2 * net/core/netprio_cgroup.c Priority Control Group
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Authors: Neil Horman <nhorman@tuxdriver.com>
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 #include <linux/module.h>
15 #include <linux/slab.h>
16 #include <linux/types.h>
17 #include <linux/string.h>
18 #include <linux/errno.h>
19 #include <linux/skbuff.h>
20 #include <linux/cgroup.h>
21 #include <linux/rcupdate.h>
22 #include <linux/atomic.h>
23 #include <net/rtnetlink.h>
24 #include <net/pkt_cls.h>
26 #include <net/netprio_cgroup.h>
28 #include <linux/fdtable.h>
30 #define PRIOMAP_MIN_SZ 128
31 #define PRIOIDX_SZ 128
33 static unsigned long prioidx_map[PRIOIDX_SZ];
34 static DEFINE_SPINLOCK(prioidx_map_lock);
36 static inline struct cgroup_netprio_state *cgrp_netprio_state(struct cgroup *cgrp)
38 return container_of(cgroup_subsys_state(cgrp, net_prio_subsys_id),
39 struct cgroup_netprio_state, css);
42 static int get_prioidx(u32 *prio)
47 spin_lock_irqsave(&prioidx_map_lock, flags);
48 prioidx = find_first_zero_bit(prioidx_map, sizeof(unsigned long) * PRIOIDX_SZ);
49 if (prioidx == sizeof(unsigned long) * PRIOIDX_SZ) {
50 spin_unlock_irqrestore(&prioidx_map_lock, flags);
53 set_bit(prioidx, prioidx_map);
54 spin_unlock_irqrestore(&prioidx_map_lock, flags);
59 static void put_prioidx(u32 idx)
63 spin_lock_irqsave(&prioidx_map_lock, flags);
64 clear_bit(idx, prioidx_map);
65 spin_unlock_irqrestore(&prioidx_map_lock, flags);
69 * Extend @dev->priomap so that it's large enough to accomodate
70 * @target_idx. @dev->priomap.priomap_len > @target_idx after successful
71 * return. Must be called under rtnl lock.
73 static int extend_netdev_table(struct net_device *dev, u32 target_idx)
75 struct netprio_map *old, *new;
76 size_t new_sz, new_len;
78 /* is the existing priomap large enough? */
79 old = rtnl_dereference(dev->priomap);
80 if (old && old->priomap_len > target_idx)
84 * Determine the new size. Let's keep it power-of-two. We start
85 * from PRIOMAP_MIN_SZ and double it until it's large enough to
86 * accommodate @target_idx.
88 new_sz = PRIOMAP_MIN_SZ;
90 new_len = (new_sz - offsetof(struct netprio_map, priomap)) /
91 sizeof(new->priomap[0]);
92 if (new_len > target_idx)
96 if (WARN_ON(new_sz < PRIOMAP_MIN_SZ))
100 /* allocate & copy */
101 new = kzalloc(new_sz, GFP_KERNEL);
103 pr_warn("Unable to alloc new priomap!\n");
108 memcpy(new->priomap, old->priomap,
109 old->priomap_len * sizeof(old->priomap[0]));
111 new->priomap_len = new_len;
113 /* install the new priomap */
114 rcu_assign_pointer(dev->priomap, new);
120 static struct cgroup_subsys_state *cgrp_css_alloc(struct cgroup *cgrp)
122 struct cgroup_netprio_state *cs;
125 cs = kzalloc(sizeof(*cs), GFP_KERNEL);
127 return ERR_PTR(-ENOMEM);
129 if (cgrp->parent && cgrp_netprio_state(cgrp->parent)->prioidx)
132 ret = get_prioidx(&cs->prioidx);
134 pr_warn("No space in priority index array\n");
144 static void cgrp_css_free(struct cgroup *cgrp)
146 struct cgroup_netprio_state *cs;
147 struct net_device *dev;
148 struct netprio_map *map;
150 cs = cgrp_netprio_state(cgrp);
152 for_each_netdev(&init_net, dev) {
153 map = rtnl_dereference(dev->priomap);
154 if (map && cs->prioidx < map->priomap_len)
155 map->priomap[cs->prioidx] = 0;
158 put_prioidx(cs->prioidx);
162 static u64 read_prioidx(struct cgroup *cgrp, struct cftype *cft)
164 return (u64)cgrp_netprio_state(cgrp)->prioidx;
167 static int read_priomap(struct cgroup *cont, struct cftype *cft,
168 struct cgroup_map_cb *cb)
170 struct net_device *dev;
171 u32 prioidx = cgrp_netprio_state(cont)->prioidx;
173 struct netprio_map *map;
176 for_each_netdev_rcu(&init_net, dev) {
177 map = rcu_dereference(dev->priomap);
178 priority = (map && prioidx < map->priomap_len) ? map->priomap[prioidx] : 0;
179 cb->fill(cb, dev->name, priority);
185 static int write_priomap(struct cgroup *cgrp, struct cftype *cft,
188 u32 prioidx = cgrp_netprio_state(cgrp)->prioidx;
189 char devname[IFNAMSIZ + 1];
190 struct net_device *dev;
191 struct netprio_map *map;
195 if (sscanf(buffer, "%"__stringify(IFNAMSIZ)"s %u", devname, &prio) != 2)
198 dev = dev_get_by_name(&init_net, devname);
204 ret = extend_netdev_table(dev, prioidx);
208 map = rtnl_dereference(dev->priomap);
210 map->priomap[prioidx] = prio;
217 static int update_netprio(const void *v, struct file *file, unsigned n)
220 struct socket *sock = sock_from_file(file, &err);
222 sock->sk->sk_cgrp_prioidx = (u32)(unsigned long)v;
226 void net_prio_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
228 struct task_struct *p;
231 cgroup_taskset_for_each(p, cgrp, tset) {
233 v = (void *)(unsigned long)task_netprioidx(p);
234 iterate_fd(p->files, 0, update_netprio, v);
239 static struct cftype ss_files[] = {
242 .read_u64 = read_prioidx,
246 .read_map = read_priomap,
247 .write_string = write_priomap,
252 struct cgroup_subsys net_prio_subsys = {
254 .css_alloc = cgrp_css_alloc,
255 .css_free = cgrp_css_free,
256 .attach = net_prio_attach,
257 .subsys_id = net_prio_subsys_id,
258 .base_cftypes = ss_files,
259 .module = THIS_MODULE,
262 * net_prio has artificial limit on the number of cgroups and
263 * disallows nesting making it impossible to co-mount it with other
264 * hierarchical subsystems. Remove the artificially low PRIOIDX_SZ
265 * limit and properly nest configuration such that children follow
266 * their parents' configurations by default and are allowed to
267 * override and remove the following.
269 .broken_hierarchy = true,
272 static int netprio_device_event(struct notifier_block *unused,
273 unsigned long event, void *ptr)
275 struct net_device *dev = ptr;
276 struct netprio_map *old;
279 * Note this is called with rtnl_lock held so we have update side
280 * protection on our rcu assignments
284 case NETDEV_UNREGISTER:
285 old = rtnl_dereference(dev->priomap);
286 RCU_INIT_POINTER(dev->priomap, NULL);
294 static struct notifier_block netprio_device_notifier = {
295 .notifier_call = netprio_device_event
298 static int __init init_cgroup_netprio(void)
302 ret = cgroup_load_subsys(&net_prio_subsys);
306 register_netdevice_notifier(&netprio_device_notifier);
312 static void __exit exit_cgroup_netprio(void)
314 struct netprio_map *old;
315 struct net_device *dev;
317 unregister_netdevice_notifier(&netprio_device_notifier);
319 cgroup_unload_subsys(&net_prio_subsys);
322 for_each_netdev(&init_net, dev) {
323 old = rtnl_dereference(dev->priomap);
324 RCU_INIT_POINTER(dev->priomap, NULL);
331 module_init(init_cgroup_netprio);
332 module_exit(exit_cgroup_netprio);
333 MODULE_LICENSE("GPL v2");