src/fb_eth.c

   1 /*
   2  * Lightweight Autonomic Network Architecture
   3  *
   4  * Eth/PHY layer. Redirects all traffic into the LANA stack.
   5  * Singleton object.
   6  *
   7  * Copyright 2011 Daniel Borkmann <dborkma@tik.ee.ethz.ch>,
   8  * Swiss federal institute of technology (ETH Zurich)
   9  * Subject to the GPL.
  10  */
  11
  12 #include <linux/module.h>
  13 #include <linux/kernel.h>
  14 #include <linux/init.h>
  15 #include <linux/notifier.h>
  16 #include <linux/if_ether.h>
  17 #include <linux/if_arp.h>
  18 #include <linux/if.h>
  19 #include <linux/etherdevice.h>
  20 #include <linux/rtnetlink.h>
  21 #include <linux/seqlock.h>
  22
  23 #include "xt_idp.h"
  24 #include "xt_engine.h"
  25 #include "xt_skb.h"
  26 #include "xt_fblock.h"
  27 #include "xt_builder.h"
  28 #include "xt_vlink.h"
  29
  30 #define IFF_VLINK_MAS   0x20000
  31 #define IFF_VLINK_DEV   0x40000
  32 #define IFF_IS_BRIDGED  0x60000
  33
  34 struct fb_eth_priv {
  35         idp_t port[2];
  36         seqlock_t lock;
  37         struct net_device *dev;
  38 };
  39
  40 static LIST_HEAD(fb_eth_devs);
  41 static DEFINE_SPINLOCK(fb_eth_devs_lock);
  42
  43 struct fb_eth_dev_node {
  44         struct list_head list;
  45         struct fblock *fb;
  46         struct net_device *dev;
  47 };
  48
  49 static inline int fb_eth_dev_is_bridged(struct net_device *dev)
  50 {
  51         return (dev->priv_flags & IFF_IS_BRIDGED) == IFF_IS_BRIDGED;
  52 }
  53
  54 static inline int fb_ethvlink_real_dev_is_hooked(struct net_device *dev)
  55 {
  56         return (dev->priv_flags & IFF_VLINK_MAS) == IFF_VLINK_MAS;
  57 }
  58
  59 static inline void fb_eth_make_dev_bridged(struct net_device *dev)
  60 {
  61         dev->priv_flags |= IFF_IS_BRIDGED;
  62 }
  63
  64 static inline void fb_eth_make_dev_unbridged(struct net_device *dev)
  65 {
  66         dev->priv_flags &= ~IFF_IS_BRIDGED;
  67 }
  68
  69 static rx_handler_result_t fb_eth_handle_frame(struct sk_buff **pskb)
  70 {
  71         unsigned int seq;
  72         struct sk_buff *skb = *pskb;
  73         struct fb_eth_dev_node *node;
  74         struct fblock *fb = NULL;
  75         struct fb_eth_priv __percpu *fb_priv_cpu;
  76
  77         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
  78                 return RX_HANDLER_PASS;
  79         if (unlikely(!is_valid_ether_addr(eth_hdr(skb)->h_source)))
  80                 goto drop;
  81         skb = skb_share_check(skb, GFP_ATOMIC);
  82         if (unlikely(!skb))
  83                 return RX_HANDLER_CONSUMED;
  84
  85         list_for_each_entry_rcu(node, &fb_eth_devs, list)
  86                 if (skb->dev == node->dev)
  87                         fb = node->fb;
  88         if (!fb)
  89                 goto drop;
  90
  91         skb_orphan(skb);
  92
  93         fb_priv_cpu = this_cpu_ptr(rcu_dereference(fb->private_data));
  94         if (fb_priv_cpu->port[TYPE_INGRESS] == IDP_UNKNOWN)
  95                 goto drop;
  96
  97         do {
  98                 seq = read_seqbegin(&fb_priv_cpu->lock);
  99                 write_next_idp_to_skb(skb, fb->idp,
 100                                       fb_priv_cpu->port[TYPE_INGRESS]);
 101         } while (read_seqretry(&fb_priv_cpu->lock, seq));
 102
 103         process_packet(skb, TYPE_INGRESS);
 104
 105         return RX_HANDLER_CONSUMED;
 106 drop:
 107         kfree_skb(skb);
 108         return RX_HANDLER_CONSUMED;
 109 }
 110
 111 static int fb_eth_netrx(const struct fblock * const fb,
 112                         struct sk_buff * const skb,
 113                         enum path_type * const dir)
 114 {
 115         struct fb_eth_priv __percpu *fb_priv_cpu;
 116         fb_priv_cpu = this_cpu_ptr(rcu_dereference(fb->private_data));
 117         write_next_idp_to_skb(skb, fb->idp, IDP_UNKNOWN);
 118         skb->dev = fb_priv_cpu->dev;
 119         dev_queue_xmit(skb);
 120         return PPE_DROPPED;
 121 }
 122
 123 static int fb_eth_event(struct notifier_block *self, unsigned long cmd,
 124                         void *args)
 125 {
 126         int ret = NOTIFY_OK;
 127         unsigned int cpu;
 128         struct fblock *fb;
 129         struct fb_eth_priv __percpu *fb_priv;
 130
 131         rcu_read_lock();
 132         fb = rcu_dereference_raw(container_of(self, struct fblock_notifier, nb)->self);
 133         fb_priv = (struct fb_eth_priv __percpu *) rcu_dereference_raw(fb->private_data);
 134         rcu_read_unlock();
 135
 136         switch (cmd) {
 137         case FBLOCK_BIND_IDP: {
 138                 int bound = 0;
 139                 struct fblock_bind_msg *msg = args;
 140                 get_online_cpus();
 141                 for_each_online_cpu(cpu) {
 142                         struct fb_eth_priv *fb_priv_cpu;
 143                         fb_priv_cpu = per_cpu_ptr(fb_priv, cpu);
 144                         if (fb_priv_cpu->port[msg->dir] == IDP_UNKNOWN) {
 145                                 write_seqlock(&fb_priv_cpu->lock);
 146                                 fb_priv_cpu->port[msg->dir] = msg->idp;
 147                                 write_sequnlock(&fb_priv_cpu->lock);
 148                                 bound = 1;
 149                         } else {
 150                                 ret = NOTIFY_BAD;
 151                                 break;
 152                         }
 153                 }
 154                 put_online_cpus();
 155                 if (bound)
 156                         printk(KERN_INFO "[%s::vlink] port %s bound to IDP%u\n",
 157                                fb->name, path_names[msg->dir], msg->idp);
 158                 } break;
 159         case FBLOCK_UNBIND_IDP: {
 160                 int unbound = 0;
 161                 struct fblock_bind_msg *msg = args;
 162                 get_online_cpus();
 163                 for_each_online_cpu(cpu) {
 164                         struct fb_eth_priv *fb_priv_cpu;
 165                         fb_priv_cpu = per_cpu_ptr(fb_priv, cpu);
 166                         if (fb_priv_cpu->port[msg->dir] == msg->idp) {
 167                                 write_seqlock(&fb_priv_cpu->lock);
 168                                 fb_priv_cpu->port[msg->dir] = IDP_UNKNOWN;
 169                                 write_sequnlock(&fb_priv_cpu->lock);
 170                                 unbound = 1;
 171                         } else {
 172                                 ret = NOTIFY_BAD;
 173                                 break;
 174                         }
 175                 }
 176                 put_online_cpus();
 177                 if (unbound)
 178                         printk(KERN_INFO "[%s::vlink] port %s unbound\n",
 179                                fb->name, path_names[msg->dir]);
 180                 } break;
 181         default:
 182                 break;
 183         }
 184
 185         return ret;
 186 }
 187
 188 static void cleanup_fb_eth(struct net_device *dev)
 189 {
 190         rtnl_lock();
 191         if (fb_eth_dev_is_bridged(dev)) {
 192                 netdev_rx_handler_unregister(dev);
 193                 fb_eth_make_dev_unbridged(dev);
 194         }
 195         rtnl_unlock();
 196 }
 197
 198 static int init_fb_eth(struct net_device *dev)
 199 {
 200         int ret = 0;
 201         rtnl_lock();
 202         ret = netdev_rx_handler_register(dev, fb_eth_handle_frame, NULL);
 203         if (ret)
 204                 ret = -EIO;
 205         else
 206                 fb_eth_make_dev_bridged(dev);
 207         rtnl_unlock();
 208         return ret;
 209 }
 210
 211 static struct fblock *fb_eth_build_fblock(struct net_device *dev)
 212 {
 213         int ret = 0;
 214         unsigned int cpu;
 215         struct fblock *fb;
 216         struct fb_eth_priv __percpu *fb_priv;
 217
 218         fb = alloc_fblock(GFP_ATOMIC);
 219         if (!fb)
 220                 return NULL;
 221
 222         fb_priv = alloc_percpu(struct fb_eth_priv);
 223         if (!fb_priv)
 224                 goto err;
 225
 226         get_online_cpus();
 227         for_each_online_cpu(cpu) {
 228                 struct fb_eth_priv *fb_priv_cpu;
 229                 fb_priv_cpu = per_cpu_ptr(fb_priv, cpu);
 230                 seqlock_init(&fb_priv_cpu->lock);
 231                 fb_priv_cpu->port[0] = IDP_UNKNOWN;
 232                 fb_priv_cpu->port[1] = IDP_UNKNOWN;
 233                 fb_priv_cpu->dev = dev;
 234         }
 235         put_online_cpus();
 236
 237         ret = init_fblock(fb, dev->name, fb_priv);
 238         if (ret)
 239                 goto err2;
 240
 241         fb->netfb_rx = fb_eth_netrx;
 242         fb->event_rx = fb_eth_event;
 243         fb->factory = NULL;
 244
 245         ret = register_fblock_namespace(fb);
 246         if (ret)
 247                 goto err3;
 248         ret = init_fb_eth(dev);
 249         if (ret)
 250                 goto err4;
 251         __module_get(THIS_MODULE);
 252         smp_wmb();
 253         return fb;
 254 err4:
 255         unregister_fblock_namespace_no_rcu(fb);
 256 err3:
 257         cleanup_fblock_ctor(fb);
 258 err2:
 259         free_percpu(fb_priv);
 260 err:
 261         kfree_fblock(fb);
 262         fb = NULL;
 263         return NULL;
 264 }
 265
 266 static void fb_eth_destroy_fblock(struct fblock *fb)
 267 {
 268         struct fb_eth_priv __percpu *fb_priv_cpu;
 269
 270         rcu_read_lock();
 271         fb_priv_cpu = this_cpu_ptr(rcu_dereference(fb->private_data));
 272         cleanup_fb_eth(fb_priv_cpu->dev);
 273         rcu_read_unlock();
 274
 275         unregister_fblock_namespace_no_rcu(fb);
 276         cleanup_fblock(fb);
 277         free_percpu(rcu_dereference_raw(fb->private_data));
 278         kfree_fblock(fb);
 279         module_put(THIS_MODULE);
 280 }
 281
 282 static struct vlink_subsys fb_eth_sys __read_mostly = {
 283         .name = "eth",
 284         .owner = THIS_MODULE,
 285         .type = VLINKNLGRP_ETHERNET,
 286         .rwsem = __RWSEM_INITIALIZER(fb_eth_sys.rwsem),
 287 };
 288
 289 static int fb_eth_start_hook_dev(struct vlinknlmsg *vhdr, struct nlmsghdr *nlh)
 290 {
 291         unsigned long flags;
 292         struct net_device *dev;
 293         struct fb_eth_dev_node *node;
 294
 295         if (vhdr->cmd != VLINKNLCMD_START_HOOK_DEVICE)
 296                 return NETLINK_VLINK_RX_NXT;
 297
 298         dev = dev_get_by_name(&init_net, vhdr->real_name);
 299         if (dev && (dev->priv_flags & IFF_VLINK_DEV) == IFF_VLINK_DEV)
 300                 goto err;
 301         else if (!dev)
 302                 return NETLINK_VLINK_RX_EMERG;
 303
 304         if (fb_eth_dev_is_bridged(dev))
 305                 goto out;
 306         if (fb_ethvlink_real_dev_is_hooked(dev))
 307                 goto out;
 308
 309         node = kmalloc(sizeof(*node), GFP_KERNEL);
 310         if (!node)
 311                 goto out;
 312         node->dev = dev;
 313         node->fb = fb_eth_build_fblock(dev);
 314         if (!node->fb) {
 315                 kfree(node);
 316                 goto out;
 317         }
 318
 319         spin_lock_irqsave(&fb_eth_devs_lock, flags);
 320         list_add_rcu(&node->list, &fb_eth_devs);
 321         spin_unlock_irqrestore(&fb_eth_devs_lock, flags);
 322
 323         printk(KERN_INFO "[lana] hook attached to carrier %s\n",
 324                vhdr->real_name);
 325 out:
 326         dev_put(dev);
 327         return NETLINK_VLINK_RX_STOP;
 328 err:
 329         dev_put(dev);
 330         return NETLINK_VLINK_RX_EMERG;
 331 }
 332
 333 static int fb_eth_stop_hook_dev(struct vlinknlmsg *vhdr, struct nlmsghdr *nlh)
 334 {
 335         unsigned long flags;
 336         struct fblock *fb = NULL;
 337         struct net_device *dev;
 338         struct fb_eth_dev_node *node;
 339
 340         if (vhdr->cmd != VLINKNLCMD_STOP_HOOK_DEVICE)
 341                 return NETLINK_VLINK_RX_NXT;
 342
 343         dev = dev_get_by_name(&init_net, vhdr->real_name);
 344         if (!dev)
 345                 return NETLINK_VLINK_RX_EMERG;
 346         if (!fb_eth_dev_is_bridged(dev))
 347                 goto err_put;
 348         if ((dev->flags & IFF_RUNNING) == IFF_RUNNING)
 349                 goto err_put;
 350
 351         spin_lock_irqsave(&fb_eth_devs_lock, flags);
 352         list_for_each_entry_rcu(node, &fb_eth_devs, list) {
 353                 if (dev == node->dev) {
 354                         fb = node->fb;
 355                         break;
 356                 }
 357         }
 358         spin_unlock_irqrestore(&fb_eth_devs_lock, flags);
 359
 360         if (!fb)
 361                 goto err_put;
 362         if (atomic_read(&fb->refcnt) > 2) {
 363                 printk(KERN_INFO "Cannot remove vlink dev! Still in use by "
 364                        "others!\n");
 365                 goto err_put;
 366         }
 367
 368         spin_lock_irqsave(&fb_eth_devs_lock, flags);
 369         list_del_rcu(&node->list);
 370         spin_unlock_irqrestore(&fb_eth_devs_lock, flags);
 371
 372         fb_eth_destroy_fblock(fb);
 373         synchronize_rcu();
 374         kfree(node);
 375
 376         printk(KERN_INFO "[lana] hook detached from carrier %s\n",
 377                vhdr->real_name);
 378
 379         dev_put(dev);
 380         return NETLINK_VLINK_RX_STOP;
 381 err_put:
 382         dev_put(dev);
 383         return NETLINK_VLINK_RX_EMERG;
 384 }
 385
 386 static struct vlink_callback fb_eth_start_hook_dev_cb =
 387         VLINK_CALLBACK_INIT(fb_eth_start_hook_dev, NETLINK_VLINK_PRIO_HIGH);
 388 static struct vlink_callback fb_eth_stop_hook_dev_cb =
 389         VLINK_CALLBACK_INIT(fb_eth_stop_hook_dev, NETLINK_VLINK_PRIO_HIGH);
 390
 391 static int __init init_fb_eth_module(void)
 392 {
 393         int ret = 0;
 394         ret = vlink_subsys_register(&fb_eth_sys);
 395         if (ret)
 396                 return ret;
 397
 398         vlink_add_callback(&fb_eth_sys, &fb_eth_start_hook_dev_cb);
 399         vlink_add_callback(&fb_eth_sys, &fb_eth_stop_hook_dev_cb);
 400
 401         printk(KERN_INFO "[lana] Ethernet vlink layer loaded!\n");
 402         return ret;
 403 }
 404
 405 static void __exit cleanup_fb_eth_module(void)
 406 {
 407         synchronize_rcu();
 408         vlink_subsys_unregister_batch(&fb_eth_sys);
 409         printk(KERN_INFO "[lana] Ethernet vlink layer removed!\n");
 410 }
 411
 412 module_init(init_fb_eth_module);
 413 module_exit(cleanup_fb_eth_module);
 414
 415 MODULE_LICENSE("GPL");
 416 MODULE_AUTHOR("Daniel Borkmann <dborkma@tik.ee.ethz.ch>");
 417 MODULE_DESCRIPTION("Ethernet virtual link layer driver");
 418