Working on vnode sequencing and locking - HEAD will destabilize for a little bit
dillon at apollo.backplane.com
Thu Aug 10 10:03:29 PDT 2006
:On 2006-08-09, Matthew Dillon <dillon at xxxxxxxxxxxxxxxxxxxx> wrote:
:> And, poof, no more indefinite blocking states in the kernel for NFS or
:> for the upcoming userland VFS or clustering.
:> There will be fine-grained range locks to maintain UNIX atomicy
:> requirements, but since they aren't going to be hard locks they won't
:> prevent basic things like ^C from working properly.
:Btw, what does "^C working properly" mean? AFAIK SUS says/implies that
:filesystem related syscalls shouldn't be interruptible, with the
:exception of read and write.
Well, maybe not ^C so much as SIGKILL (signal 9). The reality is that
the requirement never took into account things like NFS or really any
remote filesystem. This is why NFS has a shim called the 'intr' option
that allows you to ^C a filesystem op blocked in an NFS RPC.
I think a distinction has to be made between normal signal interruption,
which we clearly do not want to allow, and termination cases which
we clearly do want to allow.
<dillon at xxxxxxxxxxxxx>
More information about the Kernel