41  * TRIM is a feature which is used to notify a SSD that some previously
47  * There are two supported TRIM methods; manual and automatic.
49  * Manual TRIM:
51  * A manual TRIM is initiated by running the 'zpool trim' command.  A single
53  * managing that vdev TRIM process.  This involves iterating over all the
55  * required TRIM I/Os.
60  * the TRIM are regularly written to the pool.  This allows the TRIM to be
63  * Automatic TRIM:
65  * An automatic TRIM is enabled by setting the 'autotrim' pool property
68  * core TRIM process as a manual TRIM, but with a few key differences.
70  * 1) Automatic TRIM happens continuously in the background and operates
76  *    metaslab is disabled at a time.  Unlike manual TRIM, this means each
77  *    'vdev_autotrim' thread is responsible for issuing TRIM I/Os for its
80  * 3) There is no automatic TRIM progress information stored on disk, nor
83  * While the automatic TRIM process is highly effective it is more likely
84  * than a manual TRIM to encounter tiny ranges.  Ranges less than or equal to
86  * TRIM and are skipped.  This means small amounts of freed space may not
93  * For this reason it may be beneficial to occasionally manually TRIM a pool
94  * even when automatic TRIM is enabled.
98  * Maximum size of TRIM I/O, ranges will be chunked in to 128MiB lengths.
103  * Minimum size of TRIM I/O, extents smaller than 32Kib will be skipped.
108  * Skip uninitialized metaslabs during the TRIM process.  This option is
110  * TRIM operations are slow.  As a pool ages an increasing fraction of
113  * manual TRIM and will persist for the duration of the requested TRIM.
118  * Maximum number of queued TRIM I/Os per leaf vdev.  The number of
119  * concurrent TRIM I/Os issued to the device is controlled by the
127  * efficient TRIM operations, by allowing them to be aggregated longer,
133  * time.  This can result is larger TRIM operations, and increased memory
143  * trimmed and a range tree containing the extents to TRIM.  All provided
150 	vdev_t		*trim_vdev;		/* Leaf vdev to TRIM */
152 	zfs_range_tree_t	*trim_tree;	/* TRIM ranges (in metaslab) */
153 	trim_type_t	trim_type;		/* Manual or auto TRIM */
154 	uint64_t	trim_extent_bytes_max;	/* Maximum TRIM I/O size */
155 	uint64_t	trim_extent_bytes_min;	/* Minimum TRIM I/O size */
156 	enum trim_flag	trim_flags;		/* TRIM flags (secure) */
209  * The sync task for updating the on-disk state of a manual TRIM.  This
288  * Update the on-disk state of a manual TRIM.  This is called to request
289  * that a TRIM be started/suspended/canceled, or to change one of the
290  * TRIM options (partial, secure, rate).
316 	 * If we're activating, then preserve the requested rate and trim  in vdev_trim_change_state()
352 		spa_history_log_internal(spa, "trim", tx,  in vdev_trim_change_state()
357 		spa_history_log_internal(spa, "trim", tx,  in vdev_trim_change_state()
364 			spa_history_log_internal(spa, "trim", tx,  in vdev_trim_change_state()
370 		spa_history_log_internal(spa, "trim", tx,  in vdev_trim_change_state()
384  * The zio_done_func_t done callback for each manual TRIM issued.  It is
385  * responsible for updating the TRIM stats, reissuing failed TRIM I/Os,
386  * and limiting the number of in flight TRIM I/Os.
425  * The zio_done_func_t done callback for each automatic TRIM issued.  It
426  * is responsible for updating the TRIM stats and limiting the number of
427  * in flight TRIM I/Os.  Automatic TRIM I/Os are best effort and are
455  * The zio_done_func_t done callback for each TRIM issued via
456  * vdev_trim_simple(). It is responsible for updating the TRIM stats and
457  * limiting the number of in flight TRIM I/Os.  Simple TRIM I/Os are best
484  * Returns the average trim rate in bytes/sec for the ta->trim_vdev.
494  * Issues a physical TRIM and takes care of rate limiting (bytes/sec)
495  * and number of concurrent TRIM I/Os.
507 	 * Limit manual TRIM I/Os to the requested rate.  This does not  in vdev_trim_range()
508 	 * apply to automatic TRIM since no per vdev rate can be specified.  in vdev_trim_range()
585  * Issues TRIM I/Os for all ranges in the provided ta->trim_tree range tree.
586  * Additional parameters describing how the TRIM should be performed must
632 	 * returning. TRIM zios have lower priority over regular or syncing  in vdev_trim_ranges()
633 	 * zios, so all TRIM zios for this metaslab must complete before the  in vdev_trim_ranges()
635 	 * this metaslab could cut ahead of still queued TRIM zios for this  in vdev_trim_ranges()
674  * Calculates the completion percentage of a manual TRIM.
748  * Load from disk the vdev's manual TRIM information.  This includes the
749  * state, progress, and options provided when initiating the manual TRIM.
816 	 * Only a manual trim will be traversing the vdev sequentially.  in vdev_trim_xlate_range_add()
817 	 * For an auto trim all valid ranges should be added.  in vdev_trim_xlate_range_add()
870  * Each manual TRIM thread is responsible for trimming the unallocated
872  * over its top-level metaslabs and issuing TRIM I/O for the space described
909 	 * When a secure TRIM has been requested infer that the intent  in vdev_trim_thread()
911 	 * minimum TRIM size to prevent ranges from being skipped.  in vdev_trim_thread()
938 		 * If a partial TRIM was requested skip metaslabs which have  in vdev_trim_thread()
985 	 * check to see if it needs to restart a trim. That thread will be  in vdev_trim_thread()
1001  * Initiates a manual TRIM for the vdev_t.  Callers must hold vdev_trim_lock,
1037  * Wait for vdev trim threads which were listed to cleanly exit.
1059  * required to call vdev_trim_stop_wait() to block for all the trim threads
1073 	 * Allow cancel requests to proceed even if the trim thread has  in vdev_trim_stop()
1112  * Convenience function to stop trimming of a vdev tree and set all trim
1151  * Conditionally restarts a manual TRIM given its on-disk state.
1198  * Used by the automatic TRIM when ZFS_DEBUG_TRIM is set to verify that
1199  * every TRIM range is contained within ms_allocatable.
1213  * Each automatic TRIM thread is responsible for managing the trimming of a
1214  * top-level vdev in the pool.  No automatic TRIM state is maintained on-disk.
1216  * N.B. This behavior is different from a manual TRIM where a thread
1260 		 *    TRIM commands for a group evenly over the entire device.  in vdev_autotrim_thread()
1278 			 * or when there are no recent frees to trim.  in vdev_autotrim_thread()
1289 			 * This may happen when a manual TRIM or initialize  in vdev_autotrim_thread()
1291 			 * of a manual TRIM, the ms_trim tree will have been  in vdev_autotrim_thread()
1292 			 * vacated.  Only ranges added after the manual TRIM  in vdev_autotrim_thread()
1294 			 * These will be processed when the automatic TRIM  in vdev_autotrim_thread()
1314 			 * trim trees for a metaslab.  If the top-level vdev  in vdev_autotrim_thread()
1317 			 * and a trim tree should be constructed for each.  in vdev_autotrim_thread()
1372 			 * Issue the TRIM I/Os for all ranges covered by the  in vdev_autotrim_thread()
1373 			 * TRIM trees.  These ranges are safe to TRIM because  in vdev_autotrim_thread()
1381 				 * Always yield to a manual TRIM if one has  in vdev_autotrim_thread()
1394 				 * of the required TRIM I/Os.  in vdev_autotrim_thread()
1420 			 * Wait for couples of kicks, to ensure the trim io is  in vdev_autotrim_thread()
1643 	 * must check to see if it needs to restart a trim. That thread  in vdev_trim_l2arc_thread()
1675  * Punches out TRIM threads for the L2ARC devices in a spa and assigns them
1686 	 * Locate the spa's l2arc devices and kick off TRIM threads.  in vdev_trim_l2arc()
1694 			 * Don't attempt TRIM if the vdev is UNAVAIL or if the  in vdev_trim_l2arc()
1695 			 * cache device was not marked for whole device TRIM  in vdev_trim_l2arc()
1779 	"Max size of TRIM commands, larger will be split");
1782 	"Min size of TRIM commands, smaller will be skipped");
1788 	"Min number of txgs to aggregate frees before issuing TRIM");