cgroup: fix cgroup_path() vs rename() race
rename() will change dentry->d_name. The result of this race can
be worse than seeing partially rewritten name, but we might access
a stale pointer because rename() will re-allocate memory to hold
a longer name.
As accessing dentry->name must be protected by dentry->d_lock or
parent inode's i_mutex, while on the other hand cgroup-path() can
be called with some irq-safe spinlocks held, we can't generate
cgroup path using dentry->d_name.
Alternatively we make a copy of dentry->d_name and save it in
cgrp->name when a cgroup is created, and update cgrp->name at
rename().
v5: use flexible array instead of zero-size array.
v4: - allocate root_cgroup_name and all root_cgroup->name points to it.
- add cgroup_name() wrapper.
v3: use kfree_rcu() instead of synchronize_rcu() in user-visible path.
v2: make cgrp->name RCU safe.
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index a32f943..5068216 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -238,6 +238,8 @@
/* dummytop is a shorthand for the dummy hierarchy's top cgroup */
#define dummytop (&rootnode.top_cgroup)
+static struct cgroup_name root_cgroup_name = { .name = "/" };
+
/* This flag indicates whether tasks in the fork and exit paths should
* check for fork/exit handlers to call. This avoids us having to do
* extra work in the fork/exit path if none of the subsystems need to
@@ -859,6 +861,17 @@
return inode;
}
+static struct cgroup_name *cgroup_alloc_name(struct dentry *dentry)
+{
+ struct cgroup_name *name;
+
+ name = kmalloc(sizeof(*name) + dentry->d_name.len + 1, GFP_KERNEL);
+ if (!name)
+ return NULL;
+ strcpy(name->name, dentry->d_name.name);
+ return name;
+}
+
static void cgroup_free_fn(struct work_struct *work)
{
struct cgroup *cgrp = container_of(work, struct cgroup, free_work);
@@ -889,6 +902,7 @@
simple_xattrs_free(&cgrp->xattrs);
ida_simple_remove(&cgrp->root->cgroup_ida, cgrp->id);
+ kfree(rcu_dereference_raw(cgrp->name));
kfree(cgrp);
}
@@ -1421,6 +1435,7 @@
INIT_LIST_HEAD(&root->allcg_list);
root->number_of_cgroups = 1;
cgrp->root = root;
+ cgrp->name = &root_cgroup_name;
cgrp->top_cgroup = cgrp;
init_cgroup_housekeeping(cgrp);
list_add_tail(&cgrp->allcg_node, &root->allcg_list);
@@ -1769,49 +1784,45 @@
* @buf: the buffer to write the path into
* @buflen: the length of the buffer
*
- * Called with cgroup_mutex held or else with an RCU-protected cgroup
- * reference. Writes path of cgroup into buf. Returns 0 on success,
- * -errno on error.
+ * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
+ *
+ * We can't generate cgroup path using dentry->d_name, as accessing
+ * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
+ * inode's i_mutex, while on the other hand cgroup_path() can be called
+ * with some irq-safe spinlocks held.
*/
int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
{
- struct dentry *dentry = cgrp->dentry;
+ int ret = -ENAMETOOLONG;
char *start;
- rcu_lockdep_assert(rcu_read_lock_held() || cgroup_lock_is_held(),
- "cgroup_path() called without proper locking");
-
- if (cgrp == dummytop) {
- /*
- * Inactive subsystems have no dentry for their root
- * cgroup
- */
- strcpy(buf, "/");
- return 0;
- }
-
start = buf + buflen - 1;
-
*start = '\0';
- for (;;) {
- int len = dentry->d_name.len;
+ rcu_read_lock();
+ while (cgrp) {
+ const char *name = cgroup_name(cgrp);
+ int len;
+
+ len = strlen(name);
if ((start -= len) < buf)
- return -ENAMETOOLONG;
- memcpy(start, dentry->d_name.name, len);
- cgrp = cgrp->parent;
- if (!cgrp)
+ goto out;
+ memcpy(start, name, len);
+
+ if (!cgrp->parent)
break;
- dentry = cgrp->dentry;
- if (!cgrp->parent)
- continue;
if (--start < buf)
- return -ENAMETOOLONG;
+ goto out;
*start = '/';
+
+ cgrp = cgrp->parent;
}
+ ret = 0;
memmove(buf, start, buf + buflen - start);
- return 0;
+out:
+ rcu_read_unlock();
+ return ret;
}
EXPORT_SYMBOL_GPL(cgroup_path);
@@ -2537,13 +2548,40 @@
static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
struct inode *new_dir, struct dentry *new_dentry)
{
+ int ret;
+ struct cgroup_name *name, *old_name;
+ struct cgroup *cgrp;
+
+ /*
+ * It's convinient to use parent dir's i_mutex to protected
+ * cgrp->name.
+ */
+ lockdep_assert_held(&old_dir->i_mutex);
+
if (!S_ISDIR(old_dentry->d_inode->i_mode))
return -ENOTDIR;
if (new_dentry->d_inode)
return -EEXIST;
if (old_dir != new_dir)
return -EIO;
- return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
+
+ cgrp = __d_cgrp(old_dentry);
+
+ name = cgroup_alloc_name(new_dentry);
+ if (!name)
+ return -ENOMEM;
+
+ ret = simple_rename(old_dir, old_dentry, new_dir, new_dentry);
+ if (ret) {
+ kfree(name);
+ return ret;
+ }
+
+ old_name = cgrp->name;
+ rcu_assign_pointer(cgrp->name, name);
+
+ kfree_rcu(old_name, rcu_head);
+ return 0;
}
static struct simple_xattrs *__d_xattrs(struct dentry *dentry)
@@ -4158,6 +4196,7 @@
umode_t mode)
{
struct cgroup *cgrp;
+ struct cgroup_name *name;
struct cgroupfs_root *root = parent->root;
int err = 0;
struct cgroup_subsys *ss;
@@ -4168,9 +4207,14 @@
if (!cgrp)
return -ENOMEM;
+ name = cgroup_alloc_name(dentry);
+ if (!name)
+ goto err_free_cgrp;
+ rcu_assign_pointer(cgrp->name, name);
+
cgrp->id = ida_simple_get(&root->cgroup_ida, 1, 0, GFP_KERNEL);
if (cgrp->id < 0)
- goto err_free_cgrp;
+ goto err_free_name;
/*
* Only live parents can have children. Note that the liveliness
@@ -4276,6 +4320,8 @@
deactivate_super(sb);
err_free_id:
ida_simple_remove(&root->cgroup_ida, cgrp->id);
+err_free_name:
+ kfree(rcu_dereference_raw(cgrp->name));
err_free_cgrp:
kfree(cgrp);
return err;