Re: kern/54470: Adapt FFS to work with HP-UX UFS1 variant

[Konrad Schroder <perseant%hhhh.org@localhost>]

Do we have a way to construct a file system that works with NetBSD, 
which has a physical sector size other than DEV_BSIZE?  If so, I can 
test against it.

Note too that the macros in question exist only within the scope of the 
fsck_ffs utility.  The other userland programs that I adapted didn't 
need them, and the kernel uses macros that depend on struct ump and 
could easily be special-cased for HP-UX or other old file systems if 
need be.  Come to think of it, in fsck_ffs we could just do something like

    exter daddr_t dev_bshift;

    daddr_t dev_bshift = DEV_BSHIFT;

    #define BTODB(size) ((size) >> dev_bshift)
    #define DBTOB(size) ((size) << dev_bshift)

and change the value of dev_bsize, if necessary, in the setup.  The 
kernel would do the same, but storing the dev_bshift value in struct 
ump.  Would this address your concern?

Thanks,

-Konrad

On 8/17/2019 4:20 AM, Michael van Elst wrote:
> The following reply was made to PR kern/54470; it has been noted by GNATS.
>
> From: mlelstv%serpens.de@localhost (Michael van Elst)
> To: gnats-bugs%netbsd.org@localhost
> Cc:
> Subject: Re: kern/54470: Adapt FFS to work with HP-UX UFS1 variant
> Date: Sat, 17 Aug 2019 11:15:28 -0000 (UTC)
>
>   perseant%hhhh.org@localhost (Konrad Schroder) writes:
>   
>   >Can you be more specific?  It gives the right values for my HP-UX image,
>   >and for existing NetBSD file systems.
>   
>   >>   >+#define BTODB(fs, size) ((size) >> ((fs)->fs_fshift - (fs)->fs_fsbtodb))
>   >>   >+#define DBTOB(fs, bc)   ((bc) << ((fs)->fs_fshift - (fs)->fs_fsbtodb))
>   
>   The original BSD code used definitions like these above as a 'disk offset'
>   referred to physical block numbers. This is still used that way in
>   userland.
>   
>   NetBSD disk drivers always use logical block numbers for a block size
>   of 512 bytes (DEV_BSIZE) and filesystems ignore the superblock values when
>   translating between filesystem block numbers and disk block numbers.
>   
>   For FFS the macros FFS_FSBTODB / FFS_DBTOFSB are used to do this conversion.
>   They are differently defined for kernel and userland to address the different
>   views.
>   
>   btodb / dbtob are originally MD macros (then unified) that always used
>   a platform specific unit, i.e. DEV_BSIZE, to specify disk offsets. This
>   unit is independent of a particular filesystem and the physical blocks of
>   a specific device.
>   
>   Now what unit does di_blocks use, it's only used for reporting and quota
>   calculations ? The original code used btodb(file_length), i.e. the values
>   used the platform-wide DEV_BSIZE as a unit and that's what we still use
>   today.
>   
>   Your patch changes that and di_blocks is now using the physical block
>   size of the particular disk (recorded in the superblock) as the unit.
>   Apparently that's what HP-UX is doing.
>   
>   I see two issues: you added another view on disk offsets and you break
>   existing filesystems that use a physical sector size != 512.
>   
>   Apparently we need conditional code to handle di_blocks for the
>   different systems.
>   
>   
>   --
>   --
>                                   Michael van Elst
>   Internet: mlelstv%serpens.de@localhost
>                                   "A potential Snark may lurk in every tree."
>