On 29/05/2020 01:34, Al Viro wrote:
The way comedi compat ioctls are done is wrong. Instead of having ->compat_ioctl() copying the 32bit stuff in, then passing the kernel copies to helpers shared with native ->ioctl() and doing copyout with conversion if needed, it's playing silly buggers with creating a 64bit copy on user stack, then calling native ioctl (which copies that copy into the kernel), then fetching it from user stack, converting to 32bit variant and copying that to user. Extra headache for no good reason. And the single largest remaining pile of __put_user()/__get_user() this side of arch/*. IMO compat_alloc_user_space() should die...NOTE: this is only compile-tested - I simply don't have the hardware in question. Anyway, the branch lives in #uaccess.comedi, based at v5.7-rc1 Al Viro (10): comedi: move compat ioctl handling to native fops comedi: get rid of indirection via translated_ioctl() comedi: get rid of compat_alloc_user_space() mess in COMEDI_CHANINFO compat comedi: get rid of compat_alloc_user_space() mess in COMEDI_RANGEINFO compat comedi: get rid of compat_alloc_user_space() mess in COMEDI_INSN compat comedi: get rid of compat_alloc_user_space() mess in COMEDI_INSNLIST compat comedi: lift copy_from_user() into callers of __comedi_get_user_cmd() comedi: do_cmdtest_ioctl(): lift copyin/copyout into the caller comedi: do_cmd_ioctl(): lift copyin/copyout into the caller comedi: get rid of compat_alloc_user_space() mess in COMEDI_CMD{,TEST} compat
There is a bug in patch 05. Patch 10 doesn't seem to have been sent yet (I didn't receive it and I can't see it in the thread in the LKML archives). I've signed off on 01-04, 06-09.
These should be Cc'd to Greg KH and to de...@driverdev.osuosl.org. Cheers, Ian -- -=( Ian Abbott <abbo...@mev.co.uk> || Web: www.mev.co.uk )=- -=( MEV Ltd. is a company registered in England & Wales. )=- -=( Registered number: 02862268. Registered address: )=- -=( 15 West Park Road, Bramhall, STOCKPORT, SK7 3JZ, UK. )=-
