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 PRIOIDX_SZ 128
32 static unsigned long prioidx_map[PRIOIDX_SZ];
33 static DEFINE_SPINLOCK(prioidx_map_lock);
34 static atomic_t max_prioidx = ATOMIC_INIT(0);
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 if (atomic_read(&max_prioidx) < prioidx)
55 atomic_set(&max_prioidx, prioidx);
56 spin_unlock_irqrestore(&prioidx_map_lock, flags);
61 static void put_prioidx(u32 idx)
65 spin_lock_irqsave(&prioidx_map_lock, flags);
66 clear_bit(idx, prioidx_map);
67 spin_unlock_irqrestore(&prioidx_map_lock, flags);
70 static int extend_netdev_table(struct net_device *dev, u32 new_len)
72 size_t new_size = sizeof(struct netprio_map) +
73 ((sizeof(u32) * new_len));
74 struct netprio_map *new_priomap = kzalloc(new_size, GFP_KERNEL);
75 struct netprio_map *old_priomap;
78 old_priomap = rtnl_dereference(dev->priomap);
81 pr_warn("Unable to alloc new priomap!\n");
86 old_priomap && (i < old_priomap->priomap_len);
88 new_priomap->priomap[i] = old_priomap->priomap[i];
90 new_priomap->priomap_len = new_len;
92 rcu_assign_pointer(dev->priomap, new_priomap);
94 kfree_rcu(old_priomap, rcu);
98 static int write_update_netdev_table(struct net_device *dev)
102 struct netprio_map *map;
104 max_len = atomic_read(&max_prioidx) + 1;
105 map = rtnl_dereference(dev->priomap);
106 if (!map || map->priomap_len < max_len)
107 ret = extend_netdev_table(dev, max_len);
112 static int update_netdev_tables(void)
115 struct net_device *dev;
117 struct netprio_map *map;
120 max_len = atomic_read(&max_prioidx) + 1;
121 for_each_netdev(&init_net, dev) {
122 map = rtnl_dereference(dev->priomap);
124 * don't allocate priomap if we didn't
125 * change net_prio.ifpriomap (map == NULL),
126 * this will speed up skb_update_prio.
128 if (map && map->priomap_len < max_len) {
129 ret = extend_netdev_table(dev, max_len);
138 static struct cgroup_subsys_state *cgrp_create(struct cgroup *cgrp)
140 struct cgroup_netprio_state *cs;
143 cs = kzalloc(sizeof(*cs), GFP_KERNEL);
145 return ERR_PTR(-ENOMEM);
147 if (cgrp->parent && cgrp_netprio_state(cgrp->parent)->prioidx)
150 ret = get_prioidx(&cs->prioidx);
152 pr_warn("No space in priority index array\n");
156 ret = update_netdev_tables();
158 put_prioidx(cs->prioidx);
168 static void cgrp_destroy(struct cgroup *cgrp)
170 struct cgroup_netprio_state *cs;
171 struct net_device *dev;
172 struct netprio_map *map;
174 cs = cgrp_netprio_state(cgrp);
176 for_each_netdev(&init_net, dev) {
177 map = rtnl_dereference(dev->priomap);
178 if (map && cs->prioidx < map->priomap_len)
179 map->priomap[cs->prioidx] = 0;
182 put_prioidx(cs->prioidx);
186 static u64 read_prioidx(struct cgroup *cgrp, struct cftype *cft)
188 return (u64)cgrp_netprio_state(cgrp)->prioidx;
191 static int read_priomap(struct cgroup *cont, struct cftype *cft,
192 struct cgroup_map_cb *cb)
194 struct net_device *dev;
195 u32 prioidx = cgrp_netprio_state(cont)->prioidx;
197 struct netprio_map *map;
200 for_each_netdev_rcu(&init_net, dev) {
201 map = rcu_dereference(dev->priomap);
202 priority = (map && prioidx < map->priomap_len) ? map->priomap[prioidx] : 0;
203 cb->fill(cb, dev->name, priority);
209 static int write_priomap(struct cgroup *cgrp, struct cftype *cft,
212 char *devname = kstrdup(buffer, GFP_KERNEL);
214 u32 prioidx = cgrp_netprio_state(cgrp)->prioidx;
215 unsigned long priority;
217 struct net_device *dev;
218 struct netprio_map *map;
224 * Minimally sized valid priomap string
226 if (strlen(devname) < 3)
227 goto out_free_devname;
229 priostr = strstr(devname, " ");
231 goto out_free_devname;
234 *Separate the devname from the associated priority
235 *and advance the priostr pointer to the priority value
241 * If the priostr points to NULL, we're at the end of the passed
242 * in string, and its not a valid write
244 if (*priostr == '\0')
245 goto out_free_devname;
247 ret = kstrtoul(priostr, 10, &priority);
249 goto out_free_devname;
253 dev = dev_get_by_name(&init_net, devname);
255 goto out_free_devname;
258 ret = write_update_netdev_table(dev);
262 map = rtnl_dereference(dev->priomap);
264 map->priomap[prioidx] = priority;
275 void net_prio_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
277 struct task_struct *p;
279 cgroup_taskset_for_each(p, cgrp, tset) {
282 struct files_struct *files;
291 spin_lock(&files->file_lock);
292 fdt = files_fdtable(files);
293 for (fd = 0; fd < fdt->max_fds; fd++) {
298 file = fcheck_files(files, fd);
302 sock = sock_from_file(file, &err);
304 sock_update_netprioidx(sock->sk, p);
306 spin_unlock(&files->file_lock);
311 static struct cftype ss_files[] = {
314 .read_u64 = read_prioidx,
318 .read_map = read_priomap,
319 .write_string = write_priomap,
324 struct cgroup_subsys net_prio_subsys = {
326 .create = cgrp_create,
327 .destroy = cgrp_destroy,
328 .attach = net_prio_attach,
329 #ifdef CONFIG_NETPRIO_CGROUP
330 .subsys_id = net_prio_subsys_id,
332 .base_cftypes = ss_files,
333 .module = THIS_MODULE
336 static int netprio_device_event(struct notifier_block *unused,
337 unsigned long event, void *ptr)
339 struct net_device *dev = ptr;
340 struct netprio_map *old;
343 * Note this is called with rtnl_lock held so we have update side
344 * protection on our rcu assignments
348 case NETDEV_UNREGISTER:
349 old = rtnl_dereference(dev->priomap);
350 RCU_INIT_POINTER(dev->priomap, NULL);
358 static struct notifier_block netprio_device_notifier = {
359 .notifier_call = netprio_device_event
362 static int __init init_cgroup_netprio(void)
366 ret = cgroup_load_subsys(&net_prio_subsys);
369 #ifndef CONFIG_NETPRIO_CGROUP
371 net_prio_subsys_id = net_prio_subsys.subsys_id;
374 register_netdevice_notifier(&netprio_device_notifier);
380 static void __exit exit_cgroup_netprio(void)
382 struct netprio_map *old;
383 struct net_device *dev;
385 unregister_netdevice_notifier(&netprio_device_notifier);
387 cgroup_unload_subsys(&net_prio_subsys);
389 #ifndef CONFIG_NETPRIO_CGROUP
390 net_prio_subsys_id = -1;
395 for_each_netdev(&init_net, dev) {
396 old = rtnl_dereference(dev->priomap);
397 RCU_INIT_POINTER(dev->priomap, NULL);
404 module_init(init_cgroup_netprio);
405 module_exit(exit_cgroup_netprio);
406 MODULE_LICENSE("GPL v2");