releases/4.9.183/cgroup-use-css_tryget-instead-of-css_tryget_online-in-task_get_css.patch

   1 From 18fa84a2db0e15b02baa5d94bdb5bd509175d2f6 Mon Sep 17 00:00:00 2001
   2 From: Tejun Heo <tj@kernel.org>
   3 Date: Wed, 29 May 2019 13:46:25 -0700
   4 Subject: cgroup: Use css_tryget() instead of css_tryget_online() in task_get_css()
   5
   6 From: Tejun Heo <tj@kernel.org>
   7
   8 commit 18fa84a2db0e15b02baa5d94bdb5bd509175d2f6 upstream.
   9
  10 A PF_EXITING task can stay associated with an offline css.  If such
  11 task calls task_get_css(), it can get stuck indefinitely.  This can be
  12 triggered by BSD process accounting which writes to a file with
  13 PF_EXITING set when racing against memcg disable as in the backtrace
  14 at the end.
  15
  16 After this change, task_get_css() may return a css which was already
  17 offline when the function was called.  None of the existing users are
  18 affected by this change.
  19
  20   INFO: rcu_sched self-detected stall on CPU
  21   INFO: rcu_sched detected stalls on CPUs/tasks:
  22   ...
  23   NMI backtrace for cpu 0
  24   ...
  25   Call Trace:
  26    <IRQ>
  27    dump_stack+0x46/0x68
  28    nmi_cpu_backtrace.cold.2+0x13/0x57
  29    nmi_trigger_cpumask_backtrace+0xba/0xca
  30    rcu_dump_cpu_stacks+0x9e/0xce
  31    rcu_check_callbacks.cold.74+0x2af/0x433
  32    update_process_times+0x28/0x60
  33    tick_sched_timer+0x34/0x70
  34    __hrtimer_run_queues+0xee/0x250
  35    hrtimer_interrupt+0xf4/0x210
  36    smp_apic_timer_interrupt+0x56/0x110
  37    apic_timer_interrupt+0xf/0x20
  38    </IRQ>
  39   RIP: 0010:balance_dirty_pages_ratelimited+0x28f/0x3d0
  40   ...
  41    btrfs_file_write_iter+0x31b/0x563
  42    __vfs_write+0xfa/0x140
  43    __kernel_write+0x4f/0x100
  44    do_acct_process+0x495/0x580
  45    acct_process+0xb9/0xdb
  46    do_exit+0x748/0xa00
  47    do_group_exit+0x3a/0xa0
  48    get_signal+0x254/0x560
  49    do_signal+0x23/0x5c0
  50    exit_to_usermode_loop+0x5d/0xa0
  51    prepare_exit_to_usermode+0x53/0x80
  52    retint_user+0x8/0x8
  53
  54 Signed-off-by: Tejun Heo <tj@kernel.org>
  55 Cc: stable@vger.kernel.org # v4.2+
  56 Fixes: ec438699a9ae ("cgroup, block: implement task_get_css() and use it in bio_associate_current()")
  57 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  58
  59 ---
  60  include/linux/cgroup.h |   10 ++++++++--
  61  1 file changed, 8 insertions(+), 2 deletions(-)
  62
  63 --- a/include/linux/cgroup.h
  64 +++ b/include/linux/cgroup.h
  65 @@ -462,7 +462,7 @@ static inline struct cgroup_subsys_state
  66   *
  67   * Find the css for the (@task, @subsys_id) combination, increment a
  68   * reference on and return it.  This function is guaranteed to return a
  69 - * valid css.
  70 + * valid css.  The returned css may already have been offlined.
  71   */
  72  static inline struct cgroup_subsys_state *
  73  task_get_css(struct task_struct *task, int subsys_id)
  74 @@ -472,7 +472,13 @@ task_get_css(struct task_struct *task, i
  75         rcu_read_lock();
  76         while (true) {
  77                 css = task_css(task, subsys_id);
  78 -               if (likely(css_tryget_online(css)))
  79 +               /*
  80 +                * Can't use css_tryget_online() here.  A task which has
  81 +                * PF_EXITING set may stay associated with an offline css.
  82 +                * If such task calls this function, css_tryget_online()
  83 +                * will keep failing.
  84 +                */
  85 +               if (likely(css_tryget(css)))
  86                         break;
  87                 cpu_relax();
  88         }