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