more feedback

Signed-off-by: Paul Dagnelie <pcd@delphix.com>

module/zfs/dsl_dataset.c

@@ -3086,6 +3086,14 @@ dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
 	mutex_enter(&dd->dd_activity_lock);
 	uint64_t holds = zfs_refcount_count(&ds->ds_longholds) -
 	    (owner != NULL ? 1 : 0);
+	/*
+	 * The value of dd_activity_waiters can chance as soon as we drop the
+	 * lock, but we're fine with that; new waiters coming in or old
+	 * waiters leaving doesn't cause problems, since we're going to cancel
+	 * waiters later anyway. The goal of this check is to verify that no
+	 * non-waiters have long-holds, and all new long-holds will be
+	 * prevented because we're holding the pool config as writer.
+	 */
 	if (holds != dd->dd_activity_waiters)
 		held = B_TRUE;
 	mutex_exit(&dd->dd_activity_lock);

module/zfs/zfs_ioctl.c

@@ -4120,6 +4120,16 @@ zfs_ioc_wait_fs(const char *name, nvlist_t *innvl, nvlist_t *outnvl)
 	mutex_enter(&dd->dd_activity_lock);
 	dd->dd_activity_waiters++;
 
+	/*
+	 * We get a long-hold here so that the dsl_dataset_t and dsl_dir_t
+	 * aren't evicted while we're waiting. Normally this is prevented by
+	 * holding the pool, but we can't do that while we're waiting since
+	 * that would prevent TXGs from syncing out. Some of the functionality
+	 * of long-holds (e.g. preventing deletion) is unnecessary for this
+	 * case, since we would cancel the waiters before proceeding with a
+	 * deletion. An alternative mechanism for keeping the dataset around
+	 * could be developed but this is simpler.
+	 */
 	dsl_dataset_long_hold(ds, FTAG);
 	dsl_pool_rele(dp, FTAG);
 

tests/zfs-tests/tests/functional/cli_root/zfs_wait/zfs_wait_deleteq.ksh

@@ -28,7 +28,7 @@
 # 4. Start a background process waiting for the delete queue to empty.
 # 5. Verify that the command doesn't return immediately.
 # 6. Close the open file descriptor.
-# 7. Verify that the command returns soon after the descriptor is closedd.
+# 7. Verify that the command returns soon after the descriptor is closed.
 #
 
 function cleanup
@@ -54,4 +54,4 @@ exec 3<&-
 log_must sleep 0.5
 bkgrnd_proc_succeeded $pid
 
-log_pass "'zfs wait -t discard' and 'zpool checkpoint -dw' work."
+log_pass "'zfs wait -t discard' works."