]> Git Repo - linux.git/commitdiff
mm: fix use-after-free bug when mm->mmap is reused after being freed
authorSuren Baghdasaryan <[email protected]>
Sat, 26 Feb 2022 03:11:05 +0000 (19:11 -0800)
committerLinus Torvalds <[email protected]>
Sat, 26 Feb 2022 17:51:17 +0000 (09:51 -0800)
oom reaping (__oom_reap_task_mm) relies on a 2 way synchronization with
exit_mmap.  First it relies on the mmap_lock to exclude from unlock
path[1], page tables tear down (free_pgtables) and vma destruction.
This alone is not sufficient because mm->mmap is never reset.

For historical reasons[2] the lock is taken there is also MMF_OOM_SKIP
set for oom victims before.

The oom reaper only ever looks at oom victims so the whole scheme works
properly but process_mrelease can opearate on any task (with fatal
signals pending) which doesn't really imply oom victims.  That means
that the MMF_OOM_SKIP part of the synchronization doesn't work and it
can see a task after the whole address space has been demolished and
traverse an already released mm->mmap list.  This leads to use after
free as properly caught up by KASAN report.

Fix the issue by reseting mm->mmap so that MMF_OOM_SKIP synchronization
is not needed anymore.  The MMF_OOM_SKIP is not removed from exit_mmap
yet but it acts mostly as an optimization now.

[1] 27ae357fa82b ("mm, oom: fix concurrent munlock and oom reaper unmap, v3")
[2] 212925802454 ("mm: oom: let oom_reap_task and exit_mmap run concurrently")

[[email protected]: changelog rewrite]

Link: https://lore.kernel.org/all/[email protected]/
Link: https://lkml.kernel.org/r/[email protected]
Fixes: 64591e8605d6 ("mm: protect free_pgtables with mmap_lock write lock in exit_mmap")
Signed-off-by: Suren Baghdasaryan <[email protected]>
Reported-by: [email protected]
Suggested-by: Michal Hocko <[email protected]>
Reviewed-by: Rik van Riel <[email protected]>
Reviewed-by: Yang Shi <[email protected]>
Acked-by: Michal Hocko <[email protected]>
Cc: David Rientjes <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Cc: Johannes Weiner <[email protected]>
Cc: Roman Gushchin <[email protected]>
Cc: Rik van Riel <[email protected]>
Cc: Minchan Kim <[email protected]>
Cc: Kirill A. Shutemov <[email protected]>
Cc: Andrea Arcangeli <[email protected]>
Cc: Christian Brauner <[email protected]>
Cc: Christoph Hellwig <[email protected]>
Cc: Oleg Nesterov <[email protected]>
Cc: David Hildenbrand <[email protected]>
Cc: Jann Horn <[email protected]>
Cc: Shakeel Butt <[email protected]>
Cc: Andy Lutomirski <[email protected]>
Cc: Christian Brauner <[email protected]>
Cc: Florian Weimer <[email protected]>
Cc: Jan Engelhardt <[email protected]>
Cc: Tim Murray <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
mm/mmap.c

index 1e8fdb0b51eddcb8bce46b6f7c0648980827ea04..d445c1b9d60650b58c39c3688405567598c71fd9 100644 (file)
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -3186,6 +3186,7 @@ void exit_mmap(struct mm_struct *mm)
                vma = remove_vma(vma);
                cond_resched();
        }
+       mm->mmap = NULL;
        mmap_write_unlock(mm);
        vm_unacct_memory(nr_accounted);
 }
This page took 0.059624 seconds and 4 git commands to generate.