Date: Tue, 4 Apr 2000 22:29:29 +1000 (EST) From: Bruce Evans <bde@zeta.org.au> To: Poul-Henning Kamp <phk@critter.freebsd.dk> Cc: "Justin T. Gibbs" <gibbs@FreeBSD.org>, cvs-committers@FreeBSD.org, cvs-all@FreeBSD.org Subject: Re: cvs commit: src/sys/kern kern_mib.c vfs_bio.c src/sys/sys buf.h Message-ID: <Pine.BSF.4.21.0004042139060.2001-100000@alphplex.bde.org> In-Reply-To: <23861.954797317@critter.freebsd.dk>
next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, 3 Apr 2000, Poul-Henning Kamp wrote:
> In message <Pine.BSF.4.21.0004040649300.1047-100000@alphplex.bde.org>, Bruce Ev
> ans writes:
> >Not quite. Something obviously has to set the disk-level block number
> >(currently bio_pblkno). It should be set from a valid block number or
> >offset (bio_blkno or bio_offset). There needs to be a validity bit or
> >out-of bounds values to indicate invalid fields (I prefer the bit).
> >Callers prefer bio_blkno if it is convenient and works (i.e., in most
> >cases, until 1TB disks become common).
> Uhm Bruce, the code to implement this dual valid bit thingie will take
> up more space than we will ever save by doing it.
>
> Lets keep it simple please.
You don't understand. This _is_ simple and efficient. See the first
40 lines of dscheck() which deal with translation from a logical blkno
to slice-relative physical blkno. The duality bit handling adds only
about 28 lines to this:
if (ssp->dss_secmult == 1) { /* old */
if (biop->bio_flags & BIO_BYTEOFFSET) {
if (biop->bio_offset % (u_int64_t)DEV_BSIZE)
goto bad_bio_offset;
if (biop->bio_offset >= (u_int64_t)DEV_BSIZE << 31)
goto bad_bio_offset;
secno = biop->b_offset >> DEV_BSHIFT;
} else
secno = blkno; /* old */
... /* old */
} else { /* old */
and 2 more sets of 7 new lines for the other 2 special cases for
ssp->dss_secmult, and 7 new lines for error reporting.
The few callers that want a byte offset need 1 new line to set
BIO_BYTEOFFSET. Currently, only physio() wants this.
> I will propose we stick to just one field in struct bio: a 64bit
> byte offset, we may want to keep a blkno around in struct buf,
> but lets decide that separately.
This requires more widespread changes to convert block numbers to offsets
in all file systems, and normally gives 2 extra sets of conversions:
first to offsets in the file systems, then to sector numbers in dscheck().
With more work on the filesystems, some of the conversions can be
combined. E.g., in ffs, fs->fsbtodb gives the shift count for converting
fs-blocks to "disk" blocks. It needs to be replaced by (fs->fsbtodb + 9)
to convert directly from fs-blocks to byte offsets.
Bruce
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe cvs-all" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.21.0004042139060.2001-100000>