This allows us to unbind a cgroup subsystem from a hierarchy
which has sub-cgroups in it.
Due to some complexity, for now subsytems that use css_get/put()
can't be unbound from such a hierarchy.
Usage:
# mount -t cgroup -o cpuset,cpuacct xxx /mnt
# mkdir /mnt/tmp
# echo $$ > /mnt/tmp/tasks
(remove it from the hierarchy)
# mount -o remount,cpuset xxx /mnt
Signed-off-by: Li Zefan <lizf@xxxxxxxxxxxxxx>
---
kernel/cgroup.c | 76 ++++++++++++++++++++++++++++++++++++++++++++-----------
1 files changed, 61 insertions(+), 15 deletions(-)
diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index 46df5f8..9ce3fdb 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -1036,19 +1036,42 @@ static int hierarchy_attach_css(struct cgroup *cgrp, void *data)
return ret;
}
-static int hierarchy_update_css_sets(struct cgroup *cgrp, void *data)
+static int hierarchy_remove_css(struct cgroup *cgrp, void *data)
{
- unsigned long added_bits = (unsigned long)data;
+ unsigned long removed_bits = (unsigned long)data;
int i;
+
+ /*
+ * There might be some pointers to the cgroup_subsys_state
+ * that we are going to destroy.
+ */
+ synchronize_rcu();
+
+ for_each_set_bit(i, &removed_bits, CGROUP_SUBSYS_COUNT) {
+ subsys[i]->destroy(subsys[i], cgrp);
+ cgrp->subsys[i] = NULL;
+ }
+
+ return 0;
+}
+
+static void hierarchy_update_css_sets(struct cgroup *cgrp,
+ unsigned long bits, bool add)
+{
struct cg_cgroup_link *link;
+ int i;
write_lock(&css_set_lock);
list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
struct css_set *cg = link->cg;
struct hlist_head *hhead;
- for_each_set_bit(i, &added_bits, CGROUP_SUBSYS_COUNT)
- cg->subsys[i] = cgrp->subsys[i];
+ for_each_set_bit(i, &bits, CGROUP_SUBSYS_COUNT) {
+ if (add)
+ cg->subsys[i] = cgrp->subsys[i];
+ else
+ cg->subsys[i] = dummytop->subsys[i];
+ }
/* rehash */
hlist_del(&cg->hlist);
@@ -1056,7 +1079,21 @@ static int hierarchy_update_css_sets(struct cgroup *cgrp, void *data)
hlist_add_head(&cg->hlist, hhead);
}
write_unlock(&css_set_lock);
+}
+static int hierarchy_add_to_css_sets(struct cgroup *cgrp, void *data)
+{
+ unsigned long added_bits = (unsigned long)data;
+
+ hierarchy_update_css_sets(cgrp, added_bits, true);
+ return 0;
+}
+
+static int hierarchy_remove_from_css_sets(struct cgroup *cgrp, void *data)
+{
+ unsigned long removed_bits = (unsigned long)data;
+
+ hierarchy_update_css_sets(cgrp, removed_bits, false);
return 0;
}
@@ -1076,7 +1113,7 @@ static int hierarchy_populate_dir(struct cgroup *cgrp, void *data)
static int rebind_subsystems(struct cgroupfs_root *root,
unsigned long final_bits)
{
- unsigned long added_bits, removed_bits;
+ unsigned long added_bits, removed_bits, changed_bits;
struct cgroup *cgrp = &root->top_cgroup;
int i;
int err;
@@ -1085,6 +1122,7 @@ static int rebind_subsystems(struct cgroupfs_root *root,
removed_bits = root->actual_subsys_bits & ~final_bits;
added_bits = final_bits & ~root->actual_subsys_bits;
+ changed_bits = removed_bits | added_bits;
/* Check that any added subsystems are currently free */
for_each_set_bit(i, &added_bits, CGROUP_SUBSYS_COUNT) {
@@ -1100,14 +1138,11 @@ static int rebind_subsystems(struct cgroupfs_root *root,
}
}
- /* removing will be supported later */
- if (root->number_of_cgroups > 1 && removed_bits)
- return -EBUSY;
-
if (root->number_of_cgroups > 1) {
- for_each_set_bit(i, &added_bits, CGROUP_SUBSYS_COUNT)
+ for_each_set_bit(i, &changed_bits, CGROUP_SUBSYS_COUNT)
if (!subsys[i]->can_bind)
return -EBUSY;
+ }
for_each_set_bit(i, &added_bits, CGROUP_SUBSYS_COUNT) {
BUG_ON(cgrp->subsys[i]);
@@ -1123,7 +1158,7 @@ static int rebind_subsystems(struct cgroupfs_root *root,
if (err)
goto failed;
- cgroup_walk_hierarchy(hierarchy_update_css_sets,
+ cgroup_walk_hierarchy(hierarchy_add_to_css_sets,
(void *)added_bits, cgrp);
/* Process each subsystem */
@@ -1143,11 +1178,7 @@ static int rebind_subsystems(struct cgroupfs_root *root,
} else if (bit & removed_bits) {
/* We're removing this subsystem */
BUG_ON(ss == NULL);
- BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
- BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
mutex_lock(&ss->hierarchy_mutex);
- dummytop->subsys[i]->cgroup = dummytop;
- cgrp->subsys[i] = NULL;
if (ss->bind)
ss->bind(ss, dummytop);
subsys[i]->root = &rootnode;
@@ -1173,6 +1204,21 @@ static int rebind_subsystems(struct cgroupfs_root *root,
}
}
root->subsys_bits = root->actual_subsys_bits = final_bits;
+
+ for_each_set_bit(i, &removed_bits, CGROUP_SUBSYS_COUNT) {
+ BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
+ BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
+
+ dummytop->subsys[i]->cgroup = dummytop;
+ cgrp->subsys[i] = NULL;
+ }
+
+ cgroup_walk_hierarchy(hierarchy_remove_from_css_sets,
+ (void *)removed_bits, cgrp);
+
+ cgroup_walk_hierarchy(hierarchy_remove_css,
+ (void *)removed_bits, cgrp);
+
synchronize_rcu();
return 0;
--
1.7.0.1
_______________________________________________
Containers mailing list
Containers@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linux-foundation.org/mailman/listinfo/containers
