On Wed 02-08-17 21:33:49, Dave Chinner wrote: > On Wed, Aug 02, 2017 at 10:37:41AM +0200, Jan Kara wrote: > > The bio describing discard operation is allocated by > > __blkdev_issue_discard() which returns us a reference to it. That > > reference is never released and thus we leak this bio. Drop the bio > > reference once it completes in xlog_discard_endio(). > > > > CC: stable@xxxxxxxxxxxxxxx > > Fixes: 4560e78f40cb55bd2ea8f1ef4001c5baa88531c7 > > Signed-off-by: Jan Kara <jack@xxxxxxx> > > --- > > fs/xfs/xfs_log_cil.c | 1 + > > 1 file changed, 1 insertion(+) > > > > Note, this is untested. It's just something I've noticed when looking at the > > code. > > > > diff --git a/fs/xfs/xfs_log_cil.c b/fs/xfs/xfs_log_cil.c > > index fbe72b134bef..43aa42a3a5d3 100644 > > --- a/fs/xfs/xfs_log_cil.c > > +++ b/fs/xfs/xfs_log_cil.c > > @@ -539,6 +539,7 @@ xlog_discard_endio( > > > > INIT_WORK(&ctx->discard_endio_work, xlog_discard_endio_work); > > queue_work(xfs_discard_wq, &ctx->discard_endio_work); > > + bio_put(bio); > > } > > At a quick glance, it does seem that we are leaking the bio here. It > appears to me that ext4_process_freed_data() also has the same > problem and leaks the discard_bio.... Yes, and a fix of that prompted me to look into how XFS does things ;) Honza -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR