selftests: net: bridge: increase IGMP/MLD exclude timeout membership interval

When running the bridge IGMP/MLD selftests on debug kernels we can get
spurious errors when setting up the IGMP/MLD exclude timeout tests
because the membership interval is just 3 seconds and the setup has 2
seconds of sleep plus various validations, the one second that is left
is not enough. Increase the membership interval from 3 to 5 seconds to
make room for the setup validation and 2 seconds of sleep.

Fixes: 34d7ecb3d4f7 ("selftests: net: bridge: update IGMP/MLD membership interval value")
Reported-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: bridge: xmit: make sure we have at least eth header len bytes

syzbot triggered an uninit value[1] error in bridge device's xmit path
by sending a short (less than ETH_HLEN bytes) skb. To fix it check if
we can actually pull that amount instead of assuming.

Tested with dropwatch:
 drop at: br_dev_xmit+0xb93/0x12d0 [bridge] (0xffffffffc06739b3)
 origin: software
 timestamp: Mon May 13 11:31:53 2024 778214037 nsec
 protocol: 0x88a8
 length: 2
 original length: 2
 drop reason: PKT_TOO_SMALL

[1]
BUG: KMSAN: uninit-value in br_dev_xmit+0x61d/0x1cb0 net/bridge/br_device.c:65
 br_dev_xmit+0x61d/0x1cb0 net/bridge/br_device.c:65
 __netdev_start_xmit include/linux/netdevice.h:4903 [inline]
 netdev_start_xmit include/linux/netdevice.h:4917 [inline]
 xmit_one net/core/dev.c:3531 [inline]
 dev_hard_start_xmit+0x247/0xa20 net/core/dev.c:3547
 __dev_queue_xmit+0x34db/0x5350 net/core/dev.c:4341
 dev_queue_xmit include/linux/netdevice.h:3091 [inline]
 __bpf_tx_skb net/core/filter.c:2136 [inline]
 __bpf_redirect_common net/core/filter.c:2180 [inline]
 __bpf_redirect+0x14a6/0x1620 net/core/filter.c:2187
 ____bpf_clone_redirect net/core/filter.c:2460 [inline]
 bpf_clone_redirect+0x328/0x470 net/core/filter.c:2432
 ___bpf_prog_run+0x13fe/0xe0f0 kernel/bpf/core.c:1997
 __bpf_prog_run512+0xb5/0xe0 kernel/bpf/core.c:2238
 bpf_dispatcher_nop_func include/linux/bpf.h:1234 [inline]
 __bpf_prog_run include/linux/filter.h:657 [inline]
 bpf_prog_run include/linux/filter.h:664 [inline]
 bpf_test_run+0x499/0xc30 net/bpf/test_run.c:425
 bpf_prog_test_run_skb+0x14ea/0x1f20 net/bpf/test_run.c:1058
 bpf_prog_test_run+0x6b7/0xad0 kernel/bpf/syscall.c:4269
 __sys_bpf+0x6aa/0xd90 kernel/bpf/syscall.c:5678
 __do_sys_bpf kernel/bpf/syscall.c:5767 [inline]
 __se_sys_bpf kernel/bpf/syscall.c:5765 [inline]
 __x64_sys_bpf+0xa0/0xe0 kernel/bpf/syscall.c:5765
 x64_sys_call+0x96b/0x3b50 arch/x86/include/generated/asm/syscalls_64.h:322
 do_syscall_x64 arch/x86/entry/common.c:52 [inline]
 do_syscall_64+0xcf/0x1e0 arch/x86/entry/common.c:83
 entry_SYSCALL_64_after_hwframe+0x77/0x7f

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Reported-by: syzbot+a63a1f6a062033cf0f40@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=a63a1f6a062033cf0f40
Signed-off-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: micrel: Fix receiving the timestamp in the frame for lan8841

The blamed commit started to use the ptp workqueue to get the second
part of the timestamp. And when the port was set down, then this
workqueue is stopped. But if the config option NETWORK_PHY_TIMESTAMPING
is not enabled, then the ptp_clock is not initialized so then it would
crash when it would try to access the delayed work.
So then basically by setting up and then down the port, it would crash.
The fix consists in checking if the ptp_clock is initialized and only
then cancel the delayed work.

Fixes: cc7554954848 ("net: micrel: Change to receive timestamp in the frame for lan8841")
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: fec: remove .ndo_poll_controller to avoid deadlocks

There is a deadlock issue found in sungem driver, please refer to the
commit ac0a230f719b ("eth: sungem: remove .ndo_poll_controller to avoid
deadlocks"). The root cause of the issue is that netpoll is in atomic
context and disable_irq() is called by .ndo_poll_controller interface
of sungem driver, however, disable_irq() might sleep. After analyzing
the implementation of fec_poll_controller(), the fec driver should have
the same issue. Due to the fec driver uses NAPI for TX completions, the
.ndo_poll_controller is unnecessary to be implemented in the fec driver,
so fec_poll_controller() can be safely removed.

Fixes: 7f5c6addcdc0 ("net/fec: add poll controller function for fec nic")
Signed-off-by: Wei Fang <wei.fang@nxp.com>
Link: https://lore.kernel.org/r/20240511062009.652918-1-wei.fang@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

dpll: fix return value check for kmemdup

The return value of kmemdup() is dst->freq_supported, not
src->freq_supported. Update the check accordingly.

Fixes: 830ead5fb0c5 ("dpll: fix pin dump crash for rebound module")
Signed-off-by: Chen Ni <nichen@iscas.ac.cn>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Reviewed-by: Arkadiusz Kubalewski <arkadiusz.kubalewski@intel.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20240513032824.2410459-1-nichen@iscas.ac.cn
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'ax25-fix-issues-of-ax25_dev-and-net_device'

Duoming Zhou says:

====================
ax25: Fix issues of ax25_dev and net_device

The first patch uses kernel universal linked list to implement
ax25_dev_list, which makes the operation of the list easier.
The second and third patch fix reference count leak issues of
the object "ax25_dev" and "net_device".
====================

Link: https://lore.kernel.org/r/cover.1715247018.git.duoming@zju.edu.cn
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ax25: Fix reference count leak issue of net_device

There is a reference count leak issue of the object "net_device" in
ax25_dev_device_down(). When the ax25 device is shutting down, the
ax25_dev_device_down() drops the reference count of net_device one
or zero times depending on if we goto unlock_put or not, which will
cause memory leak.

In order to solve the above issue, decrease the reference count of
net_device after dev->ax25_ptr is set to null.

Fixes: d01ffb9eee4a ("ax25: add refcount in ax25_dev to avoid UAF bugs")
Suggested-by: Dan Carpenter <dan.carpenter@linaro.org>
Signed-off-by: Duoming Zhou <duoming@zju.edu.cn>
Reviewed-by: Dan Carpenter <dan.carpenter@linaro.org>
Link: https://lore.kernel.org/r/7ce3b23a40d9084657ba1125432f0ecc380cbc80.1715247018.git.duoming@zju.edu.cn
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ax25: Fix reference count leak issues of ax25_dev

The ax25_addr_ax25dev() and ax25_dev_device_down() exist a reference
count leak issue of the object "ax25_dev".

Memory leak issue in ax25_addr_ax25dev():

The reference count of the object "ax25_dev" can be increased multiple
times in ax25_addr_ax25dev(). This will cause a memory leak.

Memory leak issues in ax25_dev_device_down():

The reference count of ax25_dev is set to 1 in ax25_dev_device_up() and
then increase the reference count when ax25_dev is added to ax25_dev_list.
As a result, the reference count of ax25_dev is 2. But when the device is
shutting down. The ax25_dev_device_down() drops the reference count once
or twice depending on if we goto unlock_put or not, which will cause
memory leak.

As for the issue of ax25_addr_ax25dev(), it is impossible for one pointer
to be on a list twice. So add a break in ax25_addr_ax25dev(). As for the
issue of ax25_dev_device_down(), increase the reference count of ax25_dev
once in ax25_dev_device_up() and decrease the reference count of ax25_dev
after it is removed from the ax25_dev_list.

Fixes: d01ffb9eee4a ("ax25: add refcount in ax25_dev to avoid UAF bugs")
Suggested-by: Dan Carpenter <dan.carpenter@linaro.org>
Signed-off-by: Duoming Zhou <duoming@zju.edu.cn>
Reviewed-by: Dan Carpenter <dan.carpenter@linaro.org>
Link: https://lore.kernel.org/r/361bbf2a4b091e120006279ec3b382d73c4a0c17.1715247018.git.duoming@zju.edu.cn
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ax25: Use kernel universal linked list to implement ax25_dev_list

The origin ax25_dev_list implements its own single linked list,
which is complicated and error-prone. For example, when deleting
the node of ax25_dev_list in ax25_dev_device_down(), we have to
operate on the head node and other nodes separately.

This patch uses kernel universal linked list to replace original
ax25_dev_list, which make the operation of ax25_dev_list easier.

We should do "dev->ax25_ptr = ax25_dev;" and "dev->ax25_ptr = NULL;"
while holding the spinlock, otherwise the ax25_dev_device_up() and
ax25_dev_device_down() could race.

Suggested-by: Dan Carpenter <dan.carpenter@linaro.org>
Signed-off-by: Duoming Zhou <duoming@zju.edu.cn>
Reviewed-by: Dan Carpenter <dan.carpenter@linaro.org>
Link: https://lore.kernel.org/r/85bba3af651ca0e1a519da8d0d715b949891171c.1715247018.git.duoming@zju.edu.cn
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: usb: ax88179_178a: fix link status when link is set to down/up

The idea was to keep only one reset at initialization stage in order to
reduce the total delay, or the reset from usbnet_probe or the reset from
usbnet_open.

I have seen that restarting from usbnet_probe is necessary to avoid doing
too complex things. But when the link is set to down/up (for example to
configure a different mac address) the link is not correctly recovered
unless a reset is commanded from usbnet_open.

So, detect the initialization stage (first call) to not reset from
usbnet_open after the reset from usbnet_probe and after this stage, always
reset from usbnet_open too (when the link needs to be rechecked).

Apply to all the possible devices, the behavior now is going to be the same.

cc: stable@vger.kernel.org # 6.6+
Fixes: 56f78615bcb1 ("net: usb: ax88179_178a: avoid writing the mac address before first reading")
Reported-by: Isaac Ganoung <inventor500@vivaldi.net>
Reported-by: Yongqin Liu <yongqin.liu@linaro.org>
Signed-off-by: Jose Ignacio Tornos Martinez <jtornosm@redhat.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240510090846.328201-1-jtornosm@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: smc91x: Fix m68k kernel compilation for ColdFire CPU

Compiling the m68k kernel with support for the ColdFire CPU family fails
with the following error:

In file included from drivers/net/ethernet/smsc/smc91x.c:80:
drivers/net/ethernet/smsc/smc91x.c: In function ‘smc_reset’:
drivers/net/ethernet/smsc/smc91x.h:160:40: error: implicit declaration of function ‘_swapw’; did you mean ‘swap’? [-Werror=implicit-function-declaration]
  160 | #define SMC_outw(lp, v, a, r)   writew(_swapw(v), (a) + (r))
      |                                        ^~~~~~
drivers/net/ethernet/smsc/smc91x.h:904:25: note: in expansion of macro ‘SMC_outw’
  904 |                         SMC_outw(lp, x, ioaddr, BANK_SELECT);           \
      |                         ^~~~~~~~
drivers/net/ethernet/smsc/smc91x.c:250:9: note: in expansion of macro ‘SMC_SELECT_BANK’
  250 |         SMC_SELECT_BANK(lp, 2);
      |         ^~~~~~~~~~~~~~~
cc1: some warnings being treated as errors

The function _swapw() was removed in commit d97cf70af097 ("m68k: use
asm-generic/io.h for non-MMU io access functions"), but is still used in
drivers/net/ethernet/smsc/smc91x.h.

Use ioread16be() and iowrite16be() to resolve the error.

Cc: stable@vger.kernel.org
Fixes: d97cf70af097 ("m68k: use asm-generic/io.h for non-MMU io access functions")
Signed-off-by: Thorsten Blum <thorsten.blum@toblux.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://lore.kernel.org/r/20240510113054.186648-2-thorsten.blum@toblux.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

netlabel: fix RCU annotation for IPv4 options on socket creation

Xiumei reports the following splat when netlabel and TCP socket are used:

 =============================
 WARNING: suspicious RCU usage
 6.9.0-rc2+ #637 Not tainted
 -----------------------------
 net/ipv4/cipso_ipv4.c:1880 suspicious rcu_dereference_protected() usage!

 other info that might help us debug this:

 rcu_scheduler_active = 2, debug_locks = 1
 1 lock held by ncat/23333:
  #0: ffffffff906030c0 (rcu_read_lock){....}-{1:2}, at: netlbl_sock_setattr+0x25/0x1b0

 stack backtrace:
 CPU: 11 PID: 23333 Comm: ncat Kdump: loaded Not tainted 6.9.0-rc2+ #637
 Hardware name: Supermicro SYS-6027R-72RF/X9DRH-7TF/7F/iTF/iF, BIOS 3.0  07/26/2013
 Call Trace:
  <TASK>
  dump_stack_lvl+0xa9/0xc0
  lockdep_rcu_suspicious+0x117/0x190
  cipso_v4_sock_setattr+0x1ab/0x1b0
  netlbl_sock_setattr+0x13e/0x1b0
  selinux_netlbl_socket_post_create+0x3f/0x80
  selinux_socket_post_create+0x1a0/0x460
  security_socket_post_create+0x42/0x60
  __sock_create+0x342/0x3a0
  __sys_socket_create.part.22+0x42/0x70
  __sys_socket+0x37/0xb0
  __x64_sys_socket+0x16/0x20
  do_syscall_64+0x96/0x180
  ? do_user_addr_fault+0x68d/0xa30
  ? exc_page_fault+0x171/0x280
  ? asm_exc_page_fault+0x22/0x30
  entry_SYSCALL_64_after_hwframe+0x71/0x79
 RIP: 0033:0x7fbc0ca3fc1b
 Code: 73 01 c3 48 8b 0d 05 f2 1b 00 f7 d8 64 89 01 48 83 c8 ff c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa b8 29 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d d5 f1 1b 00 f7 d8 64 89 01 48
 RSP: 002b:00007fff18635208 EFLAGS: 00000246 ORIG_RAX: 0000000000000029
 RAX: ffffffffffffffda RBX: 0000000000000001 RCX: 00007fbc0ca3fc1b
 RDX: 0000000000000006 RSI: 0000000000000001 RDI: 0000000000000002
 RBP: 000055d24f80f8a0 R08: 0000000000000003 R09: 0000000000000001

R10: 0000000000020000 R11: 0000000000000246 R12: 000055d24f80f8a0
 R13: 0000000000000000 R14: 000055d24f80fb88 R15: 0000000000000000
  </TASK>

The current implementation of cipso_v4_sock_setattr() replaces IP options
under the assumption that the caller holds the socket lock; however, such
assumption is not true, nor needed, in selinux_socket_post_create() hook.

Let all callers of cipso_v4_sock_setattr() specify the "socket lock held"
condition, except selinux_socket_post_create() _ where such condition can
safely be set as true even without holding the socket lock.

Fixes: f6d8bd051c39 ("inet: add RCU protection to inet->opt")
Reported-by: Xiumei Mu <xmu@redhat.com>
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Acked-by: Casey Schaufler <casey@schaufler-ca.com>
Acked-by: Paul Moore <paul@paul-moore.com>
Link: https://lore.kernel.org/r/f4260d000a3a55b9e8b6a3b4e3fffc7da9f82d41.1715359817.git.dcaratti@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

inet: fix inet_fill_ifaddr() flags truncation

I missed that (struct ifaddrmsg)->ifa_flags was only 8bits,
while (struct in_ifaddr)->ifa_flags is 32bits.

Use a temporary 32bit variable as I did in set_ifa_lifetime()
and check_lifetime().

Fixes: 3ddc2231c810 ("inet: annotate data-races around ifa->ifa_flags")
Reported-by: Yu Watanabe <watanabe.yu@gmail.com>
Dianosed-by: Yu Watanabe <watanabe.yu@gmail.com>
Closes: https://github.com/systemd/systemd/pull/32666#issuecomment-2103977928
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Larysa Zaremba <larysa.zaremba@intel.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240510072932.2678952-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge tag 'for-netdev' of https://git./linux/kernel/git/bpf/bpf

Daniel Borkmann says:

====================
pull-request: bpf 2024-05-13

We've added 3 non-merge commits during the last 2 day(s) which contain
a total of 2 files changed, 62 insertions(+), 8 deletions(-).

The main changes are:

1) Fix a case where syzkaller found that it's unexpectedly possible
   to attach a cgroup_skb program to the sockopt hooks. The fix adds
   missing attach_type enforcement for the link_create case along
   with selftests, from Stanislav Fomichev.

* tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf:
  selftests/bpf: Add sockopt case to verify prog_type
  selftests/bpf: Extend sockopt tests to use BPF_LINK_CREATE
  bpf: Add BPF_PROG_TYPE_CGROUP_SKB attach type enforcement in BPF_LINK_CREATE
====================

Link: https://lore.kernel.org/r/20240513041845.31040-1-daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'mlx5-misc-fixes'

Tariq Toukan says:

====================
mlx5 misc fixes

This patchset provides bug fixes to mlx5 driver.

Patch 1 by Shay fixes the error flow in mlx5e_suspend().
Patch 2 by Shay aligns the peer devlink set logic with the register devlink flow.
Patch 3 by Maher solves a deadlock in lag enable/disable.
Patches 4 and 5 by Akiva address issues in command interface corner cases.

Series generated against:
commit 393ceeb9211e ("Merge branch 'there-are-some-bugfix-for-the-hns3-ethernet-driver'")
====================

Link: https://lore.kernel.org/r/20240509112951.590184-1-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/mlx5: Discard command completions in internal error

Fix use after free when FW completion arrives while device is in
internal error state. Avoid calling completion handler in this case,
since the device will flush the command interface and trigger all
completions manually.

Kernel log:
------------[ cut here ]------------
refcount_t: underflow; use-after-free.
...
RIP: 0010:refcount_warn_saturate+0xd8/0xe0
...
Call Trace:
<IRQ>
? __warn+0x79/0x120
? refcount_warn_saturate+0xd8/0xe0
? report_bug+0x17c/0x190
? handle_bug+0x3c/0x60
? exc_invalid_op+0x14/0x70
? asm_exc_invalid_op+0x16/0x20
? refcount_warn_saturate+0xd8/0xe0
cmd_ent_put+0x13b/0x160 [mlx5_core]
mlx5_cmd_comp_handler+0x5f9/0x670 [mlx5_core]
cmd_comp_notifier+0x1f/0x30 [mlx5_core]
notifier_call_chain+0x35/0xb0
atomic_notifier_call_chain+0x16/0x20
mlx5_eq_async_int+0xf6/0x290 [mlx5_core]
notifier_call_chain+0x35/0xb0
atomic_notifier_call_chain+0x16/0x20
irq_int_handler+0x19/0x30 [mlx5_core]
__handle_irq_event_percpu+0x4b/0x160
handle_irq_event+0x2e/0x80
handle_edge_irq+0x98/0x230
__common_interrupt+0x3b/0xa0
common_interrupt+0x7b/0xa0
</IRQ>
<TASK>
asm_common_interrupt+0x22/0x40

Fixes: 51d138c2610a ("net/mlx5: Fix health error state handling")
Signed-off-by: Akiva Goldberger <agoldberger@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://lore.kernel.org/r/20240509112951.590184-6-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/mlx5: Add a timeout to acquire the command queue semaphore

Prevent forced completion handling on an entry that has not yet been
assigned an index, causing an out of bounds access on idx = -22.
Instead of waiting indefinitely for the sem, blocking flow now waits for
index to be allocated or a sem acquisition timeout before beginning the
timer for FW completion.

Kernel log example:
mlx5_core 0000:06:00.0: wait_func_handle_exec_timeout:1128:(pid 185911): cmd[-22]: CREATE_UCTX(0xa04) No done completion

Fixes: 8e715cd613a1 ("net/mlx5: Set command entry semaphore up once got index free")
Signed-off-by: Akiva Goldberger <agoldberger@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://lore.kernel.org/r/20240509112951.590184-5-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/mlx5: Reload only IB representors upon lag disable/enable

On lag disable, the bond IB device along with all of its
representors are destroyed, and then the slaves' representors get reloaded.

In case the slave IB representor load fails, the eswitch error flow
unloads all representors, including ethernet representors, where the
netdevs get detached and removed from lag bond. Such flow is inaccurate
as the lag driver is not responsible for loading/unloading ethernet
representors. Furthermore, the flow described above begins by holding
lag lock to prevent bond changes during disable flow. However, when
reaching the ethernet representors detachment from lag, the lag lock is
required again, triggering the following deadlock:

Call trace:
__switch_to+0xf4/0x148
__schedule+0x2c8/0x7d0
schedule+0x50/0xe0
schedule_preempt_disabled+0x18/0x28
__mutex_lock.isra.13+0x2b8/0x570
__mutex_lock_slowpath+0x1c/0x28
mutex_lock+0x4c/0x68
mlx5_lag_remove_netdev+0x3c/0x1a0 [mlx5_core]
mlx5e_uplink_rep_disable+0x70/0xa0 [mlx5_core]
mlx5e_detach_netdev+0x6c/0xb0 [mlx5_core]
mlx5e_netdev_change_profile+0x44/0x138 [mlx5_core]
mlx5e_netdev_attach_nic_profile+0x28/0x38 [mlx5_core]
mlx5e_vport_rep_unload+0x184/0x1b8 [mlx5_core]
mlx5_esw_offloads_rep_load+0xd8/0xe0 [mlx5_core]
mlx5_eswitch_reload_reps+0x74/0xd0 [mlx5_core]
mlx5_disable_lag+0x130/0x138 [mlx5_core]
mlx5_lag_disable_change+0x6c/0x70 [mlx5_core] // hold ldev->lock
mlx5_devlink_eswitch_mode_set+0xc0/0x410 [mlx5_core]
devlink_nl_cmd_eswitch_set_doit+0xdc/0x180
genl_family_rcv_msg_doit.isra.17+0xe8/0x138
genl_rcv_msg+0xe4/0x220
netlink_rcv_skb+0x44/0x108
genl_rcv+0x40/0x58
netlink_unicast+0x198/0x268
netlink_sendmsg+0x1d4/0x418
sock_sendmsg+0x54/0x60
__sys_sendto+0xf4/0x120
__arm64_sys_sendto+0x30/0x40
el0_svc_common+0x8c/0x120
do_el0_svc+0x30/0xa0
el0_svc+0x20/0x30
el0_sync_handler+0x90/0xb8
el0_sync+0x160/0x180

Thus, upon lag enable/disable, load and unload only the IB representors
of the slaves preventing the deadlock mentioned above.

While at it, refactor the mlx5_esw_offloads_rep_load() function to have
a static helper method for its internal logic, in symmetry with the
representor unload design.

Fixes: 598fe77df855 ("net/mlx5: Lag, Create shared FDB when in switchdev mode")
Co-developed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Maher Sanalla <msanalla@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240509112951.590184-4-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/mlx5: Fix peer devlink set for SF representor devlink port

The cited patch change register devlink flow, and neglect to reflect
the changes for peer devlink set logic. Peer devlink set is
triggering a call trace if done after devl_register.[1]

Hence, align peer devlink set logic with register devlink flow.

[1]
WARNING: CPU: 4 PID: 3394 at net/devlink/core.c:155 devlink_rel_nested_in_add+0x177/0x180
CPU: 4 PID: 3394 Comm: kworker/u40:1 Not tainted 6.9.0-rc4_for_linust_min_debug_2024_04_16_14_08 #1
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
Workqueue: mlx5_vhca_event0 mlx5_vhca_state_work_handler [mlx5_core]
RIP: 0010:devlink_rel_nested_in_add+0x177/0x180
Call Trace:
 <TASK>
 ? __warn+0x78/0x120
 ? devlink_rel_nested_in_add+0x177/0x180
 ? report_bug+0x16d/0x180
 ? handle_bug+0x3c/0x60
 ? exc_invalid_op+0x14/0x70
 ? asm_exc_invalid_op+0x16/0x20
 ? devlink_port_init+0x30/0x30
 ? devlink_port_type_clear+0x50/0x50
 ? devlink_rel_nested_in_add+0x177/0x180
 ? devlink_rel_nested_in_add+0xdd/0x180
 mlx5_sf_mdev_event+0x74/0xb0 [mlx5_core]
 notifier_call_chain+0x35/0xb0
 blocking_notifier_call_chain+0x3d/0x60
 mlx5_blocking_notifier_call_chain+0x22/0x30 [mlx5_core]
 mlx5_sf_dev_probe+0x185/0x3e0 [mlx5_core]
 auxiliary_bus_probe+0x38/0x80
 ? driver_sysfs_add+0x51/0x80
 really_probe+0xc5/0x3a0
 ? driver_probe_device+0x90/0x90
 __driver_probe_device+0x80/0x160
 driver_probe_device+0x1e/0x90
 __device_attach_driver+0x7d/0x100
 bus_for_each_drv+0x80/0xd0
 __device_attach+0xbc/0x1f0
 bus_probe_device+0x86/0xa0
 device_add+0x64f/0x860
 __auxiliary_device_add+0x3b/0xa0
 mlx5_sf_dev_add+0x139/0x330 [mlx5_core]
 mlx5_sf_dev_state_change_handler+0x1e4/0x250 [mlx5_core]
 notifier_call_chain+0x35/0xb0
 blocking_notifier_call_chain+0x3d/0x60
 mlx5_vhca_state_work_handler+0x151/0x200 [mlx5_core]
 process_one_work+0x13f/0x2e0
 worker_thread+0x2bd/0x3c0
 ? rescuer_thread+0x410/0x410
 kthread+0xc4/0xf0
 ? kthread_complete_and_exit+0x20/0x20
 ret_from_fork+0x2d/0x50
 ? kthread_complete_and_exit+0x20/0x20
 ret_from_fork_asm+0x11/0x20
 </TASK>

Fixes: bf729988303a ("net/mlx5: Restore mistakenly dropped parts in register devlink flow")
Fixes: c6e77aa9dd82 ("net/mlx5: Register devlink first under devlink lock")
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240509112951.590184-3-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/mlx5e: Fix netif state handling

mlx5e_suspend cleans resources only if netif_device_present() returns
true. However, mlx5e_resume changes the state of netif, via
mlx5e_nic_enable, only if reg_state == NETREG_REGISTERED.
In the below case, the above leads to NULL-ptr Oops[1] and memory
leaks:

mlx5e_probe
 _mlx5e_resume
  mlx5e_attach_netdev
   mlx5e_nic_enable  <-- netdev not reg, not calling netif_device_attach()
  register_netdev <-- failed for some reason.
ERROR_FLOW:
 _mlx5e_suspend <-- netif_device_present return false, resources aren't freed :(

Hence, clean resources in this case as well.

[1]
BUG: kernel NULL pointer dereference, address: 0000000000000000
PGD 0 P4D 0
Oops: 0010 [#1] SMP
CPU: 2 PID: 9345 Comm: test-ovs-ct-gen Not tainted 6.5.0_for_upstream_min_debug_2023_09_05_16_01 #1
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
RIP: 0010:0x0
Code: Unable to access opcode bytes at0xffffffffffffffd6.
RSP: 0018:ffff888178aaf758 EFLAGS: 00010246
Call Trace:
 <TASK>
 ? __die+0x20/0x60
 ? page_fault_oops+0x14c/0x3c0
 ? exc_page_fault+0x75/0x140
 ? asm_exc_page_fault+0x22/0x30
 notifier_call_chain+0x35/0xb0
 blocking_notifier_call_chain+0x3d/0x60
 mlx5_blocking_notifier_call_chain+0x22/0x30 [mlx5_core]
 mlx5_core_uplink_netdev_event_replay+0x3e/0x60 [mlx5_core]
 mlx5_mdev_netdev_track+0x53/0x60 [mlx5_ib]
 mlx5_ib_roce_init+0xc3/0x340 [mlx5_ib]
 __mlx5_ib_add+0x34/0xd0 [mlx5_ib]
 mlx5r_probe+0xe1/0x210 [mlx5_ib]
 ? auxiliary_match_id+0x6a/0x90
 auxiliary_bus_probe+0x38/0x80
 ? driver_sysfs_add+0x51/0x80
 really_probe+0xc9/0x3e0
 ? driver_probe_device+0x90/0x90
 __driver_probe_device+0x80/0x160
 driver_probe_device+0x1e/0x90
 __device_attach_driver+0x7d/0x100
 bus_for_each_drv+0x80/0xd0
 __device_attach+0xbc/0x1f0
 bus_probe_device+0x86/0xa0
 device_add+0x637/0x840
 __auxiliary_device_add+0x3b/0xa0
 add_adev+0xc9/0x140 [mlx5_core]
 mlx5_rescan_drivers_locked+0x22a/0x310 [mlx5_core]
 mlx5_register_device+0x53/0xa0 [mlx5_core]
 mlx5_init_one_devl_locked+0x5c4/0x9c0 [mlx5_core]
 mlx5_init_one+0x3b/0x60 [mlx5_core]
 probe_one+0x44c/0x730 [mlx5_core]
 local_pci_probe+0x3e/0x90
 pci_device_probe+0xbf/0x210
 ? kernfs_create_link+0x5d/0xa0
 ? sysfs_do_create_link_sd+0x60/0xc0
 really_probe+0xc9/0x3e0
 ? driver_probe_device+0x90/0x90
 __driver_probe_device+0x80/0x160
 driver_probe_device+0x1e/0x90
 __device_attach_driver+0x7d/0x100
 bus_for_each_drv+0x80/0xd0
 __device_attach+0xbc/0x1f0
 pci_bus_add_device+0x54/0x80
 pci_iov_add_virtfn+0x2e6/0x320
 sriov_enable+0x208/0x420
 mlx5_core_sriov_configure+0x9e/0x200 [mlx5_core]
 sriov_numvfs_store+0xae/0x1a0
 kernfs_fop_write_iter+0x10c/0x1a0
 vfs_write+0x291/0x3c0
 ksys_write+0x5f/0xe0
 do_syscall_64+0x3d/0x90
 entry_SYSCALL_64_after_hwframe+0x46/0xb0
 CR2: 0000000000000000
 ---[ end trace 0000000000000000  ]---

Fixes: 2c3b5beec46a ("net/mlx5e: More generic netdev management API")
Signed-off-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240509112951.590184-2-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'ipv6-sr-fix-errors-during-unregister'

Hangbin Liu says:

====================
ipv6: sr: fix errors during unregister

Fix some errors in seg6 unregister path, like missing unregister functions,
incorrect unregister order, etc.
====================

Link: https://lore.kernel.org/r/20240509131812.1662197-1-liuhangbin@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ipv6: sr: fix invalid unregister error path

The error path of seg6_init() is wrong in case CONFIG_IPV6_SEG6_LWTUNNEL
is not defined. In that case if seg6_hmac_init() fails, the
genl_unregister_family() isn't called.

This issue exist since commit 46738b1317e1 ("ipv6: sr: add option to control
lwtunnel support"), and commit 5559cea2d5aa ("ipv6: sr: fix possible
use-after-free and null-ptr-deref") replaced unregister_pernet_subsys()
with genl_unregister_family() in this error path.

Fixes: 46738b1317e1 ("ipv6: sr: add option to control lwtunnel support")
Reported-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Reviewed-by: Sabrina Dubroca <sd@queasysnail.net>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240509131812.1662197-4-liuhangbin@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ipv6: sr: fix incorrect unregister order

Commit 5559cea2d5aa ("ipv6: sr: fix possible use-after-free and
null-ptr-deref") changed the register order in seg6_init(). But the
unregister order in seg6_exit() is not updated.

Fixes: 5559cea2d5aa ("ipv6: sr: fix possible use-after-free and null-ptr-deref")
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Reviewed-by: Sabrina Dubroca <sd@queasysnail.net>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240509131812.1662197-3-liuhangbin@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ipv6: sr: add missing seg6_local_exit

Currently, we only call seg6_local_exit() in seg6_init() if
seg6_local_init() failed. But forgot to call it in seg6_exit().

Fixes: d1df6fd8a1d2 ("ipv6: sr: define core operations for seg6local lightweight tunnel")
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Reviewed-by: Sabrina Dubroca <sd@queasysnail.net>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240509131812.1662197-2-liuhangbin@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: openvswitch: fix overwriting ct original tuple for ICMPv6

OVS_PACKET_CMD_EXECUTE has 3 main attributes:
 - OVS_PACKET_ATTR_KEY - Packet metadata in a netlink format.
 - OVS_PACKET_ATTR_PACKET - Binary packet content.
 - OVS_PACKET_ATTR_ACTIONS - Actions to execute on the packet.

OVS_PACKET_ATTR_KEY is parsed first to populate sw_flow_key structure
with the metadata like conntrack state, input port, recirculation id,
etc.  Then the packet itself gets parsed to populate the rest of the
keys from the packet headers.

Whenever the packet parsing code starts parsing the ICMPv6 header, it
first zeroes out fields in the key corresponding to Neighbor Discovery
information even if it is not an ND packet.

It is an 'ipv6.nd' field.  However, the 'ipv6' is a union that shares
the space between 'nd' and 'ct_orig' that holds the original tuple
conntrack metadata parsed from the OVS_PACKET_ATTR_KEY.

ND packets should not normally have conntrack state, so it's fine to
share the space, but normal ICMPv6 Echo packets or maybe other types of
ICMPv6 can have the state attached and it should not be overwritten.

The issue results in all but the last 4 bytes of the destination
address being wiped from the original conntrack tuple leading to
incorrect packet matching and potentially executing wrong actions
in case this packet recirculates within the datapath or goes back
to userspace.

ND fields should not be accessed in non-ND packets, so not clearing
them should be fine.  Executing memset() only for actual ND packets to
avoid the issue.

Initializing the whole thing before parsing is needed because ND packet
may not contain all the options.

The issue only affects the OVS_PACKET_CMD_EXECUTE path and doesn't
affect packets entering OVS datapath from network interfaces, because
in this case CT metadata is populated from skb after the packet is
already parsed.

Fixes: 9dd7f8907c37 ("openvswitch: Add original direction conntrack tuple to sw_flow_key.")
Reported-by: Antonin Bas <antonin.bas@broadcom.com>
Closes: https://github.com/openvswitch/ovs-issues/issues/327
Signed-off-by: Ilya Maximets <i.maximets@ovn.org>
Acked-by: Aaron Conole <aconole@redhat.com>
Acked-by: Eelco Chaudron <echaudro@redhat.com>
Link: https://lore.kernel.org/r/20240509094228.1035477-1-i.maximets@ovn.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

af_unix: Fix data races in unix_release_sock/unix_stream_sendmsg

A data-race condition has been identified in af_unix. In one data path,
the write function unix_release_sock() atomically writes to
sk->sk_shutdown using WRITE_ONCE. However, on the reader side,
unix_stream_sendmsg() does not read it atomically. Consequently, this
issue is causing the following KCSAN splat to occur:

BUG: KCSAN: data-race in unix_release_sock / unix_stream_sendmsg

write (marked) to 0xffff88867256ddbb of 1 bytes by task 7270 on cpu 28:
unix_release_sock (net/unix/af_unix.c:640)
unix_release (net/unix/af_unix.c:1050)
sock_close (net/socket.c:659 net/socket.c:1421)
__fput (fs/file_table.c:422)
__fput_sync (fs/file_table.c:508)
__se_sys_close (fs/open.c:1559 fs/open.c:1541)
__x64_sys_close (fs/open.c:1541)
x64_sys_call (arch/x86/entry/syscall_64.c:33)
do_syscall_64 (arch/x86/entry/common.c:?)
entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:130)

read to 0xffff88867256ddbb of 1 bytes by task 989 on cpu 14:
unix_stream_sendmsg (net/unix/af_unix.c:2273)
__sock_sendmsg (net/socket.c:730 net/socket.c:745)
____sys_sendmsg (net/socket.c:2584)
__sys_sendmmsg (net/socket.c:2638 net/socket.c:2724)
__x64_sys_sendmmsg (net/socket.c:2753 net/socket.c:2750 net/socket.c:2750)
x64_sys_call (arch/x86/entry/syscall_64.c:33)
do_syscall_64 (arch/x86/entry/common.c:?)
entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:130)

value changed: 0x01 -> 0x03

The line numbers are related to commit dd5a440a31fa ("Linux 6.9-rc7").

Commit e1d09c2c2f57 ("af_unix: Fix data races around sk->sk_shutdown.")
addressed a comparable issue in the past regarding sk->sk_shutdown.
However, it overlooked resolving this particular data path.
This patch only offending unix_stream_sendmsg() function, since the
other reads seem to be protected by unix_state_lock() as discussed in
Link: https://lore.kernel.org/all/20240508173324.53565-1-kuniyu@amazon.com/
Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Signed-off-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20240509081459.2807828-1-leitao@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: ethernet: cortina: Locking fixes

This fixes a probably long standing problem in the Cortina
Gemini ethernet driver: there are some paths in the code
where the IRQ registers are written without taking the proper
locks.

Fixes: 4d5ae32f5e1e ("net: ethernet: Add a driver for Gemini gigabit ethernet")
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240509-gemini-ethernet-locking-v1-1-afd00a528b95@linaro.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ice: Fix package download algorithm

Previously, the driver assumed that all signature segments would contain
one or more buffers to download. In the future, there will be signature
segments that will contain no buffers to download.

Correct download flow to allow for signature segments that have zero
download buffers and skip the download in this case.

Fixes: 3cbdb0343022 ("ice: Add support for E830 DDP package segment")
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Signed-off-by: Dan Nowlin <dan.nowlin@intel.com>
Signed-off-by: Paul Greenwalt <paul.greenwalt@intel.com>
Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
Tested-by: Pucha Himasekhar Reddy <himasekharx.reddy.pucha@intel.com> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Link: https://lore.kernel.org/r/20240508171908.2760776-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: ethernet: mediatek: use ADMAv1 instead of ADMAv2.0 on MT7981 and MT7986

ADMAv2.0 is plagued by RX hangs which can't easily detected and happen upon
receival of a corrupted Ethernet frame.

Use ADMAv1 instead which is also still present and usable, and doesn't
suffer from that problem.

Fixes: 197c9e9b17b1 ("net: ethernet: mtk_eth_soc: introduce support for mt7986 chipset")
Co-developed-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Link: https://lore.kernel.org/r/57cef74bbd0c243366ad1ff4221e3f72f437ec80.1715164770.git.daniel@makrotopia.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: ethernet: mediatek: split tx and rx fields in mtk_soc_data struct

Split tx and rx fields in mtk_soc_data struct. This is a preliminary
patch to roll back to ADMAv1 for MT7986 and MT7981 SoC in order to fix a
hw hang if the device receives a corrupted packet when using ADMAv2.0.

Fixes: 197c9e9b17b1 ("net: ethernet: mtk_eth_soc: introduce support for mt7986 chipset")
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Link: https://lore.kernel.org/r/70a799b1f060ec2f57883e88ccb420ac0fb0abb5.1715164770.git.daniel@makrotopia.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

selftests: net: move amt to socat for better compatibility

The test seems to expect that nc will exit after the first
received message. This is not the case with Ncat 7.94.
There are multiple versions of nc out there, switch
to socat for better compatibility.

Tell socat to exit after 128 bytes and pad the message.

Since the test sets -e make sure we don't set exit code
(|| true) and print the pass / fail rather then silently
moving over the test and just setting non-zero exit code
with no output indicating what failed.

Fixes: c08e8baea78e ("selftests: add amt interface selftest script")
Acked-by: Paolo Abeni<pabeni@redhat.com>
Tested-by: Taehee Yoo <ap420073@gmail.com>
Link: https://lore.kernel.org/r/20240509161952.3940476-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

selftests: net: add missing config for amt.sh

Test needs IPv6 multicast. smcroute currently crashes when trying
to install a route in a kernel without IPv6 multicast.

Fixes: c08e8baea78e ("selftests: add amt interface selftest script")
Link: https://lore.kernel.org/r/20240509161919.3939966-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

eth: sungem: remove .ndo_poll_controller to avoid deadlocks

Erhard reports netpoll warnings from sungem:

  netpoll_send_skb_on_dev(): eth0 enabled interrupts in poll (gem_start_xmit+0x0/0x398)
  WARNING: CPU: 1 PID: 1 at net/core/netpoll.c:370 netpoll_send_skb+0x1fc/0x20c

gem_poll_controller() disables interrupts, which may sleep.
We can't sleep in netpoll, it has interrupts disabled completely.
Strangely, gem_poll_controller() doesn't even poll the completions,
and instead acts as if an interrupt has fired so it just schedules
NAPI and exits. None of this has been necessary for years, since
netpoll invokes NAPI directly.

Fixes: fe09bb619096 ("sungem: Spring cleaning and GRO support")
Reported-and-tested-by: Erhard Furtner <erhard_f@mailbox.org>
Link: https://lore.kernel.org/all/20240428125306.2c3080ef@legion
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20240508134504.3560956-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ptp: ocp: fix DPLL functions

In ptp_ocp driver pin actions assume sma_nr starts with 1, but for DPLL
subsystem callback 0-based index was used. Fix it providing proper index.

Fixes: 09eeb3aecc6c ("ptp_ocp: implement DPLL ops")
Signed-off-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Link: https://lore.kernel.org/r/20240508132111.11545-1-vadim.fedorenko@linux.dev
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge tag 'net-6.9-rc8' of git://git./linux/kernel/git/netdev/net

Pull networking fixes from Paolo Abeni:
 "Including fixes from bluetooth and IPsec.

  The bridge patch is actually a follow-up to a recent fix in the same
  area. We have a pending v6.8 AF_UNIX regression; it should be solved
  soon, but not in time for this PR.

  Current release - regressions:

   - eth: ks8851: Queue RX packets in IRQ handler instead of disabling
     BHs

   - net: bridge: fix corrupted ethernet header on multicast-to-unicast

  Current release - new code bugs:

   - xfrm: fix possible bad pointer derferencing in error path

  Previous releases - regressionis:

   - core: fix out-of-bounds access in ops_init

   - ipv6:
      - fix potential uninit-value access in __ip6_make_skb()
      - fib6_rules: avoid possible NULL dereference in fib6_rule_action()

   - tcp: use refcount_inc_not_zero() in tcp_twsk_unique().

   - rtnetlink: correct nested IFLA_VF_VLAN_LIST attribute validation

   - rxrpc: fix congestion control algorithm

   - bluetooth:
      - l2cap: fix slab-use-after-free in l2cap_connect()
      - msft: fix slab-use-after-free in msft_do_close()

   - eth: hns3: fix kernel crash when devlink reload during
     initialization

   - eth: dsa: mv88e6xxx: add phylink_get_caps for the mv88e6320/21
     family

  Previous releases - always broken:

   - xfrm: preserve vlan tags for transport mode software GRO

   - tcp: defer shutdown(SEND_SHUTDOWN) for TCP_SYN_RECV sockets

   - eth: hns3: keep using user config after hardware reset"

* tag 'net-6.9-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (47 commits)
  net: dsa: mv88e6xxx: read cmode on mv88e6320/21 serdes only ports
  net: dsa: mv88e6xxx: add phylink_get_caps for the mv88e6320/21 family
  net: hns3: fix kernel crash when devlink reload during initialization
  net: hns3: fix port vlan filter not disabled issue
  net: hns3: use appropriate barrier function after setting a bit value
  net: hns3: release PTP resources if pf initialization failed
  net: hns3: change type of numa_node_mask as nodemask_t
  net: hns3: direct return when receive a unknown mailbox message
  net: hns3: using user configure after hardware reset
  net/smc: fix neighbour and rtable leak in smc_ib_find_route()
  ipv6: prevent NULL dereference in ip6_output()
  hsr: Simplify code for announcing HSR nodes timer setup
  ipv6: fib6_rules: avoid possible NULL dereference in fib6_rule_action()
  dt-bindings: net: mediatek: remove wrongly added clocks and SerDes
  rxrpc: Only transmit one ACK per jumbo packet received
  rxrpc: Fix congestion control algorithm
  selftests: test_bridge_neigh_suppress.sh: Fix failures due to duplicate MAC
  ipv6: Fix potential uninit-value access in __ip6_make_skb()
  net: phy: marvell-88q2xxx: add support for Rev B1 and B2
  appletalk: Improve handling of broadcast packets
  ...

Merge tag 'for-linus' of git://git./linux/kernel/git/rmk/linux

Pull ARM fix from Russell King:

 - clear stale KASan stack poison when a CPU resumes

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rmk/linux:
  ARM: 9381/1: kasan: clear stale stack poison

Merge tag 'pull-fixes' of git://git./linux/kernel/git/viro/vfs

Pull dentry leak fix from Al Viro:
 "Dentry leak fix in the qibfs driver that I forgot to send a pull
  request for ;-/

  My apologies - it actually sat in vfs.git#fixes for more than two
  months..."

* tag 'pull-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  qibfs: fix dentry leak

net: dsa: mv88e6xxx: read cmode on mv88e6320/21 serdes only ports

On the mv88e6320 and 6321 switch family, port 0/1 are serdes only ports.
Modified the mv88e6352_get_port4_serdes_cmode function to pass a port
number since the register set of the 6352 is equal on the 6320/21.

Signed-off-by: Steffen Bätz <steffen@innosonix.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Fabio Estevam <festevam@gmail.com>
Link: https://lore.kernel.org/r/20240508072944.54880-3-steffen@innosonix.de
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

net: dsa: mv88e6xxx: add phylink_get_caps for the mv88e6320/21 family

As of commit de5c9bf40c45 ("net: phylink: require supported_interfaces to
be filled")
Marvell 88e6320/21 switches fail to be probed:

...
mv88e6085 30be0000.ethernet-1:00: phylink: error: empty supported_interfaces
error creating PHYLINK: -22
...

The problem stems from the use of mv88e6185_phylink_get_caps() to get
the device capabilities.
Since there are serdes only ports 0/1 included, create a new dedicated
phylink_get_caps for the 6320 and 6321 to properly support their
set of capabilities.

Fixes: de5c9bf40c45 ("net: phylink: require supported_interfaces to be filled")
Signed-off-by: Steffen Bätz <steffen@innosonix.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Fabio Estevam <festevam@gmail.com>
Link: https://lore.kernel.org/r/20240508072944.54880-2-steffen@innosonix.de
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

Merge branch 'there-are-some-bugfix-for-the-hns3-ethernet-driver'

Jijie Shao says:

====================
There are some bugfix for the HNS3 ethernet driver
====================

Link: https://lore.kernel.org/r/20240507134224.2646246-1-shaojijie@huawei.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

net: hns3: fix kernel crash when devlink reload during initialization

The devlink reload process will access the hardware resources,
but the register operation is done before the hardware is initialized.
So, processing the devlink reload during initialization may lead to kernel
crash.

This patch fixes this by registering the devlink after
hardware initialization.

Fixes: cd6242991d2e ("net: hns3: add support for registering devlink for VF")
Fixes: 93305b77ffcb ("net: hns3: fix kernel crash when devlink reload during pf initialization")
Signed-off-by: Yonglong Liu <liuyonglong@huawei.com>
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

net: hns3: fix port vlan filter not disabled issue

According to hardware limitation, for device support modify
VLAN filter state but not support bypass port VLAN filter,
it should always disable the port VLAN filter. but the driver
enables port VLAN filter when initializing, if there is no
VLAN(except VLAN 0) id added, the driver will disable it
in service task. In most time, it works fine. But there is
a time window before the service task shceduled and net device
being registered. So if user adds VLAN at this time, the driver
will not update the VLAN filter state,  and the port VLAN filter
remains enabled.

To fix the problem, if support modify VLAN filter state but not
support bypass port VLAN filter, set the port vlan filter to "off".

Fixes: 184cd221a863 ("net: hns3: disable port VLAN filter when support function level VLAN filter control")
Fixes: 2ba306627f59 ("net: hns3: add support for modify VLAN filter state")
Signed-off-by: Yonglong Liu <liuyonglong@huawei.com>
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

net: hns3: use appropriate barrier function after setting a bit value

There is a memory barrier in followed case. When set the port down,
hclgevf_set_timmer will set DOWN in state. Meanwhile, the service task has
different behaviour based on whether the state is DOWN. Thus, to make sure
service task see DOWN, use smp_mb__after_atomic after calling set_bit().

          CPU0                        CPU1
========================== ===================================
hclgevf_set_timer_task()    hclgevf_periodic_service_task()
  set_bit(DOWN,state)         test_bit(DOWN,state)

pf also has this issue.

Fixes: ff200099d271 ("net: hns3: remove unnecessary work in hclgevf_main")
Fixes: 1c6dfe6fc6f7 ("net: hns3: remove mailbox and reset work in hclge_main")
Signed-off-by: Peiyang Wang <wangpeiyang1@huawei.com>
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

net: hns3: release PTP resources if pf initialization failed

During the PF initialization process, hclge_update_port_info may return an
error code for some reason. At this point,  the ptp initialization has been
completed. To void memory leaks, the resources that are applied by ptp
should be released. Therefore, when hclge_update_port_info returns an error
code, hclge_ptp_uninit is called to release the corresponding resources.

Fixes: eaf83ae59e18 ("net: hns3: add querying fec ability from firmware")
Signed-off-by: Peiyang Wang <wangpeiyang1@huawei.com>
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Reviewed-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

net: hns3: change type of numa_node_mask as nodemask_t

It provides nodemask_t to describe the numa node mask in kernel. To
improve transportability, change the type of numa_node_mask as nodemask_t.

Fixes: 38caee9d3ee8 ("net: hns3: Add support of the HNAE3 framework")
Signed-off-by: Peiyang Wang <wangpeiyang1@huawei.com>
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

net: hns3: direct return when receive a unknown mailbox message

Currently, the driver didn't return when receive a unknown
mailbox message, and continue checking whether need to
generate a response. It's unnecessary and may be incorrect.

Fixes: bb5790b71bad ("net: hns3: refactor mailbox response scheme between PF and VF")
Signed-off-by: Jian Shen <shenjian15@huawei.com>
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

net: hns3: using user configure after hardware reset

When a reset occurring, it's supposed to recover user's configuration.
Currently, the port info(speed, duplex and autoneg) is stored in hclge_mac
and will be scheduled updated. Consider the case that reset was happened
consecutively. During the first reset, the port info is configured with
a temporary value cause the PHY is reset and looking for best link config.
Second reset start and use pervious configuration which is not the user's.
The specific process is as follows:

+------+               +----+                +----+
| USER |               | PF |                | HW |
+---+--+               +-+--+                +-+--+
    |  ethtool --reset   |                     |
    +------------------->|    reset command    |
    |  ethtool --reset   +-------------------->|
    +------------------->|                     +---+
    |                    +---+                 |   |
    |                    |   |reset currently  |   | HW RESET
    |                    |   |and wait to do   |   |
    |                    |<--+                 |   |
    |                    | send pervious cfg   |<--+
    |                    | (1000M FULL AN_ON)  |
    |                    +-------------------->|
    |                    | read cfg(time task) |
    |                    | (10M HALF AN_OFF)   +---+
    |                    |<--------------------+   | cfg take effect
    |                    |    reset command    |<--+
    |                    +-------------------->|
    |                    |                     +---+
    |                    | send pervious cfg   |   | HW RESET
    |                    | (10M HALF AN_OFF)   |<--+
    |                    +-------------------->|
    |                    | read cfg(time task) |
    |                    |  (10M HALF AN_OFF)  +---+
    |                    |<--------------------+   | cfg take effect
    |                    |                     |   |
    |                    | read cfg(time task) |<--+
    |                    |  (10M HALF AN_OFF)  |
    |                    |<--------------------+
    |                    |                     |
    v                    v                     v

To avoid aboved situation, this patch introduced req_speed, req_duplex,
req_autoneg to store user's configuration and it only be used after
hardware reset and to recover user's configuration

Fixes: f5f2b3e4dcc0 ("net: hns3: add support for imp-controlled PHYs")
Signed-off-by: Peiyang Wang <wangpeiyang1@huawei.com>
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

net/smc: fix neighbour and rtable leak in smc_ib_find_route()

In smc_ib_find_route(), the neighbour found by neigh_lookup() and rtable
resolved by ip_route_output_flow() are not released or put before return.
It may cause the refcount leak, so fix it.

Link: https://lore.kernel.org/r/20240506015439.108739-1-guwen@linux.alibaba.com
Fixes: e5c4744cfb59 ("net/smc: add SMC-Rv2 connection establishment")
Signed-off-by: Wen Gu <guwen@linux.alibaba.com>
Link: https://lore.kernel.org/r/20240507125331.2808-1-guwen@linux.alibaba.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

ipv6: prevent NULL dereference in ip6_output()

According to syzbot, there is a chance that ip6_dst_idev()
returns NULL in ip6_output(). Most places in IPv6 stack
deal with a NULL idev just fine, but not here.

syzbot reported:

general protection fault, probably for non-canonical address 0xdffffc00000000bc: 0000 [#1] PREEMPT SMP KASAN PTI
KASAN: null-ptr-deref in range [0x00000000000005e0-0x00000000000005e7]
CPU: 0 PID: 9775 Comm: syz-executor.4 Not tainted 6.9.0-rc5-syzkaller-00157-g6a30653b604a #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 03/27/2024
 RIP: 0010:ip6_output+0x231/0x3f0 net/ipv6/ip6_output.c:237
Code: 3c 1e 00 49 89 df 74 08 4c 89 ef e8 19 58 db f7 48 8b 44 24 20 49 89 45 00 49 89 c5 48 8d 9d e0 05 00 00 48 89 d8 48 c1 e8 03 <42> 0f b6 04 38 84 c0 4c 8b 74 24 28 0f 85 61 01 00 00 8b 1b 31 ff
RSP: 0018:ffffc9000927f0d8 EFLAGS: 00010202
RAX: 00000000000000bc RBX: 00000000000005e0 RCX: 0000000000040000
RDX: ffffc900131f9000 RSI: 0000000000004f47 RDI: 0000000000004f48
RBP: 0000000000000000 R08: ffffffff8a1f0b9a R09: 1ffffffff1f51fad
R10: dffffc0000000000 R11: fffffbfff1f51fae R12: ffff8880293ec8c0
R13: ffff88805d7fc000 R14: 1ffff1100527d91a R15: dffffc0000000000
FS:  00007f135c6856c0(0000) GS:ffff8880b9400000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000020000080 CR3: 0000000064096000 CR4: 00000000003506f0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 <TASK>
  NF_HOOK include/linux/netfilter.h:314 [inline]
  ip6_xmit+0xefe/0x17f0 net/ipv6/ip6_output.c:358
  sctp_v6_xmit+0x9f2/0x13f0 net/sctp/ipv6.c:248
  sctp_packet_transmit+0x26ad/0x2ca0 net/sctp/output.c:653
  sctp_packet_singleton+0x22c/0x320 net/sctp/outqueue.c:783
  sctp_outq_flush_ctrl net/sctp/outqueue.c:914 [inline]
  sctp_outq_flush+0x6d5/0x3e20 net/sctp/outqueue.c:1212
  sctp_side_effects net/sctp/sm_sideeffect.c:1198 [inline]
  sctp_do_sm+0x59cc/0x60c0 net/sctp/sm_sideeffect.c:1169
  sctp_primitive_ASSOCIATE+0x95/0xc0 net/sctp/primitive.c:73
  __sctp_connect+0x9cd/0xe30 net/sctp/socket.c:1234
  sctp_connect net/sctp/socket.c:4819 [inline]
  sctp_inet_connect+0x149/0x1f0 net/sctp/socket.c:4834
  __sys_connect_file net/socket.c:2048 [inline]
  __sys_connect+0x2df/0x310 net/socket.c:2065
  __do_sys_connect net/socket.c:2075 [inline]
  __se_sys_connect net/socket.c:2072 [inline]
  __x64_sys_connect+0x7a/0x90 net/socket.c:2072
  do_syscall_x64 arch/x86/entry/common.c:52 [inline]
  do_syscall_64+0xf5/0x240 arch/x86/entry/common.c:83
 entry_SYSCALL_64_after_hwframe+0x77/0x7f

Fixes: 778d80be5269 ("ipv6: Add disable_ipv6 sysctl to disable IPv6 operaion on specific interface.")
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Larysa Zaremba <larysa.zaremba@intel.com>
Link: https://lore.kernel.org/r/20240507161842.773961-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

hsr: Simplify code for announcing HSR nodes timer setup

Up till now the code to start HSR announce timer, which triggers sending
supervisory frames, was assuming that hsr_netdev_notify() would be called
at least twice for hsrX interface. This was required to have different
values for old and current values of network device's operstate.

This is problematic for a case where hsrX interface is already in the
operational state when hsr_netdev_notify() is called, so timer is not
configured to trigger and as a result the hsrX is not sending supervisory
frames to HSR ring.

This error has been discovered when hsr_ping.sh script was run. To be
more specific - for the hsr1 and hsr2 the hsr_netdev_notify() was
called at least twice with different IF_OPER_{LOWERDOWN|DOWN|UP} states
assigned in hsr_check_carrier_and_operstate(hsr). As a result there was
no issue with sending supervisory frames.
However, with hsr3, the notify function was called only once with
operstate set to IF_OPER_UP and timer responsible for triggering
supervisory frames was not fired.

The solution is to use netif_oper_up() and netif_running() helper
functions to assess if network hsrX device is up.
Only then, when the timer is not already pending, it is started.
Otherwise it is deactivated.

Fixes: f421436a591d ("net/hsr: Add support for the High-availability Seamless Redundancy protocol (HSRv0)")
Signed-off-by: Lukasz Majewski <lukma@denx.de>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240507111214.3519800-1-lukma@denx.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ipv6: fib6_rules: avoid possible NULL dereference in fib6_rule_action()

syzbot is able to trigger the following crash [1],
caused by unsafe ip6_dst_idev() use.

Indeed ip6_dst_idev() can return NULL, and must always be checked.

[1]

Oops: general protection fault, probably for non-canonical address 0xdffffc0000000000: 0000 [#1] PREEMPT SMP KASAN PTI
KASAN: null-ptr-deref in range [0x0000000000000000-0x0000000000000007]
CPU: 0 PID: 31648 Comm: syz-executor.0 Not tainted 6.9.0-rc4-next-20240417-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 03/27/2024
 RIP: 0010:__fib6_rule_action net/ipv6/fib6_rules.c:237 [inline]
 RIP: 0010:fib6_rule_action+0x241/0x7b0 net/ipv6/fib6_rules.c:267
Code: 02 00 00 49 8d 9f d8 00 00 00 48 89 d8 48 c1 e8 03 42 80 3c 20 00 74 08 48 89 df e8 f9 32 bf f7 48 8b 1b 48 89 d8 48 c1 e8 03 <42> 80 3c 20 00 74 08 48 89 df e8 e0 32 bf f7 4c 8b 03 48 89 ef 4c
RSP: 0018:ffffc9000fc1f2f0 EFLAGS: 00010246
RAX: 0000000000000000 RBX: 0000000000000000 RCX: 1a772f98c8186700
RDX: 0000000000000003 RSI: ffffffff8bcac4e0 RDI: ffffffff8c1f9760
RBP: ffff8880673fb980 R08: ffffffff8fac15ef R09: 1ffffffff1f582bd
R10: dffffc0000000000 R11: fffffbfff1f582be R12: dffffc0000000000
R13: 0000000000000080 R14: ffff888076509000 R15: ffff88807a029a00
FS:  00007f55e82ca6c0(0000) GS:ffff8880b9400000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000001b31d23000 CR3: 0000000022b66000 CR4: 00000000003506f0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 <TASK>
  fib_rules_lookup+0x62c/0xdb0 net/core/fib_rules.c:317
  fib6_rule_lookup+0x1fd/0x790 net/ipv6/fib6_rules.c:108
  ip6_route_output_flags_noref net/ipv6/route.c:2637 [inline]
  ip6_route_output_flags+0x38e/0x610 net/ipv6/route.c:2649
  ip6_route_output include/net/ip6_route.h:93 [inline]
  ip6_dst_lookup_tail+0x189/0x11a0 net/ipv6/ip6_output.c:1120
  ip6_dst_lookup_flow+0xb9/0x180 net/ipv6/ip6_output.c:1250
  sctp_v6_get_dst+0x792/0x1e20 net/sctp/ipv6.c:326
  sctp_transport_route+0x12c/0x2e0 net/sctp/transport.c:455
  sctp_assoc_add_peer+0x614/0x15c0 net/sctp/associola.c:662
  sctp_connect_new_asoc+0x31d/0x6c0 net/sctp/socket.c:1099
  __sctp_connect+0x66d/0xe30 net/sctp/socket.c:1197
  sctp_connect net/sctp/socket.c:4819 [inline]
  sctp_inet_connect+0x149/0x1f0 net/sctp/socket.c:4834
  __sys_connect_file net/socket.c:2048 [inline]
  __sys_connect+0x2df/0x310 net/socket.c:2065
  __do_sys_connect net/socket.c:2075 [inline]
  __se_sys_connect net/socket.c:2072 [inline]
  __x64_sys_connect+0x7a/0x90 net/socket.c:2072
  do_syscall_x64 arch/x86/entry/common.c:52 [inline]
  do_syscall_64+0xf5/0x240 arch/x86/entry/common.c:83
 entry_SYSCALL_64_after_hwframe+0x77/0x7f

Fixes: 5e5f3f0f8013 ("[IPV6] ADDRCONF: Convert ipv6_get_saddr() to ipv6_dev_get_saddr().")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240507163145.835254-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

dt-bindings: net: mediatek: remove wrongly added clocks and SerDes

Several clocks as well as both sgmiisys phandles were added by mistake
to the Ethernet bindings for MT7988. Also, the total number of clocks
didn't match with the actual number of items listed.

This happened because the vendor driver which served as a reference uses
a high number of syscon phandles to access various parts of the SoC
which wasn't acceptable upstream. Hence several parts which have never
previously been supported (such SerDes PHY and USXGMII PCS) are going to
be implemented by separate drivers. As a result the device tree will
look much more sane.

Quickly align the bindings with the upcoming reality of the drivers
actually adding support for the remaining Ethernet-related features of
the MT7988 SoC.

Fixes: c94a9aabec36 ("dt-bindings: net: mediatek,net: add mt7988-eth binding")
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Acked-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Link: https://lore.kernel.org/r/1569290b21cc787a424469ed74456a7e976b102d.1715084326.git.daniel@makrotopia.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge tag '6.9-rc7-ksmbd-fixes' of git://git.samba.org/ksmbd

Pull smb server fixes from Steve French:
 "Five ksmbd server fixes, all also for stable

   - Three fixes related to SMB3 leases (fixes two xfstests, and a
     locking issue)

   - Unitialized variable fix

   - Socket creation fix when bindv6only is set"

* tag '6.9-rc7-ksmbd-fixes' of git://git.samba.org/ksmbd:
  ksmbd: do not grant v2 lease if parent lease key and epoch are not set
  ksmbd: use rwsem instead of rwlock for lease break
  ksmbd: avoid to send duplicate lease break notifications
  ksmbd: off ipv6only for both ipv4/ipv6 binding
  ksmbd: fix uninitialized symbol 'share' in smb2_tree_connect()

Merge tag 'fuse-fixes-6.9-final' of git://git./linux/kernel/git/mszeredi/fuse

Pull fuse fixes from Miklos Szeredi:
 "Two one-liner fixes for issues introduced in -rc1"

* tag 'fuse-fixes-6.9-final' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/fuse:
  virtiofs: include a newline in sysfs tag
  fuse: verify zero padding in fuse_backing_map

Merge tag 'exfat-for-6.9-rc8' of git://git./linux/kernel/git/linkinjeon/exfat

Pull exfat fixes from Namjae Jeon:

 - Fix xfstests generic/013 test failure with dirsync mount option

 - Initialize the reserved fields of deleted file and stream extension
   dentries to zero

* tag 'exfat-for-6.9-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/linkinjeon/exfat:
  exfat: zero the reserved fields of file and stream extension dentries
  exfat: fix timing of synchronizing bitmap and inode

Merge tag 'bcachefs-2024-05-07.2' of https://evilpiepirate.org/git/bcachefs

Pull bcachefs fixes from Kent Overstreet:

 - Various syzbot fixes; mainly small gaps in validation

 - Fix an integer overflow in fiemap() which was preventing filefrag
   from returning the full list of extents

 - Fix a refcounting bug on the device refcount, turned up by new
   assertions in the development branch

 - Fix a device removal/readd bug; write_super() was repeatedly dropping
   and retaking bch_dev->io_ref references

* tag 'bcachefs-2024-05-07.2' of https://evilpiepirate.org/git/bcachefs:
  bcachefs: Add missing sched_annotate_sleep() in bch2_journal_flush_seq_async()
  bcachefs: Fix race in bch2_write_super()
  bcachefs: BCH_SB_LAYOUT_SIZE_BITS_MAX
  bcachefs: Add missing skcipher_request_set_callback() call
  bcachefs: Fix snapshot_t() usage in bch2_fs_quota_read_inode()
  bcachefs: Fix shift-by-64 in bformat_needs_redo()
  bcachefs: Guard against unknown k.k->type in __bkey_invalid()
  bcachefs: Add missing validation for superblock section clean
  bcachefs: Fix assert in bch2_alloc_v4_invalid()
  bcachefs: fix overflow in fiemap
  bcachefs: Add a better limit for maximum number of buckets
  bcachefs: Fix lifetime issue in device iterator helpers
  bcachefs: Fix bch2_dev_lookup() refcounting
  bcachefs: Initialize bch_write_op->failed in inline data path
  bcachefs: Fix refcount put in sb_field_resize error path
  bcachefs: Inodes need extra padding for varint_decode_fast()
  bcachefs: Fix early error path in bch2_fs_btree_key_cache_exit()
  bcachefs: bucket_pos_to_bp_noerror()
  bcachefs: don't free error pointers
  bcachefs: Fix a scheduler splat in __bch2_next_write_buffer_flush_journal_buf()

Merge tag 'soc-fixes-6.9-3' of git://git./linux/kernel/git/soc/soc

Pull ARM SoC fixes from Arnd Bergmann:
 "These are a couple of last minute fixes that came in over the previous
  week, addressing:

   - A pin configuration bug on a qualcomm board that caused issues with
     ethernet and mmc

   - Two minor code fixes for misleading console output in the microchip
     firmware driver

   - A build warning in the sifive cache driver"

* tag 'soc-fixes-6.9-3' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc:
  firmware: microchip: clarify that sizes and addresses are in hex
  firmware: microchip: don't unconditionally print validation success
  arm64: dts: qcom: sa8155p-adp: fix SDHC2 CD pin configuration
  cache: sifive_ccache: Silence unused variable warning

Merge tag 'pci-v6.9-fixes-2' of git://git./linux/kernel/git/pci/pci

Pull pci fixes from Bjorn Helgaas:

 - Update kernel-parameters doc to describe "pcie_aspm=off" more
   accurately (Bjorn Helgaas)

 - Restore the parent's (not the child's) ASPM state to the parent
   during resume, which fixes a reboot during resume (Kai-Heng Feng)

* tag 'pci-v6.9-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/pci/pci:
  PCI/ASPM: Restore parent state to parent, child state to child
  PCI/ASPM: Clarify that pcie_aspm=off means leave ASPM untouched

Merge branch 'rxrpc-miscellaneous-fixes'

David Howells says:

====================
rxrpc: Miscellaneous fixes (part)

Here some miscellaneous fixes for AF_RXRPC:

 (1) Fix the congestion control algorithm to start cwnd at 4 and to not cut
     ssthresh when the peer cuts its rwind size.

 (2) Only transmit a single ACK for all the DATA packets glued together
     into a jumbo packet to reduce the number of ACKs being generated.
====================

Link: https://lore.kernel.org/r/20240503150749.1001323-1-dhowells@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

rxrpc: Only transmit one ACK per jumbo packet received

Only generate one ACK packet for all the subpackets in a jumbo packet.  If
we would like to generate more than one ACK, we prioritise them base on
their reason code, in the order, highest first:

   OutOfSeq > NoSpace > ExceedsWin > Duplicate > Requested > Delay > Idle

For the first four, we reference the lowest offending subpacket; for the
last three, the highest.

This reduces the number of ACKs we end up transmitting to one per UDP
packet transmitted to reduce network loading and packet parsing.

Fixes: 5d7edbc9231e ("rxrpc: Get rid of the Rx ring")
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Reviewed-by: Jeffrey Altman <jaltman@auristor.com <mailto:jaltman@auristor.com>>
Link: https://lore.kernel.org/r/20240503150749.1001323-3-dhowells@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

rxrpc: Fix congestion control algorithm

Make the following fixes to the congestion control algorithm:

 (1) Don't vary the cwnd starting value by the size of RXRPC_TX_SMSS since
     that's currently held constant - set to the size of a jumbo subpacket
     payload so that we can create jumbo packets on the fly.  The current
     code invariably picks 3 as the starting value.

     Further, the starting cwnd needs to be an even number because we ack
     every other packet, so set it to 4.

 (2) Don't cut ssthresh when we see an ACK come from the peer with a
     receive window (rwind) less than ssthresh.  ssthresh keeps track of
     characteristics of the connection whereas rwind may be reduced by the
     peer for any reason - and may be reduced to 0.

Fixes: 1fc4fa2ac93d ("rxrpc: Fix congestion management")
Fixes: 0851115090a3 ("rxrpc: Reduce ssthresh to peer's receive window")
Signed-off-by: David Howells <dhowells@redhat.com>
Suggested-by: Simon Wilkinson <sxw@auristor.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Reviewed-by: Jeffrey Altman <jaltman@auristor.com <mailto:jaltman@auristor.com>>
Link: https://lore.kernel.org/r/20240503150749.1001323-2-dhowells@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

selftests: test_bridge_neigh_suppress.sh: Fix failures due to duplicate MAC

When creating the topology for the test, three veth pairs are created in
the initial network namespace before being moved to one of the network
namespaces created by the test.

On systems where systemd-udev uses MACAddressPolicy=persistent (default
since systemd version 242), this will result in some net devices having
the same MAC address since they were created with the same name in the
initial network namespace. In turn, this leads to arping / ndisc6
failing since packets are dropped by the bridge's loopback filter.

Fix by creating each net device in the correct network namespace instead
of moving it there from the initial network namespace.

Reported-by: Jakub Kicinski <kuba@kernel.org>
Closes: https://lore.kernel.org/netdev/20240426074015.251854d4@kernel.org/
Fixes: 7648ac72dcd7 ("selftests: net: Add bridge neighbor suppression test")
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Link: https://lore.kernel.org/r/20240507113033.1732534-1-idosch@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ipv6: Fix potential uninit-value access in __ip6_make_skb()

As it was done in commit fc1092f51567 ("ipv4: Fix uninit-value access in
__ip_make_skb()") for IPv4, check FLOWI_FLAG_KNOWN_NH on fl6->flowi6_flags
instead of testing HDRINCL on the socket to avoid a race condition which
causes uninit-value access.

Fixes: ea30388baebc ("ipv6: Fix an uninit variable access bug in __ip6_make_skb()")
Signed-off-by: Shigeru Yoshida <syoshida@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: phy: marvell-88q2xxx: add support for Rev B1 and B2

Different revisions of the Marvell 88q2xxx phy needs different init
sequences.

Add init sequence for Rev B1 and Rev B2. Rev B2 init sequence skips one
register write.

Tested-by: Dimitri Fedrau <dima.fedrau@gmail.com>
Signed-off-by: Gregor Herburger <gregor.herburger@ew.tq-group.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

appletalk: Improve handling of broadcast packets

When a broadcast AppleTalk packet is received, prefer queuing it on the
socket whose address matches the address of the interface that received
the packet (and is listening on the correct port). Userspace
applications that handle such packets will usually send a response on
the same socket that received the packet; this fix allows the response
to be sent on the correct interface.

If a socket matching the interface's address is not found, an arbitrary
socket listening on the correct port will be used, if any. This matches
the implementation's previous behavior.

Fixes atalkd's responses to network information requests when multiple
network interfaces are configured to use AppleTalk.

Link: https://lore.kernel.org/netdev/20200722113752.1218-2-vincent.ldev@duvert.net/
Link: https://gist.github.com/VinDuv/4db433b6dce39d51a5b7847ee749b2a4
Signed-off-by: Vincent Duvert <vincent.ldev@duvert.net>
Signed-off-by: Doug Brown <doug@schmorgal.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: bridge: fix corrupted ethernet header on multicast-to-unicast

The change from skb_copy to pskb_copy unfortunately changed the data
copying to omit the ethernet header, since it was pulled before reaching
this point. Fix this by calling __skb_push/pull around pskb_copy.

Fixes: 59c878cbcdd8 ("net: bridge: fix multicast-to-unicast with fraglist GSO")
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

virtiofs: include a newline in sysfs tag

The internal tag string doesn't contain a newline. Append one when
emitting the tag via sysfs.

[Stefan] Orthogonal to the newline issue, sysfs_emit(buf, "%s", fs->tag) is
needed to prevent format string injection.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Fixes: a8f62f50b4e4 ("virtiofs: export filesystem tags through sysfs")
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

mptcp: only allow set existing scheduler for net.mptcp.scheduler

The current behavior is to accept any strings as inputs, this results in
an inconsistent result where an unexisting scheduler can be set:

  # sysctl -w net.mptcp.scheduler=notdefault
  net.mptcp.scheduler = notdefault

This patch changes this behavior by checking for existing scheduler
before accepting the input.

Fixes: e3b2870b6d22 ("mptcp: add a new sysctl scheduler")
Cc: stable@vger.kernel.org
Signed-off-by: Gregory Detal <gregory.detal@gmail.com>
Reviewed-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Tested-by: Geliang Tang <geliang@kernel.org>
Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://lore.kernel.org/r/20240506-upstream-net-20240506-mptcp-sched-exist-v1-1-2ed1529e521e@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

nfc: nci: Fix kcov check in nci_rx_work()

Commit 7e8cdc97148c ("nfc: Add KCOV annotations") added
kcov_remote_start_common()/kcov_remote_stop() pair into nci_rx_work(),
with an assumption that kcov_remote_stop() is called upon continue of
the for loop. But commit d24b03535e5e ("nfc: nci: Fix uninit-value in
nci_dev_up and nci_ntf_packet") forgot to call kcov_remote_stop() before
break of the for loop.

Reported-by: syzbot <syzbot+0438378d6f157baae1a2@syzkaller.appspotmail.com>
Closes: https://syzkaller.appspot.com/bug?extid=0438378d6f157baae1a2
Fixes: d24b03535e5e ("nfc: nci: Fix uninit-value in nci_dev_up and nci_ntf_packet")
Suggested-by: Andrey Konovalov <andreyknvl@gmail.com>
Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Link: https://lore.kernel.org/r/6d10f829-5a0c-405a-b39a-d7266f3a1a0b@I-love.SAKURA.ne.jp
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

bcachefs: Add missing sched_annotate_sleep() in bch2_journal_flush_seq_async()

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Fix race in bch2_write_super()

bch2_write_super() was looping over online devices multiple times -
dropping and retaking io_ref each time.

This meant it could race with device removal; it could increment the
sequence number on a device but fail to write it - and then if the
device was re-added, it would get confused the next time around thinking
a superblock write was silently dropped.

Fix this by taking io_ref once, and stashing pointers to online devices
in a darray.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Merge tag 'qcom-arm64-fixes-for-6.9-2' of https://git./linux/kernel/git/qcom/linux into arm/fixes

One more Qualcomm Arm64 DeviceTree fix for v6.9

On ths SA8155P automotive platform, the wrong gpio controller is defined
for the SD-card detect pin, which depending on probe ordering of things
cause ethernet to be broken. The card detect pin reference is corrected
to solve this problem.

* tag 'qcom-arm64-fixes-for-6.9-2' of https://git.kernel.org/pub/scm/linux/kernel/git/qcom/linux:
  arm64: dts: qcom: sa8155p-adp: fix SDHC2 CD pin configuration

Link: https://lore.kernel.org/r/20240427153817.1430382-1-andersson@kernel.org
Signed-off-by: Arnd Bergmann <arnd@arndb.de>

netlink: specs: Add missing bridge linkinfo attrs

Attributes for FDB learned entries were added to the if_link netlink api
for bridge linkinfo but are missing from the rt_link.yaml spec. Add the
missing attributes to the spec.

Fixes: ddd1ad68826d ("net: bridge: Add netlink knobs for number / max learned FDB entries")
Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20240503164304.87427-1-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

phonet: fix rtm_phonet_notify() skb allocation

fill_route() stores three components in the skb:

- struct rtmsg
- RTA_DST (u8)
- RTA_OIF (u32)

Therefore, rtm_phonet_notify() should use

NLMSG_ALIGN(sizeof(struct rtmsg)) +
nla_total_size(1) +
nla_total_size(4)

Fixes: f062f41d0657 ("Phonet: routing table Netlink interface")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Rémi Denis-Courmont <courmisch@gmail.com>
Link: https://lore.kernel.org/r/20240502161700.1804476-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge tag 'for-6.9-rc7-tag' of git://git./linux/kernel/git/kdave/linux

Pull btrfs fixes from David Sterba:
 "Two more fixes, both have some visible effects on user space:

   - add check if quotas are enabled when passing qgroup inheritance
     info, this affects snapper that could fail to create a snapshot

   - do check for leaf/node flag WRITTEN earlier so that nodes are
     completely validated before access, this used to be done by
     integrity checker but it's been removed and left an unhandled case"

* tag 'for-6.9-rc7-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  btrfs: make sure that WRITTEN is set on all metadata blocks
  btrfs: qgroup: do not check qgroup inherit if qgroup is disabled

Reapply "drm/qxl: simplify qxl_fence_wait"

This reverts commit 07ed11afb68d94eadd4ffc082b97c2331307c5ea.

Stephen Rostedt reports:
 "I went to run my tests on my VMs and the tests hung on boot up.
  Unfortunately, the most I ever got out was:

  [   93.607888] Testing event system initcall: OK
  [   93.667730] Running tests on all trace events:
  [   93.669757] Testing all events: OK
  [   95.631064] ------------[ cut here ]------------
  Timed out after 60 seconds"

and further debugging points to a possible circular locking dependency
between the console_owner locking and the worker pool locking.

Reverting the commit allows Steve's VM to boot to completion again.

[ This may obviously result in the "[TTM] Buffer eviction failed"
  messages again, which was the reason for that original revert. But at
  this point this seems preferable to a non-booting system... ]

Reported-and-bisected-by: Steven Rostedt <rostedt@goodmis.org>
Link: https://lore.kernel.org/all/20240502081641.457aa25f@gandalf.local.home/
Acked-by: Maxime Ripard <mripard@kernel.org>
Cc: Alex Constantino <dreaming.about.electric.sheep@gmail.com>
Cc: Maxime Ripard <mripard@kernel.org>
Cc: Timo Lindfors <timo.lindfors@iki.fi>
Cc: Dave Airlie <airlied@redhat.com>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Thomas Zimmermann <tzimmermann@suse.de>
Cc: Daniel Vetter <daniel@ffwll.ch>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

PCI/ASPM: Restore parent state to parent, child state to child

There's a typo that makes parent device uses child LNKCTL value and vice
versa. This causes Micron NVMe to trigger a reboot upon system resume.

Correct the typo to fix the issue.

Fixes: 64dbb2d70744 ("PCI/ASPM: Disable L1 before configuring L1 Substates")
Link: https://lore.kernel.org/r/20240506051602.1990743-1-kai.heng.feng@canonical.com
Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
[bhelgaas: update subject]
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>

Merge tag 'slab-for-6.9-rc7-fixes' of git://git./linux/kernel/git/vbabka/slab

Pull slab fixes from Vlastimil Babka:

 - Fix for cleanup infrastructure (Dan Carpenter)

   This makes the __free(kfree) cleanup hooks not crash on error
   pointers.

 - SLUB fix for freepointer checking (Nicolas Bouchinet)

   This fixes a recently introduced bug that manifests when
   init_on_free, CONFIG_SLAB_FREELIST_HARDENED and consistency checks
   (slub_debug=F) are all enabled, and results in false-positive
   freepointer corrupt reports for caches that store freepointer outside
   of the object area.

* tag 'slab-for-6.9-rc7-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/vbabka/slab:
  mm/slab: make __free(kfree) accept error pointers
  mm/slub: avoid zeroing outside-object freepointer for single free

Merge tag 'auxdisplay-v6.10-1' of git://git./linux/kernel/git/andy/linux-auxdisplay

Pull auxdisplay fixes from Andy Shevchenko:

 - A couple of non-critical build fixes to Character LCD library

 - Miscellaneous fixes here and there

* tag 'auxdisplay-v6.10-1' of git://git.kernel.org/pub/scm/linux/kernel/git/andy/linux-auxdisplay:
  auxdisplay: charlcd: Don't rebuild when CONFIG_PANEL_BOOT_MESSAGE=y
  auxdisplay: charlcd: Add missing MODULE_DESCRIPTION()
  auxdisplay: seg-led-gpio: Convert to platform remove callback returning void
  auxdisplay: linedisp: Group display drivers together

bcachefs: BCH_SB_LAYOUT_SIZE_BITS_MAX

Define a constant for the max superblock size, to avoid a too-large
shift.

Reported-by: syzbot+a8b0fb419355c91dda7f@syzkaller.appspotmail.com
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Add missing skcipher_request_set_callback() call

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Fix snapshot_t() usage in bch2_fs_quota_read_inode()

bch2_fs_quota_read_inode() wasn't entirely updated to the
bch2_snapshot_tree() helper, which takes rcu lock.

Reported-by: syzbot+a3a9a61224ed3b7f0010@syzkaller.appspotmail.com
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Fix shift-by-64 in bformat_needs_redo()

Ancient versions of bcachefs produced packed formats that could
represent keys that our in memory format cannot represent;
bformat_needs_redo() has some tricky shifts to check for this sort of
overflow.

Reported-by: syzbot+594427aebfefeebe91c6@syzkaller.appspotmail.com
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Guard against unknown k.k->type in __bkey_invalid()

For forwards compatibility we have to allow unknown key types, and only
run the checks that make sense against them.

Fix a missing guard on k.k->type being known.

Reported-by: syzbot+ae4dc916da3ce51f284f@syzkaller.appspotmail.com
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Add missing validation for superblock section clean

We were forgetting to check for jset entries that overrun the end of the
section - both in validate and to_text(); to_text() needs to be safe for
types that fail to validate.

Reported-by: syzbot+c48865e11e7e893ec4ab@syzkaller.appspotmail.com
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Fix assert in bch2_alloc_v4_invalid()

Reported-by: syzbot+10827fa6b176e1acf1d0@syzkaller.appspotmail.com
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: fix overflow in fiemap

filefrag (and potentially other utilities that call fiemap) sometimes
pass ULONG_MAX as the length.  fiemap_prep clamps excessively large
lengths - but the calculation of end can overflow if it occurs before
calling fiemap_prep.  When this happens, filefrag assumes it has read to
the end and exits.

Signed-off-by: Reed Riley <reed@riley.engineer>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Add a better limit for maximum number of buckets

The bucket_gens array is a single array allocation (one byte per
bucket), and kernel allocations are still limited to INT_MAX.

Check this limit to avoid failing the bucket_gens array allocation.

Reported-by: syzbot+b29f436493184ea42e2b@syzkaller.appspotmail.com
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Fix lifetime issue in device iterator helpers

bch2_get_next_dev() and bch2_get_next_online_dev() iterate over devices,
dropping and taking refs as they go; we can't access the previous device
(for ca->dev_idx) after we've dropped our ref to it, unless we take
rcu_read_lock() first.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Fix bch2_dev_lookup() refcounting

bch2_dev_lookup() is supposed to take a ref on the device it returns, but
for_each_member_device() takes refs as it iterates,
for_each_member_device_rcu() does not.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Initialize bch_write_op->failed in inline data path

Normally this is initialized in __bch2_write(), which is executed in a
loop, but the inline data path skips this.

Reported-by: syzbot+fd3ccb331eb21f05d13b@syzkaller.appspotmail.com
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Fix refcount put in sb_field_resize error path

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Inodes need extra padding for varint_decode_fast()

Reported-by: syzbot+66b9b74f6520068596a9@syzkaller.appspotmail.com
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Fix early error path in bch2_fs_btree_key_cache_exit()

Reported-by: syzbot+a35cdb62ec34d44fb062@syzkaller.appspotmail.com
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: bucket_pos_to_bp_noerror()

We don't want the assert when we're checking if the backpointer is
valid.

Reported-by: syzbot+bf7215c0525098e7747a@syzkaller.appspotmail.com
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: don't free error pointers

Reported-by: syzbot+3333603f569fc2ef258c@syzkaller.appspotmail.com
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Fix a scheduler splat in __bch2_next_write_buffer_flush_journal_buf()

We're using mutex_lock() inside a wait_event() conditional -
prepare_to_wait() has already flipped task state, so potentially
blocking ops need annotation.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

net: fix out-of-bounds access in ops_init

net_alloc_generic is called by net_alloc, which is called without any
locking. It reads max_gen_ptrs, which is changed under pernet_ops_rwsem. It
is read twice, first to allocate an array, then to set s.len, which is
later used to limit the bounds of the array access.

It is possible that the array is allocated and another thread is
registering a new pernet ops, increments max_gen_ptrs, which is then used
to set s.len with a larger than allocated length for the variable array.

Fix it by reading max_gen_ptrs only once in net_alloc_generic. If
max_gen_ptrs is later incremented, it will be caught in net_assign_generic.

Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@igalia.com>
Fixes: 073862ba5d24 ("netns: fix net_alloc_generic()")
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20240502132006.3430840-1-cascardo@igalia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

Linux 6.9-rc7

epoll: be better about file lifetimes

epoll can call out to vfs_poll() with a file pointer that may race with
the last 'fput()'. That would make f_count go down to zero, and while
the ep->mtx locking means that the resulting file pointer tear-down will
be blocked until the poll returns, it means that f_count is already
dead, and any use of it won't actually get a reference to the file any
more: it's dead regardless.

Make sure we have a valid ref on the file pointer before we call down to
vfs_poll() from the epoll routines.

Link: https://lore.kernel.org/lkml/0000000000002d631f0615918f1e@google.com/
Reported-by: syzbot+045b454ab35fd82a35fb@syzkaller.appspotmail.com
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>