Re: [PATCH RESEND] nbd: avoid losing pointer to reallocated config->socks in nbd_add_socket

[Xiubo Li]

On 2019/9/21 0:06, Eugene Syromiatnikov wrote:
> In the (very unlikely) case of config->socks reallocation success
> and nsock allocation failure config->nsock will not get updated
> with the new pointer to socks array. Fix it by updating config->socks
> right after reallocation successfulness check.
>
> Fixes: 9561a7ade0c2 ("nbd: add multi-connection support")
> Signed-off-by: Eugene Syromiatnikov <esyr@xxxxxxxxxx>
> Cc: stable@xxxxxxxxxxxxxxx # 4.10+
> ---
>   drivers/block/nbd.c | 4 ++--
>   1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/block/nbd.c b/drivers/block/nbd.c
> index a8e3815..a04c686 100644
> --- a/drivers/block/nbd.c
> +++ b/drivers/block/nbd.c
> @@ -987,14 +987,14 @@ static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
>   		sockfd_put(sock);
>   		return -ENOMEM;
>   	}
> +	config->socks = socks;
> +
>   	nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL);
>   	if (!nsock) {
>   		sockfd_put(sock);
>   		return -ENOMEM;
>   	}
>   
> -	config->socks = socks;
> -

This makes sense.

If the socks allocating successes, then the old config->socks will be 
freed by krealloc() and return the new one, but if the nsock allocating 
fails, the config->socks will hold the released memory, which may cause 
the kernel crash.

Thanks

BRs



>   	nsock->fallback_index = -1;
>   	nsock->dead = false;
>   	mutex_init(&nsock->tx_lock);