On 3/25/19 10:19 AM, James Bottomley wrote:
> On Mon, 2019-03-25 at 15:34 +0100, Arnd Bergmann wrote:
>> On big-endian architectures, the signal masks are differnet
>> between 32-bit and 64-bit tasks, so we have to use a different
>> function for reading them from user space.
>>
>> io_cqring_wait() initially got this wrong, and always interprets
>> this as a native structure. This is ok on x86 and most arm64,
>> but not on s390, ppc64be, mips64be, sparc64 and parisc.
>>
>> Signed-off-by: Arnd Bergmann <arnd@xxxxxxxx>
>> ---
>> fs/io_uring.c | 10 +++++++++-
>> 1 file changed, 9 insertions(+), 1 deletion(-)
>>
>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>> index 6aaa30580a2b..8f48d29abf76 100644
>> --- a/fs/io_uring.c
>> +++ b/fs/io_uring.c
>> @@ -1968,7 +1968,15 @@ static int io_cqring_wait(struct io_ring_ctx
>> *ctx, int min_events,
>> return 0;
>>
>> if (sig) {
>> - ret = set_user_sigmask(sig, &ksigmask, &sigsaved,
>> sigsz);
>> +#ifdef CONFIG_COMPAT
>> + if (in_compat_syscall())
>> + ret = set_compat_user_sigmask((const
>> compat_sigset_t __user *)sig,
>> + &ksigmask,
>> &sigsaved, sigsz);
>> + else
>> +#endif
>
> This looks a bit suboptimal: shouldn't in_compat_syscall() be hard
> coded to return 0 if CONFIG_COMPAT isn't defined? That way the
> compiler can do the correct optimization and we don't have to litter
> #ifdefs and worry about undefined variables and other things.
That requires the types to be valid for !CONFIG_COMPAT, as well as the
sigmask helper.
--
Jens Axboe
