Re: [PATCH] nvme: Enter string size calculation “subsysnqn”

[Christoph Hellwig]

On Thu, Dec 26, 2024 at 01:45:35PM +0300, Denis Arefev wrote:
> When memory is allocated, the size of the string 
> is calculated nvmet_subsys_alloc(...).

s/is/is in/ ?

> When memory was accessed, constant size was used.
> 
> Fixes: 95409e277d83 ("nvmet: implement unique discovery NQN")
> Reported-by: syzbot+a84181c81389771eb46a@xxxxxxxxxxxxxxxxxxxxxxxxx
> Closes: https://syzkaller.appspot.com/bug?extid=a84181c81389771eb46a
> Signed-off-by: Denis Arefev <arefev@xxxxxxxxx>
> ---
>  drivers/nvme/target/configfs.c | 10 ++++++----
>  1 file changed, 6 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/nvme/target/configfs.c b/drivers/nvme/target/configfs.c
> index eeee9e9b854c..2f74204d000e 100644
> --- a/drivers/nvme/target/configfs.c
> +++ b/drivers/nvme/target/configfs.c
> @@ -2247,14 +2247,15 @@ static struct config_group nvmet_hosts_group;
>  static ssize_t nvmet_root_discovery_nqn_show(struct config_item *item,
>  					     char *page)
>  {
> -	return snprintf(page, PAGE_SIZE, "%s\n", nvmet_disc_subsys->subsysnqn);
> +	return snprintf(page, strnlen(nvmet_disc_subsys->subsysnqn, PAGE_SIZE),
> +			"%s\n", nvmet_disc_subsys->subsysnqn);

This doesn't really make sense as %s already calculates the length
of the string internally.

>  }
>  
>  static ssize_t nvmet_root_discovery_nqn_store(struct config_item *item,
>  		const char *page, size_t count)
>  {
>  	struct list_head *entry;
> -	size_t len;
> +	size_t len, nqn_len;
>  
>  	len = strcspn(page, "\n");
>  	if (!len || len > NVMF_NQN_FIELD_LEN - 1)
> @@ -2271,8 +2272,9 @@ static ssize_t nvmet_root_discovery_nqn_store(struct config_item *item,
>  			return -EINVAL;
>  		}
>  	}
> -	memset(nvmet_disc_subsys->subsysnqn, 0, NVMF_NQN_FIELD_LEN);
> -	memcpy(nvmet_disc_subsys->subsysnqn, page, len);
> +	nqn_len = strnlen(nvmet_disc_subsys->subsysnqn, NVMF_NQN_SIZE);
> +	memset(nvmet_disc_subsys->subsysnqn, 0, nqn_len);
> +	memcpy(nvmet_disc_subsys->subsysnqn, page, nqn_len);

The memset should still happen for the entire field, only the copy
needs to me limited.

But i really wonder if we should do away with this mess and just
use kmemdup()ed allocations for the nqns.