Forcing the disk cache to be written to disk

Question

Hi all. We are looking for a way to insure the content of the icommon in the ufs on disk as we need to read it. However, calling sync is async and does not seem to provide what we're looking for. When a file is updated, created all the information is not immediately written to disk. It is kept in the unified memory and later flushed to the disk cache which then writes it down on the media. However, since we're reading the hot FS we need to force ALL file metadata AND data to be written to the media.
Seems like when I run the command 'ff', probably as a side-effect, this is happening. ff takes too long and is at the FS level. Is it a side-effect, or does 'ff' calls a specific func (truss did not reveal anything useful).
Thanks all and best regards.

Answer 1

I. What is really happening under the hood?
1.sync(2) passes execution into kernel mode
2.kernel function syssync() is called then
3.then vfs_sync(0)
4.eventually ufs_sync(vfsp,�) is called with NULL as a first argument
in case of NULL(as a 1st arg) ufs_sync() just schedules but does not necessarily completes the writing of ufs metadata before returning.
So, this behavior completely matches with one described in man section for sync(2).
II. directio does not solve your problem because it affects only the way the file data( not metadata) goes to disk. ( read directio(3C) and mount_ufs(1M) carefully )
III. What you really need
May be it seems to be a heavy weapon for you but one of the possibilities is:
     To write a loadable system call (loadable kernel module) that will
invoke ufs_sync() with proper arguments (non NULL vfsp) for mounted file
system of interest.