livepatch: fix race between fork and KLP transition
authorRik van Riel <riel@surriel.com>
Mon, 8 Aug 2022 19:00:19 +0000 (15:00 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 26 Oct 2022 10:34:30 +0000 (12:34 +0200)
commit31dc1727c10328b38a91a49ac0e48bbb6da95fbd
treeeb694c3704a2594d0e5fb53498ccb4c9ef1def42
parent36997b75bbb39c3d17cf8ca595d463c508acef63
livepatch: fix race between fork and KLP transition

commit 747f7a2901174c9afa805dddfb7b24db6f65e985 upstream.

The KLP transition code depends on the TIF_PATCH_PENDING and
the task->patch_state to stay in sync. On a normal (forward)
transition, TIF_PATCH_PENDING will be set on every task in
the system, while on a reverse transition (after a failed
forward one) first TIF_PATCH_PENDING will be cleared from
every task, followed by it being set on tasks that need to
be transitioned back to the original code.

However, the fork code copies over the TIF_PATCH_PENDING flag
from the parent to the child early on, in dup_task_struct and
setup_thread_stack. Much later, klp_copy_process will set
child->patch_state to match that of the parent.

However, the parent's patch_state may have been changed by KLP loading
or unloading since it was initially copied over into the child.

This results in the KLP code occasionally hitting this warning in
klp_complete_transition:

        for_each_process_thread(g, task) {
                WARN_ON_ONCE(test_tsk_thread_flag(task, TIF_PATCH_PENDING));
                task->patch_state = KLP_UNDEFINED;
        }

Set, or clear, the TIF_PATCH_PENDING flag in the child task
depending on whether or not it is needed at the time
klp_copy_process is called, at a point in copy_process where the
tasklist_lock is held exclusively, preventing races with the KLP
code.

The KLP code does have a few places where the state is changed
without the tasklist_lock held, but those should not cause
problems because klp_update_patch_state(current) cannot be
called while the current task is in the middle of fork,
klp_check_and_switch_task() which is called under the pi_lock,
which prevents rescheduling, and manipulation of the patch
state of idle tasks, which do not fork.

This should prevent this warning from triggering again in the
future, and close the race for both normal and reverse transitions.

Signed-off-by: Rik van Riel <riel@surriel.com>
Reported-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Petr Mladek <pmladek@suse.com>
Acked-by: Josh Poimboeuf <jpoimboe@kernel.org>
Fixes: d83a7cb375ee ("livepatch: change to a per-task consistency model")
Cc: stable@kernel.org
Signed-off-by: Petr Mladek <pmladek@suse.com>
Link: https://lore.kernel.org/r/20220808150019.03d6a67b@imladris.surriel.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
kernel/livepatch/transition.c