On Wed, Feb 19, 2020 at 5:36 PM Miklos Szeredi <miklos@xxxxxxxxxx> wrote: > > On Wed, Feb 19, 2020 at 4:28 PM Amir Goldstein <amir73il@xxxxxxxxx> wrote: > > > > On Wed, Feb 19, 2020 at 4:25 PM Miklos Szeredi <miklos@xxxxxxxxxx> wrote: > > > > While this makes sense on 64bit arch, it's going to overflow on 32bit > > > (due to i_ino being "unsigned long"). > > > > It's not clear here, but on 32bit, xinobits is 0: > > > > ofs->xino_mode = BITS_PER_LONG - 32; > > > > To the expression doesn't change i_ino. > > Correct? > > Want me to clarify that by comment or by code? > > Ah, missed that. I think no need to clarify further. > Mmm.. only it doesn't seem to be true. Seems like with xino=on xino_mode won't be 0 on 32bit. I think we do want to force disable xino on 32bit and send fix to stable - if only to keep the code in ovl_map_ino() simpler. What's more, I think there is a bug with xino_mode - it is not initialized to -1 on the default xino=off mode, so ovl_same_fs() could be wrong. I will try to look at this tomorrow. Thanks, Amir.