On Fri, Dec 3, 2021 at 8:20 PM Stan Hu <stanhu@xxxxxxxxx> wrote: > > A number of users have reported that under certain conditions using > Docker and overlay2, copy_file_range() can unexpectedly create a > 0-byte file: In https://github.com/docker/for-linux/issues/1015. > > We started seeing mysterious failures in our CI tests as a result of > files not properly being copied. > > https://github.com/docker/for-linux/issues/1015#issuecomment-841915668 > has a sample reproduction test. > > I analyzed the diff between 5.10 and 5.11 and found that if I applied > the following kernel patch, the reproduction test passes: > > https://lore.kernel.org/linux-fsdevel/20201207163255.564116-6-mszeredi@xxxxxxxxxx/#t > > This landed in this merge commit and this commit: > > 1. https://github.com/torvalds/linux/commit/92dbc9dedccb9759c7f9f2f0ae6242396376988f > 2. https://github.com/torvalds/linux/commit/82a763e61e2b601309d696d4fa514c77d64ee1be > > Could this patch be backported for kernels 5.6 to 5.10? This problem is a regression from bug (missing update size of dest file) in commit 1a980b8cbf00 ("ovl: add splice file read write helper") The commit you want to backport fixes that problem, but introduces another regression later fixed by commit 9b91b6b019fd ("ovl: fix deadlock in splice write") So you'd need to backport both to end up with the correct implementation of ovl_splice_write() Thanks, Amir.
