Add ndo_set_rx_mode_async callback that drivers can implement instead
of the legacy ndo_set_rx_mode. The legacy callback runs under the
netif_addr_lock spinlock with BHs disabled, preventing drivers from
sleeping. The async variant runs from a work queue with rtnl_lock and
netdev_lock_ops held, in fully sleepable context.
When __dev_set_rx_mode() sees ndo_set_rx_mode_async, it schedules
netdev_rx_mode_work instead of calling the driver inline. The work
function takes two snapshots of each address list (uc/mc) under
the addr_lock, then drops the lock and calls the driver with the
work copies. After the driver returns, it reconciles the snapshots
back to the real lists under the lock.
Add netif_rx_mode_sync() to opportunistically execute the pending
workqueue update inline, so that rx mode changes are committed
before returning to userspace:
- dev_change_flags (SIOCSIFFLAGS / RTM_NEWLINK)
- dev_set_promiscuity
- dev_set_allmulti
- dev_ifsioc SIOCADDMULTI / SIOCDELMULTI
- do_setlink (RTM_SETLINK)
Note that some deep hierarchies still do skip the lower updates via:
- dev_uc_sync
- dev_mc_sync
If we do end up hitting user-visible issues, we can add more calls to
netif_rx_mode_sync in specific places. But hopefully we should not,
the actual user-visible lists are still synced, it's that just HW state
that might be lagging.
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20260416185712.2155425-3-sdf@fomichev.me
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Introduce __hw_addr_list_snapshot() and __hw_addr_list_reconcile()
for use by the upcoming ndo_set_rx_mode_async callback.
The async rx_mode path needs to snapshot the device's unicast and
multicast address lists under the addr_lock, hand those snapshots
to the driver (which may sleep), and then propagate any sync_cnt
changes back to the real lists. Two identical snapshots are taken:
a work copy for the driver to pass to __hw_addr_sync_dev() and a
reference copy to compute deltas against.
__hw_addr_list_reconcile() walks the reference snapshot comparing
each entry against the work snapshot to determine what the driver
synced or unsynced. It then applies those deltas to the real list,
handling concurrent modifications:
- If the real entry was concurrently removed but the driver synced
it to hardware (delta > 0), re-insert a stale entry so the next
work run properly unsyncs it from hardware.
- If the entry still exists, apply the delta normally. An entry
whose refcount drops to zero is removed.
# dev_addr_test_snapshot_benchmark: 1024 addrs x 1000 snapshots: 89872802 ns total, 89872 ns/iter
# dev_addr_test_snapshot_benchmark.speed: slow
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20260416185712.2155425-2-sdf@fomichev.me
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
dev_put() releases a reference which didn't have a tracker.
References without a tracker are accounted in the tracking
code as "no_tracker". We can't free the tracker and then
call dev_put(). The references themselves will be fine
but the tracking code will think it's a double-release:
refcount_t: decrement hit 0; leaking memory.
IOW commit under fixes confused dev_put() (release never tracked
reference) with __dev_put() (just release the reference, skipping
the reference tracking infra).
Since __netdev_put_lock() uses dev_put() we can't feed a previously
tracked netdev ref into it. Let's flip things around.
netdev_put(dev, NULL) is the same as dev_put(dev) so make
netdev_put_lock() the real function and have __netdev_put_lock()
feed it a NULL tracker for all the cases that were untracked.
Fixes: d04686d9bc ("net: Implement netdev_nl_queue_create_doit")
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://patch.msgid.link/20260410153600.1984522-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The netif_get_rx_queue_lease_locked() API hides the locking
and the descend onto the leased queue. Making the code
harder to follow (at least to me). Remove the API and open
code the descend a bit. Most of the code now looks like:
if (!leased)
return __helper(x);
hw_rxq = ..
netdev_lock(hw_rxq->dev);
ret = __helper(x);
netdev_unlock(hw_rxq->dev);
return ret;
Of course if we have more code paths that need the wrapping
we may need to revisit. For now, IMHO, having to know what
netif_get_rx_queue_lease_locked() does is not worth the 20LoC
it saves.
Link: https://patch.msgid.link/20260408151251.72bd2482@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When a process in a container wants to setup a memory provider, it will
use the virtual netdev and a leased rxq, and call netif_mp_{open,close}_rxq
to try and restart the queue. At this point, proxy the queue restart on
the real rxq in the physical netdev.
For memory providers (io_uring zero-copy rx and devmem), it causes the
real rxq in the physical netdev to be filled from a memory provider that
has DMA mapped memory from a process within a container.
Signed-off-by: David Wei <dw@davidwei.uk>
Co-developed-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Link: https://patch.msgid.link/20260402231031.447597-7-daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Implement netdev_nl_queue_create_doit which creates a new rx queue in a
virtual netdev and then leases it to a rx queue in a physical netdev.
Example with ynl client:
# ynl --family netdev --output-json --do queue-create \
--json '{"ifindex": 8, "type": "rx", "lease": {"ifindex": 4, "queue": {"type": "rx", "id": 15}}}'
{'id': 1}
Note that the netdevice locking order is always from the virtual to
the physical device.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Co-developed-by: David Wei <dw@davidwei.uk>
Signed-off-by: David Wei <dw@davidwei.uk>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Link: https://patch.msgid.link/20260402231031.447597-3-daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Tunnel xmit functions (iptunnel_xmit, ip6tunnel_xmit) lack their own
recursion limit. When a bond device in broadcast mode has GRE tap
interfaces as slaves, and those GRE tunnels route back through the
bond, multicast/broadcast traffic triggers infinite recursion between
bond_xmit_broadcast() and ip_tunnel_xmit()/ip6_tnl_xmit(), causing
kernel stack overflow.
The existing XMIT_RECURSION_LIMIT (8) in the no-qdisc path is not
sufficient because tunnel recursion involves route lookups and full IP
output, consuming much more stack per level. Use a lower limit of 4
(IP_TUNNEL_RECURSION_LIMIT) to prevent overflow.
Add recursion detection using dev_xmit_recursion helpers directly in
iptunnel_xmit() and ip6tunnel_xmit() to cover all IPv4/IPv6 tunnel
paths including UDP encapsulated tunnels (VXLAN, Geneve, etc.).
Move dev_xmit_recursion helpers from net/core/dev.h to public header
include/linux/netdevice.h so they can be used by tunnel code.
BUG: KASAN: stack-out-of-bounds in blake2s.constprop.0+0xe7/0x160
Write of size 32 at addr ffff88810033fed0 by task kworker/0:1/11
Workqueue: mld mld_ifc_work
Call Trace:
<TASK>
__build_flow_key.constprop.0 (net/ipv4/route.c:515)
ip_rt_update_pmtu (net/ipv4/route.c:1073)
iptunnel_xmit (net/ipv4/ip_tunnel_core.c:84)
ip_tunnel_xmit (net/ipv4/ip_tunnel.c:847)
gre_tap_xmit (net/ipv4/ip_gre.c:779)
dev_hard_start_xmit (net/core/dev.c:3887)
sch_direct_xmit (net/sched/sch_generic.c:347)
__dev_queue_xmit (net/core/dev.c:4802)
bond_dev_queue_xmit (drivers/net/bonding/bond_main.c:312)
bond_xmit_broadcast (drivers/net/bonding/bond_main.c:5279)
bond_start_xmit (drivers/net/bonding/bond_main.c:5530)
dev_hard_start_xmit (net/core/dev.c:3887)
__dev_queue_xmit (net/core/dev.c:4841)
ip_finish_output2 (net/ipv4/ip_output.c:237)
ip_output (net/ipv4/ip_output.c:438)
iptunnel_xmit (net/ipv4/ip_tunnel_core.c:86)
gre_tap_xmit (net/ipv4/ip_gre.c:779)
dev_hard_start_xmit (net/core/dev.c:3887)
sch_direct_xmit (net/sched/sch_generic.c:347)
__dev_queue_xmit (net/core/dev.c:4802)
bond_dev_queue_xmit (drivers/net/bonding/bond_main.c:312)
bond_xmit_broadcast (drivers/net/bonding/bond_main.c:5279)
bond_start_xmit (drivers/net/bonding/bond_main.c:5530)
dev_hard_start_xmit (net/core/dev.c:3887)
__dev_queue_xmit (net/core/dev.c:4841)
ip_finish_output2 (net/ipv4/ip_output.c:237)
ip_output (net/ipv4/ip_output.c:438)
iptunnel_xmit (net/ipv4/ip_tunnel_core.c:86)
ip_tunnel_xmit (net/ipv4/ip_tunnel.c:847)
gre_tap_xmit (net/ipv4/ip_gre.c:779)
dev_hard_start_xmit (net/core/dev.c:3887)
sch_direct_xmit (net/sched/sch_generic.c:347)
__dev_queue_xmit (net/core/dev.c:4802)
bond_dev_queue_xmit (drivers/net/bonding/bond_main.c:312)
bond_xmit_broadcast (drivers/net/bonding/bond_main.c:5279)
bond_start_xmit (drivers/net/bonding/bond_main.c:5530)
dev_hard_start_xmit (net/core/dev.c:3887)
__dev_queue_xmit (net/core/dev.c:4841)
mld_sendpack
mld_ifc_work
process_one_work
worker_thread
</TASK>
Fixes: 745e20f1b6 ("net: add a recursion limit in xmit path")
Reported-by: Xiang Mei <xmei5@asu.edu>
Signed-off-by: Weiming Shi <bestswngs@gmail.com>
Link: https://patch.msgid.link/20260306160133.3852900-2-bestswngs@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
I imagine (tm) that as the number of per-queue configuration
options grows some of them may conflict for certain drivers.
While the drivers can obviously do all the validation locally
doing so is fairly inconvenient as the config is fed to drivers
piecemeal via different ops (for different params and NIC-wide
vs per-queue).
Add a centralized callback for validating the queue config
in queue ops. The callback gets invoked before memory provider
is installed, and in the future should also be called when ring
params are modified.
The validation is done after each layer of configuration.
Since we can't fail MP un-binding we must make sure that
the config is valid both before and after MP overrides are
applied. This is moot for now since the set of MP and device
configs are disjoint. It will matter significantly in the future,
so adding it now so that we don't forget..
Link: https://patch.msgid.link/20260122005113.2476634-6-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This reverts commit 77b9c4a438, reversing
changes made to 4515ec4ad5:
931420a2fc ("selftests/net: Add netkit container tests")
ab771c938d ("selftests/net: Make NetDrvContEnv support queue leasing")
6be87fbb27 ("selftests/net: Add env for container based tests")
61d99ce3df ("selftests/net: Add bpf skb forwarding program")
920da36341 ("netkit: Add xsk support for af_xdp applications")
eef51113f8 ("netkit: Add netkit notifier to check for unregistering devices")
b5ef109d22 ("netkit: Implement rtnl_link_ops->alloc and ndo_queue_create")
b5c3fa4a0b ("netkit: Add single device mode for netkit")
0073d2fd67 ("xsk: Proxy pool management for leased queues")
1ecea95dd3 ("xsk: Extend xsk_rcv_check validation")
804bf334d0 ("net: Proxy netdev_queue_get_dma_dev for leased queues")
0caa9a8dde ("net: Proxy net_mp_{open,close}_rxq for leased queues")
ff8889ff91 ("net, ethtool: Disallow leased real rxqs to be resized")
9e2103f361 ("net: Add lease info to queue-get response")
31127dedde ("net: Implement netdev_nl_queue_create_doit")
a5546e18f7 ("net: Add queue-create operation")
The series will conflict with io_uring work, and the code needs more
polish.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Implement netdev_nl_queue_create_doit which creates a new rx queue in a
virtual netdev and then leases it to a rx queue in a physical netdev.
Example with ynl client:
# ./pyynl/cli.py \
--spec ~/netlink/specs/netdev.yaml \
--do queue-create \
--json '{"ifindex": 8, "type": "rx", "lease": {"ifindex": 4, "queue": {"type": "rx", "id": 15}}}'
{'id': 1}
Note that the netdevice locking order is always from the virtual to
the physical device.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Co-developed-by: David Wei <dw@davidwei.uk>
Signed-off-by: David Wei <dw@davidwei.uk>
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Link: https://patch.msgid.link/20260115082603.219152-3-daniel@iogearbox.net
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Pull io_uring updates from Jens Axboe:
- Unify how task_work cancelations are detected, placing it in the
task_work running state rather than needing to check the task state
- Series cleaning up and moving the cancelation code to where it
belongs, in cancel.c
- Cleanup of waitid and futex argument handling
- Add support for mixed sized SQEs. 6.18 added support for mixed sized
CQEs, improving flexibility and efficiency of workloads that need big
CQEs. This adds similar support for SQEs, where the occasional need
for a 128b SQE doesn't necessitate having all SQEs be 128b in size
- Introduce zcrx and SQ/CQ layout queries. The former returns what zcrx
features are available. And both return the ring size information to
help with allocation size calculation for user provided rings like
IORING_SETUP_NO_MMAP and IORING_MEM_REGION_TYPE_USER
- Zcrx updates for 6.19. It includes a bunch of small patches,
IORING_REGISTER_ZCRX_CTRL and RQ flushing and David's work on sharing
zcrx b/w multiple io_uring instances
- Series cleaning up ring initializations, notable deduplicating ring
size and offset calculations. It also moves most of the checking
before doing any allocations, making the code simpler
- Add support for getsockname and getpeername, which is mostly a
trivial hookup after a bit of refactoring on the networking side
- Various fixes and cleanups
* tag 'for-6.19/io_uring-20251201' of git://git.kernel.org/pub/scm/linux/kernel/git/axboe/linux: (68 commits)
io_uring: Introduce getsockname io_uring cmd
socket: Split out a getsockname helper for io_uring
socket: Unify getsockname and getpeername implementation
io_uring/query: drop unused io_handle_query_entry() ctx arg
io_uring/kbuf: remove obsolete buf_nr_pages and update comments
io_uring/register: use correct location for io_rings_layout
io_uring/zcrx: share an ifq between rings
io_uring/zcrx: add io_fill_zcrx_offsets()
io_uring/zcrx: export zcrx via a file
io_uring/zcrx: move io_zcrx_scrub() and dependencies up
io_uring/zcrx: count zcrx users
io_uring/zcrx: add sync refill queue flushing
io_uring/zcrx: introduce IORING_REGISTER_ZCRX_CTRL
io_uring/zcrx: elide passing msg flags
io_uring/zcrx: use folio_nr_pages() instead of shift operation
io_uring/zcrx: convert to use netmem_desc
io_uring/query: introduce rings info query
io_uring/query: introduce zcrx query
io_uring: move cq/sq user offset init around
io_uring: pre-calculate scq layout
...
Need to call netdev_get_by_index_lock() from io_uring/zcrx.c, but it is
currently private to net. Export the function in linux/netdevice.h.
Signed-off-by: David Wei <dw@davidwei.uk>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add a new state NAPI_STATE_THREADED_BUSY_POLL to the NAPI state enum to
enable and disable threaded busy polling.
When threaded busy polling is enabled for a NAPI, enable
NAPI_STATE_THREADED also.
When the threaded NAPI is scheduled, set NAPI_STATE_IN_BUSY_POLL to
signal napi_complete_done not to rearm interrupts.
Whenever NAPI_STATE_THREADED_BUSY_POLL is unset, the
NAPI_STATE_IN_BUSY_POLL will be unset, napi_complete_done unsets the
NAPI_STATE_SCHED_THREADED bit also, which in turn will make the kthread
go to sleep.
Signed-off-by: Samiullah Khawaja <skhawaja@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Acked-by: Martin Karsten <mkarsten@uwaterloo.ca>
Tested-by: Martin Karsten <mkarsten@uwaterloo.ca>
Link: https://patch.msgid.link/20251028203007.575686-2-skhawaja@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The following order of calls currently deadlocks if:
- device has threaded=1; and
- NAPI has persistent config with threaded=0.
netif_napi_add_weight_config()
dev->threaded == 1
napi_kthread_create()
napi_enable()
napi_restore_config()
napi_set_threaded(0)
napi_stop_kthread()
while (NAPIF_STATE_SCHED)
msleep(20)
We deadlock because disabled NAPI has STATE_SCHED set.
Creating a thread in netif_napi_add() just to destroy it in
napi_disable() is fairly ugly in the first place. Let's read
both the device config and the NAPI config in netif_napi_add().
Fixes: e6d7626881 ("net: Update threaded state in napi config in netif_set_threaded")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Joe Damato <joe@dama.to>
Link: https://patch.msgid.link/20250809001205.1147153-4-kuba@kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Instead of using '0' and '1' for napi threaded state use an enum with
'disabled' and 'enabled' states.
Tested:
./tools/testing/selftests/net/nl_netdev.py
TAP version 13
1..7
ok 1 nl_netdev.empty_check
ok 2 nl_netdev.lo_check
ok 3 nl_netdev.page_pool_check
ok 4 nl_netdev.napi_list_check
ok 5 nl_netdev.dev_set_threaded
ok 6 nl_netdev.napi_set_threaded
ok 7 nl_netdev.nsim_rxq_reset_down
# Totals: pass:7 fail:0 xfail:0 xpass:0 skip:0 error:0
Signed-off-by: Samiullah Khawaja <skhawaja@google.com>
Link: https://patch.msgid.link/20250723013031.2911384-4-skhawaja@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Prepare for adding an enum type for NAPI threaded states by adding
netif_threaded_enable API. De-export the existing netif_set_threaded API
and only use it internally. Update existing drivers to use
netif_threaded_enable instead of the de-exported netif_set_threaded.
Note that dev_set_threaded used by mt76 debugfs file is unchanged.
Signed-off-by: Samiullah Khawaja <skhawaja@google.com>
Link: https://patch.msgid.link/20250723013031.2911384-3-skhawaja@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
A net device has a threaded sysctl that can be used to enable threaded
NAPI polling on all of the NAPI contexts under that device. Allow
enabling threaded NAPI polling at individual NAPI level using netlink.
Extend the netlink operation `napi-set` and allow setting the threaded
attribute of a NAPI. This will enable the threaded polling on a NAPI
context.
Add a test in `nl_netdev.py` that verifies various cases of threaded
NAPI being set at NAPI and at device level.
Tested
./tools/testing/selftests/net/nl_netdev.py
TAP version 13
1..7
ok 1 nl_netdev.empty_check
ok 2 nl_netdev.lo_check
ok 3 nl_netdev.page_pool_check
ok 4 nl_netdev.napi_list_check
ok 5 nl_netdev.dev_set_threaded
ok 6 nl_netdev.napi_set_threaded
ok 7 nl_netdev.nsim_rxq_reset_down
# Totals: pass:7 fail:0 xfail:0 xpass:0 skip:0 error:0
Signed-off-by: Samiullah Khawaja <skhawaja@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250710211203.3979655-1-skhawaja@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add helpers to "lock a netdev in a backward-compatible way",
which for ops-locked netdevs will mean take the instance lock.
For drivers which haven't opted into the ops locking we'll take
rtnl_lock.
The scoped foreach is dropping and re-taking the lock for each
device, even if prev and next are both under rtnl_lock.
I hope that's fine since we expect that netdev nl to be mostly
supported by modern drivers, and modern drivers should also
opt into the instance locking.
Note that these helpers are mostly needed for queue related state,
because drivers modify queue config in their ops in a non-atomic
way. Or differently put, queue changes don't have a clear-cut API
like NAPI configuration. Any state that can should just use the
instance lock directly, not the "compat" hacks.
Reviewed-by: Joe Damato <jdamato@fastly.com>
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20250408195956.412733-4-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
netdev_get_by_index_lock() performs following steps:
rcu_lock();
dev = lookup(netns, ifindex);
dev_get(dev);
rcu_unlock();
[... lock & validate the dev ...]
return dev
Validation right now only checks if the device is registered but since
the lookup is netns-aware we must also protect against the device
switching netns right after we dropped the RCU lock. Otherwise
the caller in netns1 may get a pointer to a device which has just
switched to netns2.
We can't hold the lock for the entire netns change process (because of
the NETDEV_UNREGISTER notifier), and there's no existing marking to
indicate that the netns is unlisted because of netns move, so add one.
AFAIU none of the existing netdev_get_by_index_lock() callers can
suffer from this problem (NAPI code double checks the netns membership
and other callers are either under rtnl_lock or not ns-sensitive),
so this patch does not have to be treated as a fix.
Reviewed-by: Joe Damato <jdamato@fastly.com>
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20250408195956.412733-2-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
As explained in commit f3483c8e1d ("net: rfs: hash function change"),
masking low order bits of skb_get_hash(skb) has low entropy.
A NIC with 32 RX queues uses the 5 low order bits of rss key
to select a queue. This means all packets landing to a given
queue share the same 5 low order bits.
Switch to hash_32() to reduce hash collisions.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250407163602.170356-2-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Move the more esoteric helpers for netdev instance lock to
a dedicated header. This avoids growing netdevice.h to infinity
and makes rebuilding the kernel much faster (after touching
the header with the helpers).
The main netdev_lock() / netdev_unlock() functions are used
in static inlines in netdevice.h and will probably be used
most commonly, so keep them in netdevice.h.
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20250307183006.2312761-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
