Re: [PATCH v5 02/11] block: Block Device Filtering Mechanism

[Yu Kuai <yukuai1@xxxxxxxxxxxxxxx>]

Hi,

在 2023/07/18 1:39, Sergei Shtepa 写道:
> Hi.
>
> On 7/12/23 14:34, Yu Kuai wrote:
> > Subject:
> > Re: [PATCH v5 02/11] block: Block Device Filtering Mechanism
> > From:
> > Yu Kuai <yukuai1@xxxxxxxxxxxxxxx>
> > Date:
> > 7/12/23, 14:34
> >
> > To:
> > Yu Kuai <yukuai1@xxxxxxxxxxxxxxx>, Sergei Shtepa <sergei.shtepa@xxxxxxxxx>, axboe@xxxxxxxxx, hch@xxxxxxxxxxxxx, corbet@xxxxxxx, snitzer@xxxxxxxxxx
> > CC:
> > viro@xxxxxxxxxxxxxxxxxx, brauner@xxxxxxxxxx, dchinner@xxxxxxxxxx, willy@xxxxxxxxxxxxx, dlemoal@xxxxxxxxxx, linux@xxxxxxxxxxxxxx, jack@xxxxxxx, ming.lei@xxxxxxxxxx, linux-block@xxxxxxxxxxxxxxx, linux-doc@xxxxxxxxxxxxxxx, linux-kernel@xxxxxxxxxxxxxxx, linux-fsdevel@xxxxxxxxxxxxxxx, Donald Buczek <buczek@xxxxxxxxxxxxx>, "yukuai (C)" <yukuai3@xxxxxxxxxx>
> >
> >
> > Hi,
> >
> > 在 2023/07/12 18:04, Yu Kuai 写道:
> > > Hi,
> > >
> > > 在 2023/07/11 10:02, Yu Kuai 写道:
> > >
> > > > > +static bool submit_bio_filter(struct bio *bio)
> > > > > +{
> > > > > +    if (bio_flagged(bio, BIO_FILTERED))
> > > > > +        return false;
> > > > > +
> > > > > +    bio_set_flag(bio, BIO_FILTERED);
> > > > > +    return bio->bi_bdev->bd_filter->ops->submit_bio(bio);
> > > > > +}
> > > > > +
> > > > >    static void __submit_bio(struct bio *bio)
> > > > >    {
> > > > > +    /*
> > > > > +     * If there is a filter driver attached, check if the BIO needs to go to
> > > > > +     * the filter driver first, which can then pass on the bio or consume it.
> > > > > +     */
> > > > > +    if (bio->bi_bdev->bd_filter && submit_bio_filter(bio))
> > > > > +        return;
> > > > > +
> > > > >        if (unlikely(!blk_crypto_bio_prep(&bio)))
> > > > >            return;
> > >
> > > ...
> > >
> > > > > +static void __blkfilter_detach(struct block_device *bdev)
> > > > > +{
> > > > > +    struct blkfilter *flt = bdev->bd_filter;
> > > > > +    const struct blkfilter_operations *ops = flt->ops;
> > > > > +
> > > > > +    bdev->bd_filter = NULL;
> > > > > +    ops->detach(flt);
> > > > > +    module_put(ops->owner);
> > > > > +}
> > > > > +
> > > > > +void blkfilter_detach(struct block_device *bdev)
> > > > > +{
> > > > > +    if (bdev->bd_filter) {
> > > > > +        blk_mq_freeze_queue(bdev->bd_queue);
> > >
> > > And this is not sate as well, for bio-based device, q_usage_counter is
> > > not grabbed while submit_bio_filter() is called, hence there is a risk
> > > of uaf from submit_bio_filter().
> >
> > And there is another question, can blkfilter_detach() from
> > del_gendisk/delete_partiton and ioctl concurrent? I think it's a
> > problem.
>
> Yes, it looks like if two threads execute the blkfilter_detach() function,
> then a problem is possible. The blk_mq_freeze_queue() function does not
> block threads.
> But for this, it is necessary that the IOCTL for the block device and
> its removal are performed simultaneously. Is this possible?

I think it's possible, ioctl only requires to open the disk/partition,
and it won't block disk removal:

t1:			t2:

open dev
ioctl
			remove dev
			 del_gendisk
 blkfilter_detach	  blkfilter_detach

> I suppose that using mutex bdev->bd_disk->open_mutex in
> blkfilter_ioctl_attach(), blkfilter_ioctl_detach() and
> blkfilter_ioctl_ctl() can fix the problem. What do you think?

I think it's ok, and blkfilter ioctl must check disk_live() while
holding the lock.

Thanks,
Kuai

> > Thanks,
> > Kuai
> > > Thanks,
> > > Kuai
> > >
> > > .
> .