On Sat, May 22, 2021 at 2:19 PM Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote: > > On Fri, 21 May 2021 00:44:33 -0700 Mina Almasry <almasrymina@xxxxxxxxxx> wrote: > > > The userfaultfd hugetlb tests detect a resv_huge_pages underflow. This > > happens when hugetlb_mcopy_atomic_pte() is called with !is_continue on > > an index for which we already have a page in the cache. When this > > happens, we allocate a second page, double consuming the reservation, > > and then fail to insert the page into the cache and return -EEXIST. > > > > To fix this, we first if there exists a page in the cache which already > > ^ check > > > consumed the reservation, and return -EEXIST immediately if so. > > > > Secondly, if we fail to copy the page contents while holding the > > hugetlb_fault_mutex, we will drop the mutex and return to the caller > > after allocating a page that consumed a reservation. In this case there > > may be a fault that double consumes the reservation. To handle this, we > > free the allocated page, fix the reservations, and allocate a temporary > > hugetlb page and return that to the caller. When the caller does the > > copy outside of the lock, we again check the cache, and allocate a page > > consuming the reservation, and copy over the contents. > > > > Test: > > Hacked the code locally such that resv_huge_pages underflows produce > > a warning and the copy_huge_page_from_user() always fails, then: > > > > ./tools/testing/selftests/vm/userfaultfd hugetlb_shared 10 > > 2 /tmp/kokonut_test/huge/userfaultfd_test && echo test success > > ./tools/testing/selftests/vm/userfaultfd hugetlb 10 > > 2 /tmp/kokonut_test/huge/userfaultfd_test && echo test success > > > > Both tests succeed and produce no warnings. After the test runs > > number of free/resv hugepages is correct. > > > > ... > > > > include/linux/hugetlb.h | 4 ++ > > mm/hugetlb.c | 103 ++++++++++++++++++++++++++++++++++++---- > > mm/migrate.c | 39 +++------------ > > 3 files changed, 103 insertions(+), 43 deletions(-) > > I'm assuming we want this in -stable? > Umm I'll yield to Mike. This is a transient underflow; not actually THAT serious of an issue. Sorry, I'll clarify that in the commit message for the next version. > Are we able to identify a Fixes: for this? > No, this issue has been there latent for some time. It repros as far back as 5.11 at least, which is why maybe it's not that serious to require in -stable. > It's a large change. Can we come up with some smaller and easier to > review and integrate version which we can feed into 5.13 and -stable > and do the fancier version for 5.14? > Yes. If we only do the hugetlbfs_pagecache_present() check then that gets us some 90% of the way there, the rest of the patch addresses an unlikely race. > If you don't think -stable needs this then this version will be OK as-is.
