[RESEND for nvme-6.1] nvmet-tcp: Fix NULL pointer dereference during release

[Sagi Grimberg]

On 9/20/22 16:16, zhenwei pi wrote:
> nvmet-tcp frees CMD buffers in nvmet_tcp_uninit_data_in_cmds(),
> and waits the inflight IO requests in nvmet_sq_destroy(). During wait
> the inflight IO requests, the callback nvmet_tcp_queue_response()
> is called from backend after IO complete, this leads a typical
> Use-After-Free issue like this:
> 
>   BUG: kernel NULL pointer dereference, address: 0000000000000008
>   #PF: supervisor read access in kernel mode
>   #PF: error_code(0x0000) - not-present page
>   PGD 107f80067 P4D 107f80067 PUD 10789e067 PMD 0
>   Oops: 0000 [#1] PREEMPT SMP NOPTI
>   CPU: 1 PID: 123 Comm: kworker/1:1H Kdump: loaded Tainted: G            E      6.0.0-rc2.bm.1-amd64 #15
>   Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014
>   Workqueue: nvmet_tcp_wq nvmet_tcp_io_work [nvmet_tcp]
>   RIP: 0010:shash_ahash_digest+0x2b/0x110
>   Code: 1f 44 00 00 41 57 41 56 41 55 41 54 55 48 89 fd 53 48 89 f3 48 83 ec 08 44 8b 67 30 45 85 e4 74 1c 48 8b 57 38 b8 00 10 00 00 <44> 8b 7a 08 44 29 f8 39 42 0c 0f 46 42 0c 41 39 c4 76 43 48 8b 03
>   RSP: 0018:ffffc9000051bdd8 EFLAGS: 00010206
>   RAX: 0000000000001000 RBX: ffff888100ab5470 RCX: 0000000000000000
>   RDX: 0000000000000000 RSI: ffff888100ab5470 RDI: ffff888100ab5420
>   RBP: ffff888100ab5420 R08: ffff8881024d08c8 R09: ffff888103e1b4b8
>   R10: 8080808080808080 R11: 0000000000000000 R12: 0000000000001000
>   R13: 0000000000000000 R14: ffff88813412bd4c R15: ffff8881024d0800
>   FS:  0000000000000000(0000) GS:ffff88883fa40000(0000) knlGS:0000000000000000
>   CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>   CR2: 0000000000000008 CR3: 0000000104b48000 CR4: 0000000000350ee0
>   Call Trace:
>    <TASK>
>    nvmet_tcp_io_work+0xa52/0xb52 [nvmet_tcp]
>    ? __switch_to+0x106/0x420
>    process_one_work+0x1ae/0x380
>    ? process_one_work+0x380/0x380
>    worker_thread+0x30/0x360
>    ? process_one_work+0x380/0x380
>    kthread+0xe6/0x110
>    ? kthread_complete_and_exit+0x20/0x20
>    ret_from_fork+0x1f/0x30
> 
> Suggested by Sagi, separate nvmet_tcp_uninit_data_in_cmds() into two
> steps:
>    uninit data in cmds                  <- new step 1
>    nvmet_sq_destroy();
>    cancel_work_sync(&queue->io_work);
>    free CMD buffers                     <- new step 2
> 
> Signed-off-by: zhenwei pi <pizhenwei at bytedance.com>
> ---
>   drivers/nvme/target/tcp.c | 19 ++++++++++++++++---
>   1 file changed, 16 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/nvme/target/tcp.c b/drivers/nvme/target/tcp.c
> index c07de4f4f719..70baeab6af30 100644
> --- a/drivers/nvme/target/tcp.c
> +++ b/drivers/nvme/target/tcp.c
> @@ -1406,14 +1406,26 @@ static void nvmet_tcp_uninit_data_in_cmds(struct nvmet_tcp_queue *queue)
>   	for (i = 0; i < queue->nr_cmds; i++, cmd++) {
>   		if (nvmet_tcp_need_data_in(cmd))
>   			nvmet_req_uninit(&cmd->req);
> -
> -		nvmet_tcp_free_cmd_buffers(cmd);
>   	}
>   
>   	if (!queue->nr_cmds && nvmet_tcp_need_data_in(&queue->connect)) {
>   		/* failed in connect */
> -		nvmet_tcp_finish_cmd(&queue->connect);

I think that nvmet_tcp_finish_cmd is now redundant and can be removed
if we fold it to its single remaining call-site.

Other than that,
Reviewed-by: Sagi Grimberg <sagi at grimberg.me>