From 3bd1e76265e7655e601372e51bc2413198378122 Mon Sep 17 00:00:00 2001
From: Mike Zeller <mike@mikezeller.net>
Date: Wed, 24 Oct 2018 12:27:04 -0400
Subject: [PATCH] OS-4718 ZFS actively hostile to 512e drive replacements

---
 usr/src/uts/common/fs/zfs/spa.c           | 10 +++++++---
 usr/src/uts/common/fs/zfs/sys/vdev_impl.h |  1 +
 usr/src/uts/common/fs/zfs/vdev_disk.c     | 22 +++++++++++++++++++++-
 3 files changed, 29 insertions(+), 4 deletions(-)

diff --git a/usr/src/uts/common/fs/zfs/spa.c b/usr/src/uts/common/fs/zfs/spa.c
index 599ae9faded9..ffc1363cedab 100644
--- a/usr/src/uts/common/fs/zfs/spa.c
+++ b/usr/src/uts/common/fs/zfs/spa.c
@@ -5733,11 +5733,15 @@ spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
 		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
 
 	/*
-	 * The new device cannot have a higher alignment requirement
-	 * than the top-level vdev.
+	 * The new device cannot have a higher alignment requirement than the
+	 * top-level vdev.  If this is an Advanced Format (e.g. 512e) disk, we
+	 * also need to check the fallback logical ashift value.
 	 */
-	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
+	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift &&
+	    (newvd->vdev_ashift_af == 0 ||
+	    newvd->vdev_ashift_af > oldvd->vdev_top->vdev_ashift)) {
 		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
+	}
 
 	/*
 	 * If this is an in-place replacement, update oldvd's path and devid
diff --git a/usr/src/uts/common/fs/zfs/sys/vdev_impl.h b/usr/src/uts/common/fs/zfs/sys/vdev_impl.h
index 71753cf24f7d..76acba5b3211 100644
--- a/usr/src/uts/common/fs/zfs/sys/vdev_impl.h
+++ b/usr/src/uts/common/fs/zfs/sys/vdev_impl.h
@@ -201,6 +201,7 @@ 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	*/
+	uint64_t	vdev_ashift_af;	/* adv. format fallback shift	*/
 	uint64_t	vdev_state;	/* see VDEV_STATE_* #defines	*/
 	uint64_t	vdev_prevstate;	/* used when reopening a vdev	*/
 	vdev_ops_t	*vdev_ops;	/* vdev operations		*/
diff --git a/usr/src/uts/common/fs/zfs/vdev_disk.c b/usr/src/uts/common/fs/zfs/vdev_disk.c
index e4b86b419b9d..b7bbc48aff40 100644
--- a/usr/src/uts/common/fs/zfs/vdev_disk.c
+++ b/usr/src/uts/common/fs/zfs/vdev_disk.c
@@ -530,11 +530,31 @@ vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
 		VDEV_DEBUG("vdev_disk_open(\"%s\"): "
 		    "both DKIOCGMEDIAINFO{,EXT} calls failed, %d\n",
 		    vd->vdev_path, error);
-		pbsize = DEV_BSIZE;
+		blksz = pbsize = DEV_BSIZE;
 	}
 
 	*ashift = highbit64(MAX(pbsize, SPA_MINBLOCKSIZE)) - 1;
 
+	/*
+	 * Advanced Format (512e) disks have a 4KB physical sector size, but
+	 * also report a 512 byte logical sector size (through emulation in the
+	 * firmware) to better support legacy operating systems.  While we
+	 * generally wish to create new pools with a 4KB block size, we also
+	 * need to allow people to use AF disks in their existing 512 byte
+	 * pools, even if not completely optimal.
+	 */
+	if (blksz != 0 && blksz < pbsize) {
+		/*
+		 * The logical block size is smaller than the reported physical
+		 * block size.  Record the logical ashift so that
+		 * spa_vdev_attach() can use it as a fallback.
+		 */
+		vd->vdev_ashift_af = highbit64(MAX(blksz,
+		    SPA_MINBLOCKSIZE)) - 1;
+	} else {
+		vd->vdev_ashift_af = 0;
+	}
+
 	if (vd->vdev_wholedisk == 1) {
 		int wce = 1;