PFS3AIO v3.1 Final

[Toni Wilen]

PFS3AIO 3.1 is now available from Aminet.

http://aminet.net/disk/misc/pfs3aio.lha

3.1
---

- TD64 support check false negative fixed, v3.0 rejected many TD64 capable
HD controllers.
- Always re-execute access mode test when removable drive is inserted
and before formatting.
- Set default stack size back to original default of 4000 bytes from old
AROS compiled value of 6000. Saves stackswap 6k of RAM.
- Always execute TD_MOTOR off after idle delay period, some HD controllers
only switch off activity led when TD_MOTOR off is executed.
- Previous version's (3.0) OS 3.1.4 SCSIDIRECT support was very broken and
corrupts the partition. Reformat required if used with v3.0!
- Fixed out of bounds buffer read when writing file to disk that does not
fill last disk block completely. It was harmless but can for example
cause MuGuardianAngel hits. (Original PFS3 bug/optimization)
- Return faked (larger than real blocksize) InfoData structure block size
value if drive is larger than 16G as a workaround for programs
(including WB) that calculate in use percentage using 32-bit
"in use * 100 / total" formula where multiplication overflows if size
is about 20G or larger with 512 block size.
- Removed Direct SCSI scsi_Actual validation after read or write operation.
Some SCSI controllers don't update it properly, causing pfs3aio to show
disk read or write error requester. Was added in 3.0.
- Fixed softlinks, they have never worked previously. (Piru)

26.5.2018 WARNING: Do not use OS 3.1.4 Beta HDToolBox Direct SCSI option with 3.0. It causes pfs3aio v3.0 to create incompatible partition geometry with all other pfs3(aio) versions. Fixed in v3.1.

3.0
---

- Bug fixes, including finally fixed nasty bug causing Wrong Index Block ID
(and others) if partition was >=5G and lots of small files were created
very quickly, for example unpacking archive with thousands of small files.
This bug has most likely existed since PFS3 introduced >=5G partition
support.
- Compiled with m68k optimized GCC 6
(https://github.com/bebbo/amigaos-cross-toolchain), 16k smaller than
previous pfs3aio.
- Some HD controllers modify MaxTransfer after filesystem has been
initialized, which was ignored by previous versions.
- Automatic max transfer limit is only enabled if scsi.device version is v36
or newer and older than v50.
- Limit Direct SCSI mode max transfer size to 65535 blocks (~32MB).
- Automatically attempt to increase number of buffers if all buffers are in
use and more buffers are required.
- Added support for new OS 3.1.4 filesystem mount flags. (SCSIDIRECT,
SUPERFLOPPY and DISABLENSD).
- Shows more detailed information in error requesters (access mode, start
block, end block, SCSI details if SCSI Direct)
- Includes unsupported and experimental >104G partition support. Shows
warning if enabled during formatting.
- ACTION_DIE (unmount) stability fix.
- Filename changed from pfs3_aio-handler to pfs3aio.

--

Includes all previous pfs3aio features:

- Full KS 1.x compatibility.
- 68000+ compatible.
- TD64, NSD and Direct SCSI supported and autodetected.
and more...

Souces available from https://github.com/tonioni/pfs3aio (and pfsdoctor)
Source tree also includes experimental >4G file size (48 bit file length field) support but it is not enabled in this version. Contact me if you want to test it.

>104G partition support and >4G file size support requires reformat and on-disk structures are not anymore compatible with old PFS3 or PFSDoctor versions. (https://download.abime.net/winuae/fi...perimental.zip Note: command line only, no GUI like in official pfsdoctor)

[malko]

Thank you .
Will replace my existing hardfile to use this PFS3 test version and report if something unexpected happens.

[spudje]

Euh, for me, too stupid to understand. Do I need to install this on HDDs/CFs instead of the pfs3aio_handler and still use the same id?

So I understood wrongly (some of your other posts in a different thead, regarding USB CF adapters) when I thought the bug was in WinUAE, not in PFS3AIO?

[fgh]

Awesome, I'm glad you guys were able to hunt it down!

[Retrofan]

Very wonderful . Thanks Toni and others involved

[gulliver]

Great!

Thanks.

[ShK]

@toni. If the max partition size was 104*253*253*32 = 213 021 952 blocks, how is it now?

[cyberhead97]

Toni has done a really great job with Michiel Pelt releasing this new PFS3AIO version with nice bug fixes.

I have been using an experimental version of PFS3AIO v19.0 over the weekend stress testing it and could not reproduce the wrong index errors pfsdocfor usually reports after extracting EmeraldMines_v1.0_CD.lha and also seen copying all TimeRunner games. With PFS3AIO v2.3 these two games can cause these errors.

My tests was done using 8GB (300MB DH1, 7GB DH1) and 16GB (300MB DH0, 14.1GB DH1) hdf files first installing Workbench 3.1 and all WHDLoad Games. Then extracting EmeraldMines_v1.0_CD.lha a few times and still after this I didn't get any errors .

[Toni Wilen]

Originally Posted by spudje

Euh, for me, too stupid to understand. Do I need to install this on HDDs/CFs instead of the pfs3aio_handler and still use the same id?

Yes, name isn't same just because I wanted it to be shorter when typing it in old hdtoolbox versions

So I understood wrongly (some of your other posts in a different thead, regarding USB CF adapters) when I thought the bug was in WinUAE, not in PFS3AIO?

It was bug in PFS3 (not just pfs3aio but all pfs3 versions since ages ago). It usually only happened in emulation because it is easier and much faster to copy or unpack thousands of files in emulation.

Originally Posted by ShK

@toni. If the max partition size was 104*253*253*32 = 213 021 952 blocks, how is it now?

104*253*253*32 = ~104G (original, 1k bitmap blocks, header + 253 long words)
104*509*509*32 = ~421G (2k blocks, header + 509 long words)
104*1021*1021*32 = ~1693G (4k blocks, header + 1021 long words)

(This is now much more clearly defined in struct.h)

There is always chance of some 32-bit calculation overflows in very large partitions but up to 421G mode should be safe. Testing huge partition sizes takes far too much time...

[DanyPPC]

Thanks Toni, good work !

[supaduper]

Awesome PFSaio is a must these days for any Amiga user, thanks for ridding us of those nasty bugs Toni !

[spudje]

Thanks Toni! Awesome. I got the block ID errors a lot when copying/unpacking WHDLoad slaves on CFs in emulation. And I also got in once only recently when updating iGame library on a real Amiga.

I always had some sort of a gut feeling these errors eventually resulted in an unstable setup. So this is very welcome. Although I don't look forward redoing all my Amiga setups again :/

[Jope]

Why would you redo them? Just update the handler in your RDB.

[Toni Wilen]

If partition is smaller than 5G: no need to reformat.

If partition is 5G or larger: run pfsdoctor (original version that comes with pfs3 is fine), if no errors found: no need to reformat. If errors found: reformat recommended, pfsdoctor can't fix it properly.

EDIT: Note that original pfsdoctor does not support NSD and does not autodetect TD64/Direct SCSI.

[spudje]

Oh super, that potentially limits the work a lot indeed!

[Jope]

I obviously need to put some time aside to test this with all of the SCSI controllers I have hoarded away. I'm especially interested in seeing whether the directscsi problem I had with the GVP Series II is still there.

[patrik]

Originally Posted by Jope

I obviously need to put some time aside to test this with all of the SCSI controllers I have hoarded away. I'm especially interested in seeing whether the directscsi problem I had with the GVP Series II is still there.

I am quite sure I have asked you this before, but did all versions of gvpscsi.device suffer from this, or could it be fixed by upgrading to 4.15 which Ralph Babel has made available for free?

[supaduper]

Just updated my RDB with Toni`s Ver 19 pfs3aio and its all good

Any noobs who want to do this for partitions less then 5gb in size , just extract pfs3aio from Toni`s download then put it into L: directory.
Then fire up HDtoolbox, then >, Partition Drive >select your first partition then> tick Advanced Options, >Update > input L: pfs3aio into dialog box > Click OK ( if successful it will change to Ver19 in the window) > Click OK > Save changes to drive.

Your done..enjoy

[Jope]

Originally Posted by patrik

I am quite sure I have asked you this before, but did all versions of gvpscsi.device suffer from this, or could it be fixed by upgrading to 4.15 which Ralph Babel has made available for free?

I always upgrade any GVP I happen to have with the latest on Ralph's site. :-) I probably will not have the motivation to try any other versions but the latest 4.15 and 3.15.

[Sinphaltimus]

Originally Posted by supaduper

Just updated my RDB with Toni`s Ver 19 pfs3aio and its all good

Any noobs who want to do this for partitions less then 5gb in size , just extract pfs3aio from Toni`s download then put it into L: directory.
Then fire up HDtoolbox, then >, Partition Drive >select your first partition then> tick Advanced Options, >Update > input L: pfs3aio into dialog box > Click OK > Save changes to drive.

Your done..enjoy

I'm slightly confused about the larger than 5GB partition instructions Toni put here..

If partition is smaller than 5G: no need to reformat.

If partition is 5G or larger: run pfsdoctor (original version that comes with pfs3 is fine), if no errors found: no need to reformat. If errors found: reformat recommended, pfsdoctor can't fix it properly.

EDIT: Note that original pfsdoctor does not support NSD and does not autodetect TD64/Direct SCSI.

So just to ask for clarity on this.

If the partition is bigger than 5GB, follow the same instructions as you posted for smaller than 5GB partition then run the original PFSDoctor on it and if no errors I'm good to go? If errors then I need to reformat

OR

Do I run the original PFSDoctor BEFORE doing this change over?