ifconfig up后內(nèi)核網(wǎng)絡(luò)驅(qū)動(dòng)做了什么？

背景

最近在排查一個(gè)網(wǎng)絡(luò)問題，ifconfig eth0 up 后，網(wǎng)卡link up比較慢。因此，分析了下從ifconfig up 到網(wǎng)絡(luò)驅(qū)動(dòng)的調(diào)用流程。這里順便作個(gè)記錄。

ifconfig eth0 up 調(diào)用的是busybox 的命令，因此從busybox 源碼入手，逐步分析下調(diào)用流程。代碼介紹文件位于:networking/ifenslave.c

ifconfig eth0 up

ifconfig eth0 up 和 ifconfig eth0 down 分別對(duì)應(yīng)busybox 的set_if_up()和set_if_down().

staticintset_if_down(char*ifname,intflags)
{
intres=set_if_flags(ifname,flags&~IFF_UP);
if(res)
bb_perror_msg("%s:can'tdown",ifname);
returnres;
}

staticintset_if_up(char*ifname,intflags)
{
intres=set_if_flags(ifname,flags|IFF_UP);
if(res)
bb_perror_msg("%s:can'tup",ifname);
returnres;
}

比如,當(dāng)我們敲ifconfig eth0 down時(shí),實(shí)則就是調(diào)用:

set_if_down("eth0",master_flags.ifr_flags);

set_if_flags()會(huì)將網(wǎng)卡名，up / down 標(biāo)志位flags通過ioctl命令SIOCSIFFLAGS 傳遞給內(nèi)核網(wǎng)卡驅(qū)動(dòng)。

staticintset_if_flags(char*ifname,intflags)
{
structifreqifr;

ifr.ifr_flags=flags;
returnset_ifrname_and_do_ioctl(SIOCSIFFLAGS,&ifr,ifname);
}

dev_ifsioc

接著深入到內(nèi)核代碼中，看下SIOCSIFFLAGS命令在哪里實(shí)現(xiàn)。代碼位于kernel etcoredev_ioctl.c。

staticintdev_ifsioc(structnet*net,structifreq*ifr,unsignedintcmd)
{
interr;
structnet_device*dev=__dev_get_by_name(net,ifr->ifr_name);
conststructnet_device_ops*ops;

if(!dev)
return-ENODEV;

ops=dev->netdev_ops;

switch(cmd){
caseSIOCSIFFLAGS:/*Setinterfaceflags*/
returndev_change_flags(dev,ifr->ifr_flags);

caseSIOCSIFMETRIC:/*Setthemetricontheinterface
(currentlyunused)*/
return-EOPNOTSUPP;

...................

}
returnerr;
}

dev_ifsioc()會(huì)調(diào)用__dev_get_by_name()根據(jù) 網(wǎng)卡名遍歷 net鏈表，如果匹配到則返回net_device結(jié)構(gòu)體指針。接著，SIOCSIFFLAGS會(huì)調(diào)用到dev_change_flags()，最后調(diào)用到__dev_change_flags()。

dev_change_flags

intdev_change_flags(structnet_device*dev,unsignedintflags)
{
intret;
unsignedintchanges,old_flags=dev->flags,old_gflags=dev->gflags;

ret=__dev_change_flags(dev,flags);
if(retflags)|(old_gflags^dev->gflags);
__dev_notify_flags(dev,old_flags,changes);
returnret;
}

int__dev_change_flags(structnet_device*dev,unsignedintflags)
{
unsignedintold_flags=dev->flags;
intret;

ASSERT_RTNL();

/*
*Settheflagsonourdevice.
*/

dev->flags=(flags&(IFF_DEBUG|IFF_NOTRAILERS|IFF_NOARP|
IFF_DYNAMIC|IFF_MULTICAST|IFF_PORTSEL|
IFF_AUTOMEDIA))|
(dev->flags&(IFF_UP|IFF_VOLATILE|IFF_PROMISC|
IFF_ALLMULTI));

/*
*Loadinthecorrectmulticastlistnowtheflagshavechanged.
*/

if((old_flags^flags)&IFF_MULTICAST)
dev_change_rx_flags(dev,IFF_MULTICAST);

dev_set_rx_mode(dev);

/*
*Havewedownedtheinterface.WehandleIFF_UPourselves
*accordingtouserattemptstosetit,ratherthanblindly
*settingit.
*/

ret=0;
/*兩個(gè)標(biāo)識(shí)有一個(gè)是IFF_UP*/
if((old_flags^flags)&IFF_UP)
ret=((old_flags&IFF_UP)?__dev_close:__dev_open)(dev);//通過flags判斷調(diào)用__dev_close還是__dev_open

if((flags^dev->gflags)&IFF_PROMISC){
intinc=(flags&IFF_PROMISC)?1:-1;
unsignedintold_flags=dev->flags;

dev->gflags^=IFF_PROMISC;

if(__dev_set_promiscuity(dev,inc,false)>=0)
if(dev->flags!=old_flags)
dev_set_rx_mode(dev);
}

/*NOTE:orderofsynchronizationofIFF_PROMISCandIFF_ALLMULTI
isimportant.Some(broken)driverssetIFF_PROMISC,when
IFF_ALLMULTIisrequestednotaskingusandnotreporting.
*/
if((flags^dev->gflags)&IFF_ALLMULTI){
intinc=(flags&IFF_ALLMULTI)?1:-1;

dev->gflags^=IFF_ALLMULTI;
__dev_set_allmulti(dev,inc,false);
}

returnret;
}

在__dev_change_flags(dev, flags)函數(shù)中,通過判斷flag的IFF_UP位上的值是否相反,來實(shí)現(xiàn)是調(diào)用__dev_close()還是__dev_open()來開關(guān)eth0。

__dev_close

__dev_close中會(huì)將當(dāng)前的net_device加入到等待設(shè)備關(guān)閉列表中。

staticint__dev_close(structnet_device*dev)
{
intretval;
LIST_HEAD(single);

list_add(&dev->close_list,&single);
retval=__dev_close_many(&single);
list_del(&single);

returnretval;
}

__dev_close_many

__dev_close_many通知設(shè)備正在關(guān)閉，等待未發(fā)送完的數(shù)據(jù)發(fā)送完,最后清除開啟標(biāo)記。

staticint__dev_close_many(structlist_head*head)
{
structnet_device*dev;

ASSERT_RTNL();
might_sleep();

list_for_each_entry(dev,head,close_list){
/*Temporarilydisablenetpolluntiltheinterfaceisdown*/
/*禁用netpoll*/
netpoll_poll_disable(dev);
/*通知設(shè)備正在關(guān)閉*/
call_netdevice_notifiers(NETDEV_GOING_DOWN,dev);
/*清除start標(biāo)志位*/
clear_bit(__LINK_STATE_START,&dev->state);

/*Synchronizetoscheduledpoll.Wecannottouchpolllist,it
*canbeevenondifferentcpu.Sojustclearnetif_running().
*
*dev->stop()willinvokenapi_disable()onallofit's
*napi_structinstancesonthisdevice.
*/
smp_mb__after_atomic();/*Commitnetif_running().*/
}
/*未發(fā)送完的數(shù)據(jù)發(fā)送完*/
dev_deactivate_many(head);

list_for_each_entry(dev,head,close_list){
conststructnet_device_ops*ops=dev->netdev_ops;

/*
*Callthedevicespecificclose.Thiscannotfail.
*OnlyifdeviceisUP
*
*WeallowittobecalledevenafteraDETACHhot-plug
*event.
*/
/*調(diào)用設(shè)備關(guān)閉操作*/
if(ops->ndo_stop)
ops->ndo_stop(dev);
/*標(biāo)記設(shè)備關(guān)閉*/
dev->flags&=~IFF_UP;
/*啟用netpoll*/
netpoll_poll_enable(dev);
}

return0;
}

ndo_stop

ndo_stop為關(guān)閉網(wǎng)卡時(shí)，不同網(wǎng)卡驅(qū)動(dòng)注冊(cè)的不同的關(guān)閉函數(shù)，我們以海思的網(wǎng)卡驅(qū)動(dòng)為例，分析下ndo_stop函數(shù)的實(shí)現(xiàn)。代碼位于kerneldrivers etethernethisiliconhnshns_enet.c。

hns_nic_net_stop

staticinthns_nic_net_stop(structnet_device*ndev)
{
hns_nic_net_down(ndev);

return0;
}

hns_nic_net_down

staticvoidhns_nic_net_down(structnet_device*ndev)
{
inti;
structhnae_ae_ops*ops;
structhns_nic_priv*priv=netdev_priv(ndev);

if(test_and_set_bit(NIC_STATE_DOWN,&priv->state))
return;

(void)del_timer_sync(&priv->service_timer);
netif_tx_stop_all_queues(ndev);
netif_carrier_off(ndev);
netif_tx_disable(ndev);
priv->link=0;

if(priv->phy)
phy_stop(priv->phy);

ops=priv->ae_handle->dev->ops;

if(ops->stop)
ops->stop(priv->ae_handle);

netif_tx_stop_all_queues(ndev);

for(i=priv->ae_handle->q_num-1;i>=0;i--){
hns_nic_ring_close(ndev,i);
hns_nic_ring_close(ndev,i+priv->ae_handle->q_num);

/*cleantxbuffers*/
hns_nic_tx_clr_all_bufs(priv->ring_data+i);
}
}

hns_nic_net_down()中會(huì)調(diào)用netif_carrier_off()通知內(nèi)核子系統(tǒng)網(wǎng)絡(luò)斷開。下面我們?cè)敿?xì)分析下netif_carrier_off()的實(shí)現(xiàn)。

netif_carrier_off()

voidnetif_carrier_off(structnet_device*dev)
{
/*設(shè)置網(wǎng)卡為載波斷開狀態(tài)即nocarrier狀態(tài)，上行時(shí)軟中斷下半部讀到該狀態(tài)不會(huì)進(jìn)行網(wǎng)卡收包*/
if(!test_and_set_bit(__LINK_STATE_NOCARRIER,&dev->state)){
if(dev->reg_state==NETREG_UNINITIALIZED)
return;
/*增加設(shè)備改變狀態(tài)*/
atomic_inc(&dev->carrier_changes);
/*加入事件處理隊(duì)列進(jìn)行處理*/
linkwatch_fire_event(dev);
}
}

linkwatch_fire_event()

linkwatch_fire_event()函數(shù)將設(shè)備加入到事件隊(duì)列，并且進(jìn)行事件調(diào)度，調(diào)度中會(huì)根據(jù)是否為緊急事件做不同處理。

voidlinkwatch_fire_event(structnet_device*dev)
{
/*判斷是否是緊急處理的事件*/
boolurgent=linkwatch_urgent_event(dev);
/*判斷是否是緊急處理的事件*/
if(!test_and_set_bit(__LINK_STATE_LINKWATCH_PENDING,&dev->state)){
/*添加事件到事件列表*/
linkwatch_add_event(dev);
}elseif(!urgent)
/*設(shè)備以前已經(jīng)設(shè)置了pending標(biāo)記，不是緊急事件，直接返回*/
return;
/*事件調(diào)度*/
linkwatch_schedule_work(urgent);
}

linkwatch_urgent_event()

linkwatch_urgent_event()判斷是否是否需要緊急處理。

staticboollinkwatch_urgent_event(structnet_device*dev)
{
/*設(shè)備未運(yùn)行，非緊急*/
if(!netif_running(dev))
returnfalse;
/*設(shè)備的索引號(hào)與連接索引號(hào)不等，緊急*/
if(dev->ifindex!=dev_get_iflink(dev))
returntrue;
/*設(shè)備作為teamport，緊急*/
if(dev->priv_flags&IFF_TEAM_PORT)
returntrue;
/*連接與否&&發(fā)送隊(duì)列排隊(duì)規(guī)則改變與否*/
returnnetif_carrier_ok(dev)&&qdisc_tx_changing(dev);
}

linkwatch_add_event()

linkwatch_add_event()將設(shè)備加入到事件處理鏈表。

staticvoidlinkwatch_add_event(structnet_device*dev)
{
unsignedlongflags;

spin_lock_irqsave(&lweventlist_lock,flags);
/*若未添加，則添加設(shè)備到事件列表*/
if(list_empty(&dev->link_watch_list)){
list_add_tail(&dev->link_watch_list,&lweventlist);
dev_hold(dev);
}
spin_unlock_irqrestore(&lweventlist_lock,flags);
}

linkwatch_schedule_work()

linkwatch_schedule_work()對(duì)事件處理進(jìn)行調(diào)度，緊急事件立即執(zhí)行，非緊急事件延后執(zhí)行。

staticvoidlinkwatch_schedule_work(inturgent)
{
unsignedlongdelay=linkwatch_nextevent-jiffies;
/*已經(jīng)設(shè)置了緊急標(biāo)記，則返回*/
if(test_bit(LW_URGENT,&linkwatch_flags))
return;

/*需要緊急調(diào)度*/
if(urgent){
/*之前設(shè)置了，則返回*/
if(test_and_set_bit(LW_URGENT,&linkwatch_flags))
return;
/*未設(shè)置緊急，則立即執(zhí)行*/
delay=0;
}

/*如果大于1s則立即執(zhí)行*/
if(delay>HZ)
delay=0;

/*如果設(shè)置了緊急標(biāo)記，則立即執(zhí)行*/
if(test_bit(LW_URGENT,&linkwatch_flags))
mod_delayed_work(system_wq,&linkwatch_work,0);
else
/*未設(shè)置緊急標(biāo)記，則按照delay執(zhí)行*/
schedule_delayed_work(&linkwatch_work,delay);
}

__linkwatch_run_queue()

__linkwatch_run_queue()完成對(duì)事件調(diào)度隊(duì)列中設(shè)備的處理。

staticvoid__linkwatch_run_queue(inturgent_only)
{
structnet_device*dev;
LIST_HEAD(wrk);

/*
*Limitthenumberoflinkwatcheventstoone
*persecondsothatarunawaydriverdoesnot
*causeastormofmessagesonthenetlink
*socket.Thislimitdoesnotapplytoupevents
*whilethedeviceqdiscisdown.
*/
/*已達(dá)到調(diào)度時(shí)間*/
if(!urgent_only)
linkwatch_nextevent=jiffies+HZ;
/*Limitwrap-aroundeffectondelay.*/
/*
未到達(dá)調(diào)度時(shí)間，并且下一次調(diào)度在當(dāng)前時(shí)間的1s以后
那么設(shè)置調(diào)度時(shí)間是當(dāng)前時(shí)間
*/
elseif(time_after(linkwatch_nextevent,jiffies+HZ))
linkwatch_nextevent=jiffies;
/*清除緊急標(biāo)識(shí)*/
clear_bit(LW_URGENT,&linkwatch_flags);

spin_lock_irq(&lweventlist_lock);
list_splice_init(&lweventlist,&wrk);
/*遍歷鏈表*/
while(!list_empty(&wrk)){
/*獲取設(shè)備*/
dev=list_first_entry(&wrk,structnet_device,link_watch_list);
/*從鏈表移除設(shè)備*/
list_del_init(&dev->link_watch_list);
/*未到達(dá)調(diào)度時(shí)間&&不需要緊急處理*/
if(urgent_only&&!linkwatch_urgent_event(dev)){
/*添加到鏈表尾部*/
list_add_tail(&dev->link_watch_list,&lweventlist);
/*繼續(xù)處理*/
continue;
}
spin_unlock_irq(&lweventlist_lock);
/*處理設(shè)備*/
linkwatch_do_dev(dev);
spin_lock_irq(&lweventlist_lock);
}
/*鏈表有未處理事件，則以非緊急狀態(tài)調(diào)度隊(duì)列*/
if(!list_empty(&lweventlist))
linkwatch_schedule_work(0);
spin_unlock_irq(&lweventlist_lock);
}

linkwatch_do_dev()

linkwatch_do_dev()完成對(duì)某個(gè)設(shè)備的狀態(tài)改變處理。

staticvoidlinkwatch_do_dev(structnet_device*dev)
{
/*
*Makesuretheabovereadiscompletesinceitcanbe
*rewrittenassoonasweclearthebitbelow.
*/
smp_mb__before_atomic();

/*Weareabouttohandlethisdevice,
*soneweventscanbeaccepted
*/
/*清除pending標(biāo)記*/
clear_bit(__LINK_STATE_LINKWATCH_PENDING,&dev->state);

rfc2863_policy(dev);
/*如果設(shè)備啟動(dòng)狀態(tài)*/
if(dev->flags&IFF_UP){
/*鏈路連接*/
if(netif_carrier_ok(dev))
/*啟用排隊(duì)規(guī)則*/
dev_activate(dev);
else
/*關(guān)閉排隊(duì)規(guī)則*/
dev_deactivate(dev);
/*設(shè)備狀態(tài)改變處理，執(zhí)行netdev_chain上設(shè)備狀態(tài)變更回調(diào)*/
netdev_state_change(dev);
}
dev_put(dev);
}

phy_stop()

最后，hns_nic_net_down()中會(huì)調(diào)用phy_stop()將網(wǎng)卡link down。

voidphy_stop(structphy_device*phydev)
{
mutex_lock(&phydev->lock);

if(PHY_HALTED==phydev->state)
gotoout_unlock;

if(phy_interrupt_is_valid(phydev)){
/*DisablePHYInterrupts*/
phy_config_interrupt(phydev,PHY_INTERRUPT_DISABLED);

/*Clearanypendinginterrupts*/
phy_clear_interrupt(phydev);
}

phydev->state=PHY_HALTED;

out_unlock:
mutex_unlock(&phydev->lock);

/*Cannotcallflush_scheduled_work()hereasdesiredbecause
*ofrtnl_lock(),butPHY_HALTEDshallguaranteephy_change()
*willnotreenableinterrupts.
*/
}

phy_stop()將phydev->state設(shè)置為PHY_HALTED，將網(wǎng)卡關(guān)閉。

__dev_open

__dev_open為設(shè)備啟用核心函數(shù)，該函數(shù)打開eth0，設(shè)置啟用標(biāo)記，并且設(shè)置接收模式，排隊(duì)規(guī)則等。

staticint__dev_open(structnet_device*dev)
{
conststructnet_device_ops*ops=dev->netdev_ops;
intret;

ASSERT_RTNL();
/*設(shè)備不可用*/
if(!netif_device_present(dev))
return-ENODEV;

/*Blocknetpollfromtryingtodoanyrxpathservicing.
*Ifwedon'tdothisthereisachancendo_poll_controller
*orndo_pollmayberunningwhileweopenthedevice
*/
/*禁用netpoll*/
netpoll_poll_disable(dev);
/*設(shè)備打開前通知*/
ret=call_netdevice_notifiers(NETDEV_PRE_UP,dev);
ret=notifier_to_errno(ret);
if(ret)
returnret;
/*設(shè)置設(shè)備打開標(biāo)記,設(shè)備將設(shè)置IFF_UP標(biāo)志位*/
set_bit(__LINK_STATE_START,&dev->state);
/*校驗(yàn)地址*/
if(ops->ndo_validate_addr)
ret=ops->ndo_validate_addr(dev);
/*執(zhí)行打開*/
if(!ret&&ops->ndo_open)
ret=ops->ndo_open(dev);
/*啟用netpoll*/
netpoll_poll_enable(dev);
/*失敗，清除打開標(biāo)記*/
if(ret)
clear_bit(__LINK_STATE_START,&dev->state);
/*設(shè)備打開操作*/
else{
/*設(shè)置打開標(biāo)記*/
dev->flags|=IFF_UP;
/*設(shè)置接收模式*/
dev_set_rx_mode(dev);
/*初始化排隊(duì)規(guī)則*/
dev_activate(dev);
/*加入設(shè)備數(shù)據(jù)到熵池*/
add_device_randomness(dev->dev_addr,dev->addr_len);
}

returnret;
}

hns_nic_net_open()

我們以海思的網(wǎng)卡驅(qū)動(dòng)為例，分析下ndo_open()函數(shù)的實(shí)現(xiàn)。代碼位于kerneldrivers etethernethisiliconhnshns_enet.c。

staticinthns_nic_net_open(structnet_device*ndev)
{
structhns_nic_priv*priv=netdev_priv(ndev);
structhnae_handle*h=priv->ae_handle;
intret;

if(test_bit(NIC_STATE_TESTING,&priv->state))
return-EBUSY;

priv->link=0;
netif_carrier_off(ndev);
/*設(shè)置txqueue的個(gè)數(shù)*/
ret=netif_set_real_num_tx_queues(ndev,h->q_num);
if(retq_num);
if(ret

	

	hns_nic_net_up()

	
staticinthns_nic_net_up(structnet_device*ndev)
{
structhns_nic_priv*priv=netdev_priv(ndev);
structhnae_handle*h=priv->ae_handle;
inti,j,k;
intret;
/*初始化中斷，并設(shè)置中斷函數(shù)為hns_irq_handle，每個(gè)rx和txqueue都對(duì)應(yīng)一個(gè)中斷*/
ret=hns_nic_init_irq(priv);
if(ret!=0){
netdev_err(ndev,"hnsinitirqfailed!ret=%d
",ret);
returnret;
}

for(i=0;iq_num*2;i++){
/*使能中斷，使能napi*/
ret=hns_nic_ring_open(ndev,i);
if(ret)
gotoout_has_some_queues;
}

for(k=0;kq_num;k++)
h->dev->ops->toggle_queue_status(h->qs[k],1);
/*設(shè)置mac地址*/
ret=h->dev->ops->set_mac_addr(h,ndev->dev_addr);
if(ret)
gotoout_set_mac_addr_err;
/*hns的start函數(shù)為null*/
ret=h->dev->ops->start?h->dev->ops->start(h):0;
if(ret)
gotoout_start_err;

if(priv->phy)
/*啟動(dòng)phy*/
phy_start(priv->phy);

clear_bit(NIC_STATE_DOWN,&priv->state);
/*修改time每一秒到期一次*/
(void)mod_timer(&priv->service_timer,jiffies+SERVICE_TIMER_HZ);

return0;

out_start_err:
netif_stop_queue(ndev);
out_set_mac_addr_err:
for(k=0;kq_num;k++)
h->dev->ops->toggle_queue_status(h->qs[k],0);
out_has_some_queues:
for(j=i-1;j>=0;j--)
hns_nic_ring_close(ndev,j);

set_bit(NIC_STATE_DOWN,&priv->state);

returnret;
}


	

	phy_start()

	最后會(huì)調(diào)用到phy_start()啟動(dòng)網(wǎng)卡。

	
voidphy_start(structphy_device*phydev)
{
booldo_resume=false;
interr=0;

mutex_lock(&phydev->lock);

switch(phydev->state){
casePHY_STARTING:
phydev->state=PHY_PENDING;
break;
casePHY_READY:
phydev->state=PHY_UP;
break;
casePHY_HALTED:
/*makesureinterruptsarere-enabledforthePHY*/
err=phy_enable_interrupts(phydev);
if(errstate=PHY_RESUMING;
do_resume=true;
break;
default:
break;
}
mutex_unlock(&phydev->lock);

/*ifphywassuspended,bringthephysicallinkupagain*/
if(do_resume)
phy_resume(phydev);
}


	

	





	審核編輯：劉清