[PATCH 09/12] gfs2: hold i_rwsem until AIO completes

[Christoph Hellwig]

Switch gfs from the magic i_dio_count scheme to just hold i_rwsem
until the actual I/O has completed to reduce the locking complexity
and avoid nasty bugs due to missing inode_dio_wait calls.

Note that gfs only uses i_rwsem for direct I/O writes, not for
reads so no change to the read behavior.  It might also make sense
to use the same scheme for the gfs2 internal cluster lock.

Signed-off-by: Christoph Hellwig <hch@xxxxxx>
---
 fs/gfs2/bmap.c  |  2 --
 fs/gfs2/file.c  |  6 ++++--
 fs/gfs2/glops.c | 10 ++--------
 3 files changed, 6 insertions(+), 12 deletions(-)

diff --git a/fs/gfs2/bmap.c b/fs/gfs2/bmap.c
index 08f6fbb3655e..226f4eb680c7 100644
--- a/fs/gfs2/bmap.c
+++ b/fs/gfs2/bmap.c
@@ -2181,8 +2181,6 @@ int gfs2_setattr_size(struct inode *inode, u64 newsize)
 	if (ret)
 		return ret;
 
-	inode_dio_wait(inode);
-
 	ret = gfs2_rsqa_alloc(ip);
 	if (ret)
 		goto out;
diff --git a/fs/gfs2/file.c b/fs/gfs2/file.c
index 2260cb5d31af..82a2f313a3e6 100644
--- a/fs/gfs2/file.c
+++ b/fs/gfs2/file.c
@@ -806,7 +806,8 @@ static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
 	if (offset + len > i_size_read(&ip->i_inode))
 		goto out;
 
-	ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL, 0);
+	ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL,
+			   IOMAP_DIO_RWSEM_EXCL);
 
 out:
 	gfs2_glock_dq(&gh);
@@ -923,7 +924,8 @@ static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 	return generic_write_sync(iocb, ret);
 
 out_unlock:
-	inode_unlock(inode);
+	if (ret != -EIOCBQUEUED)
+		inode_unlock(inode);
 	return ret;
 }
 
diff --git a/fs/gfs2/glops.c b/fs/gfs2/glops.c
index 4ede1f18de85..a705eeb75117 100644
--- a/fs/gfs2/glops.c
+++ b/fs/gfs2/glops.c
@@ -243,11 +243,8 @@ static void inode_go_sync(struct gfs2_glock *gl)
 	struct address_space *metamapping = gfs2_glock2aspace(gl);
 	int error;
 
-	if (isreg) {
-		if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
-			unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
-		inode_dio_wait(&ip->i_inode);
-	}
+	if (isreg && test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
+		unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
 	if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
 		goto out;
 
@@ -440,9 +437,6 @@ static int inode_go_lock(struct gfs2_holder *gh)
 			return error;
 	}
 
-	if (gh->gh_state != LM_ST_DEFERRED)
-		inode_dio_wait(&ip->i_inode);
-
 	if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
 	    (gl->gl_state == LM_ST_EXCLUSIVE) &&
 	    (gh->gh_state == LM_ST_EXCLUSIVE)) {
-- 
2.24.1