On Thu, Dec 05, 2019 at 04:58:30PM +0000, fdmanana@xxxxxxxxxx wrote: > From: Filipe Manana <fdmanana@xxxxxxxx> [...] > V2: Fixed changelog because it stated in the first paragraph that the problem > affected only extents shared within the same file, but that's not true and > contradicted the last paragraph. So fix that and state explicitly that it > happens as well for extents shared between different files. > > V3: Drop existing checksums from the log during the fast fsync path as well. > This is also needed to make the test case pass on kernels < 5.4 because > on 5.4 the full sync bit is set on the inode whenever we clone, while > before 5.4 we don't set it. The tree checker patch for detecting overlapping > checksum items detected this during a fast fsync on generic/561 while > running it in a loop for several hours. Patch replaced, thanks.
