On 11/1/21 5:50 PM, Linus Torvalds wrote: > On Mon, Nov 1, 2021 at 4:20 PM Jens Axboe <axboe@xxxxxxxxx> wrote: >> >> Yes, probably safer just to make bdev_nr_bytes() return sector_t as >> well, even if loff_t isn't strictly wrong. > > Well, that would actually change the sign of some of the existing > comparisons. Possibly changing their meaning entirely.. > > So having 'loff_t' being signed may be an odd choice for a byte size, > but it is what it is. At least the current set of cleanups seemed to > keep the type logic the same when it changed i_size_read() to be > bdev_nr_bytes() instead. > > Changing it to 'sector_t' not only doesn't make conceptual sense when > it's a byte count, it might also be dangerous. > > So my reaction was really that it wasn't obvious that bdev_nr_bytes() > did the shift in the right type.. It does happen to do that, but > historically sector_t was the smaller type. OK, I misunderstood your original email, as per Christoph's email as well. May be worth adding loff_t cast, if for nothing else just to have it stick out to the next person touching it. -- Jens Axboe
