[Bug 205163] Periodic jobs triggering panics on ZFS host with jails

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=205163

--- Comment #3 from dustinwenz at ebureau.com ---
I was checking out some of the variables as they are set at frame 7 (below). If
it's to be believed, zfs_dirty_data_max_internal is much smaller than it should
be at that point in the code (maybe due to memory corruption, or a race
condition regarding zfs_dirty_data_max_internal). That would cause the
subtraction at line 1079 to wrap to some value a bit less than uint64 max, but
cast as a long long. I'm pretty sure that isn't how the code is intended to
work. If that is a problem, disabling the vfs.zfs_dynamic_write_buffer tunable
may help, because it should force zfs_dirty_data_max_internal to be the same as
zfs_dirty_data_max.

(kgdb) frame 7
#7  0xffffffff819f54d4 in dmu_tx_wait (tx=0xfffff805786ae400) at
/usr/src/sys/modules/zfs/../../cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_tx.c:1078
1078        min_tx_time = zfs_delay_scale *
(kgdb) list
1073         * cause a divide-by-zero if it's == the max.
1074         */
1075        ASSERT3U(dirty, <, zfs_dirty_data_max_internal);
1076    
1077        now = gethrtime();
1078        min_tx_time = zfs_delay_scale *
1079            (dirty - delay_min_bytes) / (zfs_dirty_data_max_internal -
dirty);
1080        if (now > tx->tx_start + min_tx_time)
1081            return;
1082    
(kgdb) print zfs_dirty_data_max 
$1 = 3429452595
(kgdb) print zfs_dirty_data_max_internal
$2 = 331001856
(kgdb) print dirty
$3 = 336166912

-- 
You are receiving this mail because:
You are the assignee for the bug.