On Jan 8, 2015, at 7:09 AM, Phillip Susi <psusi@xxxxxxxxxx> wrote: > On 1/7/2015 11:58 PM, Andreas Dilger wrote: >>> No, it knows that the inode table needs initialized because there >>> is a flag in the group descriptor that says this inode table is >>> still uninitalized. It never reads the blocks to see if they are >>> full of zeros. mke2fs sets the flag when it does not initialize >>> the table with zeros, either by direct writes ( which it doesn't >>> do if lazy_itable_init is true, which it defaults to these days >>> ), or by discarding the blocks when the device claims to support >>> deterministic discard that zeros. >> >> That is only partially correct. While it is true that mke2fs sets >> the UNINIT flag at format time, the "lazy" part of that means there >> is a kernel thread still does the zeroing of the inode table >> blocks, but after the filesystem is mounted, for any group that >> does not have the ZEROED flag set. After that point, the "UNINIT" >> flag is an optimization to avoid reading the bitmap and unused >> blocks from disk during allocation. > > That is pretty much what I said, except that I was pointing out that > it does not *read* first to see if the disk is already zeroed, as that > would be a waste of time. It just writes out the zeros for block > groups that still have the uninit flag set. Sorry, I didn't get that from my reading, so I thought I'd clarify. I'd actually proposed that the ext4_init_inode_table() thread start by reading the itable blocks first, check them for zeroes, and only switch over to writing if it finds any non-zero data in the blocks. I think that would be a net win in some cases, and only a tiny bit of overhead (a single read) if it turns out to be wrong. >> This is needed in case the group descriptor or inode bitmap is >> corrupted, and e2fsck needs to scan the inode table for in-use >> inodes. We don't want it to find old inodes from before the >> filesystem was formatted. >> >> The ext4_init_inode_table() calls >> sb_issue_zeroout->blkdev_issue_zeroout(), so if the underlying >> storage supported deterministic zeroing of the underlying storage, >> this could be handled very efficiently. > > Again, that's pretty much what I said. Cheers, Andreas -- To unsubscribe from this list: send the line "unsubscribe linux-ide" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
