Skip to content

mac80211: fix rate_control_send_low warnings for delbas

Luis R. Rodriguez linux-wireless 3 versions – New

Get series with b4
b4 am [email protected]

Versions newest first · linked by author and title

  1. v3 Latest3 patches 0 repliesNew
  2. v23 patches 30 repliesSuperseded
  3. v11 patch 12 repliesSuperseded

Patches in v3 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/3mac80211: fix channel assumption for association done worknet/mac80211/mlme.c+3 -30 replies
2/3mac80211: wait until completely disassociated before new associationnet/mac80211/mlme.c net/mac80211/work.c+16 -10 replies
3/3mac80211: move to the home channel for disassociation when roamingnet/mac80211/mlme.c+9 -00 replies

Reviews trailers per patch

No Reviewed-by, Acked-by or Tested-by trailers for v3 yet.

Read the v3 thread

Cover letter [PATCH v3 0/3] mac80211: fix rate_control_send_low warnings for delbas

Note: this v3 is the same as v2, I just trimmed the CC list as I was not seeing the patches posted to linux-wireless. Please use the v3 thread for comments.

The following two patches:

9c87ba6 - mac80211: Fix reassociation processing (within ESS roaming) e1dd33f - cfg80211: Allow reassociation in associated state

$ git describe --contains 9c87ba6
v2.6.34-rc2~48^2~77^2~6
$ git describe --contains 9c87ba6
v2.6.34-rc2~48^2~77^2~6

Added support for cfg80211/mac80211 to cleanly roam between two BSSes on an ESS by allowing the station to authenticate to two APS at the same time, and when an association comes in for the new AP we first disassociate from the old AP and then associate with the new one.

What we forgot to take into consideration is that when we disassociate with the older AP we may need to transmit frames to that AP and those frames may actually be intended to go under a different channel and even sometimes a completely separate band than the new APs. When we TX a frame we assume the frame we want to TX however will be on the current hardware configured channel. The channel we would try to send a frame on can be different than the channel we prepared the bitrates for the peer on though. What this meant is that upon tearing down a BA agreement we would try to send a frame to a peer but not find a valid rate for that peer and generate warnings like the following:

WARNING: at include/net/mac80211.h:2677 rate_control_send_low+0xd3/0x140 [mac80211]()
Hardware name: 6460DWU
Modules linked in: ath9k mac80211 ath9k_common ath9k_hw ath cfg80211 <etc>
Pid: 898, comm: wpa_supplicant Tainted: G        W   2.6.36-rc5-wl+ #254
Call Trace:
 [<ffffffff8105fddf>] warn_slowpath_common+0x7f/0xc0
 [<ffffffff8105fe3a>] warn_slowpath_null+0x1a/0x20
 [<ffffffffa03b8573>] rate_control_send_low+0xd3/0x140 [mac80211]
 [<ffffffffa0192fa8>] ath_get_rate+0x48/0x570 [ath9k]
 [<ffffffff812b8d19>] ? put_dec+0x59/0x60
 [<ffffffffa03b839e>] rate_control_get_rate+0x8e/0x190 [mac80211]
 [<ffffffffa03c27e8>] ieee80211_tx_h_rate_ctrl+0x1a8/0x4e0 [mac80211]
 [<ffffffffa03c2ec0>] invoke_tx_handlers+0x100/0x140 [mac80211]
 [<ffffffffa03c2f85>] ieee80211_tx+0x85/0x240 [mac80211]
 [<ffffffff81479c70>] ? skb_release_data+0xd0/0xe0
 [<ffffffff8147bb0d>] ? pskb_expand_head+0x10d/0x1a0
 [<ffffffffa03c31f6>] ieee80211_xmit+0xb6/0x1d0 [mac80211]
 [<ffffffff81479db3>] ? __alloc_skb+0x83/0x170
 [<ffffffffa03c3364>] ieee80211_tx_skb+0x54/0x70 [mac80211]
 [<ffffffffa03ac0dd>] ieee80211_send_delba+0x11d/0x190 [mac80211]
 [<ffffffffa03adcf0>] ___ieee80211_stop_rx_ba_session+0xf0/0x110 [mac80211]
 [<ffffffffa03add60>] __ieee80211_stop_rx_ba_session+0x50/0x70 [mac80211]
 [<ffffffffa03ac3f3>] ieee80211_sta_tear_down_BA_sessions+0x43/0x50 [mac80211]
 [<ffffffffa03b23c3>] ieee80211_set_disassoc+0x103/0x240 [mac80211]
 [<ffffffffa03b3b2d>] ieee80211_mgd_assoc+0x8d/0x3a0 [mac80211]
 [<ffffffffa03ba66f>] ieee80211_assoc+0x4f/0x80 [mac80211]
 [<ffffffffa011e5b6>] __cfg80211_mlme_assoc+0x1f6/0x240 [cfg80211]
 [<ffffffffa011e68f>] cfg80211_mlme_assoc+0x8f/0xc0 [cfg80211]
 [<ffffffffa010afd0>] ? cfg80211_get_dev_from_ifindex+0x70/0x80 [cfg80211]
 [<ffffffffa011489a>] nl80211_associate+0x23a/0x260 [cfg80211]
 [<ffffffff812c6c6f>] ? nla_parse+0xef/0x110
 [<ffffffff814ad738>] genl_rcv_msg+0x1d8/0x210
 [<ffffffff81475cf4>] ? sock_alloc_send_pskb+0x1d4/0x330
 [<ffffffff814ad560>] ? genl_rcv_msg+0x0/0x210
 [<ffffffff814ac179>] netlink_rcv_skb+0xa9/0xd0
 [<ffffffff814ad545>] genl_rcv+0x25/0x40
 [<ffffffff814abdd8>] netlink_unicast+0x2c8/0x2e0
 [<ffffffff814acc30>] netlink_sendmsg+0x250/0x360
 [<ffffffff81472643>] sock_sendmsg+0xf3/0x120
 [<ffffffff81562dbe>] ? _raw_spin_lock+0xe/0x20
 [<ffffffff81471105>] ? move_addr_to_kernel+0x65/0x70
 [<ffffffff8147d168>] ? verify_iovec+0x88/0xe0
 [<ffffffff81472d70>] sys_sendmsg+0x240/0x3a0
 [<ffffffff81151b0a>] ? do_readv_writev+0x1aa/0x1f0
 [<ffffffff815604b0>] ? schedule+0x3c0/0xa00
 [<ffffffff81151b98>] ? vfs_writev+0x48/0x60
 [<ffffffff81151cc1>] ? sys_writev+0x51/0xb0
 [<ffffffff8100b072>] system_call_fastpath+0x16/0x1b

This can be easily reproduced with the test-roam script [1] and statically defining the ESS variable with the BSSes of an AP in 2.4 GHz band and another one on the 5 GHz band.

Since we end up authenticate to two different APs at the same time to transmit to the new AP consists of an offchannel operation and while mac80211 does have a state machine for this it assumes we don't transmit on our operating channel within each work item. If we do need to transmit to on the operating channel each work item needs to put us on the home channel for that operation as we also use the operatin channel as the target channel for the skb we are going to transmit. This series addresses this by addressing a series of possible races on the frame's set channel, prevening us from associating to a new AP if we were previously associated and haven't yet associated, and ensuring we transmit the disassocation on the operating channel.

For more information refer to:

http://code.google.com/p/chromium-os/issues/detail?id=6348
http://marc.info/?l=linux-wireless&m=128401139730423

This patch series has a fix for kernels >= v2.6.34. The fix will actually cure the warnings correctly, it ensures we tear down our BA agreements with our old APs prior to moving away from them and ensures we transmit those frames on the intended channel.

[1] http://www.kernel.org/pub/linux/kernel/people/mcgrof/scripts/test-roam

Luis R. Rodriguez (3):
mac80211: fix channel assumption for association done work
mac80211: wait until completely disassociated before new association
mac80211: move to the home channel for disassociation when roaming

 net/mac80211/mlme.c |   18 ++++++++++++++----
 net/mac80211/work.c |   14 ++++++++++++++
 2 files changed, 28 insertions(+), 4 deletions(-)