Does UFS2 send BIO_FLUSH to GEOM when update metadata (with
softupdates)?
Lev Serebryakov
lev at freebsd.org
Fri Nov 25 12:27:45 UTC 2011
Hello, Kostik.
You wrote 25 ноября 2011 г., 15:37:56:
> Harware that reports finished write and not making the data to stable
> storage, or software driver that does such thing itself are just
> utterly broken.
Ok, than any modern hardware, especially high-performance one, is
"utterly broken". We need to live with this.
Almost all this hardware provide some knobs to fix situation in
critical cases -- flags for synchronious write, etc. FFS/SU/SU_J
doesn't use these knobs. All such flags are lost on "struct buf" <-> "struct
bio" boundary.
I like patch from ivoars@, and even more like approach with new
flag for BIO and changes in ATA and SCSI drivers to support this.
And my geom_raid5, which NEEDS write cache, too.
Alexander Motin (mav@), AHCI driver author and person, who do a lot of
work in GEOM now, agree with me (I could provide you with quotes from
off-list message exchange in Russian). He agrees to add support for
this flag to AHCI/ATA layer, if it will be introduced.
IMHO, something should be done here. And this ``should be done''
is not ``somebody should done this.'' I could do coding, testing
(but not excessive performance one 00 I don't have enough spare
hardware), mav@ seems to be ready to make performance tests. But
here is one trick: FFS2/SU code is very fragile, and I will need
supervision from FFS guru. And if FFS guru thinks, that all storage
hardware in world should be fixed (in pricv of huge perofrmance
degradation), and FFS/SU is Ok as-is, it will be impossible.
--
// Black Lion AKA Lev Serebryakov <lev at serebryakov.spb.ru>
More information about the freebsd-fs
mailing list