Re: [PATCH 32/94] kernel/fork: Convert dup_mmap to use maple tree

[Liam Howlett <liam.howlett@oracle.com>]

* Suren Baghdasaryan <surenb@google.com> [210528 20:42]:
> On Wed, Apr 28, 2021 at 8:36 AM Liam Howlett <liam.howlett@oracle.com> wrote:
> >
> > Use the maple tree iterator to duplicate the mm_struct trees.
> >
> > Signed-off-by: Liam R. Howlett <Liam.Howlett@Oracle.com>
> > ---
> >  include/linux/mm.h       |  2 --
> >  include/linux/sched/mm.h |  3 +++
> >  kernel/fork.c            | 24 +++++++++++++++++++-----
> >  mm/mmap.c                |  4 ----
> >  4 files changed, 22 insertions(+), 11 deletions(-)
> >
> > diff --git a/include/linux/mm.h b/include/linux/mm.h
> > index e89bacfa9145..7f7dff6ad884 100644
> > --- a/include/linux/mm.h
> > +++ b/include/linux/mm.h
> > @@ -2498,8 +2498,6 @@ extern bool arch_has_descending_max_zone_pfns(void);
> >  /* nommu.c */
> >  extern atomic_long_t mmap_pages_allocated;
> >  extern int nommu_shrink_inode_mappings(struct inode *, size_t, size_t);
> > -/* maple_tree */
> > -void vma_store(struct mm_struct *mm, struct vm_area_struct *vma);
> >
> >  /* interval_tree.c */
> >  void vma_interval_tree_insert(struct vm_area_struct *node,
> > diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h
> > index e24b1fe348e3..76cab3aea6ab 100644
> > --- a/include/linux/sched/mm.h
> > +++ b/include/linux/sched/mm.h
> > @@ -8,6 +8,7 @@
> >  #include <linux/mm_types.h>
> >  #include <linux/gfp.h>
> >  #include <linux/sync_core.h>
> > +#include <linux/maple_tree.h>
> >
> >  /*
> >   * Routines for handling mm_structs
> > @@ -67,11 +68,13 @@ static inline void mmdrop(struct mm_struct *mm)
> >   */
> >  static inline void mmget(struct mm_struct *mm)
> >  {
> > +       mt_set_in_rcu(&mm->mm_mt);
> >         atomic_inc(&mm->mm_users);
> >  }
> >
> >  static inline bool mmget_not_zero(struct mm_struct *mm)
> >  {
> > +       mt_set_in_rcu(&mm->mm_mt);
> 
> Should you be calling mt_set_in_rcu() if atomic_inc_not_zero() failed?
> I don't think mmput() is called after mmget_not_zero() fails and
> mt_clear_in_rcu() will not be called.

Good catch, but having it the way it is will be faster with the
possibility of re-entering RCU mode if there is a race during tear down.
Entering RCU mode during tear-down mean that the nodes that are not
already freed would remain for an RCU cycle before being freed.  I don't
think it is worth checking every time this is called for such a low
payoff.  I should probably add a comment about this though.

> 
> >         return atomic_inc_not_zero(&mm->mm_users);
> >  }
> >
> > diff --git a/kernel/fork.c b/kernel/fork.c
> > index c37abaf28eb9..832416ff613e 100644
> > --- a/kernel/fork.c
> > +++ b/kernel/fork.c
> > @@ -477,7 +477,9 @@ static __latent_entropy int dup_mmap(struct mm_struct *mm,
> >         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
> >         struct rb_node **rb_link, *rb_parent;
> >         int retval;
> > -       unsigned long charge;
> > +       unsigned long charge = 0;
> > +       MA_STATE(old_mas, &oldmm->mm_mt, 0, 0);
> > +       MA_STATE(mas, &mm->mm_mt, 0, 0);
> >         LIST_HEAD(uf);
> >
> >         uprobe_start_dup_mmap();
> > @@ -511,7 +513,13 @@ static __latent_entropy int dup_mmap(struct mm_struct *mm,
> >                 goto out;
> >
> >         prev = NULL;
> > -       for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
> > +
> > +       retval = mas_entry_count(&mas, oldmm->map_count);
> > +       if (retval)
> > +               goto fail_nomem;
> > +
> > +       rcu_read_lock();
> > +       mas_for_each(&old_mas, mpnt, ULONG_MAX) {
> >                 struct file *file;
> >
> >                 if (mpnt->vm_flags & VM_DONTCOPY) {
> > @@ -525,7 +533,7 @@ static __latent_entropy int dup_mmap(struct mm_struct *mm,
> >                  */
> >                 if (fatal_signal_pending(current)) {
> >                         retval = -EINTR;
> > -                       goto out;
> > +                       goto loop_out;
> >                 }
> >                 if (mpnt->vm_flags & VM_ACCOUNT) {
> >                         unsigned long len = vma_pages(mpnt);
> > @@ -594,7 +602,9 @@ static __latent_entropy int dup_mmap(struct mm_struct *mm,
> >                 rb_parent = &tmp->vm_rb;
> >
> >                 /* Link the vma into the MT */
> > -               vma_store(mm, tmp);
> > +               mas.index = tmp->vm_start;
> > +               mas.last = tmp->vm_end - 1;
> > +               mas_store(&mas, tmp);
> >
> >                 mm->map_count++;
> >                 if (!(tmp->vm_flags & VM_WIPEONFORK))
> > @@ -604,14 +614,17 @@ static __latent_entropy int dup_mmap(struct mm_struct *mm,
> >                         tmp->vm_ops->open(tmp);
> >
> >                 if (retval)
> > -                       goto out;
> > +                       goto loop_out;
> >         }
> >         /* a new mm has just been created */
> >         retval = arch_dup_mmap(oldmm, mm);
> > +loop_out:
> >  out:
> > +       rcu_read_unlock();
> >         mmap_write_unlock(mm);
> >         flush_tlb_mm(oldmm);
> >         mmap_write_unlock(oldmm);
> > +       mas_destroy(&mas);
> >         dup_userfaultfd_complete(&uf);
> >  fail_uprobe_end:
> >         uprobe_end_dup_mmap();
> > @@ -1092,6 +1105,7 @@ static inline void __mmput(struct mm_struct *mm)
> >  {
> >         VM_BUG_ON(atomic_read(&mm->mm_users));
> >
> > +       mt_clear_in_rcu(&mm->mm_mt);
> >         uprobe_clear_state(mm);
> >         exit_aio(mm);
> >         ksm_exit(mm);
> > diff --git a/mm/mmap.c b/mm/mmap.c
> > index 929c2f9eb3f5..1bd43f4db28e 100644
> > --- a/mm/mmap.c
> > +++ b/mm/mmap.c
> > @@ -780,10 +780,6 @@ static inline void vma_mt_store(struct mm_struct *mm, struct vm_area_struct *vma
> >                 GFP_KERNEL);
> >  }
> >
> > -void vma_store(struct mm_struct *mm, struct vm_area_struct *vma) {
> > -       vma_mt_store(mm, vma);
> > -}
> > -
> >  static void
> >  __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
> >         struct vm_area_struct *prev, struct rb_node **rb_link,
> > --
> > 2.30.2