Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-or-later 2 : /* 3 : * Linux network device link state notification 4 : * 5 : * Author: 6 : * Stefan Rompf <sux@loplof.de> 7 : */ 8 : 9 : #include <linux/module.h> 10 : #include <linux/netdevice.h> 11 : #include <linux/if.h> 12 : #include <net/sock.h> 13 : #include <net/pkt_sched.h> 14 : #include <linux/rtnetlink.h> 15 : #include <linux/jiffies.h> 16 : #include <linux/spinlock.h> 17 : #include <linux/workqueue.h> 18 : #include <linux/bitops.h> 19 : #include <linux/types.h> 20 : 21 : 22 : enum lw_bits { 23 : LW_URGENT = 0, 24 : }; 25 : 26 : static unsigned long linkwatch_flags; 27 : static unsigned long linkwatch_nextevent; 28 : 29 : static void linkwatch_event(struct work_struct *dummy); 30 : static DECLARE_DELAYED_WORK(linkwatch_work, linkwatch_event); 31 : 32 : static LIST_HEAD(lweventlist); 33 : static DEFINE_SPINLOCK(lweventlist_lock); 34 : 35 2 : static unsigned char default_operstate(const struct net_device *dev) 36 : { 37 2 : if (netif_testing(dev)) 38 : return IF_OPER_TESTING; 39 : 40 2 : if (!netif_carrier_ok(dev)) 41 1 : return (dev->ifindex != dev_get_iflink(dev) ? 42 1 : IF_OPER_LOWERLAYERDOWN : IF_OPER_DOWN); 43 : 44 1 : if (netif_dormant(dev)) 45 0 : return IF_OPER_DORMANT; 46 : 47 : return IF_OPER_UP; 48 : } 49 : 50 : 51 2 : static void rfc2863_policy(struct net_device *dev) 52 : { 53 2 : unsigned char operstate = default_operstate(dev); 54 : 55 2 : if (operstate == dev->operstate) 56 : return; 57 : 58 2 : write_lock_bh(&dev_base_lock); 59 : 60 2 : switch(dev->link_mode) { 61 0 : case IF_LINK_MODE_TESTING: 62 0 : if (operstate == IF_OPER_UP) 63 0 : operstate = IF_OPER_TESTING; 64 : break; 65 : 66 0 : case IF_LINK_MODE_DORMANT: 67 0 : if (operstate == IF_OPER_UP) 68 0 : operstate = IF_OPER_DORMANT; 69 : break; 70 : case IF_LINK_MODE_DEFAULT: 71 : default: 72 : break; 73 : } 74 : 75 2 : dev->operstate = operstate; 76 : 77 2 : write_unlock_bh(&dev_base_lock); 78 : } 79 : 80 : 81 2 : void linkwatch_init_dev(struct net_device *dev) 82 : { 83 : /* Handle pre-registration link state changes */ 84 4 : if (!netif_carrier_ok(dev) || netif_dormant(dev) || 85 2 : netif_testing(dev)) 86 0 : rfc2863_policy(dev); 87 2 : } 88 : 89 : 90 2 : static bool linkwatch_urgent_event(struct net_device *dev) 91 : { 92 2 : if (!netif_running(dev)) 93 : return false; 94 : 95 0 : if (dev->ifindex != dev_get_iflink(dev)) 96 : return true; 97 : 98 0 : if (netif_is_lag_port(dev) || netif_is_lag_master(dev)) 99 : return true; 100 : 101 0 : return netif_carrier_ok(dev) && qdisc_tx_changing(dev); 102 : } 103 : 104 : 105 2 : static void linkwatch_add_event(struct net_device *dev) 106 : { 107 2 : unsigned long flags; 108 : 109 2 : spin_lock_irqsave(&lweventlist_lock, flags); 110 2 : if (list_empty(&dev->link_watch_list)) { 111 2 : list_add_tail(&dev->link_watch_list, &lweventlist); 112 2 : dev_hold(dev); 113 : } 114 2 : spin_unlock_irqrestore(&lweventlist_lock, flags); 115 2 : } 116 : 117 : 118 2 : static void linkwatch_schedule_work(int urgent) 119 : { 120 2 : unsigned long delay = linkwatch_nextevent - jiffies; 121 : 122 2 : if (test_bit(LW_URGENT, &linkwatch_flags)) 123 : return; 124 : 125 : /* Minimise down-time: drop delay for up event. */ 126 2 : if (urgent) { 127 0 : if (test_and_set_bit(LW_URGENT, &linkwatch_flags)) 128 : return; 129 : delay = 0; 130 : } 131 : 132 : /* If we wrap around we'll delay it by at most HZ. */ 133 2 : if (delay > HZ) 134 1 : delay = 0; 135 : 136 : /* 137 : * If urgent, schedule immediate execution; otherwise, don't 138 : * override the existing timer. 139 : */ 140 2 : if (test_bit(LW_URGENT, &linkwatch_flags)) 141 0 : mod_delayed_work(system_wq, &linkwatch_work, 0); 142 : else 143 2 : schedule_delayed_work(&linkwatch_work, delay); 144 : } 145 : 146 : 147 2 : static void linkwatch_do_dev(struct net_device *dev) 148 : { 149 : /* 150 : * Make sure the above read is complete since it can be 151 : * rewritten as soon as we clear the bit below. 152 : */ 153 2 : smp_mb__before_atomic(); 154 : 155 : /* We are about to handle this device, 156 : * so new events can be accepted 157 : */ 158 2 : clear_bit(__LINK_STATE_LINKWATCH_PENDING, &dev->state); 159 : 160 2 : rfc2863_policy(dev); 161 2 : if (dev->flags & IFF_UP && netif_device_present(dev)) { 162 0 : if (netif_carrier_ok(dev)) 163 0 : dev_activate(dev); 164 : else 165 0 : dev_deactivate(dev); 166 : 167 0 : netdev_state_change(dev); 168 : } 169 2 : dev_put(dev); 170 2 : } 171 : 172 2 : static void __linkwatch_run_queue(int urgent_only) 173 : { 174 : #define MAX_DO_DEV_PER_LOOP 100 175 : 176 2 : int do_dev = MAX_DO_DEV_PER_LOOP; 177 2 : struct net_device *dev; 178 2 : LIST_HEAD(wrk); 179 : 180 : /* Give urgent case more budget */ 181 2 : if (urgent_only) 182 0 : do_dev += MAX_DO_DEV_PER_LOOP; 183 : 184 : /* 185 : * Limit the number of linkwatch events to one 186 : * per second so that a runaway driver does not 187 : * cause a storm of messages on the netlink 188 : * socket. This limit does not apply to up events 189 : * while the device qdisc is down. 190 : */ 191 2 : if (!urgent_only) 192 2 : linkwatch_nextevent = jiffies + HZ; 193 : /* Limit wrap-around effect on delay. */ 194 0 : else if (time_after(linkwatch_nextevent, jiffies + HZ)) 195 0 : linkwatch_nextevent = jiffies; 196 : 197 2 : clear_bit(LW_URGENT, &linkwatch_flags); 198 : 199 2 : spin_lock_irq(&lweventlist_lock); 200 2 : list_splice_init(&lweventlist, &wrk); 201 : 202 2 : while (!list_empty(&wrk) && do_dev > 0) { 203 : 204 2 : dev = list_first_entry(&wrk, struct net_device, link_watch_list); 205 2 : list_del_init(&dev->link_watch_list); 206 : 207 2 : if (urgent_only && !linkwatch_urgent_event(dev)) { 208 0 : list_add_tail(&dev->link_watch_list, &lweventlist); 209 0 : continue; 210 : } 211 2 : spin_unlock_irq(&lweventlist_lock); 212 2 : linkwatch_do_dev(dev); 213 2 : do_dev--; 214 6 : spin_lock_irq(&lweventlist_lock); 215 : } 216 : 217 : /* Add the remaining work back to lweventlist */ 218 2 : list_splice_init(&wrk, &lweventlist); 219 : 220 2 : if (!list_empty(&lweventlist)) 221 0 : linkwatch_schedule_work(0); 222 2 : spin_unlock_irq(&lweventlist_lock); 223 2 : } 224 : 225 0 : void linkwatch_forget_dev(struct net_device *dev) 226 : { 227 0 : unsigned long flags; 228 0 : int clean = 0; 229 : 230 0 : spin_lock_irqsave(&lweventlist_lock, flags); 231 0 : if (!list_empty(&dev->link_watch_list)) { 232 0 : list_del_init(&dev->link_watch_list); 233 0 : clean = 1; 234 : } 235 0 : spin_unlock_irqrestore(&lweventlist_lock, flags); 236 0 : if (clean) 237 0 : linkwatch_do_dev(dev); 238 0 : } 239 : 240 : 241 : /* Must be called with the rtnl semaphore held */ 242 0 : void linkwatch_run_queue(void) 243 : { 244 0 : __linkwatch_run_queue(0); 245 0 : } 246 : 247 : 248 2 : static void linkwatch_event(struct work_struct *dummy) 249 : { 250 2 : rtnl_lock(); 251 4 : __linkwatch_run_queue(time_after(linkwatch_nextevent, jiffies)); 252 2 : rtnl_unlock(); 253 2 : } 254 : 255 : 256 2 : void linkwatch_fire_event(struct net_device *dev) 257 : { 258 2 : bool urgent = linkwatch_urgent_event(dev); 259 : 260 2 : if (!test_and_set_bit(__LINK_STATE_LINKWATCH_PENDING, &dev->state)) { 261 2 : linkwatch_add_event(dev); 262 0 : } else if (!urgent) 263 : return; 264 : 265 2 : linkwatch_schedule_work(urgent); 266 : } 267 : EXPORT_SYMBOL(linkwatch_fire_event);