Hello, Linus.
This pull request contains fixes for two long standing subtle bugs.
* kthread_bind() on a new kthread binds it to specific CPUs and
prevents userland from messing with the affinity or cgroup
membership. Unfortunately, for cgroup membership, there's a window
between kthread creation and kthread_bind*() invocation where the
kthread can be moved into a non-root cgroup by userland.
Depending on what controllers are in effect, this can assign the
kthread unexpected attributes. For example, in the reported case,
workqueue workers ended up in a non-root cpuset cgroups and had
their CPU affinities overridden. This broke workqueue invariants
and led to workqueue stalls.
Fixed by closing the window between kthread creation and
kthread_bind() as suggested by Oleg.
* There was a bug in cgroup mount path which could allow two competing
mount attempts to attach the same cgroup_root to two different
superblocks. This was caused by mishandling return value from
kernfs_pin_sb(). Fixed.
Thanks.
The following changes since commit b6a6759daf55dade2b65089957832759d502acfb:
cgroups: censor kernel pointer in debug files (2017-03-06 15:16:03 -0500)
are available in the git repository at:
git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup.git for-4.11-fixes
for you to fetch changes up to bfb0b80db5f9dca5ac0a5fd0edb765ee555e5a8e:
cgroup: avoid attaching a cgroup root to two different superblocks
(2017-04-11 09:00:57 +0900)
----------------------------------------------------------------
Tejun Heo (1):
cgroup, kthread: close race window where new kthreads can be migrated to
non-root cgroups
Zefan Li (1):
cgroup: avoid attaching a cgroup root to two different superblocks
include/linux/cgroup.h | 21 +++++++++++++++++++++
include/linux/sched.h | 4 ++++
kernel/cgroup/cgroup-v1.c | 2 +-
kernel/cgroup/cgroup.c | 9 +++++----
kernel/kthread.c | 3 +++
5 files changed, 34 insertions(+), 5 deletions(-)
diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h
index f6b43fb..af9c86e 100644
--- a/include/linux/cgroup.h
+++ b/include/linux/cgroup.h
@@ -570,6 +570,25 @@ static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
pr_cont_kernfs_path(cgrp->kn);
}
+static inline void cgroup_init_kthreadd(void)
+{
+ /*
+ * kthreadd is inherited by all kthreads, keep it in the root so
+ * that the new kthreads are guaranteed to stay in the root until
+ * initialization is finished.
+ */
+ current->no_cgroup_migration = 1;
+}
+
+static inline void cgroup_kthread_ready(void)
+{
+ /*
+ * This kthread finished initialization. The creator should have
+ * set PF_NO_SETAFFINITY if this kthread should stay in the root.
+ */
+ current->no_cgroup_migration = 0;
+}
+
#else /* !CONFIG_CGROUPS */
struct cgroup_subsys_state;
@@ -590,6 +609,8 @@ static inline void cgroup_free(struct task_struct *p) {}
static inline int cgroup_init_early(void) { return 0; }
static inline int cgroup_init(void) { return 0; }
+static inline void cgroup_init_kthreadd(void) {}
+static inline void cgroup_kthread_ready(void) {}
static inline bool task_under_cgroup_hierarchy(struct task_struct *task,
struct cgroup *ancestor)
diff --git a/include/linux/sched.h b/include/linux/sched.h
index d67eee8..4cf9a59 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -604,6 +604,10 @@ struct task_struct {
#ifdef CONFIG_COMPAT_BRK
unsigned brk_randomized:1;
#endif
+#ifdef CONFIG_CGROUPS
+ /* disallow userland-initiated cgroup migration */
+ unsigned no_cgroup_migration:1;
+#endif
unsigned long atomic_flags; /* Flags requiring atomic
access. */
diff --git a/kernel/cgroup/cgroup-v1.c b/kernel/cgroup/cgroup-v1.c
index 1dc22f6..12e19f0 100644
--- a/kernel/cgroup/cgroup-v1.c
+++ b/kernel/cgroup/cgroup-v1.c
@@ -1146,7 +1146,7 @@ struct dentry *cgroup1_mount(struct file_system_type
*fs_type, int flags,
* path is super cold. Let's just sleep a bit and retry.
*/
pinned_sb = kernfs_pin_sb(root->kf_root, NULL);
- if (IS_ERR(pinned_sb) ||
+ if (IS_ERR_OR_NULL(pinned_sb) ||
!percpu_ref_tryget_live(&root->cgrp.self.refcnt)) {
mutex_unlock(&cgroup_mutex);
if (!IS_ERR_OR_NULL(pinned_sb))
diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c
index 0125589..638ef75 100644
--- a/kernel/cgroup/cgroup.c
+++ b/kernel/cgroup/cgroup.c
@@ -2425,11 +2425,12 @@ ssize_t __cgroup_procs_write(struct kernfs_open_file
*of, char *buf,
tsk = tsk->group_leader;
/*
- * Workqueue threads may acquire PF_NO_SETAFFINITY and become
- * trapped in a cpuset, or RT worker may be born in a cgroup
- * with no rt_runtime allocated. Just say no.
+ * kthreads may acquire PF_NO_SETAFFINITY during initialization.
+ * If userland migrates such a kthread to a non-root cgroup, it can
+ * become trapped in a cpuset, or RT kthread may be born in a
+ * cgroup with no rt_runtime allocated. Just say no.
*/
- if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) {
+ if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) {
ret = -EINVAL;
goto out_unlock_rcu;
}
diff --git a/kernel/kthread.c b/kernel/kthread.c
index 2f26ade..26db528 100644
--- a/kernel/kthread.c
+++ b/kernel/kthread.c
@@ -20,6 +20,7 @@
#include <linux/freezer.h>
#include <linux/ptrace.h>
#include <linux/uaccess.h>
+#include <linux/cgroup.h>
#include <trace/events/sched.h>
static DEFINE_SPINLOCK(kthread_create_lock);
@@ -225,6 +226,7 @@ static int kthread(void *_create)
ret = -EINTR;
if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
+ cgroup_kthread_ready();
__kthread_parkme(self);
ret = threadfn(data);
}
@@ -538,6 +540,7 @@ int kthreadd(void *unused)
set_mems_allowed(node_states[N_MEMORY]);
current->flags |= PF_NOFREEZE;
+ cgroup_init_kthreadd();
for (;;) {
set_current_state(TASK_INTERRUPTIBLE);
