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
33 * Extend @dev->priomap so that it's large enough to accommodate
34 * @target_idx. @dev->priomap.priomap_len > @target_idx after successful
35 * return. Must be called under rtnl lock.
37 static int extend_netdev_table(struct net_device *dev, u32 target_idx)
39 struct netprio_map *old, *new;
40 size_t new_sz, new_len;
42 /* is the existing priomap large enough? */
43 old = rtnl_dereference(dev->priomap);
44 if (old && old->priomap_len > target_idx)
48 * Determine the new size. Let's keep it power-of-two. We start
49 * from PRIOMAP_MIN_SZ and double it until it's large enough to
50 * accommodate @target_idx.
52 new_sz = PRIOMAP_MIN_SZ;
54 new_len = (new_sz - offsetof(struct netprio_map, priomap)) /
55 sizeof(new->priomap[0]);
56 if (new_len > target_idx)
60 if (WARN_ON(new_sz < PRIOMAP_MIN_SZ))
65 new = kzalloc(new_sz, GFP_KERNEL);
70 memcpy(new->priomap, old->priomap,
71 old->priomap_len * sizeof(old->priomap[0]));
73 new->priomap_len = new_len;
75 /* install the new priomap */
76 rcu_assign_pointer(dev->priomap, new);
83 * netprio_prio - return the effective netprio of a cgroup-net_device pair
84 * @css: css part of the target pair
85 * @dev: net_device part of the target pair
87 * Should be called under RCU read or rtnl lock.
89 static u32 netprio_prio(struct cgroup_subsys_state *css, struct net_device *dev)
91 struct netprio_map *map = rcu_dereference_rtnl(dev->priomap);
92 int id = css->cgroup->id;
94 if (map && id < map->priomap_len)
95 return map->priomap[id];
100 * netprio_set_prio - set netprio on a cgroup-net_device pair
101 * @css: css part of the target pair
102 * @dev: net_device part of the target pair
105 * Set netprio to @prio on @css-@dev pair. Should be called under rtnl
106 * lock and may fail under memory pressure for non-zero @prio.
108 static int netprio_set_prio(struct cgroup_subsys_state *css,
109 struct net_device *dev, u32 prio)
111 struct netprio_map *map;
112 int id = css->cgroup->id;
115 /* avoid extending priomap for zero writes */
116 map = rtnl_dereference(dev->priomap);
117 if (!prio && (!map || map->priomap_len <= id))
120 ret = extend_netdev_table(dev, id);
124 map = rtnl_dereference(dev->priomap);
125 map->priomap[id] = prio;
129 static struct cgroup_subsys_state *
130 cgrp_css_alloc(struct cgroup_subsys_state *parent_css)
132 struct cgroup_subsys_state *css;
134 css = kzalloc(sizeof(*css), GFP_KERNEL);
136 return ERR_PTR(-ENOMEM);
141 static int cgrp_css_online(struct cgroup_subsys_state *css)
143 struct cgroup_subsys_state *parent_css = css->parent;
144 struct net_device *dev;
152 * Inherit prios from the parent. As all prios are set during
153 * onlining, there is no need to clear them on offline.
155 for_each_netdev(&init_net, dev) {
156 u32 prio = netprio_prio(parent_css, dev);
158 ret = netprio_set_prio(css, dev, prio);
166 static void cgrp_css_free(struct cgroup_subsys_state *css)
171 static u64 read_prioidx(struct cgroup_subsys_state *css, struct cftype *cft)
173 return css->cgroup->id;
176 static int read_priomap(struct seq_file *sf, void *v)
178 struct net_device *dev;
181 for_each_netdev_rcu(&init_net, dev)
182 seq_printf(sf, "%s %u\n", dev->name,
183 netprio_prio(seq_css(sf), dev));
188 static ssize_t write_priomap(struct kernfs_open_file *of,
189 char *buf, size_t nbytes, loff_t off)
191 char devname[IFNAMSIZ + 1];
192 struct net_device *dev;
196 if (sscanf(buf, "%"__stringify(IFNAMSIZ)"s %u", devname, &prio) != 2)
199 dev = dev_get_by_name(&init_net, devname);
205 ret = netprio_set_prio(of_css(of), dev, prio);
209 return ret ?: nbytes;
212 static int update_netprio(const void *v, struct file *file, unsigned n)
215 struct socket *sock = sock_from_file(file, &err);
217 sock->sk->sk_cgrp_prioidx = (u32)(unsigned long)v;
221 static void net_prio_attach(struct cgroup_taskset *tset)
223 struct task_struct *p;
224 struct cgroup_subsys_state *css;
226 cgroup_taskset_for_each(p, css, tset) {
227 void *v = (void *)(unsigned long)css->cgroup->id;
230 iterate_fd(p->files, 0, update_netprio, v);
235 static struct cftype ss_files[] = {
238 .read_u64 = read_prioidx,
242 .seq_show = read_priomap,
243 .write = write_priomap,
248 struct cgroup_subsys net_prio_cgrp_subsys = {
249 .css_alloc = cgrp_css_alloc,
250 .css_online = cgrp_css_online,
251 .css_free = cgrp_css_free,
252 .attach = net_prio_attach,
253 .legacy_cftypes = ss_files,
256 static int netprio_device_event(struct notifier_block *unused,
257 unsigned long event, void *ptr)
259 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
260 struct netprio_map *old;
263 * Note this is called with rtnl_lock held so we have update side
264 * protection on our rcu assignments
268 case NETDEV_UNREGISTER:
269 old = rtnl_dereference(dev->priomap);
270 RCU_INIT_POINTER(dev->priomap, NULL);
278 static struct notifier_block netprio_device_notifier = {
279 .notifier_call = netprio_device_event
282 static int __init init_cgroup_netprio(void)
284 register_netdevice_notifier(&netprio_device_notifier);
288 subsys_initcall(init_cgroup_netprio);
289 MODULE_LICENSE("GPL v2");