| Message ID | 20240821063901.650776-2-hch@lst.de (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [1/5] xfs: use kfree_rcu_mightsleep to free the perag structures | expand |
On Wed, Aug 21, 2024 at 08:38:28AM +0200, Christoph Hellwig wrote: > Using the kfree_rcu_mightsleep is simpler and removes the need for a > rcu_head in the perag structure. > > Signed-off-by: Christoph Hellwig <hch@lst.de> > --- > fs/xfs/libxfs/xfs_ag.c | 12 +----------- > fs/xfs/libxfs/xfs_ag.h | 3 --- > 2 files changed, 1 insertion(+), 14 deletions(-) > > diff --git a/fs/xfs/libxfs/xfs_ag.c b/fs/xfs/libxfs/xfs_ag.c > index 7e80732cb54708..4b5a39a83f7aed 100644 > --- a/fs/xfs/libxfs/xfs_ag.c > +++ b/fs/xfs/libxfs/xfs_ag.c > @@ -235,16 +235,6 @@ xfs_initialize_perag_data( > return error; > } > > -STATIC void > -__xfs_free_perag( > - struct rcu_head *head) > -{ > - struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head); > - > - ASSERT(!delayed_work_pending(&pag->pag_blockgc_work)); > - kfree(pag); > -} > - > /* > * Free up the per-ag resources associated with the mount structure. > */ > @@ -270,7 +260,7 @@ xfs_free_perag( > xfs_perag_rele(pag); > XFS_IS_CORRUPT(pag->pag_mount, > atomic_read(&pag->pag_active_ref) != 0); > - call_rcu(&pag->rcu_head, __xfs_free_perag); > + kfree_rcu_mightsleep(pag); I started wondering, have you seen any complaints from might_sleep when freeing pags after a failed growfs? Then I wondered if growfs_data could actually take any locks that would prevent sleeping, which led me to another question: why do growfs_{data,log} hold m_growlock but growfs_rt doesn't? Is that actually safe? I think the kfree_rcu_mightsleep conversion is ok, but I want to see if anything blows up with the rtgroups variant. --D > } > } > > diff --git a/fs/xfs/libxfs/xfs_ag.h b/fs/xfs/libxfs/xfs_ag.h > index 35de09a2516c70..d62c266c0b44d5 100644 > --- a/fs/xfs/libxfs/xfs_ag.h > +++ b/fs/xfs/libxfs/xfs_ag.h > @@ -63,9 +63,6 @@ struct xfs_perag { > /* Blocks reserved for the reverse mapping btree. */ > struct xfs_ag_resv pag_rmapbt_resv; > > - /* for rcu-safe freeing */ > - struct rcu_head rcu_head; > - > /* Precalculated geometry info */ > xfs_agblock_t block_count; > xfs_agblock_t min_block; > -- > 2.43.0 > >
On Wed, Aug 21, 2024 at 09:19:39AM -0700, Darrick J. Wong wrote: > I started wondering, have you seen any complaints from might_sleep when > freeing pags after a failed growfs? No, why would I? We're not freeing perags with a spinlock held there. > Then I wondered if growfs_data > could actually take any locks that would prevent sleeping, which led me > to another question: why do growfs_{data,log} hold m_growlock but > growfs_rt doesn't? Is that actually safe? As far as I can tell growfs_rt is missing a m_growlock critical section and right now we allow parallel calls to growfs_rt, which could lead to unexpected results.
diff --git a/fs/xfs/libxfs/xfs_ag.c b/fs/xfs/libxfs/xfs_ag.c index 7e80732cb54708..4b5a39a83f7aed 100644 --- a/fs/xfs/libxfs/xfs_ag.c +++ b/fs/xfs/libxfs/xfs_ag.c @@ -235,16 +235,6 @@ xfs_initialize_perag_data( return error; } -STATIC void -__xfs_free_perag( - struct rcu_head *head) -{ - struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head); - - ASSERT(!delayed_work_pending(&pag->pag_blockgc_work)); - kfree(pag); -} - /* * Free up the per-ag resources associated with the mount structure. */ @@ -270,7 +260,7 @@ xfs_free_perag( xfs_perag_rele(pag); XFS_IS_CORRUPT(pag->pag_mount, atomic_read(&pag->pag_active_ref) != 0); - call_rcu(&pag->rcu_head, __xfs_free_perag); + kfree_rcu_mightsleep(pag); } } diff --git a/fs/xfs/libxfs/xfs_ag.h b/fs/xfs/libxfs/xfs_ag.h index 35de09a2516c70..d62c266c0b44d5 100644 --- a/fs/xfs/libxfs/xfs_ag.h +++ b/fs/xfs/libxfs/xfs_ag.h @@ -63,9 +63,6 @@ struct xfs_perag { /* Blocks reserved for the reverse mapping btree. */ struct xfs_ag_resv pag_rmapbt_resv; - /* for rcu-safe freeing */ - struct rcu_head rcu_head; - /* Precalculated geometry info */ xfs_agblock_t block_count; xfs_agblock_t min_block;
Using the kfree_rcu_mightsleep is simpler and removes the need for a rcu_head in the perag structure. Signed-off-by: Christoph Hellwig <hch@lst.de> --- fs/xfs/libxfs/xfs_ag.c | 12 +----------- fs/xfs/libxfs/xfs_ag.h | 3 --- 2 files changed, 1 insertion(+), 14 deletions(-)