The profile introspection seq file has a locking bug when policy is viewed
from a virtual root (task in a policy namespace, introspection from the
real root is not affected).
The test for root
while (parent) {
is correct for the real root, but incorrect for tasks in a policy namespace.
This allows the task to walk backup the policy tree past its virtual root
causing it to be unlocked before the virtual root should be in the p_stop
fn.
This results in the following lockdep back trace:
[ 78.479744] [ BUG: bad unlock balance detected! ]
[ 78.479792] 3.11.0-11-generic #17 Not tainted
[ 78.479838] -------------------------------------
[ 78.479885] grep/2223 is trying to release lock (&ns->lock) at:
[ 78.479952] [<ffffffff817bf3be>] mutex_unlock+0xe/0x10
[ 78.480002] but there are no more locks to release!
[ 78.480037]
[ 78.480037] other info that might help us debug this:
[ 78.480037] 1 lock held by grep/2223:
[ 78.480037] #0: (&p->lock){+.+.+.}, at: [<ffffffff812111bd>] seq_read+0x3d/0x3d0
[ 78.480037]
[ 78.480037] stack backtrace:
[ 78.480037] CPU: 0 PID: 2223 Comm: grep Not tainted 3.11.0-11-generic #17
[ 78.480037] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
[ 78.480037] ffffffff817bf3be ffff880007763d60 ffffffff817b97ef ffff8800189d2190
[ 78.480037] ffff880007763d88 ffffffff810e1c6e ffff88001f044730 ffff8800189d2190
[ 78.480037] ffffffff817bf3be ffff880007763e00 ffffffff810e5bd6 0000000724fe56b7
[ 78.480037] Call Trace:
[ 78.480037] [<ffffffff817bf3be>] ? mutex_unlock+0xe/0x10
[ 78.480037] [<ffffffff817b97ef>] dump_stack+0x54/0x74
[ 78.480037] [<ffffffff810e1c6e>] print_unlock_imbalance_bug+0xee/0x100
[ 78.480037] [<ffffffff817bf3be>] ? mutex_unlock+0xe/0x10
[ 78.480037] [<ffffffff810e5bd6>] lock_release_non_nested+0x226/0x300
[ 78.480037] [<ffffffff817bf2fe>] ? __mutex_unlock_slowpath+0xce/0x180
[ 78.480037] [<ffffffff817bf3be>] ? mutex_unlock+0xe/0x10
[ 78.480037] [<ffffffff810e5d5c>] lock_release+0xac/0x310
[ 78.480037] [<ffffffff817bf2b3>] __mutex_unlock_slowpath+0x83/0x180
[ 78.480037] [<ffffffff817bf3be>] mutex_unlock+0xe/0x10
[ 78.480037] [<ffffffff81376c91>] p_stop+0x51/0x90
[ 78.480037] [<ffffffff81211408>] seq_read+0x288/0x3d0
[ 78.480037] [<ffffffff811e9d9e>] vfs_read+0x9e/0x170
[ 78.480037] [<ffffffff811ea8cc>] SyS_read+0x4c/0xa0
[ 78.480037] [<ffffffff817ccc9d>] system_call_fastpath+0x1a/0x1f
Signed-off-by: John Johansen <email address hidden>
Signed-off-by: Andy Whitcroft <email address hidden>
UBUNTU: SAUCE: apparmor: allocate path lookup buffers during init
BugLink: http://bugs.launchpad.net/bugs/1208988
dependency of: fix unix domain sockets to be mediated on connection needed to fix sleep in atomic context
Dynamically allocating buffers to store the path lookup slows mediation
down, and may require being able to sleep or accept failure of buffer
allocation. Handling fd inheritance during committing_creds, and unix
domain sockets can't fail nor can it sleep to do an allocation, so it
requires having preallocated buffers.
So add support for preallocated buffers and convert everything that
can be to use them.
Signed-off-by: John Johansen <email address hidden>
Signed-off-by: Andy Whitcroft <email address hidden>
This patch uses TRACE_EVENT to add tracepoints for the open(),
exec() and uselib() syscalls so that ureadahead can cheaply trace
the boot sequence to determine what to read to speed up the next.
It's not upstream because it will need to be rebased onto the syscall
trace events whenever that gets merged, and is a stop-gap.
Signed-off-by: Scott James Remnant <email address hidden>
Acked-by: Stefan Bader <email address hidden>
Acked-by: Andy Whitcroft <email address hidden>
[<email address hidden>: ported for 3.4]
Signed-off-by: Stefan Bader <email address hidden>
Signed-off-by: Andy Whitcroft <email address hidden>
This is a sync and squash to the apparmor 3 alpha 4 development snapshot
backported from 3.12. The set of patches in this squash are available
is the aa3.0-presquash branch of the dev tree.
Several of the patches in the squash have been submitted upstream.
Several more will be submitted soon, and other parts are still in
active review and development.
This squash will be updated to remove patches as they are pulled
into the upstream tree, and add new patches as they become available
to the stable apparmor dev branch.
In addition to the apparmor 3 sync this contains the backport patch
series for the 3.4 mako kernel
- apparmor: 3.10 backport revert no delay vfree()
- apparmor: 3.8 backport provide file_inode helper 496ad9aa
- apparmor: 3.6 backport revert uapi for capnames 43c422ed
- apparmor: 3.6 backport revert uapi for resnames 8a1ab315
- apparmor: 3.6 backport define kuid_t d2b31ca64
- apparmor: 3.6 backport kuid_t support for audit 2db81452
- apparmor: 3.6 backport remove const from sb_mount 808d4e3c
- apparmor: 3.4 backport revert file_mmap e5467859
- apparmor: 3.4 backport cap_mmap_addr d007794a
- apparmor: 3.4 backport fake no_new_privs 259e5e6c
- apparmor: 3.4 backport revert task audi_data 0972c74e
- apparmor: 3.4 backport alias file_open 83d49856
- UBUNTU: SAUCE: (no-up) apparmor: update config options
Signed-off-by: John Johansen <email address hidden>
Signed-off-by: Tim Gardner <email address hidden>
