Commit | Line | Data |
---|---|---|
5bc1421e NH |
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_subsys *ss, | |
27 | struct cgroup *cgrp); | |
28 | static void cgrp_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp); | |
29 | static int cgrp_populate(struct cgroup_subsys *ss, struct cgroup *cgrp); | |
30 | ||
0221cd51 | 31 | struct cgroup_subsys net_prio_subsys = { |
5bc1421e NH |
32 | .name = "net_prio", |
33 | .create = cgrp_create, | |
34 | .destroy = cgrp_destroy, | |
35 | .populate = cgrp_populate, | |
36 | #ifdef CONFIG_NETPRIO_CGROUP | |
37 | .subsys_id = net_prio_subsys_id, | |
38 | #endif | |
39 | .module = THIS_MODULE | |
40 | }; | |
41 | ||
42 | #define PRIOIDX_SZ 128 | |
43 | ||
44 | static unsigned long prioidx_map[PRIOIDX_SZ]; | |
45 | static DEFINE_SPINLOCK(prioidx_map_lock); | |
46 | static atomic_t max_prioidx = ATOMIC_INIT(0); | |
47 | ||
48 | static inline struct cgroup_netprio_state *cgrp_netprio_state(struct cgroup *cgrp) | |
49 | { | |
50 | return container_of(cgroup_subsys_state(cgrp, net_prio_subsys_id), | |
51 | struct cgroup_netprio_state, css); | |
52 | } | |
53 | ||
54 | static int get_prioidx(u32 *prio) | |
55 | { | |
56 | unsigned long flags; | |
57 | u32 prioidx; | |
58 | ||
59 | spin_lock_irqsave(&prioidx_map_lock, flags); | |
60 | prioidx = find_first_zero_bit(prioidx_map, sizeof(unsigned long) * PRIOIDX_SZ); | |
5962b35c NH |
61 | if (prioidx == sizeof(unsigned long) * PRIOIDX_SZ) { |
62 | spin_unlock_irqrestore(&prioidx_map_lock, flags); | |
63 | return -ENOSPC; | |
64 | } | |
5bc1421e NH |
65 | set_bit(prioidx, prioidx_map); |
66 | spin_unlock_irqrestore(&prioidx_map_lock, flags); | |
5bc1421e NH |
67 | atomic_set(&max_prioidx, prioidx); |
68 | *prio = prioidx; | |
69 | return 0; | |
70 | } | |
71 | ||
72 | static void put_prioidx(u32 idx) | |
73 | { | |
74 | unsigned long flags; | |
75 | ||
76 | spin_lock_irqsave(&prioidx_map_lock, flags); | |
77 | clear_bit(idx, prioidx_map); | |
78 | spin_unlock_irqrestore(&prioidx_map_lock, flags); | |
79 | } | |
80 | ||
81 | static void extend_netdev_table(struct net_device *dev, u32 new_len) | |
82 | { | |
83 | size_t new_size = sizeof(struct netprio_map) + | |
84 | ((sizeof(u32) * new_len)); | |
85 | struct netprio_map *new_priomap = kzalloc(new_size, GFP_KERNEL); | |
86 | struct netprio_map *old_priomap; | |
87 | int i; | |
88 | ||
89 | old_priomap = rtnl_dereference(dev->priomap); | |
90 | ||
91 | if (!new_priomap) { | |
92 | printk(KERN_WARNING "Unable to alloc new priomap!\n"); | |
93 | return; | |
94 | } | |
95 | ||
96 | for (i = 0; | |
97 | old_priomap && (i < old_priomap->priomap_len); | |
98 | i++) | |
99 | new_priomap->priomap[i] = old_priomap->priomap[i]; | |
100 | ||
101 | new_priomap->priomap_len = new_len; | |
102 | ||
103 | rcu_assign_pointer(dev->priomap, new_priomap); | |
104 | if (old_priomap) | |
105 | kfree_rcu(old_priomap, rcu); | |
106 | } | |
107 | ||
108 | static void update_netdev_tables(void) | |
109 | { | |
110 | struct net_device *dev; | |
111 | u32 max_len = atomic_read(&max_prioidx); | |
112 | struct netprio_map *map; | |
113 | ||
114 | rtnl_lock(); | |
115 | for_each_netdev(&init_net, dev) { | |
116 | map = rtnl_dereference(dev->priomap); | |
117 | if ((!map) || | |
118 | (map->priomap_len < max_len)) | |
119 | extend_netdev_table(dev, max_len); | |
120 | } | |
121 | rtnl_unlock(); | |
122 | } | |
123 | ||
124 | static struct cgroup_subsys_state *cgrp_create(struct cgroup_subsys *ss, | |
125 | struct cgroup *cgrp) | |
126 | { | |
127 | struct cgroup_netprio_state *cs; | |
128 | int ret; | |
129 | ||
130 | cs = kzalloc(sizeof(*cs), GFP_KERNEL); | |
131 | if (!cs) | |
132 | return ERR_PTR(-ENOMEM); | |
133 | ||
134 | if (cgrp->parent && cgrp_netprio_state(cgrp->parent)->prioidx) { | |
135 | kfree(cs); | |
136 | return ERR_PTR(-EINVAL); | |
137 | } | |
138 | ||
139 | ret = get_prioidx(&cs->prioidx); | |
140 | if (ret != 0) { | |
141 | printk(KERN_WARNING "No space in priority index array\n"); | |
142 | kfree(cs); | |
143 | return ERR_PTR(ret); | |
144 | } | |
145 | ||
146 | return &cs->css; | |
147 | } | |
148 | ||
149 | static void cgrp_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
150 | { | |
151 | struct cgroup_netprio_state *cs; | |
152 | struct net_device *dev; | |
153 | struct netprio_map *map; | |
154 | ||
155 | cs = cgrp_netprio_state(cgrp); | |
156 | rtnl_lock(); | |
157 | for_each_netdev(&init_net, dev) { | |
158 | map = rtnl_dereference(dev->priomap); | |
159 | if (map) | |
160 | map->priomap[cs->prioidx] = 0; | |
161 | } | |
162 | rtnl_unlock(); | |
163 | put_prioidx(cs->prioidx); | |
164 | kfree(cs); | |
165 | } | |
166 | ||
167 | static u64 read_prioidx(struct cgroup *cgrp, struct cftype *cft) | |
168 | { | |
169 | return (u64)cgrp_netprio_state(cgrp)->prioidx; | |
170 | } | |
171 | ||
172 | static int read_priomap(struct cgroup *cont, struct cftype *cft, | |
173 | struct cgroup_map_cb *cb) | |
174 | { | |
175 | struct net_device *dev; | |
176 | u32 prioidx = cgrp_netprio_state(cont)->prioidx; | |
177 | u32 priority; | |
178 | struct netprio_map *map; | |
179 | ||
180 | rcu_read_lock(); | |
181 | for_each_netdev_rcu(&init_net, dev) { | |
182 | map = rcu_dereference(dev->priomap); | |
183 | priority = map ? map->priomap[prioidx] : 0; | |
184 | cb->fill(cb, dev->name, priority); | |
185 | } | |
186 | rcu_read_unlock(); | |
187 | return 0; | |
188 | } | |
189 | ||
190 | static int write_priomap(struct cgroup *cgrp, struct cftype *cft, | |
191 | const char *buffer) | |
192 | { | |
193 | char *devname = kstrdup(buffer, GFP_KERNEL); | |
194 | int ret = -EINVAL; | |
195 | u32 prioidx = cgrp_netprio_state(cgrp)->prioidx; | |
196 | unsigned long priority; | |
197 | char *priostr; | |
198 | struct net_device *dev; | |
199 | struct netprio_map *map; | |
200 | ||
201 | if (!devname) | |
202 | return -ENOMEM; | |
203 | ||
204 | /* | |
205 | * Minimally sized valid priomap string | |
206 | */ | |
207 | if (strlen(devname) < 3) | |
208 | goto out_free_devname; | |
209 | ||
210 | priostr = strstr(devname, " "); | |
211 | if (!priostr) | |
212 | goto out_free_devname; | |
213 | ||
214 | /* | |
215 | *Separate the devname from the associated priority | |
216 | *and advance the priostr poitner to the priority value | |
217 | */ | |
218 | *priostr = '\0'; | |
219 | priostr++; | |
220 | ||
221 | /* | |
222 | * If the priostr points to NULL, we're at the end of the passed | |
223 | * in string, and its not a valid write | |
224 | */ | |
225 | if (*priostr == '\0') | |
226 | goto out_free_devname; | |
227 | ||
228 | ret = kstrtoul(priostr, 10, &priority); | |
229 | if (ret < 0) | |
230 | goto out_free_devname; | |
231 | ||
232 | ret = -ENODEV; | |
233 | ||
234 | dev = dev_get_by_name(&init_net, devname); | |
235 | if (!dev) | |
236 | goto out_free_devname; | |
237 | ||
238 | update_netdev_tables(); | |
239 | ret = 0; | |
240 | rcu_read_lock(); | |
241 | map = rcu_dereference(dev->priomap); | |
242 | if (map) | |
243 | map->priomap[prioidx] = priority; | |
244 | rcu_read_unlock(); | |
245 | dev_put(dev); | |
246 | ||
247 | out_free_devname: | |
248 | kfree(devname); | |
249 | return ret; | |
250 | } | |
251 | ||
252 | static struct cftype ss_files[] = { | |
253 | { | |
254 | .name = "prioidx", | |
255 | .read_u64 = read_prioidx, | |
256 | }, | |
257 | { | |
258 | .name = "ifpriomap", | |
259 | .read_map = read_priomap, | |
260 | .write_string = write_priomap, | |
261 | }, | |
262 | }; | |
263 | ||
264 | static int cgrp_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
265 | { | |
266 | return cgroup_add_files(cgrp, ss, ss_files, ARRAY_SIZE(ss_files)); | |
267 | } | |
268 | ||
269 | static int netprio_device_event(struct notifier_block *unused, | |
270 | unsigned long event, void *ptr) | |
271 | { | |
272 | struct net_device *dev = ptr; | |
273 | struct netprio_map *old; | |
274 | u32 max_len = atomic_read(&max_prioidx); | |
275 | ||
276 | /* | |
277 | * Note this is called with rtnl_lock held so we have update side | |
278 | * protection on our rcu assignments | |
279 | */ | |
280 | ||
281 | switch (event) { | |
282 | ||
283 | case NETDEV_REGISTER: | |
284 | if (max_len) | |
285 | extend_netdev_table(dev, max_len); | |
286 | break; | |
287 | case NETDEV_UNREGISTER: | |
288 | old = rtnl_dereference(dev->priomap); | |
2cfa5a04 | 289 | RCU_INIT_POINTER(dev->priomap, NULL); |
5bc1421e NH |
290 | if (old) |
291 | kfree_rcu(old, rcu); | |
292 | break; | |
293 | } | |
294 | return NOTIFY_DONE; | |
295 | } | |
296 | ||
297 | static struct notifier_block netprio_device_notifier = { | |
298 | .notifier_call = netprio_device_event | |
299 | }; | |
300 | ||
301 | static int __init init_cgroup_netprio(void) | |
302 | { | |
303 | int ret; | |
304 | ||
305 | ret = cgroup_load_subsys(&net_prio_subsys); | |
306 | if (ret) | |
307 | goto out; | |
308 | #ifndef CONFIG_NETPRIO_CGROUP | |
309 | smp_wmb(); | |
310 | net_prio_subsys_id = net_prio_subsys.subsys_id; | |
311 | #endif | |
312 | ||
313 | register_netdevice_notifier(&netprio_device_notifier); | |
314 | ||
315 | out: | |
316 | return ret; | |
317 | } | |
318 | ||
319 | static void __exit exit_cgroup_netprio(void) | |
320 | { | |
321 | struct netprio_map *old; | |
322 | struct net_device *dev; | |
323 | ||
324 | unregister_netdevice_notifier(&netprio_device_notifier); | |
325 | ||
326 | cgroup_unload_subsys(&net_prio_subsys); | |
327 | ||
328 | #ifndef CONFIG_NETPRIO_CGROUP | |
329 | net_prio_subsys_id = -1; | |
330 | synchronize_rcu(); | |
331 | #endif | |
332 | ||
333 | rtnl_lock(); | |
334 | for_each_netdev(&init_net, dev) { | |
335 | old = rtnl_dereference(dev->priomap); | |
2cfa5a04 | 336 | RCU_INIT_POINTER(dev->priomap, NULL); |
5bc1421e NH |
337 | if (old) |
338 | kfree_rcu(old, rcu); | |
339 | } | |
340 | rtnl_unlock(); | |
341 | } | |
342 | ||
343 | module_init(init_cgroup_netprio); | |
344 | module_exit(exit_cgroup_netprio); | |
345 | MODULE_LICENSE("GPL v2"); |