Re: [RESEND PATCH 2/2] ext4: ext4_inode_is_fast_symlink should use cluster size

[Theodore Ts'o <tytso@mit.edu>]

On Mon, Jan 06, 2014 at 10:47:03AM -0700, Andreas Dilger wrote:
> What about the idea to stop using the blocks count for doing the
> fast/slow symlink check, and instead use i_size?  IMHO this is far
> more robust than using the blocks count, since we've repeatedly seen
> bugs when the number of blocks allocated to an inode changes for
> done reason (e.g. xattrs, bigalloc, multi-block xattrs in the
> future).

I did see your earlier proposal on this front, but I didn't want to
this change without thinking about it a bit more closely.  In
particular, we probably would want to enforce this change in e2fsck
for a while first.

Currently, if we have a slow symlink where i_size is less than 60
bytes, both e2fsprogs and the kernel handles this case.  See the
attached file system image.

Yes, I created it synthetically, but keep in mind that that there are
other implementations of ext2/3/4 other than just in the Linux kernel.
In particular, the GNU Hurd and *BSD have their own independent
implementation of ext2.  So even if the Linux kernel has never created
a slow symlink with i_size <= 60 bytes, but that doesn't mean that
it's for certain that there are no such implementations out there in the wild.

That doesn't mean that we should never make such a change, but it does
mean that it's not something I want to do lightly.

							- Ted