ddt_object_remove (from ddt_sync_entry()) panics.

[BenEstrabaud]

Hi,

After starting a Volume block device (a similar Volume to LVM) and calling "zpool import" to import the previously created pool ontop of that Volume (single vdev pool), I get a "PANIC" from the ZFS driver in what seems to be the deduplication code:

<0>[ 1390.005410] VERIFY(ddt_object_remove(ddt, otype, oclass, dde, tx) == 0) failed
<0>[ 1390.013567] PANIC at ddt.c:1106:ddt_sync_entry()
<4>[ 1390.018797] Showing stack for process 24500
<4>[ 1390.023570] CPU: 0 PID: 24500 Comm: txg_sync Tainted: P O 4.1.5 #15
<4>[ 1390.031969] Hardware name: Newisys NDS_SB1EA/NDS_SB1EA, BIOS 9.20 07/28/2015
<4>[ 1390.031971] ffffffffa0695cff ffff881eb48d79e8 ffffffff816b4518 0000000000000000
<4>[ 1390.031972] ffff881eb48d7a18 ffff881eb48d79f8 ffffffffa0538fe4 ffff881eb48d7b88
<4>[ 1390.031973] ffffffffa05391b7 ffff881eb48d7ab8 ffffffffa069f93f 6428594649524556
<4>[ 1390.031974] Call Trace:
<4>[ 1390.031984] [] dump_stack+0x48/0x60
<4>[ 1390.031989] [] spl_dumpstack+0x44/0x50 [spl]
<4>[ 1390.031992] [] spl_panic+0xa7/0xe0 [spl]
<4>[ 1390.032009] [] ? fzap_byteswap+0x831/0x850 [zfs]
<4>[ 1390.032016] [] ? dbuf_rele+0x40/0x50 [zfs]
<4>[ 1390.032021] [] ? dmu_buf_rele+0xe/0x10 [zfs]
<4>[ 1390.032034] [] ? zap_unlockdir+0x4a/0xc0 [zfs]
<4>[ 1390.032036] [] ? spl_kmem_cache_free+0x16c/0x1e0 [spl]
<4>[ 1390.032044] [] ddt_sync+0x2e6/0xa40 [zfs]
<4>[ 1390.032057] [] ? zio_buf_free+0x251/0x12f0 [zfs]
<4>[ 1390.032068] [] ? zio_wait+0x15b/0x1e0 [zfs]
<4>[ 1390.032081] [] spa_sync+0x3bb/0xe80 [zfs]
<4>[ 1390.032083] [] ? __wake_up_common+0x59/0x90
<4>[ 1390.032084] [] ? __wake_up+0x53/0x70
<4>[ 1390.032087] [] ? getrawmonotonic64+0x3f/0xd0
<4>[ 1390.032099] [] txg_init+0x614/0x890 [zfs]
<4>[ 1390.032102] [] ? kfree+0x108/0x140
<4>[ 1390.032112] [] ? txg_init+0x240/0x890 [zfs]
<4>[ 1390.032115] [] ? __thread_create+0x160/0x1f0 [spl]
<4>[ 1390.032118] [] __thread_create+0x1d8/0x1f0 [spl]
<4>[ 1390.032120] [] ? __thread_create+0x160/0x1f0 [spl]
<4>[ 1390.032122] [] kthread+0xce/0x100
<4>[ 1390.032124] [] ? kthread_freezable_should_stop+0x70/0x70
<4>[ 1390.032125] [] ret_from_fork+0x42/0x70
<4>[ 1390.032127] [] ? kthread_freezable_should_stop+0x70/0x70
<6>[ 1428.259556] ixgbe 0000:01:00.1 eth3: NIC Link is Down

That issue only happened once after extensive export/import testing (think tens of iterations).

it seems to be like a "BUG ON" condition, where an assertion fails (ddt_object_remove(ddt, otype, oclass, dde, tx) == 0). Any idea what could be causing this?

[behlendorf]

@BenEstrabaud it would be great if you could tweak the code slightly so we can get the errno returned. This should work, then we may be able to say why.

diff --git a/module/zfs/ddt.c b/module/zfs/ddt.c
index 12c1b73..9df77cf 100644
--- a/module/zfs/ddt.c
+++ b/module/zfs/ddt.c
@@ -1103,7 +1103,7 @@ ddt_sync_entry(ddt_t *ddt, ddt_entry_t *dde, dmu_tx_t *tx,

        if (otype != DDT_TYPES &&
            (otype != ntype || oclass != nclass || total_refcnt == 0)) {
-               VERIFY(ddt_object_remove(ddt, otype, oclass, dde, tx) == 0);
+               VERIFY0(ddt_object_remove(ddt, otype, oclass, dde, tx));
                ASSERT(ddt_object_lookup(ddt, otype, oclass, dde) == ENOENT);
        }

[loli10K]

Duplicate of #1681