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          * If urgent, schedule immediate execution; otherwise, don't
 124          * override the existing timer.
 125          */
 126         if (test_bit(LW_URGENT, &linkwatch_flags))
 127                 mod_delayed_work(system_wq, &linkwatch_work, 0);
 128         else
 129                 schedule_delayed_work(&linkwatch_work, delay);
 130 }
 131
 132
 133 static void linkwatch_do_dev(struct net_device *dev)
 134 {
 135         /*
 136          * Make sure the above read is complete since it can be
 137          * rewritten as soon as we clear the bit below.
 138          */
 139         smp_mb__before_clear_bit();
 140
 141         /* We are about to handle this device,
 142          * so new events can be accepted
 143          */
 144         clear_bit(__LINK_STATE_LINKWATCH_PENDING, &dev->state);
 145
 146         rfc2863_policy(dev);
 147         if (dev->flags & IFF_UP) {
 148                 if (netif_carrier_ok(dev))
 149                         dev_activate(dev);
 150                 else
 151                         dev_deactivate(dev);
 152
 153                 netdev_state_change(dev);
 154         }
 155         dev_put(dev);
 156 }
 157
 158 static void __linkwatch_run_queue(int urgent_only)
 159 {
 160         struct net_device *dev;
 161         LIST_HEAD(wrk);
 162
 163         /*
 164          * Limit the number of linkwatch events to one
 165          * per second so that a runaway driver does not
 166          * cause a storm of messages on the netlink
 167          * socket.  This limit does not apply to up events
 168          * while the device qdisc is down.
 169          */
 170         if (!urgent_only)
 171                 linkwatch_nextevent = jiffies + HZ;
 172         /* Limit wrap-around effect on delay. */
 173         else if (time_after(linkwatch_nextevent, jiffies + HZ))
 174                 linkwatch_nextevent = jiffies;
 175
 176         clear_bit(LW_URGENT, &linkwatch_flags);
 177
 178         spin_lock_irq(&lweventlist_lock);
 179         list_splice_init(&lweventlist, &wrk);
 180
 181         while (!list_empty(&wrk)) {
 182
 183                 dev = list_first_entry(&wrk, struct net_device, link_watch_list);
 184                 list_del_init(&dev->link_watch_list);
 185
 186                 if (urgent_only && !linkwatch_urgent_event(dev)) {
 187                         list_add_tail(&dev->link_watch_list, &lweventlist);
 188                         continue;
 189                 }
 190                 spin_unlock_irq(&lweventlist_lock);
 191                 linkwatch_do_dev(dev);
 192                 spin_lock_irq(&lweventlist_lock);
 193         }
 194
 195         if (!list_empty(&lweventlist))
 196                 linkwatch_schedule_work(0);
 197         spin_unlock_irq(&lweventlist_lock);
 198 }
 199
 200 void linkwatch_forget_dev(struct net_device *dev)
 201 {
 202         unsigned long flags;
 203         int clean = 0;
 204
 205         spin_lock_irqsave(&lweventlist_lock, flags);
 206         if (!list_empty(&dev->link_watch_list)) {
 207                 list_del_init(&dev->link_watch_list);
 208                 clean = 1;
 209         }
 210         spin_unlock_irqrestore(&lweventlist_lock, flags);
 211         if (clean)
 212                 linkwatch_do_dev(dev);
 213 }
 214
 215
 216 /* Must be called with the rtnl semaphore held */
 217 void linkwatch_run_queue(void)
 218 {
 219         __linkwatch_run_queue(0);
 220 }
 221
 222
 223 static void linkwatch_event(struct work_struct *dummy)
 224 {
 225         rtnl_lock();
 226         __linkwatch_run_queue(time_after(linkwatch_nextevent, jiffies));
 227         rtnl_unlock();
 228 }
 229
 230
 231 void linkwatch_fire_event(struct net_device *dev)
 232 {
 233         bool urgent = linkwatch_urgent_event(dev);
 234
 235         if (!test_and_set_bit(__LINK_STATE_LINKWATCH_PENDING, &dev->state)) {
 236                 linkwatch_add_event(dev);
 237         } else if (!urgent)
 238                 return;
 239
 240         linkwatch_schedule_work(urgent);
 241 }
 242 EXPORT_SYMBOL(linkwatch_fire_event);