On Tue, Jun 11, 2024 at 01:05:05PM +0200, Jan Kara wrote: > On Tue 11-06-24 12:23:59, Mateusz Guzik wrote: > > On Tue, Jun 11, 2024 at 12:02:22PM +0200, Jan Kara wrote: > > > On Tue 11-06-24 06:15:40, Mateusz Guzik wrote: > > > > new_inode used to have the following: > > > > spin_lock(&inode_lock); > > > > inodes_stat.nr_inodes++; > > > > list_add(&inode->i_list, &inode_in_use); > > > > list_add(&inode->i_sb_list, &sb->s_inodes); > > > > inode->i_ino = ++last_ino; > > > > inode->i_state = 0; > > > > spin_unlock(&inode_lock); > > > > > > > > over time things disappeared, got moved around or got replaced (global > > > > inode lock with a per-inode lock), eventually this got reduced to: > > > > spin_lock(&inode->i_lock); > > > > inode->i_state = 0; > > > > spin_unlock(&inode->i_lock); > > > > > > > > But the lock acquire here does not synchronize against anyone. > > > > > > > > Additionally iget5_locked performs i_state = 0 assignment without any > > > > locks to begin with and the two combined look confusing at best. > > > > > > > > It looks like the current state is a leftover which was not cleaned up. > > > > > > > > Ideally it would be an invariant that i_state == 0 to begin with, but > > > > achieving that would require dealing with all filesystem alloc handlers > > > > one by one. > > > > > > > > In the meantime drop the misleading locking and move i_state zeroing to > > > > alloc_inode so that others don't need to deal with it by hand. > > > > > > > > Signed-off-by: Mateusz Guzik <mjguzik@xxxxxxxxx> > > > > > > Good point. But the initialization would seem more natural in > > > inode_init_always(), wouldn't it? And that will also address your "FIXME" > > > comment. > > > > > > > My point is that by the time the inode is destroyed some of the fields > > like i_state should be set to a well-known value, this one preferably > > plain 0. > > Well, i_state is set to a more or less well defined value but it is not > zero. I don't see a performance difference in whether set it to 0 on > freeing or on allocation and on allocation it is actually much easier to > find when reading the code. > I was thinking more about assertion potential, not anything perf-related, but it is a moot subject now. > > I did not patch inode_init_always because it is exported and xfs uses it > > in 2 spots, only one of which zeroing the thing immediately after. > > Another one is a little more involved, it probably would not be a > > problem as the value is set altered later anyway, but I don't want to > > mess with semantics of the func if it can be easily avoided. > > Well, I'd consider that as another good reason to actually clean this up. > Look, inode_init_always() is used in bcachefs and xfs. bcachefs sets > i_state to 0 just before calling inode_init_always(), xfs just after one > call of inode_init_always() and the call in xfs_reinit_inode() is used > only from xfs_iget_recycle() which sets i_state to I_NEW. So I claim that > moving i_state clearing to inode_init_always() will not cause any issue and > is actually desirable. > Ok, see my reply to Dave's e-mail. Just tell me how to ship this and I'll do the needful(tm).
