Re: [PATCH net] bonding: do not assume skb mac_header is set

[Jiri Pirko <jiri@resnulli.us>]

Thu, Jun 22, 2023 at 05:23:04PM CEST, edumazet@google.com wrote:
>Drivers must not assume in their ndo_start_xmit() that
>skbs have their mac_header set. skb->data is all what is needed.
>
>bonding seems to be one of the last offender as caught by syzbot:
>
>WARNING: CPU: 1 PID: 12155 at include/linux/skbuff.h:2907 skb_mac_offset include/linux/skbuff.h:2913 [inline]
>WARNING: CPU: 1 PID: 12155 at include/linux/skbuff.h:2907 bond_xmit_hash drivers/net/bonding/bond_main.c:4170 [inline]
>WARNING: CPU: 1 PID: 12155 at include/linux/skbuff.h:2907 bond_xmit_3ad_xor_slave_get drivers/net/bonding/bond_main.c:5149 [inline]
>WARNING: CPU: 1 PID: 12155 at include/linux/skbuff.h:2907 bond_3ad_xor_xmit drivers/net/bonding/bond_main.c:5186 [inline]
>WARNING: CPU: 1 PID: 12155 at include/linux/skbuff.h:2907 __bond_start_xmit drivers/net/bonding/bond_main.c:5442 [inline]
>WARNING: CPU: 1 PID: 12155 at include/linux/skbuff.h:2907 bond_start_xmit+0x14ab/0x19d0 drivers/net/bonding/bond_main.c:5470
>Modules linked in:
>CPU: 1 PID: 12155 Comm: syz-executor.3 Not tainted 6.1.30-syzkaller #0
>Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 05/25/2023
>RIP: 0010:skb_mac_header include/linux/skbuff.h:2907 [inline]
>RIP: 0010:skb_mac_offset include/linux/skbuff.h:2913 [inline]
>RIP: 0010:bond_xmit_hash drivers/net/bonding/bond_main.c:4170 [inline]
>RIP: 0010:bond_xmit_3ad_xor_slave_get drivers/net/bonding/bond_main.c:5149 [inline]
>RIP: 0010:bond_3ad_xor_xmit drivers/net/bonding/bond_main.c:5186 [inline]
>RIP: 0010:__bond_start_xmit drivers/net/bonding/bond_main.c:5442 [inline]
>RIP: 0010:bond_start_xmit+0x14ab/0x19d0 drivers/net/bonding/bond_main.c:5470
>Code: 8b 7c 24 30 e8 76 dd 1a 01 48 85 c0 74 0d 48 89 c3 e8 29 67 2e fe e9 15 ef ff ff e8 1f 67 2e fe e9 10 ef ff ff e8 15 67 2e fe <0f> 0b e9 45 f8 ff ff e8 09 67 2e fe e9 dc fa ff ff e8 ff 66 2e fe
>RSP: 0018:ffffc90002fff6e0 EFLAGS: 00010283
>RAX: ffffffff835874db RBX: 000000000000ffff RCX: 0000000000040000
>RDX: ffffc90004dcf000 RSI: 00000000000000b5 RDI: 00000000000000b6
>RBP: ffffc90002fff8b8 R08: ffffffff83586d16 R09: ffffffff83586584
>R10: 0000000000000007 R11: ffff8881599fc780 R12: ffff88811b6a7b7e
>R13: 1ffff110236d4f6f R14: ffff88811b6a7ac0 R15: 1ffff110236d4f76
>FS: 00007f2e9eb47700(0000) GS:ffff8881f6b00000(0000) knlGS:0000000000000000
>CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>CR2: 0000001b2e421000 CR3: 000000010e6d4000 CR4: 00000000003526e0
>DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
>DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
>Call Trace:
><TASK>
>[<ffffffff8471a49f>] netdev_start_xmit include/linux/netdevice.h:4925 [inline]
>[<ffffffff8471a49f>] __dev_direct_xmit+0x4ef/0x850 net/core/dev.c:4380
>[<ffffffff851d845b>] dev_direct_xmit include/linux/netdevice.h:3043 [inline]
>[<ffffffff851d845b>] packet_direct_xmit+0x18b/0x300 net/packet/af_packet.c:284
>[<ffffffff851c7472>] packet_snd net/packet/af_packet.c:3112 [inline]
>[<ffffffff851c7472>] packet_sendmsg+0x4a22/0x64d0 net/packet/af_packet.c:3143
>[<ffffffff8467a4b2>] sock_sendmsg_nosec net/socket.c:716 [inline]
>[<ffffffff8467a4b2>] sock_sendmsg net/socket.c:736 [inline]
>[<ffffffff8467a4b2>] __sys_sendto+0x472/0x5f0 net/socket.c:2139
>[<ffffffff8467a715>] __do_sys_sendto net/socket.c:2151 [inline]
>[<ffffffff8467a715>] __se_sys_sendto net/socket.c:2147 [inline]
>[<ffffffff8467a715>] __x64_sys_sendto+0xe5/0x100 net/socket.c:2147
>[<ffffffff8553071f>] do_syscall_x64 arch/x86/entry/common.c:50 [inline]
>[<ffffffff8553071f>] do_syscall_64+0x2f/0x50 arch/x86/entry/common.c:80
>[<ffffffff85600087>] entry_SYSCALL_64_after_hwframe+0x63/0xcd
>
>Fixes: 7b8fc0103bb5 ("bonding: add a vlan+srcmac tx hashing option")
>Reported-by: syzbot <syzkaller@googlegroups.com>
>Signed-off-by: Eric Dumazet <edumazet@google.com>
>Cc: Jarod Wilson <jarod@redhat.com>
>Cc: Moshe Tal <moshet@nvidia.com>
>Cc: Jussi Maki <joamaki@gmail.com>
>Cc: Jay Vosburgh <j.vosburgh@gmail.com>
>Cc: Andy Gospodarek <andy@greyhouse.net>
>Cc: Vladimir Oltean <vladimir.oltean@nxp.com>
>---
> drivers/net/bonding/bond_main.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
>diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
>index edbaa1444f8ecd9bf344a50f6f599d7eaaf4ff3e..091e035c76a6ff29facbaf1c0f26d185dc8ff5e3 100644
>--- a/drivers/net/bonding/bond_main.c
>+++ b/drivers/net/bonding/bond_main.c
>@@ -4197,7 +4197,7 @@ u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
> 		return skb->hash;
> 
> 	return __bond_xmit_hash(bond, skb, skb->data, skb->protocol,
>-				skb_mac_offset(skb), skb_network_offset(skb),
>+				0, skb_network_offset(skb),

After this change, both callers of __bond_xmit_hash() pass 0 as mhoff.
Wouldn't it make sense to remove this arg entirely here and in
bond_vlan_srcmac_hash() and bond_eth_hash()?


> 				skb_headlen(skb));
> }
> 
>-- 
>2.41.0.178.g377b9f9a00-goog
>
>