svn commit: r253687 - head/sys/net

[Alan Somers]

On Fri, Jul 26, 2013 at 1:41 PM, Adrian Chadd <adrian at freebsd.org> wrote:
> Author: adrian
> Date: Fri Jul 26 19:41:13 2013
> New Revision: 253687
> URL: http://svnweb.freebsd.org/changeset/base/253687
>
> Log:
>   Break out the static, global LACP debug options into a per-lagg unit
>   sysctl tree.
>
>   * Create a net.link.lagg.X.lacp node

I think this introduced a lock order reversal.

>   * Add a debug node under that for tx_test and rx_test
>   * Add lacp_strict_mode, defaulting to 1
>
>   tx_test and rx_test are still a bitmap of unit numbers for now.
>   At some point it would be nice to create child nodes of the lagg bundle
>   for each sub-interface, and then populate those with various knobs
>   and statistics.
>
>   Sponsored by: Netflix
>
> Modified:
>   head/sys/net/ieee8023ad_lacp.c
>   head/sys/net/ieee8023ad_lacp.h
>   head/sys/net/if_lagg.c
>   head/sys/net/if_lagg.h
>
> Modified: head/sys/net/ieee8023ad_lacp.c
> ==============================================================================
> --- head/sys/net/ieee8023ad_lacp.c      Fri Jul 26 19:11:08 2013        (r253686)
> +++ head/sys/net/ieee8023ad_lacp.c      Fri Jul 26 19:41:13 2013        (r253687)

<Extra chunks elided>
;
> @@ -765,10 +791,19 @@ lacp_attach(struct lagg_softc *sc)
>
>         lsc->lsc_hashkey = arc4random();
>         lsc->lsc_active_aggregator = NULL;
> +       lsc->lsc_strict_mode = 1;
>         LACP_LOCK_INIT(lsc);
>         TAILQ_INIT(&lsc->lsc_aggregators);
>         LIST_INIT(&lsc->lsc_ports);
>
> +       /* Create a child of the parent lagg interface */
> +       oid = SYSCTL_ADD_NODE(&sc->ctx, SYSCTL_CHILDREN(sc->sc_oid),
> +           OID_AUTO, "lacp", CTLFLAG_RD, NULL, "LACP");

This line grabs a sleepable lock, but we already had a nonsleepable
lock further up the stack, acquired in lagg_ioctl().

> +
> +       /* Attach sysctl nodes */
> +       lacp_attach_sysctl(lsc, oid);
> +       lacp_attach_sysctl_debug(lsc, oid);
> +
>         callout_init_mtx(&lsc->lsc_transit_callout, &lsc->lsc_mtx, 0);
>         callout_init_mtx(&lsc->lsc_callout, &lsc->lsc_mtx, 0);
>

Here's the warning from Witness.as well as a warning from UMA.  Many
more UMA warnings followed.

lock order reversal: (sleepable after non-sleepable)
1st 0xfffff8000252ca08 if_lagg rmlock (if_lagg rmlock) @
/usr/home/alans/freebsd/head/sys/modules/if_lagg/../../net/if_lagg.c:1040
2nd 0xffffffff814ef4e0 sysctl lock (sysctl lock) @
/usr/home/alans/freebsd/head/sys/kern/kern_sysctl.c:474
KDB: stack backtrace:
db_trace_self_wrapper() at db_trace_self_wrapper+0x2b/frame 0xfffffe00977485b0
kdb_backtrace() at kdb_backtrace+0x39/frame 0xfffffe0097748660
witness_checkorder() at witness_checkorder+0xdc2/frame 0xfffffe00977486f0
_sx_xlock() at _sx_xlock+0x75/frame 0xfffffe0097748730
sysctl_add_oid() at sysctl_add_oid+0x4a/frame 0xfffffe0097748780
lacp_attach() at lacp_attach+0xf7/frame 0xfffffe00977487f0
lagg_lacp_attach() at lagg_lacp_attach+0x88/frame 0xfffffe0097748810
lagg_ioctl() at lagg_ioctl+0x98a/frame 0xfffffe00977488f0
in_control() at in_control+0x38e/frame 0xfffffe0097748970
ifioctl() at ifioctl+0xba2/frame 0xfffffe0097748a30
kern_ioctl() at kern_ioctl+0x22b/frame 0xfffffe0097748a90
sys_ioctl() at sys_ioctl+0x13c/frame 0xfffffe0097748ae0
amd64_syscall() at amd64_syscall+0x25a/frame 0xfffffe0097748bf0
Xfast_syscall() at Xfast_syscall+0xfb/frame 0xfffffe0097748bf0
--- syscall (54, FreeBSD ELF64, sys_ioctl), rip = 0x800fa045a, rsp =
0x7fffffffe118, rbp = 0x7fffffffe1a0 ---
uma_zalloc_arg: zone "128" with the following non-sleepable locks held:
exclusive rm if_lagg rmlock (if_lagg rmlock) r = 0
(0xfffff8000252ca08) locked @
/usr/home/alans/freebsd/head/sys/modules/if_lagg/../../net/if_lagg.c:1040
KDB: stack backtrace:
db_trace_self_wrapper() at db_trace_self_wrapper+0x2b/frame 0xfffffe0097748500
kdb_backtrace() at kdb_backtrace+0x39/frame 0xfffffe00977485b0
witness_warn() at witness_warn+0x4b5/frame 0xfffffe0097748670
uma_zalloc_arg() at uma_zalloc_arg+0x3b/frame 0xfffffe00977486e0
malloc() at malloc+0x194/frame 0xfffffe0097748730
sysctl_add_oid() at sysctl_add_oid+0x11f/frame 0xfffffe0097748780
lacp_attach() at lacp_attach+0xf7/frame 0xfffffe00977487f0
lagg_lacp_attach() at lagg_lacp_attach+0x88/frame 0xfffffe0097748810
lagg_ioctl() at lagg_ioctl+0x98a/frame 0xfffffe00977488f0
in_control() at in_control+0x38e/frame 0xfffffe0097748970
ifioctl() at ifioctl+0xba2/frame 0xfffffe0097748a30
kern_ioctl() at kern_ioctl+0x22b/frame 0xfffffe0097748a90
sys_ioctl() at sys_ioctl+0x13c/frame 0xfffffe0097748ae0
amd64_syscall() at amd64_syscall+0x25a/frame 0xfffffe0097748bf0
Xfast_syscall() at Xfast_syscall+0xfb/frame 0xfffffe0097748bf0
--- syscall (54, FreeBSD ELF64, sys_ioctl), rip = 0x800fa045a, rsp =
0x7fffffffe118, rbp = 0x7fffffffe1a0 ---


# uname -a
FreeBSD alans-fbsd-head 11.0-CURRENT FreeBSD 11.0-CURRENT #49
r264887M: Thu Apr 24 17:21:48 MDT 2014
alans at ns1.eng.sldomain.com:/vmpool/obj/usr/home/alans/freebsd/head/sys/GENERIC
 amd64

To reproduce:
ifconfig tap0 create
ifconfig tap1 create
ifconfig tap2 create
ifconfig lagg0 create
ifconfig lagg0 up laggproto lacp laggport tap0 laggport tap1 laggport
tap2 192.0.0.2/24

If I create and destroy the lagg in a tight loop, while running
"ifconfig -am" in a tight loop in another terminal, I eventually hit a
general protection fault in __mtx_lock_sleep.  I think it might be
related.

Can you reproduce this?  Do you have any good ideas for a solution?

-Alan