home *** CD-ROM | disk | FTP | other *** search
- From: Stephen Henson (CSM) <shenson@camborne-school-of-mines.ac.uk>
- Date: Thu, 4 Mar 93 22:23:06 GMT
- Message-Id: <15252.9303042223@csm.ac.uk>
- To: mint <mint@terminator.rs.itd.umich.edu>
- Subject: New MiNT fcntl calls etc
-
-
- I agree with many of the sentiments mentioned here about new fcntl calls.
- FUTIME is simplicity itself to add to minixfs and an inode level truncate
- function has been waiting for support for some time.
-
- Fsync() is also a good idea because (among other reasons) there are
- two kinds of disk change. "Hard", by physically changing the disk, and
- "Soft", by a request, via for example Dlock(). The "Soft" kind should Fsync()
- the disk in question first before the change. This would avoid things like
- fsck reporting spurious errors because the cache never got written out (I've
- written minixfs to avoid this kind of thing, but at a performance cost).
-
- Filesystems ought to have a way to specify no operations can be
- performed on a given drive. For example, sometimes when a minix partition
- is recognised it can't be accessed because e.g. MAX_INCREMENT was set too
- low when minix.xfs was compiled. The best way to handle this is to inhibit
- all operations on the drive but do the usual checks when it is changed. One
- way would be to pass a 'null_fs' pointer to the filesystem at start up which
- fs->root could then set as necessary. This could be done by adding lots of
- little checks and a flag to minix.xfs but that is a bit messy.
-
- Non blocking disk I/O requests would be nice. Although low level this
- isn't too hard to do. Floppies would be a bit more painful. This would make
- multi-thread filesystems possible (there isn't much point with blocking I/O).
- Having said that some of the other device drivers need an overhaul as well.
-
- Steve.
-