On 06/16, Jay Vosburgh wrote:

Stanislav Fomichev [off-list ref] wrote:

quoted

On 06/16, Jay Vosburgh wrote:

quoted

	 Remove the ability to disable use_carrier in bonding, and remove
all code related to the old link state check that utilizes ethtool or
ioctl to determine the link state of an interface in a bond.

	To avoid acquiring RTNL many times per second, bonding's miimon
link monitor inspects link state under RCU, but not under RTNL.  However,
ethtool implementations in drivers may sleep, and therefore the ethtool or
ioctl strategy is unsuitable for use with calls into driver ethtool
functions.

	The use_carrier option was introduced in 2003, to provide
backwards compatibility for network device drivers that did not support
the then-new netif_carrier_ok/on/off system.  Today, device drivers are
expected to support netif_carrier_*, and the use_carrier backwards
compatibility logic is no longer necessary.

	Bonding now always behaves as if use_carrier=1, which relies on
netif_carrier_ok() to determine the link state of interfaces.  This has
been the default setting for use_carrier since its introduction.  For
backwards compatibility, the option itself remains, but may only be set to
1, and queries will always return 1.

Reported-by: syzbot+b8c48ea38ca27d150063@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=b8c48ea38ca27d150063
Link: https://lore.kernel.org/lkml/000000000000eb54bf061cfd666a@google.com/ (local)
Link: https://lore.kernel.org/netdev/20240718122017.d2e33aaac43a.I10ab9c9ded97163aef4e4de10985cd8f7de60d28@changeid/ (local)
Link: http://lore.kernel.org/netdev/aEt6LvBMwUMxmUyx@mini-arch (local)
Signed-off-by: Jay Vosburgh <jv@jvosburgh.net>

Acked-by: Stanislav Fomichev <sdf@fomichev.me>

Maybe better to target 'net' with the following?
Fixes: f7a11cba0ed7 ("bonding: hold ops lock around get_link")

	I targeted net-next and left the Fixes: tag off on purpose.

	First, the bug this nominally fixes is many years old, and
wasn't introduced by f7a11cba0ed7.

	More importantly, though, this patch is removing functionality
that someone theoretically could be relying on, and I don't think such
removals should happen in the middle of a stable series.  The default
setting for use_carrier (i.e., using netif_carrier) will never hit the
issue in practice, so the exposure seems to be minimal for common use.

SG, especially assuming that use_carrier has a default of 1 (so the
issue should not appear in most/default setups).