Store the RSS key inside struct igb_adapter and introduce the
igb_write_rss_key() helper function. This allows the driver to program
the E1000 registers using a persistent RSS key, instead of using a
stack-local buffer in igb_setup_mrqc().
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Piotr Kwapulinski <piotr.kwapulinski@intel.com>
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Signed-off-by: Takashi Kozu <takkozu@amazon.com>
Tested-by: Rinitha S <sx.rinitha@intel.com> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Change igc_ethtool_set_rxfh() to accept and save a userspace-provided
RSS key. When a key is provided, store it in the adapter and write the
RSSRK registers accordingly.
This can be tested using `ethtool -X <dev> hkey <key>`.
Signed-off-by: Kohei Enju <kohei@enjuk.jp>
Reviewed-by: Simon Horman <horms@kernel.org>
Tested-by: Avigail Dahan <avigailx.dahan@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Store the RSS key inside struct igc_adapter and introduce the
igc_write_rss_key() helper function. This allows the driver to program
the RSSRK registers using a persistent RSS key, instead of using a
stack-local buffer in igc_setup_mrqc().
This is a preparation patch for adding RSS key get/set support in
subsequent changes, and no functional change is intended in this patch.
Signed-off-by: Kohei Enju <kohei@enjuk.jp>
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Tested-by: Avigail Dahan <avigailx.dahan@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Kuniyuki Iwashima says:
====================
net: fib_rules: RTNL-less RTM_NEWRULE and RTM_DELRULE.
RTM_NEWRULE and RTM_DELRULE acquire rtnl_net_lock(), but this is
only for fib_unmerge() in IPv4.
Since commit d954a67a7d ("ipv4: fib_rule: Move fib4_rules_exit()
to ->exit()."), RTM_DELRULE no longer needs RTNL.
fib_unmerge() is one-time event for each netns, so we only need
RTNL for the first IPv4 rule.
This series introduces per-fib_rules_ops mutex and drops RTNL
from fib_rules code except for the first IPv4 RTM_NEWRULE.
The script below creates 1K rules in parallel in 4K netns, and
it got 20x/30x faster for IPv4/IPv6.
#!/bin/bash
N=4096
F=rules.txt
for i in $(seq $N); do ip netns add ns-$i; done
printf 'rule add from all table %d\n' {1..1024} > $F
for v in 4 6; do
echo "=== IPv${v} ==="
time { for i in $(seq $N); do nsenter \
--net=/var/run/netns/ns-$i ip -$v -batch $F & done; wait; }
done
for i in $(seq $N); do ip netns del ns-$i; done
rm -f $F
Without this series:
# ./test.sh
=== IPv4 ===
real 0m22.752s
user 0m7.834s
sys 92m46.721s
=== IPv6 ===
real 0m35.181s
user 0m8.635s
sys 142m30.479s
With this series:
# ./test.sh
=== IPv4 ===
real 0m0.918s
user 0m5.675s
sys 2m7.024s
=== IPv6 ===
real 0m1.214s
user 0m7.917s
sys 4m19.489s
====================
Link: https://patch.msgid.link/20260629181226.1929658-1-kuniyu@google.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Now, RTM_DELRULE no longer needs RTNL, and the only RTNL dependant
in RTM_NEWRULE is fib_unmerge(), which is called for the first
IPv4 rule.
Let's add fib_rules_ops.need_rtnl() and hold RTNL only for the
first IPv4 rule.
Tested:
The script below creates 1K rules in parallel in 4K netns, and
it got 20x/30x faster for IPv4/IPv6.
#!/bin/bash
N=4096
F=rules.txt
for i in $(seq $N); do ip netns add ns-$i; done
printf 'rule add from all table %d\n' {1..1024} > $F
for v in 4 6; do
echo "=== IPv${v} ==="
time { for i in $(seq $N); do nsenter \
--net=/var/run/netns/ns-$i ip -$v -batch $F & done; wait; }
done
for i in $(seq $N); do ip netns del ns-$i; done
rm -f $F
Without this series:
# ./test.sh
=== IPv4 ===
real 0m22.752s
user 0m7.834s
sys 92m46.721s
=== IPv6 ===
real 0m35.181s
user 0m8.635s
sys 142m30.479s
With this series:
# ./test.sh
=== IPv4 ===
real 0m0.918s
user 0m5.675s
sys 2m7.024s
=== IPv6 ===
real 0m1.214s
user 0m7.917s
sys 4m19.489s
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260629181226.1929658-10-kuniyu@google.com
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
We will no longer hold RTNL for RTM_NEWRULE and RMT_DELRULE
except for the first IPv4 RTM_NEWRULE.
Let's covnert __dev_get_by_name() in fib_nl2rule_rtnl() to
dev_get_by_name_rcu() and rename it to fib_nl2rule_locked().
Note that dev_get_by_name_rcu() must be called inside ops->lock
to serialise fib_rules_event() by __dev_change_net_namespace().
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260629181226.1929658-9-kuniyu@google.com
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Now, fib_rule structs are protected by per-fib_rules_ops mutex.
Let's drop ASSERT_RTNL_NET() and rtnl_dereference().
Note that fib_rules_event() iterates over net->rules_ops without
net->rules_mod_lock, but this is fine because all fib_rule users
are built-in and concurrent fib_rules_unregister() does not happen.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260629181226.1929658-8-kuniyu@google.com
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
fib_newrule() will drop RTNL except for the first IPv4 rule.
net->ipv4.fib_table_hash[] will be read with no protection,
but this is fine because fib_table is not destroyed until
netns dismantle except for the merged main/local table.
fib_unmerge() will continue to be called under RTNL, so other
readers (fib_flush() and fib_info_notify_update()) just have
to care about the concurrent hlist_add().
IPv6 and IPMR/IP6MR also take this strategy and use RCU helpers
to avoid data race against concurrent hlist_add().
Let's not use lockdep_rtnl_is_held() and rcu_dereference_rtnl()
for net->ipv4.fib_table_hash[].
Note that commit a7e5353123 ("fib_trie: Make fib_table rcu
safe") started to use the _safe version in fib_flush(), but it
is not needed thanks to RTNL.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260629181226.1929658-5-kuniyu@google.com
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
fib_newrule() will drop RTNL except for the first IPv4 rule.
Then, fib4_rule_configure() could call fib_empty_table() and create
a new IPv4 fib_table without RTNL.
Currently, net->ipv4.fib_table_hash[] is only protected by RTNL.
As a prep, let's protect net->ipv4.fib_table_hash[] with a dedicated
spinlock.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260629181226.1929658-4-kuniyu@google.com
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
IPv4 local and main route tables are merged by default to avoid
unnecessary rule lookups.
When the first IPv4 rule is created, fib_unmerge() splits the
two tables.
However, fib4_rule_configure() currently always calls fib_unmerge(),
and even fetching a table via fib_get_table() requires RTNL (or RCU).
We will drop RTNL from fib_newrule() if not needed.
Let's call fib_unmerge() only once for the first rule.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260629181226.1929658-3-kuniyu@google.com
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Ioana Ciornei says:
====================
dpaa2-switch: add support for LAG offload
This patch set adds support in dpaa2-switch for offloading upper bond
devices.
The first two patches remove the necessity to hold rtnl_lock during the
event processing workqueue by ensuring that all event were processed
before any changes in FDB layout happens.
Patch #3 updates the logic around choosing the FDB that should be used
on a switch port. This is necessary since with the addition of the LAG
offload, we need to take into account all ports which are under the same
bridge, even though not directly.
The next four patches clean up the FDB event by making them easier to
integrate with bond devices and also add the
dpaa2_switch_port_to_bridge_port() helper to be used in the LAG offload
support.
The 8th patch adds the necessary new APIs for the LAG configuration
while the next one uses them, both in the prechangeupper phase and the
changeupper one. Which ports can be part of the same LAG group is
configurable at boot time, thus we use the prechangeupper callback in
order to validate that a requested configuration can be offloaded or
not.
This set also extends the handling of FDBs and port objects so that they
are handled by the driver even on an offloaded bond device.
v3: https://lore.kernel.org/all/20260603143623.3712024-1-ioana.ciornei@nxp.com/
v2: https://lore.kernel.org/all/20260512131554.952971-1-ioana.ciornei@nxp.com/
v1: https://lore.kernel.org/all/20260506151540.1242997-1-ioana.ciornei@nxp.com/
====================
Link: https://patch.msgid.link/20260629112309.154328-1-ioana.ciornei@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Switch ports configured as part of a LAG group are not able to provide
a precise source port for all packets which reach the control interface.
The only frames which will have a precise source port are those that are
explicitly trapped, for example STP and LCAP frames. For any other
frames (for example, those which are flooded) we can only know the
ingress LAG group.
Take into account the DPAA2_ETHSW_FLC_IMPRECISE_IF_ID bit and based on
its value target the bond device or the specific source netdevice.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Link: https://patch.msgid.link/20260629112309.154328-14-ioana.ciornei@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Do not trap only STP frames to the control interface but rather trap all
link local reserved addresses. This will still be done by looking at the
destination MAC address but keeping in mind to not take into account the
last byte.
This change will benefit LACP frames which now will reach the control
interface.
While at it, change the prototype of the
dpaa2_switch_port_trap_mac_addr() function so that we directly pass a
'const u8 *' so that it matches the ether_addr_copy() used.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Link: https://patch.msgid.link/20260629112309.154328-13-ioana.ciornei@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This patch adds support for offloading port objects, VLANs and MDBs,
added on upper bond devices.
First of all, the use of the switchdev_handle_*() replication helpers
is introduced for the SWITCHDEV_PORT_OBJ_ADD/SWITCHDEV_PORT_OBJ_DEL
events. With this change, setting up the 'port_obj_info->handled = true'
is not needed anymore since it's now handled by the new helpers.
In the DPAA2 architecture, there is no difference in adding a FDB or MDB
which points towards a LAG port. Unlike other architectures, we do not
need to populate all the possible destinations which are under the LAG,
we only have to specify a single queueing destination (QDID) which
represents the LAG. This all means that handling of MDBs in bond devices
needs to have refcount mechanism as with the FDBs.
This mechanism is triggered by calling the dpaa2_switch_lag_fdb_add() /
dpaa2_switch_lag_fdb_del() functions which were added in the previous
patch.
Also change how dpaa2_switch_port_mdb_del() behaves in case the
underlying HW operation failed. Since the delete operations cannot be
stopped from a switchdev standpoint, go ahead and ignore the return code
from the dpaa2_switch_*_fdb_del() calls.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Link: https://patch.msgid.link/20260629112309.154328-12-ioana.ciornei@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This patch adds support for offloading FDB entries added on upper bond
devices.
First of all, the call to switchdev_bridge_port_offload() is updated so
that the notifier blocks needed for FDB events replay are available to
the bridge core.
Using switchdev_handle_*() helpers is also necessary because each FDB
event needs to be fanned out to any DPAA2 switch lower device. This
triggers another change in the return type used by the
dpaa2_switch_port_fdb_event() - from notifier types to regular errno
types.
Handling of the SWITCHDEV_FDB_ADD_TO_DEVICE/SWITCHDEV_FDB_DEL_TO_DEVICE
events is updated so that the newly dpaa2_switch_lag_fdb_add() /
dpaa2_switch_lag_fdb_del() functions are called anytime a port is under
a bond device. This will allow us to manage refcounting on FDB entries
which are added on the upper bond devices.
The DPAA2 switch uses shared-VLAN learning which means that the vid
parameter is not used when adding an FDB entry to HW. The current
behavior when dealing with FDB entries with the same MAC address but
different VLANs is to add the entry to HW every time while removal will
get done on the first 'bridge fdb del' command issued by the user.
The same behavior is kept also for FDBs added on bond devices by keeping
the refcount on the {vid, addr} pair while the HW operation disregards
entirely the vid parameter.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Link: https://patch.msgid.link/20260629112309.154328-11-ioana.ciornei@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This patch adds the bulk of the changes needed in order to support
offloading of an upper bond device.
First of all, handling of the NETDEV_CHANGEUPPER and
NETDEV_PRECHANGEUPPER events is extended so that the driver is capable
to handle joining or leaving an upper bond device.
All the restrictions around the LAG offload support are added in the
newly added dpaa2_switch_pre_lag_join() function.
The same events are extended to also detect if one of our upper bond
devices changes its own upper device. In this case, on each lower device
that is DPAA2 the corresponding dpaa2_switch_port_[pre]changeupper()
function will be called. This will start the process of joining the same
FDB as the one used by the bridge device.
Setting the 'offload_fwd_mark' field on the skbs is also extended to be
setup not only when the port is under a bridge but also under a bond
device that is offloaded.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Link: https://patch.msgid.link/20260629112309.154328-10-ioana.ciornei@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Add the necessary APIs to configure and control the LAG support on the
DPAA2 switch object.
- The dpsw_lag_set() function will be used to either verify that a LAG
configuration can be support or to actually apply it in HW.
- The dpsw_if_set_lag_state() will get used in the next patches to
change the per port LAG state of a specific DPSW interface.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Link: https://patch.msgid.link/20260629112309.154328-9-ioana.ciornei@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This patch consolidates the unicast and multicast management by creating
two new functions - dpaa2_switch_port_fdb_[add|del]() - which can be
used for either uc or mc addresses. Having this common entrypoint for
both types of addresses will help us in the next patches to streamline
the same addresses but on LAG ports.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Link: https://patch.msgid.link/20260629112309.154328-8-ioana.ciornei@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
In preparation for adding offloading support for upper bond devices we
have to let the switchdev framework know if a specific bridge port is
offloaded or not, even if that brport is an upper device.
For this to happen, create the dpaa2_switch_port_to_bridge_port function
which will determine the bridge port corresponding to a particular DPAA2
switch interface and use it in the switchdev_bridge_port_offload call.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Link: https://patch.msgid.link/20260629112309.154328-7-ioana.ciornei@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The dpaa2_switch_fdb_for_join() function is responsible with determining
what FDB should be used by a port as a consequence of it joining a
bridge. The rule is that all DPAA2 switch ports under the same bridge
will use the FDB of the first port which joined that bridge. Extend the
function so that the function also covers the scenario in which there is
bridged bond device.
For this to happen, in case a bond device is encountered through the
bridge ports the function needs to descend one level through its lowers
as well.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Link: https://patch.msgid.link/20260629112309.154328-4-ioana.ciornei@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The only reason why the rtnl_lock is held in the
dpaa2_switch_event_work() is so that there is no concurency between the
changeupper notifier which manages the per port FDB assignment and the
workqueue which adds / deletes addresses into that forwarding database.
To avoid this kind of concurency without a rtnl_lock, flush the event
workqueue as the last step from the pre_bridge_leave so that any
in-flight operations targeting the current FDB are finalized before the
bridge layout (and the per port FDB assignment) changes.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Link: https://patch.msgid.link/20260629112309.154328-3-ioana.ciornei@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The DPSW object does not implement strict address filtering thus any
call to the dev_mc_add() / dev_mc_del() is pointless. Remove these calls
from the dpaa2_switch_port_mdb_add() and dpaa2_switch_port_mdb_del()
functions.
And since the multicast addresses no longer reach the netdev->mc list,
there is no point in keeping the dpaa2_switch_port_lookup_address()
function which searches through that list to verify if the same address
is added multiple times.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Link: https://patch.msgid.link/20260629112309.154328-2-ioana.ciornei@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Breno Leitao says:
====================
net: do not warn on best-effort skb allocation failures
Both netconsole and netpoll keep a small preallocated pool of skbs
(skb_pool) so they can still get a buffer under memory pressure.
On the hot path they first attempt a normal GFP_ATOMIC allocation and only
fall back to the pool when that fails, keeping the pool as a last resort.
This is where the problem happens. If alloc_skb() fails, we now have
more than 100 message coming from the page=0 failure, which consumes
the scarce pool of skb, making the real issue disappear.
So the noise (memory allocation failure) deplets the SKB buffer and
crowds out the real message we were trying to deliver.
This is happening on the Meta fleet. The stack trace looks like:
pr/netcon_ext0: page allocation failure: order:0, mode:0x40820(GFP_ATOMIC|__GFP_COMP), nodemask=(null),cpuset=/,mems_allowed=0
...
dump_stack_lvl
warn_alloc
__alloc_pages_slowpath
__alloc_frozen_pages_noprof
alloc_pages_mpol
alloc_slab_page
allocate_slab
kmem_cache_alloc_node_noprof
__alloc_skb
send_udp
netconsole_write
nbcon_emit_next_record
nbcon_emit_one
nbcon_kthread_func
kthread
Solution: Do not warn if netconsole/netpoll fails to allocate these SKBs.
Pass __GFP_NOWARN on these best-effort allocations -- both the hot-path
attempt in netconsole's find_skb() and the pool refill in netpoll's
refill_skbs() -- and let the existing fallback paths do their job
quietly. The allocation will happen on SKB refill workqueue.
Given I am touching this code, if alloc_skb() fails, reschedule the
workqueue to try later.
====================
Link: https://patch.msgid.link/20260629-netpoll_no_warn-v1-0-f380f0b2cd0c@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
refill_skbs() tops up the per-netpoll skb pool with GFP_ATOMIC and
simply stops on the first allocation failure, leaving the pool partially
filled; a later refill tops it up once memory frees up. The allocation
failing is therefore an expected and fully handled condition, but
without __GFP_NOWARN the page allocator emits a warn_alloc() splat with
a full stack trace on every miss.
Pass __GFP_NOWARN so the best-effort refill stays quiet, mirroring the
same change in netconsole's find_skb().
Signed-off-by: Breno Leitao <leitao@debian.org>
Link: https://patch.msgid.link/20260629-netpoll_no_warn-v1-2-f380f0b2cd0c@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
find_skb() allocates the skb with GFP_ATOMIC as a best-effort attempt:
on failure it falls back to the preallocated skb pool and, failing that,
polls the device and retries. The allocation failing is therefore an
expected and fully handled condition, but without __GFP_NOWARN the page
allocator still emits a warn_alloc() splat with a full stack trace on
every miss, which then consumes the whole SKB pool, that would be useful
printing the real issue rather than the memory failure.
Pass __GFP_NOWARN so the best-effort allocation stays quiet and lets the
existing fallback path do its job.
Signed-off-by: Breno Leitao <leitao@debian.org>
Link: https://patch.msgid.link/20260629-netpoll_no_warn-v1-1-f380f0b2cd0c@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When a linear skb crosses a 16 KiB boundary, ioc3_start_xmit()
splits it into two buffers of lengths s1 and s2. The descriptor
advertises those lengths through B1CNT and B2CNT.
The first buffer is mapped with s1, but the second buffer is also
mapped with s1 even though the device is told to fetch s2 bytes from
it. When the lengths differ, the DMA mapping does not cover the same
region as the second descriptor buffer, which can result in incorrect
cache maintenance or a DMA fault on implementations that enforce the
mapped range.
There is a separate mismatch in the error path. If mapping the second
buffer fails, only d1 needs to be unmapped. d1 was mapped for s1 bytes,
but the driver unmaps it using the full packet length. Streaming DMA
mappings must be unmapped with the same size used for the corresponding
map operation.
Map the second buffer with s2 and unmap the first buffer with s1 when
the second mapping fails.
Cc: <stable+noautosel@kernel.org> # untested fix for ancient HW
Signed-off-by: Xu Rao <raoxu@uniontech.com>
Reviewed-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Link: https://patch.msgid.link/4E1486BC4536407E+20260629080623.908426-1-raoxu@uniontech.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
ioc3eth_remove() frees the coherent RX and TX descriptor rings before
unregistering the netdev. If the interface is running,
unregister_netdev() invokes ioc3_close() through ndo_stop.
ioc3_close() stops the device and then calls ioc3_free_rx_bufs() and
ioc3_clean_tx_ring(). Both cleanup functions access descriptors in the
rings, so the current ordering causes CPU accesses to freed coherent
memory. Until ioc3_stop() disables RX and TX DMA, the device may also
continue using the freed ring addresses.
Unregister the netdev before releasing the rings. This lets the core
close a running interface and quiesce the device while the rings are
still valid. Keep the explicit timer deletion because ndo_stop is not
called when the interface is already down.
Cc: <stable+noautosel@kernel.org> # untested fix for ancient HW
Signed-off-by: Xu Rao <raoxu@uniontech.com>
Reviewed-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Link: https://patch.msgid.link/40CD736C4911C181+20260629085053.964383-1-raoxu@uniontech.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add READ_ONCE()/WRITE_ONCE() annotations on port->is_enabled.
While this field is written under bond->mode_lock protection,
is is read without this lock being held.
Change bond_fill_info() to acquire RCU and use READ_ONCE()
to read bond->params fields that can be updated concurrently
from sysfs/procfs/rtnetlink.
Add const qualifiers to bond_uses_primary(), __agg_active_ports(),
bond_option_active_slave_get_rcu(), bond_3ad_get_active_agg_info(),
__bond_3ad_get_active_agg_info() helpers.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Jay Vosburgh <jv@jvosburgh.net>
Cc: Andrew Lunn <andrew+netdev@lunn.ch>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Link: https://patch.msgid.link/20260629173200.469953-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The UMEM state refactor made __send_pkts() use xsk->umem for Tx
address generation. At the same time, the shared-UMEM Tx setup copies the
Rx UMEM state into a Tx-local state object and resets base_addr and
next_buffer before configuring the Tx socket.
Passing that Tx-local object to xsk_configure() makes xsk->umem point to
the zero-based Tx allocator state. This breaks the BIDIRECTIONAL test once
the roles are switched: the same socket is then used for Rx validation, but
received descriptors from the other logical UMEM half are checked against
base_addr == 0. With the new UMEM bounds check, a valid address such as
base_addr + XDP_PACKET_HEADROOM is rejected as being outside the UMEM
window.
Keep xsk->umem as the shared/Rx UMEM view used for socket configuration
and Rx validation. Use the ifobject-local UMEM copy only for Tx descriptor
address generation, preserving the BIDIRECTIONAL test's intent of using
the proper logical UMEM half after the direction switch.
Reviewed-by: Jason Xing <kerneljasonxing@gmail.com>
Reviewed-by: Tushar Vyavahare <tushar.vyavahare@intel.com>
Tested-by: Tushar Vyavahare <tushar.vyavahare@intel.com>
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Link: https://patch.msgid.link/20260629191221.2700-1-maciej.fijalkowski@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
irq_of_parse_and_map() creates irqdomain mappings that should be
balanced with irq_dispose_mapping(). The driver never called
irq_dispose_mapping(), leaking mappings on probe failure and
device removal.
Fix by adding irq_dispose_mapping() in free_gfar_dev() and
expanding its loop from priv->num_grps to MAXGROUPS so the
error path also catches partially-initialized groups. All
irqinfo pointers are pre-initialized to NULL in gfar_of_init(),
making the NULL-guarded walk in free_gfar_dev() safe for every
scenario.
gfar_parse_group() itself is left as a simple parse function
with no resource management; cleanup is centralized in the
caller's error path.
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Link: https://patch.msgid.link/20260626225228.427392-1-rosenp@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
mlxsw_sp_acl_erp_delta_clear() takes 'const char *enc_key' but modifies
the memory it points to. This is a logical error in the function
declaration.
The only caller passes a non-const buffer (aentry->ht_key.enc_key), so
the const qualifier is misleading and unnecessary.
Remove const from the enc_key parameter to match the actual usage.
Signed-off-by: Evgenii Burenchev <evg28bur@yandex.ru>
Reviewed-by: Petr Machata <petrm@nvidia.com>
Link: https://patch.msgid.link/20260625114831.17386-1-evg28bur@yandex.ru
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Pull networking fixes from Jakub Kicinski:
"Including fixes from netfilter and IPsec.
Current release - regressions:
- do not acquire dev->tx_global_lock in netdev_watchdog_up()
- ethtool: keep rtnl_lock for ops using ethtool_op_get_link()
- fix deadlock in nested UP notifier events
Current release - new code bugs:
- eth:
- cn20k: fix subbank free list indexing for search order
- airoha: fix BQL underflow in shared QDMA TX ring
Previous releases - regressions:
- netfilter:
- flowtable: fix offloaded ct timeout never being extended
- nf_conncount: prevent connlimit drops for early confirmed ct
Previous releases - always broken:
- require CAP_NET_ADMIN in the originating netns when modifying
cross-netns devices
- report NAPI thread PID in the caller's pid namespace
- mac802154: fix dirty frag in in-place crypto for IOT radios
- sctp: hold socket lock when dumping endpoints in sctp_diag, avoid
an overflow
- eth: gve: fix header buffer corruption with header-split and HW-GRO
- af_key: initialize alg_key_len for IPComp states, prevent OOB read"
* tag 'net-7.2-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (213 commits)
selftests: bonding: add a test for VLAN propagation over a bonded real device
vlan: defer real device state propagation to netdev_work
net: add the driver-facing netdev_work scheduling API
net: turn the rx_mode work into a generic netdev_work facility
net: ethtool: keep rtnl_lock for ops using ethtool_op_get_link()
rxrpc: Fix rxrpc_rotate_tx_rotate() to check there's something to rotate
rxrpc: Fix leak of released call in recvmsg(MSG_PEEK)
rxrpc: Fix socket notification race
rxrpc: Fix potential infinite loop in rxrpc_recvmsg()
rxrpc: Fix oob challenge leak in cleanup after notification failure
rxrpc: Fix the reception of a reply packet before data transmission
afs: Fix uncancelled rxrpc OOB message handler
afs: Fix further netns teardown to cancel the preallocation charger
rxrpc: Fix double unlock in rxrpc_recvmsg()
rxrpc: Fix leak of connection from OOB challenge
rxrpc: Fix ACKALL packet handling
net: hns3: differentiate autoneg default values between copper and fiber
net: hns3: fix permanent link down deadlock after reset
net: hns3: refactor MAC autoneg and speed configuration
net: hns3: unify copper port ksettings configuration path
...
Pull kvm fixes from Paolo Bonzini:
"s390:
- Fix S390_USER_OPEREXEC so it can now be enabled regardless of other
unrelated capabilities
- Fix handling of the _PAGE_UNUSED pte bit that could lead to guest
memory corruption in some scenarios
- A bunch of misc gmap fixes (locking, behaviour under memory
pressure)
- Fix CMMA dirty tracking
x86:
- Tidy up some WARN_ON() and BUG_ON(), replacing them with
WARN_ON_ONCE() or KVM_BUG_ON(). All of these have obviously never
triggered, or somebody would have been annoyed earlier, but still...
- Fix missing interrupt due to stale CR8 intercept
- Add a statistic that can come in handy to debug leaks as well as
the vulnerability to a class of recently-discovered issues
- Do not ask arch/x86/kernel to export
default_cpu_present_to_apicid() just for KVM"
* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (22 commits)
x86/apic: KVM: Use cpu_physical_id() to get APIC ID of running vCPU for AVIC
KVM: x86/mmu: Expose number of shadow MMU shadow pages as a stat
KVM: x86: Unconditionally recompute CR8 intercept on PPR update
KVM: VMX: Grab vmcs12 on CR8 interception update iff vCPU is in guest mode
KVM: x86: WARN (once) if RTC pending EOI tracking goes off the rails
KVM: x86: WARN and fail kvm_set_irq() if a PIC or I/O APIC vector is invalid
KVM: x86: Bug the VM, not the kernel, if the ISR count {under,over}flows
KVM: x86/mmu: Bug the VM, not the host kernel, if KVM write-protects upper SPTEs
KVM: x86: Replace BUG_ON() with WARN_ON_ONCE() on "bad" nested GPA translation
KVM: Replace guest-triggerable BUG_ON() in ioeventfd datamatch with get_unaligned()
KVM: s390: Return failure in case of failure in kvm_s390_set_cmma_bits()
KVM: s390: selftests: Fix cmma selftest
KVM: s390: Fix cmma dirty tracking
KVM: s390: Fix locking in kvm_s390_set_mem_control()
KVM: s390: Fix handle_{sske,pfmf} under memory pressure
KVM: s390: Fix code typo in gmap_protect_asce_top_level()
KVM: s390: Do not set special large pages dirty
KVM: s390: Fix dat_peek_cmma() overflow
s390/mm: Fix handling of _PAGE_UNUSED pte bit
KVM: s390: Fix typo in UCONTROL documentation
...
Jakub Kicinski says:
====================
net: avoid nested UP notifier events
syzbot reported that recent ethtool rework leads to deadlock
on stacked devices. VLANs create nested notifications, confusing
execution context. Bringing up dummy causes vlan to bring itself
up as well. Which in turn causes bond to ask for link state -
a call chain traveling in the opposite direction.
bond (3) bond_update_speed_duplex(vlan)
| ^ v
vlan (2) UP(vlan) (4) vlan_ethtool_get_link_ksettings()
| ^ v
dummy (1) UP(dummy) (5) __ethtool_get_link_ksettings()
We locked the instance lock of dummy at (1) and will will
try to lock it again at (5) - which of course deadlocks.
For non-nested notifications this is avoided because NETDEV_UP
is always run ops-locked (so that bond asks for link using the
netif_ API which assumes instance lock already held). The nesting,
however, makes this problematic, we cannot carry the state of
the whole chain back in the opposite direction.
AFAICT vlan is the only driver which causes such issues.
So let's try a localized fix of deferring vlan auto-open
to a workqueue.
====================
Link: https://patch.msgid.link/20260624182018.2445732-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add a regression test for the VLAN notifier handling that the netdev_work
deferral fixed.
A VLAN's real device propagates its UP/DOWN, MTU and feature changes onto
the VLANs stacked on top of it. This used to be done synchronously from the
real device's notifier and deadlocked when the real device was brought up
while enslaved to a bond (instance lock held across NETDEV_UP) and the VLAN
on top was itself a bond member: the synchronous propagation re-entered the
stack and took the same instance lock again.
The test covers both halves:
- that the deferred UP/DOWN, MTU and feature propagation actually lands on
the VLAN (link state and MTU use an ops-locked dummy, i.e. the deferral
path; features use veth, which exports vlan_features to inherit), and
- that the deadlock-prone topology - a VLAN on a dummy, with the VLAN and
the dummy each enslaved to a different bond - can be built without
hanging.
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20260624182018.2445732-5-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
vlan_device_event() generates nested UP/DOWN, MTU and feature
change events. It executes an event for the VLAN device directly
from the notifier - while the locks of the lower device are held.
This causes deadlocks, for example:
bond (3) bond_update_speed_duplex(vlan)
| ^ v
vlan (2) UP(vlan) (4) vlan_ethtool_get_link_ksettings()
| ^ v
dummy (1) UP(dummy) (5) __ethtool_get_link_ksettings()
The dummy device is ops locked, vlan creates a nested event (2),
then bond wants to ask vlan for link state (3). bond uses the
"I'm already holding the instance lock" flavor of API. But in
this case the lock held refers to vlan itself. We hit vlan's
link settings trampoline (4) and call __ethtool_get_link_ksettings()
which tries to lock dummy. Deadlock. There's no clean way for us
to tell the vlan_ethtool_get_link_ksettings() that the caller
is already in lower device's critical section.
Defer the propagation to the per-netdev work facility instead:
the notifier only schedules netdev_work_sched(vlandev, VLAN_WORK_*),
and ndo_work (vlan_dev_work) applies the change later. Hopefully
nobody expects the VLAN state changes to be instantaneous.
If someone does expect the changes to be instantaneous we will
have to do the same thing Stan did for rx_mode and "strategically"
place sync calls, to make sure such delayed works are executed
after we drop the ops lock but before we drop rtnl_lock.
Stan suggests that if we need that down the line we may
consider reshaping the mechanism into "async notifications".
AFAICT only vlan does this sort of netdev open chaining,
so as a first try I think that sticking the complexity into
the vlan code makes sense.
One corner case is that we need to cancel the event if user
explicitly changes the state before work could run. Consider
the following operations with vlan0 on top of dummy0:
ip link set dev dummy0 up # queues work to up vlan0
ip link set dev vlan0 down # user explicitly downs the vlan
ndo_work # acts on the stale event
Reported-by: syzbot+09da62a8b78959ceb8bb@syzkaller.appspotmail.com
Reported-by: syzbot+cb67c392b0b8f0fd0fc1@syzkaller.appspotmail.com
Reported-by: syzbot+9bb8bd77f3966641f298@syzkaller.appspotmail.com
Fixes: 9f275c2e90 ("net: ethtool: make sure __ethtool_get_link_ksettings() is ops-locked")
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Nicolai Buchwitz <nb@tipi-net.de>
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20260624182018.2445732-4-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
With an extra event mask we can easily extend the netdev work
to also service driver-defined events. For advanced drivers
this is probably not a perfect match, but it makes running
deferred work easier in simple cases.
Expose the netdev_work facility to drivers. Add helpers
to schedule work and a dedicated ndo to perform the driver-
-scheduled actions.
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20260624182018.2445732-3-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The rx_mode update runs from a workqueue: drivers have their
ndo_set_rx_mode_async() callback executed by a single global
work item under RTNL and ops lock. This is a useful pattern.
Support multiple "events" that need to be serviced and make RX_MODE
sync the first one. Call the events "core" because later on
we will let drivers define and schedule their own.
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20260624182018.2445732-2-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Breno reports following splats on mlx5:
RTNL: assertion failed at net/core/dev.c (2241)
WARNING: net/core/dev.c:2241 at netif_state_change+0xed/0x130, CPU#5: ethtool/1335
RIP: 0010:netif_state_change+0xf9/0x130
Call Trace:
<TASK>
__linkwatch_sync_dev+0xea/0x120
ethtool_op_get_link+0xe/0x20
__ethtool_get_link+0x26/0x40
linkstate_prepare_data+0x51/0x200
ethnl_default_doit+0x213/0x470
genl_family_rcv_msg_doit+0xdd/0x110
Looks like I missed ethtool_op_get_link() trying to sync linkwatch,
which needs rtnl_lock. Not all drivers do this - bnxt doesn't,
it just returns the link state, so add an opt-in bit.
Reported-by: Breno Leitao <leitao@debian.org>
Fixes: 45079e0013 ("net: ethtool: optionally skip rtnl_lock on Netlink path for GET ops")
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Reviewed-by: Breno Leitao <leitao@debian.org>
Acked-by: Harshitha Ramamurthy <hramamurthy@google.com>
Link: https://patch.msgid.link/20260624190439.2521219-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
David Howells says:
====================
rxrpc: Miscellaneous fixes
Here are some miscellaneous AF_RXRPC fixes for more stuff found by Sashiko[1][2]:
(1) Fix ACKALL handling by adding two more call states to simplify when
ACKs are valid.
(2) Fix connection leak from AF_RXRPC recvmsg userspace OOB handling.
(3) Fix double unlock in AF_RXRPC recvmsg userspace OOB handling.
(4) Fix AFS preallocate charge to flush the waitqueue after unlistening
the socket so that any charging thread that does manage to get started
will be waited for before socket destruction.
(5) Fix AFS OOB notify handling to cancel in-progress OOB notification
handling and then to flush the workqueue it's on.
(6) Fix handling of apparent reply reception before initial transmission
starts in client call.
(7) Fix OOB challenge leak in cleanup on notification failure.
(8) Fix infinite loop in recvmsg if OOB packet available, but no calls.
(9) Fix notify vs recvmsg race where notify thinks the call is already
queued.
(10) Fix MSG_PEEK call leak for calls with no content.
(11) Fix rxrpc_rotate_tx_window() to check that there's something in the Tx
buffer before attempting to rotate it.
====================
Link: https://patch.msgid.link/20260624163819.3017002-1-dhowells@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
