net/core/netprio_cgroup.c

   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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  13
  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
  28 #include <linux/fdtable.h>
  29
  30 #define PRIOMAP_MIN_SZ          128
  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
  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)
  44 {
  45         struct netprio_map *old, *new;
  46         size_t new_sz, new_len;
  47
  48         /* is the existing priomap large enough? */
  49         old = rtnl_dereference(dev->priomap);
  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         }
  69
  70         /* allocate & copy */
  71         new = kzalloc(new_sz, GFP_KERNEL);
  72         if (!new) {
  73                 pr_warn("Unable to alloc new priomap!\n");
  74                 return -ENOMEM;
  75         }
  76
  77         if (old)
  78                 memcpy(new->priomap, old->priomap,
  79                        old->priomap_len * sizeof(old->priomap[0]));
  80
  81         new->priomap_len = new_len;
  82
  83         /* install the new priomap */
  84         rcu_assign_pointer(dev->priomap, new);
  85         if (old)
  86                 kfree_rcu(old, rcu);
  87         return 0;
  88 }
  89
  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
 135 static struct cgroup_subsys_state *cgrp_css_alloc(struct cgroup *cgrp)
 136 {
 137         struct cgroup_netprio_state *cs;
 138
 139         cs = kzalloc(sizeof(*cs), GFP_KERNEL);
 140         if (!cs)
 141                 return ERR_PTR(-ENOMEM);
 142
 143         return &cs->css;
 144 }
 145
 146 static int cgrp_css_online(struct cgroup *cgrp)
 147 {
 148         struct cgroup *parent = cgrp->parent;
 149         struct net_device *dev;
 150         int ret = 0;
 151
 152         if (!parent)
 153                 return 0;
 154
 155         rtnl_lock();
 156         /*
 157          * Inherit prios from the parent.  As all prios are set during
 158          * onlining, there is no need to clear them on offline.
 159          */
 160         for_each_netdev(&init_net, dev) {
 161                 u32 prio = netprio_prio(parent, dev);
 162
 163                 ret = netprio_set_prio(cgrp, dev, prio);
 164                 if (ret)
 165                         break;
 166         }
 167         rtnl_unlock();
 168         return ret;
 169 }
 170
 171 static void cgrp_css_free(struct cgroup *cgrp)
 172 {
 173         kfree(cgrp_netprio_state(cgrp));
 174 }
 175
 176 static u64 read_prioidx(struct cgroup *cgrp, struct cftype *cft)
 177 {
 178         return cgrp->id;
 179 }
 180
 181 static int read_priomap(struct cgroup *cont, struct cftype *cft,
 182                         struct cgroup_map_cb *cb)
 183 {
 184         struct net_device *dev;
 185
 186         rcu_read_lock();
 187         for_each_netdev_rcu(&init_net, dev)
 188                 cb->fill(cb, dev->name, netprio_prio(cont, dev));
 189         rcu_read_unlock();
 190         return 0;
 191 }
 192
 193 static int write_priomap(struct cgroup *cgrp, struct cftype *cft,
 194                          const char *buffer)
 195 {
 196         char devname[IFNAMSIZ + 1];
 197         struct net_device *dev;
 198         u32 prio;
 199         int ret;
 200
 201         if (sscanf(buffer, "%"__stringify(IFNAMSIZ)"s %u", devname, &prio) != 2)
 202                 return -EINVAL;
 203
 204         dev = dev_get_by_name(&init_net, devname);
 205         if (!dev)
 206                 return -ENODEV;
 207
 208         rtnl_lock();
 209
 210         ret = netprio_set_prio(cgrp, dev, prio);
 211
 212         rtnl_unlock();
 213         dev_put(dev);
 214         return ret;
 215 }
 216
 217 static int update_netprio(const void *v, struct file *file, unsigned n)
 218 {
 219         int err;
 220         struct socket *sock = sock_from_file(file, &err);
 221         if (sock)
 222                 sock->sk->sk_cgrp_prioidx = (u32)(unsigned long)v;
 223         return 0;
 224 }
 225
 226 static void net_prio_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
 227 {
 228         struct task_struct *p;
 229         void *v;
 230
 231         cgroup_taskset_for_each(p, cgrp, tset) {
 232                 task_lock(p);
 233                 v = (void *)(unsigned long)task_netprioidx(p);
 234                 iterate_fd(p->files, 0, update_netprio, v);
 235                 task_unlock(p);
 236         }
 237 }
 238
 239 static struct cftype ss_files[] = {
 240         {
 241                 .name = "prioidx",
 242                 .read_u64 = read_prioidx,
 243         },
 244         {
 245                 .name = "ifpriomap",
 246                 .read_map = read_priomap,
 247                 .write_string = write_priomap,
 248         },
 249         { }     /* terminate */
 250 };
 251
 252 struct cgroup_subsys net_prio_subsys = {
 253         .name           = "net_prio",
 254         .css_alloc      = cgrp_css_alloc,
 255         .css_online     = cgrp_css_online,
 256         .css_free       = cgrp_css_free,
 257         .attach         = net_prio_attach,
 258         .subsys_id      = net_prio_subsys_id,
 259         .base_cftypes   = ss_files,
 260         .module         = THIS_MODULE,
 261 };
 262
 263 static int netprio_device_event(struct notifier_block *unused,
 264                                 unsigned long event, void *ptr)
 265 {
 266         struct net_device *dev = ptr;
 267         struct netprio_map *old;
 268
 269         /*
 270          * Note this is called with rtnl_lock held so we have update side
 271          * protection on our rcu assignments
 272          */
 273
 274         switch (event) {
 275         case NETDEV_UNREGISTER:
 276                 old = rtnl_dereference(dev->priomap);
 277                 RCU_INIT_POINTER(dev->priomap, NULL);
 278                 if (old)
 279                         kfree_rcu(old, rcu);
 280                 break;
 281         }
 282         return NOTIFY_DONE;
 283 }
 284
 285 static struct notifier_block netprio_device_notifier = {
 286         .notifier_call = netprio_device_event
 287 };
 288
 289 static int __init init_cgroup_netprio(void)
 290 {
 291         int ret;
 292
 293         ret = cgroup_load_subsys(&net_prio_subsys);
 294         if (ret)
 295                 goto out;
 296
 297         register_netdevice_notifier(&netprio_device_notifier);
 298
 299 out:
 300         return ret;
 301 }
 302
 303 static void __exit exit_cgroup_netprio(void)
 304 {
 305         struct netprio_map *old;
 306         struct net_device *dev;
 307
 308         unregister_netdevice_notifier(&netprio_device_notifier);
 309
 310         cgroup_unload_subsys(&net_prio_subsys);
 311
 312         rtnl_lock();
 313         for_each_netdev(&init_net, dev) {
 314                 old = rtnl_dereference(dev->priomap);
 315                 RCU_INIT_POINTER(dev->priomap, NULL);
 316                 if (old)
 317                         kfree_rcu(old, rcu);
 318         }
 319         rtnl_unlock();
 320 }
 321
 322 module_init(init_cgroup_netprio);
 323 module_exit(exit_cgroup_netprio);
 324 MODULE_LICENSE("GPL v2");