Re: [PATCH v4 2/3] rust: block: add rnull, Rust null_blk implementation

[Andreas Hindborg <nmi@metaspace.dk>]

Andreas Hindborg <nmi@metaspace.dk> writes:

> Keith Busch <kbusch@kernel.org> writes:
>
>> On Sat, Jun 01, 2024 at 05:36:20PM +0200, Andreas Hindborg wrote:
>>> Keith Busch <kbusch@kernel.org> writes:
>>> 
>>> > On Sat, Jun 01, 2024 at 03:40:04PM +0200, Andreas Hindborg wrote:
>>> >> +impl kernel::Module for NullBlkModule {
>>> >> +    fn init(_module: &'static ThisModule) -> Result<Self> {
>>> >> +        pr_info!("Rust null_blk loaded\n");
>>> >> +        let tagset = Arc::pin_init(TagSet::try_new(1, 256, 1), flags::GFP_KERNEL)?;
>>> >> +
>>> >> +        let disk = {
>>> >> +            let block_size: u16 = 4096;
>>> >> +            if block_size % 512 != 0 || !(512..=4096).contains(&block_size) {
>>> >> +                return Err(kernel::error::code::EINVAL);
>>> >> +            }
>>> >
>>> > You've set block_size to the literal 4096, then validate its value
>>> > immediately after? Am I missing some way this could ever be invalid?
>>> 
>>> Good catch. It is because I have a patch in the outbound queue that allows setting
>>> the block size via a module parameter. The module parameter patch is not
>>> upstream yet. Once I have that up, I will send the patch with the block
>>> size config.
>>> 
>>> Do you think it is OK to have this redundancy? It would only be for a
>>> few cycles.
>>
>> It's fine, just wondering why it's there. But it also allows values like
>> 1536 and 3584, which are not valid block sizes, so I think you want the
>> check to be:
>>
>> 	if !(512..=4096).contains(&block_size) || ((block_size & (block_size - 1)) != 0)
>
> Right, that makes sense. I modeled it after the C null_blk validation
> code in `null_validate_conf`. It contains this:
>
> 	dev->blocksize = round_down(dev->blocksize, 512);
> 	dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
>
> That would have the same semantics, right? I guess I'll try to make a
> device with a 1536 block size and see what happens.

This happens:

root@debian:~# insmod /mnt/linux-build/drivers/block/null_blk/null_blk.ko bs=1536
BUG: kernel NULL pointer dereference, address: 0000000000000000
#PF: supervisor write access in kernel mode
#PF: error_code(0x0002) - not-present page
PGD 0 P4D 0
Oops: Oops: 0002 [#1] SMP
CPU: 2 PID: 291 Comm: insmod Not tainted 6.10.0-rc1+ #839

Probably a good idea with a better check.

BR Andreas