Date:   Wed, 10 Jun 2020 09:37:36 +0800
From:   Zefan Li <lizefan@...wei.com>
To:     Daniel Borkmann <daniel@...earbox.net>,
        Alexei Starovoitov <alexei.starovoitov@...il.com>,
        Brenden Blanco <bblanco@...il.com>,
        <karsten.elfenbein@...il.com>
CC:     Lu Fengqi <lufq.fnst@...fujitsu.com>, bpf <bpf@...r.kernel.org>,
        "Network Development" <netdev@...r.kernel.org>,
        Alexei Starovoitov <ast@...nel.org>, Tejun Heo <tj@...nel.org>,
        Yang Yingliang <yangyingliang@...wei.com>
Subject: Re: BUG: kernel NULL pointer dereference in
 __cgroup_bpf_run_filter_skb

On 2020/6/10 4:50, Daniel Borkmann wrote:
> [ +Zefan, + TJ ]
> 
> On 6/3/20 12:17 AM, Alexei Starovoitov wrote:
>> On Tue, Jun 2, 2020 at 2:46 PM Brenden Blanco <bblanco@...il.com> wrote:
>>> On Sat, May 30, 2020 at 12:51 AM Lu Fengqi <lufq.fnst@...fujitsu.com> wrote:
>>>>
>>>> I encountered a reproducible NULL pointer dereference using the mainline
>>>> kernel v5.7-rc7-44-g75caf310d16c(which also happened multiple times on
>>>> 5.6.14). The machine is installed with archlinux, used as a kubernetes
>>>> v1.18.3 node, and uses calico v3.13.2 as a cni plugin. I use kdump/crash
>>>> to see the value of the bpf_prog pointer in cgroup.bpf is 0x0 or 0x800.
>>>>
>>>> I am not sure whether this is caused by kernel bpf or calico? If you need
>>>> me to provide more information, please let me know. Any suggestions are
>>>> very helpful.
>>>
>>> I encountered a similar set of crashes. I was able to workaround it by
>>> disabling the systemd IPAddressDeny feature until the number of
>>> bpf-progs in use by systemd reached 0 (via lsof inspection). I hit the
>>> crash in kernels 5.4.43 through 5.7.
>>>
>>> [40188.268677] BUG: kernel NULL pointer dereference, address: 0000000000000010
>>> [40188.268736] #PF: supervisor read access in kernel mode
>>> [40188.268773] #PF: error_code(0x0000) - not-present page
>>> [40188.268819] PGD 0 P4D 0
>>> [40188.268842] Oops: 0000 [#1] SMP PTI
>>> [40188.268871] CPU: 7 PID: 2834 Comm: nfsd Tainted: P           OE
>>> 5.4.43-1-lts #1
>>> [40188.268915] Hardware name: Supermicro Super Server/X10SRi-F, BIOS
>>> 3.2 11/22/2019
>>> [40188.268970] RIP: 0010:__cgroup_bpf_run_filter_skb+0x155/0x1d0
>>> [40188.269013] Code: 48 8b 4c 24 08 4c 01 ab c8 00 00 00 48 89 4b 18
>>> 48 83 c4 10 5b 5d 41 5c 41 5d 41 5e 41 5f c3 31 c0 c3 c3 48 8b 86 38
>>> 06 00 00 <48> 8b 78 10 4c 8d 70 10 48 85 ff 74 5f 31 ed 49 8b 46 08 65
>>> 48 89
>>> [40188.269115] RSP: 0018:ffffb0c581cf3918 EFLAGS: 00010246
>>> [40188.269153] RAX: 0000000000000000 RBX: ffff8e32156bfae0 RCX: 0000000000000048
>>> [40188.269202] RDX: 0000000000000000 RSI: ffff8e31f9d1e000 RDI: ffff8e31f9bc8940
>>> [40188.269250] RBP: ffff8e31f9bc8940 R08: ffff8e3215c74a40 R09: 0000000000000001
>>> [40188.269299] R10: 0000000000000003 R11: 0000000000000000 R12: 0000000000000000
>>> [40188.269348] R13: 0000000000000000 R14: 000000000000e400 R15: 0000000000000001
>>> [40188.269391] FS:  0000000000000000(0000) GS:ffff8e321fbc0000(0000)
>>> knlGS:0000000000000000
>>> [40188.269446] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>> [40188.269481] CR2: 0000000000000010 CR3: 0000000271c0a001 CR4: 00000000001606e0
>>> [40188.269530] Call Trace:
>>> [40188.269559]  ip6_finish_output+0x68/0xa0
>>> [40188.269587]  ip6_output+0x6e/0x130
>>> [40188.269615]  ? __ip6_finish_output+0x110/0x110
>>> [40188.269648]  ip6_xmit+0x2cf/0x5e0
>>> [40188.269675]  ? ipv6_anycast_cleanup+0x50/0x50
>>> [40188.269711]  inet6_csk_xmit+0xb6/0x100
>>> [40188.269742]  __tcp_transmit_skb+0x4ff/0xb10
>>> [40188.269776]  tcp_write_xmit+0x517/0x1030
>>> [40188.269807]  __tcp_push_pending_frames+0x32/0xf0
>>> [40188.269843]  do_tcp_sendpages+0x5fa/0x630
>>> [40188.269875]  tcp_sendpage+0x48/0x80
>>> [40188.269904]  inet_sendpage+0x52/0x90
>>> [40188.269931]  kernel_sendpage+0x1a/0x30
>>> [40188.269989]  svc_send_common+0x136/0x150 [sunrpc]
>>> [40188.270044]  svc_sendto+0xd7/0x240 [sunrpc]
>>> [40188.270096]  svc_tcp_sendto+0x36/0x50 [sunrpc]
>>> [40188.271531]  svc_send+0x7b/0x150 [sunrpc]
>>> [40188.272961]  nfsd+0xe3/0x140 [nfsd]
>>> [40188.274381]  ? nfsd_destroy+0x50/0x50 [nfsd]
>>> [40188.275785]  kthread+0x117/0x130
>>> [40188.277166]  ? __kthread_bind_mask+0x60/0x60
>>> [40188.278520]  ret_from_fork+0x35/0x40
>>> [40188.279820] Modules linked in: netconsole veth macvlan xt_nat
>>> xt_MASQUERADE nf_conntrack_netlink nfnetlink xfrm_user xfrm_algo
>>> iptable_nat rpcsec_gss_krb5 nls_iso8859_1 nls_cp437 vfat fat 8021q
>>> garp mrp twofish_generic twofish_avx_x86_64 twofish_x86_64_3way
>>> twofish_x86_64 twofish_common intel_rapl_msr intel_rapl_common
>>> algif_skcipher af_alg zfs(POE) sb_edac x86_pkg_temp_thermal
>>> intel_powerclamp iTCO_wdt zunicode(POE) ipmi_ssif zavl(POE)
>>> iTCO_vendor_support icp(POE) coretemp kvm_intel kvm irqbypass
>>> zcommon(POE) znvpair(POE) intel_cstate spl(OE) intel_uncore zlua(POE)
>>> intel_rapl_perf ast drm_vram_helper pcspkr ttm ixgbe drm_kms_helper
>>> i2c_i801 joydev mei_me syscopyarea sysfillrect sysimgblt igb libphy
>>> fb_sys_fops mousedev ioatdma i2c_algo_bit mdio input_leds lpc_ich mei
>>> dca ipmi_si ipmi_devintf ipmi_msghandler evdev mac_hid
>>> acpi_power_meter ip6t_REJECT nf_reject_ipv6 xt_hl ip6t_rt ipt_REJECT
>>> nf_reject_ipv4 xt_multiport br_netfilter bridge stp llc xt_limit
>>> xt_addrtype xt_tcpudp xt_physdev
>>> [40188.279868]  xt_conntrack ip6table_filter ip6_tables
>>> nf_conntrack_netbios_ns nf_conntrack_broadcast nf_nat_ftp nf_nat
>>> nf_conntrack_ftp nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 libcrc32c
>>> iptable_filter nfsd sha256_ssse3 drm sha1_ssse3 auth_rpcgss nfs_acl
>>> lockd grace agpgart sunrpc ip_tables x_tables ext4 crc32c_generic
>>> crc16 mbcache jbd2 raid1 md_mod hid_generic usbhid hid sd_mod dm_crypt
>>> dm_mod crct10dif_pclmul crc32_pclmul crc32c_intel ghash_clmulni_intel
>>> ahci aesni_intel libahci xhci_pci crypto_simd libata xhci_hcd cryptd
>>> glue_helper scsi_mod ehci_pci ehci_hcd wmi
>>> [40188.300047] CR2: 0000000000000010
>>> [40188.301739] ---[ end trace 44ac77af42fe7195 ]---
>>>
>>> I also hit an interesting (related?) warning on one of the iterations
>>> during boot.
>>>
>>> [ 8891.070041] ------------[ cut here ]------------
>>> [ 8891.070093] percpu ref (cgroup_bpf_release_fn) <= 0 (-1) after
>>> switching to atomic
>>> [ 8891.070117] WARNING: CPU: 7 PID: 54 at lib/percpu-refcount.c:160
>>> percpu_ref_switch_to_atomic_rcu+0x12f/0x140
>>> [ 8891.070178] Modules linked in: netconsole veth macvlan xt_nat
>>> xt_MASQUERADE nf_conntrack_netlink nfnetlink xfrm_user xfrm_algo
>>> iptable_nat nls_iso8859_1 nls_cp437 vfat fat 8021q garp mrp
>>> twofish_generic twofish_avx_x86_64 twofish_x86_64_3way twofish_x86_64
>>> twofish_common algif_skcipher af_alg intel_rapl_msr intel_rapl_common
>>> zfs(POE) zunicode(POE) zavl(POE) iTCO_wdt icp(POE) iTCO_vendor_support
>>> ipmi_ssif sb_edac x86_pkg_temp_thermal zcommon(POE) intel_powerclamp
>>> znvpair(POE) kvm_intel spl(OE) zlua(POE) kvm irqbypass ast
>>> intel_cstate intel_uncore drm_vram_helper ttm intel_rapl_perf pcspkr
>>> drm_kms_helper i2c_i801 syscopyarea joydev lpc_ich sysfillrect
>>> mousedev input_leds ixgbe sysimgblt fb_sys_fops mei_me igb mei libphy
>>> ioatdma i2c_algo_bit mdio dca ipmi_si acpi_power_meter ipmi_devintf
>>> ipmi_msghandler evdev mac_hid ip6t_REJECT nf_reject_ipv6 xt_hl ip6t_rt
>>> ipt_REJECT nf_reject_ipv4 xt_multiport br_netfilter bridge stp llc
>>> xt_limit xt_addrtype xt_tcpudp xt_physdev xt_conntrack ip6table_filter
>>> [ 8891.070220]  ip6_tables nf_conntrack_netbios_ns
>>> nf_conntrack_broadcast nf_nat_ftp nf_nat nf_conntrack_ftp nf_conntrack
>>> nf_defrag_ipv6 nf_defrag_ipv4 libcrc32c iptable_filter nfsd
>>> sha256_ssse3 drm sha1_ssse3 nfs_acl lockd auth_rpcgss grace sunrpc
>>> agpgart ip_tables x_tables ext4 crc32c_generic crc16 mbcache jbd2
>>> raid1 hid_generic usbhid hid md_mod sd_mod dm_crypt dm_mod
>>> crct10dif_pclmul crc32_pclmul crc32c_intel ghash_clmulni_intel ahci
>>> libahci aesni_intel libata crypto_simd cryptd xhci_pci glue_helper
>>> ehci_pci xhci_hcd scsi_mod ehci_hcd wmi [last unloaded: coretemp]
>>> [ 8891.070664] CPU: 7 PID: 54 Comm: ksoftirqd/7 Tainted: P
>>> OE     5.4.43-1-lts #1
>>> [ 8891.070691] Hardware name: Supermicro Super Server/X10SRi-F, BIOS
>>> 3.2 11/22/2019
>>> [ 8891.070721] RIP: 0010:percpu_ref_switch_to_atomic_rcu+0x12f/0x140
>>> [ 8891.070745] Code: eb 99 80 3d 33 da eb 00 00 0f 85 4d ff ff ff 48
>>> 8b 55 d8 48 8b 75 e8 48 c7 c7 20 1c 50 8a c6 05 17 da eb 00 01 e8 0f
>>> 4c c3 ff <0f> 0b e9 2b ff ff ff 0f 0b eb a2 90 90 90 90 90 90 8d 8c 16
>>> ef be
>>> [ 8891.070812] RSP: 0018:ffffbf72c027fe00 EFLAGS: 00010286
>>> [ 8891.070833] RAX: 0000000000000000 RBX: 8000000000000002 RCX: 0000000000000000
>>> [ 8891.070857] RDX: 0000000000000046 RSI: ffffffff8acd7b46 RDI: 0000000000000246
>>> [ 8891.070885] RBP: ffffa0b458b1f8e8 R08: 000008161d131aa4 R09: 0000000000000046
>>> [ 8891.070913] R10: 0000000080000007 R11: ffffffff8acd7b2b R12: 00003ebe60014fc8
>>> [ 8891.070938] R13: ffffa0b45fbeb350 R14: ffffa0b45b953c00 R15: ffffa0b45b953c00
>>> [ 8891.070968] FS:  0000000000000000(0000) GS:ffffa0b45fbc0000(0000)
>>> knlGS:0000000000000000
>>> [ 8891.071001] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>> [ 8891.071025] CR2: 00007fae6b351e10 CR3: 000000082b21c005 CR4: 00000000001606e0
>>> [ 8891.071058] Call Trace:
>>> [ 8891.071075]  rcu_core+0x1ba/0x4e0
>>> [ 8891.071093]  __do_softirq+0xe9/0x2dc
>>> [ 8891.071110]  run_ksoftirqd+0x26/0x40
>>> [ 8891.072222]  smpboot_thread_fn+0xc5/0x160
>>> [ 8891.073320]  ? smpboot_unregister_percpu_thread+0x60/0x60
>>> [ 8891.074423]  kthread+0x117/0x130
>>> [ 8891.075542]  ? __kthread_bind_mask+0x60/0x60
>>> [ 8891.076672]  ret_from_fork+0x35/0x40
>>> [ 8891.077815] ---[ end trace 727b9af96c86f011 ]---
>>
>> Thanks for the reporting.
>> We've seen very similar stack trace due to out of order cgroup destroy.
>> But it was fixed in
>> commit e10360f815ca ("bpf: cgroup: prevent out-of-order release of cgroup bpf")
>> Sounds like it wasn't fixed completely?
>> If somebody can reproduce could you please
>> revert both the fix e10360f815ca and offending
>> commit 4bfc0bb2c60e ("bpf: decouple the lifetime of cgroup_bpf from
>> cgroup itself")
>> and see whether it reproduces ?
>> It will help us narrow down the problem.
> 
> Recent updates in [0] suggest that reverting [1] would fix the panic.
> 
> Zefan, please take a look.
> 
>   [0] https://bugzilla.kernel.org/show_bug.cgi?id=208003
>   [1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=090e28b229af92dc5b40786ca673999d59e73056
> 

Haven't looked into it, but my guess is the bug is in somewhere else,
and the above commit just unveiled it.

Powered by blists - more mailing lists