bonding: fix missed rcu protection
Subsystems (3)
PATCH RESEND 1/1 bonding: fix missed rcu protection
When removing the rcu_read_lock in bond_ethtool_get_ts_info(), I didn't notice it could be called via setsockopt, which doesn't hold rcu lock, as syzbot pointed:
stack backtrace: CPU: 0 PID: 3599 Comm: syz-executor317 Not tainted 5.18.0-rc5-syzkaller-01392-g01f4685797a5 #0 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 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 during the whole slave dev get_ts_info period.
---
Sorry, I hit send too quick in last patch and forgot to add net prefix. So resend the patch in this mail.
--- drivers/net/bonding/bond_main.c | 12 +++++++++--- 1 file changed, 9 insertions(+), 3 deletions(-)
drivers/net/bonding/bond_main.c +9 -3
@@ -5591,16 +5591,20 @@ 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); + 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 +5612,9 @@ static int bond_ethtool_get_ts_info(struct net_device *bond_dev,SOF_TIMESTAMPING_SOFTWARE ; info->phc_index =-1 ; -return 0 ; +out: +rcu_read_unlock (); +return ret; }static const struct ethtool_ops bond_ethtool_ops = {
Signature · 1 line
2.35.1
Quoted 27 lines · Hangbin Liu
When removing the
rcu_read_lockinbond_ethtool_get_ts_info(), I didn't notice it could be called via setsockopt, which doesn't hold rcu lock, as syzbot pointed:stack backtrace: CPU: 0 PID: 3599 Comm: syz-executor317 Not tainted 5.18.0-rc5-syzkaller-01392-g01f4685797a5 #0 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 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
Quoted 6 lines
Fix it by adding
rcu_read_lockduring the whole slave devget_ts_infoperiod.Reported-by [email protected]Fixes aa6034678e87 ("bonding: use rcu_dereference_rtnl when get bonding active slave")Suggested-by Vladimir Oltean <[email protected]>Signed-off-by Hangbin Liu <[email protected]>
Quoted 27 lines · Hangbin Liu
When removing the
rcu_read_lockinbond_ethtool_get_ts_info(), I didn't notice it could be called via setsockopt, which doesn't hold rcu lock, as syzbot pointed:stack backtrace: CPU: 0 PID: 3599 Comm: syz-executor317 Not tainted 5.18.0-rc5-syzkaller-01392-g01f4685797a5 #0 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 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
Quoted 7 lines
Fix it by adding
rcu_read_lockduring the whole slave devget_ts_infoperiod.Reported-by [email protected]Fixes aa6034678e87 ("bonding: use rcu_dereference_rtnl when get bonding active slave")Suggested-by Vladimir Oltean <[email protected]>Signed-off-by Hangbin Liu <[email protected]>---
General feedback, the whole stack trace kind of clutters the commit message and you could trim most of it. Otherwise the patch looks good to me.
Jakub Kicinski
maintainerQuoted 16 lines · On Fri, 13 May 2022 18:33:50 +0800 Hangbin Liu wrote:
drivers/net/bonding/bond_main.c +2
-0@@ -5591,16 +5591,20 @@ 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;
drivers/net/bonding/bond_main.c static int bond_ethtool_get_ts_info(struct net_device *bond_dev, · line 5609
Quoted 10 lines
drivers/net/bonding/bond_main.c +4
-2if (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; } }
Can't ->get_ts_info sleep now? It'd be a little sad to force it to be atomic just because of one upper dev trying to be fancy. Maybe all we need to do is to take a ref on the real_dev?
Also please add a Link: to the previous discussion, it'd have been useful to get the context in which Vladimir suggested this.
Quoted 3 lines · Jakub Kicinski
Can't ->get_ts_info sleep now? It'd be a little sad to force it to be atomic just because of one upper dev trying to be fancy. Maybe all we need to do is to take a ref on the
real_dev?
Do you mean
drivers/net/bonding/bond_main.c +10 -3
@@ -5591,16 +5591,20 @@ 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 ; real_dev =bond_option_active_slave_get_rcu (bond);if (real_dev) { +dev_hold (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 +5612,10 @@ static int bond_ethtool_get_ts_info(struct net_device *bond_dev,SOF_TIMESTAMPING_SOFTWARE ; info->phc_index =-1 ; -return 0 ; +out: +if (real_dev) +dev_put (real_dev); +return ret; }
This look OK to me.
Vladimir, Jay, WDYT?
Quoted 3 lines
Also please add a Link: to the previous discussion, it'd have been useful to get the context in which Vladimir suggested this.
OK, I will.
Thanks
Hangbin
Paolo Abeni
maintainerQuoted 12 lines · Hangbin Liu
Quoted 3 lines · Jakub Kicinski
Can't ->get_ts_info sleep now? It'd be a little sad to force it to be atomic just because of one upper dev trying to be fancy. Maybe all we need to do is to take a ref on the
real_dev?Do you mean
drivers/net/bonding/bond_main.c +0
-0
drivers/net/bonding/bond_main.c static int bond_ethtool_get_ts_info(struct net_device *bond_dev, · line 5595
Quoted 5 lines
drivers/net/bonding/bond_main.c +1
-0@@ -5591,16 +5591,20 @@ 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 ;
You additionally need something alike the following:
rcu_read_lock();
drivers/net/bonding/bond_main.c static int bond_ethtool_get_ts_info(struct net_device *bond_dev, · line 5598
Quoted 3 lines
drivers/net/bonding/bond_main.c +1
-0real_dev =bond_option_active_slave_get_rcu (bond);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 5608
Quoted 12 lines
drivers/net/bonding/bond_main.c +4
-2ops = 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 5609
Quoted 1 line
drivers/net/bonding/bond_main.c +0
-0}
... or you will hit the initial RCU splat. Overall this will not put atomicy constraint on get_ts_info.
Cheers,
Paol
Quoted 26 lines
drivers/net/bonding/bond_main.c +4
-1@@ -5608,7 +5612,10 @@ static int bond_ethtool_get_ts_info(struct net_device *bond_dev,SOF_TIMESTAMPING_SOFTWARE ; info->phc_index =-1 ; -return 0 ; +out: +if (real_dev) +dev_put (real_dev); +return ret; }This look OK to me.
Vladimir, Jay, WDYT?
Quoted 3 lines
Also please add a Link: to the previous discussion, it'd have been useful to get the context in which Vladimir suggested this.
OK, I will.
Thanks
Hangbin
Quoted 21 lines · Paolo Abeni
Quoted 12 lines · Hangbin Liu
Quoted 3 lines · Jakub Kicinski
Can't ->get_ts_info sleep now? It'd be a little sad to force it to be atomic just because of one upper dev trying to be fancy. Maybe all we need to do is to take a ref on the
real_dev?Do you mean
drivers/net/bonding/bond_main.c +0
-0drivers/net/bonding/bond_main.c static int bond_ethtool_get_ts_info(struct net_device *bond_dev, · line 5595
Quoted 5 lines
drivers/net/bonding/bond_main.c +1
-0@@ -5591,16 +5591,20 @@ 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 ;You additionally need something alike the following:
Quoted 1 line
rcu_read_lock();
Thanks Paolo, I only thought the real_dev ref and forgot the initial problem is the rcu warning...
Hangbin