Date: Thu, 25 Sep 1997 22:44:22 +0000 (GMT) From: Terry Lambert <tlambert@primenet.com> To: kato@migmatite.eps.nagoya-u.ac.jp (KATO Takenori) Cc: bde@zeta.org.au, current@FreeBSD.ORG Subject: Re: Daily SNAPshots at current.freebsd.org shut down for now. Message-ID: <199709252244.PAA05902@usr04.primenet.com> In-Reply-To: <199709251645.BAA07216@gneiss.eps.nagoya-u.ac.jp> from "KATO Takenori" at Sep 26, 97 01:45:50 am
next in thread | previous in thread | raw e-mail | index | archive | help
> > clustering is 2 too many for a production system :-) (non-clustering for
> > /dev/vn must be forced independently of any mount options and sysctls,
> > except possibly for debugging).
>
> How does mount() in vfs_syscall.c know special device is /dev/vn?
It needs to be a per device flag, only asserted by the vn device.
> When mount() can know specical device is /dev/vn, then
>
> if (special device is /dev/vn)
> mp->mnt_flag |= MNT_NOCLUSTER;
>
> works, and doclusterread and doclusterwrite can be eliminated.
The easiest place to add this would be as a falg bit in d_flags;
put it under D_CHAIN in sys/disklabel.h:
#define D_CHAIN 0x10 /* can do back-back transfers */
#define D_UNCLUSTERED 0x20 /* do not cluster{read|write} */
This flag is to be set in the disklabel (look how D_ECC is handled).
Personally, I'd prefer to have it an attribute of the "real" device,
but that would mean changing d_type into a bitmap, and adding a flag
in sys/conf.h and changing the dev/vn/vn.c from:
static struct bdevsw vn_bdevsw =
{ vnopen, vnclose, vnstrategy, vnioctl, /*15*/
vndump, vnsize, D_DISK, "vn", &vn_cdevsw, -1 };
to:
static struct bdevsw vn_bdevsw =
{ vnopen, vnclose, vnstrategy, vnioctl, /*15*/
vndump, vnsize, D_DISK|D_UNCLUSTERED,
"vn", &vn_cdevsw, -1 };
The alternative of adding:
#define D_CCD 4 /* a concatenated disk*/
Would require too much overhead to check for a d_type of D_CCD in addition
to a d_type of D_DISK literally *everywhere*.
Either change would mean touching a lot more files than the disklabel
hack... but it *is* a hack.
On the other hand, Bruce did ask "there's time to go back and do it right
later?" (or something to that effect), so he might champion a large change
(bit checking of d_type instead of value checking, everywhere it's
checked currently).
Anyway, that'd be my preference for implementation.
Terry Lambert
terry@lambert.org
---
Any opinions in this posting are my own and not those of my present
or previous employers.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199709252244.PAA05902>