Subject: Re: SCSI MMC device abstraction and UDF patch for review
To: None <reinoud@netbsd.org>
From: YAMAMOTO Takashi <yamt@mwd.biglobe.ne.jp>
List: tech-kern
Date: 12/27/2005 07:22:05
> > > ftp://ftp.NetBSD.org/pub/NetBSD/misc/reinoud/src-diffs-udf.20051222.gz
> >
> > - it seems to do double-caching. ie. at both of udf and block device.
> > is it intended?
>
> Yes and no. For UDF it is better to use caching in udf since blocks can
> move around esp. on recordables; every disc write changes the location of
> the block so its easier to have it cached on the file's vnode on its
> logical block instead of a moving buffer on disc. Setting B_NOCACHE in the
> buffers passed to the block device is prolly advisable yes.
>
> The two stubs for reading are to be integrated in a safeguard mechanism
> that orders read and write requests in such a way that it doesn't inflict
> adressing issues due to percularities in writing to recordables.
it also involves redundant memcpy, which can't be solved by B_NOCACHE.
> > - "bp->b_lblkno = bp->b_blkno;" in udf_strategy is only for printf?
>
> not intentionally. During developement of the code i read that when the
> logical block number wasn't equal to the blocknumber that it was then
> allready translated with VOP_BMAP and thus could be directed to the
> device's strategy routine. This might be in imported FS though. To assure
> that *all* reads and writes are passed trough UDF i made this explicit
> assurance that all buffers are allways BMAP'd and thus passed trough UDF's
> strategy. This might be based on old or incomplete information but i'd
> rather be safe in this than get surprised later by other parts of the FS
> stack making this assumption.
i'd say "safe" == "hide bugs".
> > - is udf sector size ("lb_size" in udf_strategy)
> > always smaller than or equal to page size?
>
> no... lb_size can be every multiple of 512 in the UDF spec though trough
> limitations in NetBSD the code only accepts lb_sizes in powers of 2. See
> the check in udf_mount(). Most common are 512 and 2048 in practice. Other
> values are more to safeguad implementations to make assumptions on this.
> Would/could this pose problems ?
if it's larger than page size, page-aligned i/o can't be be represented
in the unit.
YAMAMOTO Takashi