Re: [RFC][PATCH 0/4] Check O_* flags set with fcntl() on anon_inode files.

[Matt Helsley <matthltc@us.ibm.com>]

On Sun, Feb 14, 2010 at 03:52:05AM +0000, Al Viro wrote:
> On Sat, Feb 13, 2010 at 04:27:43PM -0800, Matt Helsley wrote:
> > anon_inode interfaces often do not support flags that can be set
> > by fcntl(). Right now using fcntl() to set these flags falsely
> > reports success for things like O_ASYNC yet SIGIO is not delivered.
> > 
> > I relied on the flags allowed by the syscalls that create
> > these files to determine the flags that are allowed to be set by
> > fcntl().
> > 
> > Each patch checks flags for one anonymous inode interface:
> > 
> > [PATCH 1/4] signalfd
> > [PATCH 2/4] timerfd
> > [PATCH 3/4] epoll
> > [PATCH 4/4] eventfd
> > 
> > I did not check the perf, kvm-vm, or kvm-vcpu uses of anon_inodes.
> 
> Er...  O_ASYNC is silently ignored for regular files as well, so any
> userland code that tries to rely on fcntl() rejecting it is and always
> had been badly b0rken.

Of course. Did you mean to imply that the kernel shouldn't bother to
reject these, or were you merely making an observation?

Cheers,
	-Matt Helsley