Fwd: DELETE support in the VOP_STRATEGY(9)?
Dag-Erling Smørgrav
des at des.no
Tue Dec 8 19:02:40 UTC 2015
Konstantin Belousov <kostikbel at gmail.com> writes:
> Exactly. I completely agree with the statement above, and think that
> this is the only thing that should be implemented. There could be a
> request, most likely an filesystem-level ioctl, which punches holes in
> the file. Its effect on the file state should be the same as if the
> seek was done between writes, if the filesystem supports the ioctl.
> More, if supported, the lseek(SEEK_HOLE/SEEK_DATA) behaviour should
> be consistent with the request. The later might mean that we should
> restrict the interface to only accept ranges at block boundaries.
Having spent a couple of weeks recently trying to get code that uses
SEEK_HOLE to work reliably, my advice is to burn it down and salt the
earth. Wait, am I thinking out loud again? What I meant to say is that
there is no such thing as consistency where SEEK_HOLE is concerned, and
it might as well just be an alias for (SEEK_END, 0).
DES
--
Dag-Erling Smørgrav - des at des.no
More information about the freebsd-hackers
mailing list