On Mon, Dec 21, 2020 at 11:36 AM Antoine Tenart <aten...@kernel.org> wrote:
>
> Two race conditions can be triggered in xps, resulting in various oops
> and invalid memory accesses:
>
> 1. Calling netdev_set_num_tc while netif_set_xps_queue:
>
> - netdev_set_num_tc sets dev->tc_num.
>
> - netif_set_xps_queue uses dev->tc_num as one of the parameters to
> compute the size of new_dev_maps when allocating it. dev->tc_num is
> also used to access the map, and the compiler may generate code to
> retrieve this field multiple times in the function.
>
> If new_dev_maps is allocated using dev->tc_num and then dev->tc_num
> is set to a higher value through netdev_set_num_tc, later accesses to
> new_dev_maps in netif_set_xps_queue could lead to accessing memory
> outside of new_dev_maps; triggering an oops.
>
> One way of triggering this is to set an iface up (for which the
> driver uses netdev_set_num_tc in the open path, such as bnx2x) and
> writing to xps_cpus or xps_rxqs in a concurrent thread. With the
> right timing an oops is triggered.
>
> 2. Calling netif_set_xps_queue while netdev_set_num_tc is running:
>
> 2.1. netdev_set_num_tc starts by resetting the xps queues,
> dev->tc_num isn't updated yet.
>
> 2.2. netif_set_xps_queue is called, setting up the maps with the
> *old* dev->num_tc.
>
> 2.3. dev->tc_num is updated.
>
> 2.3. Later accesses to the map leads to out of bound accesses and
> oops.
>
> A similar issue can be found with netdev_reset_tc.
>
> The fix can't be to only link the size of the maps to them, as
> invalid configuration could still occur. The reset then set logic in
> both netdev_set_num_tc and netdev_reset_tc must be protected by a
> lock.
>
> Both issues have the same fix: netif_set_xps_queue, netdev_set_num_tc
> and netdev_reset_tc should be mutually exclusive.
>
> This patch fixes those races by:
>
> - Reworking netif_set_xps_queue by moving the xps_map_mutex up so the
> access of dev->num_tc is done under the lock.
>
> - Using xps_map_mutex in both netdev_set_num_tc and netdev_reset_tc for
> the reset and set logic:
>
> + As xps_map_mutex was taken in the reset path, netif_reset_xps_queues
> had to be reworked to offer an unlocked version (as well as
> netdev_unbind_all_sb_channels which calls it).
>
> + cpus_read_lock was taken in the reset path as well, and is always
> taken before xps_map_mutex. It had to be moved out of the unlocked
> version as well.
>
> This is why the patch is a little bit longer, and moves
> netdev_unbind_sb_channel up in the file.
>
> Fixes: 184c449f91fe ("net: Add support for XPS with QoS via traffic classes")
> Signed-off-by: Antoine Tenart <aten...@kernel.org>
Looking over this patch it seems kind of obvious that extending the
xps_map_mutex is making things far more complex then they need to be.
Applying the rtnl_mutex would probably be much simpler. Although as I
think you have already discovered we need to apply it to the store,
and show for this interface. In addition we probably need to perform
similar locking around traffic_class_show in order to prevent it from
generating a similar error.
