Re: [RFC net-next 00/28] ndo_ioctl rework

[Arnd Bergmann]

On Sun, Nov 8, 2020 at 1:06 AM Jakub Kicinski <kuba@xxxxxxxxxx> wrote:
>
> On Fri,  6 Nov 2020 23:17:15 +0100 Arnd Bergmann wrote:
> > Any suggestions on how to proceed? I think the ndo_siocdevprivate
> > change is the most interesting here, and I would like to get
> > that merged.
>
> Splitting out / eliminating ioctl pass-thry in general seems like
> a nice clean up. How did you get the ndo_eth_ioctl patch done, was
> it manual work?

I had done a fairly detailed manual search through all drivers
with a .ndo_do_ioctl function to find the SIOCDEVPRIVATE users,
based on a simple shell script. After I had converted all of those,
I realized that 80% of the remaining ndo_do_ioctl users were
in drivers/net/ethernet, implementing the same five commands,
so I did a 'git grep -wl ndo_do_ioctl drivers/net/ethernet/ |
sed -i "s:ndo_do_ioctl:ndo_eth_ioctl"' and converted the rest
by hand.

FWIW, this is what remains afterwards:

$ git grep -w ndo_do_ioctl
Documentation/networking/netdevices.rst:ndo_do_ioctl:
drivers/char/pcmcia/synclink_cs.c:      .ndo_do_ioctl   = hdlcdev_ioctl,
drivers/net/appletalk/cops.c:        .ndo_do_ioctl           = cops_ioctl,
drivers/net/appletalk/ltpc.c:   .ndo_do_ioctl           = ltpc_ioctl,
drivers/net/bonding/bond_main.c:        .ndo_do_ioctl           = bond_do_ioctl,
drivers/net/wan/c101.c: .ndo_do_ioctl   = c101_ioctl,
drivers/net/wan/cosa.c: .ndo_do_ioctl   = cosa_net_ioctl,
drivers/net/wan/farsync.c:      .ndo_do_ioctl   = fst_ioctl,
drivers/net/wan/fsl_ucc_hdlc.c: .ndo_do_ioctl   = uhdlc_ioctl,
drivers/net/wan/hdlc_fr.c:      .ndo_do_ioctl   = pvc_ioctl,
drivers/net/wan/hostess_sv11.c: .ndo_do_ioctl   = hostess_ioctl,
drivers/net/wan/ixp4xx_hss.c:   .ndo_do_ioctl   = hss_hdlc_ioctl,
drivers/net/wan/lmc/lmc_main.c: .ndo_do_ioctl   = lmc_ioctl,
drivers/net/wan/n2.c:   .ndo_do_ioctl   = n2_ioctl,
drivers/net/wan/pc300too.c:     .ndo_do_ioctl   = pc300_ioctl,
drivers/net/wan/pci200syn.c:    .ndo_do_ioctl   = pci200_ioctl,
drivers/net/wan/sealevel.c:     .ndo_do_ioctl   = sealevel_ioctl,
drivers/net/wan/wanxl.c:        .ndo_do_ioctl   = wanxl_ioctl,
drivers/tty/synclink.c: .ndo_do_ioctl   = hdlcdev_ioctl,
drivers/tty/synclink_gt.c:      .ndo_do_ioctl   = hdlcdev_ioctl,
drivers/tty/synclinkmp.c:       .ndo_do_ioctl   = hdlcdev_ioctl,
include/linux/netdevice.h: * int (*ndo_do_ioctl)(struct net_device
*dev, struct ifreq *ifr, int cmd);
include/linux/netdevice.h:      int
(*ndo_do_ioctl)(struct net_device *dev,
net/appletalk/aarp.c:   if (!(ops->ndo_do_ioctl(iface->dev, &atreq,
SIOCSIFADDR))) {
net/appletalk/aarp.c:           ops->ndo_do_ioctl(iface->dev, &atreq,
SIOCGIFADDR);
net/bridge/br_device.c: .ndo_do_ioctl            = br_dev_ioctl,
net/core/dev_ioctl.c:   if (ops->ndo_do_ioctl) {
net/core/dev_ioctl.c:                   return ops->ndo_do_ioctl(dev, ifr, cmd);
net/ieee802154/socket.c:        if (dev->type == ARPHRD_IEEE802154 &&
dev->netdev_ops->ndo_do_ioctl)
net/ieee802154/socket.c:                ret =
dev->netdev_ops->ndo_do_ioctl(dev, &ifr, cmd);
net/mac802154/iface.c:  .ndo_do_ioctl           = mac802154_wpan_ioctl,

> > For the wireless drivers, removing the old drivers
> > instead of just the dead code might be an alternative, depending
> > on whether anyone thinks there might still be users.
>
> Dunno if you want to dig into removal with a series like this,
> anything using ioctls will be pretty old (with the exception
> of what you separated into ndo_eth_ioctl). You may get bogged
> down.

Ok

      Arnd