commit c312828c37 upstream.
bpf_cgroup_from_id() is basically a wrapper to cgroup_get_from_id(),
that is relying on kernfs to determine the right cgroup associated to
the target id.
As a kfunc, it has the potential to be attached to any function through
BPF, particularly in contexts where certain locks are held.
However, kernfs is not using an irq safe spinlock for kernfs_idr_lock,
that means any kernfs function that is acquiring this lock can be
interrupted and potentially hit bpf_cgroup_from_id() in the process,
triggering a deadlock.
For example, it is really easy to trigger a lockdep splat between
kernfs_idr_lock and rq->_lock, attaching a small BPF program to
__set_cpus_allowed_ptr_locked() that just calls bpf_cgroup_from_id():
=====================================================
WARNING: HARDIRQ-safe -> HARDIRQ-unsafe lock order detected
6.7.0-rc7-virtme #5 Not tainted
-----------------------------------------------------
repro/131 [HC0[0]:SC0[0]:HE0:SE1] is trying to acquire:
ffffffffb2dc4578 (kernfs_idr_lock){+.+.}-{2:2}, at: kernfs_find_and_get_node_by_id+0x1d/0x80
and this task is already holding:
ffff911cbecaf218 (&rq->__lock){-.-.}-{2:2}, at: task_rq_lock+0x50/0xc0
which would create a new lock dependency:
(&rq->__lock){-.-.}-{2:2} -> (kernfs_idr_lock){+.+.}-{2:2}
but this new dependency connects a HARDIRQ-irq-safe lock:
(&rq->__lock){-.-.}-{2:2}
... which became HARDIRQ-irq-safe at:
lock_acquire+0xbf/0x2b0
_raw_spin_lock_nested+0x2e/0x40
scheduler_tick+0x5d/0x170
update_process_times+0x9c/0xb0
tick_periodic+0x27/0xe0
tick_handle_periodic+0x24/0x70
__sysvec_apic_timer_interrupt+0x64/0x1a0
sysvec_apic_timer_interrupt+0x6f/0x80
asm_sysvec_apic_timer_interrupt+0x1a/0x20
memcpy+0xc/0x20
arch_dup_task_struct+0x15/0x30
copy_process+0x1ce/0x1eb0
kernel_clone+0xac/0x390
kernel_thread+0x6f/0xa0
kthreadd+0x199/0x230
ret_from_fork+0x31/0x50
ret_from_fork_asm+0x1b/0x30
to a HARDIRQ-irq-unsafe lock:
(kernfs_idr_lock){+.+.}-{2:2}
... which became HARDIRQ-irq-unsafe at:
...
lock_acquire+0xbf/0x2b0
_raw_spin_lock+0x30/0x40
__kernfs_new_node.isra.0+0x83/0x280
kernfs_create_root+0xf6/0x1d0
sysfs_init+0x1b/0x70
mnt_init+0xd9/0x2a0
vfs_caches_init+0xcf/0xe0
start_kernel+0x58a/0x6a0
x86_64_start_reservations+0x18/0x30
x86_64_start_kernel+0xc5/0xe0
secondary_startup_64_no_verify+0x178/0x17b
other info that might help us debug this:
Possible interrupt unsafe locking scenario:
CPU0 CPU1
---- ----
lock(kernfs_idr_lock);
local_irq_disable();
lock(&rq->__lock);
lock(kernfs_idr_lock);
<Interrupt>
lock(&rq->__lock);
*** DEADLOCK ***
Prevent this deadlock condition converting kernfs_idr_lock to a raw irq
safe spinlock.
The performance impact of this change should be negligible and it also
helps to prevent similar deadlock conditions with any other subsystems
that may depend on kernfs.
Fixes: 332ea1f697 ("bpf: Add bpf_cgroup_from_id() kfunc")
Cc: stable <stable@kernel.org>
Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20231229074916.53547-1-andrea.righi@canonical.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
22269c75ab