Re: [PATCH 12/15] driver core: Implement tagged directory support for device classes.

Tejun Heo [off-list ref] writes:

It's a bit scary tho.  Working inode->i_dentry or dentry->d_alias
crosses multiple sb's.  sysfs isn't too greedy about dcache/icache.
Only open files and directories hold them and only single copy of
sysfs_dirent is there for most nodes.  Wouldn't it be better to stay on
the safer side and use separate inode hierarchy?

To do that I believe we would need to ensure sysfs does not use 
the inode->i_mutex lock except to keep the VFS layer out.  Allowing us
to safely change the directory structure, without holding it.

You raise a good point about inode->i_dentry and dentry->d_alias.
Generally they are used by fat like filesystems but I am starting to
see uses in generic pieces of code.  I don't see any problems today
but yes it would be good to do the refactoring to allow us to duplicate
the inodes. 

Eric

Hello, Eric.

Eric W. Biederman wrote:

Tejun Heo [off-list ref] writes:

quoted

It's a bit scary tho.  Working inode->i_dentry or dentry->d_alias
crosses multiple sb's.  sysfs isn't too greedy about dcache/icache.
Only open files and directories hold them and only single copy of
sysfs_dirent is there for most nodes.  Wouldn't it be better to stay on
the safer side and use separate inode hierarchy?

To do that I believe we would need to ensure sysfs does not use 
the inode->i_mutex lock except to keep the VFS layer out.  Allowing us
to safely change the directory structure, without holding it.

I don't think sysfs is depending on i_mutex anymore but I need to go
through the code to make sure.

You raise a good point about inode->i_dentry and dentry->d_alias.
Generally they are used by fat like filesystems but I am starting to
see uses in generic pieces of code.  I don't see any problems today
but yes it would be good to do the refactoring to allow us to duplicate
the inodes. 

Yeah, I can't spot any place which can cause actual problem yet but it's
still scary as we're breaking a vfs assumption and even if it's not a
problem now, future seemingly unrelated changes can break things subtly.

Thanks.

-- 
tejun

Tejun Heo wrote:

Hello, Eric.

Eric W. Biederman wrote:

quoted

Tejun Heo [off-list ref] writes:

quoted

It's a bit scary tho.  Working inode->i_dentry or dentry->d_alias
crosses multiple sb's.  sysfs isn't too greedy about dcache/icache.
Only open files and directories hold them and only single copy of
sysfs_dirent is there for most nodes.  Wouldn't it be better to stay on
the safer side and use separate inode hierarchy?

To do that I believe we would need to ensure sysfs does not use 
the inode->i_mutex lock except to keep the VFS layer out.  Allowing us
to safely change the directory structure, without holding it.

I don't think sysfs is depending on i_mutex anymore but I need to go
through the code to make sure.

quoted

You raise a good point about inode->i_dentry and dentry->d_alias.
Generally they are used by fat like filesystems but I am starting to
see uses in generic pieces of code.  I don't see any problems today
but yes it would be good to do the refactoring to allow us to duplicate
the inodes. 

Yeah, I can't spot any place which can cause actual problem yet but it's
still scary as we're breaking a vfs assumption and even if it's not a
problem now, future seemingly unrelated changes can break things subtly.

Okay, one small problem spotted.  It seems invalidate_inodes() can fail
which will make generic_shutdown_super() complain.  It's not a fatal
failure tho.

-- 
tejun