[Ocfs2-devel] [PATCH v2] ocfs2: flush truncate log when main bitmap run out of space

[Joseph Qi]

On 18/12/28 16:29, Jia Guo wrote:
> Truncate log problem has been described in
> commit 50308d813bf2 ("ocfs2: Try to free truncate log when meeting
> ENOSPC in write.") and
> commit 2070ad1aebff ("ocfs2: retry on ENOSPC if sufficient space in
> truncate log"), but the following situations cannot be solved:
> 
> case 1:
> Main bitmap has been updated, but the transaction of the deleted blocks has
> not been committed completely. In this case, function
> ocfs2_reserve_cluster_bitmap_bits() returns success while function
> __ocfs2_claim_clusters return ENOSPC because we cannot reuse the deleted
> blocks before the transaction committed(test by function
> ocfs2_test_bg_bit_allocatable()). One can reproduce this by following
> steps:
> a, prepare a file which size is 50G, and volume still have 30G free space
> b, open the file with O_TRUNC flag
> c, sleep 5 seconds
> d, fallocate a 50G file, fallocate will fail.
> 
> case 2:
> Main bitmap doesn't have enough free bits, so does truncate log. But main
> bitmap plus truncate log has enough free bits. We are not gonging to try
> flush truncate log in this case, which is not reasonable.
> 
> So force commit the transaction when flushing the truncate log for case 1.
> For case 2, the value of osb->truncated_clusters doesn't seem to make
> sense, do the flush whenever we run out of space seems to be more
> reasonable.
> 
> Signed-off-by: Jia Guo <guojia12 at huawei.com>
> Reviewed-by: Yiwen Jiang <jiangyiwen at huawei.com>
> Reviewed-by: Gang He <ghe at suse.com>

Acked-by: Joseph Qi <jiangqi903 at gmail.com>