Date: Wed, 8 Jan 1997 09:55:52 -0700 (MST) From: Terry Lambert <terry@lambert.org> To: sos@FreeBSD.org Cc: terry@lambert.org, current@FreeBSD.org Subject: Re: Any specfs specialists out there ?? Message-ID: <199701081655.JAA16505@phaeton.artisoft.com> In-Reply-To: <199701080737.IAA01307@ravenock.cybercity.dk> from "sos@FreeBSD.org" at Jan 8, 97 08:37:48 am
next in thread | previous in thread | raw e-mail | index | archive | help
> > > I faked this by using 2*DEV_BSIZE which works just dandy on my > > > 1K sec disks. > > > > > > Any good ideas ?? > > > > What is this for? > > It just for me having disks with a 1K sector size, and without this > it fails :(. Ah. I faced the same problem when I added Unicode support to FFS, since it bumped the minimum directory block size to 1k. You can either make sure the minimal FS block size is 8k or higher, or for a 4k, make the min frag size 1k anyway. Really, specfs should not use the DEV_BSIZE as a manifest value; it should get the block size out of the structure for the device, and for your particular device, that should be 1k. I would be very careful with 512b boundry access on such a device in any case (ie: even if you make the change to specfs, you should ensure that no FS is attempting to access in units of smaller than the block size -- the directory block size may be a problem here...). > > Specfs will be dying soon if devfs becomes standard (specfs and struct > > fsops is a crock). > > Yeah, but I wouldn't hold my breath.... After specfs is gone, the pipe code (and by extension, the LOCAL domain socket code) will be the only real users of struct fsops dereference; they should be moved into the FS code and follow the path down. After that, we should be able to drastically shrink struct file for open file instances in the system open file table, and recover multiple pointer dereferences worth of speed on all operations. 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?199701081655.JAA16505>