对于arp协议，内核默认的STALE状态超时时长为60秒钟。
struct neigh_table arp_tbl = { .family = AF_INET, .key_len = 4, .protocol = cpu_to_be16(ETH_P_IP), ... .id = "arp_cache", .parms = { ... .data = { ... [NEIGH_VAR_GC_STALETIME] = 60 * HZ,通过PROC文件gc_stale_time可查看和修改此时长，如下，修改为10分钟。
/ # cat /proc/sys/net/ipv4/neigh/eth0/gc_stale_time 60 / # echo 600 > /proc/sys/net/ipv4/neigh/eth0/gc_stale_time / # / # cat /proc/sys/net/ipv4/neigh/eth0/gc_stale_time 600内核中静态变量neigh_sysctl_table定义了gc_stale_time的PROC文件信息。
static struct neigh_sysctl_table { struct ctl_table_header *sysctl_header; struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1]; } neigh_sysctl_template __read_mostly = { .neigh_vars = { ... NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),除了以上的PROC文件外，还可使用ip ntable命令查看和修改设备的邻居表参数。
# ip ntable show dev eth0 inet arp_cache dev eth0 refcnt 12 reachable 28884 base_reachable 30000 retrans 1000 gc_stale 60000 delay_probe 5000 queue 31 app_probes 0 ucast_probes 3 mcast_probes 3 anycast_delay 1000 proxy_delay 800 proxy_queue 64 locktime 1000与PROC文件不同，这里显示的gc_stale时间单位为毫秒。如下将设备eth0的邻居表参数gc_stale_time修改为10分钟。
# ip ntable change name arp_cache dev eth0 gc_stale 600000内核函数neigh_init负责以上ip ntable change命令的处理。
static int __init neigh_init(void) { ... rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, 0);如下为neightbl_set的实现，函数nla_get_msecs读取IP命令行设置的gc_stale的毫秒值参数。
static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh, struct netlink_ext_ack *extack) { struct neigh_table *tbl; struct nlattr *tb[NDTA_MAX+1]; if (tb[NDTA_PARMS]) { struct neigh_parms *p; p = lookup_neigh_parms(tbl, net, ifindex); ... for (i = 1; i <= NDTPA_MAX; i++) { if (tbp[i] == NULL) continue; switch (i) { ... case NDTPA_GC_STALETIME: NEIGH_VAR_SET(p, GC_STALETIME, nla_get_msecs(tbp[i])); break;对于arp协议，宏NEIGH_VAR_SET将修改全局变量arp_tbl的成员parms的data数组，具体为以NEIGH_VAR_GC_STALETIME为所对应的成员的值。函数nla_get_msecs将命令行输入的毫秒值转换为内核使用的jiffies值。
#define NEIGH_VAR_SET(p, attr, val) neigh_var_set(p, NEIGH_VAR_ ## attr, val) static inline void neigh_var_set(struct neigh_parms *p, int index, int val) { set_bit(index, p->data_state); p->data[index] = val; }显示命令ip ntable show由内核中的函数neightbl_fill_parms填充值，对于gc_stale的值，由nla_put_msecs填充。
static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms) { ... if ((parms->dev && ... nla_put_msecs(skb, NDTPA_GC_STALETIME, NEIGH_VAR(parms, GC_STALETIME), NDTPA_PAD) ||如下函数nla_put_msecs，其需要将内核使用gc_stale的jiffies表示的值转换为ip ntable show显示时的毫秒值，通过jiffies_to_msecs实现。
static inline int nla_put_msecs(struct sk_buff *skb, int attrtype, unsigned long njiffies, int padattr) { u64 tmp = jiffies_to_msecs(njiffies); return nla_put_64bit(skb, attrtype, sizeof(u64), &tmp, padattr); }在邻居表初始化时，创建了一个延迟工作项，注册的处理函数为neigh_periodic_work。
void neigh_table_init(int index, struct neigh_table *tbl) { tbl->parms.reachable_time = neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME)); ... INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work); queue_delayed_work(system_power_efficient_wq, &tbl->gc_work, tbl->parms.reachable_time);对于arp，其执行周期初始化为30秒钟。
struct neigh_table arp_tbl = { ... .parms = { .tbl = &arp_tbl, .reachable_time = 30 * HZ, .data = { [NEIGH_VAR_MCAST_PROBES] = 3, [NEIGH_VAR_UCAST_PROBES] = 3, [NEIGH_VAR_RETRANS_TIME] = 1 * HZ, [NEIGH_VAR_BASE_REACHABLE_TIME] = 30 * HZ,如下处理函数neigh_periodic_work，如果邻居项的引用计数为1，并且，空闲了gc_stale定义的时长没有被使用，内核将释放此邻居项。
static void neigh_periodic_work(struct work_struct *work) { struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work); struct neighbour *n; struct neighbour __rcu **np; struct neigh_hash_table *nht; ... nht = rcu_dereference_protected(tbl->nht, lockdep_is_held(&tbl->lock)); ... for (i = 0 ; i < (1 << nht->hash_shift); i++) { np = &nht->hash_buckets[i]; while ((n = rcu_dereference_protected(*np, lockdep_is_held(&tbl->lock))) != NULL) { ... if (refcount_read(&n->refcnt) == 1 && (state == NUD_FAILED || time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) { *np = n->next; neigh_mark_dead(n); write_unlock(&n->lock); neigh_cleanup_and_release(n); continue; }内核版本 5.0
