On Fri, Feb 15, 2019 at 6:10 AM Jann Horn <ja...@google.com> wrote: > > On Thu, Feb 14, 2019 at 11:21 PM David Miller <da...@davemloft.net> wrote: > > > > From: Jann Horn <ja...@google.com> > > Date: Thu, 14 Feb 2019 22:26:22 +0100 > > > > > On Thu, Feb 14, 2019 at 6:13 PM David Miller <da...@davemloft.net> wrote: > > >> > > >> From: Jann Horn <ja...@google.com> > > >> Date: Wed, 13 Feb 2019 22:45:59 +0100 > > >> > > >> > The basic idea behind ->pagecnt_bias is: If we pre-allocate the maximum > > >> > number of references that we might need to create in the fastpath > > >> > later, > > >> > the bump-allocation fastpath only has to modify the non-atomic bias > > >> > value > > >> > that tracks the number of extra references we hold instead of the > > >> > atomic > > >> > refcount. The maximum number of allocations we can serve (under the > > >> > assumption that no allocation is made with size 0) is nc->size, so > > >> > that's > > >> > the bias used. > > >> > > > >> > However, even when all memory in the allocation has been given away, a > > >> > reference to the page is still held; and in the `offset < 0` slowpath, > > >> > the > > >> > page may be reused if everyone else has dropped their references. > > >> > This means that the necessary number of references is actually > > >> > `nc->size+1`. > > >> > > > >> > Luckily, from a quick grep, it looks like the only path that can call > > >> > page_frag_alloc(fragsz=1) is TAP with the IFF_NAPI_FRAGS flag, which > > >> > requires CAP_NET_ADMIN in the init namespace and is only intended to be > > >> > used for kernel testing and fuzzing. > > >> > > > >> > To test for this issue, put a `WARN_ON(page_ref_count(page) == 0)` in > > >> > the > > >> > `offset < 0` path, below the virt_to_page() call, and then repeatedly > > >> > call > > >> > writev() on a TAP device with > > >> > IFF_TAP|IFF_NO_PI|IFF_NAPI_FRAGS|IFF_NAPI, > > >> > with a vector consisting of 15 elements containing 1 byte each. > > >> > > > >> > Signed-off-by: Jann Horn <ja...@google.com> > > >> > > >> Applied and queued up for -stable. > > > > > > I had sent a v2 at Alexander Duyck's request an hour before you > > > applied the patch (with a minor difference that, in Alexander's > > > opinion, might be slightly more efficient). I guess the net tree > > > doesn't work like the mm tree, where patches can get removed and > > > replaced with newer versions? So if Alexander wants that change > > > (s/size/PAGE_FRAG_CACHE_MAX_SIZE/ in the refcount), someone has to > > > send that as a separate patch? > > > > Yes, please send a follow-up. Sorry about that. > > @Alexander Do you want to do that? It was your idea and I don't think > I can reasonably judge the usefulness of the change.
I'll take care of it. I'm kind of annoyed that you resubmitted this to netdev before anyone had a chance to even provide review comments though. As is this doesn't really address the issue anyway since the bigger issue is the data alignment issue that I pointed out. I'll have patches for both ready shortly. - Alex
