|
From: Rob L. <ro...@la...> - 2013-05-19 04:26:28
|
On 05/17/2013 03:36:46 AM, Dominique Martinet wrote: > Hi, > > Dominique Martinet wrote on Wed, Mar 20, 2013 : > > I've seen that mmap in 9p has been made read-only in the client for > > quite a while (commit 14b8869ff in feb 2008), and the commit message > > says it was because this would eventually cause a kernel BUG() call. > > > > I've also seen some threads in mail archives concerning writable > > mmap and cache here: > > http://thread.gmane.org/gmane.linux.kernel/1096376 > > and later > > http://www.spinics.net/lists/linux-fsdevel/msg42311.html > > > > And the code (fs/9p/vfs_file.c) looks like to me writable mmap is > > available only for cache=fscache or loose. > > > > > > So question: > > Would a writable mmap be possible in 9p without cache altogether, > or are > > they incompatible at a theorical/protocol level? > > > Allowing myself to repost this (with the revert commit author in > direct > copy) as it didn't get any reply. I'm interested in seeing this work. Has anybody tried any of the filesystem stress tests on this? The leader seems to be "xfstests": http://xfs.org/index.php/Getting_the_latest_source_code#XFS_tests But there's a bunch of others: http://dbench.samba.org/ https://lwn.net/Articles/326926/ http://www.bullopensource.org/ext4/ext4_test_plan.html And so on. That might be better than just enabling it and hoping it works. Instead actually trying to make it fail under load... Rob |