On Thu, Sep 16, 2021 at 04:16:55PM +0200, Jan Kara wrote: > On Wed 15-09-21 20:35:45, Ming Lei wrote: > > When running ->fallocate(), blkdev_fallocate() should hold > > mapping->invalidate_lock to prevent page cache from being > > accessed, otherwise stale data may be read in page cache. > > > > Without this patch, blktests block/009 fails sometimes. With > > this patch, block/009 can pass always. > > > > Cc: Jan Kara <jack@xxxxxxx> > > Signed-off-by: Ming Lei <ming.lei@xxxxxxxxxx> > > Interesting. Looking at block/009 testcase I'm somewhat struggling to see > how it could trigger a situation where stale data would be seen. Do you > have some independent read accesses to the block device while the testcase > is running? That would explain it and yes, this patch should fix that. Yeah, the issue can be reproduced quite easily if independent read is run. However, sometimes it can be triggered without this read, since systemd-udev or reading partition table may do that too. > > BTW, you'd need to rebase this against current Linus' tree - Christoph has > moved this code to block/... OK. > > Also with the invalidate_lock held, there's no need for the second > truncate_bdev_range() call anymore. No pages can be created in the > discarded area while you are holding the invalidate_lock. Good catch, will remove the 2nd truncate_bdev_range() in V2. Thanks, Ming
