Re: [PATCH] minor optimization for EXT3

[Alex Tomas]

>>>>> Andi Kleen (AK) writes:

 AK> Alex Tomas <[e-mail]> writes:
 >> +			if (i == start + inodes_per_buffer) {
 >> +				/* all inodes (but our) are free. so, we skip I/O */

 AK> Won't this make undeletion a lot harder? Deleted inodes will now be trashed
 AK> at will, so you cannot use their contents anymore. 

AFAIK ext3 doesn't support undeletion at all

 AK> Also dtimes in free inodes  can be now lost, can't they? Did you check 
 AK> if that causes problems in fsck?  [my understanding was that ext2/3 fsck relies on the 
 AK> dtime to make some heuristics when recovering files work better]

freed inodes will be lost. I've checked filesystem by fsck after lots of creations/removals.
it seems OK.

 AK> Maybe it should be an mount option so that users can trade performance against
 AK> better recoverability.

well, I'm not sure we really need it