| Message ID | 20230620235726.3873043-1-surenb@google.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [1/3] mm: change vma_start_read to fail if VMA got detached from under it | expand |
On Tue, Jun 20, 2023 at 11:57 PM Suren Baghdasaryan <surenb@google.com> wrote: > > Current implementation of vma_start_read() checks VMA for being locked > before taking vma->vm_lock and then checks that again. This mechanism > fails to detect a case when the VMA gets write-locked, modified and > unlocked after the first check but before the vma->vm_lock is obtained. > While this is not strictly a problem (vma_start_read would not produce > a false unlocked result), this allows it to successfully lock a VMA which > got detached from the VMA tree while vma_start_read was locking it. > New condition checks for any change in vma->vm_lock_seq after we obtain > vma->vm_lock and will cause vma_start_read() to fail if the above race > occurs. Just a friendly ping for feedback. I know most people were busy fixing the stack expansion problem and these patches are not urgent, so no rush. If nobody reviews them, I'll ping again next week. > > Fixes: 5e31275cc997 ("mm: add per-VMA lock and helper functions to control it") > Suggested-by: Matthew Wilcox <willy@infradead.org> > Signed-off-by: Suren Baghdasaryan <surenb@google.com> > --- > include/linux/mm.h | 19 ++++++++++--------- > 1 file changed, 10 insertions(+), 9 deletions(-) > > diff --git a/include/linux/mm.h b/include/linux/mm.h > index 27ce77080c79..8410da79c570 100644 > --- a/include/linux/mm.h > +++ b/include/linux/mm.h > @@ -639,23 +639,24 @@ static inline void vma_numab_state_free(struct vm_area_struct *vma) {} > */ > static inline bool vma_start_read(struct vm_area_struct *vma) > { > - /* Check before locking. A race might cause false locked result. */ > - if (vma->vm_lock_seq == READ_ONCE(vma->vm_mm->mm_lock_seq)) > + int vm_lock_seq = READ_ONCE(vma->vm_lock_seq); > + > + /* > + * Check if VMA is locked before taking vma->vm_lock. A race or > + * mm_lock_seq overflow might cause false locked result. > + */ > + if (vm_lock_seq == READ_ONCE(vma->vm_mm->mm_lock_seq)) > return false; > > if (unlikely(down_read_trylock(&vma->vm_lock->lock) == 0)) > return false; > > - /* > - * Overflow might produce false locked result. > - * False unlocked result is impossible because we modify and check > - * vma->vm_lock_seq under vma->vm_lock protection and mm->mm_lock_seq > - * modification invalidates all existing locks. > - */ > - if (unlikely(vma->vm_lock_seq == READ_ONCE(vma->vm_mm->mm_lock_seq))) { > + /* Fail if VMA was write-locked after we checked it earlier */ > + if (unlikely(vm_lock_seq != READ_ONCE(vma->vm_lock_seq))) { > up_read(&vma->vm_lock->lock); > return false; > } > + > return true; > } > > -- > 2.41.0.162.gfafddb0af9-goog >
diff --git a/include/linux/mm.h b/include/linux/mm.h index 27ce77080c79..8410da79c570 100644 --- a/include/linux/mm.h +++ b/include/linux/mm.h @@ -639,23 +639,24 @@ static inline void vma_numab_state_free(struct vm_area_struct *vma) {} */ static inline bool vma_start_read(struct vm_area_struct *vma) { - /* Check before locking. A race might cause false locked result. */ - if (vma->vm_lock_seq == READ_ONCE(vma->vm_mm->mm_lock_seq)) + int vm_lock_seq = READ_ONCE(vma->vm_lock_seq); + + /* + * Check if VMA is locked before taking vma->vm_lock. A race or + * mm_lock_seq overflow might cause false locked result. + */ + if (vm_lock_seq == READ_ONCE(vma->vm_mm->mm_lock_seq)) return false; if (unlikely(down_read_trylock(&vma->vm_lock->lock) == 0)) return false; - /* - * Overflow might produce false locked result. - * False unlocked result is impossible because we modify and check - * vma->vm_lock_seq under vma->vm_lock protection and mm->mm_lock_seq - * modification invalidates all existing locks. - */ - if (unlikely(vma->vm_lock_seq == READ_ONCE(vma->vm_mm->mm_lock_seq))) { + /* Fail if VMA was write-locked after we checked it earlier */ + if (unlikely(vm_lock_seq != READ_ONCE(vma->vm_lock_seq))) { up_read(&vma->vm_lock->lock); return false; } + return true; }
Current implementation of vma_start_read() checks VMA for being locked before taking vma->vm_lock and then checks that again. This mechanism fails to detect a case when the VMA gets write-locked, modified and unlocked after the first check but before the vma->vm_lock is obtained. While this is not strictly a problem (vma_start_read would not produce a false unlocked result), this allows it to successfully lock a VMA which got detached from the VMA tree while vma_start_read was locking it. New condition checks for any change in vma->vm_lock_seq after we obtain vma->vm_lock and will cause vma_start_read() to fail if the above race occurs. Fixes: 5e31275cc997 ("mm: add per-VMA lock and helper functions to control it") Suggested-by: Matthew Wilcox <willy@infradead.org> Signed-off-by: Suren Baghdasaryan <surenb@google.com> --- include/linux/mm.h | 19 ++++++++++--------- 1 file changed, 10 insertions(+), 9 deletions(-)