freezer: Add CHECKPOINTING state to safeguard container checkpoint]
RFC PATCH 1/1 freezer: Add CHECKPOINTING state to safeguard container checkpoint]
----
Sorry for the double post - I got Pavel's email incorrect earlier.
Pavel: your email under FREEZER in MAINTAINERS is out of date...
----
Quoted 3 lines · Matt Helsley
The CHECKPOINTING state prevents userspace from unfreezing tasks until
sys_checkpoint()is finished. When doing container checkpoint userspace will do:
[...]
Quoted 4 lines
/*
+ * cgroup freezer state changes made without the aid of the cgroup filesystem
+ * must go through this function to ensure proper locking is observed.
+ */
Review
Quoted 10 lines
(quoted diff) +10
-0+static int freezer_checkpointing(struct task_struct *task, + enum freezer_state next_state) +{ + struct freezer *freezer; + enum freezer_state state; + + task_lock(task); + freezer = task_freezer(task); + spin_lock_irq(&freezer->lock); + state = freezer->state;
After "echo FROZEN > /freezer/0/freezer_state", it is likely that freezer->state remains CGROUP_FREEZING --
Then, looking at freezer_read():
...
if (state == CGROUP_FREEZING) {
/* We change from FREEZING to FROZEN lazily if the cgroup was
* only partially frozen when we exitted write. */
update_freezer_state(cgroup, freezer);
state = freezer->state; }
...
IOW, usually, only if someone reads the freezer_state file will the freezer->status reliably reflect the state.
So, I searched for where else freezer->state is consulted, and found:
int cgroup_frozen(struct task_struct *task)
{
struct freezer *freezer;
enum freezer_state state;
task_lock(task);
freezer = task_freezer(task);
state = freezer->state;
task_unlock(task);
return state == CGROUP_FROZEN;}
Which is called (only?) from kernel/power/process.c:thaw_tasks(). Here, too, unless some read from the freezer_state file, this test will incorrectly fail.
Perhaps introduce a freezer_unlazy_state(freezer) helper, or a a freezer_get_state(freezer) helper to use in those places ?
Review
Quoted 8 lines
(quoted diff) +8
-0+ if ((state == CGROUP_FROZEN && next_state == CGROUP_CHECKPOINTING) || + (state == CGROUP_CHECKPOINTING && next_state == CGROUP_FROZEN)) + freezer->state = next_state; + spin_unlock_irq(&freezer->lock); + task_unlock(task); + + return state; +}
[...]
Oren.