commit b67fbebd4cf980aecbcc750e1462128bffe8ae15 upstream.
Some drivers rely on having all VMAs through which a PFN might be accessible listed in the rmap for correctness. However, on X86, it was possible for a VMA with stale TLB entries to not be listed in the rmap.
This was fixed in mainline with commit b67fbebd4cf9 ("mmu_gather: Force tlb-flush VM_PFNMAP vmas"), but that commit relies on preceding refactoring in commit 18ba064e42df3 ("mmu_gather: Let there be one tlb_{start,end}_vma() implementation") and commit 1e9fdf21a4339 ("mmu_gather: Remove per arch tlb_{start,end}_vma()").
This patch provides equivalent protection without needing that refactoring, by forcing a TLB flush between removing PTEs in unmap_vmas() and the call to unlink_file_vma() in free_pgtables().
[This is a stable-specific rewrite of the upstream commit!] Signed-off-by: Jann Horn jannh@google.com --- mm/mmap.c | 12 ++++++++++++ 1 file changed, 12 insertions(+)
diff --git a/mm/mmap.c b/mm/mmap.c index 031fca1a7c65e..ce13266a55c69 100644 --- a/mm/mmap.c +++ b/mm/mmap.c @@ -2639,6 +2639,18 @@ static void unmap_region(struct mm_struct *mm, tlb_gather_mmu(&tlb, mm); update_hiwater_rss(mm); unmap_vmas(&tlb, vma, start, end); + + /* + * Ensure we have no stale TLB entries by the time this mapping is + * removed from the rmap. + * Note that we don't have to worry about nested flushes here because + * we're holding the mm semaphore for removing the mapping - so any + * concurrent flush in this region has to be coming through the rmap, + * and we synchronize against that using the rmap lock. + */ + if ((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) != 0) + tlb_flush_mmu(&tlb); + free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, next ? next->vm_start : USER_PGTABLES_CEILING); tlb_finish_mmu(&tlb);
base-commit: addc9003c2e895fe8a068a66de1de6fdb4c6ac60