mm: migrate: fix an incorrect call of prep_transhuge_page()
authorZi Yan <zi.yan@cs.rutgers.edu>
Thu, 30 Nov 2017 00:11:12 +0000 (16:11 -0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 5 Dec 2017 10:26:29 +0000 (11:26 +0100)
commit13167cf417ddf105d2822eebe775547ed35d1a08
tree1278aa6a1e9bdc8b7ac081ac7dafdca51eb8a590
parent8a0bb9ebaa8b8faee61f095757662fe5d7fd8da6
mm: migrate: fix an incorrect call of prep_transhuge_page()

commit 40a899ed16486455f964e46d1af31fd4fded21c1 upstream.

In https://lkml.org/lkml/2017/11/20/411, Andrea reported that during
memory hotplug/hot remove prep_transhuge_page() is called incorrectly on
non-THP pages for migration, when THP is on but THP migration is not
enabled.  This leads to a bad state of target pages for migration.

By inspecting the code, if called on a non-THP, prep_transhuge_page()
will

 1) change the value of the mapping of (page + 2), since it is used for
    THP deferred list;

 2) change the lru value of (page + 1), since it is used for THP's dtor.

Both can lead to data corruption of these two pages.

Andrea said:
 "Pragmatically and from the point of view of the memory_hotplug subsys,
  the effect is a kernel crash when pages are being migrated during a
  memory hot remove offline and migration target pages are found in a
  bad state"

This patch fixes it by only calling prep_transhuge_page() when we are
certain that the target page is THP.

Link: http://lkml.kernel.org/r/20171121021855.50525-1-zi.yan@sent.com
Fixes: 8135d8926c08 ("mm: memory_hotplug: memory hotremove supports thp migration")
Signed-off-by: Zi Yan <zi.yan@cs.rutgers.edu>
Reported-by: Andrea Reale <ar@linux.vnet.ibm.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: "Jérôme Glisse" <jglisse@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
include/linux/migrate.h