On Thu, Aug 10 2017, Olga Kornievskaia wrote: > On Wed, Aug 9, 2017 at 8:55 AM, David Howells <dhowells@xxxxxxxxxx> wrote: >> You may also want to flush any outstanding dirty data and wait for in-progress >> operations. > > Sorry for the delayed response but I've been thinking about it as this > is a tricky one (for me at least). > > Even currently, each file system needs a way to deal with flushing > cached data to storage in the situation where creds might have expired > in between when the kernel returned control back to the user but > before all of buffered writes are flushed. NFS4.1 has wording in the > spec for using machine credentials in that case. > > At the VFS layer, there no what to tell which dirty data belongs to > which user. Flushing all data under the superblock seems like a bad > idea? NFS flushes data when the file descriptor is closed. So as long as the user does have any open-for-write file descriptors, their data should be safe. Purging credentials while you still have open-for-write file descriptors is probably not a good idea. This is not the case if you "nocto" mount option is used, but that is recommended only for read-mostly mounts. NeilBrown
Attachment:
signature.asc
Description: PGP signature
