[ Upstream commit 7ad18ff6449cbd6beb26b53128ddf56d2685aa93 ]
Commit 38ec4944b593 ("gro: ensure frag0 meets IP header alignment")
did the right thing, but missed the fact that napi_gro_frags() logics
calls for skb_gro_reset_offset() *before* pulling Ethernet header
to the skb linear space.
That said, the introduced check for frag0 address being aligned to 4
always fails for it as Ethernet header is obviously 14 bytes long,
and in case with NET_IP_ALIGN its start is not aligned to 4.
Fix this by adding @nhoff argument to skb_gro_reset_offset() which
tells if an IP header is placed right at the start of frag0 or not.
This restores Fast GRO for napi_gro_frags() that became very slow
after the mentioned commit, and preserves the introduced check to
avoid silent unaligned accesses.
From v1 [0]:
- inline tiny skb_gro_reset_offset() to let the code be optimized
more efficively (esp. for the !NET_IP_ALIGN case) (Eric);
- pull in Reviewed-by from Eric.
[0] https://lore.kernel.org/netdev/20210418114200.5839-1-alobakin@pm.me
Fixes: 38ec4944b593 ("gro: ensure frag0 meets IP header alignment")
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 38ec4944b593fd90c5ef42aaaa53e66ae5769d04 upstream.
After commit 0f6925b3e8da ("virtio_net: Do not pull payload in skb->head")
Guenter Roeck reported one failure in his tests using sh architecture.
After much debugging, we have been able to spot silent unaligned accesses
in inet_gro_receive()
The issue at hand is that upper networking stacks assume their header
is word-aligned. Low level drivers are supposed to reserve NET_IP_ALIGN
bytes before the Ethernet header to make that happen.
This patch hardens skb_gro_reset_offset() to not allow frag0 fast-path
if the fragment is not properly aligned.
Some arches like x86, arm64 and powerpc do not care and define NET_IP_ALIGN
as 0, this extra check will be a NOP for them.
Note that if frag0 is not used, GRO will call pskb_may_pull()
as many times as needed to pull network and transport headers.
Fixes: 0f6925b3e8da ("virtio_net: Do not pull payload in skb->head")
Fixes: 78a478d0efd9 ("gro: Inline skb_gro_header and cache frag0 virtual address")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Guenter Roeck <linux@roeck-us.net>
Cc: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Tested-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 3a5ca857079ea022e0b1b17fc154f7ad7dbc150f upstream.
When a non-initial netns is destroyed, the usual policy is to delete
all virtual network interfaces contained, but move physical interfaces
back to the initial netns. This keeps the physical interface visible
on the system.
CAN devices are somewhat special, as they define rtnl_link_ops even
if they are physical devices. If a CAN interface is moved into a
non-initial netns, destroying that netns lets the interface vanish
instead of moving it back to the initial netns. default_device_exit()
skips CAN interfaces due to having rtnl_link_ops set. Reproducer:
ip netns add foo
ip link set can0 netns foo
ip netns delete foo
WARNING: CPU: 1 PID: 84 at net/core/dev.c:11030 ops_exit_list+0x38/0x60
CPU: 1 PID: 84 Comm: kworker/u4:2 Not tainted 5.10.19 #1
Workqueue: netns cleanup_net
[<c010e700>] (unwind_backtrace) from [<c010a1d8>] (show_stack+0x10/0x14)
[<c010a1d8>] (show_stack) from [<c086dc10>] (dump_stack+0x94/0xa8)
[<c086dc10>] (dump_stack) from [<c086b938>] (__warn+0xb8/0x114)
[<c086b938>] (__warn) from [<c086ba10>] (warn_slowpath_fmt+0x7c/0xac)
[<c086ba10>] (warn_slowpath_fmt) from [<c0629f20>] (ops_exit_list+0x38/0x60)
[<c0629f20>] (ops_exit_list) from [<c062a5c4>] (cleanup_net+0x230/0x380)
[<c062a5c4>] (cleanup_net) from [<c0142c20>] (process_one_work+0x1d8/0x438)
[<c0142c20>] (process_one_work) from [<c0142ee4>] (worker_thread+0x64/0x5a8)
[<c0142ee4>] (worker_thread) from [<c0148a98>] (kthread+0x148/0x14c)
[<c0148a98>] (kthread) from [<c0100148>] (ret_from_fork+0x14/0x2c)
To properly restore physical CAN devices to the initial netns on owning
netns exit, introduce a flag on rtnl_link_ops that can be set by drivers.
For CAN devices setting this flag, default_device_exit() considers them
non-virtual, applying the usual namespace move.
The issue was introduced in the commit mentioned below, as at that time
CAN devices did not have a dellink() operation.
Fixes: e008b5fc8dc7 ("net: Simplfy default_device_exit and improve batching.")
Link: https://lore.kernel.org/r/20210302122423.872326-1-martin@strongswan.org
Signed-off-by: Martin Willi <martin@strongswan.org>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 3b23a32a63219f51a5298bc55a65ecee866e79d0 upstream.
dev_ifsioc_locked() is called with only RCU read lock, so when
there is a parallel writer changing the mac address, it could
get a partially updated mac address, as shown below:
Thread 1 Thread 2
// eth_commit_mac_addr_change()
memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
// dev_ifsioc_locked()
memcpy(ifr->ifr_hwaddr.sa_data,
dev->dev_addr,...);
Close this race condition by guarding them with a RW semaphore,
like netdev_get_name(). We can not use seqlock here as it does not
allow blocking. The writers already take RTNL anyway, so this does
not affect the slow path. To avoid bothering existing
dev_set_mac_address() callers in drivers, introduce a new wrapper
just for user-facing callers on ioctl and rtnetlink paths.
Note, bonding also changes slave mac addresses but that requires
a separate patch due to the complexity of bonding code.
Fixes: 3710becf8a58 ("net: RCU locking for simple ioctl()")
Reported-by: "Gong, Sishuai" <sishuai@purdue.edu>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Cong Wang <cong.wang@bytedance.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 8dc1c444df193701910f5e80b5d4caaf705a8fb0 upstream.
Commit c80794323e82 ("net: Fix packet reordering caused by GRO and
listified RX cooperation") had the unfortunate effect of adding
latencies in common workloads.
Before the patch, GRO packets were immediately passed to
upper stacks.
After the patch, we can accumulate quite a lot of GRO
packets (depdending on NAPI budget).
My fix is counting in napi->rx_count number of segments
instead of number of logical packets.
Fixes: c80794323e82 ("net: Fix packet reordering caused by GRO and listified RX cooperation")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Bisected-by: John Sperbeck <jsperbeck@google.com>
Tested-by: Jian Yang <jianyang@google.com>
Cc: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Saeed Mahameed <saeedm@nvidia.com>
Reviewed-by: Edward Cree <ecree.xilinx@gmail.com>
Reviewed-by: Alexander Lobakin <alobakin@pm.me>
Link: https://lore.kernel.org/r/20210204213146.4192368-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In the virtualization scenario, you can use pvipi to optimize the sending of
IPI interrupts and reduce the number of vmexit.
rps_using_pvipi test
Test guest: 32 core, 64G memory, virtio-net 8 queues, rss interrupt is bound
to 24-31 core. Observe the number of vmexit of 31 cores.
1. 1000000pps
100 iperf instances, 10000pps per instance.
disable rps_using_pvipi 68077 vmexit/s
enable rps_using_pvipi 21138 vmexit/s
2. 2000000pps
1000 iperf instances, 2000pps per instance.
disable rps_using_pvipi 40879 vmexit/s
enable rps_using_pvipi 9195 vmexit/s
3. 2000000pps
2000 iperf instances, 1000pps per instance.
disable rps_using_pvipi 24746 vmexit/s
enable rps_using_pvipi 7292 vmexit/s
Enabling rps_using_pvipi can greatly reduce the number of vmexit.
Signed-off-by: yongduan <yongduan@tencent.com>
Reviewed-by: Jiang Biao <benbjiang@tencent.com>
In the virtualization scenario, you can use pvipi to optimize the sending of
IPI interrupts and reduce the number of vmexit.
rps_using_pvipi test
Test guest: 32 core, 64G memory, virtio-net 8 queues, rss interrupt is bound
to 24-31 core. Observe the number of vmexit of 31 cores.
1. 1000000pps
100 iperf instances, 10000pps per instance.
disable rps_using_pvipi 68077 vmexit/s
enable rps_using_pvipi 21138 vmexit/s
2. 2000000pps
1000 iperf instances, 2000pps per instance.
disable rps_using_pvipi 40879 vmexit/s
enable rps_using_pvipi 9195 vmexit/s
3. 2000000pps
2000 iperf instances, 1000pps per instance.
disable rps_using_pvipi 24746 vmexit/s
enable rps_using_pvipi 7292 vmexit/s
Enabling rps_using_pvipi can greatly reduce the number of vmexit.
Signed-off-by: yongduan <yongduan@tencent.com>
Reviewed-by: Jiang Biao <benbjiang@tencent.com>
[ Upstream commit e1b9efe6baebe79019a2183176686a0e709388ae ]
When a netdev is enslaved to a bridge, its parent identifier is queried.
This is done so that packets that were already forwarded in hardware
will not be forwarded again by the bridge device between netdevs
belonging to the same hardware instance.
The operation fails when the netdev is an upper of netdevs with
different parent identifiers.
Instead of failing the enslavement, have dev_get_port_parent_id() return
'-EOPNOTSUPP' which will signal the bridge to skip the query operation.
Other callers of the function are not affected by this change.
Fixes: 7e1146e8c10c ("net: devlink: introduce devlink_compat_switch_id_get() helper")
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reported-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 96e97bc07e90f175a8980a22827faf702ca4cb30 ]
napi_disable() makes sure to set the NAPI_STATE_NPSVC bit to prevent
netpoll from accessing rings before init is complete. However, the
same is not done for fresh napi instances in netif_napi_add(),
even though we expect NAPI instances to be added as disabled.
This causes crashes during driver reconfiguration (enabling XDP,
changing the channel count) - if there is any printk() after
netif_napi_add() but before napi_enable().
To ensure memory ordering is correct we need to use RCU accessors.
Reported-by: Rob Sherwood <rsher@fb.com>
Fixes: 2d8bff12699a ("netpoll: Close race condition between poll_one_napi and napi_disable")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 6570bc79c0dfff0f228b7afd2de720fb4e84d61d upstream.
Commit 323ebb61e32b4 ("net: use listified RX for handling GRO_NORMAL
skbs") made use of listified skb processing for the users of
napi_gro_frags().
The same technique can be used in a way more common napi_gro_receive()
to speed up non-merged (GRO_NORMAL) skbs for a wide range of drivers
including gro_cells and mac80211 users.
This slightly changes the return value in cases where skb is being
dropped by the core stack, but it seems to have no impact on related
drivers' functionality.
gro_normal_batch is left untouched as it's very individual for every
single system configuration and might be tuned in manual order to
achieve an optimal performance.
Signed-off-by: Alexander Lobakin <alobakin@dlink.ru>
Acked-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Hyunsoon Kim <h10.kim@samsung.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 7df5cb75cfb8acf96c7f2342530eb41e0c11f4c3 ]
IRQs are disabled when freeing skbs in input queue.
Use the IRQ safe variant to free skbs here.
Fixes: 145dd5f9c88f ("net: flush the softnet backlog in process context")
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 11d6011c2cf29f7c8181ebde6c8bc0c4d83adcd7 ]
Sequence counters write paths are critical sections that must never be
preempted, and blocking, even for CONFIG_PREEMPTION=n, is not allowed.
Commit 5dbe7c178d3f ("net: fix kernel deadlock with interface rename and
netdev name retrieval.") handled a deadlock, observed with
CONFIG_PREEMPTION=n, where the devnet_rename seqcount read side was
infinitely spinning: it got scheduled after the seqcount write side
blocked inside its own critical section.
To fix that deadlock, among other issues, the commit added a
cond_resched() inside the read side section. While this will get the
non-preemptible kernel eventually unstuck, the seqcount reader is fully
exhausting its slice just spinning -- until TIF_NEED_RESCHED is set.
The fix is also still broken: if the seqcount reader belongs to a
real-time scheduling policy, it can spin forever and the kernel will
livelock.
Disabling preemption over the seqcount write side critical section will
not work: inside it are a number of GFP_KERNEL allocations and mutex
locking through the drivers/base/ :: device_rename() call chain.
>From all the above, replace the seqcount with a rwsem.
Fixes: 5dbe7c178d3f (net: fix kernel deadlock with interface rename and netdev name retrieval.)
Fixes: 30e6c9fa93cf (net: devnet_rename_seq should be a seqcount)
Fixes: c91f6df2db49 (sockopt: Change getsockopt() of SO_BINDTODEVICE to return an interface name)
Cc: <stable@vger.kernel.org>
Reported-by: kbuild test robot <lkp@intel.com> [ v1 missing up_read() on error exit ]
Reported-by: Dan Carpenter <dan.carpenter@oracle.com> [ v1 missing up_read() on error exit ]
Signed-off-by: Ahmed S. Darwish <a.darwish@linutronix.de>
Reviewed-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 2da2b32fd9346009e9acdb68c570ca8d3966aba7 ]
CONFIG_PREEMPTION is selected by CONFIG_PREEMPT and by CONFIG_PREEMPT_RT.
Both PREEMPT and PREEMPT_RT require the same functionality which today
depends on CONFIG_PREEMPT.
Update the comment to use CONFIG_PREEMPTION.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: David S. Miller <davem@davemloft.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: netdev@vger.kernel.org
Link: https://lore.kernel.org/r/20191015191821.11479-22-bigeasy@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit c0bbbdc32febd4f034ecbf3ea17865785b2c0652 ]
__netif_receive_skb_core may change the skb pointer passed into it (e.g.
in rx_handler). The original skb may be freed as a result of this
operation.
The callers of __netif_receive_skb_core may further process original skb
by using pt_prev pointer returned by __netif_receive_skb_core thus
leading to unpleasant effects.
The solution is to pass skb by reference into __netif_receive_skb_core.
v2: Added Fixes tag and comment regarding ppt_prev and skb invariant.
Fixes: 88eb1944e18c ("net: core: propagate SKB lists through packet_type lookup")
Signed-off-by: Boris Sukholitko <boris.sukholitko@broadcom.com>
Acked-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit dd912306ff008891c82cd9f63e8181e47a9cb2fb ]
syzbot managed to trigger a recursive NETDEV_FEAT_CHANGE event
between bonding master and slave. I managed to find a reproducer
for this:
ip li set bond0 up
ifenslave bond0 eth0
brctl addbr br0
ethtool -K eth0 lro off
brctl addif br0 bond0
ip li set br0 up
When a NETDEV_FEAT_CHANGE event is triggered on a bonding slave,
it captures this and calls bond_compute_features() to fixup its
master's and other slaves' features. However, when syncing with
its lower devices by netdev_sync_lower_features() this event is
triggered again on slaves when the LRO feature fails to change,
so it goes back and forth recursively until the kernel stack is
exhausted.
Commit 17b85d29e82c intentionally lets __netdev_update_features()
return -1 for such a failure case, so we have to just rely on
the existing check inside netdev_sync_lower_features() and skip
NETDEV_FEAT_CHANGE event only for this specific failure case.
Fixes: fd867d51f889 ("net/core: generic support for disabling netdev features down stack")
Reported-by: syzbot+e73ceacfd8560cc8a3ca@syzkaller.appspotmail.com
Reported-by: syzbot+c2fb6f9ddcea95ba49b5@syzkaller.appspotmail.com
Cc: Jarod Wilson <jarod@redhat.com>
Cc: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Jann Horn <jannh@google.com>
Reviewed-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit a4837980fd9fa4c70a821d11831698901baef56b ]
For HZ < 1000 timeout 2000us rounds up to 1 jiffy but expires randomly
because next timer interrupt could come shortly after starting softirq.
For commonly used CONFIG_HZ=1000 nothing changes.
Fixes: 7acf8a1e8a28 ("Replace 2 jiffies with sysctl netdev_budget_usecs to enable softirq tuning")
Reported-by: Dmitry Yakunin <zeil@yandex-team.ru>
Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Alexander Lobakin