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 | ||
31 | struct cgroup_subsys net_prio_subsys = { | |
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); | |
61 | set_bit(prioidx, prioidx_map); | |
62 | spin_unlock_irqrestore(&prioidx_map_lock, flags); | |
63 | if (prioidx == sizeof(unsigned long) * PRIOIDX_SZ) | |
64 | return -ENOSPC; | |
65 | ||
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); | |
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_subsys *ss, | |
124 | struct cgroup *cgrp) | |
125 | { | |
126 | struct cgroup_netprio_state *cs; | |
127 | int ret; | |
128 | ||
129 | cs = kzalloc(sizeof(*cs), GFP_KERNEL); | |
130 | if (!cs) | |
131 | return ERR_PTR(-ENOMEM); | |
132 | ||
133 | if (cgrp->parent && cgrp_netprio_state(cgrp->parent)->prioidx) { | |
134 | kfree(cs); | |
135 | return ERR_PTR(-EINVAL); | |
136 | } | |
137 | ||
138 | ret = get_prioidx(&cs->prioidx); | |
139 | if (ret != 0) { | |
140 | printk(KERN_WARNING "No space in priority index array\n"); | |
141 | kfree(cs); | |
142 | return ERR_PTR(ret); | |
143 | } | |
144 | ||
145 | return &cs->css; | |
146 | } | |
147 | ||
148 | static void cgrp_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
149 | { | |
150 | struct cgroup_netprio_state *cs; | |
151 | struct net_device *dev; | |
152 | struct netprio_map *map; | |
153 | ||
154 | cs = cgrp_netprio_state(cgrp); | |
155 | rtnl_lock(); | |
156 | for_each_netdev(&init_net, dev) { | |
157 | map = rtnl_dereference(dev->priomap); | |
158 | if (map) | |
159 | map->priomap[cs->prioidx] = 0; | |
160 | } | |
161 | rtnl_unlock(); | |
162 | put_prioidx(cs->prioidx); | |
163 | kfree(cs); | |
164 | } | |
165 | ||
166 | static u64 read_prioidx(struct cgroup *cgrp, struct cftype *cft) | |
167 | { | |
168 | return (u64)cgrp_netprio_state(cgrp)->prioidx; | |
169 | } | |
170 | ||
171 | static int read_priomap(struct cgroup *cont, struct cftype *cft, | |
172 | struct cgroup_map_cb *cb) | |
173 | { | |
174 | struct net_device *dev; | |
175 | u32 prioidx = cgrp_netprio_state(cont)->prioidx; | |
176 | u32 priority; | |
177 | struct netprio_map *map; | |
178 | ||
179 | rcu_read_lock(); | |
180 | for_each_netdev_rcu(&init_net, dev) { | |
181 | map = rcu_dereference(dev->priomap); | |
182 | priority = map ? map->priomap[prioidx] : 0; | |
183 | cb->fill(cb, dev->name, priority); | |
184 | } | |
185 | rcu_read_unlock(); | |
186 | return 0; | |
187 | } | |
188 | ||
189 | static int write_priomap(struct cgroup *cgrp, struct cftype *cft, | |
190 | const char *buffer) | |
191 | { | |
192 | char *devname = kstrdup(buffer, GFP_KERNEL); | |
193 | int ret = -EINVAL; | |
194 | u32 prioidx = cgrp_netprio_state(cgrp)->prioidx; | |
195 | unsigned long priority; | |
196 | char *priostr; | |
197 | struct net_device *dev; | |
198 | struct netprio_map *map; | |
199 | ||
200 | if (!devname) | |
201 | return -ENOMEM; | |
202 | ||
203 | /* | |
204 | * Minimally sized valid priomap string | |
205 | */ | |
206 | if (strlen(devname) < 3) | |
207 | goto out_free_devname; | |
208 | ||
209 | priostr = strstr(devname, " "); | |
210 | if (!priostr) | |
211 | goto out_free_devname; | |
212 | ||
213 | /* | |
214 | *Separate the devname from the associated priority | |
215 | *and advance the priostr poitner to the priority value | |
216 | */ | |
217 | *priostr = '\0'; | |
218 | priostr++; | |
219 | ||
220 | /* | |
221 | * If the priostr points to NULL, we're at the end of the passed | |
222 | * in string, and its not a valid write | |
223 | */ | |
224 | if (*priostr == '\0') | |
225 | goto out_free_devname; | |
226 | ||
227 | ret = kstrtoul(priostr, 10, &priority); | |
228 | if (ret < 0) | |
229 | goto out_free_devname; | |
230 | ||
231 | ret = -ENODEV; | |
232 | ||
233 | dev = dev_get_by_name(&init_net, devname); | |
234 | if (!dev) | |
235 | goto out_free_devname; | |
236 | ||
237 | update_netdev_tables(); | |
238 | ret = 0; | |
239 | rcu_read_lock(); | |
240 | map = rcu_dereference(dev->priomap); | |
241 | if (map) | |
242 | map->priomap[prioidx] = priority; | |
243 | rcu_read_unlock(); | |
244 | dev_put(dev); | |
245 | ||
246 | out_free_devname: | |
247 | kfree(devname); | |
248 | return ret; | |
249 | } | |
250 | ||
251 | static struct cftype ss_files[] = { | |
252 | { | |
253 | .name = "prioidx", | |
254 | .read_u64 = read_prioidx, | |
255 | }, | |
256 | { | |
257 | .name = "ifpriomap", | |
258 | .read_map = read_priomap, | |
259 | .write_string = write_priomap, | |
260 | }, | |
261 | }; | |
262 | ||
263 | static int cgrp_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
264 | { | |
265 | return cgroup_add_files(cgrp, ss, ss_files, ARRAY_SIZE(ss_files)); | |
266 | } | |
267 | ||
268 | static int netprio_device_event(struct notifier_block *unused, | |
269 | unsigned long event, void *ptr) | |
270 | { | |
271 | struct net_device *dev = ptr; | |
272 | struct netprio_map *old; | |
273 | u32 max_len = atomic_read(&max_prioidx); | |
274 | ||
275 | /* | |
276 | * Note this is called with rtnl_lock held so we have update side | |
277 | * protection on our rcu assignments | |
278 | */ | |
279 | ||
280 | switch (event) { | |
281 | ||
282 | case NETDEV_REGISTER: | |
283 | if (max_len) | |
284 | extend_netdev_table(dev, max_len); | |
285 | break; | |
286 | case NETDEV_UNREGISTER: | |
287 | old = rtnl_dereference(dev->priomap); | |
288 | rcu_assign_pointer(dev->priomap, NULL); | |
289 | if (old) | |
290 | kfree_rcu(old, rcu); | |
291 | break; | |
292 | } | |
293 | return NOTIFY_DONE; | |
294 | } | |
295 | ||
296 | static struct notifier_block netprio_device_notifier = { | |
297 | .notifier_call = netprio_device_event | |
298 | }; | |
299 | ||
300 | static int __init init_cgroup_netprio(void) | |
301 | { | |
302 | int ret; | |
303 | ||
304 | ret = cgroup_load_subsys(&net_prio_subsys); | |
305 | if (ret) | |
306 | goto out; | |
307 | #ifndef CONFIG_NETPRIO_CGROUP | |
308 | smp_wmb(); | |
309 | net_prio_subsys_id = net_prio_subsys.subsys_id; | |
310 | #endif | |
311 | ||
312 | register_netdevice_notifier(&netprio_device_notifier); | |
313 | ||
314 | out: | |
315 | return ret; | |
316 | } | |
317 | ||
318 | static void __exit exit_cgroup_netprio(void) | |
319 | { | |
320 | struct netprio_map *old; | |
321 | struct net_device *dev; | |
322 | ||
323 | unregister_netdevice_notifier(&netprio_device_notifier); | |
324 | ||
325 | cgroup_unload_subsys(&net_prio_subsys); | |
326 | ||
327 | #ifndef CONFIG_NETPRIO_CGROUP | |
328 | net_prio_subsys_id = -1; | |
329 | synchronize_rcu(); | |
330 | #endif | |
331 | ||
332 | rtnl_lock(); | |
333 | for_each_netdev(&init_net, dev) { | |
334 | old = rtnl_dereference(dev->priomap); | |
335 | rcu_assign_pointer(dev->priomap, NULL); | |
336 | if (old) | |
337 | kfree_rcu(old, rcu); | |
338 | } | |
339 | rtnl_unlock(); | |
340 | } | |
341 | ||
342 | module_init(init_cgroup_netprio); | |
343 | module_exit(exit_cgroup_netprio); | |
344 | MODULE_LICENSE("GPL v2"); |