xref: /openbmc/linux/net/core/netprio_cgroup.c (revision 95c96174)
1 /*
2  * net/core/netprio_cgroup.c	Priority Control Group
3  *
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.
8  *
9  * Authors:	Neil Horman <nhorman@tuxdriver.com>
10  */
11 
12 #include <linux/module.h>
13 #include <linux/slab.h>
14 #include <linux/types.h>
15 #include <linux/string.h>
16 #include <linux/errno.h>
17 #include <linux/skbuff.h>
18 #include <linux/cgroup.h>
19 #include <linux/rcupdate.h>
20 #include <linux/atomic.h>
21 #include <net/rtnetlink.h>
22 #include <net/pkt_cls.h>
23 #include <net/sock.h>
24 #include <net/netprio_cgroup.h>
25 
26 static struct cgroup_subsys_state *cgrp_create(struct cgroup *cgrp);
27 static void cgrp_destroy(struct cgroup *cgrp);
28 static int cgrp_populate(struct cgroup_subsys *ss, struct cgroup *cgrp);
29 
30 struct cgroup_subsys net_prio_subsys = {
31 	.name		= "net_prio",
32 	.create		= cgrp_create,
33 	.destroy	= cgrp_destroy,
34 	.populate	= cgrp_populate,
35 #ifdef CONFIG_NETPRIO_CGROUP
36 	.subsys_id	= net_prio_subsys_id,
37 #endif
38 	.module		= THIS_MODULE
39 };
40 
41 #define PRIOIDX_SZ 128
42 
43 static unsigned long prioidx_map[PRIOIDX_SZ];
44 static DEFINE_SPINLOCK(prioidx_map_lock);
45 static atomic_t max_prioidx = ATOMIC_INIT(0);
46 
47 static inline struct cgroup_netprio_state *cgrp_netprio_state(struct cgroup *cgrp)
48 {
49 	return container_of(cgroup_subsys_state(cgrp, net_prio_subsys_id),
50 			    struct cgroup_netprio_state, css);
51 }
52 
53 static int get_prioidx(u32 *prio)
54 {
55 	unsigned long flags;
56 	u32 prioidx;
57 
58 	spin_lock_irqsave(&prioidx_map_lock, flags);
59 	prioidx = find_first_zero_bit(prioidx_map, sizeof(unsigned long) * PRIOIDX_SZ);
60 	if (prioidx == sizeof(unsigned long) * PRIOIDX_SZ) {
61 		spin_unlock_irqrestore(&prioidx_map_lock, flags);
62 		return -ENOSPC;
63 	}
64 	set_bit(prioidx, prioidx_map);
65 	spin_unlock_irqrestore(&prioidx_map_lock, flags);
66 	atomic_set(&max_prioidx, prioidx);
67 	*prio = prioidx;
68 	return 0;
69 }
70 
71 static void put_prioidx(u32 idx)
72 {
73 	unsigned long flags;
74 
75 	spin_lock_irqsave(&prioidx_map_lock, flags);
76 	clear_bit(idx, prioidx_map);
77 	spin_unlock_irqrestore(&prioidx_map_lock, flags);
78 }
79 
80 static void extend_netdev_table(struct net_device *dev, u32 new_len)
81 {
82 	size_t new_size = sizeof(struct netprio_map) +
83 			   ((sizeof(u32) * new_len));
84 	struct netprio_map *new_priomap = kzalloc(new_size, GFP_KERNEL);
85 	struct netprio_map *old_priomap;
86 	int i;
87 
88 	old_priomap  = rtnl_dereference(dev->priomap);
89 
90 	if (!new_priomap) {
91 		printk(KERN_WARNING "Unable to alloc new priomap!\n");
92 		return;
93 	}
94 
95 	for (i = 0;
96 	     old_priomap && (i < old_priomap->priomap_len);
97 	     i++)
98 		new_priomap->priomap[i] = old_priomap->priomap[i];
99 
100 	new_priomap->priomap_len = new_len;
101 
102 	rcu_assign_pointer(dev->priomap, new_priomap);
103 	if (old_priomap)
104 		kfree_rcu(old_priomap, rcu);
105 }
106 
107 static void update_netdev_tables(void)
108 {
109 	struct net_device *dev;
110 	u32 max_len = atomic_read(&max_prioidx) + 1;
111 	struct netprio_map *map;
112 
113 	rtnl_lock();
114 	for_each_netdev(&init_net, dev) {
115 		map = rtnl_dereference(dev->priomap);
116 		if ((!map) ||
117 		    (map->priomap_len < max_len))
118 			extend_netdev_table(dev, max_len);
119 	}
120 	rtnl_unlock();
121 }
122 
123 static struct cgroup_subsys_state *cgrp_create(struct cgroup *cgrp)
124 {
125 	struct cgroup_netprio_state *cs;
126 	int ret;
127 
128 	cs = kzalloc(sizeof(*cs), GFP_KERNEL);
129 	if (!cs)
130 		return ERR_PTR(-ENOMEM);
131 
132 	if (cgrp->parent && cgrp_netprio_state(cgrp->parent)->prioidx) {
133 		kfree(cs);
134 		return ERR_PTR(-EINVAL);
135 	}
136 
137 	ret = get_prioidx(&cs->prioidx);
138 	if (ret != 0) {
139 		printk(KERN_WARNING "No space in priority index array\n");
140 		kfree(cs);
141 		return ERR_PTR(ret);
142 	}
143 
144 	return &cs->css;
145 }
146 
147 static void cgrp_destroy(struct cgroup *cgrp)
148 {
149 	struct cgroup_netprio_state *cs;
150 	struct net_device *dev;
151 	struct netprio_map *map;
152 
153 	cs = cgrp_netprio_state(cgrp);
154 	rtnl_lock();
155 	for_each_netdev(&init_net, dev) {
156 		map = rtnl_dereference(dev->priomap);
157 		if (map)
158 			map->priomap[cs->prioidx] = 0;
159 	}
160 	rtnl_unlock();
161 	put_prioidx(cs->prioidx);
162 	kfree(cs);
163 }
164 
165 static u64 read_prioidx(struct cgroup *cgrp, struct cftype *cft)
166 {
167 	return (u64)cgrp_netprio_state(cgrp)->prioidx;
168 }
169 
170 static int read_priomap(struct cgroup *cont, struct cftype *cft,
171 			struct cgroup_map_cb *cb)
172 {
173 	struct net_device *dev;
174 	u32 prioidx = cgrp_netprio_state(cont)->prioidx;
175 	u32 priority;
176 	struct netprio_map *map;
177 
178 	rcu_read_lock();
179 	for_each_netdev_rcu(&init_net, dev) {
180 		map = rcu_dereference(dev->priomap);
181 		priority = map ? map->priomap[prioidx] : 0;
182 		cb->fill(cb, dev->name, priority);
183 	}
184 	rcu_read_unlock();
185 	return 0;
186 }
187 
188 static int write_priomap(struct cgroup *cgrp, struct cftype *cft,
189 			 const char *buffer)
190 {
191 	char *devname = kstrdup(buffer, GFP_KERNEL);
192 	int ret = -EINVAL;
193 	u32 prioidx = cgrp_netprio_state(cgrp)->prioidx;
194 	unsigned long priority;
195 	char *priostr;
196 	struct net_device *dev;
197 	struct netprio_map *map;
198 
199 	if (!devname)
200 		return -ENOMEM;
201 
202 	/*
203 	 * Minimally sized valid priomap string
204 	 */
205 	if (strlen(devname) < 3)
206 		goto out_free_devname;
207 
208 	priostr = strstr(devname, " ");
209 	if (!priostr)
210 		goto out_free_devname;
211 
212 	/*
213 	 *Separate the devname from the associated priority
214 	 *and advance the priostr poitner to the priority value
215 	 */
216 	*priostr = '\0';
217 	priostr++;
218 
219 	/*
220 	 * If the priostr points to NULL, we're at the end of the passed
221 	 * in string, and its not a valid write
222 	 */
223 	if (*priostr == '\0')
224 		goto out_free_devname;
225 
226 	ret = kstrtoul(priostr, 10, &priority);
227 	if (ret < 0)
228 		goto out_free_devname;
229 
230 	ret = -ENODEV;
231 
232 	dev = dev_get_by_name(&init_net, devname);
233 	if (!dev)
234 		goto out_free_devname;
235 
236 	update_netdev_tables();
237 	ret = 0;
238 	rcu_read_lock();
239 	map = rcu_dereference(dev->priomap);
240 	if (map)
241 		map->priomap[prioidx] = priority;
242 	rcu_read_unlock();
243 	dev_put(dev);
244 
245 out_free_devname:
246 	kfree(devname);
247 	return ret;
248 }
249 
250 static struct cftype ss_files[] = {
251 	{
252 		.name = "prioidx",
253 		.read_u64 = read_prioidx,
254 	},
255 	{
256 		.name = "ifpriomap",
257 		.read_map = read_priomap,
258 		.write_string = write_priomap,
259 	},
260 };
261 
262 static int cgrp_populate(struct cgroup_subsys *ss, struct cgroup *cgrp)
263 {
264 	return cgroup_add_files(cgrp, ss, ss_files, ARRAY_SIZE(ss_files));
265 }
266 
267 static int netprio_device_event(struct notifier_block *unused,
268 				unsigned long event, void *ptr)
269 {
270 	struct net_device *dev = ptr;
271 	struct netprio_map *old;
272 
273 	/*
274 	 * Note this is called with rtnl_lock held so we have update side
275 	 * protection on our rcu assignments
276 	 */
277 
278 	switch (event) {
279 	case NETDEV_UNREGISTER:
280 		old = rtnl_dereference(dev->priomap);
281 		RCU_INIT_POINTER(dev->priomap, NULL);
282 		if (old)
283 			kfree_rcu(old, rcu);
284 		break;
285 	}
286 	return NOTIFY_DONE;
287 }
288 
289 static struct notifier_block netprio_device_notifier = {
290 	.notifier_call = netprio_device_event
291 };
292 
293 static int __init init_cgroup_netprio(void)
294 {
295 	int ret;
296 
297 	ret = cgroup_load_subsys(&net_prio_subsys);
298 	if (ret)
299 		goto out;
300 #ifndef CONFIG_NETPRIO_CGROUP
301 	smp_wmb();
302 	net_prio_subsys_id = net_prio_subsys.subsys_id;
303 #endif
304 
305 	register_netdevice_notifier(&netprio_device_notifier);
306 
307 out:
308 	return ret;
309 }
310 
311 static void __exit exit_cgroup_netprio(void)
312 {
313 	struct netprio_map *old;
314 	struct net_device *dev;
315 
316 	unregister_netdevice_notifier(&netprio_device_notifier);
317 
318 	cgroup_unload_subsys(&net_prio_subsys);
319 
320 #ifndef CONFIG_NETPRIO_CGROUP
321 	net_prio_subsys_id = -1;
322 	synchronize_rcu();
323 #endif
324 
325 	rtnl_lock();
326 	for_each_netdev(&init_net, dev) {
327 		old = rtnl_dereference(dev->priomap);
328 		RCU_INIT_POINTER(dev->priomap, NULL);
329 		if (old)
330 			kfree_rcu(old, rcu);
331 	}
332 	rtnl_unlock();
333 }
334 
335 module_init(init_cgroup_netprio);
336 module_exit(exit_cgroup_netprio);
337 MODULE_LICENSE("GPL v2");
338