On Fri, Aug 21, 2020 at 7:33 PM Bob Peterson <rpete...@redhat.com> wrote:
> Before this patch, function gfs2_evict_inode would start a transaction
> in order to write some unknown number of revokes. Instead of calculating
> the value, it used sdp->sd_jdesc->jd_blocks, the number of blocks in the
> entire journal. You can fit a lot of revokes in a block (in a 4K block
> you can fit 503 of them). For a 512MB journal has 131072 blocks, it would
> ask for 261 free blocks.
>
> This patch changes it to calculate the number of revokes based on the
> nrpages values of the address space plus glock address space. Note that
> since it does this after clearing out the ail list, we don't need to
> use the number of (old) revokes in the calculation. We calculate it from
> the number of standing new items that need revoking.
>
> Signed-off-by: Bob Peterson <rpete...@redhat.com>
> ---
> fs/gfs2/super.c | 7 ++++++-
> 1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/fs/gfs2/super.c b/fs/gfs2/super.c
> index 9f4d9e7be839..80ac446f0110 100644
> --- a/fs/gfs2/super.c
> +++ b/fs/gfs2/super.c
> @@ -1336,6 +1336,7 @@ static void gfs2_evict_inode(struct inode *inode)
> struct gfs2_inode *ip = GFS2_I(inode);
> struct gfs2_holder gh;
> struct address_space *metamapping;
> + int nr_revokes;
> int error;
>
> if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
> @@ -1434,7 +1435,11 @@ static void gfs2_evict_inode(struct inode *inode)
> write_inode_now(inode, 1);
> gfs2_ail_flush(ip->i_gl, 0);
>
> - error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
> + nr_revokes = inode->i_mapping->nrpages + metamapping->nrpages;
We could surely use a more realistic number of revokes here, but this
approach really doesn't make any sense, sorry.
> + if (!nr_revokes)
> + goto out_unlock;
> +
> + error = gfs2_trans_begin(sdp, 0, nr_revokes);
> if (error)
> goto out_unlock;
> /* Needs to be done before glock release & also in a transaction */
> --
> 2.26.2
Thanks,
Andreas
