[Top] [All Lists]

Re: XFS filesystem structure document

To: Barry Naujok <bnaujok@xxxxxxxxxxxxxxxxx>
Subject: Re: XFS filesystem structure document
From: Stewart Smith <stewart@xxxxxxxxxxxxxxxx>
Date: Fri, 06 Oct 2006 20:49:41 +1000
Cc: xfs@xxxxxxxxxxx
In-reply-to: <200610050632.QAA03593@xxxxxxxxxxxxxxxxxxxxxxx>
References: <200610050632.QAA03593@xxxxxxxxxxxxxxxxxxxxxxx>
Sender: xfs-bounce@xxxxxxxxxxx
On Thu, 2006-10-05 at 16:38 +1000, Barry Naujok wrote:
> I have attached a PDF document describing the on-disk layout of the XFS
> filesystem for review. It's a first draft and some sections are still
> incomplete. When it is in a suitable state, it will be published on the
> oss.sgi.com website.

Heya Barry, long time no see!

Hope things are well.

a few comments and suggestions:

 perhaps expand introduction to have a bit about history of XFS and
explicitly state that this is on-disk format documentation

- state what endian everything is stored in (or if both, later describe

Common XFS Types
first time mentioning Allocation Groups, write as two words, instead of
perhaps describe (explicitly) that all xfs types are xfs_TYPE_t, but may
be structs.
also why some are xfs_dBLAH_t (i.e. what d stands for - or the
hysterical raisins).

Perhaps first it needs explaining of the idea of subvolumes? journal,
data and realtime? 

Allocation Groups
edit (add the and): "Immediately after a mkfs.xfs, the primary AG has
the following disk layout *AND* the subsequent AGs do not have any
inodes allocated:"

probably need to have quick 1 sentence explanation of interpretation of
sector and block.

may want to explicitly state that these structs don't have padding.
maybe sizeof() as well? so as to know how much "free" space in a block
with a struct of these? and what this "free" space should be initialized
to (if anything).

In list of fields with descriptions, would be also useful to list type
next to it (along with definition such as "unsigned 64bit int").

Mentions "real time device" when previously was described as sub volume.

perhaps explicitly mention of sb_logblocks is/isn't set for external

"The value must be 4 including the following flags:" except there's a
lot of flags there. perhaps needs to be more precise.

on really really really old file systems? Can't you also disable
unwritten extents with an mkfs option, or am i just dreaming (err...
nightmaring :) ?

Would be good to state if sb_inodesize can be bigger than sb_blocksize
and if so, what the value of sb_inopblock would be - or if in this case
the FS is not valid.

encoding of sb_fname, or if it's just a bunch of bytes and we don't

shouldn't log2 have the 2 as subscript not super?

for sb_logsectlog and sb_logsectsize, what should be values if internal

AG Free List
May need to more explictly state that the free list is always full, and
in what situations it isn't, and how that's represented on disk.

looks like a good start though!

(look forward to reading the journaling part especilaly)
Stewart Smith (stewart@xxxxxxxxxxxxxxxx)

Attachment: signature.asc
Description: This is a digitally signed message part

<Prev in Thread] Current Thread [Next in Thread>