Import vdev ashift optimization from FreeBSD

Many modern devices use physical allocation units that are much
larger than the minimum logical allocation size accessible by
external commands. Two prevalent examples of this are 512e disk
drives (512b logical sector, 4K physical sector) and flash devices
(512b logical sector, 4K or larger allocation block size, and 128k
or larger erase block size). Operations that modify less than the
physical sector size result in a costly read-modify-write or garbage
collection sequence on these devices.

Simply exporting the true physical sector of the device to ZFS would
yield optimal performance, but has two serious drawbacks:

 1. Existing pools created with devices that have different logical
    and physical block sizes, but were configured to use the logical
    block size (e.g. because the OS version used for pool construction
    reported the logical block size instead of the physical block
    size) will suddenly find that the vdev allocation size has
    increased. This can be easily tolerated for active members of
    the array, but ZFS would prevent replacement of a vdev with
    another identical device because it now appears that the smaller
    allocation size required by the pool is not supported by the new
    device.

 2. The device's physical block size may be too large to be supported
    by ZFS. The optimal allocation size for the vdev may be quite
    large. For example, a RAID controller may export a vdev that
    requires read-modify-write cycles unless accessed using 64k
    aligned/sized requests. ZFS currently has an 8k minimum block
    size limit.

Reporting both the logical and physical allocation sizes for vdevs
solves these problems. A device may be used so long as the logical
block size is compatible with the configuration. By comparing the
logical and physical block sizes, new configurations can be optimized
and administrators can be notified of any existing pools that are
sub-optimal.

Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Co-authored-by: Matthew Macy <mmacy@freebsd.org>
Signed-off-by: Matt Macy <mmacy@FreeBSD.org>
Closes #10619

cmd/zpool/zpool_main.c

@@ -2082,7 +2082,7 @@ print_status_config(zpool_handle_t *zhp, status_cbdata_t *cb, const char *name,
     nvlist_t *nv, int depth, boolean_t isspare, vdev_rebuild_stat_t *vrs)
 {
 	nvlist_t **child, *root;
-	uint_t c, i, children;
+	uint_t c, i, vsc, children;
 	pool_scan_stat_t *ps = NULL;
 	vdev_stat_t *vs;
 	char rbuf[6], wbuf[6], cbuf[6];
@@ -2099,7 +2099,7 @@ print_status_config(zpool_handle_t *zhp, status_cbdata_t *cb, const char *name,
 		children = 0;
 
 	verify(nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
-	    (uint64_t **)&vs, &c) == 0);
+	    (uint64_t **)&vs, &vsc) == 0);
 
 	verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
 
@@ -2199,6 +2199,10 @@ print_status_config(zpool_handle_t *zhp, status_cbdata_t *cb, const char *name,
 			(void) printf(gettext("unsupported feature(s)"));
 			break;
 
+		case VDEV_AUX_ASHIFT_TOO_BIG:
+			(void) printf(gettext("unsupported minimum blocksize"));
+			break;
+
 		case VDEV_AUX_SPARED:
 			verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
 			    &spare_cb.cb_guid) == 0);
@@ -8105,6 +8109,15 @@ status_callback(zpool_handle_t *zhp, void *data)
 		    "'zpool clear'.\n"));
 		break;
 
+	case ZPOOL_STATUS_NON_NATIVE_ASHIFT:
+		(void) printf(gettext("status: One or more devices are "
+		    "configured to use a non-native block size.\n"
+		    "\tExpect reduced performance.\n"));
+		(void) printf(gettext("action: Replace affected devices with "
+		    "devices that support the\n\tconfigured block size, or "
+		    "migrate data to a properly configured\n\tpool.\n"));
+		break;
+
 	case ZPOOL_STATUS_HOSTID_MISMATCH:
 		printf_color(ANSI_BOLD, gettext("status: "));
 		printf_color(ANSI_YELLOW, gettext("Mismatch between pool hostid"

include/libzfs.h

@@ -390,6 +390,7 @@ typedef enum {
 	ZPOOL_STATUS_REMOVED_DEV,	/* removed device */
 	ZPOOL_STATUS_REBUILDING,	/* device being rebuilt */
 	ZPOOL_STATUS_REBUILD_SCRUB,	/* recommend scrubbing the pool */
+	ZPOOL_STATUS_NON_NATIVE_ASHIFT,	/* (e.g. 512e dev with ashift of 9) */
 
 	/*
 	 * Finally, the following indicates a healthy pool.

include/os/freebsd/spl/sys/mod_os.h

@@ -78,6 +78,12 @@
 #define	param_set_slop_shift_args(var) \
     CTLTYPE_INT, &var, 0, param_set_slop_shift, "I"
 
+#define	param_set_min_auto_ashift_args(var) \
+    CTLTYPE_U64, &var, 0, param_set_min_auto_ashift, "QU"
+
+#define	param_set_max_auto_ashift_args(var) \
+    CTLTYPE_U64, &var, 0, param_set_max_auto_ashift, "QU"
+
 #include <sys/kernel.h>
 #define	module_init(fn)							\
 static void \

include/sys/fs/zfs.h

@@ -873,6 +873,7 @@ typedef enum vdev_aux {
 	VDEV_AUX_EXTERNAL_PERSIST,	/* persistent forced fault	*/
 	VDEV_AUX_ACTIVE,	/* vdev active on a different host	*/
 	VDEV_AUX_CHILDREN_OFFLINE, /* all children are offline		*/
+	VDEV_AUX_ASHIFT_TOO_BIG, /* vdev's min block size is too large   */
 } vdev_aux_t;
 
 /*
@@ -1068,8 +1069,17 @@ typedef struct vdev_stat {
 	uint64_t	vs_trim_state;		/* vdev_trim_state_t */
 	uint64_t	vs_trim_action_time;	/* time_t */
 	uint64_t	vs_rebuild_processed;	/* bytes rebuilt */
+	uint64_t	vs_configured_ashift;   /* TLV vdev_ashift */
+	uint64_t	vs_logical_ashift;	/* vdev_logical_ashift  */
+	uint64_t	vs_physical_ashift;	/* vdev_physical_ashift */
 } vdev_stat_t;
 
+/* BEGIN CSTYLED */
+#define	VDEV_STAT_VALID(field, uint64_t_field_count) \
+    ((uint64_t_field_count * sizeof (uint64_t)) >=	 \
+     (offsetof(vdev_stat_t, field) + sizeof (((vdev_stat_t *)NULL)->field)))
+/* END CSTYLED */
+
 /*
  * Extended stats
  *

include/sys/vdev.h

@@ -94,6 +94,7 @@ extern void vdev_rele(vdev_t *);
 extern int vdev_metaslab_init(vdev_t *vd, uint64_t txg);
 extern void vdev_metaslab_fini(vdev_t *vd);
 extern void vdev_metaslab_set_size(vdev_t *);
+extern void vdev_ashift_optimize(vdev_t *);
 extern void vdev_expand(vdev_t *vd, uint64_t txg);
 extern void vdev_split(vdev_t *vd);
 extern void vdev_deadman(vdev_t *vd, char *tag);

include/sys/vdev_impl.h

@@ -69,7 +69,7 @@ extern uint32_t zfs_vdev_async_write_max_active;
  * Virtual device operations
  */
 typedef int	vdev_open_func_t(vdev_t *vd, uint64_t *size, uint64_t *max_size,
-    uint64_t *ashift);
+    uint64_t *ashift, uint64_t *pshift);
 typedef void	vdev_close_func_t(vdev_t *vd);
 typedef uint64_t vdev_asize_func_t(vdev_t *vd, uint64_t psize);
 typedef void	vdev_io_start_func_t(zio_t *zio);
@@ -216,6 +216,25 @@ struct vdev {
 	uint64_t	vdev_min_asize;	/* min acceptable asize		*/
 	uint64_t	vdev_max_asize;	/* max acceptable asize		*/
 	uint64_t	vdev_ashift;	/* block alignment shift	*/
+
+	/*
+	 * Logical block alignment shift
+	 *
+	 * The smallest sized/aligned I/O supported by the device.
+	 */
+	uint64_t	vdev_logical_ashift;
+	/*
+	 * Physical block alignment shift
+	 *
+	 * The device supports logical I/Os with vdev_logical_ashift
+	 * size/alignment, but optimum performance will be achieved by
+	 * aligning/sizing requests to vdev_physical_ashift.  Smaller
+	 * requests may be inflated or incur device level read-modify-write
+	 * operations.
+	 *
+	 * May be 0 to indicate no preference (i.e. use vdev_logical_ashift).
+	 */
+	uint64_t	vdev_physical_ashift;
 	uint64_t	vdev_state;	/* see VDEV_STATE_* #defines	*/
 	uint64_t	vdev_prevstate;	/* used when reopening a vdev	*/
 	vdev_ops_t	*vdev_ops;	/* vdev operations		*/
@@ -586,6 +605,14 @@ extern int vdev_obsolete_counts_are_precise(vdev_t *vd, boolean_t *are_precise);
  */
 int vdev_checkpoint_sm_object(vdev_t *vd, uint64_t *sm_obj);
 
+/*
+ * Vdev ashift optimization tunables
+ */
+extern uint64_t zfs_vdev_min_auto_ashift;
+extern uint64_t zfs_vdev_max_auto_ashift;
+int param_set_min_auto_ashift(ZFS_MODULE_PARAM_ARGS);
+int param_set_max_auto_ashift(ZFS_MODULE_PARAM_ARGS);
+
 #ifdef	__cplusplus
 }
 #endif

lib/libzfs/libzfs_status.c

@@ -43,6 +43,7 @@
 
 #include <libzfs.h>
 #include <libzutil.h>
+#include <stdlib.h>
 #include <string.h>
 #include <unistd.h>
 #include <sys/systeminfo.h>
@@ -94,57 +95,69 @@ static char *zfs_msgid_table[] = {
 
 /* ARGSUSED */
 static int
-vdev_missing(uint64_t state, uint64_t aux, uint64_t errs)
+vdev_missing(vdev_stat_t *vs, uint_t vsc)
 {
-	return (state == VDEV_STATE_CANT_OPEN &&
+	return (vs->vs_state == VDEV_STATE_CANT_OPEN &&
-	    aux == VDEV_AUX_OPEN_FAILED);
+	    vs->vs_aux == VDEV_AUX_OPEN_FAILED);
 }
 
 /* ARGSUSED */
 static int
-vdev_faulted(uint64_t state, uint64_t aux, uint64_t errs)
+vdev_faulted(vdev_stat_t *vs, uint_t vsc)
 {
-	return (state == VDEV_STATE_FAULTED);
+	return (vs->vs_state == VDEV_STATE_FAULTED);
 }
 
 /* ARGSUSED */
 static int
-vdev_errors(uint64_t state, uint64_t aux, uint64_t errs)
+vdev_errors(vdev_stat_t *vs, uint_t vsc)
 {
-	return (state == VDEV_STATE_DEGRADED || errs != 0);
+	return (vs->vs_state == VDEV_STATE_DEGRADED ||
+	    vs->vs_read_errors != 0 || vs->vs_write_errors != 0 ||
+	    vs->vs_checksum_errors != 0);
 }
 
 /* ARGSUSED */
 static int
-vdev_broken(uint64_t state, uint64_t aux, uint64_t errs)
+vdev_broken(vdev_stat_t *vs, uint_t vsc)
 {
-	return (state == VDEV_STATE_CANT_OPEN);
+	return (vs->vs_state == VDEV_STATE_CANT_OPEN);
 }
 
 /* ARGSUSED */
 static int
-vdev_offlined(uint64_t state, uint64_t aux, uint64_t errs)
+vdev_offlined(vdev_stat_t *vs, uint_t vsc)
 {
-	return (state == VDEV_STATE_OFFLINE);
+	return (vs->vs_state == VDEV_STATE_OFFLINE);
 }
 
 /* ARGSUSED */
 static int
-vdev_removed(uint64_t state, uint64_t aux, uint64_t errs)
+vdev_removed(vdev_stat_t *vs, uint_t vsc)
 {
-	return (state == VDEV_STATE_REMOVED);
+	return (vs->vs_state == VDEV_STATE_REMOVED);
+}
+
+static int
+vdev_non_native_ashift(vdev_stat_t *vs, uint_t vsc)
+{
+	if (getenv("ZPOOL_STATUS_NON_NATIVE_ASHIFT_IGNORE") != NULL)
+		return (0);
+
+	return (VDEV_STAT_VALID(vs_physical_ashift, vsc) &&
+	    vs->vs_configured_ashift < vs->vs_physical_ashift);
 }
 
 /*
  * Detect if any leaf devices that have seen errors or could not be opened.
  */
 static boolean_t
-find_vdev_problem(nvlist_t *vdev, int (*func)(uint64_t, uint64_t, uint64_t))
+find_vdev_problem(nvlist_t *vdev, int (*func)(vdev_stat_t *, uint_t),
+    boolean_t ignore_replacing)
 {
 	nvlist_t **child;
 	vdev_stat_t *vs;
-	uint_t c, children;
+	uint_t c, vsc, children;
-	char *type;
 
 	/*
 	 * Ignore problems within a 'replacing' vdev, since we're presumably in
@@ -152,23 +165,25 @@ find_vdev_problem(nvlist_t *vdev, int (*func)(uint64_t, uint64_t, uint64_t))
 	 * out again.  We'll pick up the fact that a resilver is happening
 	 * later.
 	 */
-	verify(nvlist_lookup_string(vdev, ZPOOL_CONFIG_TYPE, &type) == 0);
+	if (ignore_replacing == B_TRUE) {
-	if (strcmp(type, VDEV_TYPE_REPLACING) == 0)
+		char *type;
-		return (B_FALSE);
+
+		verify(nvlist_lookup_string(vdev, ZPOOL_CONFIG_TYPE,
+		    &type) == 0);
+		if (strcmp(type, VDEV_TYPE_REPLACING) == 0)
+			return (B_FALSE);
+	}
 
 	if (nvlist_lookup_nvlist_array(vdev, ZPOOL_CONFIG_CHILDREN, &child,
 	    &children) == 0) {
 		for (c = 0; c < children; c++)
-			if (find_vdev_problem(child[c], func))
+			if (find_vdev_problem(child[c], func, ignore_replacing))
 				return (B_TRUE);
 	} else {
 		verify(nvlist_lookup_uint64_array(vdev, ZPOOL_CONFIG_VDEV_STATS,
-		    (uint64_t **)&vs, &c) == 0);
+		    (uint64_t **)&vs, &vsc) == 0);
 
-		if (func(vs->vs_state, vs->vs_aux,
+		if (func(vs, vsc) != 0)
-		    vs->vs_read_errors +
-		    vs->vs_write_errors +
-		    vs->vs_checksum_errors))
 			return (B_TRUE);
 	}
 
@@ -178,7 +193,7 @@ find_vdev_problem(nvlist_t *vdev, int (*func)(uint64_t, uint64_t, uint64_t))
 	if (nvlist_lookup_nvlist_array(vdev, ZPOOL_CONFIG_L2CACHE, &child,
 	    &children) == 0) {
 		for (c = 0; c < children; c++)
-			if (find_vdev_problem(child[c], func))
+			if (find_vdev_problem(child[c], func, ignore_replacing))
 				return (B_TRUE);
 	}
 
@@ -362,15 +377,15 @@ check_status(nvlist_t *config, boolean_t isimport, zpool_errata_t *erratap)
 	 * Bad devices in non-replicated config.
 	 */
 	if (vs->vs_state == VDEV_STATE_CANT_OPEN &&
-	    find_vdev_problem(nvroot, vdev_faulted))
+	    find_vdev_problem(nvroot, vdev_faulted, B_TRUE))
 		return (ZPOOL_STATUS_FAULTED_DEV_NR);
 
 	if (vs->vs_state == VDEV_STATE_CANT_OPEN &&
-	    find_vdev_problem(nvroot, vdev_missing))
+	    find_vdev_problem(nvroot, vdev_missing, B_TRUE))
 		return (ZPOOL_STATUS_MISSING_DEV_NR);
 
 	if (vs->vs_state == VDEV_STATE_CANT_OPEN &&
-	    find_vdev_problem(nvroot, vdev_broken))
+	    find_vdev_problem(nvroot, vdev_broken, B_TRUE))
 		return (ZPOOL_STATUS_CORRUPT_LABEL_NR);
 
 	/*
@@ -392,31 +407,37 @@ check_status(nvlist_t *config, boolean_t isimport, zpool_errata_t *erratap)
 	/*
 	 * Missing devices in a replicated config.
 	 */
-	if (find_vdev_problem(nvroot, vdev_faulted))
+	if (find_vdev_problem(nvroot, vdev_faulted, B_TRUE))
 		return (ZPOOL_STATUS_FAULTED_DEV_R);
-	if (find_vdev_problem(nvroot, vdev_missing))
+	if (find_vdev_problem(nvroot, vdev_missing, B_TRUE))
 		return (ZPOOL_STATUS_MISSING_DEV_R);
-	if (find_vdev_problem(nvroot, vdev_broken))
+	if (find_vdev_problem(nvroot, vdev_broken, B_TRUE))
 		return (ZPOOL_STATUS_CORRUPT_LABEL_R);
 
 	/*
 	 * Devices with errors
 	 */
-	if (!isimport && find_vdev_problem(nvroot, vdev_errors))
+	if (!isimport && find_vdev_problem(nvroot, vdev_errors, B_TRUE))
 		return (ZPOOL_STATUS_FAILING_DEV);
 
 	/*
 	 * Offlined devices
 	 */
-	if (find_vdev_problem(nvroot, vdev_offlined))
+	if (find_vdev_problem(nvroot, vdev_offlined, B_TRUE))
 		return (ZPOOL_STATUS_OFFLINE_DEV);
 
 	/*
 	 * Removed device
 	 */
-	if (find_vdev_problem(nvroot, vdev_removed))
+	if (find_vdev_problem(nvroot, vdev_removed, B_TRUE))
 		return (ZPOOL_STATUS_REMOVED_DEV);
 
+	/*
+	 * Suboptimal, but usable, ashift configuration.
+	 */
+	if (find_vdev_problem(nvroot, vdev_non_native_ashift, B_FALSE))
+		return (ZPOOL_STATUS_NON_NATIVE_ASHIFT);
+
 	/*
 	 * Informational errata available.
 	 */

man/man5/zfs-module-parameters.5

@@ -486,6 +486,29 @@ Default limit for metaslab size.
 Default value: \fB29\fR [meaning (1 << 29) = 512MB].
 .RE
 
+.sp
+.ne 2
+.na
+\fBzfs_vdev_max_auto_ashift\fR (ulong)
+.ad
+.RS 12n
+Maximum ashift used when optimizing for logical -> physical sector size on new
+top-level vdevs.
+.sp
+Default value: \fBASHIFT_MAX\fR (16).
+.RE
+
+.sp
+.ne 2
+.na
+\fBzfs_vdev_min_auto_ashift\fR (ulong)
+.ad
+.RS 12n
+Minimum ashift used when creating new top-level vdevs.
+.sp
+Default value: \fBASHIFT_MIN\fR (9).
+.RE
+
 .sp
 .ne 2
 .na

man/man8/zpool.8

@@ -470,6 +470,12 @@ The maximum time in milliseconds that
 .Nm zpool import
 will wait for an expected device to be available.
 .El
+.Bl -tag -width "ZPOOL_STATUS_NON_NATIVE_ASHIFT_IGNORE"
+.It Ev ZPOOL_STATUS_NON_NATIVE_ASHIFT_IGNORE
+If set, suppress warning about non-native vdev ashift in
+.Nm zpool status .
+The value is not used, only the presence or absence of the variable matters.
+.El
 .Bl -tag -width "ZPOOL_VDEV_NAME_GUID"
 .It Ev ZPOOL_VDEV_NAME_GUID
 Cause

module/os/freebsd/zfs/sysctl_os.c

@@ -52,6 +52,7 @@ __FBSDID("$FreeBSD$");
 #include <sys/spa.h>
 #include <sys/spa_impl.h>
 #include <sys/vdev.h>
+#include <sys/vdev_impl.h>
 #include <sys/dmu.h>
 #include <sys/dsl_dir.h>
 #include <sys/dsl_dataset.h>
@@ -518,56 +519,53 @@ SYSCTL_INT(_vfs_zfs, OID_AUTO, space_map_ibs, CTLFLAG_RWTUN,
 
 
 /* vdev.c */
-#ifdef notyet
+int
-extern uint64_t zfs_max_auto_ashift;
+param_set_min_auto_ashift(SYSCTL_HANDLER_ARGS)
-extern uint64_t zfs_min_auto_ashift;
-
-static int
-sysctl_vfs_zfs_max_auto_ashift(SYSCTL_HANDLER_ARGS)
 {
 	uint64_t val;
 	int err;
 
-	val = zfs_max_auto_ashift;
+	val = zfs_vdev_min_auto_ashift;
 	err = sysctl_handle_64(oidp, &val, 0, req);
 	if (err != 0 || req->newptr == NULL)
-		return (err);
+		return (SET_ERROR(err));
 
-	if (val > ASHIFT_MAX || val < zfs_min_auto_ashift)
+	if (val < ASHIFT_MIN || val > zfs_vdev_max_auto_ashift)
-		return (EINVAL);
+		return (SET_ERROR(EINVAL));
 
-	zfs_max_auto_ashift = val;
+	zfs_vdev_min_auto_ashift = val;
 
 	return (0);
 }
-SYSCTL_PROC(_vfs_zfs, OID_AUTO, max_auto_ashift,
+
-    CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof (uint64_t),
+int
-    sysctl_vfs_zfs_max_auto_ashift, "QU",
+param_set_max_auto_ashift(SYSCTL_HANDLER_ARGS)
-    "Max ashift used when optimising for logical -> physical sectors size on "
-    "new top-level vdevs.");
-static int
-sysctl_vfs_zfs_min_auto_ashift(SYSCTL_HANDLER_ARGS)
 {
 	uint64_t val;
 	int err;
 
-	val = zfs_min_auto_ashift;
+	val = zfs_vdev_max_auto_ashift;
 	err = sysctl_handle_64(oidp, &val, 0, req);
 	if (err != 0 || req->newptr == NULL)
-		return (err);
+		return (SET_ERROR(err));
 
-	if (val < ASHIFT_MIN || val > zfs_max_auto_ashift)
+	if (val > ASHIFT_MAX || val < zfs_vdev_min_auto_ashift)
-		return (EINVAL);
+		return (SET_ERROR(EINVAL));
 
-	zfs_min_auto_ashift = val;
+	zfs_vdev_max_auto_ashift = val;
 
 	return (0);
 }
-SYSCTL_PROC(_vfs_zfs, OID_AUTO, min_auto_ashift,
+
-    CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof (uint64_t),
+SYSCTL_PROC(_vfs_zfs, OID_AUTO, min_auto_ashift, CTLTYPE_U64 | CTLFLAG_RWTUN,
-    sysctl_vfs_zfs_min_auto_ashift, "QU",
+    &zfs_vdev_min_auto_ashift, sizeof (zfs_vdev_min_auto_ashift),
-    "Min ashift used when creating new top-level vdevs.");
+    param_set_min_auto_ashift, "QU",
-#endif
+    "Min ashift used when creating new top-level vdev. (LEGACY)");
+SYSCTL_PROC(_vfs_zfs, OID_AUTO, max_auto_ashift, CTLTYPE_U64 | CTLFLAG_RWTUN,
+    &zfs_vdev_max_auto_ashift, sizeof (zfs_vdev_max_auto_ashift),
+    param_set_max_auto_ashift, "QU",
+    "Max ashift used when optimizing for logical -> physical sector size on "
+    "new top-level vdevs. (LEGACY)");
 
 /*
  * Since the DTL space map of a vdev is not expected to have a lot of

module/os/freebsd/zfs/vdev_file.c

@@ -83,7 +83,7 @@ vdev_file_open_mode(spa_mode_t spa_mode)
 
 static int
 vdev_file_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
-    uint64_t *ashift)
+    uint64_t *logical_ashift, uint64_t *physical_ashift)
 {
 	vdev_file_t *vf;
 	zfs_file_t *fp;
@@ -167,7 +167,8 @@ skip_open:
 	}
 
 	*max_psize = *psize = zfa.zfa_size;
-	*ashift = SPA_MINBLOCKSHIFT;
+	*logical_ashift = SPA_MINBLOCKSHIFT;
+	*physical_ashift = SPA_MINBLOCKSHIFT;
 
 	return (0);
 }

module/os/freebsd/zfs/vdev_geom.c

@@ -802,7 +802,7 @@ vdev_geom_open_by_path(vdev_t *vd, int check_guid)
 
 static int
 vdev_geom_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
-    uint64_t *logical_ashift)
+    uint64_t *logical_ashift, uint64_t *physical_ashift)
 {
 	struct g_provider *pp;
 	struct g_consumer *cp;
@@ -949,11 +949,12 @@ skip_open:
 	 * transfer size.
 	 */
 	*logical_ashift = highbit(MAX(pp->sectorsize, SPA_MINBLOCKSIZE)) - 1;
-#ifdef notyet
+	*physical_ashift = 0;
-	if (pp->stripesize > (1 << *logical_ashift) && ISP2(pp->stripesize) &&
+	if (pp->stripesize && pp->stripesize > (1 << *logical_ashift) &&
-	    pp->stripesize <= (1 << ASHIFT_MAX) && pp->stripeoffset == 0)
+	    ISP2(pp->stripesize) && pp->stripesize <= (1 << ASHIFT_MAX) &&
+	    pp->stripeoffset == 0)
 		*physical_ashift = highbit(pp->stripesize) - 1;
-#endif
+
 	/*
 	 * Clear the nowritecache settings, so that on a vdev_reopen()
 	 * we will try again.

module/os/linux/zfs/vdev_disk.c

@@ -159,7 +159,7 @@ vdev_disk_error(zio_t *zio)
 
 static int
 vdev_disk_open(vdev_t *v, uint64_t *psize, uint64_t *max_psize,
-    uint64_t *ashift)
+    uint64_t *logical_ashift, uint64_t *physical_ashift)
 {
 	struct block_device *bdev;
 	fmode_t mode = vdev_bdev_mode(spa_mode(v->vdev_spa));
@@ -270,7 +270,10 @@ vdev_disk_open(vdev_t *v, uint64_t *psize, uint64_t *max_psize,
 	struct request_queue *q = bdev_get_queue(vd->vd_bdev);
 
 	/*  Determine the physical block size */
-	int block_size = bdev_physical_block_size(vd->vd_bdev);
+	int physical_block_size = bdev_physical_block_size(vd->vd_bdev);
+
+	/*  Determine the logical block size */
+	int logical_block_size = bdev_logical_block_size(vd->vd_bdev);
 
 	/* Clear the nowritecache bit, causes vdev_reopen() to try again. */
 	v->vdev_nowritecache = B_FALSE;
@@ -291,7 +294,11 @@ vdev_disk_open(vdev_t *v, uint64_t *psize, uint64_t *max_psize,
 	*max_psize = bdev_max_capacity(vd->vd_bdev, v->vdev_wholedisk);
 
 	/* Based on the minimum sector size set the block size */
-	*ashift = highbit64(MAX(block_size, SPA_MINBLOCKSIZE)) - 1;
+	*physical_ashift = highbit64(MAX(physical_block_size,
+	    SPA_MINBLOCKSIZE)) - 1;
+
+	*logical_ashift = highbit64(MAX(logical_block_size,
+	    SPA_MINBLOCKSIZE)) - 1;
 
 	return (0);
 }
@@ -824,3 +831,43 @@ char *zfs_vdev_scheduler = "unused";
 module_param_call(zfs_vdev_scheduler, param_set_vdev_scheduler,
     param_get_charp, &zfs_vdev_scheduler, 0644);
 MODULE_PARM_DESC(zfs_vdev_scheduler, "I/O scheduler");
+
+int
+param_set_min_auto_ashift(const char *buf, zfs_kernel_param_t *kp)
+{
+	uint64_t val;
+	int error;
+
+	error = kstrtoull(buf, 0, &val);
+	if (error < 0)
+		return (SET_ERROR(error));
+
+	if (val < ASHIFT_MIN || val > zfs_vdev_max_auto_ashift)
+		return (SET_ERROR(-EINVAL));
+
+	error = param_set_ulong(buf, kp);
+	if (error < 0)
+		return (SET_ERROR(error));
+
+	return (0);
+}
+
+int
+param_set_max_auto_ashift(const char *buf, zfs_kernel_param_t *kp)
+{
+	uint64_t val;
+	int error;
+
+	error = kstrtoull(buf, 0, &val);
+	if (error < 0)
+		return (SET_ERROR(error));
+
+	if (val > ASHIFT_MAX || val < zfs_vdev_min_auto_ashift)
+		return (SET_ERROR(-EINVAL));
+
+	error = param_set_ulong(buf, kp);
+	if (error < 0)
+		return (SET_ERROR(error));
+
+	return (0);
+}

module/os/linux/zfs/vdev_file.c

@@ -75,7 +75,7 @@ vdev_file_open_mode(spa_mode_t spa_mode)
 
 static int
 vdev_file_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
-    uint64_t *ashift)
+    uint64_t *logical_ashift, uint64_t *physical_ashift)
 {
 	vdev_file_t *vf;
 	zfs_file_t *fp;
@@ -159,7 +159,8 @@ skip_open:
 	}
 
 	*max_psize = *psize = zfa.zfa_size;
-	*ashift = SPA_MINBLOCKSHIFT;
+	*logical_ashift = SPA_MINBLOCKSHIFT;
+	*physical_ashift = SPA_MINBLOCKSHIFT;
 
 	return (0);
 }

module/zfs/arc.c

@@ -9283,6 +9283,8 @@ l2arc_add_vdev(spa_t *spa, vdev_t *vd)
 
 	ASSERT(!l2arc_vdev_present(vd));
 
+	vdev_ashift_optimize(vd);
+
 	/*
 	 * Create a new l2arc device entry.
 	 */

module/zfs/spa.c

@@ -5747,6 +5747,7 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
 		for (int c = 0; error == 0 && c < rvd->vdev_children; c++) {
 			vdev_t *vd = rvd->vdev_child[c];
 
+			vdev_ashift_optimize(vd);
 			vdev_metaslab_set_size(vd);
 			vdev_expand(vd, txg);
 		}

module/zfs/spa_config.c

@@ -576,8 +576,10 @@ spa_config_update(spa_t *spa, int what)
 			    (tvd->vdev_islog && tvd->vdev_removing))
 				continue;
 
-			if (tvd->vdev_ms_array == 0)
+			if (tvd->vdev_ms_array == 0) {
+				vdev_ashift_optimize(tvd);
 				vdev_metaslab_set_size(tvd);
+			}
 			vdev_expand(tvd, txg);
 		}
 	}

module/zfs/vdev.c

@@ -110,6 +110,9 @@ int zfs_vdev_standard_sm_blksz = (1 << 17);
  */
 int zfs_nocacheflush = 0;
 
+uint64_t zfs_vdev_max_auto_ashift = ASHIFT_MAX;
+uint64_t zfs_vdev_min_auto_ashift = ASHIFT_MIN;
+
 /*PRINTFLIKE2*/
 void
 vdev_dbgmsg(vdev_t *vd, const char *fmt, ...)
@@ -1176,6 +1179,8 @@ vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
 	mvd->vdev_max_asize = cvd->vdev_max_asize;
 	mvd->vdev_psize = cvd->vdev_psize;
 	mvd->vdev_ashift = cvd->vdev_ashift;
+	mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
+	mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
 	mvd->vdev_state = cvd->vdev_state;
 	mvd->vdev_crtxg = cvd->vdev_crtxg;
 
@@ -1207,7 +1212,8 @@ vdev_remove_parent(vdev_t *cvd)
 	    mvd->vdev_ops == &vdev_replacing_ops ||
 	    mvd->vdev_ops == &vdev_spare_ops);
 	cvd->vdev_ashift = mvd->vdev_ashift;
-
+	cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
+	cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
 	vdev_remove_child(mvd, cvd);
 	vdev_remove_child(pvd, mvd);
 
@@ -1677,7 +1683,8 @@ vdev_open(vdev_t *vd)
 	uint64_t osize = 0;
 	uint64_t max_osize = 0;
 	uint64_t asize, max_asize, psize;
-	uint64_t ashift = 0;
+	uint64_t logical_ashift = 0;
+	uint64_t physical_ashift = 0;
 
 	ASSERT(vd->vdev_open_thread == curthread ||
 	    spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
@@ -1707,8 +1714,8 @@ vdev_open(vdev_t *vd)
 		return (SET_ERROR(ENXIO));
 	}
 
-	error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift);
+	error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
-
+	    &logical_ashift, &physical_ashift);
 	/*
 	 * Physical volume size should never be larger than its max size, unless
 	 * the disk has shrunk while we were reading it or the device is buggy
@@ -1823,6 +1830,17 @@ vdev_open(vdev_t *vd)
 		return (SET_ERROR(EINVAL));
 	}
 
+	vd->vdev_physical_ashift =
+	    MAX(physical_ashift, vd->vdev_physical_ashift);
+	vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift);
+	vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift);
+
+	if (vd->vdev_logical_ashift > ASHIFT_MAX) {
+		vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
+		    VDEV_AUX_ASHIFT_TOO_BIG);
+		return (SET_ERROR(EDOM));
+	}
+
 	if (vd->vdev_asize == 0) {
 		/*
 		 * This is the first-ever open, so use the computed values.
@@ -1830,9 +1848,6 @@ vdev_open(vdev_t *vd)
 		 */
 		vd->vdev_asize = asize;
 		vd->vdev_max_asize = max_asize;
-		if (vd->vdev_ashift == 0) {
-			vd->vdev_ashift = ashift; /* use detected value */
-		}
 		if (vd->vdev_ashift != 0 && (vd->vdev_ashift < ASHIFT_MIN ||
 		    vd->vdev_ashift > ASHIFT_MAX)) {
 			vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
@@ -1841,16 +1856,17 @@ vdev_open(vdev_t *vd)
 		}
 	} else {
 		/*
-		 * Detect if the alignment requirement has increased.
+		 * Make sure the alignment required hasn't increased.
-		 * We don't want to make the pool unavailable, just
-		 * post an event instead.
 		 */
-		if (ashift > vd->vdev_top->vdev_ashift &&
+		if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
 		    vd->vdev_ops->vdev_op_leaf) {
 			zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT,
 			    spa, vd, NULL, NULL, 0, 0);
-		}
+			vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
+			    VDEV_AUX_BAD_LABEL);
+			return (SET_ERROR(EDOM));
 
+		}
 		vd->vdev_max_asize = max_asize;
 	}
 
@@ -2428,6 +2444,35 @@ vdev_metaslab_set_size(vdev_t *vd)
 	ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT);
 }
 
+/*
+ * Maximize performance by inflating the configured ashift for top level
+ * vdevs to be as close to the physical ashift as possible while maintaining
+ * administrator defined limits and ensuring it doesn't go below the
+ * logical ashift.
+ */
+void
+vdev_ashift_optimize(vdev_t *vd)
+{
+	if (vd == vd->vdev_top) {
+		if (vd->vdev_ashift < vd->vdev_physical_ashift) {
+			vd->vdev_ashift = MIN(
+			    MAX(zfs_vdev_max_auto_ashift, vd->vdev_ashift),
+			    MAX(zfs_vdev_min_auto_ashift,
+			    vd->vdev_physical_ashift));
+		} else {
+			/*
+			 * Unusual case where logical ashift > physical ashift
+			 * so we can't cap the calculated ashift based on max
+			 * ashift as that would cause failures.
+			 * We still check if we need to increase it to match
+			 * the min ashift.
+			 */
+			vd->vdev_ashift = MAX(zfs_vdev_min_auto_ashift,
+			    vd->vdev_ashift);
+		}
+	}
+}
+
 void
 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
 {
@@ -4083,6 +4128,11 @@ vdev_get_stats_ex(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx)
 			    1ULL << tvd->vdev_ms_shift);
 		}
 
+		vs->vs_configured_ashift = vd->vdev_top != NULL
+		    ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
+		vs->vs_logical_ashift = vd->vdev_logical_ashift;
+		vs->vs_physical_ashift = vd->vdev_physical_ashift;
+
 		/*
 		 * Report fragmentation and rebuild progress for top-level,
 		 * non-auxiliary, concrete devices.
@@ -5028,4 +5078,13 @@ ZFS_MODULE_PARAM(zfs_vdev, vdev_, validate_skip, INT, ZMOD_RW,
 
 ZFS_MODULE_PARAM(zfs, zfs_, nocacheflush, INT, ZMOD_RW,
 	"Disable cache flushes");
+
+ZFS_MODULE_PARAM_CALL(zfs_vdev, zfs_vdev_, min_auto_ashift,
+	param_set_min_auto_ashift, param_get_ulong, ZMOD_RW,
+	"Minimum ashift used when creating new top-level vdevs");
+
+ZFS_MODULE_PARAM_CALL(zfs_vdev, zfs_vdev_, max_auto_ashift,
+	param_set_max_auto_ashift, param_get_ulong, ZMOD_RW,
+	"Maximum ashift used when optimizing for logical -> physical sector "
+	"size on new top-level vdevs");
 /* END CSTYLED */

module/zfs/vdev_indirect.c

@@ -950,11 +950,12 @@ vdev_indirect_close(vdev_t *vd)
 /* ARGSUSED */
 static int
 vdev_indirect_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
-    uint64_t *ashift)
+    uint64_t *logical_ashift, uint64_t *physical_ashift)
 {
 	*psize = *max_psize = vd->vdev_asize +
 	    VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
-	*ashift = vd->vdev_ashift;
+	*logical_ashift = vd->vdev_ashift;
+	*physical_ashift = vd->vdev_physical_ashift;
 	return (0);
 }
 

module/zfs/vdev_mirror.c

@@ -366,7 +366,7 @@ vdev_mirror_map_init(zio_t *zio)
 
 static int
 vdev_mirror_open(vdev_t *vd, uint64_t *asize, uint64_t *max_asize,
-    uint64_t *ashift)
+    uint64_t *logical_ashift, uint64_t *physical_ashift)
 {
 	int numerrors = 0;
 	int lasterror = 0;
@@ -389,7 +389,9 @@ vdev_mirror_open(vdev_t *vd, uint64_t *asize, uint64_t *max_asize,
 
 		*asize = MIN(*asize - 1, cvd->vdev_asize - 1) + 1;
 		*max_asize = MIN(*max_asize - 1, cvd->vdev_max_asize - 1) + 1;
-		*ashift = MAX(*ashift, cvd->vdev_ashift);
+		*logical_ashift = MAX(*logical_ashift, cvd->vdev_ashift);
+		*physical_ashift = MAX(*physical_ashift,
+		    vd->vdev_physical_ashift);
 	}
 
 	if (numerrors == vd->vdev_children) {

module/zfs/vdev_missing.c

@@ -45,7 +45,7 @@
 /* ARGSUSED */
 static int
 vdev_missing_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
-    uint64_t *ashift)
+    uint64_t *ashift, uint64_t *pshift)
 {
 	/*
 	 * Really this should just fail.  But then the root vdev will be in the
@@ -56,6 +56,7 @@ vdev_missing_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
 	*psize = 0;
 	*max_psize = 0;
 	*ashift = 0;
+	*pshift = 0;
 	return (0);
 }
 

module/zfs/vdev_raidz.c

@@ -1554,7 +1554,7 @@ vdev_raidz_reconstruct(raidz_map_t *rm, const int *t, int nt)
 
 static int
 vdev_raidz_open(vdev_t *vd, uint64_t *asize, uint64_t *max_asize,
-    uint64_t *ashift)
+    uint64_t *logical_ashift, uint64_t *physical_ashift)
 {
 	vdev_t *cvd;
 	uint64_t nparity = vd->vdev_nparity;
@@ -1583,7 +1583,9 @@ vdev_raidz_open(vdev_t *vd, uint64_t *asize, uint64_t *max_asize,
 
 		*asize = MIN(*asize - 1, cvd->vdev_asize - 1) + 1;
 		*max_asize = MIN(*max_asize - 1, cvd->vdev_max_asize - 1) + 1;
-		*ashift = MAX(*ashift, cvd->vdev_ashift);
+		*logical_ashift = MAX(*logical_ashift, cvd->vdev_ashift);
+		*physical_ashift = MAX(*physical_ashift,
+		    cvd->vdev_physical_ashift);
 	}
 
 	*asize *= vd->vdev_children;

module/zfs/vdev_root.c

@@ -82,7 +82,7 @@ too_many_errors(vdev_t *vd, uint64_t numerrors)
 
 static int
 vdev_root_open(vdev_t *vd, uint64_t *asize, uint64_t *max_asize,
-    uint64_t *ashift)
+    uint64_t *ashift, uint64_t *pshift)
 {
 	spa_t *spa = vd->vdev_spa;
 	int lasterror = 0;
@@ -116,6 +116,7 @@ vdev_root_open(vdev_t *vd, uint64_t *asize, uint64_t *max_asize,
 	*asize = 0;
 	*max_asize = 0;
 	*ashift = 0;
+	*pshift = 0;
 
 	return (0);
 }