Re: Kernel thoughts of a Linux user

[Maneesh Soni]

On Mon, Nov 22, 2004 at 10:54:01PM +0800, Adam J. Richter wrote:
> I wrote:
> >        Please correct me if I am wrong, but, as far as I can tell,
> >in 2.6.10-rc2-bk6, a struct dentry is held for each node in the sysfs
> >tree at all times.  I infer this from noticing that sysfs_drop_dentry
> >and sysfs_hash_and_remove in fs/sysfs/inode.c only seem to be called
> >on operations to delete a node.  If I've missed something and the dentry
> >structures are all or mostly released, I would love to be corrected about
> >it as that would be really good news to me.
> 

The new code doesnot pin the leaf nodes but still pins (by having an extra 
ref count will creation) the non-leaf nodes or directory dentry and inode. 
The non-directory (leaf nodes) dentry and inodes are freed as per the normal 
VFS/dcache cache eviction policies. Dentries not in use, will be pruned under 
memory pressure and eventually evict inodes also.

> 	I should correct myself, although this correction suggests
> that sysfs currently uses slightly _more_ memory than I previously
> thought in the case of my computer (1100 directories and 2305
> non-directories in sysfs).
> 
> 	In 2.6.10-rc2-bk6, it looks like sysfs releases the dname
> structures as well in the case of a file (attribute) or symlink,
> but keeps these structures *and* a struct inode for every directory
> (kobject).  So, it looks like the non-swappable memory usage of my
> /sys is actually about 900kB.
> 
> 			directories		non-directories
> 	dentry		144			0
> 	inode		344			0
> 	sysfs_dirent	36			36
> 
> 	Bytes per:	524			36
> 	#of nodes:	1100			2305
> 	Subtotal:	576,400			82,980
> 
> 	Total:		659,380 bytes
> 
> 
> 	Perhaps the code that allows non-directories in sysfs to free
> their inode and dname structures will in the future be extended to allow
> directories do so also, which would reduce that total to 122kB.

Extending this to directories introduces lots of race conditions in 
maintaining the coherancy between VFS dentry tree and sysfs_dirent
based tree. It could be doable but not without complications.

Thanks
Maneesh

-- 
Maneesh Soni
Linux Technology Center, 
IBM Austin
email: maneesh@xxxxxxxxxx
Phone: 1-512-838-1896 Fax: 
T/L : 6781896
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/