devcgroup: relax white-list protection down to RCU
Pavel Emelyanov [Fri, 25 Jul 2008 08:47:07 +0000 (01:47 -0700)]
Currently this list is protected with a simple spinlock, even for reading
from one.  This is OK, but can be better.

Actually I want it to be better very much, since after replacing the
OpenVZ device permissions engine with the cgroup-based one I noticed, that
we set 12 default device permissions for each newly created container (for
/dev/null, full, terminals, ect devices), and people sometimes have up to
20 perms more, so traversing the ~30-40 elements list under a spinlock
doesn't seem very good.

Here's the RCU protection for white-list - dev_whitelist_item-s are added
and removed under the devcg->lock, but are looked up in permissions
checking under the rcu_read_lock.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

security/device_cgroup.c

index 236fffa..9da3532 100644 (file)
@@ -41,6 +41,7 @@ struct dev_whitelist_item {
        short type;
        short access;
        struct list_head list;
+       struct rcu_head rcu;
 };
 
 struct dev_cgroup {
@@ -133,11 +134,19 @@ static int dev_whitelist_add(struct dev_cgroup *dev_cgroup,
        }
 
        if (whcopy != NULL)
-               list_add_tail(&whcopy->list, &dev_cgroup->whitelist);
+               list_add_tail_rcu(&whcopy->list, &dev_cgroup->whitelist);
        spin_unlock(&dev_cgroup->lock);
        return 0;
 }
 
+static void whitelist_item_free(struct rcu_head *rcu)
+{
+       struct dev_whitelist_item *item;
+
+       item = container_of(rcu, struct dev_whitelist_item, rcu);
+       kfree(item);
+}
+
 /*
  * called under cgroup_lock()
  * since the list is visible to other tasks, we need the spinlock also
@@ -161,8 +170,8 @@ static void dev_whitelist_rm(struct dev_cgroup *dev_cgroup,
 remove:
                walk->access &= ~wh->access;
                if (!walk->access) {
-                       list_del(&walk->list);
-                       kfree(walk);
+                       list_del_rcu(&walk->list);
+                       call_rcu(&walk->rcu, whitelist_item_free);
                }
        }
        spin_unlock(&dev_cgroup->lock);
@@ -269,15 +278,15 @@ static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft,
        struct dev_whitelist_item *wh;
        char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
 
-       spin_lock(&devcgroup->lock);
-       list_for_each_entry(wh, &devcgroup->whitelist, list) {
+       rcu_read_lock();
+       list_for_each_entry_rcu(wh, &devcgroup->whitelist, list) {
                set_access(acc, wh->access);
                set_majmin(maj, wh->major);
                set_majmin(min, wh->minor);
                seq_printf(m, "%c %s:%s %s\n", type_to_char(wh->type),
                           maj, min, acc);
        }
-       spin_unlock(&devcgroup->lock);
+       rcu_read_unlock();
 
        return 0;
 }
@@ -510,8 +519,8 @@ int devcgroup_inode_permission(struct inode *inode, int mask)
        if (!dev_cgroup)
                return 0;
 
-       spin_lock(&dev_cgroup->lock);
-       list_for_each_entry(wh, &dev_cgroup->whitelist, list) {
+       rcu_read_lock();
+       list_for_each_entry_rcu(wh, &dev_cgroup->whitelist, list) {
                if (wh->type & DEV_ALL)
                        goto acc_check;
                if ((wh->type & DEV_BLOCK) && !S_ISBLK(inode->i_mode))
@@ -527,10 +536,10 @@ acc_check:
                        continue;
                if ((mask & MAY_READ) && !(wh->access & ACC_READ))
                        continue;
-               spin_unlock(&dev_cgroup->lock);
+               rcu_read_unlock();
                return 0;
        }
-       spin_unlock(&dev_cgroup->lock);
+       rcu_read_unlock();
 
        return -EPERM;
 }
@@ -545,7 +554,7 @@ int devcgroup_inode_mknod(int mode, dev_t dev)
        if (!dev_cgroup)
                return 0;
 
-       spin_lock(&dev_cgroup->lock);
+       rcu_read_lock();
        list_for_each_entry(wh, &dev_cgroup->whitelist, list) {
                if (wh->type & DEV_ALL)
                        goto acc_check;
@@ -560,9 +569,9 @@ int devcgroup_inode_mknod(int mode, dev_t dev)
 acc_check:
                if (!(wh->access & ACC_MKNOD))
                        continue;
-               spin_unlock(&dev_cgroup->lock);
+               rcu_read_unlock();
                return 0;
        }
-       spin_unlock(&dev_cgroup->lock);
+       rcu_read_unlock();
        return -EPERM;
 }