Re: bin/45933: fsck_ffs can't find wapbl journal on 4k-block disks

[Michael van Elst <mlelstv%serpens.de@localhost>]

The following reply was made to PR bin/45933; it has been noted by GNATS.

From: Michael van Elst <mlelstv%serpens.de@localhost>
To: David Krinsky <krinsky%bantha.org@localhost>
Cc: gnats-bugs%NetBSD.org@localhost, mlelstv%NetBSD.org@localhost, 
gnats-admin%netbsd.org@localhost,
        netbsd-bugs%netbsd.org@localhost
Subject: Re: bin/45933: fsck_ffs can't find wapbl journal on 4k-block disks
Date: Tue, 15 Jul 2014 06:33:58 +0200

 On Sun, Jul 13, 2014 at 06:38:51PM -0400, David Krinsky wrote:
 
 > # dumpfs -s /dev/xbd4
 > wapbl version 0x1       location 1      flags 0x0
 > wapbl loc0 268435456    loc1 131072     loc2 512        loc3 0
 
 location = 1     UFS_WAPBL_JOURNALLOC_END_PARTITION
 loc0 = 268435456 Journal placed at this physical offset 
 loc1 = 131072    Journal size is 128k sectors.
 loc2 = 512       Physical sector size 512 bytes.
 
 So WAPBL says it uses 512 byte sectors.
 
 
 > # drvctl -p xbd4        
 >                         <key>sector-size</key>  
 >                         <integer>0x200</integer>
                                 
 The disk driver also reports 512 byte sectors.
 
 However, fsck uses this value only to find the superblock and does
 a recalculation of the sector size based on the superblock contents,
 so that it can support filesystem images.
 
 dev_bsize = sblock->fs_fsize / FFS_FSBTODB(sblock, 1);
 
 with
 
 > fsize   4096    shift   12      mask    0xfffff000
 > frag    8       shift   3       fsbtodb 0
 
 this evaluates to:
 
             4096             / (1 << 0)
 
 The problem is that the fsbtodb value in the superblock is wrong.
 The value holds the shift value between filesystem block size
 (which is really the fragsize not the blocksize!) and
 the physical sector size. So the superblock says the physical
 sector size is 4096.
 
 Did you create the filesystem with newfs -S 4096 or on the Dom0 ?
 That would explain why the superblock has the wrong value.
 
 Kernel code works because it ignores the fsbtodb value. It also
 created the journal at the first mount and stored its location
 based on the actual driver values.
 
 
 Greetings,
 -- 
                                 Michael van Elst
 Internet: mlelstv%serpens.de@localhost
                                 "A potential Snark may lurk in every tree."