From: Tom Caputi <tcaputi@datto.com>
Date: Tue, 16 Oct 2018 21:00:55 +0000 (-0400)
Subject: Fix lock inversion in txg_sync_thread()
X-Git-Tag: zfs-0.8.0~332
X-Git-Url: https://git.proxmox.com/?a=commitdiff_plain;h=a783dd96843699744e0543b579b5f50c9023432f;p=mirror_zfs.git

Fix lock inversion in txg_sync_thread()

This patch fixes a lock inversion issue in txg_sync_thread() where
the code would attempt hold the spa config lock as a reader while
holding tx->tx_sync_lock. This races with spa_vdev_remove() which
attempts to hold the tx->tx_sync_lock to assign a new tx (via
spa_history_log_internal()) while holding the spa config lock as a
writer.

Reviewed-by: Serapheim Dimitropoulos <serapheim.dimitro@delphix.com>
Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Co-authored-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes #8010
---

diff --git a/module/zfs/txg.c b/module/zfs/txg.c
index aad1a8e57..db0f60cd1 100644
--- a/module/zfs/txg.c
+++ b/module/zfs/txg.c
@@ -562,22 +562,22 @@ txg_sync_thread(void *arg)
 		tx->tx_quiesced_txg = 0;
 		tx->tx_syncing_txg = txg;
 		DTRACE_PROBE2(txg__syncing, dsl_pool_t *, dp, uint64_t, txg);
-		txg_stat_t *ts = spa_txg_history_init_io(spa, txg, dp);
 		cv_broadcast(&tx->tx_quiesce_more_cv);
 
 		dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n",
 		    txg, tx->tx_quiesce_txg_waiting, tx->tx_sync_txg_waiting);
 		mutex_exit(&tx->tx_sync_lock);
 
+		txg_stat_t *ts = spa_txg_history_init_io(spa, txg, dp);
 		start = ddi_get_lbolt();
 		spa_sync(spa, txg);
 		delta = ddi_get_lbolt() - start;
+		spa_txg_history_fini_io(spa, ts);
 
 		mutex_enter(&tx->tx_sync_lock);
 		tx->tx_synced_txg = txg;
 		tx->tx_syncing_txg = 0;
 		DTRACE_PROBE2(txg__synced, dsl_pool_t *, dp, uint64_t, txg);
-		spa_txg_history_fini_io(spa, ts);
 		cv_broadcast(&tx->tx_sync_done_cv);
 
 		/*