
+		spin_lock(&fs_info->trans_lock);
+		trans = fs_info->running_transaction;
+		if (trans)
+			refcount_inc(&trans->use_count);
+		spin_unlock(&fs_info->trans_lock);
+
+		ASSERT(trans);
+		if (trans) {
+			if (atomic_dec_and_test(&trans->pending_ordered))
+				wake_up(&trans->pending_wait);
+			btrfs_put_transaction(trans);
+		}
+	}
+
 	spin_lock(&root->ordered_extent_lock);
 	list_del_init(&entry->root_extent_list);
 	root->nr_ordered_extents--;
@@ -775,6 +806,34 @@ out:
 }
 
 /*
+ * Adds all ordered extents to the given list. The list ends up sorted by the
+ * file_offset of the ordered extents.
+ */
+void btrfs_get_ordered_extents_for_logging(struct btrfs_inode *inode,
+					   struct list_head *list)
+{
+	struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
+	struct rb_node *n;
+
+	ASSERT(inode_is_locked(&inode->vfs_inode));
+
+	spin_lock_irq(&tree->lock);
+	for (n = rb_first(&tree->tree); n; n = rb_next(n)) {
+		struct btrfs_ordered_extent *ordered;
+
+		ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
+
+		if (test_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
+			continue;
+
+		ASSERT(list_empty(&ordered->log_list));
+		list_add_tail(&ordered->log_list, list);
+		refcount_inc(&ordered->refs);
+	}
+	spin_unlock_irq(&tree->lock);
+}
+
+/*
  * lookup and return any extent before 'file_offset'.  NULL is returned
  * if none is found
  */
diff --git a/fs/btrfs/ordered-data.h b/fs/btrfs/ordered-data.h
index d61ea9c880a3..644258a7dfb1 100644
--- a/fs/btrfs/ordered-data.h
+++ b/fs/btrfs/ordered-data.h
@@ -56,6 +56,12 @@ enum {
 	BTRFS_ORDERED_TRUNCATED,
 	/* Regular IO for COW */
 	BTRFS_ORDERED_REGULAR,
+	/* Used during fsync to track already logged extents */
+	BTRFS_ORDERED_LOGGED,
+	/* We have already logged all the csums of the ordered extent */
+	BTRFS_ORDERED_LOGGED_CSUM,
+	/* We wait for this extent to complete in the current transaction */
+	BTRFS_ORDERED_PENDING,
 };
 
 struct btrfs_ordered_extent {
@@ -104,6 +110,9 @@ struct btrfs_ordered_extent {
 	/* list of checksums for insertion when the extent io is done */
 	struct list_head list;
 
+	/* used for fast fsyncs */
+	struct list_head log_list;
+
 	/* used to wait for the BTRFS_ORDERED_COMPLETE bit */
 	wait_queue_head_t wait;
 
@@ -174,6 +183,8 @@ struct btrfs_ordered_extent *btrfs_lookup_ordered_range(
 		struct btrfs_inode *inode,
 		u64 file_offset,
 		u64 len);
+void btrfs_get_ordered_extents_for_logging(struct btrfs_inode *inode,
+					   struct list_head *list);
 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
 			   u8 *sum, int len);
 u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr,
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index d2fc292ac61b..f8265dbec9c3 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -292,6 +292,8 @@ loop:
 	}
 
 	cur_trans->fs_info = fs_info;
+	atomic_set(&cur_trans->pending_ordered, 0);
+	init_waitqueue_head(&cur_trans->pending_wait);
 	atomic_set(&cur_trans->num_writers, 1);
 	extwriter_counter_init(cur_trans, type);
 	init_waitqueue_head(&cur_trans->writer_wait);
@@ -2165,6 +2167,14 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 
 	btrfs_wait_delalloc_flush(trans);
 
+	/*
+	 * Wait for all ordered extents started by a fast fsync that joined this
+	 * transaction. Otherwise if this transaction commits before the ordered
+	 * extents complete we lose logged data after a power failure.
+	 */
+	wait_event(cur_trans->pending_wait,
+		   atomic_read(&cur_trans->pending_ordered) == 0);
+
 	btrfs_scrub_pause(fs_info);
 	/*
 	 * Ok now we need to make sure to block out any other joins while we
diff --git a/fs/btrfs/transaction.h b/fs/btrfs/transaction.h
index d60b055b8695..8241c050ba71 100644
--- a/fs/btrfs/transaction.h
+++ b/fs/btrfs/transaction.h
@@ -85,6 +85,13 @@ struct btrfs_transaction {
 	spinlock_t dropped_roots_lock;
 	struct btrfs_delayed_ref_root delayed_refs;
 	struct btrfs_fs_info *fs_info;
+
+	/*
+	 * Number of ordered extents the transaction must wait for before
+	 * committing. These are ordered extents started by a fast fsync.
+	 */
+	atomic_t pending_ordered;
+	wait_queue_head_t pending_wait;
 };
 
 #define __TRANS_FREEZABLE	(1U << 0)
diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c
index e86fe7bbba82..e35cfe8cf68b 100644
--- a/fs/btrfs/tree-log.c
+++ b/fs/btrfs/tree-log.c
@@ -96,8 +96,6 @@ enum {
 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
 			   struct btrfs_root *root, struct btrfs_inode *inode,
 			   int inode_only,
-			   const loff_t start,
-			   const loff_t end,
 			   struct btrfs_log_ctx *ctx);
 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
 			     struct btrfs_root *root,
@@ -4080,10 +4078,14 @@ static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
 static int log_extent_csums(struct btrfs_trans_handle *trans,
 			    struct btrfs_inode *inode,
 			    struct btrfs_root *log_root,
-			    const struct extent_map *em)
+			    const struct extent_map *em,
+			    struct btrfs_log_ctx *ctx)
 {
+	struct btrfs_ordered_extent *ordered;
 	u64 csum_offset;
 	u64 csum_len;
+	u64 mod_start = em->mod_start;
+	u64 mod_len = em->mod_len;
 	LIST_HEAD(ordered_sums);
 	int ret = 0;
 
@@ -4092,13 +4094,71 @@ static int log_extent_csums(struct btrfs_trans_handle *trans,
 	    em->block_start == EXTENT_MAP_HOLE)
 		return 0;
 
+	list_for_each_entry(ordered, &ctx->ordered_extents, log_list) {
+		const u64 ordered_end = ordered->file_offset + ordered->num_bytes;
+		const u64 mod_end = mod_start + mod_len;
+		struct btrfs_ordered_sum *sums;
+
+		if (mod_len == 0)
+			break;
+
+		if (ordered_end <= mod_start)
+			continue;
+		if (mod_end <= ordered->file_offset)
+			break;
+
+		/*
+		 * We are going to copy all the csums on this ordered extent, so
+		 * go ahead and adjust mod_start and mod_len in case this ordered
+		 * extent has already been logged.
+		 */
+		if (ordered->file_offset > mod_start) {
+			if (ordered_end >= mod_end)
+				mod_len = ordered->file_offset - mod_start;
+			/*
+			 * If we have this case
+			 *
+			 * |--------- logged extent ---------|
+			 *       |----- ordered extent ----|
+			 *
+			 * Just don't mess with mod_start and mod_len, we'll
+			 * just end up logging more csums than we need and it
+			 * will be ok.
+			 */
+		} else {
+			if (ordered_end < mod_end) {
+				mod_len = mod_end - ordered_end;
+				mod_start = ordered_end;
+			} else {
+				mod_len = 0;
+			}
+		}
+
+		/*
+		 * To keep us from looping for the above case of an ordered
+		 * extent that falls inside of the logged extent.
+		 */
+		if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags))
+			continue;
+
+		list_for_each_entry(sums, &ordered->list, list) {
+			ret = log_csums(trans, inode, log_root, sums);
+			if (ret)
+				return ret;
+		}
+	}
+
+	/* We're done, found all csums in the ordered extents. */
+	if (mod_len == 0)
+		return 0;
+
 	/* If we're compressed we have to save the entire range of csums. */
 	if (em->compress_type) {
 		csum_offset = 0;
 		csum_len = max(em->block_len, em->orig_block_len);
 	} else {
-		csum_offset = em->mod_start - em->start;
-		csum_len = em->mod_len;
+		csum_offset = mod_start - em->start;
+		csum_len = mod_len;
 	}
 
 	/* block start is already adjusted for the file extent offset. */
@@ -4138,7 +4198,7 @@ static int log_one_extent(struct btrfs_trans_handle *trans,
 	int ret;
 	int extent_inserted = 0;
 
-	ret = log_extent_csums(trans, inode, log, em);
+	ret = log_extent_csums(trans, inode, log, em, ctx);
 	if (ret)
 		return ret;
 
@@ -4340,10 +4400,10 @@ static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
 				     struct btrfs_root *root,
 				     struct btrfs_inode *inode,
 				     struct btrfs_path *path,
-				     struct btrfs_log_ctx *ctx,
-				     const u64 start,
-				     const u64 end)
+				     struct btrfs_log_ctx *ctx)
 {
+	struct btrfs_ordered_extent *ordered;
+	struct btrfs_ordered_extent *tmp;
 	struct extent_map *em, *n;
 	struct list_head extents;
 	struct extent_map_tree *tree = &inode->extent_tree;
@@ -4357,23 +4417,6 @@ static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
 	test_gen = root->fs_info->last_trans_committed;
 
 	list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
-		/*
-		 * Skip extents outside our logging range. It's important to do
-		 * it for correctness because if we don't ignore them, we may
-		 * log them before their ordered extent completes, and therefore
-		 * we could log them without logging their respective checksums
-		 * (the checksum items are added to the csum tree at the very
-		 * end of btrfs_finish_ordered_io()). Also leave such extents
-		 * outside of our range in the list, since we may have another
-		 * ranged fsync in the near future that needs them. If an extent
-		 * outside our range corresponds to a hole, log it to avoid
-		 * leaving gaps between extents (fsck will complain when we are
-		 * not using the NO_HOLES feature).
-		 */
-		if ((em->start > end || em->start + em->len <= start) &&
-		    em->block_start != EXTENT_MAP_HOLE)
-			continue;
-
 		list_del_init(&em->list);
 		/*
 		 * Just an arbitrary number, this can be really CPU intensive
@@ -4432,8 +4475,32 @@ process:
 	btrfs_release_path(path);
 	if (!ret)
 		ret = btrfs_log_prealloc_extents(trans, inode, path);
+	if (ret)
+		return ret;
 
-	return ret;
+	/*
+	 * We have logged all extents successfully, now make sure the commit of
+	 * the current transaction waits for the ordered extents to complete
+	 * before it commits and wipes out the log trees, otherwise we would
+	 * lose data if an ordered extents completes after the transaction
+	 * commits and a power failure happens after the transaction commit.
+	 */
+	list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
+		list_del_init(&ordered->log_list);
+		set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags);
+
+		if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
+			spin_lock_irq(&inode->ordered_tree.lock);
+			if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
+				set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
+				atomic_inc(&trans->transaction->pending_ordered);
+			}
+			spin_unlock_irq(&inode->ordered_tree.lock);
+		}
+		btrfs_put_ordered_extent(ordered);
+	}
+
+	return 0;
 }
 
 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
@@ -4839,7 +4906,7 @@ static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
 					ret = btrfs_log_inode(trans, root,
 						      BTRFS_I(inode),
 						      LOG_OTHER_INODE_ALL,
-						      0, LLONG_MAX, ctx);
+						      ctx);
 					btrfs_add_delayed_iput(inode);
 				}
 			}
@@ -4897,7 +4964,7 @@ static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
 		 * log with the new name before we unpin it.
 		 */
 		ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
-				      LOG_OTHER_INODE, 0, LLONG_MAX, ctx);
+				      LOG_OTHER_INODE, ctx);
 		if (ret) {
 			btrfs_add_delayed_iput(inode);
 			continue;
@@ -5110,8 +5177,6 @@ next_key:
 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
 			   struct btrfs_root *root, struct btrfs_inode *inode,
 			   int inode_only,
-			   const loff_t start,
-			   const loff_t end,
 			   struct btrfs_log_ctx *ctx)
 {
 	struct btrfs_path *path;
@@ -5290,7 +5355,7 @@ log_extents:
 	}
 	if (fast_search) {
 		ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
-						ctx, start, end);
+						ctx);
 		if (ret) {
 			err = ret;
 			goto out_unlock;
@@ -5299,31 +5364,8 @@ log_extents:
 		struct extent_map *em, *n;
 
 		write_lock(&em_tree->lock);
-		/*
-		 * We can't just remove every em if we're called for a ranged
-		 * fsync - that is, one that doesn't cover the whole possible
-		 * file range (0 to LLONG_MAX). This is because we can have
-		 * em's that fall outside the range we're logging and therefore
-		 * their ordered operations haven't completed yet
-		 * (btrfs_finish_ordered_io() not invoked yet). This means we
-		 * didn't get their respective file extent item in the fs/subvol
-		 * tree yet, and need to let the next fast fsync (one which
-		 * consults the list of modified extent maps) find the em so
-		 * that it logs a matching file extent item and waits for the
-		 * respective ordered operation to complete (if it's still
-		 * running).
-		 *
-		 * Removing every em outside the range we're logging would make
-		 * the next fast fsync not log their matching file extent items,
-		 * therefore making us lose data after a log replay.
-		 */
-		list_for_each_entry_safe(em, n, &em_tree->modified_extents,
-					 list) {
-			const u64 mod_end = em->mod_start + em->mod_len - 1;
-
-			if (em->mod_start >= start && mod_end <= end)
-				list_del_init(&em->list);
-		}
+		list_for_each_entry_safe(em, n, &em_tree->modified_extents, list)
+			list_del_init(&em->list);
 		write_unlock(&em_tree->lock);
 	}
 
@@ -5604,7 +5646,7 @@ process_leaf:
 			if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
 				log_mode = LOG_INODE_ALL;
 			ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
-					      log_mode, 0, LLONG_MAX, ctx);
+					      log_mode, ctx);
 			if (!ret &&
 			    btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
 				ret = 1;
@@ -5748,7 +5790,7 @@ static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
 			if (ctx)
 				ctx->log_new_dentries = false;
 			ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
-					      LOG_INODE_ALL, 0, LLONG_MAX, ctx);
+					      LOG_INODE_ALL, ctx);
 			if (!ret &&
 			    btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
 				ret = 1;
@@ -5799,8 +5841,7 @@ static int log_new_ancestors(struct btrfs_trans_handle *trans,
 
 		if (BTRFS_I(inode)->generation > last_committed)
 			ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
-					      LOG_INODE_EXISTS,
-					      0, LLONG_MAX, ctx);
+					      LOG_INODE_EXISTS, ctx);
 		btrfs_add_delayed_iput(inode);
 		if (ret)
 			return ret;
@@ -5855,7 +5896,7 @@ static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
 
 		if (inode->generation > fs_info->last_trans_committed) {
 			ret = btrfs_log_inode(trans, root, inode,
-					LOG_INODE_EXISTS, 0, LLONG_MAX, ctx);
+					      LOG_INODE_EXISTS, ctx);
 			if (ret)
 				break;
 		}
@@ -5963,8 +6004,6 @@ out:
 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
 				  struct btrfs_inode *inode,
 				  struct dentry *parent,
-				  const loff_t start,
-				  const loff_t end,
 				  int inode_only,
 				  struct btrfs_log_ctx *ctx)
 {
@@ -6017,7 +6056,7 @@ static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
 	if (ret)
 		goto end_no_trans;
 
-	ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
+	ret = btrfs_log_inode(trans, root, inode, inode_only, ctx);
 	if (ret)
 		goto end_trans;
 
@@ -6113,15 +6152,13 @@ end_no_trans:
  */
 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
 			  struct dentry *dentry,
-			  const loff_t start,
-			  const loff_t end,
 			  struct btrfs_log_ctx *ctx)
 {
 	struct dentry *parent = dget_parent(dentry);
 	int ret;
 
 	ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
-				     start, end, LOG_INODE_ALL, ctx);
+				     LOG_INODE_ALL, ctx);
 	dput(parent);
 
 	return ret;
@@ -6416,7 +6453,6 @@ void btrfs_log_new_name(struct btrfs_trans_handle *trans,
 	 * we don't need to worry about getting a log committed that has an
 	 * inconsistent state after a rename operation.
 	 */
-	btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
-			       LOG_INODE_EXISTS, &ctx);
+	btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx);
 }
 
diff --git a/fs/btrfs/tree-log.h b/fs/btrfs/tree-log.h
index ddfc6789d9bf..731bd9c029f5 100644
--- a/fs/btrfs/tree-log.h
+++ b/fs/btrfs/tree-log.h
@@ -19,6 +19,8 @@ struct btrfs_log_ctx {
 	bool logging_new_name;
 	struct inode *inode;
 	struct list_head list;
+	/* Only used for fast fsyncs. */
+	struct list_head ordered_extents;
 };
 
 static inline void btrfs_init_log_ctx(struct btrfs_log_ctx *ctx,
@@ -30,6 +32,20 @@ static inline void btrfs_init_log_ctx(struct btrfs_log_ctx *ctx,
 	ctx->logging_new_name = false;
 	ctx->inode = inode;
 	INIT_LIST_HEAD(&ctx->list);
+	INIT_LIST_HEAD(&ctx->ordered_extents);
+}
+
+static inline void btrfs_release_log_ctx_extents(struct btrfs_log_ctx *ctx)
+{
+	struct btrfs_ordered_extent *ordered;
+	struct btrfs_ordered_extent *tmp;
+
+	ASSERT(inode_is_locked(ctx->inode));
+
+	list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
+		list_del_init(&ordered->log_list);
+		btrfs_put_ordered_extent(ordered);
+	}
 }
 
 static inline void btrfs_set_log_full_commit(struct btrfs_trans_handle *trans)
@@ -51,8 +67,6 @@ int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
 int btrfs_recover_log_trees(struct btrfs_root *tree_root);
 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
 			  struct dentry *dentry,
-			  const loff_t start,
-			  const loff_t end,
 			  struct btrfs_log_ctx *ctx);
 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
 				 struct btrfs_root *root,