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 static struct cgroup_subsys_state *cgrp_create(struct cgroup *cgrp);
  29 static void cgrp_destroy(struct cgroup *cgrp);
  30 static int cgrp_populate(struct cgroup_subsys *ss, struct cgroup *cgrp);
  31
  32 struct cgroup_subsys net_prio_subsys = {
  33         .name           = "net_prio",
  34         .create         = cgrp_create,
  35         .destroy        = cgrp_destroy,
  36         .populate       = cgrp_populate,
  37 #ifdef CONFIG_NETPRIO_CGROUP
  38         .subsys_id      = net_prio_subsys_id,
  39 #endif
  40         .module         = THIS_MODULE
  41 };
  42
  43 #define PRIOIDX_SZ 128
  44
  45 static unsigned long prioidx_map[PRIOIDX_SZ];
  46 static DEFINE_SPINLOCK(prioidx_map_lock);
  47 static atomic_t max_prioidx = ATOMIC_INIT(0);
  48
  49 static inline struct cgroup_netprio_state *cgrp_netprio_state(struct cgroup *cgrp)
  50 {
  51         return container_of(cgroup_subsys_state(cgrp, net_prio_subsys_id),
  52                             struct cgroup_netprio_state, css);
  53 }
  54
  55 static int get_prioidx(u32 *prio)
  56 {
  57         unsigned long flags;
  58         u32 prioidx;
  59
  60         spin_lock_irqsave(&prioidx_map_lock, flags);
  61         prioidx = find_first_zero_bit(prioidx_map, sizeof(unsigned long) * PRIOIDX_SZ);
  62         if (prioidx == sizeof(unsigned long) * PRIOIDX_SZ) {
  63                 spin_unlock_irqrestore(&prioidx_map_lock, flags);
  64                 return -ENOSPC;
  65         }
  66         set_bit(prioidx, prioidx_map);
  67         spin_unlock_irqrestore(&prioidx_map_lock, flags);
  68         atomic_set(&max_prioidx, prioidx);
  69         *prio = prioidx;
  70         return 0;
  71 }
  72
  73 static void put_prioidx(u32 idx)
  74 {
  75         unsigned long flags;
  76
  77         spin_lock_irqsave(&prioidx_map_lock, flags);
  78         clear_bit(idx, prioidx_map);
  79         spin_unlock_irqrestore(&prioidx_map_lock, flags);
  80 }
  81
  82 static void extend_netdev_table(struct net_device *dev, u32 new_len)
  83 {
  84         size_t new_size = sizeof(struct netprio_map) +
  85                            ((sizeof(u32) * new_len));
  86         struct netprio_map *new_priomap = kzalloc(new_size, GFP_KERNEL);
  87         struct netprio_map *old_priomap;
  88         int i;
  89
  90         old_priomap  = rtnl_dereference(dev->priomap);
  91
  92         if (!new_priomap) {
  93                 pr_warn("Unable to alloc new priomap!\n");
  94                 return;
  95         }
  96
  97         for (i = 0;
  98              old_priomap && (i < old_priomap->priomap_len);
  99              i++)
 100                 new_priomap->priomap[i] = old_priomap->priomap[i];
 101
 102         new_priomap->priomap_len = new_len;
 103
 104         rcu_assign_pointer(dev->priomap, new_priomap);
 105         if (old_priomap)
 106                 kfree_rcu(old_priomap, rcu);
 107 }
 108
 109 static void update_netdev_tables(void)
 110 {
 111         struct net_device *dev;
 112         u32 max_len = atomic_read(&max_prioidx) + 1;
 113         struct netprio_map *map;
 114
 115         rtnl_lock();
 116         for_each_netdev(&init_net, dev) {
 117                 map = rtnl_dereference(dev->priomap);
 118                 if ((!map) ||
 119                     (map->priomap_len < max_len))
 120                         extend_netdev_table(dev, max_len);
 121         }
 122         rtnl_unlock();
 123 }
 124
 125 static struct cgroup_subsys_state *cgrp_create(struct cgroup *cgrp)
 126 {
 127         struct cgroup_netprio_state *cs;
 128         int ret;
 129
 130         cs = kzalloc(sizeof(*cs), GFP_KERNEL);
 131         if (!cs)
 132                 return ERR_PTR(-ENOMEM);
 133
 134         if (cgrp->parent && cgrp_netprio_state(cgrp->parent)->prioidx) {
 135                 kfree(cs);
 136                 return ERR_PTR(-EINVAL);
 137         }
 138
 139         ret = get_prioidx(&cs->prioidx);
 140         if (ret != 0) {
 141                 pr_warn("No space in priority index array\n");
 142                 kfree(cs);
 143                 return ERR_PTR(ret);
 144         }
 145
 146         return &cs->css;
 147 }
 148
 149 static void cgrp_destroy(struct cgroup *cgrp)
 150 {
 151         struct cgroup_netprio_state *cs;
 152         struct net_device *dev;
 153         struct netprio_map *map;
 154
 155         cs = cgrp_netprio_state(cgrp);
 156         rtnl_lock();
 157         for_each_netdev(&init_net, dev) {
 158                 map = rtnl_dereference(dev->priomap);
 159                 if (map)
 160                         map->priomap[cs->prioidx] = 0;
 161         }
 162         rtnl_unlock();
 163         put_prioidx(cs->prioidx);
 164         kfree(cs);
 165 }
 166
 167 static u64 read_prioidx(struct cgroup *cgrp, struct cftype *cft)
 168 {
 169         return (u64)cgrp_netprio_state(cgrp)->prioidx;
 170 }
 171
 172 static int read_priomap(struct cgroup *cont, struct cftype *cft,
 173                         struct cgroup_map_cb *cb)
 174 {
 175         struct net_device *dev;
 176         u32 prioidx = cgrp_netprio_state(cont)->prioidx;
 177         u32 priority;
 178         struct netprio_map *map;
 179
 180         rcu_read_lock();
 181         for_each_netdev_rcu(&init_net, dev) {
 182                 map = rcu_dereference(dev->priomap);
 183                 priority = map ? map->priomap[prioidx] : 0;
 184                 cb->fill(cb, dev->name, priority);
 185         }
 186         rcu_read_unlock();
 187         return 0;
 188 }
 189
 190 static int write_priomap(struct cgroup *cgrp, struct cftype *cft,
 191                          const char *buffer)
 192 {
 193         char *devname = kstrdup(buffer, GFP_KERNEL);
 194         int ret = -EINVAL;
 195         u32 prioidx = cgrp_netprio_state(cgrp)->prioidx;
 196         unsigned long priority;
 197         char *priostr;
 198         struct net_device *dev;
 199         struct netprio_map *map;
 200
 201         if (!devname)
 202                 return -ENOMEM;
 203
 204         /*
 205          * Minimally sized valid priomap string
 206          */
 207         if (strlen(devname) < 3)
 208                 goto out_free_devname;
 209
 210         priostr = strstr(devname, " ");
 211         if (!priostr)
 212                 goto out_free_devname;
 213
 214         /*
 215          *Separate the devname from the associated priority
 216          *and advance the priostr poitner to the priority value
 217          */
 218         *priostr = '\0';
 219         priostr++;
 220
 221         /*
 222          * If the priostr points to NULL, we're at the end of the passed
 223          * in string, and its not a valid write
 224          */
 225         if (*priostr == '\0')
 226                 goto out_free_devname;
 227
 228         ret = kstrtoul(priostr, 10, &priority);
 229         if (ret < 0)
 230                 goto out_free_devname;
 231
 232         ret = -ENODEV;
 233
 234         dev = dev_get_by_name(&init_net, devname);
 235         if (!dev)
 236                 goto out_free_devname;
 237
 238         update_netdev_tables();
 239         ret = 0;
 240         rcu_read_lock();
 241         map = rcu_dereference(dev->priomap);
 242         if (map)
 243                 map->priomap[prioidx] = priority;
 244         rcu_read_unlock();
 245         dev_put(dev);
 246
 247 out_free_devname:
 248         kfree(devname);
 249         return ret;
 250 }
 251
 252 static struct cftype ss_files[] = {
 253         {
 254                 .name = "prioidx",
 255                 .read_u64 = read_prioidx,
 256         },
 257         {
 258                 .name = "ifpriomap",
 259                 .read_map = read_priomap,
 260                 .write_string = write_priomap,
 261         },
 262 };
 263
 264 static int cgrp_populate(struct cgroup_subsys *ss, struct cgroup *cgrp)
 265 {
 266         return cgroup_add_files(cgrp, ss, ss_files, ARRAY_SIZE(ss_files));
 267 }
 268
 269 static int netprio_device_event(struct notifier_block *unused,
 270                                 unsigned long event, void *ptr)
 271 {
 272         struct net_device *dev = ptr;
 273         struct netprio_map *old;
 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         case NETDEV_UNREGISTER:
 282                 old = rtnl_dereference(dev->priomap);
 283                 RCU_INIT_POINTER(dev->priomap, NULL);
 284                 if (old)
 285                         kfree_rcu(old, rcu);
 286                 break;
 287         }
 288         return NOTIFY_DONE;
 289 }
 290
 291 static struct notifier_block netprio_device_notifier = {
 292         .notifier_call = netprio_device_event
 293 };
 294
 295 static int __init init_cgroup_netprio(void)
 296 {
 297         int ret;
 298
 299         ret = cgroup_load_subsys(&net_prio_subsys);
 300         if (ret)
 301                 goto out;
 302 #ifndef CONFIG_NETPRIO_CGROUP
 303         smp_wmb();
 304         net_prio_subsys_id = net_prio_subsys.subsys_id;
 305 #endif
 306
 307         register_netdevice_notifier(&netprio_device_notifier);
 308
 309 out:
 310         return ret;
 311 }
 312
 313 static void __exit exit_cgroup_netprio(void)
 314 {
 315         struct netprio_map *old;
 316         struct net_device *dev;
 317
 318         unregister_netdevice_notifier(&netprio_device_notifier);
 319
 320         cgroup_unload_subsys(&net_prio_subsys);
 321
 322 #ifndef CONFIG_NETPRIO_CGROUP
 323         net_prio_subsys_id = -1;
 324         synchronize_rcu();
 325 #endif
 326
 327         rtnl_lock();
 328         for_each_netdev(&init_net, dev) {
 329                 old = rtnl_dereference(dev->priomap);
 330                 RCU_INIT_POINTER(dev->priomap, NULL);
 331                 if (old)
 332                         kfree_rcu(old, rcu);
 333         }
 334         rtnl_unlock();
 335 }
 336
 337 module_init(init_cgroup_netprio);
 338 module_exit(exit_cgroup_netprio);
 339 MODULE_LICENSE("GPL v2");