| |||||||||||||
The 2010 Linux Storage and Filesystem Summit, day 1The 2010 Linux Storage and Filesystem Summit, day 1Posted Aug 12, 2010 12:48 UTC (Thu) by jlayton (subscriber, #31672)Parent article: The 2010 Linux Storage and Filesystem Summit, day 1
> The final lightning talk came from Boaz Harrosh, who talked about "stable
> pages." Currently, pages which are currently under writeback can be > modified by filesystem code. That's potentially a data integrity problem, > and it can be fatal in situations where, for example, checksums of page > contents are being made.
This would be a boon for network filesystems too. Right now, it's rather nasty to deal with things like signing in NFS and CIFS. We have to take a checksum of the packet contents, but by the time you do that the page contents can change.
This is especially a problem with CIFS as the server will often drop the connection if the packet integrity seems to be compromised, and dropped connections with CIFS can mean the loss of a lot of state (open files, locks, etc).
(Log in to post comments)
The 2010 Linux Storage and Filesystem Summit, day 1 Posted Apr 5, 2011 23:13 UTC (Tue) by butlerm (subscriber, #13312) [Link]
This sounds like a good idea, but suppose a file had several pages that were mapped into some processes' memory with PROT_WRITE access. Wouldn't this require the filesystem to do the following?:
(1) temporarily mark those pages as write only while writeback was in progress
|
|||||||||||||