[External] Re: [PATCH] nvme-rdma: handle nvme completion data length

[zhenwei pi]

On 10/22/20 5:55 PM, Chao Leng wrote:
> 
> 
> On 2020/10/22 16:38, zhenwei pi wrote:
>> Hit a kernel warning:
>> refcount_t: underflow; use-after-free.
>> WARNING: CPU: 0 PID: 0 at lib/refcount.c:28
>>
>> RIP: 0010:refcount_warn_saturate+0xd9/0xe0
>> Call Trace:
>>   <IRQ>
>>   nvme_rdma_recv_done+0xf3/0x280 [nvme_rdma]
>>   __ib_process_cq+0x76/0x150 [ib_core]
>>   ...
>>
>> The reason is that a zero bytes message received from target, and the
>> host side continues to process without length checking, then the
>> previous CQE is processed twice.
>>
>> Handle data length, ignore zero bytes message, and try to recovery for
>> corrupted CQE case.
>>
>> Signed-off-by: zhenwei pi <pizhenwei at bytedance.com>
>> ---
>>   drivers/nvme/host/rdma.c | 11 +++++++++++
>>   1 file changed, 11 insertions(+)
>>
>> diff --git a/drivers/nvme/host/rdma.c b/drivers/nvme/host/rdma.c
>> index 9e378d0a0c01..9f5112040d43 100644
>> --- a/drivers/nvme/host/rdma.c
>> +++ b/drivers/nvme/host/rdma.c
>> @@ -1767,6 +1767,17 @@ static void nvme_rdma_recv_done(struct ib_cq 
>> *cq, struct ib_wc *wc)
>>           return;
>>       }
>> +    if (unlikely(!wc->byte_len)) {
>> +        /* zero bytes message could be ignored */
>> +        return;
>> +    } else if (unlikely(wc->byte_len < len)) {
>> +        /* Corrupted completion, try to recovry */
>> +        dev_err(queue->ctrl->ctrl.device,
>> +            "Unexpected nvme completion length(%d)\n", wc->byte_len);
>> +        nvme_rdma_error_recovery(queue->ctrl);
>> +        return;
>> +    }
> !wc->byte_len and wc->byte_len < len may be the same type of anomaly.
> Why do different error handling?
> In which scenario zero bytes message received from target? fault inject 
> test or normal test/run?

Zero bytes message could be used as transport layer keep alive mechanism 
(I's also developing target side transport layer keep alive now. To 
reclaim resource, target side needs to close dead connections even kato 
is set as 0).

>> +
>>       ib_dma_sync_single_for_cpu(ibdev, qe->dma, len, DMA_FROM_DEVICE);
>>       /*
>>        * AEN requests are special as they don't time out and can
>>

-- 
zhenwei pi