fork: reset mm->pinned_vm

mm->pinned_vm counts pages of mm's address space that were permanently
pinned in memory by increasing their reference counter. The counter was
introduced by commit bc3e53f682d9 ("mm: distinguish between mlocked and
pinned pages"), while before it locked_vm had been used for such pages.

Obviously, we should reset the counter on fork if !CLONE_VM, just like
we do with locked_vm, but currently we don't. Let's fix it.

This patch will fix the contents of /proc/pid/status:VmPin.

ib_umem_get[infiniband] and perf_mmap still check pinned_vm against
RLIMIT_MEMLOCK.  It's left from the times when pinned pages were accounted
under locked_vm, but today it looks wrong.  It isn't clear how we should
deal with it.

We still have some drivers accounting pinned pages under mm->locked_vm -
this is what commit bc3e53f682d9 was fighting against.  It's
infiniband/usnic and vfio.

Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Roland Dreier <roland@kernel.org>
Cc: Sean Hefty <sean.hefty@intel.com>
Cc: Hal Rosenstock <hal.rosenstock@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

diff --git a/kernel/fork.c b/kernel/fork.c
index 418b52a..5a547a5 100644 (file)
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -543,6 +543,7 @@ static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
        atomic_long_set(&mm->nr_ptes, 0);
        mm->map_count = 0;
        mm->locked_vm = 0;
+       mm->pinned_vm = 0;
        memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
        spin_lock_init(&mm->page_table_lock);
        mm_init_cpumask(mm);