On Sat, Mar 26, 2022 at 4:29 AM Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote: > > On Fri, Mar 25, 2022 at 2:52 AM Muchun Song <songmuchun@xxxxxxxxxxxxx> wrote: > > > > We can see that we put the dentry (ffff88807ebda0f8) into > > the list_lru (ffff888011bd47f0). But we do not allocate struct > > list_lru_one for the memcg (ffff88801c530000). Then it panics. > > Hmm. > > Looking at memcg_slab_pre_alloc_hook(), I note that it will return > success without doing the LRU checking for several cases. > > So since you can reproduce the problem, I would suggest you add some > debug code to __d_alloc() that prints out something big if it gets a > dentry but you can't look up the list_lru_one() for that dentry. > > Hmm? > > The only other situation I can think of is if dentry->d_sb were to > change during the dentry lifetime, but I don't think that can happen. > The only assignment I can find with "git grep" is that > > dentry->d_sb = sb; > > in __d_alloc(), and while it's possible my grep pattern was bogus, it > sounds unlikely. > I have found the root cause, it was caused by kfence. Here is the fix patch [1]. [1] https://lore.kernel.org/all/20220327051853.57647-1-songmuchun@xxxxxxxxxxxxx/ Thanks.
