fuse.getattr called by every block when do DIRECT_IO read with 4K block

[kk47]

When I use the fio to test the 4k read performance, I found the getattr was called in every block read, it seems the fuse filesystem with c++ libfuse do not have this question.

The fio command:
fio -numjobs=1 -iodepth=1 -ioengine=libaio -direct=1 -rw=read -size=4M -bs=4K -time_based -runtime=20 -name=Fio-1 -directory=/mnt

and the dd command got the same result
dd if=/mnt/testfile of=/dev/null bs=4K count=10 iflag=direct

cat /mnt/.accesslog

2024.09.11 09:25:42.345608 [uid:0,gid:0,pid:874844] read (1026,4096,1122304): (4096) - OK <0.000028>
2024.09.11 09:25:42.350034 [uid:0,gid:0,pid:874844] getattr (1026): (1026,[-rw-r--r--:0100644,1,0,0,1726017921,1726017922,1726017922,4194304]) - OK <0.004051>
2024.09.11 09:25:42.354292 [uid:0,gid:0,pid:874844] read (1026,4096,1126400): (4096) - OK <0.000031>
2024.09.11 09:25:42.362388 [uid:0,gid:0,pid:874844] getattr (1026): (1026,[-rw-r--r--:0100644,1,0,0,1726017921,1726017922,1726017922,4194304]) - OK <0.005230>
2024.09.11 09:25:42.363005 [uid:0,gid:0,pid:874844] read (1026,4096,1130496): (4096) - OK <0.000017>
2024.09.11 09:25:42.373989 [uid:0,gid:0,pid:874844] getattr (1026): (1026,[-rw-r--r--:0100644,1,0,0,1726017921,1726017922,1726017922,4194304]) - OK <0.010789>
2024.09.11 09:25:42.375317 [uid:0,gid:0,pid:874844] read (1026,4096,1134592): (4096) - OK <0.000024>
2024.09.11 09:25:42.394042 [uid:0,gid:0,pid:874844] getattr (1026): (1026,[-rw-r--r--:0100644,1,0,0,1726017921,1726017922,1726017922,4194304]) - OK <0.016077>
2024.09.11 09:25:42.394414 [uid:0,gid:0,pid:874844] read (1026,4096,1138688): (4096) - OK <0.000048>
2024.09.11 09:25:42.400903 [uid:0,gid:0,pid:874844] getattr (1026): (1026,[-rw-r--r--:0100644,1,0,0,1726017921,1726017922,1726017922,4194304]) - OK <0.005750>
2024.09.11 09:25:42.401618 [uid:0,gid:0,pid:874844] read (1026,4096,1142784): (4096) - OK <0.000017>
2024.09.11 09:25:42.407509 [uid:0,gid:0,pid:874844] getattr (1026): (1026,[-rw-r--r--:0100644,1,0,0,1726017921,1726017922,1726017922,4194304]) - OK <0.005622>
2024.09.11 09:25:42.408256 [uid:0,gid:0,pid:874844] read (1026,4096,1146880): (4096) - OK <0.000024>
2024.09.11 09:25:42.409010 [uid:0,gid:0,pid:874844] flush (1026,9,CAE4DB8115C4BE94) - OK <0.000004>

I want to known which component called the getattr, go-fuse, vfs, or someone else, and why the getattr before block read is necessary.

[davies]

it's called by kfuse kernel module.

[kk47]

it's called by kfuse kernel module.
Thanks for your reply, I have checked the source code of Linux-4.19 kernel fs/fuse and found these codes:

 fs/fuse/file.c

static ssize_t __fuse_direct_read(struct fuse_io_priv *io,
				  struct iov_iter *iter,
				  loff_t *ppos)
{
	ssize_t res;
	struct inode *inode = file_inode(io->iocb->ki_filp);

	if (is_bad_inode(inode))
		return -EIO;

	res = fuse_direct_io(io, iter, ppos, 0);

	fuse_invalidate_attr(inode); 

	return res;
}


fs/fuse/dir.c
/*
 * Mark the attributes as stale, so that at the next call to
 * ->getattr() they will be fetched from userspace
 */
void fuse_invalidate_attr(struct inode *inode)
{
	get_fuse_inode(inode)->i_time = 0;
}

is the fuse_invalidate_attr function cause the GetAttr called after every 4k block read ?