Commit 9e8db59132 ("net: avoid false positives in untrusted gso
validation") added a '&& skb->network_header' check before flow-dissecting
GSO packets without VIRTIO_NET_HDR_F_NEEDS_CSUM in
__virtio_net_hdr_to_skb(), because some callers (such as tun_get_user(),
tun_xdp_one(), virtnet_receive_done(), and raw_verify_header()) called
virtio_net_hdr_*_to_skb() before initializing skb->network_header and
skb->dev.
Because __alloc_skb() and __build_skb_around() zero-initialize
skb->network_header to 0 (unlike mac_header and transport_header which
are initialized to ~0U), those four callers always had
skb->network_header == 0 and bypassed flow dissection in
__virtio_net_hdr_to_skb(). More generally, skb->network_header is an
offset from skb->head (where 0 is also a valid offset whenever
skb_headroom(skb) is 0), not a boolean flag.
Whenever the 'if (gso_type && skb->network_header)' branch was skipped,
the fallback 'else if (gso_type)' only pulled nh_min_len + thlen (40 bytes
for TCPv4) without dissecting the packet, without validating ip_proto or
n_proto, and without setting skb->transport_header.
If the packet has a malformed network header, it is not rejected and a
subsequent skb_probe_transport_header() also fails, leaving
skb->transport_header at ~0U (0xffff). Similarly, if an IPv4 packet
carries IP options (ihl > 5) or an IPv6 packet carries extension headers,
pulling only nh_min_len + thlen can leave the TCP header outside
skb->head. In both cases, tcp_hdrlen(skb) in skb_gso_transport_seglen()
reads out-of-bounds:
BUG: KASAN: slab-out-of-bounds in skb_gso_transport_seglen
Read of size 2 by task poc/133
skb_gso_transport_seglen (net/core/gso.c:155)
skb_gso_validate_mac_len (net/core/gso.c:270)
tbf_enqueue (net/sched/sch_tbf.c:260)
dev_qdisc_enqueue (net/core/dev.c:4227)
__dev_queue_xmit (net/core/dev.c:4884)
In addition, checking virtio_net_hdr_match_proto() only inside
'if (!skb->protocol)' before flow dissection both skipped validation when
skb->protocol was pre-set by the caller and rejected VLAN-tagged frames
whose outer L2 protocol is ETH_P_8021Q or ETH_P_8021AD.
Fix this by:
1. Initializing skb->dev and skb->network_header (plus skb->protocol for
IFF_TUN) before virtio_net_hdr_*_to_skb() in tun_get_user(),
tun_xdp_one(), virtnet_receive_done(), and raw_verify_header(). In
tun_get_user(), drop the redundant skb_reset_mac_header(skb) in the
IFF_TUN case since __virtio_net_hdr_to_skb() unconditionally resets
mac_header.
2. Removing '&& skb->network_header' and the unvalidated
'else if (gso_type)' fallback in __virtio_net_hdr_to_skb() so all GSO
packets without VIRTIO_NET_HDR_F_NEEDS_CSUM are flow-dissected, have
their transport header pulled into linear data, and have
skb->transport_header set.
3. Moving the virtio_net_hdr_match_proto() check to after
skb_flow_dissect_flow_keys_basic(), validating keys.basic.n_proto
against hdr_gso_type.
Fixes: 9e8db59132 ("net: avoid false positives in untrusted gso validation")
Fixes: d5be7f632b ("net: validate untrusted gso packets without csum offload")
Fixes: 924a9bc362 ("net: check if protocol extracted by virtio_net_hdr_set_proto is correct")
Reported-by: Weiming Shi <bestswngs@gmail.com>
Closes: https://lore.kernel.org/netdev/20260927163117.746432-2-bestswngs@gmail.com/
Signed-off-by: Eric Dumazet <edumazet@kernel.org>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Link: https://patch.msgid.link/20261001191140.2818991-3-edumazet@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tun_get_user() uses tun->align both as skb headroom and when choosing how
much packet data to keep linear. OVS can propagate an oversized headroom
request from another port to TUN or TAP.
When align is larger than the usable space in a one-page skb head,
SKB_MAX_HEAD(align) underflows and the result becomes negative when stored
in good_linear. That value later wraps when assigned to the size_t linear
variable, and tun_alloc_skb() can place skb->data outside the allocated
head.
Bound the headroom stored by TUN to the one-page skb-head budget and the
largest non-sentinel 16-bit skb header offset. Leave one linear byte for
raw TUN and a complete Ethernet header for TAP, including NET_IP_ALIGN.
Also pull the raw-TUN protocol byte and the TAP Ethernet header before
accessing them, so these checks remain safe for nonlinear skbs supplied by
other allocation paths.
Fixes: eaea34b23c ("net/tun: implement ndo_set_rx_headroom")
Cc: stable@vger.kernel.org
Signed-off-by: Asim Viladi Oglu Manizada <manizada@pm.me>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20260812012139.2134643-1-manizada@pm.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This commit prevents tail-drop when IFF_BACKPRESSURE is set, a qdisc is
present and the ptr_ring becomes full. Once the ring reaches capacity
after a produce attempt, the netdev queue is stopped instead of dropping
subsequent packets. Without the flag, or if no qdisc is present, the
previous tail-drop behavior is preserved.
IFF_BACKPRESSURE is added to TUN_FEATURES here and not in the patch that
defines it, so that TUNSETIFF honours the flag only once the
implementation behind it is complete.
The unconditional version of this behavior was reverted because it caused
a significant throughput drop in an IPv6 multicast testcase on Brett
Sheffield's librecast testbed [1]: with 8 iperf3 TCP threads sending, the
throughput dropped from 13.5 Gbit/s to 9.13 Gbit/s. This is why the queue
stopping is now gated on IFF_BACKPRESSURE.
If producing an entry fails anyway due to a race, tun_net_xmit() drops
the packet. Such rare races are expected because LLTX is enabled and the
transmit path operates without the usual locking.
The queue state is only touched while the device is running. The stop
itself would be harmless during teardown, as tun_net_close() sets the
same bit, but the re-check below it wakes the queue again and must not
clear that stop.
A later TUNSETIFF can clear the flag again while the device has at most
one queue. Past that point tun_set_iff() returns before it writes
tun->flags, which is how it already treats every other TUN_FEATURES bit.
For the case where the flag does change, tun_set_iff() calls
tun_force_wake_queue() for the attached tfiles, so that no queue stays
stopped without a consumer that would wake it.
The __tun_wake_queue() function of the consumer races with the producer
for waking/stopping the netdev queue, which could result in a stalled
queue. Therefore, an smp_mb__after_atomic() is introduced that pairs
with the smp_mb() of the consumer. It follows the principle of store
buffering described in tools/memory-model/Documentation/recipes.txt:
- The producer in tun_net_xmit() first sets __QUEUE_STATE_DRV_XOFF,
followed by an smp_mb__after_atomic() (= smp_mb()), and then reads the
ring with __ptr_ring_check_produce().
- The consumer in __tun_wake_queue() first writes zero to the ring in
__ptr_ring_consume(), followed by an smp_mb(), and then reads the queue
status with netif_tx_queue_stopped().
=> Following the aforementioned principle, it is impossible for the
producer to see a full ring (and therefore not wake the queue on the
re-check) while the consumer simultaneously fails to see a stopped
queue (and therefore also does not wake it).
tun_net_xmit() holds only the producer_lock and can not reset cons_cnt,
which the consumer_lock protects, so the wake on the re-check leaves
stale credit behind. That is accepted as best-effort, the re-check rarely
succeeds and the next drain corrects the count.
The documentation in tuntap.rst is updated accordingly.
Benchmarks:
My own benchmarks show a slight regression in raw transmission performance
when using two sending threads. Packet loss also occurs only in the
two-thread sending case; no packet loss was observed with a single
sending thread.
Test setup:
AMD Ryzen 5 5600X at 4.3 GHz, 3200 MHz RAM, isolated QEMU threads;
Average over 50 runs @ 100,000,000 packets. SRSO and spectre v2
mitigations disabled.
Note for tap+vhost-net:
XDP drop program active in VM -> ~2.5x faster; slower for tap due to
more syscalls (high utilization of entry_SYSRETQ_unsafe_stack in perf)
+--------------------------+--------------+----------------+----------+
| 1 thread | Stock | Patched with | diff |
| sending | | fq_codel qdisc | |
+------------+-------------+--------------+----------------+----------+
| TAP | Received | 1.132 Mpps | 1.123 Mpps | -0.8% |
| +-------------+--------------+----------------+----------+
| | Lost/s | 3.765 Mpps | 0 pps | |
+------------+-------------+--------------+----------------+----------+
| TAP | Received | 3.857 Mpps | 3.901 Mpps | +1.1% |
| +-------------+--------------+----------------+----------+
| +vhost-net | Lost/s | 0.802 Mpps | 0 pps | |
+------------+-------------+--------------+----------------+----------+
+--------------------------+--------------+----------------+----------+
| 2 threads | Stock | Patched with | diff |
| sending | | fq_codel qdisc | |
+------------+-------------+--------------+----------------+----------+
| TAP | Received | 1.115 Mpps | 1.081 Mpps | -3.0% |
| +-------------+--------------+----------------+----------+
| | Lost/s | 8.490 Mpps | 391 pps | |
+------------+-------------+--------------+----------------+----------+
| TAP | Received | 3.664 Mpps | 3.555 Mpps | -3.0% |
| +-------------+--------------+----------------+----------+
| +vhost-net | Lost/s | 5.330 Mpps | 938 pps | |
+------------+-------------+--------------+----------------+----------+
[1] https://lore.kernel.org/netdev/akVnoOYQOrt8k-Gu@karahi.librecast.net/
Co-developed-by: Tim Gebauer <tim.gebauer@tu-dortmund.de>
Signed-off-by: Tim Gebauer <tim.gebauer@tu-dortmund.de>
Signed-off-by: Simon Schippers <simon.schippers@tu-dortmund.de>
Link: https://lore.kernel.org/netdev/akVnoOYQOrt8k-Gu@karahi.librecast.net/
Link: https://patch.msgid.link/20260803183641.96882-6-simon.schippers@tu-dortmund.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add tun_wake_queue() to tun.c and export it for use by vhost-net. The
function validates that the file belongs to a device implemented by
drivers/net/tun.c, in IFF_TUN as well as in IFF_TAP mode, and that the
tfile exists, dereferences the tun_struct under RCU, and delegates to
__tun_wake_queue().
vhost_net_buf_produce() now calls tun_wake_queue() after a successful
batched consume of the ring to allow the netdev subqueue to be woken up.
The point is to allow the queue to be stopped when it gets full, which is
required for traffic shaping, implemented by the following
"stop tail-drop when IFF_BACKPRESSURE is set".
As __tun_wake_queue() returns early unless IFF_BACKPRESSURE is set, a
tun/tap device that does not opt in only pays for the added check.
macvtap and ipvtap rings, which get_tap_ptr_ring() accepts too, are
unaffected: their producer is the tap_handle_frame() rx_handler and not
ndo_start_xmit, so stopping a netdev TX queue would not hold it back.
drivers/net/tap.c has no netdev_ops of its own either. No
tap_wake_queue() is needed.
cons_cnt and the wake decision are best-effort and are not reverted by
ptr_ring_unconsume(), so vhost_net_buf_unproduce() can leave the subqueue
woken over a full ring. The producer re-stops it on the next packet, and
that path only runs from vhost_net_stop_vq() and vhost_net_set_backend(),
when the consumer is going away, so a stopped queue is the correct end
state rather than a stall.
Co-developed-by: Tim Gebauer <tim.gebauer@tu-dortmund.de>
Signed-off-by: Tim Gebauer <tim.gebauer@tu-dortmund.de>
Signed-off-by: Simon Schippers <simon.schippers@tu-dortmund.de>
Link: https://patch.msgid.link/20260803183641.96882-4-simon.schippers@tu-dortmund.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Introduce tun_ring_consume() that wraps ptr_ring_consume() and calls
__tun_wake_queue(). The latter wakes the stopped netdev subqueue once
half of the ring capacity has been consumed, tracked via the new
cons_cnt field in tun_file. As a safety net, the queue is also woken on
the last consumed entry if it leaves the ring empty. The point is to
allow the queue to be stopped when it gets full, which is required for
traffic shaping, implemented by the following "stop tail-drop when
IFF_BACKPRESSURE is set".
__tun_wake_queue() returns early unless IFF_BACKPRESSURE is set, so for a
tun/tap device that does not opt in only the added check on the consume
path remains.
Every site that clears __QUEUE_STATE_DRV_XOFF now checks netif_running()
under a ring lock that tun_net_close() takes, so that none of them undoes
its stop. The core sets it before it calls ndo_open() and clears it
before it calls ndo_stop(), so it is false for exactly as long as the
device is down. IFF_UP would not do, it is only cleared after ndo_stop()
returns.
Some implementation details:
- tun_ring_recv() replaces ptr_ring_consume() with tun_ring_consume()
to properly wake the queue.
- __tun_wake_queue() returns early for a device that is not running, so a
stop from tun_net_close() is not mistaken for backpressure, and it only
wakes if the tfile still owns its slot in tun->tfiles[]. A detached
tfile keeps its queue_index, which __tun_detach() may already have
handed to the tfile that took over the slot.
- lockdep_assert_held() enforces the documented consumer_lock
precondition of __tun_wake_queue().
- __tun_detach() locks the tx_ring.consumer_lock to avoid races with
the consumer on the queue_index, and that of tfile across the hand-over
of the slot, which makes the ownership check above exact.
- The ptr_ring_consume() call in tun_queue_purge() is not replaced with
tun_ring_consume(). Instead __tun_detach() wakes the netdev queue for
the ntfile taking it over, to avoid a possible stall. The queue is only
woken if the ring of the ntfile is empty, as otherwise the consumer
wakes it after consuming the remaining entries. This does not matter
for tun_detach_all(), as it is called during device teardown and no
tfile takes over any queue.
- That wake sits after synchronize_net() and tun_queue_purge(), so it can
not be undone by a concurrent tun_net_xmit() or __tun_wake_queue().
- Ensure detached queues are woken on re-attach by calling the new
tun_force_wake_queue() helper from tun_attach(), and reuse it across
the existing wake paths. Unlike __tun_wake_queue() it ignores
IFF_BACKPRESSURE, so a queue can not stay stopped after the flag is
cleared. It does honour netif_running(), but it always clears cons_cnt,
so no old count is left over when the queue is stopped again.
- tun_net_close() takes and releases both ring locks of every tfile
before netif_tx_stop_all_queues(), so that its stop is the last write
to __QUEUE_STATE_DRV_XOFF.
- The aforementioned upcoming patch explains the pairing of the smp_mb()
of __tun_wake_queue().
Co-developed-by: Tim Gebauer <tim.gebauer@tu-dortmund.de>
Signed-off-by: Tim Gebauer <tim.gebauer@tu-dortmund.de>
Signed-off-by: Simon Schippers <simon.schippers@tu-dortmund.de>
Link: https://patch.msgid.link/20260803183641.96882-3-simon.schippers@tu-dortmund.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This reverts commit d4c22d70d7.
There is no netdev queue left to wake after reverting
commit 1d6e569b7d ("tun/tap & vhost-net: avoid ptr_ring tail-drop
when a qdisc is present").
Signed-off-by: Simon Schippers <simon.schippers@tu-dortmund.de>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Link: https://patch.msgid.link/20260728092240.250257-5-simon.schippers@tu-dortmund.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This reverts commit baf808fe4f.
There is no netdev queue left to wake after reverting
commit 1d6e569b7d ("tun/tap & vhost-net: avoid ptr_ring tail-drop
when a qdisc is present").
Signed-off-by: Simon Schippers <simon.schippers@tu-dortmund.de>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Link: https://patch.msgid.link/20260728092240.250257-4-simon.schippers@tu-dortmund.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This reverts commit 1d6e569b7d.
The commit stops the netdev queue when the ptr_ring is full instead of
dropping the packet. My own tests showed no relevant regression, but on
Brett Sheffield's librecast testbed an IPv6 multicast testcase got
slower. With 8 iperf3 TCP threads sending, the throughput dropped from
13.5 Gbit/s to 9.13 Gbit/s.
Reported-by: Brett Sheffield <brett@librecast.net>
Closes: https://lore.kernel.org/netdev/akVnoOYQOrt8k-Gu@karahi.librecast.net/
Signed-off-by: Simon Schippers <simon.schippers@tu-dortmund.de>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Link: https://patch.msgid.link/20260728092240.250257-2-simon.schippers@tu-dortmund.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Update tun_fill_info() to read device configuration fields (flags, owner,
group, numqueues, numdisabled) locklessly using READ_ONCE().
Annotate all writes to these fields in the control paths with WRITE_ONCE()
to prevent data races, as these fields can be modified concurrently via
ioctls (TUNSETPERSIST, TUNSETOWNER, TUNSETGROUP, TUNSETIFF) or queue
attaching/detaching.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20260706163517.2415530-1-edumazet@google.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Cross-merge networking fixes after downstream PR (net-7.1-rc8).
Conflicts:
drivers/net/ethernet/wangxun/txgbe/txgbe_aml.c
f67aead16e ("net: txgbe: rework service event handling")
57d39faed4 ("net: txgbe: improve functions of AML 40G devices")
net/rds/info.c
512db8267b ("rds: mark snapshot pages dirty in rds_info_getsockopt()")
6e94eeb2a2 ("rds: convert to getsockopt_iter")
Adjacent changes:
include/net/sock.h
1ee90b77b7 ("net: guard timestamp cmsgs to real error queue skbs")
f0de88303d ("net: make is_skb_wmem() available to modules")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tun_put_user() declares an on-stack struct virtio_net_hdr_v1_hash_tunnel
without zeroing it. For a non-tunnel skb, virtio_net_hdr_tnl_from_skb()
only initializes the first 10 bytes (sizeof(struct virtio_net_hdr)),
leaving bytes 10..23 (num_buffers and the hash/tunnel fields) as stack
garbage.
An unprivileged user can set the vnet header size to 24 with
TUNSETVNETHDRSZ, so __tun_vnet_hdr_put() copies all 24 bytes of the
partially-initialized struct to userspace, leaking 14 bytes of kernel
stack on every read of a non-tunnel packet.
Fix it the same way tun_get_user() already does by zeroing the whole
header right after declaration.
Fixes: 288f304351 ("tun: enable gso over UDP tunnel support.")
Reported-by: Weiming Shi <bestswngs@gmail.com>
Signed-off-by: Xiang Mei <xmei5@asu.edu>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20260607054428.3050243-1-xmei5@asu.edu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When build_skb() fails in tun_xdp_one(), the function sets ret to
-ENOMEM and jumps to the out label, which returns without freeing the
page that vhost_net_build_xdp() allocated for the frame. As with the
short-frame rejection path, tun_sendmsg() discards the per-buffer error
and still returns total_len, so vhost_tx_batch() takes the success path
and never frees the page. Each build_skb() failure in a batch leaks one
page-frag chunk.
Free the page before taking the error path, matching the put_page() the
other error exits of tun_xdp_one() already perform.
Fixes: 043d222f93 ("tuntap: accept an array of XDP buffs through sendmsg()")
Reported-by: Xiang Mei <xmei5@asu.edu>
Signed-off-by: Weiming Shi <bestswngs@gmail.com>
Reviewed-by: Dongli Zhang <dongli.zhang@oracle.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20260521163312.1479805-2-bestswngs@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tun_xdp_one() returns -EINVAL on a frame shorter than ETH_HLEN without
freeing the page that vhost_net_build_xdp() allocated for it.
tun_sendmsg() discards that -EINVAL and still returns total_len, so
vhost_tx_batch() takes the success path and never frees the page; each
short frame in a batch leaks one page-frag chunk.
A local process that can open /dev/net/tun and /dev/vhost-net can hit
this path: it attaches a tun/tap device as the vhost-net backend and
feeds TX descriptors whose length minus the virtio-net header is below
ETH_HLEN. Each kick leaks the page-frag chunks for that batch, and a
tight submission loop exhausts host memory and triggers an OOM panic.
Free the page before returning -EINVAL, matching the XDP-program error
path in the same function.
Fixes: 049584807f ("tun: add missing verification for short frame")
Reported-by: Xiang Mei <xmei5@asu.edu>
Signed-off-by: Weiming Shi <bestswngs@gmail.com>
Reviewed-by: Dongli Zhang <dongli.zhang@oracle.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20260520160020.375349-2-bestswngs@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This commit prevents tail-drop when a qdisc is present and the ptr_ring
becomes full. Once the ring reaches capacity after a produce attempt,
the netdev queue is stopped instead of dropping subsequent packets.
If no qdisc is present, the previous tail-drop behavior is preserved.
If producing an entry fails anyway due to a race, tun_net_xmit() drops
the packet. Such races are expected because LLTX is enabled and the
transmit path operates without the usual locking.
The __tun_wake_queue() function of the consumer races with the producer
for waking/stopping the netdev queue, which could result in a stalled
queue. Therefore, an smp_mb__after_atomic() is introduced that pairs
with the smp_mb() of the consumer. It follows the principle of store
buffering described in tools/memory-model/Documentation/recipes.txt:
- The producer in tun_net_xmit() first sets __QUEUE_STATE_DRV_XOFF,
followed by an smp_mb__after_atomic() (= smp_mb()), and then reads the
ring with __ptr_ring_check_produce().
- The consumer in __tun_wake_queue() first writes zero to the ring in
__ptr_ring_consume(), followed by an smp_mb(), and then reads the queue
status with netif_tx_queue_stopped().
=> Following the aforementioned principle, it is impossible for the
producer to see a full ring (and therefore not wake the queue on the
re-check) while the consumer simultaneously fails to see a stopped
queue (and therefore also does not wake it).
Benchmarks:
The benchmarks show a slight regression in raw transmission performance
when using two sending threads. Packet loss also occurs only in the
two-thread sending case; no packet loss was observed with a single
sending thread.
Test setup:
AMD Ryzen 5 5600X at 4.3 GHz, 3200 MHz RAM, isolated QEMU threads;
Average over 50 runs @ 100,000,000 packets. SRSO and spectre v2
mitigations disabled.
Note for tap+vhost-net:
XDP drop program active in VM -> ~2.5x faster; slower for tap due to
more syscalls (high utilization of entry_SYSRETQ_unsafe_stack in perf)
+--------------------------+--------------+----------------+----------+
| 1 thread | Stock | Patched with | diff |
| sending | | fq_codel qdisc | |
+------------+-------------+--------------+----------------+----------+
| TAP | Received | 1.132 Mpps | 1.123 Mpps | -0.8% |
| +-------------+--------------+----------------+----------+
| | Lost/s | 3.765 Mpps | 0 pps | |
+------------+-------------+--------------+----------------+----------+
| TAP | Received | 3.857 Mpps | 3.901 Mpps | +1.1% |
| +-------------+--------------+----------------+----------+
| +vhost-net | Lost/s | 0.802 Mpps | 0 pps | |
+------------+-------------+--------------+----------------+----------+
+--------------------------+--------------+----------------+----------+
| 2 threads | Stock | Patched with | diff |
| sending | | fq_codel qdisc | |
+------------+-------------+--------------+----------------+----------+
| TAP | Received | 1.115 Mpps | 1.081 Mpps | -3.0% |
| +-------------+--------------+----------------+----------+
| | Lost/s | 8.490 Mpps | 391 pps | |
+------------+-------------+--------------+----------------+----------+
| TAP | Received | 3.664 Mpps | 3.555 Mpps | -3.0% |
| +-------------+--------------+----------------+----------+
| +vhost-net | Lost/s | 5.330 Mpps | 938 pps | |
+------------+-------------+--------------+----------------+----------+
Co-developed-by: Tim Gebauer <tim.gebauer@tu-dortmund.de>
Signed-off-by: Tim Gebauer <tim.gebauer@tu-dortmund.de>
Signed-off-by: Simon Schippers <simon.schippers@tu-dortmund.de>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Link: https://patch.msgid.link/20260510151529.43895-5-simon.schippers@tu-dortmund.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add tun_wake_queue() to tun.c and export it for use by vhost-net. The
function validates that the file belongs to a tun/tap device and that
the tfile exists, dereferences the tun_struct under RCU, and delegates
to __tun_wake_queue().
vhost_net_buf_produce() now calls tun_wake_queue() after a successful
batched consume of the ring to allow the netdev subqueue to be woken up.
The point is to allow the queue to be stopped when it gets full, which
is required for traffic shaping - implemented by the following
"avoid ptr_ring tail-drop when a qdisc is present".
Without the corresponding queue stopping, this patch alone causes no
throughput regression for a tap+vhost-net setup sending to a qemu VM:
3.857 Mpps to 3.891 Mpps.
Details: AMD Ryzen 5 5600X at 4.3 GHz, 3200 MHz RAM, isolated QEMU
threads, XDP drop program active in VM, pktgen sender; Avg over
50 runs @ 100,000,000 packets. SRSO and spectre v2 mitigations disabled.
Co-developed-by: Tim Gebauer <tim.gebauer@tu-dortmund.de>
Signed-off-by: Tim Gebauer <tim.gebauer@tu-dortmund.de>
Signed-off-by: Simon Schippers <simon.schippers@tu-dortmund.de>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Link: https://patch.msgid.link/20260510151529.43895-3-simon.schippers@tu-dortmund.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Introduce tun_ring_consume() that wraps ptr_ring_consume() and calls
__tun_wake_queue(). The latter wakes the stopped netdev subqueue once
half of the ring capacity has been consumed, tracked via the new
cons_cnt field in tun_file. As a safety net, the queue is also woken on
the last consumed entry if it leaves the ring empty. The point is to
allow the queue to be stopped when it gets full, which is required for
traffic shaping - implemented by the following "avoid ptr_ring tail-drop
when a qdisc is present".
Some implementation details:
- tun_ring_recv() replaces ptr_ring_consume() with tun_ring_consume()
to properly wake the queue.
- __tun_detach() locks the tx_ring.consumer_lock to avoid races with
the consumer on the queue_index.
- The ptr_ring_consume() call in tun_queue_purge() is not replaced with
tun_ring_consume(). Instead, within the same tx_ring.consumer_lock
in __tun_detach(), the netdev queue is woken for the ntfile taking
it over, to avoid a possible stall. This does not matter for
tun_detach_all(), as it is called during device teardown and no tfile
takes over any queue.
- Reset cons_cnt in tun_attach() so the half-ring wake threshold is
valid for the new ring size after ptr_ring_resize().
- tun_queue_resize() wakes all queues after resizing with the proper
tx_ring.consumer_lock and resets the cons_cnt to avoid a possible
stale queue.
- The aforementioned upcoming patch explains the pairing of the smp_mb()
of __tun_wake_queue().
Without the corresponding queue stopping, this patch alone causes no
regression for a tap setup sending to a qemu VM: 1.132 Mpps
to 1.134 Mpps.
Details: AMD Ryzen 5 5600X at 4.3 GHz, 3200 MHz RAM, isolated QEMU
threads, pktgen sender; Avg over 50 runs @ 100,000,000 packets;
SRSO and spectre v2 mitigations disabled.
Co-developed-by: Tim Gebauer <tim.gebauer@tu-dortmund.de>
Signed-off-by: Tim Gebauer <tim.gebauer@tu-dortmund.de>
Signed-off-by: Simon Schippers <simon.schippers@tu-dortmund.de>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Link: https://patch.msgid.link/20260510151529.43895-2-simon.schippers@tu-dortmund.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
sk_filter_trim_cap will soon return the reason by value,
do the same for sk_filter_reason().
$ scripts/bloat-o-meter -t vmlinux.old vmlinux.new
add/remove: 0/0 grow/shrink: 0/2 up/down: 0/-21 (-21)
Function old new delta
sock_queue_rcv_skb_reason 128 126 -2
tun_net_xmit 1146 1127 -19
Total: Before=29722661, After=29722640, chg -0.00%
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20260409145625.2306224-4-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This was done entirely with mindless brute force, using
git grep -l '\<k[vmz]*alloc_objs*(.*, GFP_KERNEL)' |
xargs sed -i 's/\(alloc_objs*(.*\), GFP_KERNEL)/\1)/'
to convert the new alloc_obj() users that had a simple GFP_KERNEL
argument to just drop that argument.
Note that due to the extreme simplicity of the scripting, any slightly
more complex cases spread over multiple lines would not be triggered:
they definitely exist, but this covers the vast bulk of the cases, and
the resulting diff is also then easier to check automatically.
For the same reason the 'flex' versions will be done as a separate
conversion.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This is the result of running the Coccinelle script from
scripts/coccinelle/api/kmalloc_objs.cocci. The script is designed to
avoid scalar types (which need careful case-by-case checking), and
instead replace kmalloc-family calls that allocate struct or union
object instances:
Single allocations: kmalloc(sizeof(TYPE), ...)
are replaced with: kmalloc_obj(TYPE, ...)
Array allocations: kmalloc_array(COUNT, sizeof(TYPE), ...)
are replaced with: kmalloc_objs(TYPE, COUNT, ...)
Flex array allocations: kmalloc(struct_size(PTR, FAM, COUNT), ...)
are replaced with: kmalloc_flex(*PTR, FAM, COUNT, ...)
(where TYPE may also be *VAR)
The resulting allocations no longer return "void *", instead returning
"TYPE *".
Signed-off-by: Kees Cook <kees@kernel.org>
Replace the strcpy() calls that copy the device name into ifr->ifr_name
with strscpy() to avoid potential overflows and guarantee NULL termination.
Destination is ifr->ifr_name (size IFNAMSIZ).
Tested in QEMU (BusyBox rootfs):
- Created TUN devices via TUNSETIFF helper
- Set addresses and brought links up
- Verified long interface names are safely truncated (IFNAMSIZ-1)
Signed-off-by: Miguel García <miguelgarciaroman8@gmail.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250812082244.60240-1-miguelgarciaroman8@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit cc34acd577 ("docs: net: document new locking reality")
introduced netif_ vs dev_ function semantics: the former expects locked
netdev, the latter takes care of the locking. We don't strictly
follow this semantics on either side, but there are more dev_xxx handlers
now that don't fit. Rename them to netif_xxx where appropriate.
netif_get_mac_address is used only by tun/tap, so move it into
NETDEV_INTERNAL namespace.
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20250717172333.1288349-3-sdf@fomichev.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add a new SKB drop reason (SKB_DROP_REASON_PFMEMALLOC) to track packets
dropped due to memory pressure. In production environments, we've observed
memory exhaustion reported by memory layer stack traces, but these drops
were not properly tracked in the SKB drop reason infrastructure.
While most network code paths now properly report pfmemalloc drops, some
protocol-specific socket implementations still use sk_filter() without
drop reason tracking:
- Bluetooth L2CAP sockets
- CAIF sockets
- IUCV sockets
- Netlink sockets
- SCTP sockets
- Unix domain sockets
These remaining cases represent less common paths and could be converted
in a follow-up patch if needed. The current implementation provides
significantly improved observability into memory pressure events in the
network stack, especially for key protocols like TCP and UDP, helping to
diagnose problems in production environments.
Reported-by: Matt Fleming <mfleming@cloudflare.com>
Signed-off-by: Jesper Dangaard Brouer <hawk@kernel.org>
Link: https://patch.msgid.link/175268316579.2407873.11634752355644843509.stgit@firesoul
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Paolo Abeni says:
====================
virtio: introduce GSO over UDP tunnel
Some virtualized deployments use UDP tunnel pervasively and are impacted
negatively by the lack of GSO support for such kind of traffic in the
virtual NIC driver.
The virtio_net specification recently introduced support for GSO over
UDP tunnel, this series updates the virtio implementation to support
such a feature.
Currently the kernel virtio support limits the feature space to 64,
while the virtio specification allows for a larger number of features.
Specifically the GSO-over-UDP-tunnel-related virtio features use bits
65-69.
The first four patches in this series rework the virtio and vhost
feature support to cope with up to 128 bits. The limit is set by
a define and could be easily raised in future, as needed.
This implementation choice is aimed at keeping the code churn as
limited as possible. For the same reason, only the virtio_net driver is
reworked to leverage the extended feature space; all other
virtio/vhost drivers are unaffected, but could be upgraded to support
the extended features space in a later time.
The last four patches bring in the actual GSO over UDP tunnel support.
As per specification, some additional fields are introduced into the
virtio net header to support the new offload. The presence of such
fields depends on the negotiated features.
New helpers are introduced to convert the UDP-tunneled skb metadata to
an extended virtio net header and vice versa. Such helpers are used by
the tun and virtio_net driver to cope with the newly supported offloads.
Tested with basic stream transfer with all the possible permutations of
host kernel/qemu/guest kernel with/without GSO over UDP tunnel support.
====================
Link: https://patch.msgid.link/cover.1751874094.git.pabeni@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add new tun features to represent the newly introduced virtio
GSO over UDP tunnel offload. Allows detection and selection of
such features via the existing TUNSETOFFLOAD ioctl and compute
the expected virtio header size and tunnel header offset using
the current netdev features, so that we can plug almost seamless
the newly introduced virtio helpers to serialize the extended
virtio header.
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
---
v6 -> v7:
- rebased
v4 -> v5:
- encapsulate the guest feature guessing in a tun helper
- dropped irrelevant check on xdp buff headroom
- do not remove unrelated black line
- avoid line len > 80 char
v3 -> v4:
- virtio tnl-related fields are at fixed offset, cleanup
the code accordingly.
- use netdev features instead of flags bit to check for
the configured offload
- drop packet in case of enabled features/configured hdr
size mismatch
v2 -> v3:
- cleaned-up uAPI comments
- use explicit struct layout instead of raw buf.
With f95f0f95cfb7("net, xdp: Introduce xdp_init_buff utility routine"),
buffer length could be stored as frame size so there's no need to have
a dedicated tun_xdp_hdr structure. We can simply store virtio net
header instead.
Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Link: https://patch.msgid.link/20250701010352.74515-1-jasowang@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Move this API to the canonical timer_*() namespace.
[ tglx: Redone against pre rc1 ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/aB2X0jCKQO56WdMt@gmail.com
Convert callers of dev_set_mac_address_user() to use struct
sockaddr_storage. Add sanity checks on dev->addr_len usage.
Signed-off-by: Kees Cook <kees@kernel.org>
Acked-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Link: https://patch.msgid.link/20250521204619.2301870-8-kees@kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
timer_delete[_sync]() replaces del_timer[_sync](). Convert the whole tree
over and remove the historical wrapper inlines.
Conversion was done with coccinelle plus manual fixups where necessary.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQQ6NaUOruQGUkvPdG4raS+Z+3y5EwUCZ8qHFwAKCRAraS+Z+3y5
E6QRAQCI/irYxpT6nBgT3ejyO6CIHjbHjdNP5KRkE8JT61zmJgD8Diet4dwwOxLK
c1Pq5Jnl9KLpEPG99TcjtH3c++N3fww=
=ltml
-----END PGP SIGNATURE-----
Merge tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next
Martin KaFai Lau says:
====================
pull-request: bpf-next 2025-03-06
We've added 6 non-merge commits during the last 13 day(s) which contain
a total of 6 files changed, 230 insertions(+), 56 deletions(-).
The main changes are:
1) Add XDP metadata support for tun driver, from Marcus Wichelmann.
* tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next:
selftests/bpf: Fix file descriptor assertion in open_tuntap helper
selftests/bpf: Add test for XDP metadata support in tun driver
selftests/bpf: Refactor xdp_context_functional test and bpf program
selftests/bpf: Move open_tuntap to network helpers
net: tun: Enable transfer of XDP metadata to skb
net: tun: Enable XDP metadata support
====================
Link: https://patch.msgid.link/20250307055335.441298-1-martin.lau@linux.dev
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When the XDP metadata area was used, it is expected that the same
metadata can also be accessed from TC, as can be read in the description
of the bpf_xdp_adjust_meta helper function. In the tun driver, this was
not yet implemented.
To make this work, the skb that is being built on XDP_PASS should know
of the current size of the metadata area. This is ensured by adding
calls to skb_metadata_set. For the tun_xdp_one code path, an additional
check is necessary to handle the case where the externally initialized
xdp_buff has no metadata support (xdp->data_meta == xdp->data + 1).
More information about this feature can be found in the commit message
of commit de8f3a83b0 ("bpf: add meta pointer for direct access").
Signed-off-by: Marcus Wichelmann <marcus.wichelmann@hetzner-cloud.de>
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Link: https://patch.msgid.link/20250305213438.3863922-3-marcus.wichelmann@hetzner-cloud.de
Enable the support for the bpf_xdp_adjust_meta helper function for XDP
buffers initialized by the tun driver. This allows to reserve a metadata
area that is useful to pass any information from one XDP program to
another one, for example when using tail-calls.
Whether this helper function can be used in an XDP program depends on
how the xdp_buff was initialized. Most net drivers initialize the
xdp_buff in a way, that allows bpf_xdp_adjust_meta to be used. In case
of the tun driver, this is currently not the case.
There are two code paths in the tun driver that lead to a
bpf_prog_run_xdp and where metadata support should be enabled:
1. tun_build_skb, which is called by tun_get_user and is used when
writing packets from userspace into the device. In this case, the
xdp_buff created in tun_build_skb has no support for
bpf_xdp_adjust_meta and calls of that helper function result in
ENOTSUPP.
For this code path, it's sufficient to set the meta_valid argument of
the xdp_prepare_buff call. The reserved headroom is large enough
already.
2. tun_xdp_one, which is called by tun_sendmsg which again is called by
other drivers (e.g. vhost_net). When the TUN_MSG_PTR mode is used,
another driver may pass a batch of xdp_buffs to the tun driver. In
this case, that other driver is the one initializing the xdp_buff.
See commit 043d222f93 ("tuntap: accept an array of XDP buffs
through sendmsg()") for details.
For now, the vhost_net driver is the only one using TUN_MSG_PTR and
it already initializes the xdp_buffs with metadata support and
sufficient headroom. But the tun driver disables it again, so the
xdp_set_data_meta_invalid call has to be removed.
Signed-off-by: Marcus Wichelmann <marcus.wichelmann@hetzner-cloud.de>
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Acked-by: Jason Wang <jasowang@redhat.com>
Link: https://patch.msgid.link/20250305213438.3863922-2-marcus.wichelmann@hetzner-cloud.de
The vnet handling code will be reused by tap.
Functions are renamed to ensure that their names contain "vnet" to
clarify that they are part of the decoupled vnet handling code.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250207-tun-v6-5-fb49cf8b103e@daynix.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Decouple the vnet handling code so that we can reuse it for tap.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250207-tun-v6-4-fb49cf8b103e@daynix.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Decouple vnet-related functions from tun_struct so that we can reuse
them for tap in the future.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250207-tun-v6-3-fb49cf8b103e@daynix.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
hdr_len is repeatedly used so keep it in a local variable.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250207-tun-v6-2-fb49cf8b103e@daynix.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Check IS_ENABLED(CONFIG_TUN_VNET_CROSS_LE) to save some lines and make
future changes easier.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250207-tun-v6-1-fb49cf8b103e@daynix.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This reverts commit 3ca459eaba.
The blamed commit caused a regression when neither tun->owner nor
tun->group is set. This is intended to be allowed, but now requires
CAP_NET_ADMIN.
Discussion in the referenced thread pointed out that the original
issue that prompted this patch can be resolved in userspace.
The relaxed access control may also make a device accessible when it
previously wasn't, while existing users may depend on it to not be.
This is a clean pure git revert, except for fixing the indentation on
the gid_valid line that checkpatch correctly flagged.
Fixes: 3ca459eaba ("tun: fix group permission check")
Link: https://lore.kernel.org/netdev/CAFqZXNtkCBT4f+PwyVRmQGoT3p1eVa01fCG_aNtpt6dakXncUg@mail.gmail.com/
Signed-off-by: Willem de Bruijn <willemb@google.com>
Cc: Ondrej Mosnacek <omosnace@redhat.com>
Cc: Stas Sergeev <stsp2@yandex.ru>
Link: https://patch.msgid.link/20250204161015.739430-1-willemdebruijn.kernel@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Currently tun checks the group permission even if the user have matched.
Besides going against the usual permission semantic, this has a
very interesting implication: if the tun group is not among the
supplementary groups of the tun user, then effectively no one can
access the tun device. CAP_SYS_ADMIN still can, but its the same as
not setting the tun ownership.
This patch relaxes the group checking so that either the user match
or the group match is enough. This avoids the situation when no one
can access the device even though the ownership is properly set.
Also I simplified the logic by removing the redundant inversions:
tun_not_capable() --> !tun_capable()
Signed-off-by: Stas Sergeev <stsp2@yandex.ru>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Link: https://patch.msgid.link/20241205073614.294773-1-stsp2@yandex.ru
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
alloc_ltalkdev in net/appletalk/dev.c is dead since
commit 00f3696f75 ("net: appletalk: remove cops support")
Removing it (and it's helper) leaves dev.c and if_ltalk.h empty;
remove them and the Makefile entry.
tun.c was including that if_ltalk.h but actually wanted
the uapi version for LTALK_ALEN, fix up the path.
Signed-off-by: Dr. David Alan Gilbert <linux@treblig.org>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
no_llseek had been defined to NULL two years ago, in commit 868941b144
("fs: remove no_llseek")
To quote that commit,
At -rc1 we'll need do a mechanical removal of no_llseek -
git grep -l -w no_llseek | grep -v porting.rst | while read i; do
sed -i '/\<no_llseek\>/d' $i
done
would do it.
Unfortunately, that hadn't been done. Linus, could you do that now, so
that we could finally put that thing to rest? All instances are of the
form
.llseek = no_llseek,
so it's obviously safe.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
