Re: [PATCH net v3] net: introduce CAN specific pointer in the struct net_device
Hi Jakub, On Mon, Feb 22, 2021 at 05:30:12PM -0800, Jakub Kicinski wrote: > On Mon, 22 Feb 2021 16:02:51 +0100 Oleksij Rempel wrote: > > Since 20dd3850bcf8 ("can: Speed up CAN frame receiption by using > > ml_priv") the CAN framework uses per device specific data in the AF_CAN > > protocol. For this purpose the struct net_device->ml_priv is used. Later > > the ml_priv usage in CAN was extended for other users, one of them being > > CAN_J1939. > > > > Later in the kernel ml_priv was converted to an union, used by other > > drivers. E.g. the tun driver started storing it's stats pointer. > > > > Since tun devices can claim to be a CAN device, CAN specific protocols > > will wrongly interpret this pointer, which will cause system crashes. > > Mostly this issue is visible in the CAN_J1939 stack. > > > > To fix this issue, we request a dedicated CAN pointer within the > > net_device struct. > > > > Reported-by: syzbot+5138c4dd15a0401be...@syzkaller.appspotmail.com > > Fixes: 20dd3850bcf8 ("can: Speed up CAN frame receiption by using ml_priv") > > Fixes: ffd956eef69b ("can: introduce CAN midlayer private and allocate it > > automatically") > > Fixes: 9d71dd0c7009 ("can: add support of SAE J1939 protocol") > > Fixes: 497a5757ce4e ("tun: switch to net core provided statistics counters") > > Signed-off-by: Oleksij Rempel > > > diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h > > index ddf4cfc12615..6e25c6f0f190 100644 > > --- a/include/linux/netdevice.h > > +++ b/include/linux/netdevice.h > > @@ -1584,6 +1584,16 @@ enum netdev_priv_flags { > > #define IFF_L3MDEV_RX_HANDLER IFF_L3MDEV_RX_HANDLER > > #define IFF_LIVE_RENAME_OK IFF_LIVE_RENAME_OK > > > > +/** > > + * enum netdev_ml_priv_type - &struct net_device ml_priv_type > > + * > > + * This enum specifies the type of the struct net_device::ml_priv pointer. > > + */ > > kdoc (scripts/kernel-doc -none include/linux/netdevice.h) is not happy > about the fact enum values are not defined. Perhaps they will be > sufficiently self-explanatory to not bother documenting? > > Maybe just: > > /* Specifies the type of the struct net_device::ml_priv pointer */ > > ? sounds good, done. > > +enum netdev_ml_priv_type { > > + ML_PRIV_NONE, > > + ML_PRIV_CAN, > > +}; > > + > > /** > > * struct net_device - The DEVICE structure. > > * > > @@ -1779,6 +1789,7 @@ enum netdev_priv_flags { > > * @nd_net:Network namespace this network device > > is inside > > * > > * @ml_priv: Mid-layer private > > + @ml_priv_type: Mid-layer private type > > missing '*' at the start of the line done > > * @lstats:Loopback statistics > > * @tstats:Tunnel statistics > > * @dstats:Dummy statistics > > @@ -2094,8 +2105,10 @@ struct net_device { > > possible_net_t nd_net; > > > > /* mid-layer private */ > > + void*ml_priv; > > + enum netdev_ml_priv_typeml_priv_type; > > + > > union { > > - void*ml_priv; > > struct pcpu_lstats __percpu *lstats; > > struct pcpu_sw_netstats __percpu*tstats; > > struct pcpu_dstats __percpu *dstats; > > @@ -2286,6 +2299,29 @@ static inline void netdev_reset_rx_headroom(struct > > net_device *dev) > > netdev_set_rx_headroom(dev, -1); > > } > > > > +static inline void *netdev_get_ml_priv(struct net_device *dev, > > + enum netdev_ml_priv_type type) > > +{ > > + if (dev->ml_priv_type != type) > > + return NULL; > > + > > + return dev->ml_priv; > > +} > > + > > +static inline void netdev_set_ml_priv(struct net_device *dev, > > + void *ml_priv, > > + enum netdev_ml_priv_type type) > > +{ > > + WARN_ONCE(dev->ml_priv_type && dev->ml_priv_type != type, > > + "Overwriting already set ml_priv_type (%u) with different > > ml_priv_type (%u)!\n", > > + dev->ml_priv_type, type); > > + WARN_ONCE(!dev->ml_priv_type && dev->ml_priv, > > + "Overwriting already set ml_priv and ml_priv_type is > > ML_PRIV_NONE!\n"); > > nit: do we need the _ONCE() this helper should be used on control path > and relatively rarely, no? I have no strong opinion right now. Changed to WARN() > > + dev->ml_priv = ml_priv; > > + dev->ml_priv_type = type; > > +} > > + > > /* > > * Net namespace inlines > > */ > > > @@ -454,6 +455,7 @@ static int j1939_sk_bind(struct socket *sock, struct > > sockaddr *uaddr, int len) > > j1939_local_ecu_put(priv, jsk->addr.src_name, jsk->addr.sa); > > } else { > > struct net_device *ndev; > > + struct can_ml_priv *can_ml; > > nit: rev xmas treei done > > > > > ndev = dev_get_by_index(net, addr->can_ifindex); >
Re: [PATCH net v3] net: introduce CAN specific pointer in the struct net_device
On Mon, 22 Feb 2021 16:02:51 +0100 Oleksij Rempel wrote: > Since 20dd3850bcf8 ("can: Speed up CAN frame receiption by using > ml_priv") the CAN framework uses per device specific data in the AF_CAN > protocol. For this purpose the struct net_device->ml_priv is used. Later > the ml_priv usage in CAN was extended for other users, one of them being > CAN_J1939. > > Later in the kernel ml_priv was converted to an union, used by other > drivers. E.g. the tun driver started storing it's stats pointer. > > Since tun devices can claim to be a CAN device, CAN specific protocols > will wrongly interpret this pointer, which will cause system crashes. > Mostly this issue is visible in the CAN_J1939 stack. > > To fix this issue, we request a dedicated CAN pointer within the > net_device struct. > > Reported-by: syzbot+5138c4dd15a0401be...@syzkaller.appspotmail.com > Fixes: 20dd3850bcf8 ("can: Speed up CAN frame receiption by using ml_priv") > Fixes: ffd956eef69b ("can: introduce CAN midlayer private and allocate it > automatically") > Fixes: 9d71dd0c7009 ("can: add support of SAE J1939 protocol") > Fixes: 497a5757ce4e ("tun: switch to net core provided statistics counters") > Signed-off-by: Oleksij Rempel > diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h > index ddf4cfc12615..6e25c6f0f190 100644 > --- a/include/linux/netdevice.h > +++ b/include/linux/netdevice.h > @@ -1584,6 +1584,16 @@ enum netdev_priv_flags { > #define IFF_L3MDEV_RX_HANDLERIFF_L3MDEV_RX_HANDLER > #define IFF_LIVE_RENAME_OK IFF_LIVE_RENAME_OK > > +/** > + * enum netdev_ml_priv_type - &struct net_device ml_priv_type > + * > + * This enum specifies the type of the struct net_device::ml_priv pointer. > + */ kdoc (scripts/kernel-doc -none include/linux/netdevice.h) is not happy about the fact enum values are not defined. Perhaps they will be sufficiently self-explanatory to not bother documenting? Maybe just: /* Specifies the type of the struct net_device::ml_priv pointer */ ? > +enum netdev_ml_priv_type { > + ML_PRIV_NONE, > + ML_PRIV_CAN, > +}; > + > /** > * struct net_device - The DEVICE structure. > * > @@ -1779,6 +1789,7 @@ enum netdev_priv_flags { > * @nd_net:Network namespace this network device is inside > * > * @ml_priv: Mid-layer private > + @ml_priv_type: Mid-layer private type missing '*' at the start of the line > * @lstats:Loopback statistics > * @tstats:Tunnel statistics > * @dstats:Dummy statistics > @@ -2094,8 +2105,10 @@ struct net_device { > possible_net_t nd_net; > > /* mid-layer private */ > + void*ml_priv; > + enum netdev_ml_priv_typeml_priv_type; > + > union { > - void*ml_priv; > struct pcpu_lstats __percpu *lstats; > struct pcpu_sw_netstats __percpu*tstats; > struct pcpu_dstats __percpu *dstats; > @@ -2286,6 +2299,29 @@ static inline void netdev_reset_rx_headroom(struct > net_device *dev) > netdev_set_rx_headroom(dev, -1); > } > > +static inline void *netdev_get_ml_priv(struct net_device *dev, > +enum netdev_ml_priv_type type) > +{ > + if (dev->ml_priv_type != type) > + return NULL; > + > + return dev->ml_priv; > +} > + > +static inline void netdev_set_ml_priv(struct net_device *dev, > + void *ml_priv, > + enum netdev_ml_priv_type type) > +{ > + WARN_ONCE(dev->ml_priv_type && dev->ml_priv_type != type, > + "Overwriting already set ml_priv_type (%u) with different > ml_priv_type (%u)!\n", > + dev->ml_priv_type, type); > + WARN_ONCE(!dev->ml_priv_type && dev->ml_priv, > + "Overwriting already set ml_priv and ml_priv_type is > ML_PRIV_NONE!\n"); nit: do we need the _ONCE() this helper should be used on control path and relatively rarely, no? > + dev->ml_priv = ml_priv; > + dev->ml_priv_type = type; > +} > + > /* > * Net namespace inlines > */ > @@ -454,6 +455,7 @@ static int j1939_sk_bind(struct socket *sock, struct > sockaddr *uaddr, int len) > j1939_local_ecu_put(priv, jsk->addr.src_name, jsk->addr.sa); > } else { > struct net_device *ndev; > + struct can_ml_priv *can_ml; nit: rev xmas tree > > ndev = dev_get_by_index(net, addr->can_ifindex); > if (!ndev) {
[PATCH net v3] net: introduce CAN specific pointer in the struct net_device
Since 20dd3850bcf8 ("can: Speed up CAN frame receiption by using ml_priv") the CAN framework uses per device specific data in the AF_CAN protocol. For this purpose the struct net_device->ml_priv is used. Later the ml_priv usage in CAN was extended for other users, one of them being CAN_J1939. Later in the kernel ml_priv was converted to an union, used by other drivers. E.g. the tun driver started storing it's stats pointer. Since tun devices can claim to be a CAN device, CAN specific protocols will wrongly interpret this pointer, which will cause system crashes. Mostly this issue is visible in the CAN_J1939 stack. To fix this issue, we request a dedicated CAN pointer within the net_device struct. Reported-by: syzbot+5138c4dd15a0401be...@syzkaller.appspotmail.com Fixes: 20dd3850bcf8 ("can: Speed up CAN frame receiption by using ml_priv") Fixes: ffd956eef69b ("can: introduce CAN midlayer private and allocate it automatically") Fixes: 9d71dd0c7009 ("can: add support of SAE J1939 protocol") Fixes: 497a5757ce4e ("tun: switch to net core provided statistics counters") Signed-off-by: Oleksij Rempel --- drivers/net/can/dev/dev.c | 4 +++- drivers/net/can/slcan.c| 4 +++- drivers/net/can/vcan.c | 2 +- drivers/net/can/vxcan.c| 6 +- include/linux/can/can-ml.h | 12 include/linux/netdevice.h | 38 +- net/can/af_can.c | 34 ++ net/can/j1939/main.c | 22 -- net/can/j1939/socket.c | 13 - net/can/proc.c | 19 +-- 10 files changed, 88 insertions(+), 66 deletions(-) diff --git a/drivers/net/can/dev/dev.c b/drivers/net/can/dev/dev.c index d9281ae853f8..311d8564d611 100644 --- a/drivers/net/can/dev/dev.c +++ b/drivers/net/can/dev/dev.c @@ -239,6 +239,7 @@ void can_setup(struct net_device *dev) struct net_device *alloc_candev_mqs(int sizeof_priv, unsigned int echo_skb_max, unsigned int txqs, unsigned int rxqs) { + struct can_ml_priv *can_ml; struct net_device *dev; struct can_priv *priv; int size; @@ -270,7 +271,8 @@ struct net_device *alloc_candev_mqs(int sizeof_priv, unsigned int echo_skb_max, priv = netdev_priv(dev); priv->dev = dev; - dev->ml_priv = (void *)priv + ALIGN(sizeof_priv, NETDEV_ALIGN); + can_ml = (void *)priv + ALIGN(sizeof_priv, NETDEV_ALIGN); + can_set_ml_priv(dev, can_ml); if (echo_skb_max) { priv->echo_skb_max = echo_skb_max; diff --git a/drivers/net/can/slcan.c b/drivers/net/can/slcan.c index a1bd1be09548..30c8d53c9745 100644 --- a/drivers/net/can/slcan.c +++ b/drivers/net/can/slcan.c @@ -516,6 +516,7 @@ static struct slcan *slc_alloc(void) int i; char name[IFNAMSIZ]; struct net_device *dev = NULL; + struct can_ml_priv *can_ml; struct slcan *sl; int size; @@ -538,7 +539,8 @@ static struct slcan *slc_alloc(void) dev->base_addr = i; sl = netdev_priv(dev); - dev->ml_priv = (void *)sl + ALIGN(sizeof(*sl), NETDEV_ALIGN); + can_ml = (void *)sl + ALIGN(sizeof(*sl), NETDEV_ALIGN); + can_set_ml_priv(dev, can_ml); /* Initialize channel control data */ sl->magic = SLCAN_MAGIC; diff --git a/drivers/net/can/vcan.c b/drivers/net/can/vcan.c index 39ca14b0585d..067705e2850b 100644 --- a/drivers/net/can/vcan.c +++ b/drivers/net/can/vcan.c @@ -153,7 +153,7 @@ static void vcan_setup(struct net_device *dev) dev->addr_len = 0; dev->tx_queue_len = 0; dev->flags = IFF_NOARP; - dev->ml_priv= netdev_priv(dev); + can_set_ml_priv(dev, netdev_priv(dev)); /* set flags according to driver capabilities */ if (echo) diff --git a/drivers/net/can/vxcan.c b/drivers/net/can/vxcan.c index f9a524c5f6d6..8861a7d875e7 100644 --- a/drivers/net/can/vxcan.c +++ b/drivers/net/can/vxcan.c @@ -141,6 +141,8 @@ static const struct net_device_ops vxcan_netdev_ops = { static void vxcan_setup(struct net_device *dev) { + struct can_ml_priv *can_ml; + dev->type = ARPHRD_CAN; dev->mtu= CANFD_MTU; dev->hard_header_len= 0; @@ -149,7 +151,9 @@ static void vxcan_setup(struct net_device *dev) dev->flags = (IFF_NOARP|IFF_ECHO); dev->netdev_ops = &vxcan_netdev_ops; dev->needs_free_netdev = true; - dev->ml_priv= netdev_priv(dev) + ALIGN(sizeof(struct vxcan_priv), NETDEV_ALIGN); + + can_ml = netdev_priv(dev) + ALIGN(sizeof(struct vxcan_priv), NETDEV_ALIGN); + can_set_ml_priv(dev, can_ml); } /* forward declaration for rtnl_create_link() */ diff --git a/include/linux/can/can-ml.h b/include/linux/can/can-ml.h index 2f5d731ae251..8afa92d15a66 100644 --- a/include/linux/can/can-ml.h +++ b/include/linux/can/