| Message ID | 20230926072541.564177-1-markzhang@nvidia.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Headers | show |
| Series | [rdma-next] RDMA/cma: Initialize traffic_class to 0 for multicast path records | expand |
On Tue, Sep 26, 2023 at 10:25:41AM +0300, Mark Zhang wrote: > Initialize traffic_class to 0 so that it wouldn't have a random value, > which causes a random IP DSCP in RoCEv2. It will be great to see call trace which explains how. I think that ib.rec.sl has same issue. diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c index c343edf2f664..1e2cd7c8716e 100644 --- a/drivers/infiniband/core/cma.c +++ b/drivers/infiniband/core/cma.c @@ -4968,7 +4968,7 @@ static int cma_iboe_join_multicast(struct rdma_id_private *id_priv, int err = 0; struct sockaddr *addr = (struct sockaddr *)&mc->addr; struct net_device *ndev = NULL; - struct ib_sa_multicast ib; + struct ib_sa_multicast ib = {}; enum ib_gid_type gid_type; bool send_only; Thanks > > Fixes: b5de0c60cc30 ("RDMA/cma: Fix use after free race in roce multicast join") > Signed-off-by: Mark Zhang <markzhang@nvidia.com> > --- > drivers/infiniband/core/cma.c | 1 + > 1 file changed, 1 insertion(+) > > diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c > index c343edf2f664..d3a72f4b9863 100644 > --- a/drivers/infiniband/core/cma.c > +++ b/drivers/infiniband/core/cma.c > @@ -4990,6 +4990,7 @@ static int cma_iboe_join_multicast(struct rdma_id_private *id_priv, > ib.rec.rate = IB_RATE_PORT_CURRENT; > ib.rec.hop_limit = 1; > ib.rec.mtu = iboe_get_mtu(ndev->mtu); > + ib.rec.traffic_class = 0; > > if (addr->sa_family == AF_INET) { > if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) { > -- > 2.37.1 >
On 9/26/2023 7:09 PM, Leon Romanovsky wrote: > On Tue, Sep 26, 2023 at 10:25:41AM +0300, Mark Zhang wrote: >> Initialize traffic_class to 0 so that it wouldn't have a random value, >> which causes a random IP DSCP in RoCEv2. > > It will be great to see call trace which explains how. I think that > ib.rec.sl has same issue. Yes ib.rec.sl and ib.rec.flow_label should have the same issue, I just not able to reproduce as somehow they are 0 in my setups. These fields will be used to generate the ah: cma_iboe_join_multicast cma_make_mc_event ib_init_ah_from_mcmember > diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c > index c343edf2f664..1e2cd7c8716e 100644 > --- a/drivers/infiniband/core/cma.c > +++ b/drivers/infiniband/core/cma.c > @@ -4968,7 +4968,7 @@ static int cma_iboe_join_multicast(struct rdma_id_private *id_priv, > int err = 0; > struct sockaddr *addr = (struct sockaddr *)&mc->addr; > struct net_device *ndev = NULL; > - struct ib_sa_multicast ib; > + struct ib_sa_multicast ib = {}; > enum ib_gid_type gid_type; > bool send_only; I think this patch is great. So what should I do now? Send this as v1, or you will do it? Thanks > Thanks > >> >> Fixes: b5de0c60cc30 ("RDMA/cma: Fix use after free race in roce multicast join") >> Signed-off-by: Mark Zhang <markzhang@nvidia.com> >> --- >> drivers/infiniband/core/cma.c | 1 + >> 1 file changed, 1 insertion(+) >> >> diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c >> index c343edf2f664..d3a72f4b9863 100644 >> --- a/drivers/infiniband/core/cma.c >> +++ b/drivers/infiniband/core/cma.c >> @@ -4990,6 +4990,7 @@ static int cma_iboe_join_multicast(struct rdma_id_private *id_priv, >> ib.rec.rate = IB_RATE_PORT_CURRENT; >> ib.rec.hop_limit = 1; >> ib.rec.mtu = iboe_get_mtu(ndev->mtu); >> + ib.rec.traffic_class = 0; >> >> if (addr->sa_family == AF_INET) { >> if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) { >> -- >> 2.37.1 >>
On Wed, Sep 27, 2023 at 10:30:47AM +0800, Mark Zhang wrote: > On 9/26/2023 7:09 PM, Leon Romanovsky wrote: > > On Tue, Sep 26, 2023 at 10:25:41AM +0300, Mark Zhang wrote: > > > Initialize traffic_class to 0 so that it wouldn't have a random value, > > > which causes a random IP DSCP in RoCEv2. > > > > It will be great to see call trace which explains how. I think that > > ib.rec.sl has same issue. > > Yes ib.rec.sl and ib.rec.flow_label should have the same issue, I just not > able to reproduce as somehow they are 0 in my setups. > > These fields will be used to generate the ah: > cma_iboe_join_multicast > cma_make_mc_event > ib_init_ah_from_mcmember > > > diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c > > index c343edf2f664..1e2cd7c8716e 100644 > > --- a/drivers/infiniband/core/cma.c > > +++ b/drivers/infiniband/core/cma.c > > @@ -4968,7 +4968,7 @@ static int cma_iboe_join_multicast(struct rdma_id_private *id_priv, > > int err = 0; > > struct sockaddr *addr = (struct sockaddr *)&mc->addr; > > struct net_device *ndev = NULL; > > - struct ib_sa_multicast ib; > > + struct ib_sa_multicast ib = {}; > > enum ib_gid_type gid_type; > > bool send_only; > > I think this patch is great. So what should I do now? Send this as v1, or > you will do it? Send v1 with updated commit message. Thanks > > Thanks > > > Thanks > > > > > > > > Fixes: b5de0c60cc30 ("RDMA/cma: Fix use after free race in roce multicast join") > > > Signed-off-by: Mark Zhang <markzhang@nvidia.com> > > > --- > > > drivers/infiniband/core/cma.c | 1 + > > > 1 file changed, 1 insertion(+) > > > > > > diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c > > > index c343edf2f664..d3a72f4b9863 100644 > > > --- a/drivers/infiniband/core/cma.c > > > +++ b/drivers/infiniband/core/cma.c > > > @@ -4990,6 +4990,7 @@ static int cma_iboe_join_multicast(struct rdma_id_private *id_priv, > > > ib.rec.rate = IB_RATE_PORT_CURRENT; > > > ib.rec.hop_limit = 1; > > > ib.rec.mtu = iboe_get_mtu(ndev->mtu); > > > + ib.rec.traffic_class = 0; > > > if (addr->sa_family == AF_INET) { > > > if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) { > > > -- > > > 2.37.1 > > > >
diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c index c343edf2f664..d3a72f4b9863 100644 --- a/drivers/infiniband/core/cma.c +++ b/drivers/infiniband/core/cma.c @@ -4990,6 +4990,7 @@ static int cma_iboe_join_multicast(struct rdma_id_private *id_priv, ib.rec.rate = IB_RATE_PORT_CURRENT; ib.rec.hop_limit = 1; ib.rec.mtu = iboe_get_mtu(ndev->mtu); + ib.rec.traffic_class = 0; if (addr->sa_family == AF_INET) { if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
Initialize traffic_class to 0 so that it wouldn't have a random value, which causes a random IP DSCP in RoCEv2. Fixes: b5de0c60cc30 ("RDMA/cma: Fix use after free race in roce multicast join") Signed-off-by: Mark Zhang <markzhang@nvidia.com> --- drivers/infiniband/core/cma.c | 1 + 1 file changed, 1 insertion(+)