Re: [PATCH v3 3/3] userfaultfd: use per-vma locks in userfaultfd operations

[Lokesh Gidra]

On Tue, Feb 6, 2024 at 9:05 AM Liam R. Howlett <Liam.Howlett@oraclecom> wrote:
>
> * Lokesh Gidra <lokeshgidra@xxxxxxxxxx> [240205 20:10]:
> > All userfaultfd operations, except write-protect, opportunistically use
> > per-vma locks to lock vmas. On failure, attempt again inside mmap_lock
> > critical section.
> >
> > Write-protect operation requires mmap_lock as it iterates over multiple
> > vmas.
> >
> > Signed-off-by: Lokesh Gidra <lokeshgidra@xxxxxxxxxx>
> > ---
> >  fs/userfaultfd.c              |  13 +-
> >  include/linux/mm.h            |  16 +++
> >  include/linux/userfaultfd_k.h |   5 +-
> >  mm/memory.c                   |  48 +++++++
> >  mm/userfaultfd.c              | 242 +++++++++++++++++++++-------------
> >  5 files changed, 222 insertions(+), 102 deletions(-)
> >
> > diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
> > index c00a021bcce4..60dcfafdc11a 100644
> > --- a/fs/userfaultfd.c
> > +++ b/fs/userfaultfd.c
> > @@ -2005,17 +2005,8 @@ static int userfaultfd_move(struct userfaultfd_ctx *ctx,
> >               return -EINVAL;
> >
> >       if (mmget_not_zero(mm)) {
> > -             mmap_read_lock(mm);
> > -
> > -             /* Re-check after taking map_changing_lock */
> > -             down_read(&ctx->map_changing_lock);
> > -             if (likely(!atomic_read(&ctx->mmap_changing)))
> > -                     ret = move_pages(ctx, mm, uffdio_move.dst, uffdio_move.src,
> > -                                      uffdio_move.len, uffdio_move.mode);
> > -             else
> > -                     ret = -EAGAIN;
> > -             up_read(&ctx->map_changing_lock);
> > -             mmap_read_unlock(mm);
> > +             ret = move_pages(ctx, uffdio_move.dst, uffdio_move.src,
> > +                              uffdio_move.len, uffdio_move.mode);
> >               mmput(mm);
> >       } else {
> >               return -ESRCH;
> > diff --git a/include/linux/mm.h b/include/linux/mm.h
> > index 0d1f98ab0c72..e69dfe2edcce 100644
> > --- a/include/linux/mm.h
> > +++ b/include/linux/mm.h
> > @@ -753,6 +753,11 @@ static inline void release_fault_lock(struct vm_fault *vmf)
> >               mmap_read_unlock(vmf->vma->vm_mm);
> >  }
> >
> > +static inline void unlock_vma(struct mm_struct *mm, struct vm_area_struct *vma)
> > +{
> > +     vma_end_read(vma);
> > +}
> > +
> >  static inline void assert_fault_locked(struct vm_fault *vmf)
> >  {
> >       if (vmf->flags & FAULT_FLAG_VMA_LOCK)
> > @@ -774,6 +779,9 @@ static inline void vma_assert_write_locked(struct vm_area_struct *vma)
> >               { mmap_assert_write_locked(vma->vm_mm); }
> >  static inline void vma_mark_detached(struct vm_area_struct *vma,
> >                                    bool detached) {}
> > +static inline void vma_acquire_read_lock(struct vm_area_struct *vma) {
> > +     mmap_assert_locked(vma->vm_mm);
> > +}
> >
> >  static inline struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
> >               unsigned long address)
> > @@ -786,6 +794,11 @@ static inline void release_fault_lock(struct vm_fault *vmf)
> >       mmap_read_unlock(vmf->vma->vm_mm);
> >  }
> >
> > +static inline void unlock_vma(struct mm_struct *mm, struct vm_area_struct *vma)
> > +{
> > +     mmap_read_unlock(mm);
> > +}
> > +
>
> Instead of passing two variables and only using one based on
> configuration of kernel build, why not use vma->vm_mm to
> mmap_read_unlock() and just pass the vma?
>
> It is odd to call unlock_vma() which maps to mmap_read_unlock().  Could
> we have this abstraction depend on CONFIG_PER_VMA_LOCK in uffd so that
> reading the code remains clear?  You seem to have pretty much two
> versions of each function already.  If you do that, then we can leave
> unlock_vma() undefined if !CONFIG_PER_VMA_LOCK.
>
> >  static inline void assert_fault_locked(struct vm_fault *vmf)
> >  {
> >       mmap_assert_locked(vmf->vma->vm_mm);
> > @@ -794,6 +807,9 @@ static inline void assert_fault_locked(struct vm_fault *vmf)
> >  #endif /* CONFIG_PER_VMA_LOCK */
> >
> >  extern const struct vm_operations_struct vma_dummy_vm_ops;
> > +extern struct vm_area_struct *lock_vma(struct mm_struct *mm,
> > +                                    unsigned long address,
> > +                                    bool prepare_anon);
> >
> >  /*
> >   * WARNING: vma_init does not initialize vma->vm_lock.
> > diff --git a/include/linux/userfaultfd_k.h b/include/linux/userfaultfd_k.h
> > index 3210c3552976..05d59f74fc88 100644
> > --- a/include/linux/userfaultfd_k.h
> > +++ b/include/linux/userfaultfd_k.h
> > @@ -138,9 +138,8 @@ extern long uffd_wp_range(struct vm_area_struct *vma,
> >  /* move_pages */
> >  void double_pt_lock(spinlock_t *ptl1, spinlock_t *ptl2);
> >  void double_pt_unlock(spinlock_t *ptl1, spinlock_t *ptl2);
> > -ssize_t move_pages(struct userfaultfd_ctx *ctx, struct mm_struct *mm,
> > -                unsigned long dst_start, unsigned long src_start,
> > -                unsigned long len, __u64 flags);
> > +ssize_t move_pages(struct userfaultfd_ctx *ctx, unsigned long dst_start,
> > +                unsigned long src_start, unsigned long len, __u64 flags);
> >  int move_pages_huge_pmd(struct mm_struct *mm, pmd_t *dst_pmd, pmd_t *src_pmd, pmd_t dst_pmdval,
> >                       struct vm_area_struct *dst_vma,
> >                       struct vm_area_struct *src_vma,
> > diff --git a/mm/memory.c b/mm/memory.c
> > index b05fd28dbce1..393ab3b0d6f3 100644
> > --- a/mm/memory.c
> > +++ b/mm/memory.c
> > @@ -5760,8 +5760,56 @@ struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
> >       count_vm_vma_lock_event(VMA_LOCK_ABORT);
> >       return NULL;
> >  }
> > +
> > +static void vma_acquire_read_lock(struct vm_area_struct *vma)
> > +{
> > +     /*
> > +      * We cannot use vma_start_read() as it may fail due to false locked
> > +      * (see comment in vma_start_read()). We can avoid that by directly
> > +      * locking vm_lock under mmap_lock, which guarantees that nobody could
> > +      * have locked the vma for write (vma_start_write()).
> > +      */
> > +     mmap_assert_locked(vma->vm_mm);
> > +     down_read(&vma->vm_lock->lock);
> > +}
> >  #endif /* CONFIG_PER_VMA_LOCK */
> >
> > +/*
> > + * lock_vma() - Lookup and lock VMA corresponding to @address.
>
> Missing arguments in the comment
>
> > + * @prepare_anon: If true, then prepare the VMA (if anonymous) with anon_vma.
> > + *
> > + * Should be called without holding mmap_lock. VMA should be unlocked after use
> > + * with unlock_vma().
> > + *
> > + * Return: A locked VMA containing @address, NULL of no VMA is found, or
> > + * -ENOMEM if anon_vma couldn't be allocated.
> > + */
> > +struct vm_area_struct *lock_vma(struct mm_struct *mm,
> > +                             unsigned long address,
> > +                             bool prepare_anon)
> > +{
> > +     struct vm_area_struct *vma;
> > +
> > +     vma = lock_vma_under_rcu(mm, address);
> > +
>
> Nit: extra new line
>
> > +     if (vma)
> > +             return vma;
> > +
> > +     mmap_read_lock(mm);
> > +     vma = vma_lookup(mm, address);
> > +     if (vma) {
> > +             if (prepare_anon && vma_is_anonymous(vma) &&
> > +                 anon_vma_prepare(vma))
> > +                     vma = ERR_PTR(-ENOMEM);
> > +             else
> > +                     vma_acquire_read_lock(vma);
> > +     }
> > +
> > +     if (IS_ENABLED(CONFIG_PER_VMA_LOCK) || !vma || PTR_ERR(vma) == -ENOMEM)
> > +             mmap_read_unlock(mm);
> > +     return vma;
> > +}
> > +
>
> It is also very odd that lock_vma() may, in fact, be locking the mm.  It
> seems like there is a layer of abstraction missing here, where your code
> would either lock the vma or lock the mm - like you had before, but
> without the confusing semantics of unlocking with a flag.  That is, we
> know what to do to unlock based on CONFIG_PER_VMA_LOCK, but it isn't
> always used.
>
> Maybe my comments were not clear on what I was thinking on the locking
> plan.  I was thinking that, in the CONFIG_PER_VMA_LOCK case, you could
> have a lock_vma() which does the per-vma locking which you can use in
> your code.  You could call lock_vma() in some uffd helper function that
> would do what is required (limit checking, etc) and return a locked vma.
>
> The counterpart of that would be another helper function that would do
> what was required under the mmap_read lock (limit check, etc).  The
> unlocking would be entirely config dependant as you have today.
>
> Just write the few functions you have twice: once for per-vma lock
> support, once without it.  Since we now can ensure the per-vma lock is
> taken in the per-vma lock path (or it failed), then you don't need to
> mmap_locked boolean you had in the previous version.  You solved the
> unlock issue already, but it should be abstracted so uffd calls the
> underlying unlock vs vma_unlock() doing an mmap_read_unlock() - because
> that's very confusing to see.
>
> I'd drop the vma from the function names that lock the mm or the vma as
> well.
>
> Thanks,
> Liam

I got it now. I'll make the changes in the next version.

Would it be ok to make lock_vma()/unlock_vma() (in case of
CONFIG_PER_VMA_LOCK) also be defined in mm/userfaultfd.c? The reason I
say this is because first there are no other users of these functions.
And also due to what Jann pointed out about anon_vma.
lock_vma_under_rcu() (rightly) only checks for private+anonymous case
and not private+file-backed case. So lock_vma() implementation is
getting very userfaultfd specific IMO.