Re: [PATCH] fs: proc: use down_read_killable in proc_pid_cmdline_read()

[Yang Shi]

On 2/23/18 11:45 AM, Alexey Dobriyan wrote:
> On Fri, Feb 23, 2018 at 11:42:34AM -0800, Yang Shi wrote:
> > On 2/23/18 11:33 AM, Alexey Dobriyan wrote:
> > > On Wed, Feb 21, 2018 at 03:13:10PM -0800, Yang Shi wrote:
> > >
> > > > > > > 2) access_remote_vm() et al will do the same ->mmap_sem, and
> > > > > > Yes, it does. But, __access_remote_vm() is called by access_process_vm()
> > > > > > too, which is used by much more places, i.e. ptrace, so I was not sure
> > > > > > if it is preferred to convert to killable version. So, I leave it untouched.
> > > > > Yeah, but ->mmap_sem is taken 3 times per /proc/*/cmdline read
> > > > > and your scalability tests should trigger next backtrace right away.
> > > > Yes, however, I didn't run into it if mmap_sem is acquired earlier.
> > > >
> > > > How about defining a killable version, like
> > > > __access_remote_vm_killable() which use down_read_killable(), then the
> > > > killable version can be used by proc/*/cmdline? There might be other
> > > > users in the future.
> > > It would be a disaster as interfaces multiply.
> > Might be not that bad.
> Maybe.
>
> But you need to explain why there is no backtrace several lines later:
>
> 	access_remote_vm
> 	__access_remote_vm
> 	down_read(&mm->mmap_sem)

I think it might be because:

        CPU A                                  CPU B
                                         read /proc/*/cmdline
                                         get_mm
acquire mmap_sem
munmap(300G)                 try to acquire mmap_sem --> go to sleep
release mmap_sem
                                          got mmap_sem
                                          release mmap_sem

                                          access_remote_vm
                                          put_mm


The munmap might happen right before access_remote_vm(), but I just 
didn't run into it for the time being. It may be hit on another machine 
or with some changes to the test cases.

BTW, even the hung I met happened occassionally, not very often. So, the 
access_remote_vm() hit sounds less often. But, I agree it is still 
possible in theory.

Regards,
Yang