On Tue, 2020-05-19 at 17:27 +0800, Zheng Bin wrote:
> Use the following command to test nfsv4(size of file1M is 1MB):
> mount -t nfs -o vers=4.0,actimeo=60 127.0.0.1/dir1 /mnt
> cp file1M /mnt
> du -h /mnt/file1M -->0 within 60s, then 1M
>
> When write is done(cp file1M /mnt), will call this:
> nfs_writeback_done
> nfs4_write_done
> nfs4_write_done_cb
> nfs_writeback_update_inode
> nfs_post_op_update_inode_force_wcc_locked(change, ctime,
> mtime
> nfs_post_op_update_inode_force_wcc_locked
> nfs_set_cache_invalid
> nfs_refresh_inode_locked
> nfs_update_inode
>
> nfsd write response contains change, ctime, mtime, the flag will be
> clear after nfs_update_inode. Howerver, write response does not
> contain
> space_used, previous open response contains space_used whose value is
> 0,
> so inode->i_blocks is still 0.
>
> nfs_getattr -->called by "du -h"
> do_update |= force_sync || nfs_attribute_cache_expired -->false in
> 60s
> cache_validity = READ_ONCE(NFS_I(inode)->cache_validity)
> do_update |= cache_validity & (NFS_INO_INVALID_ATTR -->false
> if (do_update) {
> __nfs_revalidate_inode
> }
>
> Within 60s, does not send getattr request to nfsd, thus "du -h
> /mnt/file1M"
> is 0.
>
> Fixes: 16e143751727 ("NFS: More fine grained attribute tracking")
> Signed-off-by: Zheng Bin <zhengbin13@xxxxxxxxxx>
> ---
> fs/nfs/inode.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/fs/nfs/inode.c b/fs/nfs/inode.c
> index b9d0921cb4fe..bdfb98d5f45b 100644
> --- a/fs/nfs/inode.c
> +++ b/fs/nfs/inode.c
> @@ -1764,7 +1764,8 @@ int
> nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct
> nfs_fa
> status = nfs_post_op_update_inode_locked(inode, fattr,
> NFS_INO_INVALID_CHANGE
> | NFS_INO_INVALID_CTIME
> - | NFS_INO_INVALID_MTIME);
> + | NFS_INO_INVALID_MTIME
> + | NFS_INO_INVALID_OTHER);
> return status;
> }
>
NACK. Please add a NFS_INO_INVALID_BLOCKS for this case instead of
setting INVALID_OTHER.
Quite frankly, "space used" is an anachronism from the days when
everything was stored on local disk on your workstation, before the
days of automated tiering, dedup and/or compression. It should not
matter to applications and so we definitely want to be able to optimise
it away in statx() calls that don't explicitly ask for it.
--
Trond Myklebust
Linux NFS client maintainer, Hammerspace
trond.myklebust@xxxxxxxxxxxxxxx
