On Thu, Aug 15, 2019 at 02:26:03PM -0700, Andrew Morton wrote: > On Thu, 15 Aug 2019 09:34:34 -0700 "Darrick J. Wong" > <darrick.w...@oracle.com> wrote: > > > While active, the media backing a swap file is leased to the kernel. > > Userspace has no business writing to it. Make sure we can't do this. > > I don't think this tests the case where a file was already open for > writing and someone does swapon(that file)? > > And then does swapoff(that file), when writes should start working again? > > Ditto all the above, with s/open/mmap/.
Heh, ok. I'll start working on a C program to do that. > Do we handle (and test!) the case where there's unwritten dirty > pagecache at the time of swapon()? Ditto pte-dirty MAP_SHARED pages? One of the tests I wrote for iomap_swapfile_activate way back when checks that. The iomap version calls vfs_fsync, but AFAICT generic_swapfile_activate doesn't do that. --D
