Created attachment 1701925 [details]
Shell script to reproduce the issue

# uname -a
Linux hp-dl388g8-04.rhts.eng.pek2.redhat.com 4.18.0-226.el8.x86_64 #1 SMP Wed Jul 15 07:40:39 UTC 2020 x86_64 x86_64 x86_64 GNU/Linux


# nmcli -o con show br_test
connection.id:                          br_test
connection.uuid:                        96c53115-7475-47e8-be16-23c1083a1838
connection.type:                        bridge
connection.interface-name:              br_test
connection.timestamp:                   1595342459
connection.autoconnect-slaves:          1 (yes)
connection.lldp:                        disable
802-3-ethernet.cloned-mac-address:      AA:CF:A8:2D:6A:48
802-3-ethernet.mtu:                     1500
ipv4.method:                            manual
ipv4.addresses:                         192.168.0.254/24
ipv4.dhcp-client-id:                    mac
ipv6.method:                            manual
ipv6.addresses:                         2000::a/64
ipv6.addr-gen-mode:                     eui64
ipv6.dhcp-duid:                         ll
ipv6.dhcp-iaid:                         mac
bridge.stp:                             no


It seems not to reproduce initially when activating/creating the device. However, it could be reproduced by externally clearing multicast_snooping (via sysfs), and then reactivate the profil: 

[1595339576.9233] platform-linux: sysctl: setting 'net:/sys/class/net/br_test/bridge/multicast_snooping' to '1' (current value is '0')
[1595339576.9233] platform-linux: sysctl: failed to set 'bridge/multicast_snooping' to '1': (17) File exists





With the attached script I am sometimes able to reproduce this. It does not always happen, but when it happens you see

  SETTING SNOOPING:
  ./rh1859244-reproducer.sh: line 24: echo: write error: File exists
  SETTING SNOOPING DONE
  0



Seems upstream kernel reworked this code heavily in https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=19e3a9c90c53479fecaa02307bf2db5ab8b3ffe3



Reassigning to Kernel for investigation.

The root cause is a race condition between br_multicast_set_hash_max() and br_multicast_toggle()...

A setting of sysfs variable bridge/hash_max calls br_multicast_set_hash_max() that calls br_mdb_rehash() to copy old hash table to new hashtable. The hashtable is protected by RCU and the old one freed after grace period. If you set bridge/multicast_snooping that calls br_multicast_toggle() before grace period the old hash table still exists and -EEXIST is returned.

The mentioned commit removes the custom hash-table implementation in facor of Linux generic rhashtable. After this refactoring the mentioned race-condition is no longer possible.

Simplified reproducer:

#!/bin/sh

ip link add br_test type bridge
ip link set up br_test

n=1
while true; do
        ip link set down br_test
        echo 0 > /sys/class/net/br_test/bridge/multicast_snooping
        sleep 0.1
        ip link set up br_test
        echo 4096 > /sys/class/net/br_test/bridge/hash_max
        echo 1 > /sys/class/net/br_test/bridge/multicast_snooping || break
        n=$((n+1))
        test "$n" -eq 100 && break
done

echo "Number of attempts: $n"

ip link del br_test

Without proposed patch-set:

[root@el8lat ~]# ./test.sh 
./test.sh: line 13: echo: write error: File exists
Number of attempts: 2

With proposed patch-set:

[root@el8lat ~]# ./test.sh 
Number of attempts: 100

reproduce with the reproducer in comment5

./test.sh 
./test.sh: line 12: echo: write error: File exists
Number of attempts: 2


# nmcli -v
nmcli tool, version 1.26.0-8.el8
# uname -a
Linux dell-per740-16.rhts.eng.pek2.redhat.com 4.18.0-240.el8.x86_64 #1 SMP Wed Sep 23 05:13:10 EDT 2020 x86_64 x86_64 x86_64 GNU/Linux

Patch(es) available on kernel-4.18.0-240.8.el8.dt4

# uname -a
Linux dell-per740-26.rhts.eng.pek2.redhat.com 4.18.0-240.8.el8.dt4.x86_64 #1 SMP Fri Oct 30 14:11:00 EDT 2020 x86_64 x86_64 x86_64 GNU/Linux# cat test.sh 
ip link add br_test type bridge
ip link set up br_test

n=1
while true; do
        ip link set down br_test
        echo 0 > /sys/class/net/br_test/bridge/multicast_snooping
        sleep 0.1
        ip link set up br_test
        echo 4096 > /sys/class/net/br_test/bridge/hash_max
        echo 1 > /sys/class/net/br_test/bridge/multicast_snooping || break
        n=$((n+1))
        test "$n" -eq 100 && break
done

echo "Number of attempts: $n"

ip link del br_test

[root@dell-per740-26 ~]# ./test.sh 
Number of attempts: 100


run bridge case , didn't find new issue

Patch(es) available on kernel-4.18.0-242.el8

# uname -r
4.18.0-242.el8.x86_64

[root@dell-per740-16 ~]# ./test.sh 
Number of attempts: 100
[root@dell-per740-16 ~]# cat test.sh 
#!/bin/sh

ip link add br_test type bridge
ip link set up br_test

n=1
while true; do
        ip link set down br_test
        echo 0 > /sys/class/net/br_test/bridge/multicast_snooping
        sleep 0.1
        ip link set up br_test
        echo 4096 > /sys/class/net/br_test/bridge/hash_max
        echo 1 > /sys/class/net/br_test/bridge/multicast_snooping || break
        n=$((n+1))
        test "$n" -eq 100 && break
done

echo "Number of attempts: $n"

ip link del br_test

run bridge case , didn't find new issue
https://beaker.engineering.redhat.com/jobs/4709353
https://beaker.engineering.redhat.com/jobs/4709354

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory (Important: kernel security, bug fix, and enhancement update), and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://access.redhat.com/errata/RHSA-2021:1578