Re: [4.19.y] 32-bit overflow in __blkdev_issue_discard()

[Greg KH <greg@kroah.com>]

On Thu, Jan 30, 2020 at 03:10:42PM +0300, Konstantin Khlebnikov wrote:
> Please consider including into 4.19 upstream commits
> 
> ba5d73851e71847ba7f7f4c27a1a6e1f5ab91c79
> ("block: cleanup __blkdev_issue_discard()")
> 
> and
> 
> 4800bf7bc8c725e955fcbc6191cc872f43f506d3
> ("block: fix 32 bit overflow in __blkdev_issue_discard()")

THis patch does not apply to the 4.19 tree :(

> Overflow of unsigned long "req_sects" (fixed in second patch)
> actually exist here much longer.
> 
> And 4.19 commit 744889b7cbb56a64f957e65ade7cb65fe3f35714
> ("block: don't deal with discard limit in blkdev_issue_discard()")
> make it worse by replacing
> 
> req_sects = min_t(sector_t, nr_sects, q->limits.max_discard_sectors);
> 
> with
> 
> unsigned int req_sects = nr_sects;
> 
> 
> because now discard length isn't cut by max_discard_sectors it easily overflows.
> As a result BLKDISCARD fails unexpectedly:
> 
> ioctl(3, BLKDISCARD, [0, 0x20000000000])  = -1 EOPNOTSUPP (Operation not supported)

I don't understand.  Can you provide backported and working patches for
the 4.19.y series so that I can apply them that way to show exactly what
you have changed here?

thanks,

greg k-h