kern/153695: [PATCH] [ZFS] Booting from zpool created on 4k-sector drive doesn't work

Tue Jan 4 23:50:09 UTC 2011

>Number:         153695
>Category:       kern
>Synopsis:       [PATCH] [ZFS] Booting from zpool created on 4k-sector drive doesn't work
>Confidential:   no
>Severity:       serious
>Priority:       low
>Responsible:    freebsd-bugs
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Tue Jan 04 23:50:09 UTC 2011
>Closed-Date:
>Last-Modified:
>Originator:     Emil Smolenski
>Release:        FreeBSD 8.2-PRERELEASE
>Organization:
>Environment:
FreeBSD 8.2-PRERELEASE amd64
>Description:
There is a hack to force zpool creation with minimum sector size equal to 4k:

# gnop create -S 4096 ${DEV0}
# zpool create tank ${DEV0}.nop
# zpool export tank
# gnop destroy ${DEV0}.nop
# zpool import tank

Zpool created this way is faster on problematic 4k sector drives which lies about its sector size (like WD EARS). This hack works perfectly fine when system is running. Gnop layer is created only for "zpool create" command -- ZFS stores information about sector size in its metadata. After zpool creation one can export the pool, remove gnop layer and reimport the pool. Difference can be seen in the output from the zdb command:

- on 512 sector device (2**9 = 512):
% zdb tank |grep ashift
ashift=9

- on 4096 sector device (2**12 = 4096):
% zdb tank |grep ashift
ashift=12

This change is permanent. The only possibility to change the value of  
ashift is: zpool destroy/create and restoring pool from backup.

But there is one problem: I cannot boot from such pool. Error message:

ZFS: i/o error - all block copies unavailable
ZFS: can't read MOS
ZFS: unexpected object set type 0

This is standard configuration with GPT scheme.

# gpart show da0
=>        34  2930211565  da0  GPT  (1.4T)
           34          30       - free -  (15K)
           64         128    1  freebsd-boot  (64K)
          192     4194304    2  freebsd-swap  (2.0G)
      4194496     8388608    3  freebsd-zfs  (4.0G)
     12583104  2917628495       - free -  (1.4T)

# zpool status tank
   pool: tank
  state: ONLINE
  scrub: none requested
config:

         NAME         STATE     READ WRITE CKSUM
         tank         ONLINE       0     0     0
           gpt/tank0  ONLINE       0     0     0

# zdb -uuu tank
Uberblock

         magic = 0000000000bab10c
         version = 15
         txg = 2838
         guid_sum = 12371721502612965633
         timestamp = 1292860198 UTC = Mon Dec 20 15:49:58 2010
         rootbp = [L0 DMU objset] 800L/200P DVA[0]=<0:2041000:1000>  
DVA[1]=<0:30062000:1000> DVA[2]=<0:ee0bd000:1000> fletcher4 lzjb LE contiguous birth=2838 fill=374 cksum=c9605617d:4e2cf0a8c94:f6decb77086a:210752c3aee4a8

This PR is similar to message which I send to freebsd-fs@ mailing list: http://lists.freebsd.org/pipermail/freebsd-fs/2010-December/010350.html
>How-To-Repeat:
Install FreeBSD 8.2-PRERELEASE like described here: http://wiki.freebsd.org/RootOnZFS/GPTZFSBoot with following exceptions:

1. Add "-b 32K" to "gpart add -s 64K -t freebsd-boot ad0" command for proper 4k alignment.
2. Issue "gnop create -S 4096 ad0" command before creating zpool.
3. Use .nop device to create zpool.
4. Export the zpool, remove the gnop layer and reimport the zpool.
>Fix:
Attached patch is far from being perfect. I posted it here because maybe it could help someone skilled enough in tracking down this issue and provide proper solution.

With this patch applied:
- I can boot from single disk zpool,
- I can boot from mirrored zpool,
- I can't boot from raidz,
- I can't boot from mirrored zpool created on HP SmartArray P400 (it may be related to bug described in this PR: kern/151910

Patch attached with submission follows:

diff -ruN sys.orig/boot/zfs/zfsimpl.c sys/boot/zfs/zfsimpl.c

--- sys.orig/boot/zfs/zfsimpl.c	2010-12-29 13:55:38.195215000 +0100
+++ sys/boot/zfs/zfsimpl.c	2010-12-29 13:24:39.825206000 +0100
@@ -770,7 +770,7 @@
 	const char *pool_name;
 	const unsigned char *vdevs;
 	int i, rc, is_newer;
-	char upbuf[1024];
+	vdev_phys_t upbuf;
 	const struct uberblock *up;
 
 	/*
@@ -921,21 +921,21 @@
 	 * the contents of the pool.
 	 */
 	for (i = 0;
-	     i < VDEV_UBERBLOCK_RING >> UBERBLOCK_SHIFT;
+	     i < VDEV_UBERBLOCK_COUNT(vdev);
 	     i++) {
-		off = offsetof(vdev_label_t, vl_uberblock);
-		off += i << UBERBLOCK_SHIFT;
+		off = VDEV_UBERBLOCK_OFFSET(vdev, i);
 		BP_ZERO(&bp);
 		DVA_SET_OFFSET(&bp.blk_dva[0], off);
-		BP_SET_LSIZE(&bp, 1 << UBERBLOCK_SHIFT);
-		BP_SET_PSIZE(&bp, 1 << UBERBLOCK_SHIFT);
+		DVA_SET_ASIZE(&bp.blk_dva[0], VDEV_UBERBLOCK_SIZE(vdev));
+		BP_SET_LSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
+		BP_SET_PSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
 		BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
 		BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
 		ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
-		if (vdev_read_phys(vdev, &bp, upbuf, off, 0))
+		if (vdev_read_phys(vdev, &bp, &upbuf, off, 0))
 			continue;
 
-		up = (const struct uberblock *) upbuf;
+		up = (const struct uberblock *) &upbuf;
 		if (up->ub_magic != UBERBLOCK_MAGIC)
 			continue;
 		if (up->ub_txg < spa->spa_txg)
diff -ruN sys.orig/cddl/boot/zfs/zfsimpl.h sys/cddl/boot/zfs/zfsimpl.h
--- sys.orig/cddl/boot/zfs/zfsimpl.h	2010-12-29 13:55:58.870864000 +0100
+++ sys/cddl/boot/zfs/zfsimpl.h	2010-12-29 13:24:29.014796000 +0100
@@ -324,7 +324,7 @@
 #define	VDEV_UBERBLOCK_RING	(128 << 10)
 
 #define	VDEV_UBERBLOCK_SHIFT(vd)	\
-	MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT)
+	MAX((vd)->v_ashift, UBERBLOCK_SHIFT)
 #define	VDEV_UBERBLOCK_COUNT(vd)	\
 	(VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd))
 #define	VDEV_UBERBLOCK_OFFSET(vd, n)	\


>Release-Note:
>Audit-Trail:
>Unformatted:
