new helper: iterate_fd()
[linux-2.6-block.git] / net / core / netprio_cgroup.c
CommitLineData
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
32static unsigned long prioidx_map[PRIOIDX_SZ];
33static DEFINE_SPINLOCK(prioidx_map_lock);
34static atomic_t max_prioidx = ATOMIC_INIT(0);
35
36static 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
42static 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
61static 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 70static 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;
76 int i;
77
78 old_priomap = rtnl_dereference(dev->priomap);
79
80 if (!new_priomap) {
e005d193 81 pr_warn("Unable to alloc new priomap!\n");
ef209f15 82 return -ENOMEM;
5bc1421e
NH
83 }
84
85 for (i = 0;
86 old_priomap && (i < old_priomap->priomap_len);
87 i++)
88 new_priomap->priomap[i] = old_priomap->priomap[i];
89
90 new_priomap->priomap_len = new_len;
91
92 rcu_assign_pointer(dev->priomap, new_priomap);
93 if (old_priomap)
94 kfree_rcu(old_priomap, rcu);
ef209f15
G
95 return 0;
96}
97
98static int write_update_netdev_table(struct net_device *dev)
99{
100 int ret = 0;
101 u32 max_len;
102 struct netprio_map *map;
103
ef209f15
G
104 max_len = atomic_read(&max_prioidx) + 1;
105 map = rtnl_dereference(dev->priomap);
106 if (!map || map->priomap_len < max_len)
107 ret = extend_netdev_table(dev, max_len);
ef209f15
G
108
109 return ret;
5bc1421e
NH
110}
111
ef209f15 112static int update_netdev_tables(void)
5bc1421e 113{
ef209f15 114 int ret = 0;
5bc1421e 115 struct net_device *dev;
ef209f15 116 u32 max_len;
5bc1421e
NH
117 struct netprio_map *map;
118
119 rtnl_lock();
ef209f15 120 max_len = atomic_read(&max_prioidx) + 1;
5bc1421e
NH
121 for_each_netdev(&init_net, dev) {
122 map = rtnl_dereference(dev->priomap);
ef209f15
G
123 /*
124 * don't allocate priomap if we didn't
125 * change net_prio.ifpriomap (map == NULL),
126 * this will speed up skb_update_prio.
127 */
128 if (map && map->priomap_len < max_len) {
129 ret = extend_netdev_table(dev, max_len);
130 if (ret < 0)
131 break;
132 }
5bc1421e
NH
133 }
134 rtnl_unlock();
ef209f15 135 return ret;
5bc1421e
NH
136}
137
761b3ef5 138static struct cgroup_subsys_state *cgrp_create(struct cgroup *cgrp)
5bc1421e
NH
139{
140 struct cgroup_netprio_state *cs;
ef209f15 141 int ret = -EINVAL;
5bc1421e
NH
142
143 cs = kzalloc(sizeof(*cs), GFP_KERNEL);
144 if (!cs)
145 return ERR_PTR(-ENOMEM);
146
ef209f15
G
147 if (cgrp->parent && cgrp_netprio_state(cgrp->parent)->prioidx)
148 goto out;
5bc1421e
NH
149
150 ret = get_prioidx(&cs->prioidx);
ef209f15 151 if (ret < 0) {
e005d193 152 pr_warn("No space in priority index array\n");
ef209f15
G
153 goto out;
154 }
155
156 ret = update_netdev_tables();
157 if (ret < 0) {
158 put_prioidx(cs->prioidx);
159 goto out;
5bc1421e
NH
160 }
161
162 return &cs->css;
ef209f15
G
163out:
164 kfree(cs);
165 return ERR_PTR(ret);
5bc1421e
NH
166}
167
761b3ef5 168static void cgrp_destroy(struct cgroup *cgrp)
5bc1421e
NH
169{
170 struct cgroup_netprio_state *cs;
171 struct net_device *dev;
172 struct netprio_map *map;
173
174 cs = cgrp_netprio_state(cgrp);
175 rtnl_lock();
176 for_each_netdev(&init_net, dev) {
177 map = rtnl_dereference(dev->priomap);
91c68ce2 178 if (map && cs->prioidx < map->priomap_len)
5bc1421e
NH
179 map->priomap[cs->prioidx] = 0;
180 }
181 rtnl_unlock();
182 put_prioidx(cs->prioidx);
183 kfree(cs);
184}
185
186static u64 read_prioidx(struct cgroup *cgrp, struct cftype *cft)
187{
188 return (u64)cgrp_netprio_state(cgrp)->prioidx;
189}
190
191static int read_priomap(struct cgroup *cont, struct cftype *cft,
192 struct cgroup_map_cb *cb)
193{
194 struct net_device *dev;
195 u32 prioidx = cgrp_netprio_state(cont)->prioidx;
196 u32 priority;
197 struct netprio_map *map;
198
199 rcu_read_lock();
200 for_each_netdev_rcu(&init_net, dev) {
201 map = rcu_dereference(dev->priomap);
91c68ce2 202 priority = (map && prioidx < map->priomap_len) ? map->priomap[prioidx] : 0;
5bc1421e
NH
203 cb->fill(cb, dev->name, priority);
204 }
205 rcu_read_unlock();
206 return 0;
207}
208
209static int write_priomap(struct cgroup *cgrp, struct cftype *cft,
210 const char *buffer)
211{
212 char *devname = kstrdup(buffer, GFP_KERNEL);
213 int ret = -EINVAL;
214 u32 prioidx = cgrp_netprio_state(cgrp)->prioidx;
215 unsigned long priority;
216 char *priostr;
217 struct net_device *dev;
218 struct netprio_map *map;
219
220 if (!devname)
221 return -ENOMEM;
222
223 /*
224 * Minimally sized valid priomap string
225 */
226 if (strlen(devname) < 3)
227 goto out_free_devname;
228
229 priostr = strstr(devname, " ");
230 if (!priostr)
231 goto out_free_devname;
232
233 /*
234 *Separate the devname from the associated priority
0f307323 235 *and advance the priostr pointer to the priority value
5bc1421e
NH
236 */
237 *priostr = '\0';
238 priostr++;
239
240 /*
241 * If the priostr points to NULL, we're at the end of the passed
242 * in string, and its not a valid write
243 */
244 if (*priostr == '\0')
245 goto out_free_devname;
246
247 ret = kstrtoul(priostr, 10, &priority);
248 if (ret < 0)
249 goto out_free_devname;
250
251 ret = -ENODEV;
252
253 dev = dev_get_by_name(&init_net, devname);
254 if (!dev)
255 goto out_free_devname;
256
476ad154 257 rtnl_lock();
ef209f15
G
258 ret = write_update_netdev_table(dev);
259 if (ret < 0)
260 goto out_put_dev;
261
476ad154 262 map = rtnl_dereference(dev->priomap);
5bc1421e
NH
263 if (map)
264 map->priomap[prioidx] = priority;
ef209f15
G
265
266out_put_dev:
476ad154 267 rtnl_unlock();
5bc1421e
NH
268 dev_put(dev);
269
270out_free_devname:
271 kfree(devname);
272 return ret;
273}
274
c3c073f8
AV
275static int update_netprio(const void *v, struct file *file, unsigned n)
276{
277 int err;
278 struct socket *sock = sock_from_file(file, &err);
279 if (sock)
280 sock->sk->sk_cgrp_prioidx = (u32)(unsigned long)v;
281 return 0;
282}
283
406a3c63
JF
284void net_prio_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
285{
286 struct task_struct *p;
c3c073f8 287 void *v;
406a3c63
JF
288
289 cgroup_taskset_for_each(p, cgrp, tset) {
406a3c63 290 task_lock(p);
c3c073f8
AV
291 v = (void *)(unsigned long)task_netprioidx(p);
292 iterate_fd(p->files, 0, update_netprio, v);
406a3c63
JF
293 task_unlock(p);
294 }
406a3c63
JF
295}
296
5bc1421e
NH
297static struct cftype ss_files[] = {
298 {
299 .name = "prioidx",
300 .read_u64 = read_prioidx,
301 },
302 {
303 .name = "ifpriomap",
304 .read_map = read_priomap,
305 .write_string = write_priomap,
306 },
4baf6e33 307 { } /* terminate */
5bc1421e
NH
308};
309
676f7c8f
TH
310struct cgroup_subsys net_prio_subsys = {
311 .name = "net_prio",
312 .create = cgrp_create,
313 .destroy = cgrp_destroy,
406a3c63 314 .attach = net_prio_attach,
676f7c8f
TH
315#ifdef CONFIG_NETPRIO_CGROUP
316 .subsys_id = net_prio_subsys_id,
317#endif
4baf6e33 318 .base_cftypes = ss_files,
676f7c8f
TH
319 .module = THIS_MODULE
320};
5bc1421e
NH
321
322static int netprio_device_event(struct notifier_block *unused,
323 unsigned long event, void *ptr)
324{
325 struct net_device *dev = ptr;
326 struct netprio_map *old;
5bc1421e
NH
327
328 /*
329 * Note this is called with rtnl_lock held so we have update side
330 * protection on our rcu assignments
331 */
332
333 switch (event) {
5bc1421e
NH
334 case NETDEV_UNREGISTER:
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 break;
340 }
341 return NOTIFY_DONE;
342}
343
344static struct notifier_block netprio_device_notifier = {
345 .notifier_call = netprio_device_event
346};
347
348static int __init init_cgroup_netprio(void)
349{
350 int ret;
351
352 ret = cgroup_load_subsys(&net_prio_subsys);
353 if (ret)
354 goto out;
355#ifndef CONFIG_NETPRIO_CGROUP
356 smp_wmb();
357 net_prio_subsys_id = net_prio_subsys.subsys_id;
358#endif
359
360 register_netdevice_notifier(&netprio_device_notifier);
361
362out:
363 return ret;
364}
365
366static void __exit exit_cgroup_netprio(void)
367{
368 struct netprio_map *old;
369 struct net_device *dev;
370
371 unregister_netdevice_notifier(&netprio_device_notifier);
372
373 cgroup_unload_subsys(&net_prio_subsys);
374
375#ifndef CONFIG_NETPRIO_CGROUP
376 net_prio_subsys_id = -1;
377 synchronize_rcu();
378#endif
379
380 rtnl_lock();
381 for_each_netdev(&init_net, dev) {
382 old = rtnl_dereference(dev->priomap);
2cfa5a04 383 RCU_INIT_POINTER(dev->priomap, NULL);
5bc1421e
NH
384 if (old)
385 kfree_rcu(old, rcu);
386 }
387 rtnl_unlock();
388}
389
390module_init(init_cgroup_netprio);
391module_exit(exit_cgroup_netprio);
392MODULE_LICENSE("GPL v2");