From: Johannes Berg <[email protected]>
My previous fix here to fix the deadlock left a race where
the exact same deadlock (see the original commit referenced
below) can still happen if cfg80211_destroy_ifaces() already
runs while nl80211_netlink_notify() is still marking some
interfaces as nl_owner_dead.
The race happens because we have two loops here - first we
dev_close() all the netdevs, and then we destroy them. If we
also have two netdevs (first one need only be a wdev though)
then we can find one during the first iteration, close it,
and go to the second iteration -- but then find two, and try
to destroy also the one we didn't close yet.
Fix this by only iterating once.
Change-Id: Ie56cd0ef3f0d2108bb8a25c8bb5efced15e6a909
Reported-by: Toke Høiland-Jørgensen <[email protected]>
Fixes: ea6b2098dd02 ("cfg80211: fix locking in netlink owner interface destruction")
Signed-off-by: Johannes Berg <[email protected]>
---
net/wireless/core.c | 17 ++++-------------
1 file changed, 4 insertions(+), 13 deletions(-)
diff --git a/net/wireless/core.c b/net/wireless/core.c
index ff74549b1022..d151a433388c 100644
--- a/net/wireless/core.c
+++ b/net/wireless/core.c
@@ -5,7 +5,7 @@
* Copyright 2006-2010 Johannes Berg <[email protected]>
* Copyright 2013-2014 Intel Mobile Communications GmbH
* Copyright 2015-2017 Intel Deutschland GmbH
- * Copyright (C) 2018-2021 Intel Corporation
+ * Copyright (C) 2018-2022 Intel Corporation
*/
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
@@ -332,29 +332,20 @@ static void cfg80211_event_work(struct work_struct *work)
void cfg80211_destroy_ifaces(struct cfg80211_registered_device *rdev)
{
struct wireless_dev *wdev, *tmp;
- bool found = false;
ASSERT_RTNL();
- list_for_each_entry(wdev, &rdev->wiphy.wdev_list, list) {
+ list_for_each_entry_safe(wdev, tmp, &rdev->wiphy.wdev_list, list) {
if (wdev->nl_owner_dead) {
if (wdev->netdev)
dev_close(wdev->netdev);
- found = true;
- }
- }
-
- if (!found)
- return;
- wiphy_lock(&rdev->wiphy);
- list_for_each_entry_safe(wdev, tmp, &rdev->wiphy.wdev_list, list) {
- if (wdev->nl_owner_dead) {
+ wiphy_lock(&rdev->wiphy);
cfg80211_leave(rdev, wdev);
rdev_del_virtual_intf(rdev, wdev);
+ wiphy_unlock(&rdev->wiphy);
}
}
- wiphy_unlock(&rdev->wiphy);
}
static void cfg80211_destroy_iface_wk(struct work_struct *work)
--
2.34.1
On Tue, 2022-02-01 at 16:35 +0200, Kalle Valo wrote:
> Johannes Berg <[email protected]> writes:
>
> > From: Johannes Berg <[email protected]>
> >
> > My previous fix here to fix the deadlock left a race where
> > the exact same deadlock (see the original commit referenced
> > below) can still happen if cfg80211_destroy_ifaces() already
> > runs while nl80211_netlink_notify() is still marking some
> > interfaces as nl_owner_dead.
> >
> > The race happens because we have two loops here - first we
> > dev_close() all the netdevs, and then we destroy them. If we
> > also have two netdevs (first one need only be a wdev though)
> > then we can find one during the first iteration, close it,
> > and go to the second iteration -- but then find two, and try
> > to destroy also the one we didn't close yet.
> >
> > Fix this by only iterating once.
> >
> > Change-Id: Ie56cd0ef3f0d2108bb8a25c8bb5efced15e6a909
>
> You forgot the Change-Id.
>
Wtf, sorry, guess I sent it from a place with the wrong script. I'll
strip it when I apply it, I guess
johannes
Johannes Berg <[email protected]> writes:
> From: Johannes Berg <[email protected]>
>
> My previous fix here to fix the deadlock left a race where
> the exact same deadlock (see the original commit referenced
> below) can still happen if cfg80211_destroy_ifaces() already
> runs while nl80211_netlink_notify() is still marking some
> interfaces as nl_owner_dead.
>
> The race happens because we have two loops here - first we
> dev_close() all the netdevs, and then we destroy them. If we
> also have two netdevs (first one need only be a wdev though)
> then we can find one during the first iteration, close it,
> and go to the second iteration -- but then find two, and try
> to destroy also the one we didn't close yet.
>
> Fix this by only iterating once.
>
> Change-Id: Ie56cd0ef3f0d2108bb8a25c8bb5efced15e6a909
> Reported-by: Toke Høiland-Jørgensen <[email protected]>
> Fixes: ea6b2098dd02 ("cfg80211: fix locking in netlink owner interface destruction")
> Signed-off-by: Johannes Berg <[email protected]>
Looks like this fixed the crash I was seeing - thanks!
Tested-by: Toke Høiland-Jørgensen <[email protected]>
Johannes Berg <[email protected]> writes:
> From: Johannes Berg <[email protected]>
>
> My previous fix here to fix the deadlock left a race where
> the exact same deadlock (see the original commit referenced
> below) can still happen if cfg80211_destroy_ifaces() already
> runs while nl80211_netlink_notify() is still marking some
> interfaces as nl_owner_dead.
>
> The race happens because we have two loops here - first we
> dev_close() all the netdevs, and then we destroy them. If we
> also have two netdevs (first one need only be a wdev though)
> then we can find one during the first iteration, close it,
> and go to the second iteration -- but then find two, and try
> to destroy also the one we didn't close yet.
>
> Fix this by only iterating once.
>
> Change-Id: Ie56cd0ef3f0d2108bb8a25c8bb5efced15e6a909
You forgot the Change-Id.
--
https://patchwork.kernel.org/project/linux-wireless/list/
https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches