[PATCH V2 3/3] nvmet-tcp: fix memory leak when performing a controller reset

Sagi Grimberg sagi at grimberg.me
Tue Nov 16 02:11:13 PST 2021


>>> If a reset controller is executed while the initiator
>>> is performing some I/O the driver may leak the memory allocated
>>> for the commands' iovec.
>>>
>>> Make sure that nvmet_tcp_uninit_data_in_cmds() releases
>>> all the memory.
>>>
>>> Signed-off-by: Maurizio Lombardi <mlombard at redhat.com>
>>> ---
>>>    drivers/nvme/target/tcp.c | 5 ++++-
>>>    1 file changed, 4 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/nvme/target/tcp.c b/drivers/nvme/target/tcp.c
>>> index 028b47ff0f0f..5a238fa067cd 100644
>>> --- a/drivers/nvme/target/tcp.c
>>> +++ b/drivers/nvme/target/tcp.c
>>> @@ -1427,7 +1427,10 @@ 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_tcp_finish_cmd(cmd);
>>> +			nvmet_req_uninit(&cmd->req);
>>> +
>>> +		nvmet_tcp_unmap_pdu_iovec(cmd);
>>> +		nvmet_tcp_free_iovec(cmd);
>>
>> I understand how unmap_pdu_iovec is re-entrance safe, but sgl_free
>> called from free_iovec is not. What prevents it from being called
>> for commands that have already freed their sgls?
> 
> The commands that have already freed their sgls
> will have the pointers set to NULL,
> calling sgl_free() against a NULL pointer is safe (it does nothing, like kfree())
> but if you prefer I will add a check.

Oh, right.

So,
Reviewed-by: Sagi Grimberg <sagi at grimberg.me>



More information about the Linux-nvme mailing list