Re: [PATCH v2 2/2] mm: adds NOSIGBUS extension to mmap()

O Mon, Jun 28, 2021 at 10:27:23AM -0400, Vivek Goyal wrote:

On Fri, Jun 04, 2021 at 12:43:22AM -0700, Ming Lin wrote:

quoted

Adds new flag MAP_NOSIGBUS of mmap() to specify the behavior of
"don't SIGBUS on fault". Right now, this flag is only allowed
for private mapping.

For MAP_NOSIGBUS mapping, map in the zero page on read fault
or fill a freshly allocated page with zeroes on write fault.

I am wondering if this could be of limited use for me if MAP_NOSIGBUS
were to be supported for shared mappings as well.

V1 did support shared mapping.
https://lkml.org/lkml/2021/6/1/1078

And V0 even supported unmapping the zero page for later write.
https://github.com/minggr/linux/commit/77f3722b94ff33cafe0a72c1bf1b8fa374adb29f

We may support shared mapping if there is a real use case.
As Hugh mentioned:

And by restricting to MAP_PRIVATE, you would allow for adding a
proper MAP_SHARED implementation later, if it's thought useful
(that being the implementation which can subsequently unmap a
zero page to let new page cache be mapped).

See https://lkml.org/lkml/2021/6/1/1258

Ming

When virtiofs is run with dax enabled, then it is possible that if
a file is shared between two guests, then one guest truncates the
file and second guest tries to do load/store operation. Given current
kvm architecture, there is no mechanism to propagate SIGBUS to guest
process, instead KVM retries page fault infinitely and guest cpu/process
hangs.

Ideally we want this error to propagate all the way back into the
guest and to the guest process but that solution is not in place yet.

https://lore.kernel.org/kvm/20200406190951.GA19259@redhat.com/ (local)

In the absense of a proper solution, one could think of mapping
shared file on host with MAP_NOSIGBUS, and hopefully that means
kvm will be able to resolve fault to a zero filled page and guest
will not hang. But this means that data sharing between two processes
is now broken. Writes by process A will not be visible to process B
in another once this situation happens, IIUC.

So if we were to MAP_NOSIGBUS, guest will not hang but failures resulting
from ftruncate will be silent and will be noticed sometime later. I guess
not exactly a very pleasant scenario...

Thanks
Vivek