Netdev List
 help / color / mirror / Atom feed
From: Stanislav Fomichev <sdf.kernel@gmail.com>
To: Jakub Kicinski <kuba@kernel.org>
Cc: davem@davemloft.net, netdev@vger.kernel.org, edumazet@google.com,
	 pabeni@redhat.com, andrew+netdev@lunn.ch, horms@kernel.org,
	 michael.chan@broadcom.com, joshwash@google.com,
	tariqt@nvidia.com, haiyangz@microsoft.com,
	 linux@armlinux.org.uk, maxime.chevallier@bootlin.com,
	willemb@google.com,  ernis@linux.microsoft.com,
	kory.maincent@bootlin.com, danieller@nvidia.com,
	 idosch@nvidia.com
Subject: Re: [PATCH net-next 13/14] net: ethtool: optionally skip rtnl_lock on IOCTL path
Date: Mon, 1 Jun 2026 08:17:56 -0700	[thread overview]
Message-ID: <ah2hvWMz70qAYmsG@devvm7509.cco0.facebook.com> (raw)
In-Reply-To: <20260528231637.251822-14-kuba@kernel.org>

On 05/28, Jakub Kicinski wrote:
> Convert the IOCTL path similarly to how we converted Netlink.
> The device lookup gets a little hairy. We could take rtnl_lock
> unconditionally and drop it before calling the driver (this would
> avoid the reference + liveness check). But I think being able
> to make progress even if rtnl is dead-locked is quite useful.
> 
> First extra concern is handling features. List all the cmds which
> modify features and always take rtnl_lock. We could fold this list
> into ethtool_ioctl_needs_rtnl() but seems cleaner to keep
> ethtool_ioctl_needs_rtnl() driver-related. If a driver changed
> features and we were not holding rtnl_lock - warn about it.
> It can only happen on buggy ops locked drivers (buggy because
> they should have set appropriate "I need rtnl for op X" bit).
> 
> Second wrinkle is the PHY ID hack which drops the locks while
> sleeping. Convert its static "busy" variable which used to
> be protected by rtnl_lock to a field in struct ethtool_netdev_state.
> This feature is about identifying an adapter or a port within
> a system, so being able to blink multiple LEDs at the same
> time is likely not very useful in practice. But it's the simplest
> fix, we can add a mutex if someone thinks a system should only
> be ID'ing one port at a time.
> 
> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
> ---
>  include/linux/ethtool.h |  2 +
>  net/ethtool/ioctl.c     | 98 ++++++++++++++++++++++++++++++-----------
>  2 files changed, 74 insertions(+), 26 deletions(-)
> 
> diff --git a/include/linux/ethtool.h b/include/linux/ethtool.h
> index 4f15221119e2..35ee57a0e5fa 100644
> --- a/include/linux/ethtool.h
> +++ b/include/linux/ethtool.h
> @@ -1375,6 +1375,7 @@ int ethtool_virtdev_set_link_ksettings(struct net_device *dev,
>   *			within RTNL.
>   * @rss_indir_user_size: Number of user provided entries for the default
>   *			 (context 0) indirection table.
> + * @phys_id_busy:	Loop blinking the device LED is running.
>   * @wol_enabled:	Wake-on-LAN is enabled
>   * @module_fw_flash_in_progress: Module firmware flashing is in progress.
>   */
> @@ -1382,6 +1383,7 @@ struct ethtool_netdev_state {
>  	struct xarray		rss_ctx;
>  	struct mutex		rss_lock;
>  	u32			rss_indir_user_size;
> +	unsigned		phys_id_busy:1;
>  	unsigned		wol_enabled:1;
>  	unsigned		module_fw_flash_in_progress:1;
>  };
> diff --git a/net/ethtool/ioctl.c b/net/ethtool/ioctl.c
> index 6c3a7e8644ae..aea087d62fe9 100644
> --- a/net/ethtool/ioctl.c
> +++ b/net/ethtool/ioctl.c
> @@ -543,7 +543,7 @@ static int ethtool_get_link_ksettings(struct net_device *dev,
>  	int err = 0;
>  	struct ethtool_link_ksettings link_ksettings;
>  
> -	ASSERT_RTNL();
> +	netdev_ops_assert_locked(dev);
>  	if (!dev->ethtool_ops->get_link_ksettings)
>  		return -EOPNOTSUPP;
>  
> @@ -600,7 +600,7 @@ static int ethtool_set_link_ksettings(struct net_device *dev,
>  	struct ethtool_link_ksettings link_ksettings = {};
>  	int err;
>  
> -	ASSERT_RTNL();
> +	netdev_ops_assert_locked(dev);
>  
>  	if (!dev->ethtool_ops->set_link_ksettings)
>  		return -EOPNOTSUPP;
> @@ -674,7 +674,7 @@ static int ethtool_get_settings(struct net_device *dev, void __user *useraddr)
>  	struct ethtool_cmd cmd;
>  	int err;
>  
> -	ASSERT_RTNL();
> +	netdev_ops_assert_locked(dev);
>  	if (!dev->ethtool_ops->get_link_ksettings)
>  		return -EOPNOTSUPP;
>  
> @@ -710,7 +710,7 @@ static int ethtool_set_settings(struct net_device *dev, void __user *useraddr)
>  	struct ethtool_cmd cmd;
>  	int ret;
>  
> -	ASSERT_RTNL();
> +	netdev_ops_assert_locked(dev);
>  
>  	if (copy_from_user(&cmd, useraddr, sizeof(cmd)))
>  		return -EFAULT;
> @@ -2451,10 +2451,10 @@ void ethtool_puts(u8 **data, const char *str)
>  }
>  EXPORT_SYMBOL(ethtool_puts);
>  
> -static int ethtool_phys_id(struct net_device *dev, void __user *useraddr)
> +static int ethtool_phys_id(struct net_device *dev, void __user *useraddr,
> +			   bool has_rtnl_lock)
>  {
>  	struct ethtool_value id;
> -	static bool busy;
>  	const struct ethtool_ops *ops = dev->ethtool_ops;
>  	netdevice_tracker dev_tracker;
>  	int rc;
> @@ -2462,7 +2462,7 @@ static int ethtool_phys_id(struct net_device *dev, void __user *useraddr)
>  	if (!ops->set_phys_id)
>  		return -EOPNOTSUPP;
>  
> -	if (busy)
> +	if (dev->ethtool->phys_id_busy)
>  		return -EBUSY;
>  
>  	if (copy_from_user(&id, useraddr, sizeof(id)))
> @@ -2472,13 +2472,14 @@ static int ethtool_phys_id(struct net_device *dev, void __user *useraddr)
>  	if (rc < 0)
>  		return rc;
>  
> -	/* Drop the RTNL lock while waiting, but prevent reentry or
> +	/* Drop the locks while waiting, but prevent reentry or
>  	 * removal of the device.
>  	 */
> -	busy = true;
> +	dev->ethtool->phys_id_busy = true;
>  	netdev_hold(dev, &dev_tracker, GFP_KERNEL);
>  	netdev_unlock_ops(dev);
> -	rtnl_unlock();
> +	if (has_rtnl_lock)
> +		rtnl_unlock();
>  
>  	if (rc == 0) {
>  		/* Driver will handle this itself */
> @@ -2491,22 +2492,25 @@ static int ethtool_phys_id(struct net_device *dev, void __user *useraddr)
>  		u64 i = 0;
>  
>  		do {
> -			rtnl_lock();
> +			if (has_rtnl_lock)
> +				rtnl_lock();
>  			netdev_lock_ops(dev);
>  			rc = ops->set_phys_id(dev,
>  				    (i++ & 1) ? ETHTOOL_ID_OFF : ETHTOOL_ID_ON);
>  			netdev_unlock_ops(dev);
> -			rtnl_unlock();
> +			if (has_rtnl_lock)
> +				rtnl_unlock();
>  			if (rc)
>  				break;
>  			schedule_timeout_interruptible(interval);
>  		} while (!signal_pending(current) && (!id.data || i < count));
>  	}
>  
> -	rtnl_lock();
> +	if (has_rtnl_lock)
> +		rtnl_lock();
>  	netdev_lock_ops(dev);
>  	netdev_put(dev, &dev_tracker);
> -	busy = false;
> +	dev->ethtool->phys_id_busy = false;
>  
>  	(void) ops->set_phys_id(dev, ETHTOOL_ID_INACTIVE);
>  	return rc;
> @@ -3259,7 +3263,8 @@ static int ethtool_set_fecparam(struct net_device *dev, void __user *useraddr)
>  static int
>  dev_ethtool_locked(struct net *net, struct net_device *dev,
>  		   void __user *useraddr,
> -		   u32 ethcmd, struct ethtool_devlink_compat *devlink_state)
> +		   u32 ethcmd, struct ethtool_devlink_compat *devlink_state,
> +		   bool has_rtnl_lock)
>  {
>  	u32 sub_cmd;
>  	int rc;
> @@ -3315,6 +3320,8 @@ dev_ethtool_locked(struct net *net, struct net_device *dev,
>  			return -EPERM;
>  	}
>  
> +	netdev_ops_assert_locked(dev);
> +
>  	if (dev->dev.parent)
>  		pm_runtime_get_sync(dev->dev.parent);
>  
> @@ -3402,7 +3409,7 @@ dev_ethtool_locked(struct net *net, struct net_device *dev,
>  		rc = ethtool_get_strings(dev, useraddr);
>  		break;
>  	case ETHTOOL_PHYS_ID:
> -		rc = ethtool_phys_id(dev, useraddr);
> +		rc = ethtool_phys_id(dev, useraddr, has_rtnl_lock);
>  		break;
>  	case ETHTOOL_GSTATS:
>  		rc = ethtool_get_stats(dev, useraddr);
> @@ -3549,8 +3556,12 @@ dev_ethtool_locked(struct net *net, struct net_device *dev,
>  	if (dev->ethtool_ops->complete)
>  		dev->ethtool_ops->complete(dev);
>  
> -	if (old_features != dev->features)
> -		netdev_features_change(dev);
> +	if (old_features != dev->features) {
> +		if (has_rtnl_lock)
> +			netdev_features_change(dev);
> +		else
> +			netdev_WARN(dev, "unexpected device features change with ethtool cmd %u", ethcmd);
> +	}
>  out:
>  	if (dev->dev.parent)
>  		pm_runtime_put(dev->dev.parent);
> @@ -3558,25 +3569,60 @@ dev_ethtool_locked(struct net *net, struct net_device *dev,
>  	return rc;
>  }
>  
> +/* Commands that may toggle dev->features in net/ethtool/ioctl.c and so
> + * call into __netdev_update_features(), which still requires rtnl_lock.
> + * Driver-decided SET commands that may chain into rtnl-only helpers are
> + * covered by ethtool_ioctl_needs_rtnl()/ETHTOOL_OP_NEEDS_RTNL_*.
> + */
> +static bool ethtool_cmd_changes_features(u32 ethcmd)
> +{
> +	switch (ethcmd) {
> +	case ETHTOOL_SFEATURES:
> +	case ETHTOOL_SFLAGS:
> +	case ETHTOOL_STXCSUM:
> +	case ETHTOOL_SRXCSUM:
> +	case ETHTOOL_SSG:
> +	case ETHTOOL_STSO:
> +	case ETHTOOL_SGSO:
> +	case ETHTOOL_SGRO:
> +		return true;

GVE's gve_get_ethtool_stats (ETHTOOL_GSTATS) still has ASSERT_RTNL and
is now reachable with only ops lock?

  reply	other threads:[~2026-06-01 15:17 UTC|newest]

Thread overview: 21+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-05-28 23:16 [PATCH net-next 00/14] net: ethtool: let ops locked drivers run without rtnl_lock Jakub Kicinski
2026-05-28 23:16 ` [PATCH net-next 01/14] net: ethtool: cmis_cdb: hold instance lock for ops locked devices Jakub Kicinski
2026-05-29 11:25   ` Jakub Sitnicki
2026-05-28 23:16 ` [PATCH net-next 02/14] net: ethtool: make sure __ethtool_get_link_ksettings() is ops-locked Jakub Kicinski
2026-05-28 23:16 ` [PATCH net-next 03/14] net: ethtool: serialize broadcast notification sequence allocation Jakub Kicinski
2026-05-28 23:16 ` [PATCH net-next 04/14] net: ethtool: relax ethnl_req_get_phydev() locking assertion Jakub Kicinski
2026-05-29  8:43   ` Maxime Chevallier
2026-05-29 14:27     ` Jakub Kicinski
2026-05-28 23:16 ` [PATCH net-next 05/14] net: ethtool: make dev->hwprov ops-protected Jakub Kicinski
2026-05-28 23:16 ` [PATCH net-next 06/14] net: ethtool: optionally skip rtnl_lock on Netlink path for GET ops Jakub Kicinski
2026-05-28 23:16 ` [PATCH net-next 07/14] net: ethtool: optionally skip rtnl_lock on Netlink path for SET ops Jakub Kicinski
2026-05-28 23:16 ` [PATCH net-next 08/14] net: ethtool: optionally skip rtnl_lock in cable test handlers Jakub Kicinski
2026-05-28 23:16 ` [PATCH net-next 09/14] net: ethtool: optionally skip rtnl_lock in ethnl_tsinfo_dumpit() Jakub Kicinski
2026-05-28 23:16 ` [PATCH net-next 10/14] net: ethtool: optionally skip rtnl_lock in ethnl_act_module_fw_flash() Jakub Kicinski
2026-05-28 23:16 ` [PATCH net-next 11/14] net: ethtool: optionally skip rtnl_lock in RSS context handlers Jakub Kicinski
2026-05-28 23:16 ` [PATCH net-next 12/14] net: ethtool: ioctl: concentrate the locking Jakub Kicinski
2026-05-28 23:16 ` [PATCH net-next 13/14] net: ethtool: optionally skip rtnl_lock on IOCTL path Jakub Kicinski
2026-06-01 15:17   ` Stanislav Fomichev [this message]
2026-06-01 19:10     ` Jakub Kicinski
2026-05-28 23:16 ` [PATCH net-next 14/14] docs: net: ethtool: document ops-locked drivers and op_needs_rtnl Jakub Kicinski
2026-05-29  7:41 ` [syzbot ci] Re: net: ethtool: let ops locked drivers run without rtnl_lock syzbot ci

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=ah2hvWMz70qAYmsG@devvm7509.cco0.facebook.com \
    --to=sdf.kernel@gmail.com \
    --cc=andrew+netdev@lunn.ch \
    --cc=danieller@nvidia.com \
    --cc=davem@davemloft.net \
    --cc=edumazet@google.com \
    --cc=ernis@linux.microsoft.com \
    --cc=haiyangz@microsoft.com \
    --cc=horms@kernel.org \
    --cc=idosch@nvidia.com \
    --cc=joshwash@google.com \
    --cc=kory.maincent@bootlin.com \
    --cc=kuba@kernel.org \
    --cc=linux@armlinux.org.uk \
    --cc=maxime.chevallier@bootlin.com \
    --cc=michael.chan@broadcom.com \
    --cc=netdev@vger.kernel.org \
    --cc=pabeni@redhat.com \
    --cc=tariqt@nvidia.com \
    --cc=willemb@google.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox