Re: io_uring: BUG: kernel NULL pointer dereference

[Stefano Garzarella <sgarzare@redhat.com>]

On Mon, May 25, 2020 at 08:10:27AM -0600, Jens Axboe wrote:
> On 5/25/20 7:45 AM, Stefano Garzarella wrote:
> > On Mon, May 25, 2020 at 12:30:51PM +0200, Stefano Garzarella wrote:
> >> Hi Jens,
> >> using fio and io_uring engine with SQPOLL and IOPOLL enabled, I had the
> >> following issue that happens after 4/5 seconds fio has started.
> >> Initially I had this issue on Linux v5.7-rc6, but I just tried also
> >> Linux v5.7-rc7:
> >>
> >> [   75.343479] nvme nvme0: pci function 0000:04:00.0
> >> [   75.355110] nvme nvme0: 16/0/15 default/read/poll queues
> >> [   75.364946]  nvme0n1: p1
> >> [   82.739285] BUG: kernel NULL pointer dereference, address: 00000000000003b0
> >> [   82.747054] #PF: supervisor read access in kernel mode
> >> [   82.752785] #PF: error_code(0x0000) - not-present page
> >> [   82.758516] PGD 800000046c042067 P4D 800000046c042067 PUD 461fcf067 PMD 0 
> >> [   82.766186] Oops: 0000 [#1] SMP PTI
> >> [   82.770076] CPU: 2 PID: 1307 Comm: io_uring-sq Not tainted 5.7.0-rc7 #11
> >> [   82.777939] Hardware name: Dell Inc. PowerEdge R430/03XKDV, BIOS 1.2.6 06/08/2015
> >> [   82.786290] RIP: 0010:task_numa_work+0x4f/0x2c0
> >> [   82.791341] Code: 18 4c 8b 25 e3 f0 8e 01 49 8b 9f 00 08 00 00 4d 8b af c8 00 00 00 49 39 c7 0f 85 e8 01 00 00 48 89 6d 00 41 f6 47 24 04 75 67 <48> 8b ab b0 03 00 00 48 85 ed 75 16 8b 3d 6f 68 94 01 e8 aa fb 04
> >> [   82.812296] RSP: 0018:ffffaaa98415be10 EFLAGS: 00010246
> >> [   82.818123] RAX: ffff953ee36b8000 RBX: 0000000000000000 RCX: 0000000000000000
> >> [   82.826083] RDX: 0000000000000001 RSI: ffff953ee36b8000 RDI: ffff953ee36b8dc8
> >> [   82.834042] RBP: ffff953ee36b8dc8 R08: 00000000001200db R09: ffff9542e3ad2e08
> >> [   82.842002] R10: ffff9542ecd20070 R11: 0000000000000000 R12: 00000000fffca35b
> >> [   82.849962] R13: 000000012a06a949 R14: ffff9542e3ad2c00 R15: ffff953ee36b8000
> >> [   82.857922] FS:  0000000000000000(0000) GS:ffff953eefc40000(0000) knlGS:0000000000000000
> >> [   82.866948] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> >> [   82.873357] CR2: 00000000000003b0 CR3: 000000046bbd0002 CR4: 00000000001606e0
> >> [   82.881316] Call Trace:
> >> [   82.884046]  task_work_run+0x68/0xa0
> >> [   82.888026]  io_sq_thread+0x252/0x3d0
> >> [   82.892111]  ? finish_wait+0x80/0x80
> >> [   82.896097]  kthread+0xf9/0x130
> >> [   82.899598]  ? __ia32_sys_io_uring_enter+0x370/0x370
> >> [   82.905134]  ? kthread_park+0x90/0x90
> >> [   82.909217]  ret_from_fork+0x35/0x40
> >> [   82.913203] Modules linked in: nvme nvme_core xt_CHECKSUM xt_MASQUERADE xt_conntrack ipt_REJECT nf_reject_ipv4 tun bridge stp llc ip6table_mangle ip6table_nat iptable_mangle iptable_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 ebtable_filter ebtables ip6table_filter ip6_tables iptable_filter rfkill sunrpc intel_rapl_msr intel_rapl_common sb_edac x86_pkg_temp_thermal intel_powerclamp coretemp kvm_intel kvm irqbypass crct10dif_pclmul iTCO_wdt crc32_pclmul dcdbas ghash_clmulni_intel iTCO_vendor_support intel_cstate intel_uncore pcspkr intel_rapl_perf ipmi_ssif ixgbe mei_me mdio tg3 dca mei lpc_ich ipmi_si acpi_power_meter ipmi_devintf ipmi_msghandler ip_tables xfs libcrc32c mgag200 drm_kms_helper drm_vram_helper drm_ttm_helper ttm drm megaraid_sas crc32c_intel i2c_algo_bit wmi
> >> [   82.990613] CR2: 00000000000003b0
> >> [   82.994307] ---[ end trace 6d1725e8f60fece7 ]---
> >> [   83.039157] RIP: 0010:task_numa_work+0x4f/0x2c0
> >> [   83.044211] Code: 18 4c 8b 25 e3 f0 8e 01 49 8b 9f 00 08 00 00 4d 8b af c8 00 00 00 49 39 c7 0f 85 e8 01 00 00 48 89 6d 00 41 f6 47 24 04 75 67 <48> 8b ab b0 03 00 00 48 85 ed 75 16 8b 3d 6f 68 94 01 e8 aa fb 04
> >> [   83.065165] RSP: 0018:ffffaaa98415be10 EFLAGS: 00010246
> >> [   83.070993] RAX: ffff953ee36b8000 RBX: 0000000000000000 RCX: 0000000000000000
> >> [   83.078953] RDX: 0000000000000001 RSI: ffff953ee36b8000 RDI: ffff953ee36b8dc8
> >> [   83.086913] RBP: ffff953ee36b8dc8 R08: 00000000001200db R09: ffff9542e3ad2e08
> >> [   83.094873] R10: ffff9542ecd20070 R11: 0000000000000000 R12: 00000000fffca35b
> >> [   83.102833] R13: 000000012a06a949 R14: ffff9542e3ad2c00 R15: ffff953ee36b8000
> >> [   83.110793] FS:  0000000000000000(0000) GS:ffff953eefc40000(0000) knlGS:0000000000000000
> >> [   83.119821] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> >> [   83.126230] CR2: 00000000000003b0 CR3: 000000046bbd0002 CR4: 00000000001606e0
> >> [  113.113624] nvme nvme0: I/O 219 QID 19 timeout, aborting
> >> [  113.120135] nvme nvme0: Abort status: 0x0
> >>
> >> Steps I did:
> >>
> >>   $ modprobe nvme poll_queues=15
> >>   $ fio fio_iou.job
> >>
> >> This is the fio_iou.job that I used:
> >>
> >>   [global]
> >>   filename=/dev/nvme0n1
> >>   ioengine=io_uring
> >>   direct=1
> >>   runtime=60
> >>   ramp_time=5
> >>   gtod_reduce=1
> >>
> >>   cpus_allowed=4
> >>
> >>   [job1]
> >>   rw=randread
> >>   bs=4K
> >>   iodepth=1
> >>   registerfiles
> >>   sqthread_poll=1
> >>   sqthread_poll_cpu=2
> >>   hipri
> >>
> >> I'll try to bisect, but I have some suspicions about:
> >> b41e98524e42 io_uring: add per-task callback handler
> > 
> > I confirm, the bisection ended with this:
> > b41e98524e424d104aa7851d54fd65820759875a is the first bad commit
> 
> I think the odd part here is that task_tick_numa() checks for a
> valid mm, and queues work if the task has it. But for the sqpoll
> kthread, the mm can come and go. By the time the task work is run,
> the mm is gone and we oops on current->mm == NULL.
> 
> I think the below should fix it:
> 
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 538ba5d94e99..24a8557f001f 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -2908,7 +2908,8 @@ static void task_tick_numa(struct rq *rq, struct task_struct *curr)
>  	/*
>  	 * We don't care about NUMA placement if we don't have memory.
>  	 */
> -	if (!curr->mm || (curr->flags & PF_EXITING) || work->next != work)
> +	if (!curr->mm || (curr->flags & (PF_EXITING | PF_KTHREAD)) ||
> +	    work->next != work)
>  		return;
>  
>  	/*
> 

Great Jens! With this patch applied, everything works properly!

Pavel, Hillf, thanks both for the comments, this patch solves the problem I
have described.

Thanks,
Stefano