| Message ID | 20230609033115.3738692-1-renmingshuai@huawei.com (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | [v3] net/sched: Set the flushing flags to false to prevent an infinite loop and add one test to tdc | expand |
| Context | Check | Description |
|---|---|---|
| netdev/tree_selection | success | Not a local patch |
On Fri, Jun 09, 2023 at 11:31:15AM +0800, renmingshuai wrote: > >On 08/06/2023 09:32, renmingshuai wrote: > >>> On 07/06/2023 01:19, renmingshuai wrote: > >>>>> On 06/06/2023 11:45, renmingshuai wrote: > >>>>>> When a new chain is added by using tc, one soft lockup alarm will > >>>>>> be > >>>>>> generated after delete the prio 0 filter of the chain. To > >>>>>> reproduce > >>>>>> the problem, perform the following steps: > >>>>>> (1) tc qdisc add dev eth0 root handle 1: htb default 1 > >>>>>> (2) tc chain add dev eth0 > >>>>>> (3) tc filter del dev eth0 chain 0 parent 1: prio 0 > >>>>>> (4) tc filter add dev eth0 chain 0 parent 1: > >>>>> > >>>>> This seems like it could be added to tdc or 3 and 4 must be run in > >>>>> parallel? > >>>> 3 and 4 do not need to be run inparallel. When a new chain is added > >>>> by the > >>>> way as step 1 and the step 3 is completed, this problem always > >>>> occurs > >>>> whenever step 4 is run. > >>> > >>> Got it, > >>> The test still hangs with the provided patch. > >>> > >>> + tc qdisc add dev lo root handle 1: htb default 1 > >>> + tc chain add dev lo > >>> + tc filter del dev lo chain 0 parent 1: prio 0 > >>> [ 68.790030][ T6704] [+] > >>> [ 68.790060][ T6704] chain refcnt 2 > >>> [ 68.790951][ T6704] [-] > >>> + tc filter add dev lo chain 0 parent 1: > >>> <hangs> > >>> > >>> Also please add this test to tdc, it should be straightforward. > >>> > >> Sorry for not testing before. I forgot that the chain->refcnt was > >> increased by 1 when tcf_chain_get() is called in tc_del_tfilter(). > >> The value of chain->refcnt is 2 after chain flush. The test > >> result is as follows: > >> [root@localhost ~]# tc qdisc add dev eth2 root handle 1: htb default 1 > >> [root@localhost ~]# tc chain add dev eth2 > >> [root@localhost ~]# tc filter del dev eth2 chain 0 parent 1: prio 0 > >> [root@localhost ~]# tc filter add dev eth2 chain 0 parent 1: > >> Error: Filter kind and protocol must be specified. > >> We have an error talking to the kernel > >> > >> And I have add this test to tdc: > >> [root@localhost tc-testing]# ./tdc.py -f tc-tests/filters/tests.json > >> ok 7 c2b4 - Adding a new fiter after deleting a filter in a chain does > >> not cause an infinite loop > >> > >> Fixes: 726d061286ce ("net: sched: prevent insertion of new classifiers during chain flush") > >> Signed-off-by: renmingshuai <renmingshuai@huawei.com> > > > >Please respin with the following applied: > > > >diff --git > >a/tools/testing/selftests/tc-testing/tc-tests/filters/tests.json > >b/tools/testing/selftests/tc-testing/tc-tests/filters/tests.json > >index c759c3db9a37..361235ad574b 100644 > >--- a/tools/testing/selftests/tc-testing/tc-tests/filters/tests.json > >+++ b/tools/testing/selftests/tc-testing/tc-tests/filters/tests.json > >@@ -125,25 +125,5 @@ > > "teardown": [ > > "$TC qdisc del dev $DEV2 ingress" > > ] > >- }, > >- { > >- "id": "c2b4", > >- "name": "Adding a new fiter after deleting a filter in a chain > >does not cause an infinite loop", > >- "category": [ > >- "filter", > >- "prio" > >- ], > >- "setup": [ > >- "$TC qdisc add dev $DEV1 root handle 1: htb default 1", > >- "$TC chain add dev $DEV1" > >- ], > >- "cmdUnderTest": "$TC filter del dev $DEV1 chain 0 parent 1: > >prio 0", > >- "expExitCode": "0", > >- "verifyCmd": "$TC filter add dev $DEV1 chain 0 parent 1:", > >- "matchPattern": "Error: Filter kind and protocol must be > >specified.", > >- "matchCount": "1", > >- "teardown": [ > >- "$TC qdisc del dev $DEV1 root handle 1: htb default 1" > >- ] > > } > > ] > >diff --git > >a/tools/testing/selftests/tc-testing/tc-tests/infra/filters.json > >b/tools/testing/selftests/tc-testing/tc-tests/infra/filters. > >json > >new file mode 100644 > >index 000000000000..55d6f209c388 > >--- /dev/null > >+++ b/tools/testing/selftests/tc-testing/tc-tests/infra/filters.json > >@@ -0,0 +1,24 @@ > >+[ > >+ { > >+ "id": "c2b4", > >+ "name": "Adding a new filter after flushing empty chain doesnt > >cause an infinite loop", > >+ "category": [ > >+ "filter", > >+ "chain" > >+ ], > >+ "setup": [ > >+ "$IP link add dev $DUMMY type dummy || /bin/true", > >+ "$TC qdisc add dev $DUMMY root handle 1: htb default 1", > >+ "$TC chain add dev $DUMMY", > >+ "$TC filter del dev $DUMMY chain 0 parent 1: prio 0" > >+ ], > >+ "cmdUnderTest": "$TC filter add dev $DUMMY chain 0 parent 1:", > >+ "expExitCode": "2", > >+ "verifyCmd": "$TC chain ls dev $DUMMY", > >+ "matchPattern": "chain parent 1: chain 0", > >+ "matchCount": "1", > >+ "teardown": [ > >+ "$TC qdisc del dev $DUMMY root handle 1: htb default 1" > >+ ] > >+ } > >+] > > Ok. The new test is passed. > [root@localhost tc-testing]# ./tdc.py -f tc-tests/infra/filter.json > Test c2b4: Adding a new filter after flushing empty chain doesn't cause an infinite loop > All test results: > 1..1 > ok 1 c2b4 - Adding a new filter after flushing empty chain doesn't cause an infinite loop > > Fixes: 726d061286ce ("net: sched: prevent insertion of new classifiers during chain flush") > Signed-off-by: renmingshuai <renmingshuai@huawei.com> Hi renmingshuai, the above text, in it's entirety, does not meet the needs of a patch submission for the Linux kernel. In particular: * All the context from previous discussion (the bits prefixed by '>') should be removed. * The patch description should consist of just that, a description of the patch. Any relevant test results are also welcome here. * Information about changes from previous versions should be provided in the form of a changelog, typically below the scissors (---). A recent example is here: - https://lore.kernel.org/all/20230607162353.3631199-1-mtottenh@akamai.com/ [PATCH v3] net/sched: act_pedit: Parse L3 Header for L4 offset Also, please consider using spaces and capital letters in your name in the from and signed-off parts of your submission. e.g.: Mingshuai Ren <...> or Ren Mingshuai <...> ...
diff --git a/net/sched/cls_api.c b/net/sched/cls_api.c index 2621550bfddc..3ea054e03fbf 100644 --- a/net/sched/cls_api.c +++ b/net/sched/cls_api.c @@ -2442,6 +2442,13 @@ static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n, tfilter_notify_chain(net, skb, block, q, parent, n, chain, RTM_DELTFILTER, extack); tcf_chain_flush(chain, rtnl_held); + /* Set the flushing flags to false to prevent an infinite loop + * when a new filter is added. + */ + mutex_lock(&chain->filter_chain_lock); + if (chain->refcnt == 2) + chain->flushing = false; + mutex_unlock(&chain->filter_chain_lock); err = 0; goto errout; } diff --git a/tools/testing/selftests/tc-testing/tc-tests/infra/filter.json b/tools/testing/selftests/tc-testing/tc-tests/infra/filter.json new file mode 100644 index 000000000000..db3b42aaa4fa --- /dev/null +++ b/tools/testing/selftests/tc-testing/tc-tests/infra/filter.json @@ -0,0 +1,25 @@ +[ + { + "id": "c2b4", + "name": "Adding a new filter after flushing empty chain doesn't cause an infinite loop", + "category": [ + "filter", + "chain" + ], + "setup": [ + "$IP link add dev $DUMMY type dummy || /bin/true", + "$TC qdisc add dev $DUMMY root handle 1: htb default 1", + "$TC chain add dev $DUMMY", + "$TC filter del dev $DUMMY chain 0 parent 1: prio 0" + ], + "cmdUnderTest": "$TC filter add dev $DUMMY chain 0 parent 1:", + "expExitCode": "2", + "verifyCmd": "$TC chain ls dev $DUMMY", + "matchPattern": "chain parent 1: chain 0", + "matchCount": "1", + "teardown": [ + "$TC qdisc del dev $DUMMY root handle 1: htb default 1", + "$IP link del dev $DUMMY type dummy" + ] + } +]