On Wed, Jul 08, 2020 at 02:56:07PM +0200, Carlos Maiolino wrote: > change xlog_ticket_alloc() to use default kmem_cache_zalloc(), and > modify its callers to pass MM flags as arguments. > > Signed-off-by: Carlos Maiolino <cmaiolino@xxxxxxxxxx> > --- > fs/xfs/xfs_log.c | 7 ++++--- > fs/xfs/xfs_log_cil.c | 2 +- > fs/xfs/xfs_log_priv.h | 2 +- > 3 files changed, 6 insertions(+), 5 deletions(-) > > diff --git a/fs/xfs/xfs_log.c b/fs/xfs/xfs_log.c > index 00fda2e8e7380..6d40d479e34a1 100644 > --- a/fs/xfs/xfs_log.c > +++ b/fs/xfs/xfs_log.c > @@ -433,7 +433,8 @@ xfs_log_reserve( > XFS_STATS_INC(mp, xs_try_logspace); > > ASSERT(*ticp == NULL); > - tic = xlog_ticket_alloc(log, unit_bytes, cnt, client, permanent, 0); > + tic = xlog_ticket_alloc(log, unit_bytes, cnt, client, permanent, > + (GFP_KERNEL | __GFP_NOFAIL)); > *ticp = tic; xfs_log_reserve() is called either from transaction reservation which is under memalloc_nofs context, or from the CIL with explicit GFP_NOFS, or from the unmount path which is GFP_KERNEL but is holding various filesystem locks. I suspect that this patch should just remove the gfp flags from xlog_ticket_alloc() and just unconditionally use GFP_NOFS | __GFP_NOFAIL fo allocating the ticket. That would clean this up quite a bit.... Cheers, Dave. -- Dave Chinner david@xxxxxxxxxxxxx
