On 6/19/22 1:25 AM, Christoph Hellwig wrote: > On Thu, Jun 16, 2022 at 10:04:37PM -0500, Mike Christie wrote: >> If NDOB is set we don't have a buffer. We will then crash when trying to >> access the t_data_sg. This has us allocate a page to use for the data >> buffer that gets passed to vfs_iter_write. > > But only due to the last patch, so this should be reordered before > that. I didn't get that part. You can hit this bug with what is upstream right now. You don't need patch 4. > > I also think this should just use ZERO_PAGE() or even better switch to > FALLOC_FL_ZERO_RANGE as a first pass. > Ok.
