nfsd: improvements for wake_up_bit/var
Subsystems (2)
cover nfsd: improvements for wake_up_bit/var
I've been digging into wake_up_bit and wake_up_var recently. There are a lot of places where the required barriers aren't quite right.
This patch fixes them up for nfsd. The bugs are mostly minor, though the rp_locked on might be a credible problem on weakly ordered hosts (e.g. power64).
NeilBrown
[PATCH 1/2] nfsd: use clear_and_wake_up_bit()
[PATCH 2/2] nfsd: avoid races with wake_up_var()
PATCH 1/2 nfsd: use clear_and_wake_up_bit()
nfsd has two places that open-code clear_and_wake_up_bit(). One has the required memory barriers. The other does not.
Change both to use clear_and_wake_up_bit() so with have the barriers without the noise.
--- fs/nfsd/nfs4recover.c | 5 +---- fs/nfsd/nfs4state.c | 3 +-- 2 files changed, 2 insertions(+), 6 deletions(-)
fs/nfsd/nfs4recover.c +1 -4
@@ -1895,10 +1895,7 @@ nfsd4_cltrack_upcall_lock(struct nfs4_client *clp)static void nfsd4_cltrack_upcall_unlock (struct nfs4_client *clp) { -smp_mb__before_atomic (); -clear_bit (NFSD4_CLIENT_UPCALL_LOCK , &clp->cl_flags); -smp_mb__after_atomic (); -wake_up_bit (&clp->cl_flags,NFSD4_CLIENT_UPCALL_LOCK ); +clear_and_wake_up_bit (&clp->cl_flags,NFSD4_CLIENT_UPCALL_LOCK ); } static void
fs/nfsd/nfs4state.c +1 -2
@@ -3076,8 +3076,7 @@ nfsd4_cb_getattr_release(struct nfsd4_callback *cb) container_of(ncf, struct nfs4_delegation, dl_cb_fattr);nfs4_put_stid (&dp->dl_stid); -clear_bit (CB_GETATTR_BUSY , &ncf->ncf_cb_flags); -wake_up_bit (&ncf->ncf_cb_flags,CB_GETATTR_BUSY ); +clear_and_wake_up_bit (&ncf->ncf_cb_flags,CB_GETATTR_BUSY ); }static const struct nfsd4_callback_ops nfsd4_cb_recall_any_ops = {
Signature · 1 line
2.44.0
PATCH 2/2 nfsd: avoid races with wake_up_var()
In reply to NeilBrownwake_up_var() needs a barrier after the important change is made in the var and before wake_up_var() is called, else it is possible that a wake up won't be sent when it should.
In each case here the var is changed in an "atomic" manner, so smb_mb__after_atomic() is sufficient.
In one case the important change (removing the lease) is performed *after* the wake_up, which is backwards. The code survives in part because the wait_var_event is given a timeout.
This patch adds the required barriers and calls destroy_delegation() *before* waking any threads waiting for the delegation to be destroyed.
--- fs/nfsd/nfs4state.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
fs/nfsd/nfs4state.c +4 -1
@@ -4700,6 +4700,7 @@ void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)if (so !=NULL ) { cstate->replay_owner =NULL ;atomic_set (&so->so_replay.rp_locked,RP_UNLOCKED ); +smb_mb__after_atomic ();wake_up_var (&so->so_replay.rp_locked);nfs4_put_stateowner (so); }
@@ -5000,6 +5001,7 @@ move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)* so tell them to stop waiting. */ atomic_set (&oo->oo_owner.so_replay.rp_locked,RP_UNHASHED ); +smb_mb__after_atomic ();wake_up_var (&oo->oo_owner.so_replay.rp_locked);wait_event (close_wq,refcount_read (&s->st_stid.sc_count) ==2 );
@@ -7469,8 +7471,9 @@ nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, goto put_stateid;trace_nfsd_deleg_return (stateid); -wake_up_var (d_inode (cstate->current_fh.fh_dentry));destroy_delegation (dp); +smb_mb__after_atomic (); +wake_up_var (d_inode (cstate->current_fh.fh_dentry)); put_stateid:nfs4_put_stid (&dp->dl_stid); out:
Signature · 1 line
2.44.0
Hi Neil,
Do we need a barrier in nfs4_disable_swap() as well ?
10830 set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
10831 clear_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state);
Quoted 1 line
Quoted 1 line
Quoted 1 line
HERE>>>>
10832 wake_up_var(&clp->cl_state);
Regards,
Santosh
Quoted 13 lines · NeilBrown
I've been digging into
wake_up_bitandwake_up_varrecently. There are a lot of places where the required barriers aren't quite right.This patch fixes them up for nfsd. The bugs are mostly minor, though the
rp_lockedon might be a credible problem on weakly ordered hosts (e.g. power64).NeilBrown
[PATCH 1/2] nfsd: use
clear_and_wake_up_bit()
[PATCH 2/2] nfsd: avoid races withwake_up_var()
Quoted 7 lines · Santosh Pradhan
Hi Neil,
Do we need a barrier innfs4_disable_swap()as well ?10830
set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
10831clear_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state);Quoted 1 line
Quoted 1 line
Quoted 1 line
HERE>>>>
10832
wake_up_var(&clp->cl_state);
Probably but that is in fs/nfs/, not fs/nfsd/.
Thanks,
NeilBrown
Quoted 19 lines
Regards,
SantoshQuoted 13 lines · NeilBrown
I've been digging into
wake_up_bitandwake_up_varrecently. There are a lot of places where the required barriers aren't quite right.This patch fixes them up for nfsd. The bugs are mostly minor, though the
rp_lockedon might be a credible problem on weakly ordered hosts (e.g. power64).NeilBrown
[PATCH 1/2] nfsd: use
clear_and_wake_up_bit()
[PATCH 2/2] nfsd: avoid races withwake_up_var()
Quoted 11 lines · NeilBrown
I've been digging into
wake_up_bitandwake_up_varrecently. There are a lot of places where the required barriers aren't quite right.This patch fixes them up for nfsd. The bugs are mostly minor, though the
rp_lockedon might be a credible problem on weakly ordered hosts (e.g. power64).NeilBrown
[PATCH 1/2] nfsd: use
clear_and_wake_up_bit()
[PATCH 2/2] nfsd: avoid races withwake_up_var()
Nice catches.
Chuck Lever
AppliedQuoted 8 lines · NeilBrown
I've been digging into
wake_up_bitandwake_up_varrecently. There are a lot of places where the required barriers aren't quite right.This patch fixes them up for nfsd. The bugs are mostly minor, though the
rp_lockedon might be a credible problem on weakly ordered hosts (e.g. power64).[...]
Applied to nfsd-next for v6.12, thanks!
[1/2] nfsd: use clear_and_wake_up_bit()
commit: 2b9a19d16beda1b2ca5edab47d74b73d4d958b12
[2/2] nfsd: avoid races with wake_up_var()
commit: a2bf7d13821603fb90c3f6e695bd5fb4ee19de71Both of these patches threw compilation errors. I corrected those issues before applying. Please check my work.
checkpatch.pl complained about the lack of comment on the added barriers. I felt the code was self-explanatory so made no change.
Signature · 1 line
Chuck Lever
Quoted 12 lines · Chuck Lever
Quoted 8 lines · NeilBrown
I've been digging into
wake_up_bitandwake_up_varrecently. There are a lot of places where the required barriers aren't quite right.This patch fixes them up for nfsd. The bugs are mostly minor, though the
rp_lockedon might be a credible problem on weakly ordered hosts (e.g. power64).[...]
Applied to nfsd-next for v6.12, thanks!
Quoted 7 lines
[1/2] nfsd: use
clear_and_wake_up_bit()commit: 2b9a19d16beda1b2ca5edab47d74b73d4d958b12 [2/2] nfsd: avoid races with wake_up_var() commit: a2bf7d13821603fb90c3f6e695bd5fb4ee19de71Both of these patches threw compilation errors. I corrected those issues before applying. Please check my work.
Thanks for that ... I really shouldn't post patches in the afternoon - I'm not thinking as clearly :-(
Quoted 4 lines
checkpatch.pl complained about the lack of comment on the added barriers. I felt the code was self-explanatory so made no change.
I'd really like a wake_up_var_after_atomic() but Linus wasn't keen on going in that direction.
if/when my improvements land
https://lore.kernel.org/all/[email protected]/
we could change rp_locked to an "int" and use
store_release_wake_up()
which includes that required barrier.
NeilBrown