Re: [syzbot] [kernel?] BUG: unable to handle kernel NULL pointer dereference in hrtimer_active

[Thomas Gleixner]

Cc: +netdev

On Mon, Feb 12 2024 at 02:25, syzbot wrote:
> HEAD commit:    4a7bbe7519b6 Merge tag 'scsi-fixes' of git://git.kernel.or..
> git tree:       upstream
> console output: https://syzkaller.appspot.com/x/log.txt?x=10476de0180000
> kernel config:  https://syzkaller.appspot.com/x/.config?x=264238120cdb2bda
> dashboard link: https://syzkaller.appspot.com/bug?extid=039399a9b96297ddedca
> compiler:       aarch64-linux-gnu-gcc (Debian 12.2.0-14) 12.2.0, GNU ld (GNU Binutils for Debian) 2.40
> userspace arch: arm64
>
> Unfortunately, I don't have any reproducer for this issue yet.
>
> Downloadable assets:
> disk image (non-bootable): https://storage.googleapis.com/syzbot-assets/384ffdcca292/non_bootable_disk-4a7bbe75.raw.xz
> vmlinux: https://storage.googleapis.com/syzbot-assets/ee3f97d1ed38/vmlinux-4a7bbe75.xz
> kernel image: https://storage.googleapis.com/syzbot-assets/eb6f9f8f9f37/Image-4a7bbe75.gz.xz
>
> IMPORTANT: if you fix the issue, please add the following tag to the commit:
> Reported-by: syzbot+039399a9b96297ddedca@xxxxxxxxxxxxxxxxxxxxxxxxx
>
> infiniband syz0: set active
> Unable to handle kernel NULL pointer dereference at virtual address 0000000000000010
> CPU: 1 PID: 3665 Comm: syz-executor.0 Not tainted 6.8.0-rc3-syzkaller-00279-g4a7bbe7519b6 #0
> pc : __seqprop_raw_spinlock_sequence include/linux/seqlock.h:226 [inline]

That's:

        seq = raw_read_seqcount_begin(&base->seq);

where base == NULL. That can only happen when hrtimer_cancel() is
invoked with a non-initialized timer.

> pc : hrtimer_active+0x4/0x60 kernel/time/hrtimer.c:1614
> lr : hrtimer_try_to_cancel+0x1c/0xf8 kernel/time/hrtimer.c:1331
> sp : ffff800082c63300
> x29: ffff800082c63300 x28: 0000000000000000 x27: 0000000000000000
> x26: 0000000000000340 x25: 0000000000000000 x24: f3ff00001ab7e9e0
> x23: 0000000000000000 x22: 000061100fc019e9 x21: 0000000000000009
> x20: 0000000000000000 x19: fbff00001abf9920 x18: 0000000000000000
> x17: 0000000000000000 x16: 0000000000000000 x15: ffff80008144d28c
> x14: ffff80008144d20c x13: ffff80008144d28c x12: ffff80008144d20c
> x11: ffff800080011558 x10: ffff800081907d14 x9 : ffff8000819078a4
> x8 : ffff800082c63408 x7 : 0000000000000000 x6 : ffff800080026d20
> x5 : f2ff000033f2c800 x4 : 000061100fc019e9 x3 : 0000000000000340
> x2 : 0000000000000000 x1 : 000000000000000d x0 : fbff00001abf9920
> Call trace:
>  hrtimer_active+0x4/0x60 kernel/time/hrtimer.c:1613
>  hrtimer_cancel+0x1c/0x38 kernel/time/hrtimer.c:1446
>  napi_disable+0x5c/0x11c net/core/dev.c:6502
>  veth_napi_del_range+0x64/0x1d8 drivers/net/veth.c:1109
>  veth_napi_del drivers/net/veth.c:1129 [inline]
>  veth_set_features+0x68/0x98 drivers/net/veth.c:1580
>  __netdev_update_features+0x200/0x6ec net/core/dev.c:9872
>  netdev_update_features+0x28/0x6c net/core/dev.c:9946
>  veth_xdp_set drivers/net/veth.c:1681 [inline]
>  veth_xdp+0x108/0x224 drivers/net/veth.c:1694
>  dev_xdp_install+0x64/0xf8 net/core/dev.c:9243
>  dev_xdp_attach+0x250/0x52c net/core/dev.c:9395
>  dev_change_xdp_fd+0x16c/0x218 net/core/dev.c:9643
>  do_setlink+0xdd0/0xf14 net/core/rtnetlink.c:3132
>  rtnl_group_changelink net/core/rtnetlink.c:3452 [inline]
>  __rtnl_newlink+0x460/0x898 net/core/rtnetlink.c:3711
>  rtnl_newlink+0x50/0x7c net/core/rtnetlink.c:3748
>  rtnetlink_rcv_msg+0x12c/0x380 net/core/rtnetlink.c:6615
>  netlink_rcv_skb+0x5c/0x128 net/netlink/af_netlink.c:2543
>  rtnetlink_rcv+0x18/0x24 net/core/rtnetlink.c:6633
>  netlink_unicast_kernel net/netlink/af_netlink.c:1341 [inline]
>  netlink_unicast+0x2f4/0x360 net/netlink/af_netlink.c:1367
>  netlink_sendmsg+0x1a4/0x3e8 net/netlink/af_netlink.c:1908
>  sock_sendmsg_nosec net/socket.c:730 [inline]
>  __sock_sendmsg+0x54/0x60 net/socket.c:745
>  ____sys_sendmsg+0x274/0x2ac net/socket.c:2584
>  ___sys_sendmsg+0xac/0x100 net/socket.c:2638
>  __sys_sendmsg+0x84/0xe0 net/socket.c:2667
>  __do_sys_sendmsg net/socket.c:2676 [inline]
>  __se_sys_sendmsg net/socket.c:2674 [inline]
>  __arm64_sys_sendmsg+0x24/0x30 net/socket.c:2674

So something in that syzbot test case manages to tear down a napi
context which has not yet been fully initialized. While the rest of
napi_disable() does not care much as long as neither NAPIF_STATE_SCHED
nor NAPIF_STATE_NPSVC are set in napi->state, hrtimer_cancel() pretty
much cares as demonstrated by the NULL pointer dereference.

While it would be trivial to harden the hrtimer code for the case that a
non-initialized hrtimer is canceled, I wonder whether this invocation of
napi_disable() is harmless (aside of the hrtimer issue) or if there are
some hidden subtle issues with that.

Thanks,

        tglx