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 | ||
4a6ee25c | 30 | #define PRIOMAP_MIN_SZ 128 |
5bc1421e NH |
31 | |
32 | static inline struct cgroup_netprio_state *cgrp_netprio_state(struct cgroup *cgrp) | |
33 | { | |
34 | return container_of(cgroup_subsys_state(cgrp, net_prio_subsys_id), | |
35 | struct cgroup_netprio_state, css); | |
36 | } | |
37 | ||
4a6ee25c TH |
38 | /* |
39 | * Extend @dev->priomap so that it's large enough to accomodate | |
40 | * @target_idx. @dev->priomap.priomap_len > @target_idx after successful | |
41 | * return. Must be called under rtnl lock. | |
42 | */ | |
43 | static int extend_netdev_table(struct net_device *dev, u32 target_idx) | |
5bc1421e | 44 | { |
4a6ee25c TH |
45 | struct netprio_map *old, *new; |
46 | size_t new_sz, new_len; | |
5bc1421e | 47 | |
4a6ee25c | 48 | /* is the existing priomap large enough? */ |
52bca930 | 49 | old = rtnl_dereference(dev->priomap); |
4a6ee25c TH |
50 | if (old && old->priomap_len > target_idx) |
51 | return 0; | |
52 | ||
53 | /* | |
54 | * Determine the new size. Let's keep it power-of-two. We start | |
55 | * from PRIOMAP_MIN_SZ and double it until it's large enough to | |
56 | * accommodate @target_idx. | |
57 | */ | |
58 | new_sz = PRIOMAP_MIN_SZ; | |
59 | while (true) { | |
60 | new_len = (new_sz - offsetof(struct netprio_map, priomap)) / | |
61 | sizeof(new->priomap[0]); | |
62 | if (new_len > target_idx) | |
63 | break; | |
64 | new_sz *= 2; | |
65 | /* overflowed? */ | |
66 | if (WARN_ON(new_sz < PRIOMAP_MIN_SZ)) | |
67 | return -ENOSPC; | |
68 | } | |
5bc1421e | 69 | |
4a6ee25c TH |
70 | /* allocate & copy */ |
71 | new = kzalloc(new_sz, GFP_KERNEL); | |
52bca930 | 72 | if (!new) { |
e005d193 | 73 | pr_warn("Unable to alloc new priomap!\n"); |
ef209f15 | 74 | return -ENOMEM; |
5bc1421e NH |
75 | } |
76 | ||
52bca930 TH |
77 | if (old) |
78 | memcpy(new->priomap, old->priomap, | |
79 | old->priomap_len * sizeof(old->priomap[0])); | |
5bc1421e | 80 | |
52bca930 | 81 | new->priomap_len = new_len; |
5bc1421e | 82 | |
4a6ee25c | 83 | /* install the new priomap */ |
52bca930 TH |
84 | rcu_assign_pointer(dev->priomap, new); |
85 | if (old) | |
86 | kfree_rcu(old, rcu); | |
ef209f15 G |
87 | return 0; |
88 | } | |
89 | ||
666b0ebe TH |
90 | /** |
91 | * netprio_prio - return the effective netprio of a cgroup-net_device pair | |
92 | * @cgrp: cgroup part of the target pair | |
93 | * @dev: net_device part of the target pair | |
94 | * | |
95 | * Should be called under RCU read or rtnl lock. | |
96 | */ | |
97 | static u32 netprio_prio(struct cgroup *cgrp, struct net_device *dev) | |
98 | { | |
99 | struct netprio_map *map = rcu_dereference_rtnl(dev->priomap); | |
100 | ||
101 | if (map && cgrp->id < map->priomap_len) | |
102 | return map->priomap[cgrp->id]; | |
103 | return 0; | |
104 | } | |
105 | ||
106 | /** | |
107 | * netprio_set_prio - set netprio on a cgroup-net_device pair | |
108 | * @cgrp: cgroup part of the target pair | |
109 | * @dev: net_device part of the target pair | |
110 | * @prio: prio to set | |
111 | * | |
112 | * Set netprio to @prio on @cgrp-@dev pair. Should be called under rtnl | |
113 | * lock and may fail under memory pressure for non-zero @prio. | |
114 | */ | |
115 | static int netprio_set_prio(struct cgroup *cgrp, struct net_device *dev, | |
116 | u32 prio) | |
117 | { | |
118 | struct netprio_map *map; | |
119 | int ret; | |
120 | ||
121 | /* avoid extending priomap for zero writes */ | |
122 | map = rtnl_dereference(dev->priomap); | |
123 | if (!prio && (!map || map->priomap_len <= cgrp->id)) | |
124 | return 0; | |
125 | ||
126 | ret = extend_netdev_table(dev, cgrp->id); | |
127 | if (ret) | |
128 | return ret; | |
129 | ||
130 | map = rtnl_dereference(dev->priomap); | |
131 | map->priomap[cgrp->id] = prio; | |
132 | return 0; | |
133 | } | |
134 | ||
92fb9748 | 135 | static struct cgroup_subsys_state *cgrp_css_alloc(struct cgroup *cgrp) |
5bc1421e NH |
136 | { |
137 | struct cgroup_netprio_state *cs; | |
88d642fa TH |
138 | |
139 | if (cgrp->parent && cgrp->parent->id) | |
140 | return ERR_PTR(-EINVAL); | |
5bc1421e NH |
141 | |
142 | cs = kzalloc(sizeof(*cs), GFP_KERNEL); | |
143 | if (!cs) | |
144 | return ERR_PTR(-ENOMEM); | |
145 | ||
5bc1421e NH |
146 | return &cs->css; |
147 | } | |
148 | ||
92fb9748 | 149 | static void cgrp_css_free(struct cgroup *cgrp) |
5bc1421e | 150 | { |
88d642fa | 151 | struct cgroup_netprio_state *cs = cgrp_netprio_state(cgrp); |
5bc1421e | 152 | struct net_device *dev; |
5bc1421e | 153 | |
5bc1421e | 154 | rtnl_lock(); |
666b0ebe TH |
155 | for_each_netdev(&init_net, dev) |
156 | WARN_ON_ONCE(netprio_set_prio(cgrp, dev, 0)); | |
5bc1421e | 157 | rtnl_unlock(); |
5bc1421e NH |
158 | kfree(cs); |
159 | } | |
160 | ||
161 | static u64 read_prioidx(struct cgroup *cgrp, struct cftype *cft) | |
162 | { | |
88d642fa | 163 | return cgrp->id; |
5bc1421e NH |
164 | } |
165 | ||
166 | static int read_priomap(struct cgroup *cont, struct cftype *cft, | |
167 | struct cgroup_map_cb *cb) | |
168 | { | |
169 | struct net_device *dev; | |
5bc1421e NH |
170 | |
171 | rcu_read_lock(); | |
666b0ebe TH |
172 | for_each_netdev_rcu(&init_net, dev) |
173 | cb->fill(cb, dev->name, netprio_prio(cont, dev)); | |
5bc1421e NH |
174 | rcu_read_unlock(); |
175 | return 0; | |
176 | } | |
177 | ||
178 | static int write_priomap(struct cgroup *cgrp, struct cftype *cft, | |
179 | const char *buffer) | |
180 | { | |
6d5759dd | 181 | char devname[IFNAMSIZ + 1]; |
5bc1421e | 182 | struct net_device *dev; |
6d5759dd TH |
183 | u32 prio; |
184 | int ret; | |
5bc1421e | 185 | |
6d5759dd TH |
186 | if (sscanf(buffer, "%"__stringify(IFNAMSIZ)"s %u", devname, &prio) != 2) |
187 | return -EINVAL; | |
5bc1421e NH |
188 | |
189 | dev = dev_get_by_name(&init_net, devname); | |
190 | if (!dev) | |
6d5759dd | 191 | return -ENODEV; |
5bc1421e | 192 | |
476ad154 | 193 | rtnl_lock(); |
6d5759dd | 194 | |
666b0ebe | 195 | ret = netprio_set_prio(cgrp, dev, prio); |
ef209f15 | 196 | |
476ad154 | 197 | rtnl_unlock(); |
5bc1421e | 198 | dev_put(dev); |
5bc1421e NH |
199 | return ret; |
200 | } | |
201 | ||
c3c073f8 AV |
202 | static int update_netprio(const void *v, struct file *file, unsigned n) |
203 | { | |
204 | int err; | |
205 | struct socket *sock = sock_from_file(file, &err); | |
206 | if (sock) | |
207 | sock->sk->sk_cgrp_prioidx = (u32)(unsigned long)v; | |
208 | return 0; | |
209 | } | |
210 | ||
406a3c63 JF |
211 | void net_prio_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) |
212 | { | |
213 | struct task_struct *p; | |
c3c073f8 | 214 | void *v; |
406a3c63 JF |
215 | |
216 | cgroup_taskset_for_each(p, cgrp, tset) { | |
406a3c63 | 217 | task_lock(p); |
c3c073f8 AV |
218 | v = (void *)(unsigned long)task_netprioidx(p); |
219 | iterate_fd(p->files, 0, update_netprio, v); | |
406a3c63 JF |
220 | task_unlock(p); |
221 | } | |
406a3c63 JF |
222 | } |
223 | ||
5bc1421e NH |
224 | static struct cftype ss_files[] = { |
225 | { | |
226 | .name = "prioidx", | |
227 | .read_u64 = read_prioidx, | |
228 | }, | |
229 | { | |
230 | .name = "ifpriomap", | |
231 | .read_map = read_priomap, | |
232 | .write_string = write_priomap, | |
233 | }, | |
4baf6e33 | 234 | { } /* terminate */ |
5bc1421e NH |
235 | }; |
236 | ||
676f7c8f TH |
237 | struct cgroup_subsys net_prio_subsys = { |
238 | .name = "net_prio", | |
92fb9748 TH |
239 | .css_alloc = cgrp_css_alloc, |
240 | .css_free = cgrp_css_free, | |
406a3c63 | 241 | .attach = net_prio_attach, |
676f7c8f | 242 | .subsys_id = net_prio_subsys_id, |
4baf6e33 | 243 | .base_cftypes = ss_files, |
8c7f6edb TH |
244 | .module = THIS_MODULE, |
245 | ||
246 | /* | |
247 | * net_prio has artificial limit on the number of cgroups and | |
248 | * disallows nesting making it impossible to co-mount it with other | |
249 | * hierarchical subsystems. Remove the artificially low PRIOIDX_SZ | |
250 | * limit and properly nest configuration such that children follow | |
251 | * their parents' configurations by default and are allowed to | |
252 | * override and remove the following. | |
253 | */ | |
254 | .broken_hierarchy = true, | |
676f7c8f | 255 | }; |
5bc1421e NH |
256 | |
257 | static int netprio_device_event(struct notifier_block *unused, | |
258 | unsigned long event, void *ptr) | |
259 | { | |
260 | struct net_device *dev = ptr; | |
261 | struct netprio_map *old; | |
5bc1421e NH |
262 | |
263 | /* | |
264 | * Note this is called with rtnl_lock held so we have update side | |
265 | * protection on our rcu assignments | |
266 | */ | |
267 | ||
268 | switch (event) { | |
5bc1421e NH |
269 | case NETDEV_UNREGISTER: |
270 | old = rtnl_dereference(dev->priomap); | |
2cfa5a04 | 271 | RCU_INIT_POINTER(dev->priomap, NULL); |
5bc1421e NH |
272 | if (old) |
273 | kfree_rcu(old, rcu); | |
274 | break; | |
275 | } | |
276 | return NOTIFY_DONE; | |
277 | } | |
278 | ||
279 | static struct notifier_block netprio_device_notifier = { | |
280 | .notifier_call = netprio_device_event | |
281 | }; | |
282 | ||
283 | static int __init init_cgroup_netprio(void) | |
284 | { | |
285 | int ret; | |
286 | ||
287 | ret = cgroup_load_subsys(&net_prio_subsys); | |
288 | if (ret) | |
289 | goto out; | |
5bc1421e NH |
290 | |
291 | register_netdevice_notifier(&netprio_device_notifier); | |
292 | ||
293 | out: | |
294 | return ret; | |
295 | } | |
296 | ||
297 | static void __exit exit_cgroup_netprio(void) | |
298 | { | |
299 | struct netprio_map *old; | |
300 | struct net_device *dev; | |
301 | ||
302 | unregister_netdevice_notifier(&netprio_device_notifier); | |
303 | ||
304 | cgroup_unload_subsys(&net_prio_subsys); | |
305 | ||
5bc1421e NH |
306 | rtnl_lock(); |
307 | for_each_netdev(&init_net, dev) { | |
308 | old = rtnl_dereference(dev->priomap); | |
2cfa5a04 | 309 | RCU_INIT_POINTER(dev->priomap, NULL); |
5bc1421e NH |
310 | if (old) |
311 | kfree_rcu(old, rcu); | |
312 | } | |
313 | rtnl_unlock(); | |
314 | } | |
315 | ||
316 | module_init(init_cgroup_netprio); | |
317 | module_exit(exit_cgroup_netprio); | |
318 | MODULE_LICENSE("GPL v2"); |