Add TRIM support

UNMAP/TRIM support is a frequently-requested feature to help
prevent performance from degrading on SSDs and on various other
SAN-like storage back-ends.  By issuing UNMAP/TRIM commands for
sectors which are no longer allocated the underlying device can
often more efficiently manage itself.

This TRIM implementation is modeled on the `zpool initialize`
feature which writes a pattern to all unallocated space in the
pool.  The new `zpool trim` command uses the same vdev_xlate()
code to calculate what sectors are unallocated, the same per-
vdev TRIM thread model and locking, and the same basic CLI for
a consistent user experience.  The core difference is that
instead of writing a pattern it will issue UNMAP/TRIM commands
for those extents.

The zio pipeline was updated to accommodate this by adding a new
ZIO_TYPE_TRIM type and associated spa taskq.  This new type makes
is straight forward to add the platform specific TRIM/UNMAP calls
to vdev_disk.c and vdev_file.c.  These new ZIO_TYPE_TRIM zios are
handled largely the same way as ZIO_TYPE_READs or ZIO_TYPE_WRITEs.
This makes it possible to largely avoid changing the pipieline,
one exception is that TRIM zio's may exceed the 16M block size
limit since they contain no data.

In addition to the manual `zpool trim` command, a background
automatic TRIM was added and is controlled by the 'autotrim'
property.  It relies on the exact same infrastructure as the
manual TRIM.  However, instead of relying on the extents in a
metaslab's ms_allocatable range tree, a ms_trim tree is kept
per metaslab.  When 'autotrim=on', ranges added back to the
ms_allocatable tree are also added to the ms_free tree.  The
ms_free tree is then periodically consumed by an autotrim
thread which systematically walks a top level vdev's metaslabs.

Since the automatic TRIM will skip ranges it considers too small
there is value in occasionally running a full `zpool trim`.  This
may occur when the freed blocks are small and not enough time
was allowed to aggregate them.  An automatic TRIM and a manual
`zpool trim` may be run concurrently, in which case the automatic
TRIM will yield to the manual TRIM.

Reviewed-by: Jorgen Lundman <[email protected]>
Reviewed-by: Tim Chase <[email protected]>
Reviewed-by: Matt Ahrens <[email protected]>
Reviewed-by: George Wilson <[email protected]>
Reviewed-by: Serapheim Dimitropoulos <[email protected]>
Contributions-by: Saso Kiselkov <[email protected]>
Contributions-by: Tim Chase <[email protected]>
Contributions-by: Chunwei Chen <[email protected]>
Signed-off-by: Brian Behlendorf <[email protected]>
Closes #8419 
Closes #598 

1 files changed, 6 insertions, 139 deletions

diff --git a/module/zfs/vdev_initialize.c b/module/zfs/vdev_initialize.c
index bca2db7a4..b15901326 100644
--- a/module/zfs/vdev_initialize.c
+++ b/module/zfs/vdev_initialize.c
@@ -34,12 +34,6 @@
 #include <sys/dmu_tx.h>
 
 /*
- * Maximum number of metaslabs per group that can be initialized
- * simultaneously.
- */
-int max_initialize_ms = 3;
-
-/*
  * Value that is written to disk during initialization.
  */
 #ifdef _ILP32
@@ -132,7 +126,7 @@ vdev_initialize_change_state(vdev_t *vd, vdev_initializing_state_t new_state)
 	dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
 	VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
 	dsl_sync_task_nowait(spa_get_dsl(spa), vdev_initialize_zap_update_sync,
-	    guid, 2, ZFS_SPACE_CHECK_RESERVED, tx);
+	    guid, 2, ZFS_SPACE_CHECK_NONE, tx);
 
 	switch (new_state) {
 	case VDEV_INITIALIZE_ACTIVE:
@@ -251,49 +245,6 @@ vdev_initialize_write(vdev_t *vd, uint64_t start, uint64_t size, abd_t *data)
 }
 
 /*
- * Translate a logical range to the physical range for the specified vdev_t.
- * This function is initially called with a leaf vdev and will walk each
- * parent vdev until it reaches a top-level vdev. Once the top-level is
- * reached the physical range is initialized and the recursive function
- * begins to unwind. As it unwinds it calls the parent's vdev specific
- * translation function to do the real conversion.
- */
-void
-vdev_xlate(vdev_t *vd, const range_seg_t *logical_rs, range_seg_t *physical_rs)
-{
-	/*
-	 * Walk up the vdev tree
-	 */
-	if (vd != vd->vdev_top) {
-		vdev_xlate(vd->vdev_parent, logical_rs, physical_rs);
-	} else {
-		/*
-		 * We've reached the top-level vdev, initialize the
-		 * physical range to the logical range and start to
-		 * unwind.
-		 */
-		physical_rs->rs_start = logical_rs->rs_start;
-		physical_rs->rs_end = logical_rs->rs_end;
-		return;
-	}
-
-	vdev_t *pvd = vd->vdev_parent;
-	ASSERT3P(pvd, !=, NULL);
-	ASSERT3P(pvd->vdev_ops->vdev_op_xlate, !=, NULL);
-
-	/*
-	 * As this recursive function unwinds, translate the logical
-	 * range into its physical components by calling the
-	 * vdev specific translate function.
-	 */
-	range_seg_t intermediate = { { { 0, 0 } } };
-	pvd->vdev_ops->vdev_op_xlate(vd, physical_rs, &intermediate);
-
-	physical_rs->rs_start = intermediate.rs_start;
-	physical_rs->rs_end = intermediate.rs_end;
-}
-
-/*
  * Callback to fill each ABD chunk with zfs_initialize_value. len must be
  * divisible by sizeof (uint64_t), and buf must be 8-byte aligned. The ABD
  * allocation will guarantee these for us.
@@ -363,81 +314,6 @@ vdev_initialize_ranges(vdev_t *vd, abd_t *data)
 }
 
 static void
-vdev_initialize_mg_wait(metaslab_group_t *mg)
-{
-	ASSERT(MUTEX_HELD(&mg->mg_ms_initialize_lock));
-	while (mg->mg_initialize_updating) {
-		cv_wait(&mg->mg_ms_initialize_cv, &mg->mg_ms_initialize_lock);
-	}
-}
-
-static void
-vdev_initialize_mg_mark(metaslab_group_t *mg)
-{
-	ASSERT(MUTEX_HELD(&mg->mg_ms_initialize_lock));
-	ASSERT(mg->mg_initialize_updating);
-
-	while (mg->mg_ms_initializing >= max_initialize_ms) {
-		cv_wait(&mg->mg_ms_initialize_cv, &mg->mg_ms_initialize_lock);
-	}
-	mg->mg_ms_initializing++;
-	ASSERT3U(mg->mg_ms_initializing, <=, max_initialize_ms);
-}
-
-/*
- * Mark the metaslab as being initialized to prevent any allocations
- * on this metaslab. We must also track how many metaslabs are currently
- * being initialized within a metaslab group and limit them to prevent
- * allocation failures from occurring because all metaslabs are being
- * initialized.
- */
-static void
-vdev_initialize_ms_mark(metaslab_t *msp)
-{
-	ASSERT(!MUTEX_HELD(&msp->ms_lock));
-	metaslab_group_t *mg = msp->ms_group;
-
-	mutex_enter(&mg->mg_ms_initialize_lock);
-
-	/*
-	 * To keep an accurate count of how many threads are initializing
-	 * a specific metaslab group, we only allow one thread to mark
-	 * the metaslab group at a time. This ensures that the value of
-	 * ms_initializing will be accurate when we decide to mark a metaslab
-	 * group as being initialized. To do this we force all other threads
-	 * to wait till the metaslab's mg_initialize_updating flag is no
-	 * longer set.
-	 */
-	vdev_initialize_mg_wait(mg);
-	mg->mg_initialize_updating = B_TRUE;
-	if (msp->ms_initializing == 0) {
-		vdev_initialize_mg_mark(mg);
-	}
-	mutex_enter(&msp->ms_lock);
-	msp->ms_initializing++;
-	mutex_exit(&msp->ms_lock);
-
-	mg->mg_initialize_updating = B_FALSE;
-	cv_broadcast(&mg->mg_ms_initialize_cv);
-	mutex_exit(&mg->mg_ms_initialize_lock);
-}
-
-static void
-vdev_initialize_ms_unmark(metaslab_t *msp)
-{
-	ASSERT(!MUTEX_HELD(&msp->ms_lock));
-	metaslab_group_t *mg = msp->ms_group;
-	mutex_enter(&mg->mg_ms_initialize_lock);
-	mutex_enter(&msp->ms_lock);
-	if (--msp->ms_initializing == 0) {
-		mg->mg_ms_initializing--;
-		cv_broadcast(&mg->mg_ms_initialize_cv);
-	}
-	mutex_exit(&msp->ms_lock);
-	mutex_exit(&mg->mg_ms_initialize_lock);
-}
-
-static void
 vdev_initialize_calculate_progress(vdev_t *vd)
 {
 	ASSERT(spa_config_held(vd->vdev_spa, SCL_CONFIG, RW_READER) ||
@@ -535,9 +411,8 @@ vdev_initialize_load(vdev_t *vd)
 	return (err);
 }
 
-
 /*
- * Convert the logical range into a physcial range and add it to our
+ * Convert the logical range into a physical range and add it to our
  * avl tree.
  */
 void
@@ -618,7 +493,8 @@ vdev_initialize_thread(void *arg)
 			ms_count = vd->vdev_top->vdev_ms_count;
 		}
 
-		vdev_initialize_ms_mark(msp);
+		spa_config_exit(spa, SCL_CONFIG, FTAG);
+		metaslab_disable(msp);
 		mutex_enter(&msp->ms_lock);
 		VERIFY0(metaslab_load(msp));
 
@@ -626,16 +502,8 @@ vdev_initialize_thread(void *arg)
 		    vd);
 		mutex_exit(&msp->ms_lock);
 
-		spa_config_exit(spa, SCL_CONFIG, FTAG);
 		error = vdev_initialize_ranges(vd, deadbeef);
-
-		/*
-		 * Wait for the outstanding IO to be synced to prevent
-		 * newly allocated blocks from being overwritten.
-		 */
-		txg_wait_synced(spa_get_dsl(spa), 0);
-
-		vdev_initialize_ms_unmark(msp);
+		metaslab_enable(msp, B_TRUE);
 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
 
 		range_tree_vacate(vd->vdev_initialize_tree, NULL, NULL);
@@ -853,12 +721,11 @@ vdev_initialize_restart(vdev_t *vd)
 }
 
 #if defined(_KERNEL)
-EXPORT_SYMBOL(vdev_initialize_restart);
-EXPORT_SYMBOL(vdev_xlate);
 EXPORT_SYMBOL(vdev_initialize);
 EXPORT_SYMBOL(vdev_initialize_stop);
 EXPORT_SYMBOL(vdev_initialize_stop_all);
 EXPORT_SYMBOL(vdev_initialize_stop_wait);
+EXPORT_SYMBOL(vdev_initialize_restart);
 
 /* CSTYLED */
 module_param(zfs_initialize_value, ulong, 0644);