Skip to contentSkip to conversation

bonding: fix missed rcu protection

PATCH Superseded Hangbin Liu 5 participants 7 messages May 17, 2022 – May 19, 2022

Subsystems (3)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 bonding: fix missed rcu protection

When removing the rcu_read_lock in bond_ethtool_get_ts_info() as discussed [1], I didn't notice it could be called via setsockopt, which doesn't hold rcu lock, as syzbot pointed:

  Call Trace:
   <TASK>
   __dump_stack lib/dump_stack.c:88 [inline]
   dump_stack_lvl+0xcd/0x134 lib/dump_stack.c:106
   bond_option_active_slave_get_rcu include/net/bonding.h:353 [inline]
   bond_ethtool_get_ts_info+0x32c/0x3a0 drivers/net/bonding/bond_main.c:5595
   __ethtool_get_ts_info+0x173/0x240 net/ethtool/common.c:554
   ethtool_get_phc_vclocks+0x99/0x110 net/ethtool/common.c:568
   sock_timestamping_bind_phc net/core/sock.c:869 [inline]
   sock_set_timestamping+0x3a3/0x7e0 net/core/sock.c:916
   sock_setsockopt+0x543/0x2ec0 net/core/sock.c:1221
   __sys_setsockopt+0x55e/0x6a0 net/socket.c:2223
   __do_sys_setsockopt net/socket.c:2238 [inline]
   __se_sys_setsockopt net/socket.c:2235 [inline]
   __x64_sys_setsockopt+0xba/0x150 net/socket.c:2235
   do_syscall_x64 arch/x86/entry/common.c:50 [inline]
   do_syscall_64+0x35/0xb0 arch/x86/entry/common.c:80
   entry_SYSCALL_64_after_hwframe+0x44/0xae
  RIP: 0033:0x7f8902c8eb39

Fix it by adding rcu_read_lock and take a ref on the real_dev.

[1] https://lore.kernel.org/netdev/27565.1642742439@famine/

Reported-by [email protected]
Fixes aa6034678e87 ("bonding: use rcu_dereference_rtnl when get bonding active slave")

---
v2: add ref on the real_dev as Jakub and Paolo suggested.

---
 drivers/net/bonding/bond_main.c | 17 ++++++++++++++---
 1 file changed, 14 insertions(+), 3 deletions(-)
drivers/net/bonding/bond_main.c +14 -3
index 38e152548126..fcaa5ccea7af 100644
@@ -5591,24 +5591,35 @@ static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
 	const struct ethtool_ops *ops;
 	struct net_device *real_dev;
 	struct phy_device *phydev;
+	int ret = 0;
 
+	rcu_read_lock();
 	real_dev = bond_option_active_slave_get_rcu(bond);
 	if (real_dev) {
+		dev_hold(real_dev);
+		rcu_read_unlock();
 		ops = real_dev->ethtool_ops;
 		phydev = real_dev->phydev;
 
 		if (phy_has_tsinfo(phydev)) {
-			return phy_ts_info(phydev, info);
+			ret = phy_ts_info(phydev, info);
+			goto out;
 		} else if (ops->get_ts_info) {
-			return ops->get_ts_info(real_dev, info);
+			ret = ops->get_ts_info(real_dev, info);
+			goto out;
 		}
+	} else {
+		rcu_read_unlock();
 	}
 
 	info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE |
 				SOF_TIMESTAMPING_SOFTWARE;
 	info->phc_index = -1;
 
-	return 0;
+out:
+	if (real_dev)
+		dev_put(real_dev);
+	return ret;
 }
 
 static const struct ethtool_ops bond_ethtool_ops = {
Signature · 1 line
2.35.1
#1 lore raw

---
RESEND, list still didn't receive my last version

The diffstat is slightly larger but IMO a slightly more readable version. When I was reading v2 I found myself jumping around.
I only compile tested it, so YMMV.

If this amount of change is too much v2 from Hangbin looks correct to me.

 drivers/net/bonding/bond_main.c | 31 ++++++++++++++++++++-----------
 1 file changed, 20 insertions(+), 11 deletions(-)
drivers/net/bonding/bond_main.c +20 -11
index 38e152548126..f9d27b63c454 100644
@@ -5591,23 +5591,32 @@ static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
 	const struct ethtool_ops *ops;
 	struct net_device *real_dev;
 	struct phy_device *phydev;
+	int ret = 0;
 
+	rcu_read_lock();
 	real_dev = bond_option_active_slave_get_rcu(bond);
-	if (real_dev) {
-		ops = real_dev->ethtool_ops;
-		phydev = real_dev->phydev;
-
-		if (phy_has_tsinfo(phydev)) {
-			return phy_ts_info(phydev, info);
-		} else if (ops->get_ts_info) {
-			return ops->get_ts_info(real_dev, info);
-		}
-	}
+	if (real_dev)
+		dev_hold(real_dev);
+	rcu_read_unlock();
+
+	if (!real_dev)
+		goto software;
 
+	ops = real_dev->ethtool_ops;
+	phydev = real_dev->phydev;
+
+	if (phy_has_tsinfo(phydev))
+		ret = phy_ts_info(phydev, info);
+	else if (ops->get_ts_info)
+		ret = ops->get_ts_info(real_dev, info);
+
+	dev_put(real_dev);
+	return ret;
+
+software:
 	info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE |
 				SOF_TIMESTAMPING_SOFTWARE;
 	info->phc_index = -1;
-
 	return 0;
 }
 
Signature · 1 line
2.27.0
#2 lore raw
Quoted 6 lines · Jonathan Toppins

---
RESEND, list still didn't receive my last version

The diffstat is slightly larger but IMO a slightly more readable version. When I was reading v2 I found myself jumping around.

Hi Jon,

Thanks for the commit. But..

Quoted 26 lines
drivers/net/bonding/bond_main.c +6 -10
index 38e152548126..f9d27b63c454 100644
@@ -5591,23 +5591,32 @@ static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
 	const struct ethtool_ops *ops;
 	struct net_device *real_dev;
 	struct phy_device *phydev;
+	int ret = 0;
 
+	rcu_read_lock();
 	real_dev = bond_option_active_slave_get_rcu(bond);
-	if (real_dev) {
-		ops = real_dev->ethtool_ops;
-		phydev = real_dev->phydev;
-
-		if (phy_has_tsinfo(phydev)) {
-			return phy_ts_info(phydev, info);
-		} else if (ops->get_ts_info) {
-			return ops->get_ts_info(real_dev, info);
-		}
-	}
+	if (real_dev)
+		dev_hold(real_dev);
+	rcu_read_unlock();
+

drivers/net/bonding/bond_main.c static int bond_ethtool_get_ts_info(struct net_device *bond_dev, · line 5611

Quoted 10 lines
drivers/net/bonding/bond_main.c +9 -0
+	if (!real_dev)
+		goto software;
 
+	ops = real_dev->ethtool_ops;
+	phydev = real_dev->phydev;
+
+	if (phy_has_tsinfo(phydev))
+		ret = phy_ts_info(phydev, info);
+	else if (ops->get_ts_info)
+		ret = ops->get_ts_info(real_dev, info);
	else {
		dev_put(real_dev);
		goto software;
	}

Here we need another check and goto software if !phy_has_tsinfo() and no ops->get_ts_info. With this change we also have 2 goto and dev_put().

drivers/net/bonding/bond_main.c static int bond_ethtool_get_ts_info(struct net_device *bond_dev, · line 5621

Quoted 11 lines
drivers/net/bonding/bond_main.c +5 -1
+
+	dev_put(real_dev);
+	return ret;
+
+software:
 	info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE |
 				SOF_TIMESTAMPING_SOFTWARE;
 	info->phc_index = -1;
-
 	return 0;
 }

As Jakub remind, dev_hold() and dev_put() can take NULL now. So how about this new patch:

drivers/net/bonding/bond_main.c +12 -3
index 38e152548126..b5c5196e03ee 100644
@@ -5591,16 +5591,23 @@ static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
 	const struct ethtool_ops *ops;
 	struct net_device *real_dev;
 	struct phy_device *phydev;
+	int ret = 0;
 
+	rcu_read_lock();
 	real_dev = bond_option_active_slave_get_rcu(bond);
+	dev_hold(real_dev);
+	rcu_read_unlock();
+
 	if (real_dev) {
 		ops = real_dev->ethtool_ops;
 		phydev = real_dev->phydev;
 
 		if (phy_has_tsinfo(phydev)) {
-			return phy_ts_info(phydev, info);
+			ret = phy_ts_info(phydev, info);
+			goto out;
 		} else if (ops->get_ts_info) {
-			return ops->get_ts_info(real_dev, info);
+			ret = ops->get_ts_info(real_dev, info);
+			goto out;
 		}
 	}
 
@@ -5608,7 +5615,9 @@ static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
 				SOF_TIMESTAMPING_SOFTWARE;
 	info->phc_index = -1;
 
-	return 0;
+out:
+	dev_put(real_dev);
+	return ret;
 }
 

Thanks
Hangbin

#3 lore raw
Quoted 53 lines · Hangbin Liu
Quoted 6 lines · Jonathan Toppins

---
RESEND, list still didn't receive my last version

The diffstat is slightly larger but IMO a slightly more readable version. When I was reading v2 I found myself jumping around.

Hi Jon,

Thanks for the commit. But..

Quoted 26 lines
drivers/net/bonding/bond_main.c +6 -10
index 38e152548126..f9d27b63c454 100644
@@ -5591,23 +5591,32 @@ static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
  	const struct ethtool_ops *ops;
  	struct net_device *real_dev;
  	struct phy_device *phydev;
+	int ret = 0;
 
+	rcu_read_lock();
  	real_dev = bond_option_active_slave_get_rcu(bond);
-	if (real_dev) {
-		ops = real_dev->ethtool_ops;
-		phydev = real_dev->phydev;
-
-		if (phy_has_tsinfo(phydev)) {
-			return phy_ts_info(phydev, info);
-		} else if (ops->get_ts_info) {
-			return ops->get_ts_info(real_dev, info);
-		}
-	}
+	if (real_dev)
+		dev_hold(real_dev);
+	rcu_read_unlock();
+

drivers/net/bonding/bond_main.c static int bond_ethtool_get_ts_info(struct net_device *bond_dev, · line 5611

Quoted 10 lines
drivers/net/bonding/bond_main.c +9 -0
+	if (!real_dev)
+		goto software;
 
+	ops = real_dev->ethtool_ops;
+	phydev = real_dev->phydev;
+
+	if (phy_has_tsinfo(phydev))
+		ret = phy_ts_info(phydev, info);
+	else if (ops->get_ts_info)
+		ret = ops->get_ts_info(real_dev, info);
	else {
		dev_put(real_dev);
		goto software;
	}
Quoted 2 lines

Here we need another check and goto software if !phy_has_tsinfo() and no ops->get_ts_info. With this change we also have 2 goto and dev_put().

Ah yes. I cannot think of a way to make this simpler. The patch below looks good.

Quoted 59 lines

drivers/net/bonding/bond_main.c static int bond_ethtool_get_ts_info(struct net_device *bond_dev, · line 5621

Quoted 12 lines
drivers/net/bonding/bond_main.c +5 -1
+
+	dev_put(real_dev);
+	return ret;
+
+software:
  	info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE |
  				SOF_TIMESTAMPING_SOFTWARE;
  	info->phc_index = -1;
-
  	return 0;
  }

As Jakub remind, dev_hold() and dev_put() can take NULL now. So how about this new patch:

drivers/net/bonding/bond_main.c +12 -3
index 38e152548126..b5c5196e03ee 100644
@@ -5591,16 +5591,23 @@ static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
  	const struct ethtool_ops *ops;
  	struct net_device *real_dev;
  	struct phy_device *phydev;
+	int ret = 0;
 
+	rcu_read_lock();
  	real_dev = bond_option_active_slave_get_rcu(bond);
+	dev_hold(real_dev);
+	rcu_read_unlock();
+
  	if (real_dev) {
  		ops = real_dev->ethtool_ops;
  		phydev = real_dev->phydev;
 
  		if (phy_has_tsinfo(phydev)) {
-			return phy_ts_info(phydev, info);
+			ret = phy_ts_info(phydev, info);
+			goto out;
  		} else if (ops->get_ts_info) {
-			return ops->get_ts_info(real_dev, info);
+			ret = ops->get_ts_info(real_dev, info);
+			goto out;
  		}
  	}
 
@@ -5608,7 +5615,9 @@ static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
  				SOF_TIMESTAMPING_SOFTWARE;
  	info->phc_index = -1;
 
-	return 0;
+out:
+	dev_put(real_dev);
+	return ret;
  }
 

Thanks
Hangbin

#4 lore raw
In reply to Jonathan Toppins
Quoted 10 lines · Jonathan Toppins

---
RESEND, list still didn't receive my last version

The diffstat is slightly larger but IMO a slightly more readable version. When I was reading v2 I found myself jumping around.
I only compile tested it, so YMMV.

If this amount of change is too much v2 from Hangbin looks correct to me.

Seems to be too big of a change for what the issue is, yes, sorry.

maintainer
#5 lore raw
In reply to Hangbin Liu
Quoted 19 lines · On Tue, 17 May 2022 16:23:12 +0800 Hangbin Liu wrote:
drivers/net/bonding/bond_main.c +9 -2
+	rcu_read_lock();
 	real_dev = bond_option_active_slave_get_rcu(bond);
 	if (real_dev) {
+		dev_hold(real_dev);
+		rcu_read_unlock();
 		ops = real_dev->ethtool_ops;
 		phydev = real_dev->phydev;
 
 		if (phy_has_tsinfo(phydev)) {
-			return phy_ts_info(phydev, info);
+			ret = phy_ts_info(phydev, info);
+			goto out;
 		} else if (ops->get_ts_info) {
-			return ops->get_ts_info(real_dev, info);
+			ret = ops->get_ts_info(real_dev, info);
+			goto out;
 		}
+	} else {
+		rcu_read_unlock();

drivers/net/bonding/bond_main.c static int bond_ethtool_get_ts_info(struct net_device *bond_dev, · line 5621

Quoted 10 lines
drivers/net/bonding/bond_main.c +3 -1
 	}
 
 	info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE |
 				SOF_TIMESTAMPING_SOFTWARE;
 	info->phc_index = -1;
 
-	return 0;
+out:
+	if (real_dev)
+		dev_put(real_dev);

dev_hold() and dev_put() can take NULL these days, for better or worse. I think the code simplification is worth making use of that, even tho it will make the backport slightly more tricky (perhaps make a not of this in the commit message).

#6 lore raw
Quoted 4 lines · On Tue, 17 May 2022 10:54:45 -0700 Jakub Kicinski <[email protected]> wrote:

dev_hold() and dev_put() can take NULL these days, for better or worse. I think the code simplification is worth making use of that, even tho it will make the backport slightly more tricky (perhaps make a not of this in the commit message).

Since that is so, would be worth having coccinelle script to cleanup existing code.

See scripts/coccinelle/ifnullfree.cocci for similar example.