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 | ||
e005d193 JP |
12 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
13 | ||
5bc1421e NH |
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> | |
25 | #include <net/sock.h> | |
26 | #include <net/netprio_cgroup.h> | |
27 | ||
406a3c63 JF |
28 | #include <linux/fdtable.h> |
29 | ||
5bc1421e NH |
30 | #define PRIOIDX_SZ 128 |
31 | ||
32 | static unsigned long prioidx_map[PRIOIDX_SZ]; | |
33 | static DEFINE_SPINLOCK(prioidx_map_lock); | |
34 | static atomic_t max_prioidx = ATOMIC_INIT(0); | |
35 | ||
36 | static inline struct cgroup_netprio_state *cgrp_netprio_state(struct cgroup *cgrp) | |
37 | { | |
38 | return container_of(cgroup_subsys_state(cgrp, net_prio_subsys_id), | |
39 | struct cgroup_netprio_state, css); | |
40 | } | |
41 | ||
42 | static int get_prioidx(u32 *prio) | |
43 | { | |
44 | unsigned long flags; | |
45 | u32 prioidx; | |
46 | ||
47 | spin_lock_irqsave(&prioidx_map_lock, flags); | |
48 | prioidx = find_first_zero_bit(prioidx_map, sizeof(unsigned long) * PRIOIDX_SZ); | |
5962b35c NH |
49 | if (prioidx == sizeof(unsigned long) * PRIOIDX_SZ) { |
50 | spin_unlock_irqrestore(&prioidx_map_lock, flags); | |
51 | return -ENOSPC; | |
52 | } | |
5bc1421e | 53 | set_bit(prioidx, prioidx_map); |
b761c9b1 G |
54 | if (atomic_read(&max_prioidx) < prioidx) |
55 | atomic_set(&max_prioidx, prioidx); | |
5bc1421e | 56 | spin_unlock_irqrestore(&prioidx_map_lock, flags); |
5bc1421e NH |
57 | *prio = prioidx; |
58 | return 0; | |
59 | } | |
60 | ||
61 | static void put_prioidx(u32 idx) | |
62 | { | |
63 | unsigned long flags; | |
64 | ||
65 | spin_lock_irqsave(&prioidx_map_lock, flags); | |
66 | clear_bit(idx, prioidx_map); | |
67 | spin_unlock_irqrestore(&prioidx_map_lock, flags); | |
68 | } | |
69 | ||
ef209f15 | 70 | static int extend_netdev_table(struct net_device *dev, u32 new_len) |
5bc1421e NH |
71 | { |
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; | |
5bc1421e NH |
76 | |
77 | old_priomap = rtnl_dereference(dev->priomap); | |
78 | ||
79 | if (!new_priomap) { | |
e005d193 | 80 | pr_warn("Unable to alloc new priomap!\n"); |
ef209f15 | 81 | return -ENOMEM; |
5bc1421e NH |
82 | } |
83 | ||
f05ba7fc SB |
84 | if (old_priomap) |
85 | memcpy(new_priomap->priomap, old_priomap->priomap, | |
86 | old_priomap->priomap_len * | |
87 | sizeof(old_priomap->priomap[0])); | |
5bc1421e NH |
88 | |
89 | new_priomap->priomap_len = new_len; | |
90 | ||
91 | rcu_assign_pointer(dev->priomap, new_priomap); | |
92 | if (old_priomap) | |
93 | kfree_rcu(old_priomap, rcu); | |
ef209f15 G |
94 | return 0; |
95 | } | |
96 | ||
97 | static int write_update_netdev_table(struct net_device *dev) | |
98 | { | |
99 | int ret = 0; | |
100 | u32 max_len; | |
101 | struct netprio_map *map; | |
102 | ||
ef209f15 G |
103 | max_len = atomic_read(&max_prioidx) + 1; |
104 | map = rtnl_dereference(dev->priomap); | |
105 | if (!map || map->priomap_len < max_len) | |
106 | ret = extend_netdev_table(dev, max_len); | |
ef209f15 G |
107 | |
108 | return ret; | |
5bc1421e NH |
109 | } |
110 | ||
761b3ef5 | 111 | static struct cgroup_subsys_state *cgrp_create(struct cgroup *cgrp) |
5bc1421e NH |
112 | { |
113 | struct cgroup_netprio_state *cs; | |
ef209f15 | 114 | int ret = -EINVAL; |
5bc1421e NH |
115 | |
116 | cs = kzalloc(sizeof(*cs), GFP_KERNEL); | |
117 | if (!cs) | |
118 | return ERR_PTR(-ENOMEM); | |
119 | ||
ef209f15 G |
120 | if (cgrp->parent && cgrp_netprio_state(cgrp->parent)->prioidx) |
121 | goto out; | |
5bc1421e NH |
122 | |
123 | ret = get_prioidx(&cs->prioidx); | |
ef209f15 | 124 | if (ret < 0) { |
e005d193 | 125 | pr_warn("No space in priority index array\n"); |
ef209f15 G |
126 | goto out; |
127 | } | |
128 | ||
5bc1421e | 129 | return &cs->css; |
ef209f15 G |
130 | out: |
131 | kfree(cs); | |
132 | return ERR_PTR(ret); | |
5bc1421e NH |
133 | } |
134 | ||
761b3ef5 | 135 | static void cgrp_destroy(struct cgroup *cgrp) |
5bc1421e NH |
136 | { |
137 | struct cgroup_netprio_state *cs; | |
138 | struct net_device *dev; | |
139 | struct netprio_map *map; | |
140 | ||
141 | cs = cgrp_netprio_state(cgrp); | |
142 | rtnl_lock(); | |
143 | for_each_netdev(&init_net, dev) { | |
144 | map = rtnl_dereference(dev->priomap); | |
91c68ce2 | 145 | if (map && cs->prioidx < map->priomap_len) |
5bc1421e NH |
146 | map->priomap[cs->prioidx] = 0; |
147 | } | |
148 | rtnl_unlock(); | |
149 | put_prioidx(cs->prioidx); | |
150 | kfree(cs); | |
151 | } | |
152 | ||
153 | static u64 read_prioidx(struct cgroup *cgrp, struct cftype *cft) | |
154 | { | |
155 | return (u64)cgrp_netprio_state(cgrp)->prioidx; | |
156 | } | |
157 | ||
158 | static int read_priomap(struct cgroup *cont, struct cftype *cft, | |
159 | struct cgroup_map_cb *cb) | |
160 | { | |
161 | struct net_device *dev; | |
162 | u32 prioidx = cgrp_netprio_state(cont)->prioidx; | |
163 | u32 priority; | |
164 | struct netprio_map *map; | |
165 | ||
166 | rcu_read_lock(); | |
167 | for_each_netdev_rcu(&init_net, dev) { | |
168 | map = rcu_dereference(dev->priomap); | |
91c68ce2 | 169 | priority = (map && prioidx < map->priomap_len) ? map->priomap[prioidx] : 0; |
5bc1421e NH |
170 | cb->fill(cb, dev->name, priority); |
171 | } | |
172 | rcu_read_unlock(); | |
173 | return 0; | |
174 | } | |
175 | ||
176 | static int write_priomap(struct cgroup *cgrp, struct cftype *cft, | |
177 | const char *buffer) | |
178 | { | |
179 | char *devname = kstrdup(buffer, GFP_KERNEL); | |
180 | int ret = -EINVAL; | |
181 | u32 prioidx = cgrp_netprio_state(cgrp)->prioidx; | |
182 | unsigned long priority; | |
183 | char *priostr; | |
184 | struct net_device *dev; | |
185 | struct netprio_map *map; | |
186 | ||
187 | if (!devname) | |
188 | return -ENOMEM; | |
189 | ||
190 | /* | |
191 | * Minimally sized valid priomap string | |
192 | */ | |
193 | if (strlen(devname) < 3) | |
194 | goto out_free_devname; | |
195 | ||
196 | priostr = strstr(devname, " "); | |
197 | if (!priostr) | |
198 | goto out_free_devname; | |
199 | ||
200 | /* | |
201 | *Separate the devname from the associated priority | |
0f307323 | 202 | *and advance the priostr pointer to the priority value |
5bc1421e NH |
203 | */ |
204 | *priostr = '\0'; | |
205 | priostr++; | |
206 | ||
207 | /* | |
208 | * If the priostr points to NULL, we're at the end of the passed | |
209 | * in string, and its not a valid write | |
210 | */ | |
211 | if (*priostr == '\0') | |
212 | goto out_free_devname; | |
213 | ||
214 | ret = kstrtoul(priostr, 10, &priority); | |
215 | if (ret < 0) | |
216 | goto out_free_devname; | |
217 | ||
218 | ret = -ENODEV; | |
219 | ||
220 | dev = dev_get_by_name(&init_net, devname); | |
221 | if (!dev) | |
222 | goto out_free_devname; | |
223 | ||
476ad154 | 224 | rtnl_lock(); |
ef209f15 G |
225 | ret = write_update_netdev_table(dev); |
226 | if (ret < 0) | |
227 | goto out_put_dev; | |
228 | ||
476ad154 | 229 | map = rtnl_dereference(dev->priomap); |
5bc1421e NH |
230 | if (map) |
231 | map->priomap[prioidx] = priority; | |
ef209f15 G |
232 | |
233 | out_put_dev: | |
476ad154 | 234 | rtnl_unlock(); |
5bc1421e NH |
235 | dev_put(dev); |
236 | ||
237 | out_free_devname: | |
238 | kfree(devname); | |
239 | return ret; | |
240 | } | |
241 | ||
c3c073f8 AV |
242 | static int update_netprio(const void *v, struct file *file, unsigned n) |
243 | { | |
244 | int err; | |
245 | struct socket *sock = sock_from_file(file, &err); | |
246 | if (sock) | |
247 | sock->sk->sk_cgrp_prioidx = (u32)(unsigned long)v; | |
248 | return 0; | |
249 | } | |
250 | ||
406a3c63 JF |
251 | void net_prio_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) |
252 | { | |
253 | struct task_struct *p; | |
c3c073f8 | 254 | void *v; |
406a3c63 JF |
255 | |
256 | cgroup_taskset_for_each(p, cgrp, tset) { | |
406a3c63 | 257 | task_lock(p); |
c3c073f8 AV |
258 | v = (void *)(unsigned long)task_netprioidx(p); |
259 | iterate_fd(p->files, 0, update_netprio, v); | |
406a3c63 JF |
260 | task_unlock(p); |
261 | } | |
406a3c63 JF |
262 | } |
263 | ||
5bc1421e NH |
264 | static struct cftype ss_files[] = { |
265 | { | |
266 | .name = "prioidx", | |
267 | .read_u64 = read_prioidx, | |
268 | }, | |
269 | { | |
270 | .name = "ifpriomap", | |
271 | .read_map = read_priomap, | |
272 | .write_string = write_priomap, | |
273 | }, | |
4baf6e33 | 274 | { } /* terminate */ |
5bc1421e NH |
275 | }; |
276 | ||
676f7c8f TH |
277 | struct cgroup_subsys net_prio_subsys = { |
278 | .name = "net_prio", | |
279 | .create = cgrp_create, | |
280 | .destroy = cgrp_destroy, | |
406a3c63 | 281 | .attach = net_prio_attach, |
676f7c8f | 282 | .subsys_id = net_prio_subsys_id, |
4baf6e33 | 283 | .base_cftypes = ss_files, |
8c7f6edb TH |
284 | .module = THIS_MODULE, |
285 | ||
286 | /* | |
287 | * net_prio has artificial limit on the number of cgroups and | |
288 | * disallows nesting making it impossible to co-mount it with other | |
289 | * hierarchical subsystems. Remove the artificially low PRIOIDX_SZ | |
290 | * limit and properly nest configuration such that children follow | |
291 | * their parents' configurations by default and are allowed to | |
292 | * override and remove the following. | |
293 | */ | |
294 | .broken_hierarchy = true, | |
676f7c8f | 295 | }; |
5bc1421e NH |
296 | |
297 | static int netprio_device_event(struct notifier_block *unused, | |
298 | unsigned long event, void *ptr) | |
299 | { | |
300 | struct net_device *dev = ptr; | |
301 | struct netprio_map *old; | |
5bc1421e NH |
302 | |
303 | /* | |
304 | * Note this is called with rtnl_lock held so we have update side | |
305 | * protection on our rcu assignments | |
306 | */ | |
307 | ||
308 | switch (event) { | |
5bc1421e NH |
309 | case NETDEV_UNREGISTER: |
310 | old = rtnl_dereference(dev->priomap); | |
2cfa5a04 | 311 | RCU_INIT_POINTER(dev->priomap, NULL); |
5bc1421e NH |
312 | if (old) |
313 | kfree_rcu(old, rcu); | |
314 | break; | |
315 | } | |
316 | return NOTIFY_DONE; | |
317 | } | |
318 | ||
319 | static struct notifier_block netprio_device_notifier = { | |
320 | .notifier_call = netprio_device_event | |
321 | }; | |
322 | ||
323 | static int __init init_cgroup_netprio(void) | |
324 | { | |
325 | int ret; | |
326 | ||
327 | ret = cgroup_load_subsys(&net_prio_subsys); | |
328 | if (ret) | |
329 | goto out; | |
5bc1421e NH |
330 | |
331 | register_netdevice_notifier(&netprio_device_notifier); | |
332 | ||
333 | out: | |
334 | return ret; | |
335 | } | |
336 | ||
337 | static void __exit exit_cgroup_netprio(void) | |
338 | { | |
339 | struct netprio_map *old; | |
340 | struct net_device *dev; | |
341 | ||
342 | unregister_netdevice_notifier(&netprio_device_notifier); | |
343 | ||
344 | cgroup_unload_subsys(&net_prio_subsys); | |
345 | ||
5bc1421e NH |
346 | rtnl_lock(); |
347 | for_each_netdev(&init_net, dev) { | |
348 | old = rtnl_dereference(dev->priomap); | |
2cfa5a04 | 349 | RCU_INIT_POINTER(dev->priomap, NULL); |
5bc1421e NH |
350 | if (old) |
351 | kfree_rcu(old, rcu); | |
352 | } | |
353 | rtnl_unlock(); | |
354 | } | |
355 | ||
356 | module_init(init_cgroup_netprio); | |
357 | module_exit(exit_cgroup_netprio); | |
358 | MODULE_LICENSE("GPL v2"); |