ethtool: set_channels: add a few more checks
Jakub Kicinski
maintainercover ethtool: set_channels: add a few more checks
There seems to be a few more things we can check in the core before we call drivers' ethtool_ops->set_channels. Adding the checks to the core simplifies the drivers. This set only includes changes to the NFP driver as an example.
There is a small risk in the first patch that someone actually purposefully accepts a strange configuration without RX or TX channels, but I couldn't find such a driver in the tree.
Jakub Kicinski (3):
ethtool: check if there is at least one channel for TX/RX in the core
nfp: don't check lack of RX/TX channels
ethtool: don't call set_channels in drivers if config didn't change
.../ethernet/netronome/nfp/nfp_net_ethtool.c | 3 +-- net/ethtool/channels.c | 20 +++++++++++++++++-- net/ethtool/ioctl.c | 11 ++++++++++ 3 files changed, 30 insertions(+), 4 deletions(-)
Signature · 1 line
2.25.4
Jakub Kicinski
maintainerPATCH 1/3 ethtool: check if there is at least one channel for TX/RX in the core
Having a channel config with no ability to RX or TX traffic is clearly wrong. Check for this in the core so the drivers don't have to.
--- net/ethtool/channels.c | 20 ++++++++++++++++++-- net/ethtool/ioctl.c | 5 +++++ 2 files changed, 23 insertions(+), 2 deletions(-)
net/ethtool/channels.c +18 -2
@@ -129,13 +129,13 @@ int ethnl_set_channels(struct sk_buff *skb, struct genl_info *info) {struct nlattr *tb[ETHTOOL_A_CHANNELS_MAX +1 ];unsigned int from_channel, old_total, i; +bool mod = false, mod_combined = false;struct ethtool_channels channels = {};struct ethnl_req_info req_info = {};const struct nlattr *err_attr;const struct ethtool_ops *ops;struct net_device *dev;u32 max_rx_in_use =0 ; -bool mod = false;int ret; ret = nlmsg_parse(info->nlhdr, GENL_HDRLEN, tb,
@@ -170,7 +170,8 @@ int ethnl_set_channels(struct sk_buff *skb, struct genl_info *info)ethnl_update_u32 (&channels.other_count, tb[ETHTOOL_A_CHANNELS_OTHER_COUNT ], &mod);ethnl_update_u32 (&channels.combined_count, - tb[ETHTOOL_A_CHANNELS_COMBINED_COUNT ], &mod); + tb[ETHTOOL_A_CHANNELS_COMBINED_COUNT ], &mod_combined); + mod |= mod_combined; ret =0 ;if (!mod) goto out_ops;
@@ -193,6 +194,21 @@ int ethnl_set_channels(struct sk_buff *skb, struct genl_info *info) goto out_ops; } +/* ensure there is at least one RX and one TX channel */ +if (!channels.combined_count && !channels.rx_count) + err_attr = tb[ETHTOOL_A_CHANNELS_RX_COUNT ]; +else if (!channels.combined_count && !channels.tx_count) + err_attr = tb[ETHTOOL_A_CHANNELS_TX_COUNT ]; +else + err_attr =NULL ; +if (err_attr) { +if (mod_combined) + err_attr = tb[ETHTOOL_A_CHANNELS_COMBINED_COUNT ]; + ret = -EINVAL ; +NL_SET_ERR_MSG_ATTR (info->extack, err_attr,"requested channel counts would result in no RX or TX channel being configured" ); + goto out_ops; + } +/* ensure the new Rx count fits within the configured Rx flow * indirection table settings */
net/ethtool/ioctl.c +5 -0
@@ -1676,6 +1676,11 @@ static noinline_for_stack int ethtool_set_channels(struct net_device *dev, channels.other_count > curr.max_other)return -EINVAL ; +/* ensure there is at least one RX and one TX channel */ +if (!channels.combined_count && + (!channels.rx_count || !channels.tx_count)) +return -EINVAL ; +/* ensure the new Rx count fits within the configured Rx flow * indirection table settings */ if (netif_is_rxfh_configured(dev) &&
Signature · 1 line
2.25.4
Re: [PATCH net-next 1/3] ethtool: check if there is at least one channel for TX/RX in the core
Quoted 5 lines · Jakub Kicinski
Having a channel config with no ability to RX or TX traffic is clearly wrong. Check for this in the core so the drivers don't have to.
Signed-off-by Jakub Kicinski <[email protected]>
Only one formal problem:
net/ethtool/channels.c int ethnl_set_channels(struct sk_buff *skb, struct genl_info *info) · line 177
Quoted 10 lines
net/ethtool/channels.c +2
-1@@ -170,7 +170,8 @@ int ethnl_set_channels(struct sk_buff *skb, struct genl_info *info)ethnl_update_u32 (&channels.other_count, tb[ETHTOOL_A_CHANNELS_OTHER_COUNT ], &mod);ethnl_update_u32 (&channels.combined_count, - tb[ETHTOOL_A_CHANNELS_COMBINED_COUNT ], &mod); + tb[ETHTOOL_A_CHANNELS_COMBINED_COUNT ], &mod_combined); + mod |= mod_combined; ret =0 ;if (!mod) goto out_ops;
Bitwise or should do the right thing but using "|=" on bool variables looks strange.
Michal
Jakub Kicinski
maintainerRe: [PATCH net-next 1/3] ethtool: check if there is at least one channel for TX/RX in the core
Quoted 8 lines · On Fri, 15 May 2020 22:56:56 +0200 Michal Kubecek wrote:
Quoted 5 lines · Jakub Kicinski
Having a channel config with no ability to RX or TX traffic is clearly wrong. Check for this in the core so the drivers don't have to.
Signed-off-by Jakub Kicinski <[email protected]>Reviewed-by Michal Kubecek <[email protected]>
Thanks for the reviews!
Quoted 5 lines
Quoted 5 lines
net/ethtool/channels.c +0
-1@@ -170,7 +170,8 @@ int ethnl_set_channels(struct sk_buff *skb, struct genl_info *info)ethnl_update_u32 (&channels.other_count, tb[ETHTOOL_A_CHANNELS_OTHER_COUNT ], &mod); ethnl_update_u32(&channels.combined_count, - tb[ETHTOOL_A_CHANNELS_COMBINED_COUNT ], &mod);
Quoted 8 lines
net/ethtool/channels.c int ethnl_set_channels(struct sk_buff *skb, struct genl_info *info) · line 177
Quoted 5 lines
net/ethtool/channels.c +2
-0+ tb[ETHTOOL_A_CHANNELS_COMBINED_COUNT], &mod_combined); + mod |= mod_combined; ret =0 ;if (!mod) goto out_ops;Bitwise or should do the right thing but using "|=" on bool variables looks strange.
Interesting, I never thought about it in this way.
Jakub Kicinski
maintainerPATCH 2/3 nfp: don't check lack of RX/TX channels
In reply to Jakub KicinskiCore will now perform this check.
--- drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c +1 -2
@@ -1438,8 +1438,7 @@ static int nfp_net_set_channels(struct net_device *netdev,unsigned int total_rx, total_tx;/* Reject unsupported */ -if (!channel->combined_count || - channel->other_count !=NFP_NET_NON_Q_VECTORS || +if (channel->other_count !=NFP_NET_NON_Q_VECTORS || (channel->rx_count && channel->tx_count))return -EINVAL ;
Signature · 1 line
2.25.4
Re: [PATCH net-next 2/3] nfp: don't check lack of RX/TX channels
Quoted 3 lines · Jakub Kicinski
Core will now perform this check.
Signed-off-by Jakub Kicinski <[email protected]>
Quoted 21 lines
--- drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c +1
-2@@ -1438,8 +1438,7 @@ static int nfp_net_set_channels(struct net_device *netdev,unsigned int total_rx, total_tx;/* Reject unsupported */ -if (!channel->combined_count || - channel->other_count !=NFP_NET_NON_Q_VECTORS || +if (channel->other_count !=NFP_NET_NON_Q_VECTORS || (channel->rx_count && channel->tx_count))return -EINVAL ;Signature · 1 line
2.25.4
Re: [PATCH net-next 2/3] nfp: don't check lack of RX/TX channels
In reply to Jakub KicinskiQuoted 3 lines · Jakub Kicinski
Core will now perform this check.
Signed-off-by Jakub Kicinski <[email protected]>
Quoted 21 lines
--- drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c +1
-2@@ -1438,8 +1438,7 @@ static int nfp_net_set_channels(struct net_device *netdev,unsigned int total_rx, total_tx;/* Reject unsupported */ -if (!channel->combined_count || - channel->other_count !=NFP_NET_NON_Q_VECTORS || +if (channel->other_count !=NFP_NET_NON_Q_VECTORS || (channel->rx_count && channel->tx_count))return -EINVAL ;Signature · 1 line
2.25.4
Jakub Kicinski
maintainerPATCH 3/3 ethtool: don't call set_channels in drivers if config didn't change
In reply to Jakub KicinskiDon't call drivers if nothing changed. Netlink code already contains this logic.
--- net/ethtool/ioctl.c | 6 ++++++ 1 file changed, 6 insertions(+)
net/ethtool/ioctl.c +6 -0
@@ -1669,6 +1669,12 @@ static noinline_for_stack int ethtool_set_channels(struct net_device *dev, dev->ethtool_ops->get_channels (dev, &curr); +if (channels.rx_count == curr.rx_count && + channels.tx_count == curr.tx_count && + channels.combined_count == curr.combined_count && + channels.other_count == curr.other_count) +return 0 ; +/* ensure new counts are within the maximums */ if (channels.rx_count > curr.max_rx || channels.tx_count > curr.max_tx ||
Signature · 1 line
2.25.4
Re: [PATCH net-next 3/3] ethtool: don't call set_channels in drivers if config didn't change
Quoted 4 lines · Jakub Kicinski
Don't call drivers if nothing changed. Netlink code already contains this logic.
Signed-off-by Jakub Kicinski <[email protected]>
Quoted 24 lines
--- net/ethtool/ioctl.c | 6 ++++++ 1 file changed, 6 insertions(+)net/ethtool/ioctl.c +6
-0@@ -1669,6 +1669,12 @@ static noinline_for_stack int ethtool_set_channels(struct net_device *dev, dev->ethtool_ops->get_channels (dev, &curr); +if (channels.rx_count == curr.rx_count && + channels.tx_count == curr.tx_count && + channels.combined_count == curr.combined_count && + channels.other_count == curr.other_count) +return 0 ; +/* ensure new counts are within the maximums */ if (channels.rx_count > curr.max_rx || channels.tx_count > curr.max_tx ||Signature · 1 line
2.25.4
David Miller
maintainerFrom: Jakub Kicinski <[email protected]>
Date: Fri, 15 May 2020 12:48:59 -0700
Quoted 8 lines
There seems to be a few more things we can check in the core before we call drivers'
ethtool_ops->set_channels. Adding the checks to the core simplifies the drivers. This set only includes changes to the NFP driver as an example.There is a small risk in the first patch that someone actually purposefully accepts a strange configuration without RX or TX channels, but I couldn't find such a driver in the tree.
Series applied, thanks Jakub.
And for the record I accept logical 'or' of booleans as valid :-)
Quoted 8 lines · David Miller
From: Jakub Kicinski <[email protected]>
Date: Fri, 15 May 2020 12:48:59 -0700Quoted 5 lines
There seems to be a few more things we can check in the core before we call drivers'
ethtool_ops->set_channels. Adding the checks to the core simplifies the drivers. This set only includes changes to the NFP driver as an example.
Quoted 7 lines
Quoted 3 lines
There is a small risk in the first patch that someone actually purposefully accepts a strange configuration without RX or TX channels, but I couldn't find such a driver in the tree.
Series applied, thanks Jakub.
And for the record I accept logical 'or' of booleans as valid :-)
For the record, my point was that "|=" is a bitwise operator, not logical.
Michal