If we want to do the correct permission check, the ioctl handler needs to check R/W bit by itself based on how the ioctl is implemented. Or follow up how Linux encode the needed permission into each IOCTL: https://github.com/torvalds/linux/blob/master/include/uapi/asm-generic/ioctl.h#L85-L91 and let's VFS layer do the check for each driver.
On Fri, Jan 28, 2022 at 5:14 PM Petro Karashchenko < petro.karashche...@gmail.com> wrote: > Hello team, > > Recently I have noticed that there are many places in code where > register_driver() is called with non-zero mode with file operation > structures that have neither read nor write APIs implemented. For > example "ret = register_driver(path, &opamp_fops, 0444, dev);" while > opamp_fops has only "opamp_open", "opamp_close" and "opamp_ioctl" > implemented. I made a PR to fix it > https://github.com/apache/incubator-nuttx/pull/5347 and change mode > from "0444" to "0000", but want to ask if anyone sees any drawback in > such an approach? Maybe I'm missing something? > > Best regards, > Petro >
