Re: [list] 8052ce2d97: WARNING: CPU: 0 PID: 1 at lib/list_debug.c:60 __list_del_entry+0x4d/0x71

From: Kees Cook
Date: Sat Oct 22 2016 - 02:08:49 EST


On Sat, Sep 3, 2016 at 5:35 PM, kernel test robot <xiaolong.ye@xxxxxxxxx> wrote:
>
> FYI, we noticed the following commit:
>
> https://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git dev.2016.08.19a
> commit 8052ce2d9771ab5a818307f3abbaf33bba82a631 ("list: Split list_add() debug checking into separate function")
>
> in testcase: boot
>
> on test machine: qemu-system-i386 -enable-kvm -m 360M
>
> caused below changes:

Okay, I've finally decoded this test result. There is nothing scary
here. I will explain below...

> +------------------------------------------------------------+------------+------------+
> | | 0a33d912a3 | 8052ce2d97 |
> +------------------------------------------------------------+------------+------------+
> | boot_successes | 0 | 0 |
> | boot_failures | 211 | 18 |

First, the prior commit (before my change) also never successfully booted.

> | WARNING:at_lib/list_debug.c:#__list_del_entry | 211 | 18 |

In both cases, list_del threw a warning every time.

> | calltrace:init | 209 | 18 |
> | WARNING:at_lib/list_debug.c:#__list_add | 209 | |

Before the change, subsequent list_adds would also throw a warning
(since memory had gone ahead and gotten corrupted in the prior
list_del calls). After the change, this stopped happening since the
corruption got isolated.

> | calltrace:i8042_init | 209 | 17 |
> | calltrace:SyS_reboot | 192 | 4 |
> | EIP_is_at__mutex_unlock_slowpath | 29 | |
> | Kernel_panic-not_syncing:softlockup:hung_tasks | 138 | |
> | BUG:workqueue_lockup-pool_cpus=#flags=#nice=#stuck_for#s | 56 | |
> | EIP_is_at_lock_release | 7 | |
> | EIP_is_at_lock_acquire | 28 | |
> | EIP_is_at_raw_spin_unlock_irq | 30 | |
> | EIP_is_at_mutex_lock_nested | 21 | |
> | EIP_is_at_lock_is_held | 10 | |
> | EIP_is_at_device_shutdown | 3 | |
> | EIP_is_at___might_sleep | 5 | |
> | EIP_is_at__list_del_entry | 4 | 1 |
> | EIP_is_at_get_device | 1 | |
> | BUG:unable_to_handle_kernel | 7 | 5 |
> | Oops | 7 | 5 |
> | calltrace:SyS_ipc | 4 | 4 |
> | Kernel_panic-not_syncing:Fatal_exception | 7 | 5 |
> | EIP_is_at_pm_runtime_barrier | 1 | |
> | EIP_is_at_lock_acquired | 1 | |
> | EIP_is_at__might_sleep | 1 | |
> | BUG:workqueue_lockup-pool_cpus=#-#flags=#nice=#stuck_for#s | 1 | |
> | WARNING:at_arch/x86/mm/dump_pagetables.c:#note_page | 26 | |
> | calltrace:mark_rodata_ro | 26 | |
> | WARNING:at_lib/list_debug.c:#__list_add_valid | 0 | 17 |
> | calltrace:serio_handle_event | 0 | 17 |
> | calltrace:userio_misc_init | 0 | 17 |
> | calltrace:mousedev_init | 0 | 17 |
> | calltrace:evdev_init | 0 | 17 |
> | calltrace:cmos_init | 0 | 17 |
> | calltrace:test_init | 0 | 17 |
> | calltrace:pca_isa_driver_init | 0 | 17 |
> | calltrace:vim2m_init | 0 | 17 |
> | calltrace:fm_drv_init | 0 | 17 |
> | calltrace:test_power_init | 0 | 17 |
> | calltrace:raid#_init | 0 | 17 |
> | calltrace:dm_init | 0 | 17 |
> | calltrace:vhci_init | 0 | 17 |
> | calltrace:dcdrbu_init | 0 | 17 |
> | calltrace:dcdbas_init | 0 | 17 |
> | calltrace:uhid_init | 0 | 17 |
> | calltrace:vhost_net_init | 0 | 17 |
> | calltrace:dummy_stm_init | 0 | 17 |
> | calltrace:stm_heartbeat_init | 0 | 17 |
> | calltrace:binder_init | 0 | 17 |
> | calltrace:teql_init | 0 | 17 |
> | calltrace:ipip_init | 0 | 17 |
> | calltrace:ipgre_init | 0 | 17 |
> | calltrace:sit_init | 0 | 17 |
> | calltrace:ip6_tunnel_init | 0 | 17 |
> | calltrace:ip6gre_init | 0 | 17 |
> | calltrace:irlan_init | 0 | 17 |
> | calltrace:pm_qos_power_init | 0 | 17 |
> | calltrace:ubi_init | 0 | 17 |

In most cases, the new patch allowed the kernel to boot further (more
complete call traces), before the system ultimately falls over (since
there IS some corruption happening somewhere else in -next).

> | calltrace:SyS_setsockopt | 0 | 2 |
> | calltrace:SyS_socketcall | 0 | 2 |
> | WARNING:at_include/linux/kref.h:#kobject_get | 0 | 4 |
> | WARNING:at_kernel/locking/lockdep.c:#__lock_acquire | 0 | 4 |
> | BUG:kernel_hang_in_test_stage | 0 | 4 |
> | calltrace:fw_cfg_sysfs_init | 0 | 7 |
> +------------------------------------------------------------+------------+------------+
>
>
>
> [ 152.680291] g_acm_ms usbip-vudc.0: failed to start g_acm_ms: -22
> [ 152.689209] usbip-vudc: probe of usbip-vudc.0 failed with error -22
> [ 152.698509] ------------[ cut here ]------------
> [ 152.709786] WARNING: CPU: 0 PID: 1 at lib/list_debug.c:60 __list_del_entry+0x4d/0x71
> [ 152.722785] list_del corruption. next->prev should be cc0230d4, but was 6b6b6b6b
> [ 152.732828] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.8.0-rc1-00033-g8052ce2 #1
> [ 152.742576] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Debian-1.8.2-1 04/01/2014
> [ 152.754081] 00000000 00200246 c0033e78 c13c456b c0033ea4 c13e72ac c0033e90 c103fd02
> [ 152.765720] 0000003c 6b6b6b6b cc0230d4 cc02300c c0033eac c103fd41 00000009 00000000
> [ 152.776864] c0033ea4 c1f7bc14 c0033ec0 c0033ecc c13e72ac c1f7bace 0000003c c1f7bc14
> [ 152.788472] Call Trace:
> [ 152.791964] [<c13c456b>] dump_stack+0x75/0xa9
> [ 152.797835] [<c13e72ac>] ? __list_del_entry+0x4d/0x71
> [ 152.804221] [<c103fd02>] __warn+0xad/0xc4
> [ 152.809694] [<c103fd41>] warn_slowpath_fmt+0x28/0x2d
> [ 152.816501] [<c13e72ac>] __list_del_entry+0x4d/0x71
> [ 152.822946] [<c15027a8>] device_pm_remove+0x2b/0x5a
> [ 152.829560] [<c14f6317>] device_del+0x134/0x19a
> [ 152.835558] [<c14f5c38>] ? dev_set_name+0x14/0x16
> [ 152.842135] [<c14f9d7c>] platform_device_del+0x16/0x5d
> [ 152.848909] [<c2207515>] init+0xe5/0x112
> [ 152.854149] [<c2207430>] ? usbip_host_init+0xe5/0xe5
> [ 152.860796] [<c21cec35>] do_one_initcall+0x7e/0x101
> [ 152.867316] [<c10564b7>] ? parse_args+0x190/0x265
> [ 152.873497] [<c21ced88>] ? kernel_init_freeable+0xd0/0x174
> [ 152.880923] [<c21ceda4>] kernel_init_freeable+0xec/0x174
> [ 152.887499] [<c1be421b>] kernel_init+0x8/0xd0
> [ 152.893658] [<c1bec3ae>] ret_from_kernel_thread+0xe/0x30
> [ 152.900759] [<c1be4213>] ? rest_init+0x111/0x111
> [ 152.907334] ---[ end trace 2003739b08d25430 ]---
> [ 152.920040] ------------[ cut here ]------------
>
>
>
>
>
>
> Thanks,
> Kernel Test Robot

As a result, this appears to validate the secondary purpose of the
patch: avoiding further memory corruption when lists have been stomped
on. So, this is Working As Intended. :)

The reason 0day sent an email is because the fingerprint of the boot
failure changed, not because my change _caused_ a boot failure. So,
CONFIG_DEBUG_LIST was already reporting the problem, it's just that
with my patch the system could survive a little longer before falling
over.

-Kees

--
Kees Cook
Nexus Security