Re: [PATCH v2] block: Improve limiting the bio size

[Changheun Lee <nanich.lee@samsung.com>]

> On 4/26/21 9:12 PM, Changheun Lee wrote:
> > Actually I checked "bio->bi_disk" at first as below. It works well.
> 
> Agreed that the bi_disk pointer needs to be checked.
> 
> > I think "bi_bdev->bd_disk" checking might be needed too.
> 
> bio_max_size() occurs in the hot path. Since if-tests have a runtime
> cost I'd like to keep the number of if-tests inside bio_max_size() to a
> minimum.
> 
> I only found one bd_disk assignment in the kernel tree, namely inside
> bdev_alloc(). The two bdev_alloc() calls I found inside the kernel tree
> pass a valid disk pointer to bdev_alloc(). So I think it is not
> necessary to check the disk pointer inside bio_max_size().
> 
> Thanks,
> 
> Bart.

I think bio->bi_bdev should be null in bio_copy_kern(), or bio_map_kern()
logically. Because it is simple buffer mapping in driver layer. Actually
bio->bi_bdev is null in my test environment. So bio->bi_bdev check only is
enough as Bart's patch.

Actually I don't know why NULL pointer dereference is occurred with Bart's
patch in blk_rq_map_kern(). And same problem have not occured yet in my
test environment with Bart's patch.
Maybe I missed something, or missunderstood?

Thanks,

Changheun Lee.