Received-SPF: pass (google.com: domain of linux-ext4-owner@vger.kernel.org designates 2620:137:e000::3:1 as permitted sender) client-ip=2620:137:e000::3:1;
From:   "Lu, Davina" <davinalu@amazon.com>
To:     Theodore Ts'o <tytso@mit.edu>,
        "Kiselev, Oleg" <okiselev@amazon.com>,
        hazem ahmed mohamed <hazem.ahmed.abuelfotoh@gmail.com>
CC:     "linux-ext4@vger.kernel.org" <linux-ext4@vger.kernel.org>
Subject: [PATCH 1/1] ext4: remove the bottleneck of ext4-rsv-conversion work
 queue
Thread-Topic: [PATCH 1/1] ext4: remove the bottleneck of ext4-rsv-conversion
 work queue
Thread-Index: AdoBRuwfXYXJC+v+S0ufPaUWI5yXuQ==
Date:   Tue, 17 Oct 2023 22:11:50 +0000
Message-ID: <18c6a1302c9a4a94a44e7262b97c8750@amazon.com>
Accept-Language: en-US
Content-Language: en-US
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: quoted-printable
MIME-Version: 1.0
Precedence: bulk

=20

When dioread_nolock and delay_alloc are both enaled, the
bio_endio() will trigger ext4-rsv-conversion work queue to do
ext4_do_flush_completed_IO(). The current work queue is
one-by-one updating for EXT4_IO_END_UNWRITTEN extend block at
io_end->list_vec which added by ext4_writepages().
So if the BIO has high performance, and only one thread to do
EXT4 flush will be an bottleneck. So we simple allow more thread
and with a semaphore protection, since the "ext4-rsv-conversion"
this workqueue is only for updating the EXT4_IO_END_UNWRITTEN
extend block(only exist on dioread_unlock and delay_alloc options
are set).
---
 fs/ext4/ext4.h    | 1 +
 fs/ext4/extents.c | 2 ++
 fs/ext4/super.c   | 3 ++-
 3 files changed, 5 insertions(+), 1 deletion(-)

diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 0a2d55faa095..15d8d7a1810e 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -1134,6 +1134,7 @@ struct ext4_inode_info {
 	atomic_t i_unwritten; /* Nr. of inflight conversions pending */
=20
 	spinlock_t i_block_reservation_lock;
+	struct rw_semaphore i_rsv_unwritten_sem;
=20
 	/*
 	 * Transactions that contain inode's metadata needed to complete
diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
index e4115d338f10..dbd3f69853cf 100644
--- a/fs/ext4/extents.c
+++ b/fs/ext4/extents.c
@@ -4807,6 +4807,7 @@ int ext4_convert_unwritten_extents(handle_t *handle, =
struct inode *inode,
 				break;
 			}
 		}
+		down_write(&EXT4_I(inode)->i_rsv_unwritten_sem);
 		ret =3D ext4_map_blocks(handle, inode, &map,
 				      EXT4_GET_BLOCKS_IO_CONVERT_EXT);
 		if (ret <=3D 0)
@@ -4815,6 +4816,7 @@ int ext4_convert_unwritten_extents(handle_t *handle, =
struct inode *inode,
 				     "ext4_ext_map_blocks returned %d",
 				     inode->i_ino, map.m_lblk,
 				     map.m_len, ret);
+		up_write(&EXT4_I(inode)->i_rsv_unwritten_sem);
 		ret2 =3D ext4_mark_inode_dirty(handle, inode);
 		if (credits) {
 			ret3 =3D ext4_journal_stop(handle);
diff --git a/fs/ext4/super.c b/fs/ext4/super.c
index c94ebf704616..af2af5173424 100644
--- a/fs/ext4/super.c
+++ b/fs/ext4/super.c
@@ -1466,6 +1466,7 @@ static void init_once(void *foo)
 	INIT_LIST_HEAD(&ei->i_orphan);
 	init_rwsem(&ei->xattr_sem);
 	init_rwsem(&ei->i_data_sem);
+	init_rwsem(&ei->i_rsv_unwritten_sem);
 	inode_init_once(&ei->vfs_inode);
 	ext4_fc_init_inode(&ei->vfs_inode);
 }
@@ -5452,7 +5453,7 @@ static int __ext4_fill_super(struct fs_context *fc, s=
truct super_block *sb)
 	 * concurrency isn't really necessary.  Limit it to 1.
 	 */
 	EXT4_SB(sb)->rsv_conversion_wq =3D
-		alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
+		alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
 	if (!EXT4_SB(sb)->rsv_conversion_wq) {
 		printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
 		err =3D -ENOMEM;
--=20
2.40.1