On Tue, Nov 15, 2016 at 05:47:26AM +0100, Dmitry Vyukov wrote: > Hello, > > The following program triggers BUG in munlock_vma_pages_range: > > // autogenerated by syzkaller (http://github.com/google/syzkaller) > #include <sys/mman.h> > > int main() > { > mmap((void*)0x20105000ul, 0xc00000ul, 0x2ul, 0x2172ul, -1, 0); > mremap((void*)0x201fd000ul, 0x4000ul, 0xc00000ul, 0x3ul, 0x203f0000ul); > return 0; > } > > > page:ffffea0001847cc0 count:0 mapcount:1 mapping:dead000000000400 > index:0x20400 compound_mapcount: 1 > flags: 0x5fffc0000000000() > page dumped because: VM_BUG_ON_PAGE(PageMlocked(page)) Ughh.. mlock() again. This should fix the issue: >From cc6bc279ca713a19288610c31f8256dc885c5d7e Mon Sep 17 00:00:00 2001 From: "Kirill A. Shutemov" <kirill.shutemov@xxxxxxxxxxxxxxx> Date: Tue, 15 Nov 2016 16:11:12 +0300 Subject: [PATCH] thp: fix corner case of munlock() of PTE-mapped THPs The following program triggers BUG() in munlock_vma_pages_range(): // autogenerated by syzkaller (http://github.com/google/syzkaller) #include <sys/mman.h> int main() { mmap((void*)0x20105000ul, 0xc00000ul, 0x2ul, 0x2172ul, -1, 0); mremap((void*)0x201fd000ul, 0x4000ul, 0xc00000ul, 0x3ul, 0x203f0000ul); return 0; } The test-case construct the situation when munlock_vma_pages_range() finds PTE-mapped THP-head in the middle of page table and, by mistake, skips HPAGE_PMD_NR pages after that. As result, on the next iteration it hits the middle of PMD-mapped THP and gets upset seeing mlocked tail page. The solution is only skip HPAGE_PMD_NR pages if the THP was mlocked during munlock_vma_page(). It would guarantee that the page is PMD-mapped as we never mlock PTE-mapeed THPs. Signed-off-by: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx> Reported-by: Dmitry Vyukov <dvyukov@xxxxxxxxxx> Fixes: e90309c9f772 ("thp: allow mlocked THP again") Cc: stable@xxxxxxxxxxxxxxx # v4.5+ --- mm/mlock.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) diff --git a/mm/mlock.c b/mm/mlock.c index 145a4258ddbc..cdbed8aaa426 100644 --- a/mm/mlock.c +++ b/mm/mlock.c @@ -190,10 +190,13 @@ unsigned int munlock_vma_page(struct page *page) */ spin_lock_irq(zone_lru_lock(zone)); - nr_pages = hpage_nr_pages(page); - if (!TestClearPageMlocked(page)) + if (!TestClearPageMlocked(page)) { + /* Potentially, PTE-mapped THP: do not skip the rest PTEs */ + nr_pages = 1; goto unlock_out; + } + nr_pages = hpage_nr_pages(page); __mod_zone_page_state(zone, NR_MLOCK, -nr_pages); if (__munlock_isolate_lru_page(page, true)) { -- Kirill A. Shutemov -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>