Re: io_uring NULL pointer dereference on Linux v5.4-rc1

[Jens Axboe <axboe@kernel.dk>]

On 10/9/19 11:46 AM, Stefan Hajnoczi wrote:
> On Wed, Oct 09, 2019 at 05:27:44AM -0600, Jens Axboe wrote:
>> On 10/9/19 3:23 AM, Stefan Hajnoczi wrote:
>>> I hit this NULL pointer dereference when running qemu-iotests 052 (raw)
>>> on both ext4 and XFS on dm-thin/luks.  The kernel is Linux v5.4-rc1 but
>>> I haven't found any obvious fixes in Jens' tree, so it's likely that
>>> this bug is still present:
>>>
>>> BUG: kernel NULL pointer dereference, address: 0000000000000102
>>> #PF: supervisor read access in kernel mode
>>> #PF: error_code(0x0000) - not-present page
>>> PGD 0 P4D 0
>>> Oops: 0000 [#1] SMP PTI
>>> CPU: 2 PID: 6656 Comm: qemu-io Not tainted 5.4.0-rc1 #1
>>> Hardware name: LENOVO 20BTS1N70V/20BTS1N70V, BIOS N14ET37W (1.15 ) 09/06/2016
>>> RIP: 0010:__queue_work+0x1f/0x3b0
>>> Code: eb df 66 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 41 57 49 89 f7 41 56 41 89 fe 41 55 41 89 fd 41 54 55 48 89 d5 53 48 83 ec 10 <f6> 86 02 01 00 00 01 0f 85 bc 02 00 00 49 bc eb 83 b5 80 46 86 c8
>>> RSP: 0018:ffffbef4884bbd58 EFLAGS: 00010082
>>> RAX: 0000000000000246 RBX: 0000000000000246 RCX: 0000000000000000
>>> RDX: ffff9903901f4460 RSI: 0000000000000000 RDI: 0000000000000040
>>> RBP: ffff9903901f4460 R08: ffff9903901fb040 R09: ffff990398614700
>>> R10: 0000000000000030 R11: 0000000000000000 R12: 0000000000000000
>>> R13: 0000000000000040 R14: 0000000000000040 R15: 0000000000000000
>>> FS:  00007f7d2a4e4a80(0000) GS:ffff9903a5a80000(0000) knlGS:0000000000000000
>>> CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>> CR2: 0000000000000102 CR3: 0000000203da8004 CR4: 00000000003606e0
>>> DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
>>> DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
>>> Call Trace:
>>>    ? __io_queue_sqe+0xa1/0x200
>>>    queue_work_on+0x36/0x40
>>>    __io_queue_sqe+0x16e/0x200
>>>    io_ring_submit+0xd2/0x230
>>>    ? percpu_ref_resurrect+0x46/0x70
>>>    ? __io_uring_register+0x207/0xa30
>>>    ? __schedule+0x286/0x700
>>>    __x64_sys_io_uring_enter+0x1a3/0x280
>>>    ? __x64_sys_io_uring_register+0x64/0xb0
>>>    do_syscall_64+0x5b/0x180
>>>    entry_SYSCALL_64_after_hwframe+0x44/0xa9
>>> RIP: 0033:0x7f7d3439f1fd
>>> Code: 00 c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 5b 8c 0c 00 f7 d8 64 89 01 48
>>> RSP: 002b:00007f7d2918d408 EFLAGS: 00000216 ORIG_RAX: 00000000000001aa
>>> RAX: ffffffffffffffda RBX: 00007f7d2918d4f0 RCX: 00007f7d3439f1fd
>>> RDX: 0000000000000000 RSI: 0000000000000001 RDI: 000000000000000a
>>> RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000008
>>> R10: 0000000000000000 R11: 0000000000000216 R12: 00005616e3c32ab8
>>> R13: 00005616e3c32b78 R14: 00005616e3c32ab0 R15: 0000000000000001
>>> Modules linked in: fuse ccm xt_CHECKSUM xt_MASQUERADE tun bridge stp llc nf_conntrack_netbios_ns nf_conntrack_broadcast xt_CT ip6t_rpfilter ip6t_REJECT nf_reject_ipv6 ipt_REJECT nf_reject_ipv4 xt_conntrack ebtable_nat ip6table_nat ip6table_mangle ip6table_raw ip6table_security iptable_nat nf_nat iptable_mangle iptable_raw iptable_security nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 ip_set nfnetlink ebtable_filter ebtables ip6table_filter ip6_tables iptable_filter ip_tables sunrpc vfat fat intel_rapl_msr rmi_smbus iwlmvm rmi_core intel_rapl_common x86_pkg_temp_thermal intel_powerclamp coretemp mac80211 snd_hda_codec_realtek snd_hda_codec_generic snd_hda_codec_hdmi kvm_intel snd_hda_intel kvm snd_intel_nhlt snd_hda_codec snd_usb_audio irqbypass uvcvideo snd_hda_core snd_usbmidi_lib snd_rawmidi iTCO_wdt snd_hwdep libarc4 intel_cstate cdc_ether intel_uncore videobuf2_vmalloc iwlwifi mei_wdt mei_hdcp iTCO_vendor_support snd_seq videobuf2_memops usbnet videobuf2_v4l2 snd_seq_device
>>>    intel_rapl_perf pcspkr videobuf2_common joydev wmi_bmof snd_pcm cfg80211 r8152 videodev intel_pch_thermal i2c_i801 mii mc thinkpad_acpi snd_timer mei_me ledtrig_audio snd lpc_ich mei soundcore rfkill binfmt_misc xfs dm_thin_pool dm_persistent_data dm_bio_prison libcrc32c dm_crypt i915 i2c_algo_bit drm_kms_helper drm crct10dif_pclmul crc32_pclmul crc32c_intel ghash_clmulni_intel serio_raw wmi video
>>> CR2: 0000000000000102
>>> ---[ end trace 2ac747acabe218da ]---
>>> RIP: 0010:__queue_work+0x1f/0x3b0
>>> Code: eb df 66 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 41 57 49 89 f7 41 56 41 89 fe 41 55 41 89 fd 41 54 55 48 89 d5 53 48 83 ec 10 <f6> 86 02 01 00 00 01 0f 85 bc 02 00 00 49 bc eb 83 b5 80 46 86 c8
>>> RSP: 0018:ffffbef4884bbd58 EFLAGS: 00010082
>>> RAX: 0000000000000246 RBX: 0000000000000246 RCX: 0000000000000000
>>> RDX: ffff9903901f4460 RSI: 0000000000000000 RDI: 0000000000000040
>>> RBP: ffff9903901f4460 R08: ffff9903901fb040 R09: ffff990398614700
>>> R10: 0000000000000030 R11: 0000000000000000 R12: 0000000000000000
>>> R13: 0000000000000040 R14: 0000000000000040 R15: 0000000000000000
>>> FS:  00007f7d2a4e4a80(0000) GS:ffff9903a5a80000(0000) knlGS:0000000000000000
>>> CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>> CR2: 0000000000000102 CR3: 0000000203da8004 CR4: 00000000003606e0
>>> DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
>>> DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
>>>
>>> Unfortunately I don't have time to find the root cause.  What I've
>>> figured out so far is:
>>>
>>>     bool queue_work_on(int cpu, struct workqueue_struct *wq,
>>>                        struct work_struct *work)
>>>     {
>>>         bool ret = false;
>>>         unsigned long flags;
>>>
>>>         local_irq_save(flags);
>>>
>>>         if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) {
>>>                                                        ~~~~~~~~~~~~~~~~~~~~
>>>
>>> The address of work is 0x102 so this line causes a page fault when it
>>> tries to access the data field (offset 0).
>>>
>>> The caller provided the 0x102 pointer so let's see where it comes from:
>>>
>>>     static int __io_queue_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req,
>>>                               struct sqe_submit *s, bool force_nonblock)
>>>     {
>>>         ...
>>>         if (!io_add_to_prev_work(list, req)) {
>>>             if (list)
>>>                 atomic_inc(&list->cnt);
>>>             INIT_WORK(&req->work, io_sq_wq_submit_work);
>>>             io_queue_async_work(ctx, req);
>>> 	  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>>
>>> and queue_work() is called here:
>>>
>>>     static inline void io_queue_async_work(struct io_ring_ctx *ctx,
>>>                                            struct io_kiocb *req)
>>>     {
>>>         int rw = 0;
>>>
>>>         if (req->submit.sqe) {
>>>             switch (req->submit.sqe->opcode) {
>>>             case IORING_OP_WRITEV:
>>>             case IORING_OP_WRITE_FIXED:
>>>                 rw = !(req->rw.ki_flags & IOCB_DIRECT);
>>>                 break;
>>>             }
>>>         }
>>>
>>>         queue_work(ctx->sqo_wq[rw], &req->work);
>>>         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>>
>>> I must be missing something though because it seems impossible to get
>>> this far if req is NULL.  INIT_WORK() would have Oopsed already.  Also,
>>> offsetof(struct io_kiocb, work) is 0xa0 according to pahole(1) so we
>>> still haven't reached the 0x102 offset from the Oops report.
>>>
>>> Any ideas?
>>
>> This is new in 5.4-rc1?
> 
> I didn't hit it with 5.3, but I hit other issues so I'm not sure if this
> bug exists in older kernels.
> 
>> And how are you reproducing it?
> 
>    $ git clone -b io_uring https://github.com/stefanha/qemu
>    $ cd qemu
>    $ ./configure --target-list=x86_64-softmmu
>    $ make -j$(nproc)
>    $ (cd tests/qemu-iotests && ./check -i io_uring 052)
> 
> You can mount the file system of your choice at
> tests/qemu-iotests/scratch/ before running the test.
> 
> You can view the test case at tests/qemu-iotests/052.

Thanks, that's useful. Need to look closer into this, but seems wrong
that we're killing the workqueue for SCM_RIGHTS removal. We just need to
sync it. Does this work for you?


diff --git a/fs/io_uring.c b/fs/io_uring.c
index 8a0381f1a43b..a8755582c688 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -2920,8 +2920,12 @@ static void io_finish_async(struct io_ring_ctx *ctx)
 static void io_destruct_skb(struct sk_buff *skb)
 {
 	struct io_ring_ctx *ctx = skb->sk->sk_user_data;
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(ctx->sqo_wq); i++)
+		if (ctx->sqo_wq[i])
+			flush_workqueue(ctx->sqo_wq[i]);
 
-	io_finish_async(ctx);
 	unix_destruct_scm(skb);
 }
 

-- 
Jens Axboe