Gerd Bayer says:
====================
net/smc: Fix effective buffer size
commit 0227f058aa ("net/smc: Unbind r/w buffer size from clcsock
and make them tunable") started to derive the effective buffer size for
SMC connections inconsistently in case a TCP fallback was used and
memory consumption of SMC with the default settings was doubled when
a connection negotiated SMC. That was not what we want.
This series consolidates the resulting effective buffer size that is
used with SMC sockets, which is based on Jan Karcher's effort (see
[1]). For all TCP exchanges (in particular in case of a fall back when
no SMC connection was possible) the values from net.ipv4.tcp_[rw]mem
are used. If SMC succeeds in establishing a SMC connection, the newly
introduced values from net.smc.[rw]mem are used.
net.smc.[rw]mem is initialized to 64kB, respectively. Internal test
have show this to be a good compromise between throughput/latency
and memory consumption. Also net.smc.[rw]mem is now decoupled completely
from any tuning through net.ipv4.tcp_[rw]mem.
If a user chose to tune a socket's receive or send buffer size with
setsockopt, this tuning is now consistently applied to either fall-back
TCP or proper SMC connections over the socket.
Thanks,
Gerd
v2 - v3:
- Rebase to and resolve conflict of second patch with latest net/master.
v1 - v2:
- In second patch, use sock_net() helper as suggested by Tony and demanded
by kernel test robot.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Tuning of the effective buffer size through setsockopts was working for
SMC traffic only but not for TCP fall-back connections even before
commit 0227f058aa ("net/smc: Unbind r/w buffer size from clcsock and
make them tunable"). That change made it apparent that TCP fall-back
connections would use net.smc.[rw]mem as buffer size instead of
net.ipv4_tcp_[rw]mem.
Amend the code that copies attributes between the (TCP) clcsock and the
SMC socket and adjust buffer sizes appropriately:
- Copy over sk_userlocks so that both sockets agree on whether tuning
via setsockopt is active.
- When falling back to TCP use sk_sndbuf or sk_rcvbuf as specified with
setsockopt. Otherwise, use the sysctl value for TCP/IPv4.
- Likewise, use either values from setsockopt or from sysctl for SMC
(duplicated) on successful SMC connect.
In smc_tcp_listen_work() drop the explicit copy of buffer sizes as that
is taken care of by the attribute copy.
Fixes: 0227f058aa ("net/smc: Unbind r/w buffer size from clcsock and make them tunable")
Reviewed-by: Wenjia Zhang <wenjia@linux.ibm.com>
Reviewed-by: Tony Lu <tonylu@linux.alibaba.com>
Signed-off-by: Gerd Bayer <gbayer@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 0227f058aa ("net/smc: Unbind r/w buffer size from clcsock
and make them tunable") introduced the net.smc.rmem and net.smc.wmem
sysctls to specify the size of buffers to be used for SMC type
connections. This created a regression for users that specified the
buffer size via setsockopt() as the effective buffer size was now
doubled.
Re-introduce the division by 2 in the SMC buffer create code and level
this out by duplicating the net.smc.[rw]mem values used for initializing
sk_rcvbuf/sk_sndbuf at socket creation time. This gives users of both
methods (setsockopt or sysctl) the effective buffer size that they
expect.
Initialize net.smc.[rw]mem from its own constant of 64kB, respectively.
Internal performance tests show that this value is a good compromise
between throughput/latency and memory consumption. Also, this decouples
it from any tuning that was done to net.ipv4.tcp_[rw]mem[1] before the
module for SMC protocol was loaded. Check that no more than INT_MAX / 2
is assigned to net.smc.[rw]mem, in order to avoid any overflow condition
when that is doubled for use in sk_sndbuf or sk_rcvbuf.
While at it, drop the confusing sk_buf_size variable from
__smc_buf_create and name "compressed" buffer size variables more
consistently.
Background:
Before the commit mentioned above, SMC's buffer allocator in
__smc_buf_create() always used half of the sockets' sk_rcvbuf/sk_sndbuf
value as initial value to search for appropriate buffers. If the search
resorted to using a bigger buffer when all buffers of the specified
size were busy, the duplicate of the used effective buffer size is
stored back to sk_rcvbuf/sk_sndbuf.
When available, buffers of exactly the size that a user had specified as
input to setsockopt() were used, despite setsockopt()'s documentation in
"man 7 socket" talking of a mandatory duplication:
[...]
SO_SNDBUF
Sets or gets the maximum socket send buffer in bytes.
The kernel doubles this value (to allow space for book‐
keeping overhead) when it is set using setsockopt(2),
and this doubled value is returned by getsockopt(2).
The default value is set by the
/proc/sys/net/core/wmem_default file and the maximum
allowed value is set by the /proc/sys/net/core/wmem_max
file. The minimum (doubled) value for this option is
2048.
[...]
Fixes: 0227f058aa ("net/smc: Unbind r/w buffer size from clcsock and make them tunable")
Co-developed-by: Jan Karcher <jaka@linux.ibm.com>
Signed-off-by: Jan Karcher <jaka@linux.ibm.com>
Reviewed-by: Wenjia Zhang <wenjia@linux.ibm.com>
Reviewed-by: Tony Lu <tonylu@linux.alibaba.com>
Signed-off-by: Gerd Bayer <gbayer@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Vladimir Oltean says:
====================
Fix ENETC probing after 6fffbc7ae1 ("PCI: Honor firmware's device disabled status")
I'm not sure who should take this patch set (net maintainers or PCI
maintainers). Everyone could pick up just their part, and that would
work (no compile time dependencies). However, the entire series needs
ACK from both sides and Rob for sure.
v1 at:
https://lore.kernel.org/netdev/20230521115141.2384444-1-vladimir.oltean@nxp.com/
====================
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since commit 6fffbc7ae1 ("PCI: Honor firmware's device disabled
status"), this is redundant and does nothing, because enetc_pf_probe()
no longer even gets called.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The workaround implemented in commit 3222b5b613 ("net: enetc:
initialize RFS/RSS memories for unused ports too") is no longer
effective after commit 6fffbc7ae1 ("PCI: Honor firmware's device
disabled status"). Thus, it has introduced a regression and we see AER
errors being reported again:
$ ip link set sw2p0 up && dhclient -i sw2p0 && ip addr show sw2p0
fsl_enetc 0000:00:00.2 eno2: configuring for fixed/internal link mode
fsl_enetc 0000:00:00.2 eno2: Link is Up - 2.5Gbps/Full - flow control rx/tx
mscc_felix 0000:00:00.5 swp2: configuring for fixed/sgmii link mode
mscc_felix 0000:00:00.5 swp2: Link is Up - 1Gbps/Full - flow control off
sja1105 spi2.2 sw2p0: configuring for phy/rgmii-id link mode
sja1105 spi2.2 sw2p0: Link is Up - 1Gbps/Full - flow control off
pcieport 0000:00:1f.0: AER: Multiple Corrected error received: 0000:00:00.0
pcieport 0000:00:1f.0: AER: can't find device of ID0000
Rob's suggestion is to reimplement the enetc driver workaround as a
PCI fixup, and to modify the PCI core to run the fixups for all PCI
functions. This change handles the first part.
We refactor the common code in enetc_psi_create() and enetc_psi_destroy(),
and use the PCI fixup only for those functions for which enetc_pf_probe()
won't get called. This avoids some work being done twice for the PFs
which are enabled.
Fixes: 6fffbc7ae1 ("PCI: Honor firmware's device disabled status")
Link: https://lore.kernel.org/netdev/CAL_JsqLsVYiPLx2kcHkDQ4t=hQVCR7NHziDwi9cCFUFhx48Qow@mail.gmail.com/
Suggested-by: Rob Herring <robh@kernel.org>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The blamed commit has broken probing on
arch/arm64/boot/dts/freescale/fsl-ls1028a.dtsi when &enetc_port0
(PCI function 0) has status = "disabled".
Background: pci_scan_slot() has logic to say that if the function 0 of a
device is absent, the entire device is absent and we can skip the other
functions entirely. Traditionally, this has meant that
pci_bus_read_dev_vendor_id() returns an error code for that function.
However, since the blamed commit, there is an extra confounding
condition: function 0 of the device exists and has a valid vendor id,
but it is disabled in the device tree. In that case, pci_scan_slot()
would incorrectly skip the entire device instead of just that function.
In the case of NXP LS1028A, status = "disabled" does not mean that the
PCI function's config space is not available for reading. It is, but the
Ethernet port is just not functionally useful with a particular SerDes
protocol configuration (0x9999) due to pinmuxing constraints of the Soc.
So, pci_scan_slot() skips all other functions on the ENETC ECAM
(enetc_port1, enetc_port2, enetc_mdio_pf3 etc) when just enetc_port0 had
to not be probed.
There is an additional regression introduced by the change, caused by
its fundamental premise. The enetc driver needs to run code for all PCI
functions, regardless of whether they're enabled or not in the device
tree. That is no longer possible if the driver's probe function is no
longer called. But Rob recommends that we move the of_device_is_available()
detection to dev->match_driver, and this makes the PCI fixups still run
on all functions, while just probing drivers for those functions that
are enabled. So, a separate change in the enetc driver will have to move
the workarounds to a PCI fixup.
Fixes: 6fffbc7ae1 ("PCI: Honor firmware's device disabled status")
Link: https://lore.kernel.org/netdev/CAL_JsqLsVYiPLx2kcHkDQ4t=hQVCR7NHziDwi9cCFUFhx48Qow@mail.gmail.com/
Suggested-by: Rob Herring <robh@kernel.org>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add fdir_fltr_lock locking in unprotected places.
The change in iavf_fdir_is_dup_fltr adds a spinlock around a loop which
iterates over all filters and looks for a duplicate. The filter can be
removed from list and freed from memory at the same time it's being
compared. All other places where filters are deleted are already
protected with spinlock.
The remaining changes protect adapter->fdir_active_fltr variable so now
all its uses are under a spinlock.
Fixes: 527691bf06 ("iavf: Support IPv4 Flow Director filters")
Signed-off-by: Piotr Gardocki <piotrx.gardocki@intel.com>
Tested-by: Rafal Romanowski <rafal.romanowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230807205011.3129224-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Access to shared variables through hrtimer requires locking in order
to protect the variables because actions to write into these variables
(oper_gate_closed, admin_gate_closed, and qbv_transition) might potentially
occur simultaneously. This patch provides a locking mechanisms to avoid
such scenarios.
Fixes: 175c241288 ("igc: Fix TX Hang issue when QBV Gate is closed")
Suggested-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Muhammad Husaini Zulkifli <muhammad.husaini.zulkifli@intel.com>
Tested-by: Naama Meir <naamax.meir@linux.intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Link: https://lore.kernel.org/r/20230807205129.3129346-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
-----BEGIN PGP SIGNATURE-----
iQEzBAABCAAdFiEEGhZs6bAKwk/OTgTpSD+KveBX+j4FAmTRPIkACgkQSD+KveBX
+j4ozgf/aS4GktVhN0DkTksN3K8D4QSrYYR+hiW4e7o7xz1K32qw+GRiC2r/FSGh
XHzOGybuj7+V9TxcOb8NRSqKVtpci4MChQTrWzGutqwtcxU18SyDSo/kEmHfqkT+
kuH+PDpDpNtcCwr+z3Cb+M22ZjpqwZnWdxfKa9rG+ur9QPTUnBY1+MGGYn6eeMnC
DD7HiB+q7YnCNbsFHJNp4ZeUVsTWO4gD6aOUkUhXDaBkBTpKwrCUQ+dNMm2AG1/z
3v4VdtB28BHV6o4QSop7HYk1DEnpTjZt/R+FnC6bZpIu7bDYVqw9p97EoPMvSlMv
RgorORXPr30WpvWcH6ff2y/A0MmhIw==
=nDa5
-----END PGP SIGNATURE-----
Merge tag 'mlx5-fixes-2023-08-07' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux
Saeed Mahameed says:
====================
mlx5 fixes 2023-08-07
This series provides bug fixes to mlx5 driver.
* tag 'mlx5-fixes-2023-08-07' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux:
net/mlx5e: Add capability check for vnic counters
net/mlx5: Reload auxiliary devices in pci error handlers
net/mlx5: Skip clock update work when device is in error state
net/mlx5: LAG, Check correct bucket when modifying LAG
net/mlx5e: Unoffload post act rule when handling FIB events
net/mlx5: Fix devlink controller number for ECVF
net/mlx5: Allow 0 for total host VFs
net/mlx5: Return correct EC_VF function ID
net/mlx5: DR, Fix wrong allocation of modify hdr pattern
net/mlx5e: TC, Fix internal port memory leak
net/mlx5e: Take RTNL lock when needed before calling xdp_set_features()
====================
Link: https://lore.kernel.org/r/20230807212607.50883-1-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When externel_lb and reset are executed together, a deadlock may
occur:
[ 3147.217009] INFO: task kworker/u321:0:7 blocked for more than 120 seconds.
[ 3147.230483] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[ 3147.238999] task:kworker/u321:0 state:D stack: 0 pid: 7 ppid: 2 flags:0x00000008
[ 3147.248045] Workqueue: hclge hclge_service_task [hclge]
[ 3147.253957] Call trace:
[ 3147.257093] __switch_to+0x7c/0xbc
[ 3147.261183] __schedule+0x338/0x6f0
[ 3147.265357] schedule+0x50/0xe0
[ 3147.269185] schedule_preempt_disabled+0x18/0x24
[ 3147.274488] __mutex_lock.constprop.0+0x1d4/0x5dc
[ 3147.279880] __mutex_lock_slowpath+0x1c/0x30
[ 3147.284839] mutex_lock+0x50/0x60
[ 3147.288841] rtnl_lock+0x20/0x2c
[ 3147.292759] hclge_reset_prepare+0x68/0x90 [hclge]
[ 3147.298239] hclge_reset_subtask+0x88/0xe0 [hclge]
[ 3147.303718] hclge_reset_service_task+0x84/0x120 [hclge]
[ 3147.309718] hclge_service_task+0x2c/0x70 [hclge]
[ 3147.315109] process_one_work+0x1d0/0x490
[ 3147.319805] worker_thread+0x158/0x3d0
[ 3147.324240] kthread+0x108/0x13c
[ 3147.328154] ret_from_fork+0x10/0x18
In externel_lb process, the hns3 driver call napi_disable()
first, then the reset happen, then the restore process of the
externel_lb will fail, and will not call napi_enable(). When
doing externel_lb again, napi_disable() will be double call,
cause a deadlock of rtnl_lock().
This patch use the HNS3_NIC_STATE_DOWN state to protect the
calling of napi_disable() and napi_enable() in externel_lb
process, just as the usage in ndo_stop() and ndo_start().
Fixes: 04b6ba1435 ("net: hns3: add support for external loopback test")
Signed-off-by: Yonglong Liu <liuyonglong@huawei.com>
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Link: https://lore.kernel.org/r/20230807113452.474224-5-shaojijie@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In some configure flow of hns3 driver, for example, change mtu, it will
disable MAC through firmware before configuration. But firmware disables
MAC asynchronously. The rx traffic may be not stopped in this case.
So fixes it by waiting until mac link is down.
Fixes: a9775bb64a ("net: hns3: fix set and get link ksettings issue")
Signed-off-by: Jie Wang <wangjie125@huawei.com>
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Link: https://lore.kernel.org/r/20230807113452.474224-4-shaojijie@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Some nic configurations could only be performed after link is down. So this
patch refactor this API for reuse.
Signed-off-by: Jie Wang <wangjie125@huawei.com>
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Link: https://lore.kernel.org/r/20230807113452.474224-3-shaojijie@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Restore the mac pause state to user configuration when autoneg is disabled
Signed-off-by: Jian Shen <shenjian15@huawei.com>
Signed-off-by: Peiyang Wang <wangpeiyang1@huawei.com>
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Link: https://lore.kernel.org/r/20230807113452.474224-2-shaojijie@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Change the new (unreleased) SO_PEERPIDFD sockopt to return ENODATA
rather than ESRCH if a socket type does not support remote peer-PID
queries.
Currently, SO_PEERPIDFD returns ESRCH when the socket in question is
not an AF_UNIX socket. This is quite unexpected, given that one would
assume ESRCH means the peer process already exited and thus cannot be
found. However, in that case the sockopt actually returns EINVAL (via
pidfd_prepare()). This is rather inconsistent with other syscalls, which
usually return ESRCH if a given PID refers to a non-existant process.
This changes SO_PEERPIDFD to return ENODATA instead. This is also what
SO_PEERGROUPS returns, and thus keeps a consistent behavior across
sockopts.
Note that this code is returned in 2 cases: First, if the socket type is
not AF_UNIX, and secondly if the socket was not yet connected. In both
cases ENODATA seems suitable.
Signed-off-by: David Rheinsberg <david@readahead.eu>
Reviewed-by: Christian Brauner <brauner@kernel.org>
Acked-by: Luca Boccassi <bluca@debian.org>
Fixes: 7b26952a91 ("net: core: add getsockopt SO_PEERPIDFD")
Link: https://lore.kernel.org/r/20230807081225.816199-1-david@readahead.eu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Update MAINTAINERS entries with a valid email address as the Microchip
one is no longer valid.
Acked-by: Conor Dooley <conor.dooley@microchip.com>
Acked-by: Nicolas Ferre <nicolas.ferre@microchip.com>
Signed-off-by: Claudiu Beznea <claudiu.beznea@tuxon.dev>
Acked-by: Sebastian Reichel <sre@kernel.org>
Link: https://lore.kernel.org/r/20230804050007.235799-1-claudiu.beznea@tuxon.dev
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Syzkaller reported the following issue:
=======================================
Too BIG xdp->frame_sz = 131072
WARNING: CPU: 0 PID: 5020 at net/core/filter.c:4121
____bpf_xdp_adjust_tail net/core/filter.c:4121 [inline]
WARNING: CPU: 0 PID: 5020 at net/core/filter.c:4121
bpf_xdp_adjust_tail+0x466/0xa10 net/core/filter.c:4103
...
Call Trace:
<TASK>
bpf_prog_4add87e5301a4105+0x1a/0x1c
__bpf_prog_run include/linux/filter.h:600 [inline]
bpf_prog_run_xdp include/linux/filter.h:775 [inline]
bpf_prog_run_generic_xdp+0x57e/0x11e0 net/core/dev.c:4721
netif_receive_generic_xdp net/core/dev.c:4807 [inline]
do_xdp_generic+0x35c/0x770 net/core/dev.c:4866
tun_get_user+0x2340/0x3ca0 drivers/net/tun.c:1919
tun_chr_write_iter+0xe8/0x210 drivers/net/tun.c:2043
call_write_iter include/linux/fs.h:1871 [inline]
new_sync_write fs/read_write.c:491 [inline]
vfs_write+0x650/0xe40 fs/read_write.c:584
ksys_write+0x12f/0x250 fs/read_write.c:637
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x38/0xb0 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x63/0xcd
xdp->frame_sz > PAGE_SIZE check was introduced in commit c8741e2bfe
("xdp: Allow bpf_xdp_adjust_tail() to grow packet size"). But Jesper
Dangaard Brouer <jbrouer@redhat.com> noted that after introducing the
xdp_init_buff() which all XDP driver use - it's safe to remove this
check. The original intend was to catch cases where XDP drivers have
not been updated to use xdp.frame_sz, but that is not longer a concern
(since xdp_init_buff).
Running the initial syzkaller repro it was discovered that the
contiguous physical memory allocation is used for both xdp paths in
tun_get_user(), e.g. tun_build_skb() and tun_alloc_skb(). It was also
stated by Jesper Dangaard Brouer <jbrouer@redhat.com> that XDP can
work on higher order pages, as long as this is contiguous physical
memory (e.g. a page).
Reported-and-tested-by: syzbot+f817490f5bd20541b90a@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/000000000000774b9205f1d8a80d@google.com/T/
Link: https://syzkaller.appspot.com/bug?extid=f817490f5bd20541b90a
Link: https://lore.kernel.org/all/20230725155403.796-1-andrew.kanner@gmail.com/T/
Fixes: 43b5169d83 ("net, xdp: Introduce xdp_init_buff utility routine")
Signed-off-by: Andrew Kanner <andrew.kanner@gmail.com>
Acked-by: Jesper Dangaard Brouer <hawk@kernel.org>
Acked-by: Jason Wang <jasowang@redhat.com>
Link: https://lore.kernel.org/r/20230803190316.2380231-1-andrew.kanner@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Using the syzkaller repro with reduced packet size it was discovered
that XDP_PACKET_HEADROOM is not checked in tun_can_build_skb(),
although pad may be incremented in tun_build_skb(). This may end up
with exceeding the PAGE_SIZE limit in tun_build_skb().
Jason Wang <jasowang@redhat.com> proposed to count XDP_PACKET_HEADROOM
always (e.g. without rcu_access_pointer(tun->xdp_prog)) in
tun_can_build_skb() since there's a window during which XDP program
might be attached between tun_can_build_skb() and tun_build_skb().
Fixes: 7df13219d7 ("tun: reserve extra headroom only when XDP is set")
Link: https://syzkaller.appspot.com/bug?extid=f817490f5bd20541b90a
Signed-off-by: Andrew Kanner <andrew.kanner@gmail.com>
Link: https://lore.kernel.org/r/20230803185947.2379988-1-andrew.kanner@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Jason A. Donenfeld says:
====================
wireguard fixes for 6.5-rc6
Just one patch this time, somewhat late in the cycle:
1) Fix an off-by-one calculation for the maximum node depth size in the
allowedips trie data structure, and also adjust the self-tests to hit
this case so it doesn't regress again in the future.
====================
Link: https://lore.kernel.org/r/20230807132146.2191597-1-Jason@zx2c4.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In the allowedips self-test, nodes are inserted into the tree, but it
generated an even amount of nodes, but for checking maximum node depth,
there is of course the root node, which makes the total number
necessarily odd. With two few nodes added, it never triggered the
maximum depth check like it should have. So, add 129 nodes instead of
128 nodes, and do so with a more straightforward scheme, starting with
all the bits set, and shifting over one each time. Then increase the
maximum depth to 129, and choose a better name for that variable to
make it clear that it represents depth as opposed to bits.
Cc: stable@vger.kernel.org
Fixes: e7096c131e ("net: WireGuard secure network tunnel")
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Link: https://lore.kernel.org/r/20230807132146.2191597-2-Jason@zx2c4.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
BUG_ON(!vlan_info) is triggered in unregister_vlan_dev() with
following testcase:
# ip netns add ns1
# ip netns exec ns1 ip link add bond0 type bond mode 0
# ip netns exec ns1 ip link add bond_slave_1 type veth peer veth2
# ip netns exec ns1 ip link set bond_slave_1 master bond0
# ip netns exec ns1 ip link add link bond_slave_1 name vlan10 type vlan id 10 protocol 802.1ad
# ip netns exec ns1 ip link add link bond0 name bond0_vlan10 type vlan id 10 protocol 802.1ad
# ip netns exec ns1 ip link set bond_slave_1 nomaster
# ip netns del ns1
The logical analysis of the problem is as follows:
1. create ETH_P_8021AD protocol vlan10 for bond_slave_1:
register_vlan_dev()
vlan_vid_add()
vlan_info_alloc()
__vlan_vid_add() // add [ETH_P_8021AD, 10] vid to bond_slave_1
2. create ETH_P_8021AD protocol bond0_vlan10 for bond0:
register_vlan_dev()
vlan_vid_add()
__vlan_vid_add()
vlan_add_rx_filter_info()
if (!vlan_hw_filter_capable(dev, proto)) // condition established because bond0 without NETIF_F_HW_VLAN_STAG_FILTER
return 0;
if (netif_device_present(dev))
return dev->netdev_ops->ndo_vlan_rx_add_vid(dev, proto, vid); // will be never called
// The slaves of bond0 will not refer to the [ETH_P_8021AD, 10] vid.
3. detach bond_slave_1 from bond0:
__bond_release_one()
vlan_vids_del_by_dev()
list_for_each_entry(vid_info, &vlan_info->vid_list, list)
vlan_vid_del(dev, vid_info->proto, vid_info->vid);
// bond_slave_1 [ETH_P_8021AD, 10] vid will be deleted.
// bond_slave_1->vlan_info will be assigned NULL.
4. delete vlan10 during delete ns1:
default_device_exit_batch()
dev->rtnl_link_ops->dellink() // unregister_vlan_dev() for vlan10
vlan_info = rtnl_dereference(real_dev->vlan_info); // real_dev of vlan10 is bond_slave_1
BUG_ON(!vlan_info); // bond_slave_1->vlan_info is NULL now, bug is triggered!!!
Add S-VLAN tag related features support to bond driver. So the bond driver
will always propagate the VLAN info to its slaves.
Fixes: 8ad227ff89 ("net: vlan: add 802.1ad support")
Suggested-by: Ido Schimmel <idosch@idosch.org>
Signed-off-by: Ziyang Xuan <william.xuanziyang@huawei.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Link: https://lore.kernel.org/r/20230802114320.4156068-1-william.xuanziyang@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add missing capability check for each of the vnic counters exposed by
devlink health reporter, and thus avoid unexpected behavior due to
invalid access to registers.
While at it, read only the exact number of bits for each counter whether
it was 32 bits or 64 bits.
Fixes: b0bc615df4 ("net/mlx5: Add vnic devlink health reporter to PFs/VFs")
Fixes: a33682e4e7 ("net/mlx5e: Expose catastrophic steering error counters")
Signed-off-by: Lama Kayal <lkayal@nvidia.com>
Reviewed-by: Gal Pressman <gal@nvidia.com>
Reviewed-by: Rahul Rameshbabu <rrameshbabu@nvidia.com>
Reviewed-by: Maher Sanalla <msanalla@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Handling pci errors should fully teardown and load back auxiliary
devices, same as done through mlx5 health recovery flow.
Fixes: 72ed5d5624 ("net/mlx5: Suspend auxiliary devices only in case of PCI device suspend")
Signed-off-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
When device is in error state, marked by the flag
MLX5_DEVICE_STATE_INTERNAL_ERROR, the HW and PCI may not be accessible
and so clock update work should be skipped. Furthermore, such access
through PCI in error state, after calling mlx5_pci_disable_device() can
result in failing to recover from pci errors.
Fixes: ef9814deaf ("net/mlx5e: Add HW timestamping (TS) support")
Reported-and-tested-by: Ganesh G R <ganeshgr@linux.ibm.com>
Closes: https://lore.kernel.org/netdev/9bdb9b9d-140a-7a28-f0de-2e64e873c068@nvidia.com
Signed-off-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Aya Levin <ayal@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Cited patch introduced buckets in hash mode, but missed to update
the ports/bucket check when modifying LAG.
Fix the check.
Fixes: 352899f384 ("net/mlx5: Lag, use buckets in hash mode")
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Maor Gottlieb <maorg@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
The controller number for ECVFs is always 0, because the ECPF must be
the eswitch owner for EC VFs to be enabled.
Fixes: dc13180824 ("net/mlx5: Enable devlink port for embedded cpu VF vports")
Signed-off-by: Daniel Jurgens <danielj@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
When querying eswitch functions 0 is a valid number of host VFs. After
introducing ARM SRIOV falling through to getting the max value from PCI
results in using the total VFs allowed on the ARM for the host.
Fixes: 86eec50bea ("net/mlx5: Support querying max VFs from device");
Signed-off-by: Daniel Jurgens <danielj@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
The ECVF function ID range is 1..max_ec_vfs. Currently
mlx5_vport_to_func_id returns 0..max_ec_vfs - 1. Which
results in a syndrome when querying the caps with more
recent firmware, or reading incorrect caps with older
firmware that supports EC VFs.
Fixes: 9ac0b12824 ("net/mlx5: Update vport caps query/set for EC VFs")
Signed-off-by: Daniel Jurgens <danielj@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Fixing wrong calculation of the modify hdr pattern size,
where the previously calculated number would not be enough
to accommodate the required number of actions.
Fixes: da5d0027d6 ("net/mlx5: DR, Add cache for modify header pattern")
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Erez Shitrit <erezsh@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
The flow rule can be splited, and the extra post_act rules are added
to post_act table. It's possible to trigger memleak when the rule
forwards packets from internal port and over tunnel, in the case that,
for example, CT 'new' state offload is allowed. As int_port object is
assigned to the flow attribute of post_act rule, and its refcnt is
incremented by mlx5e_tc_int_port_get(), but mlx5e_tc_int_port_put() is
not called, the refcnt is never decremented, then int_port is never
freed.
The kmemleak reports the following error:
unreferenced object 0xffff888128204b80 (size 64):
comm "handler20", pid 50121, jiffies 4296973009 (age 642.932s)
hex dump (first 32 bytes):
01 00 00 00 19 00 00 00 03 f0 00 00 04 00 00 00 ................
98 77 67 41 81 88 ff ff 98 77 67 41 81 88 ff ff .wgA.....wgA....
backtrace:
[<00000000e992680d>] kmalloc_trace+0x27/0x120
[<000000009e945a98>] mlx5e_tc_int_port_get+0x3f3/0xe20 [mlx5_core]
[<0000000035a537f0>] mlx5e_tc_add_fdb_flow+0x473/0xcf0 [mlx5_core]
[<0000000070c2cec6>] __mlx5e_add_fdb_flow+0x7cf/0xe90 [mlx5_core]
[<000000005cc84048>] mlx5e_configure_flower+0xd40/0x4c40 [mlx5_core]
[<000000004f8a2031>] mlx5e_rep_indr_offload.isra.0+0x10e/0x1c0 [mlx5_core]
[<000000007df797dc>] mlx5e_rep_indr_setup_tc_cb+0x90/0x130 [mlx5_core]
[<0000000016c15cc3>] tc_setup_cb_add+0x1cf/0x410
[<00000000a63305b4>] fl_hw_replace_filter+0x38f/0x670 [cls_flower]
[<000000008bc9e77c>] fl_change+0x1fd5/0x4430 [cls_flower]
[<00000000e7f766e4>] tc_new_tfilter+0x867/0x2010
[<00000000e101c0ef>] rtnetlink_rcv_msg+0x6fc/0x9f0
[<00000000e1111d44>] netlink_rcv_skb+0x12c/0x360
[<0000000082dd6c8b>] netlink_unicast+0x438/0x710
[<00000000fc568f70>] netlink_sendmsg+0x794/0xc50
[<0000000016e92590>] sock_sendmsg+0xc5/0x190
So fix this by moving int_port cleanup code to the flow attribute
free helper, which is used by all the attribute free cases.
Fixes: 8300f22526 ("net/mlx5e: Create new flow attr for multi table actions")
Signed-off-by: Jianbo Liu <jianbol@nvidia.com>
Reviewed-by: Vlad Buslov <vladbu@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
ionic_start_queues_reconfig returns an error code if txrx_init fails.
Handle this error code in the relevant places.
This fixes a corner case where the device could get left in a detached
state if the CMB reconfig fails and the attempt to clean up the mess
also fails. Note that calling netif_device_attach when the netdev is
already attached does not lead to unexpected behavior.
Change goto name "errout" to "err_out" to maintain consistency across
goto statements.
Fixes: 40bc471dc7 ("ionic: add tx/rx-push support with device Component Memory Buffers")
Fixes: 6f7d6f0fd7 ("ionic: pull reset_queues into tx_timeout handler")
Signed-off-by: Nitya Sunkad <nitya.sunkad@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In case rhashtable_lookup_insert_fast() fails inside vxlan_vni_add(), the
allocated percpu vni stats are not freed on the error path.
Introduce vxlan_vni_free() which would work as a nice wrapper to free
vxlan_vni_node resources properly.
Found by Linux Verification Center (linuxtesting.org).
Fixes: 4095e0e132 ("drivers: vxlan: vnifilter: per vni stats")
Suggested-by: Ido Schimmel <idosch@idosch.org>
Signed-off-by: Fedor Pchelkin <pchelkin@ispras.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
syzbot/KCSAN reported data-races in macsec whenever dev->stats fields
are updated.
It appears all of these updates can happen from multiple cpus.
Adopt SMP safe DEV_STATS_INC() to update dev->stats fields.
Fixes: c09440f7dc ("macsec: introduce IEEE 802.1AE driver")
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
TLS records end with a 16B tag. For TLS device offload we only
need to make space for this tag in the stream, the device will
generate and replace it with the actual calculated tag.
Long time ago the code would just re-reference the head frag
which mostly worked but was suboptimal because it prevented TCP
from combining the record into a single skb frag. I'm not sure
if it was correct as the first frag may be shorter than the tag.
The commit under fixes tried to replace that with using the page
frag and if the allocation failed rolling back the data, if record
was long enough. It achieves better fragment coalescing but is
also buggy.
We don't roll back the iterator, so unless we're at the end of
send we'll skip the data we designated as tag and start the
next record as if the rollback never happened.
There's also the possibility that the record was constructed
with MSG_MORE and the data came from a different syscall and
we already told the user space that we "got it".
Allocate a single dummy page and use it as fallback.
Found by code inspection, and proven by forcing allocation
failures.
Fixes: e7b159a48b ("net/tls: remove the record tail optimization")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
dccp_sendmsg() reads dp->dccps_mss_cache before locking the socket.
Same thing in do_dccp_getsockopt().
Add READ_ONCE()/WRITE_ONCE() annotations,
and change dccp_sendmsg() to check again dccps_mss_cache
after socket is locked.
Fixes: 7c657876b6 ("[DCCP]: Initial implementation")
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20230803163021.2958262-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Matthieu Baerts says:
====================
mptcp: more fixes for v6.5
Here is a new batch of fixes related to MPTCP for v6.5 and older.
Patches 1 and 2 fix issues with MPTCP Join selftest when manually
launched with '-i' parameter to use 'ip mptcp' tool instead of the
dedicated one (pm_nl_ctl). The issues have been there since v5.18.
Thank you Andrea for your first contributions to MPTCP code in the
upstream kernel!
Patch 3 avoids corrupting the data stream when trying to reset
connections that have fallen back to TCP. This can happen from v6.1.
Patch 4 fixes a race when doing a disconnect() and an accept() in
parallel on a listener socket. The issue only happens in rare cases if
the user is really unlucky since a fix that landed in v6.3 but
backported up to v6.1.
====================
Link: https://lore.kernel.org/r/20230803-upstream-net-20230803-misc-fixes-6-5-v1-0-6671b1ab11cc@tessares.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Despite commit 0ad529d9fd ("mptcp: fix possible divide by zero in
recvmsg()"), the mptcp protocol is still prone to a race between
disconnect() (or shutdown) and accept.
The root cause is that the mentioned commit checks the msk-level
flag, but mptcp_stream_accept() does acquire the msk-level lock,
as it can rely directly on the first subflow lock.
As reported by Christoph than can lead to a race where an msk
socket is accepted after that mptcp_subflow_queue_clean() releases
the listener socket lock and just before it takes destructive
actions leading to the following splat:
BUG: kernel NULL pointer dereference, address: 0000000000000012
PGD 5a4ca067 P4D 5a4ca067 PUD 37d4c067 PMD 0
Oops: 0000 [#1] PREEMPT SMP
CPU: 2 PID: 10955 Comm: syz-executor.5 Not tainted 6.5.0-rc1-gdc7b257ee5dd #37
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.11.0-2.el7 04/01/2014
RIP: 0010:mptcp_stream_accept+0x1ee/0x2f0 include/net/inet_sock.h:330
Code: 0a 09 00 48 8b 1b 4c 39 e3 74 07 e8 bc 7c 7f fe eb a1 e8 b5 7c 7f fe 4c 8b 6c 24 08 eb 05 e8 a9 7c 7f fe 49 8b 85 d8 09 00 00 <0f> b6 40 12 88 44 24 07 0f b6 6c 24 07 bf 07 00 00 00 89 ee e8 89
RSP: 0018:ffffc90000d07dc0 EFLAGS: 00010293
RAX: 0000000000000000 RBX: ffff888037e8d020 RCX: ffff88803b093300
RDX: 0000000000000000 RSI: ffffffff833822c5 RDI: ffffffff8333896a
RBP: 0000607f82031520 R08: ffff88803b093300 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000003e83 R12: ffff888037e8d020
R13: ffff888037e8c680 R14: ffff888009af7900 R15: ffff888009af6880
FS: 00007fc26d708640(0000) GS:ffff88807dd00000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000012 CR3: 0000000066bc5001 CR4: 0000000000370ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
<TASK>
do_accept+0x1ae/0x260 net/socket.c:1872
__sys_accept4+0x9b/0x110 net/socket.c:1913
__do_sys_accept4 net/socket.c:1954 [inline]
__se_sys_accept4 net/socket.c:1951 [inline]
__x64_sys_accept4+0x20/0x30 net/socket.c:1951
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x47/0xa0 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x6e/0xd8
Address the issue by temporary removing the pending request socket
from the accept queue, so that racing accept() can't touch them.
After depleting the msk - the ssk still exists, as plain TCP sockets,
re-insert them into the accept queue, so that later inet_csk_listen_stop()
will complete the tcp socket disposal.
Fixes: 2a6a870e44 ("mptcp: stops worker on unaccepted sockets at listener close")
Cc: stable@vger.kernel.org
Reported-by: Christoph Paasch <cpaasch@apple.com>
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/423
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Link: https://lore.kernel.org/r/20230803-upstream-net-20230803-misc-fixes-6-5-v1-4-6671b1ab11cc@tessares.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Since the blamed commit, the MPTCP protocol unconditionally sends
TCP resets on all the subflows on disconnect().
That fits full-blown MPTCP sockets - to implement the fastclose
mechanism - but causes unexpected corruption of the data stream,
caught as sporadic self-tests failures.
Fixes: d21f834855 ("mptcp: use fastclose on more edge scenarios")
Cc: stable@vger.kernel.org
Tested-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/419
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Link: https://lore.kernel.org/r/20230803-upstream-net-20230803-misc-fixes-6-5-v1-3-6671b1ab11cc@tessares.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
mptcp_join 'implicit EP' test currently fails when using ip mptcp:
$ ./mptcp_join.sh -iI
<snip>
001 implicit EP creation[fail] expected '10.0.2.2 10.0.2.2 id 1 implicit' found '10.0.2.2 id 1 rawflags 10 '
Error: too many addresses or duplicate one: -22.
ID change is prevented[fail] expected '10.0.2.2 10.0.2.2 id 1 implicit' found '10.0.2.2 id 1 rawflags 10 '
modif is allowed[fail] expected '10.0.2.2 10.0.2.2 id 1 signal' found '10.0.2.2 id 1 signal '
This happens because of two reasons:
- iproute v6.3.0 does not support the implicit flag, fixed with
iproute2-next commit 3a2535a41854 ("mptcp: add support for implicit
flag")
- pm_nl_check_endpoint wrongly expects the ip address to be repeated two
times in iproute output, and does not account for a final whitespace
in it.
This fixes the issue trimming the whitespace in the output string and
removing the double address in the expected string.
Fixes: 69c6ce7b6e ("selftests: mptcp: add implicit endpoint test case")
Cc: stable@vger.kernel.org
Signed-off-by: Andrea Claudi <aclaudi@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Link: https://lore.kernel.org/r/20230803-upstream-net-20230803-misc-fixes-6-5-v1-2-6671b1ab11cc@tessares.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
mptcp_join 'delete and re-add' test fails when using ip mptcp:
$ ./mptcp_join.sh -iI
<snip>
002 delete and re-add before delete[ ok ]
mptcp_info subflows=1 [ ok ]
Error: argument "ADDRESS" is wrong: invalid for non-zero id address
after delete[fail] got 2:2 subflows expected 1
This happens because endpoint delete includes an ip address while id is
not 0, contrary to what is indicated in the ip mptcp man page:
"When used with the delete id operation, an IFADDR is only included when
the ID is 0."
This fixes the issue using the $addr variable in pm_nl_del_endpoint()
only when id is 0.
Fixes: 34aa6e3bcc ("selftests: mptcp: add ip mptcp wrappers")
Cc: stable@vger.kernel.org
Signed-off-by: Andrea Claudi <aclaudi@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Link: https://lore.kernel.org/r/20230803-upstream-net-20230803-misc-fixes-6-5-v1-1-6671b1ab11cc@tessares.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Florian Westphal says:
====================
tunnels: fix ipv4 pmtu icmp checksum
The checksum of the generated ipv4 icmp pmtud message is
only correct if the skb that causes the icmp error generation
is linear.
Fix this and add a selftest for this.
====================
Link: https://lore.kernel.org/r/20230803152653.29535-1-fw@strlen.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
TCP might get stuck if a nonlinear skb exceeds the path MTU,
icmp error contains an incorrect icmp checksum in that case.
Extend the existing test for vxlan to also send at least 1MB worth of
data via TCP in addition to the existing 'large icmp packet adds
route exception'.
On my test VM this fails due to 0-size output file without
"tunnels: fix kasan splat when generating ipv4 pmtu error".
Signed-off-by: Florian Westphal <fw@strlen.de>
Link: https://lore.kernel.org/r/20230803152653.29535-3-fw@strlen.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
If we try to emit an icmp error in response to a nonliner skb, we get
BUG: KASAN: slab-out-of-bounds in ip_compute_csum+0x134/0x220
Read of size 4 at addr ffff88811c50db00 by task iperf3/1691
CPU: 2 PID: 1691 Comm: iperf3 Not tainted 6.5.0-rc3+ #309
[..]
kasan_report+0x105/0x140
ip_compute_csum+0x134/0x220
iptunnel_pmtud_build_icmp+0x554/0x1020
skb_tunnel_check_pmtu+0x513/0xb80
vxlan_xmit_one+0x139e/0x2ef0
vxlan_xmit+0x1867/0x2760
dev_hard_start_xmit+0x1ee/0x4f0
br_dev_queue_push_xmit+0x4d1/0x660
[..]
ip_compute_csum() cannot deal with nonlinear skbs, so avoid it.
After this change, splat is gone and iperf3 is no longer stuck.
Fixes: 4cb47a8644 ("tunnels: PMTU discovery support for directly bridged IP packets")
Signed-off-by: Florian Westphal <fw@strlen.de>
Link: https://lore.kernel.org/r/20230803152653.29535-2-fw@strlen.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When the tagging protocol in current use is "ocelot-8021q" and we unbind
the driver, we see this splat:
$ echo '0000:00:00.2' > /sys/bus/pci/drivers/fsl_enetc/unbind
mscc_felix 0000:00:00.5 swp0: left promiscuous mode
sja1105 spi2.0: Link is Down
DSA: tree 1 torn down
mscc_felix 0000:00:00.5 swp2: left promiscuous mode
sja1105 spi2.2: Link is Down
DSA: tree 3 torn down
fsl_enetc 0000:00:00.2 eno2: left promiscuous mode
mscc_felix 0000:00:00.5: Link is Down
------------[ cut here ]------------
RTNL: assertion failed at net/dsa/tag_8021q.c (409)
WARNING: CPU: 1 PID: 329 at net/dsa/tag_8021q.c:409 dsa_tag_8021q_unregister+0x12c/0x1a0
Modules linked in:
CPU: 1 PID: 329 Comm: bash Not tainted 6.5.0-rc3+ #771
pc : dsa_tag_8021q_unregister+0x12c/0x1a0
lr : dsa_tag_8021q_unregister+0x12c/0x1a0
Call trace:
dsa_tag_8021q_unregister+0x12c/0x1a0
felix_tag_8021q_teardown+0x130/0x150
felix_teardown+0x3c/0xd8
dsa_tree_teardown_switches+0xbc/0xe0
dsa_unregister_switch+0x168/0x260
felix_pci_remove+0x30/0x60
pci_device_remove+0x4c/0x100
device_release_driver_internal+0x188/0x288
device_links_unbind_consumers+0xfc/0x138
device_release_driver_internal+0xe0/0x288
device_driver_detach+0x24/0x38
unbind_store+0xd8/0x108
drv_attr_store+0x30/0x50
---[ end trace 0000000000000000 ]---
------------[ cut here ]------------
RTNL: assertion failed at net/8021q/vlan_core.c (376)
WARNING: CPU: 1 PID: 329 at net/8021q/vlan_core.c:376 vlan_vid_del+0x1b8/0x1f0
CPU: 1 PID: 329 Comm: bash Tainted: G W 6.5.0-rc3+ #771
pc : vlan_vid_del+0x1b8/0x1f0
lr : vlan_vid_del+0x1b8/0x1f0
dsa_tag_8021q_unregister+0x8c/0x1a0
felix_tag_8021q_teardown+0x130/0x150
felix_teardown+0x3c/0xd8
dsa_tree_teardown_switches+0xbc/0xe0
dsa_unregister_switch+0x168/0x260
felix_pci_remove+0x30/0x60
pci_device_remove+0x4c/0x100
device_release_driver_internal+0x188/0x288
device_links_unbind_consumers+0xfc/0x138
device_release_driver_internal+0xe0/0x288
device_driver_detach+0x24/0x38
unbind_store+0xd8/0x108
drv_attr_store+0x30/0x50
DSA: tree 0 torn down
This was somewhat not so easy to spot, because "ocelot-8021q" is not the
default tagging protocol, and thus, not everyone who tests the unbinding
path may have switched to it beforehand. The default
felix_tag_npi_teardown() does not require rtnl_lock() to be held.
Fixes: 7c83a7c539 ("net: dsa: add a second tagger for Ocelot switches based on tag_8021q")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://lore.kernel.org/r/20230803134253.2711124-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
