sched: Fix static_key race with sched_feat()

As pointed out by Andi Kleen, the usage of static keys can be racy in
sched_feat_disable() vs. sched_feat_enable(). Currently, we first check the
value of keys->enabled, and subsequently update the branch direction. This,
can be racy and can potentially leave the keys in an inconsistent state.

Take the i_mutex around these calls to resolve the race.

Reported-by: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Jason Baron <jbaron@akamai.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: rostedt@goodmis.org
Link: http://lkml.kernel.org/r/9d7780c83db26683955cd01e6bc654ee2586e67f.1404315388.git.jbaron@akamai.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 2f96081..8705125 100644 (file)
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -245,6 +245,7 @@ sched_feat_write(struct file *filp, const char __user *ubuf,
        char buf[64];
        char *cmp;
        int i;
+       struct inode *inode;
 
        if (cnt > 63)
                cnt = 63;
@@ -255,7 +256,11 @@ sched_feat_write(struct file *filp, const char __user *ubuf,
        buf[cnt] = 0;
        cmp = strstrip(buf);
 
+       /* Ensure the static_key remains in a consistent state */
+       inode = file_inode(filp);
+       mutex_lock(&inode->i_mutex);
        i = sched_feat_set(cmp);
+       mutex_unlock(&inode->i_mutex);
        if (i == __SCHED_FEAT_NR)
                return -EINVAL;