net/core/link_watch.c

   1 /*
   2  * Linux network device link state notification
   3  *
   4  * Author:
   5  *     Stefan Rompf <sux@loplof.de>
   6  *
   7  * This program is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public License
   9  * as published by the Free Software Foundation; either version
  10  * 2 of the License, or (at your option) any later version.
  11  *
  12  */
  13
  14 #include <linux/module.h>
  15 #include <linux/netdevice.h>
  16 #include <linux/if.h>
  17 #include <net/sock.h>
  18 #include <net/pkt_sched.h>
  19 #include <linux/rtnetlink.h>
  20 #include <linux/jiffies.h>
  21 #include <linux/spinlock.h>
  22 #include <linux/workqueue.h>
  23 #include <linux/bitops.h>
  24 #include <asm/types.h>
  25
  26
  27 enum lw_bits {
  28         LW_URGENT = 0,
  29 };
  30
  31 static unsigned long linkwatch_flags;
  32 static unsigned long linkwatch_nextevent;
  33
  34 static void linkwatch_event(struct work_struct *dummy);
  35 static DECLARE_DELAYED_WORK(linkwatch_work, linkwatch_event);
  36
  37 static LIST_HEAD(lweventlist);
  38 static DEFINE_SPINLOCK(lweventlist_lock);
  39
  40 static unsigned char default_operstate(const struct net_device *dev)
  41 {
  42         if (!netif_carrier_ok(dev))
  43                 return (dev->ifindex != dev->iflink ?
  44                         IF_OPER_LOWERLAYERDOWN : IF_OPER_DOWN);
  45
  46         if (netif_dormant(dev))
  47                 return IF_OPER_DORMANT;
  48
  49         return IF_OPER_UP;
  50 }
  51
  52
  53 static void rfc2863_policy(struct net_device *dev)
  54 {
  55         unsigned char operstate = default_operstate(dev);
  56
  57         if (operstate == dev->operstate)
  58                 return;
  59
  60         write_lock_bh(&dev_base_lock);
  61
  62         switch(dev->link_mode) {
  63         case IF_LINK_MODE_DORMANT:
  64                 if (operstate == IF_OPER_UP)
  65                         operstate = IF_OPER_DORMANT;
  66                 break;
  67
  68         case IF_LINK_MODE_DEFAULT:
  69         default:
  70                 break;
  71         }
  72
  73         dev->operstate = operstate;
  74
  75         write_unlock_bh(&dev_base_lock);
  76 }
  77
  78
  79 static bool linkwatch_urgent_event(struct net_device *dev)
  80 {
  81         if (!netif_running(dev))
  82                 return false;
  83
  84         if (dev->ifindex != dev->iflink)
  85                 return true;
  86
  87         return netif_carrier_ok(dev) && qdisc_tx_changing(dev);
  88 }
  89
  90
  91 static void linkwatch_add_event(struct net_device *dev)
  92 {
  93         unsigned long flags;
  94
  95         spin_lock_irqsave(&lweventlist_lock, flags);
  96         if (list_empty(&dev->link_watch_list)) {
  97                 list_add_tail(&dev->link_watch_list, &lweventlist);
  98                 dev_hold(dev);
  99         }
 100         spin_unlock_irqrestore(&lweventlist_lock, flags);
 101 }
 102
 103
 104 static void linkwatch_schedule_work(int urgent)
 105 {
 106         unsigned long delay = linkwatch_nextevent - jiffies;
 107
 108         if (test_bit(LW_URGENT, &linkwatch_flags))
 109                 return;
 110
 111         /* Minimise down-time: drop delay for up event. */
 112         if (urgent) {
 113                 if (test_and_set_bit(LW_URGENT, &linkwatch_flags))
 114                         return;
 115                 delay = 0;
 116         }
 117
 118         /* If we wrap around we'll delay it by at most HZ. */
 119         if (delay > HZ)
 120                 delay = 0;
 121
 122         /*
 123          * This is true if we've scheduled it immeditately or if we don't
 124          * need an immediate execution and it's already pending.
 125          */
 126         if (schedule_delayed_work(&linkwatch_work, delay) == !delay)
 127                 return;
 128
 129         /* Don't bother if there is nothing urgent. */
 130         if (!test_bit(LW_URGENT, &linkwatch_flags))
 131                 return;
 132
 133         /* It's already running which is good enough. */
 134         if (!__cancel_delayed_work(&linkwatch_work))
 135                 return;
 136
 137         /* Otherwise we reschedule it again for immediate execution. */
 138         schedule_delayed_work(&linkwatch_work, 0);
 139 }
 140
 141
 142 static void linkwatch_do_dev(struct net_device *dev)
 143 {
 144         /*
 145          * Make sure the above read is complete since it can be
 146          * rewritten as soon as we clear the bit below.
 147          */
 148         smp_mb__before_clear_bit();
 149
 150         /* We are about to handle this device,
 151          * so new events can be accepted
 152          */
 153         clear_bit(__LINK_STATE_LINKWATCH_PENDING, &dev->state);
 154
 155         rfc2863_policy(dev);
 156         if (dev->flags & IFF_UP) {
 157                 if (netif_carrier_ok(dev))
 158                         dev_activate(dev);
 159                 else
 160                         dev_deactivate(dev);
 161
 162                 netdev_state_change(dev);
 163         }
 164         dev_put(dev);
 165 }
 166
 167 static void __linkwatch_run_queue(int urgent_only)
 168 {
 169         struct net_device *dev;
 170         LIST_HEAD(wrk);
 171
 172         /*
 173          * Limit the number of linkwatch events to one
 174          * per second so that a runaway driver does not
 175          * cause a storm of messages on the netlink
 176          * socket.  This limit does not apply to up events
 177          * while the device qdisc is down.
 178          */
 179         if (!urgent_only)
 180                 linkwatch_nextevent = jiffies + HZ;
 181         /* Limit wrap-around effect on delay. */
 182         else if (time_after(linkwatch_nextevent, jiffies + HZ))
 183                 linkwatch_nextevent = jiffies;
 184
 185         clear_bit(LW_URGENT, &linkwatch_flags);
 186
 187         spin_lock_irq(&lweventlist_lock);
 188         list_splice_init(&lweventlist, &wrk);
 189
 190         while (!list_empty(&wrk)) {
 191
 192                 dev = list_first_entry(&wrk, struct net_device, link_watch_list);
 193                 list_del_init(&dev->link_watch_list);
 194
 195                 if (urgent_only && !linkwatch_urgent_event(dev)) {
 196                         list_add_tail(&dev->link_watch_list, &lweventlist);
 197                         continue;
 198                 }
 199                 spin_unlock_irq(&lweventlist_lock);
 200                 linkwatch_do_dev(dev);
 201                 spin_lock_irq(&lweventlist_lock);
 202         }
 203
 204         if (!list_empty(&lweventlist))
 205                 linkwatch_schedule_work(0);
 206         spin_unlock_irq(&lweventlist_lock);
 207 }
 208
 209 void linkwatch_forget_dev(struct net_device *dev)
 210 {
 211         unsigned long flags;
 212         int clean = 0;
 213
 214         spin_lock_irqsave(&lweventlist_lock, flags);
 215         if (!list_empty(&dev->link_watch_list)) {
 216                 list_del_init(&dev->link_watch_list);
 217                 clean = 1;
 218         }
 219         spin_unlock_irqrestore(&lweventlist_lock, flags);
 220         if (clean)
 221                 linkwatch_do_dev(dev);
 222 }
 223
 224
 225 /* Must be called with the rtnl semaphore held */
 226 void linkwatch_run_queue(void)
 227 {
 228         __linkwatch_run_queue(0);
 229 }
 230
 231
 232 static void linkwatch_event(struct work_struct *dummy)
 233 {
 234         rtnl_lock();
 235         __linkwatch_run_queue(time_after(linkwatch_nextevent, jiffies));
 236         rtnl_unlock();
 237 }
 238
 239
 240 void linkwatch_fire_event(struct net_device *dev)
 241 {
 242         bool urgent = linkwatch_urgent_event(dev);
 243
 244         if (!test_and_set_bit(__LINK_STATE_LINKWATCH_PENDING, &dev->state)) {
 245                 linkwatch_add_event(dev);
 246         } else if (!urgent)
 247                 return;
 248
 249         linkwatch_schedule_work(urgent);
 250 }
 251 EXPORT_SYMBOL(linkwatch_fire_event);