cgroup: allocate 2x cgrp_cset_links when setting up a new root
authorTejun Heo <tj@kernel.org>
Thu, 3 Mar 2016 14:58:01 +0000 (09:58 -0500)
committerTejun Heo <tj@kernel.org>
Thu, 3 Mar 2016 14:58:01 +0000 (09:58 -0500)
During prep, cgroup_setup_root() allocates cgrp_cset_links matching
the number of existing css_sets to later link the new root.  This is
fine for now as the only operation which can happen inbetween is
rebind_subsystems() and rebinding of empty subsystems doesn't create
new css_sets.

However, while not yet allowed, with the recent reimplementation,
rebind_subsystems() can rebind subsystems with descendant csses and
thus can create new css_sets.  This patch makes cgroup_setup_root()
allocate 2x of the existing css_sets so that later use of live
subsystem rebinding doesn't blow up.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Zefan Li <lizefan@huawei.com>
kernel/cgroup.c

index 58e02e9aa9701941e8ad3e7e21881983d9547cf1..40ed329482ddb6bd5d96ab45df2dce605c5af59f 100644 (file)
@@ -1915,10 +1915,11 @@ static int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
        /*
         * We're accessing css_set_count without locking css_set_lock here,
         * but that's OK - it can only be increased by someone holding
-        * cgroup_lock, and that's us. The worst that can happen is that we
-        * have some link structures left over
+        * cgroup_lock, and that's us.  Later rebinding may disable
+        * controllers on the default hierarchy and thus create new csets,
+        * which can't be more than the existing ones.  Allocate 2x.
         */
-       ret = allocate_cgrp_cset_links(css_set_count, &tmp_links);
+       ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links);
        if (ret)
                goto cancel_ref;