On Wed, 2016-04-13 at 13:55 -0700, Stephen Hemminger wrote:
> On Tue, 12 Apr 2016 15:22:29 -0700
> Eric Dumazet <[email protected]> wrote:
>
> > From: Eric Dumazet <[email protected]>
> >
> > Lets prepare for a possibility to have 64bit inode numbers for sockets,
> > even if the kernel currently enforces 32bit numbers.
> >
> > Presumably, if both kernel and userland are 64bit (no 32bit emulation),
> > kernel could switch to 64bit inode numbers soon.
> >
> > Signed-off-by: Eric Dumazet <[email protected]>
>
> Why not use ino_t rather than __u64 which is really intended for kernel/user
> abi.
> Then make code generic on sizeof inode type
Well, maybe because code dealing with ino_t is very often buggy ;)
For example iproute2 makes the assumption an ino_t can be printed using
%lu format. This does not work with a 32bit binary.
static void bpf_info_loop(int *fds, struct bpf_map_aux *aux)
{
int i, tfd[BPF_MAP_ID_MAX];
printf("ver: %d\nobj: %s\ndev: %lu\nino: %lu\nmaps: %u\n",
aux->uds_ver, aux->obj_name, aux->obj_st.st_dev,
aux->obj_st.st_ino, aux->num_ent);
...
No strong preference from my side.
