On 08/04/19 14:03, Jan Kara wrote: <> > Hum, you are right. Ext4 is buggy in this regard. I've fixed the race for > page fault in ea3d7209ca01 "ext4: fix races between page faults and hole > punching" but didn't realize the problem is there for buffered reads as > well. I'll think how we can fix this. Thanks for noticing this! > This is very interesting. Please CC me if you have devised a test for this? For a long time I want to enhance fsstress --verify that each writer always writes the long address it writes to as data (ie. file looks like an increasing long counter) And readers want to see only this pattern or zero. > Honza > Thanks Boaz
