On Wed, Jun 14, 2017 at 11:43 -0400, Ted Unangst wrote: > Mike Belopuhov wrote: > > still looking forward to replies to the original set of changes. > > i'm a little in between. on the one hand, yes, ok, it's good that we don't > leave corrupted buffers around with bad data. on the other hand, don't we want > to learn about these problems and fix them? i don't think the change is wrong, > but it seems like it covers up another issue.
Device drivers do not consider such situations as issues. Yes, they're edge cases that we can't normally trigger, but they're not bugs in drivers since over the years developers have deliberately put such code there. So I'm not entirely sure what do you think this is papering over. There's a clear violation of contract between buffer cache and the filesystem: FFS asked for 16k, got 16k plus resid of 20k which is weird to say the least.
