mm: shmem_zero_setup skip security check and lockdep conflict with XFS
authorHugh Dickins <hughd@google.com>
Sun, 14 Jun 2015 16:48:09 +0000 (09:48 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 18 Jun 2015 06:40:19 +0000 (20:40 -1000)
commit66fc13039422ba7df2d01a8ee0873e4ef965b50b
tree0826a86f9b90bf179b462e9d01b7887400a48ee0
parent0f57d86787d8b1076ea8f9cbdddda2a46d534a27
mm: shmem_zero_setup skip security check and lockdep conflict with XFS

It appears that, at some point last year, XFS made directory handling
changes which bring it into lockdep conflict with shmem_zero_setup():
it is surprising that mmap() can clone an inode while holding mmap_sem,
but that has been so for many years.

Since those few lockdep traces that I've seen all implicated selinux,
I'm hoping that we can use the __shmem_file_setup(,,,S_PRIVATE) which
v3.13's commit c7277090927a ("security: shmem: implement kernel private
shmem inodes") introduced to avoid LSM checks on kernel-internal inodes:
the mmap("/dev/zero") cloned inode is indeed a kernel-internal detail.

This also covers the !CONFIG_SHMEM use of ramfs to support /dev/zero
(and MAP_SHARED|MAP_ANONYMOUS).  I thought there were also drivers
which cloned inode in mmap(), but if so, I cannot locate them now.

Reported-and-tested-by: Prarit Bhargava <prarit@redhat.com>
Reported-and-tested-by: Daniel Wagner <wagi@monom.org>
Reported-and-tested-by: Morten Stevens <mstevens@fedoraproject.org>
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/shmem.c