rocker: do not modify fdb table in rocker_port_fdb() when preparing transactions
authorSimon Horman <simon.horman@netronome.com>
Thu, 21 May 2015 03:40:15 +0000 (12:40 +0900)
committerDavid S. Miller <davem@davemloft.net>
Thu, 21 May 2015 21:20:54 +0000 (17:20 -0400)
commit42e9488971f26c07719e99e33d14811488d31632
treeaddbdb12c7bce8b191d4c7b85a57868023da1b73
parent3098ac39630c8908fb13cd3137a4629c04df9fe3
rocker: do not modify fdb table in rocker_port_fdb() when preparing transactions

rocker_port_fdb_flush() may be called be called with
trans == SWITCHDEV_TRANS_PREPARE and then trans == SWITCHDEV_TRANS_COMMIT from
switchdev_port_attr_set() via switchdev_port_obj_add().

Adding the new entry to the FDB table when trans == SWITCHDEV_TRANS_PREPARE
may result in a memory leak because when trans == SWITCHDEV_TRANS_PREPARE
rocker_flow_tbl_bridge() will allocate memory when called via
rocker_port_fdb_learn(). However, when trans == SWITCHDEV_TRANS_COMMIT
the presence of the FDB entry in the FDB table causes
rocker_port_fdb() to set the ROCKER_OP_FLAG_REFRESH flag which results
in rocker_port_fdb_learn() skipping the call to rocker_flow_tbl_bridge()
which would free the memory allocated by it when
trans == SWITCHDEV_TRANS_PREPARE.

ip link add br0 type bridge
ip link set up dev eth0
ip link set dev eth0 master br0
bridge fdb add 52:54:00:12:35:08 dev eth0
bridge fdb add 52:54:00:12:35:09 dev eth0
[    2.600730] ------------[ cut here ]------------
[    2.601002] kernel BUG at drivers/net/ethernet/rocker/rocker.c:4369!
[    2.601373] invalid opcode: 0000 [#1] SMP
[    2.601963] Modules linked in:
[    2.602355] CPU: 0 PID: 64 Comm: bridge Not tainted 4.1.0-rc3-01048-g6d0f50c50211-dirty #1075
[    2.602721] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.8.0-0-g4c59f5d-20150219_092859-nilsson.home.kraxel.org 04/01/2014
[    2.602721] task: ffff880019facef0 ti: ffff88001f96c000 task.ti: ffff88001f96c000
[    2.602721] RIP: 0010:[<ffffffff811f1470>]  [<ffffffff811f1470>] rocker_port_obj_add+0x150/0x160
[    2.602721] RSP: 0018:ffff88001f96fa98  EFLAGS: 00000212
[    2.602721] RAX: ffff880019d4fa68 RBX: ffff88001f96fb18 RCX: 0000000000000000
[    2.602721] RDX: ffff880019d4f000 RSI: ffff88001f96fb18 RDI: ffff880019d4f000
[    2.602721] RBP: 0000000000000001 R08: 0000000000000000 R09: ffff88001f904620
[    2.602721] R10: ffff88001f96fb60 R11: ffff880019e9d100 R12: ffff88001f96fb18
[    2.602721] R13: ffff880019d4f680 R14: ffff88001f904610 R15: ffff8800198f7b80
[    2.602721] FS:  00007f3eee917700(0000) GS:ffff88001b000000(0000) knlGS:0000000000000000
[    2.602721] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[    2.602721] CR2: 00007f3eee4a15cb CR3: 000000001f933000 CR4: 00000000000006b0
[    2.602721] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[    2.602721] DR3: 0000000000000000 DR6: 0000000000000000 DR7: 0000000000000000
[    2.602721] Stack:
[    2.602721]  0000000000000000 ffff88001f96fb18 ffff880019d4f000 ffff88001f96fb18
[    2.602721]  ffff880019d4f000 ffffffff81332105 ffff88001f96fb50 ffffffff814464c0
[    2.602721]  ffff88001f96fb18 ffff88001f904600 ffff880019d4f000 ffffffff813326e5
[    2.602721] Call Trace:
[    2.602721]  [<ffffffff81332105>] ? __switchdev_port_obj_add+0x25/0x90
[    2.602721]  [<ffffffff813326e5>] ? switchdev_port_obj_add+0x25/0xc0
[    2.602721]  [<ffffffff813327b1>] ? switchdev_port_fdb_add+0x31/0x40
[    2.602721]  [<ffffffff8123911f>] ? rtnl_fdb_add+0xff/0x1e0
[    2.602721]  [<ffffffff81237d8e>] ? rtnetlink_rcv_msg+0x7e/0x250
[    2.602721]  [<ffffffff8121d1ce>] ? __skb_recv_datagram+0xfe/0x4b0
[    2.602721]  [<ffffffff81237d10>] ? rtnetlink_rcv+0x30/0x30
[    2.602721]  [<ffffffff81247958>] ? netlink_rcv_skb+0xa8/0xd0
[    2.602721]  [<ffffffff81237cff>] ? rtnetlink_rcv+0x1f/0x30
[    2.602721]  [<ffffffff81247220>] ? netlink_unicast+0x150/0x200
[    2.602721]  [<ffffffff81247714>] ? netlink_sendmsg+0x374/0x3e0
[    2.602721]  [<ffffffff8120f8df>] ? sock_sendmsg+0xf/0x30
[    2.602721]  [<ffffffff8120ffd3>] ? ___sys_sendmsg+0x1f3/0x200
[    2.602721]  [<ffffffff812100e5>] ? ___sys_recvmsg+0x105/0x140
[    2.602721]  [<ffffffff810a36f0>] ? SyS_readahead+0x90/0x90
[    2.602721]  [<ffffffff81098dfd>] ? filemap_map_pages+0x1ed/0x210
[    2.602721]  [<ffffffff810b77fc>] ? handle_mm_fault+0x5fc/0xe50
[    2.602721]  [<ffffffff81210ef9>] ? __sys_sendmsg+0x39/0x70
[    2.602721]  [<ffffffff8133ce17>] ? system_call_fastpath+0x12/0x6a
[    2.602721] Code: b7 8f a0 06 00 00 48 83 bf 88 06 00 00 00 74 1d 48 83 c4 08 89 ee 4c 89 ef 5b 5d 41 5c 41 5d 0f b7 c9 45 31 c0 e9 51 db ff ff 90 <0f> 0b b8 ea ff ff ff e9 cf fe ff ff 0f 1f 40 00 41 57 41 56 b9
[    2.602721] RIP  [<ffffffff811f1470>] rocker_port_obj_add+0x150/0x160
[    2.602721]  RSP <ffff88001f96fa98>
[    2.615848] ---[ end trace 4f7b4f1c98077108 ]---

The above is resolved by not adding the new FDB entry to the FDB table
if trans == SWITCHDEV_TRANS_PREPARE.

For symmetry this patch also skips deleting FDB entries from the FDB
table trans == SWITCHDEV_TRANS_PREPARE. However, my analysis is that
this never occurs as trans is always SWITCHDEV_TRANS_NONE when removing
FDB entries.

Fixes: c4f20321d968 ("rocker: support prepare-commit transaction model")
Acked-by: Scott Feldman <sfeldma@gmail.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/rocker/rocker.c