mbox series

[net-next,0/5] net: remove RTNL from fib_seq_sum()

Message ID 20241009184405.3752829-1-edumazet@google.com (mailing list archive)
Headers show
Series net: remove RTNL from fib_seq_sum() | expand

Message

Eric Dumazet Oct. 9, 2024, 6:44 p.m. UTC
This series is inspired by a syzbot report showing
rtnl contention and one thread blocked in:

7 locks held by syz-executor/10835:
  #0: ffff888033390420 (sb_writers#8){.+.+}-{0:0}, at: file_start_write include/linux/fs.h:2931 [inline] 
  #0: ffff888033390420 (sb_writers#8){.+.+}-{0:0}, at: vfs_write+0x224/0xc90 fs/read_write.c:679
  #1: ffff88806df6bc88 (&of->mutex){+.+.}-{3:3}, at: kernfs_fop_write_iter+0x1ea/0x500 fs/kernfs/file.c:325
  #2: ffff888026fcf3c8 (kn->active#50){.+.+}-{0:0}, at: kernfs_fop_write_iter+0x20e/0x500 fs/kernfs/file.c:326
  #3: ffffffff8f56f848 (nsim_bus_dev_list_lock){+.+.}-{3:3}, at: new_device_store+0x1b4/0x890 drivers/net/netdevsim/bus.c:166
  #4: ffff88805e0140e8 (&dev->mutex){....}-{3:3}, at: device_lock include/linux/device.h:1014 [inline] 
  #4: ffff88805e0140e8 (&dev->mutex){....}-{3:3}, at: __device_attach+0x8e/0x520 drivers/base/dd.c:1005
  #5: ffff88805c5fb250 (&devlink->lock_key#55){+.+.}-{3:3}, at: nsim_drv_probe+0xcb/0xb80 drivers/net/netdevsim/dev.c:1534
  #6: ffffffff8fcd1748 (rtnl_mutex){+.+.}-{3:3}, at: fib_seq_sum+0x31/0x290 net/core/fib_notifier.c:46

This is not a bug fix, unless I am mistaken, thus targeting net-next.


Eric Dumazet (5):
  fib: rules: use READ_ONCE()/WRITE_ONCE() on ops->fib_rules_seq
  ipv4: use READ_ONCE()/WRITE_ONCE() on net->ipv4.fib_seq
  ipv6: use READ_ONCE()/WRITE_ONCE() on fib6_table->fib_seq
  ipmr: use READ_ONCE() to read net->ipv[46].ipmr_seq
  net: do not acquire rtnl in fib_seq_sum()

 include/net/fib_notifier.h |  2 +-
 include/net/fib_rules.h    |  2 +-
 include/net/ip6_fib.h      |  8 ++++----
 include/net/ip_fib.h       |  4 ++--
 include/net/netns/ipv4.h   |  2 +-
 net/core/fib_notifier.c    |  2 --
 net/core/fib_rules.c       | 14 ++++++++------
 net/ipv4/fib_notifier.c    | 10 +++++-----
 net/ipv4/fib_rules.c       |  2 +-
 net/ipv4/ipmr.c            | 10 ++++------
 net/ipv6/fib6_notifier.c   |  2 +-
 net/ipv6/fib6_rules.c      |  2 +-
 net/ipv6/ip6_fib.c         | 14 +++++++-------
 net/ipv6/ip6mr.c           | 10 ++++------
 14 files changed, 40 insertions(+), 44 deletions(-)

Comments

David Ahern Oct. 9, 2024, 9:16 p.m. UTC | #1
On 10/9/24 12:44 PM, Eric Dumazet wrote:
> This series is inspired by a syzbot report showing
> rtnl contention and one thread blocked in:
> 
> 7 locks held by syz-executor/10835:
>   #0: ffff888033390420 (sb_writers#8){.+.+}-{0:0}, at: file_start_write include/linux/fs.h:2931 [inline] 
>   #0: ffff888033390420 (sb_writers#8){.+.+}-{0:0}, at: vfs_write+0x224/0xc90 fs/read_write.c:679
>   #1: ffff88806df6bc88 (&of->mutex){+.+.}-{3:3}, at: kernfs_fop_write_iter+0x1ea/0x500 fs/kernfs/file.c:325
>   #2: ffff888026fcf3c8 (kn->active#50){.+.+}-{0:0}, at: kernfs_fop_write_iter+0x20e/0x500 fs/kernfs/file.c:326
>   #3: ffffffff8f56f848 (nsim_bus_dev_list_lock){+.+.}-{3:3}, at: new_device_store+0x1b4/0x890 drivers/net/netdevsim/bus.c:166
>   #4: ffff88805e0140e8 (&dev->mutex){....}-{3:3}, at: device_lock include/linux/device.h:1014 [inline] 
>   #4: ffff88805e0140e8 (&dev->mutex){....}-{3:3}, at: __device_attach+0x8e/0x520 drivers/base/dd.c:1005
>   #5: ffff88805c5fb250 (&devlink->lock_key#55){+.+.}-{3:3}, at: nsim_drv_probe+0xcb/0xb80 drivers/net/netdevsim/dev.c:1534
>   #6: ffffffff8fcd1748 (rtnl_mutex){+.+.}-{3:3}, at: fib_seq_sum+0x31/0x290 net/core/fib_notifier.c:46
> 
> This is not a bug fix, unless I am mistaken, thus targeting net-next.
> 
> 
> Eric Dumazet (5):
>   fib: rules: use READ_ONCE()/WRITE_ONCE() on ops->fib_rules_seq
>   ipv4: use READ_ONCE()/WRITE_ONCE() on net->ipv4.fib_seq
>   ipv6: use READ_ONCE()/WRITE_ONCE() on fib6_table->fib_seq
>   ipmr: use READ_ONCE() to read net->ipv[46].ipmr_seq
>   net: do not acquire rtnl in fib_seq_sum()
> 
>  include/net/fib_notifier.h |  2 +-
>  include/net/fib_rules.h    |  2 +-
>  include/net/ip6_fib.h      |  8 ++++----
>  include/net/ip_fib.h       |  4 ++--
>  include/net/netns/ipv4.h   |  2 +-
>  net/core/fib_notifier.c    |  2 --
>  net/core/fib_rules.c       | 14 ++++++++------
>  net/ipv4/fib_notifier.c    | 10 +++++-----
>  net/ipv4/fib_rules.c       |  2 +-
>  net/ipv4/ipmr.c            | 10 ++++------
>  net/ipv6/fib6_notifier.c   |  2 +-
>  net/ipv6/fib6_rules.c      |  2 +-
>  net/ipv6/ip6_fib.c         | 14 +++++++-------
>  net/ipv6/ip6mr.c           | 10 ++++------
>  14 files changed, 40 insertions(+), 44 deletions(-)
> 

For the set:
Reviewed-by: David Ahern <dsahern@kernel.org>
patchwork-bot+netdevbpf@kernel.org Oct. 11, 2024, 10:50 p.m. UTC | #2
Hello:

This series was applied to netdev/net-next.git (main)
by Jakub Kicinski <kuba@kernel.org>:

On Wed,  9 Oct 2024 18:44:00 +0000 you wrote:
> This series is inspired by a syzbot report showing
> rtnl contention and one thread blocked in:
> 
> 7 locks held by syz-executor/10835:
>   #0: ffff888033390420 (sb_writers#8){.+.+}-{0:0}, at: file_start_write include/linux/fs.h:2931 [inline]
>   #0: ffff888033390420 (sb_writers#8){.+.+}-{0:0}, at: vfs_write+0x224/0xc90 fs/read_write.c:679
>   #1: ffff88806df6bc88 (&of->mutex){+.+.}-{3:3}, at: kernfs_fop_write_iter+0x1ea/0x500 fs/kernfs/file.c:325
>   #2: ffff888026fcf3c8 (kn->active#50){.+.+}-{0:0}, at: kernfs_fop_write_iter+0x20e/0x500 fs/kernfs/file.c:326
>   #3: ffffffff8f56f848 (nsim_bus_dev_list_lock){+.+.}-{3:3}, at: new_device_store+0x1b4/0x890 drivers/net/netdevsim/bus.c:166
>   #4: ffff88805e0140e8 (&dev->mutex){....}-{3:3}, at: device_lock include/linux/device.h:1014 [inline]
>   #4: ffff88805e0140e8 (&dev->mutex){....}-{3:3}, at: __device_attach+0x8e/0x520 drivers/base/dd.c:1005
>   #5: ffff88805c5fb250 (&devlink->lock_key#55){+.+.}-{3:3}, at: nsim_drv_probe+0xcb/0xb80 drivers/net/netdevsim/dev.c:1534
>   #6: ffffffff8fcd1748 (rtnl_mutex){+.+.}-{3:3}, at: fib_seq_sum+0x31/0x290 net/core/fib_notifier.c:46
> 
> [...]

Here is the summary with links:
  - [net-next,1/5] fib: rules: use READ_ONCE()/WRITE_ONCE() on ops->fib_rules_seq
    https://git.kernel.org/netdev/net-next/c/a716ff52bebf
  - [net-next,2/5] ipv4: use READ_ONCE()/WRITE_ONCE() on net->ipv4.fib_seq
    https://git.kernel.org/netdev/net-next/c/16207384d292
  - [net-next,3/5] ipv6: use READ_ONCE()/WRITE_ONCE() on fib6_table->fib_seq
    https://git.kernel.org/netdev/net-next/c/e60ea4544776
  - [net-next,4/5] ipmr: use READ_ONCE() to read net->ipv[46].ipmr_seq
    https://git.kernel.org/netdev/net-next/c/055202b16c58
  - [net-next,5/5] net: do not acquire rtnl in fib_seq_sum()
    https://git.kernel.org/netdev/net-next/c/2698acd6ea47

You are awesome, thank you!