On 6/3/25 5:49 PM, Keith Busch wrote: > On Tue, Jun 03, 2025 at 12:54:05PM -0600, Jens Axboe wrote: >> On 6/3/25 12:47 PM, Caleb Sander Mateos wrote: >>> blk_rq_integrity_map_user() creates the ubuf iter with ITER_DEST for >>> write-direction operations and ITER_SOURCE for read-direction ones. >>> This is backwards; writes use the user buffer as a source for metadata >>> and reads use it as a destination. Switch to the rq_data_dir() helper, >>> which maps writes to ITER_SOURCE (WRITE) and reads to ITER_DEST(READ). >> >> Was going to ask "how did this ever work without splats", but looks like >> a fairly recent change AND it's for integrity which isn't widely used. >> But it does show a gap in testing for sure. > > The change is good and correct, but it doesn't look like normal tests > would find a problem here. The iter direction in this path only adds the > FOLL_WRITE flag, which appears to just check for writable access. Unless > you're specifically testing something using read-only PTE's, a test > wouldn't have triggered an early error. ? Ah I missed that - yeah no way we would've spotted this one other than under really funky configurations/setups. -- Jens Axboe
