Illumos #2671: zpool import should not fail if vdev ashift has increased
authorGeorge Wilson <gwilson@delphix.com>
Thu, 3 May 2012 12:49:19 +0000 (05:49 -0700)
committerBrian Behlendorf <behlendorf1@llnl.gov>
Thu, 15 Nov 2012 19:05:59 +0000 (11:05 -0800)
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Eric Schrock <eric.schrock@delphix.com>
Reviewed by: Richard Elling <richard.elling@richardelling.com>
Reviewed by: Gordon Ross <gwr@nexenta.com>
Reviewed by: Garrett D'Amore <garrett@damore.org>
Approved by: Richard Lowe <richlowe@richlowe.net>

Refererces to Illumos issue:
      https://www.illumos.org/issues/2671

This patch has been slightly modified from the upstream Illumos
version.  In the upstream implementation a warning message is
logged to the console.  To prevent pointless console noise this
notification is now posted as a "ereport.fs.zfs.vdev.bad_ashift"
event.

The event indicates a non-optimial (but entirely safe) ashift
value was used to create the pool.  Depending on your workload
this may impact pool performance.  Unfortunately, the only way
to correct the issue is to recreate the pool with a new ashift.

NOTE: The unrelated fix to the comment in zpool_main.c appears
in the upstream commit and was preserved for consistnecy.

Ported-by: Cyril Plisko <cyril.plisko@mountall.com>
Reworked-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #955

cmd/zpool/zpool_main.c
include/sys/fm/fs/zfs.h
module/zfs/vdev.c
module/zfs/zfs_fm.c

index 5661732..96798c4 100644 (file)
@@ -3736,7 +3736,7 @@ print_dedup_stats(nvlist_t *config)
 
        /*
         * If the pool was faulted then we may not have been able to
-        * obtain the config. Otherwise, if have anything in the dedup
+        * obtain the config. Otherwise, if we have anything in the dedup
         * table continue processing the stats.
         */
        if (nvlist_lookup_uint64_array(config, ZPOOL_CONFIG_DDT_OBJ_STATS,
index 15803c0..d5c6004 100644 (file)
@@ -47,6 +47,7 @@ extern "C" {
 #define        FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM      "vdev.bad_guid_sum"
 #define        FM_EREPORT_ZFS_DEVICE_TOO_SMALL         "vdev.too_small"
 #define        FM_EREPORT_ZFS_DEVICE_BAD_LABEL         "vdev.bad_label"
+#define        FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT        "vdev.bad_ashift"
 #define        FM_EREPORT_ZFS_DEVICE_REMOVE            "vdev.remove"
 #define        FM_EREPORT_ZFS_DEVICE_CLEAR             "vdev.clear"
 #define        FM_EREPORT_ZFS_DEVICE_CHECK             "vdev.check"
@@ -71,6 +72,7 @@ extern "C" {
 #define        FM_EREPORT_PAYLOAD_ZFS_VDEV_DEVID       "vdev_devid"
 #define        FM_EREPORT_PAYLOAD_ZFS_VDEV_FRU         "vdev_fru"
 #define        FM_EREPORT_PAYLOAD_ZFS_VDEV_STATE       "vdev_state"
+#define        FM_EREPORT_PAYLOAD_ZFS_VDEV_ASHIFT      "vdev_ashift"
 #define        FM_EREPORT_PAYLOAD_ZFS_PARENT_GUID      "parent_guid"
 #define        FM_EREPORT_PAYLOAD_ZFS_PARENT_TYPE      "parent_type"
 #define        FM_EREPORT_PAYLOAD_ZFS_PARENT_PATH      "parent_path"
index 7d6d527..e0d82e6 100644 (file)
@@ -1271,13 +1271,16 @@ vdev_open(vdev_t *vd)
                vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
        } else {
                /*
-                * Make sure the alignment requirement hasn't increased.
+                * Detect if the alignment requirement has increased.
+                * We don't want to make the pool unavailable, just
+                * post an event instead.
                 */
-               if (ashift > vd->vdev_top->vdev_ashift) {
-                       vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
-                           VDEV_AUX_BAD_LABEL);
-                       return (EINVAL);
+               if (ashift > vd->vdev_top->vdev_ashift &&
+                   vd->vdev_ops->vdev_op_leaf) {
+                       zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT,
+                           spa, vd, NULL, 0, 0);
                }
+
                vd->vdev_max_asize = max_asize;
        }
 
index db6a831..820291b 100644 (file)
@@ -267,6 +267,10 @@ zfs_ereport_start(nvlist_t **ereport_out, nvlist_t **detector_out,
                        fm_payload_set(ereport,
                            FM_EREPORT_PAYLOAD_ZFS_VDEV_FRU,
                            DATA_TYPE_STRING, vd->vdev_fru, NULL);
+               if (vd->vdev_ashift)
+                       fm_payload_set(ereport,
+                           FM_EREPORT_PAYLOAD_ZFS_VDEV_ASHIFT,
+                           DATA_TYPE_UINT64, vd->vdev_ashift, NULL);
 
                if (pvd != NULL) {
                        fm_payload_set(ereport,