On Wed, Mar 24, 2021 at 09:34:58AM +0100, lukas@xxxxxxxxxxx wrote: > Hi, > I came across chapter in XFS documentation "12.4.1 xfs_db AGI Example" > quoting bellow: > --- > recs[1-85] = [startino,freecount,free]1:[96,0,0] 2:[160,0,0] 3:[224,0,0] > 4:[288,0,0] > 5:[352,0,0] 6:[416,0,0] 7:[480,0,0] > 8:[544,0,0] > 9:[608,0,0] 10:[672,0,0] 11:[736,0,0] > 12:[800,0,0] > ... > 85:[5792,9,0xff80000000000000] > > Most of the inode chunks on this filesystem are totally full, since the free > value is zero. > This means that we ought to expect inode 160 to be linked somewhere in the > directory structure. > However, notice that 0xff80000000000000 in record 85 — this means that we > would expect inode 5856 > to be free. Moving on to the free inode B+tree, we see that this is indeed > the case: > --- > > As there are 9 inodes free in the last chunk of 64 inodes it gives me first > free inode 5847 (5792+55), > on the other hand inode 5856 is also free as it's last inode in the chunk. > > My question is do I understand correctly that the first free inode in that > AG is 5847? Oops, yes, you are correct. The first free inode is 5847, not 5856 as stated in 12.4.1. > Thanks, bellow possible patch. It looks reasonable, but the patch needs to have a Signed-off-by tag with your name and email (per kernel patch submission rules[1]) before I can proceed with review. [1] https://www.kernel.org/doc/html/latest/process/submitting-patches.html#sign-your-work-the-developer-s-certificate-of-origin > --- > diff --git a/design/XFS_Filesystem_Structure/allocation_groups.asciidoc > b/design/XFS_Filesystem_Structure/allocation_groups.asciidoc > index 992615d..cdc8545 100644 > --- a/design/XFS_Filesystem_Structure/allocation_groups.asciidoc > +++ b/design/XFS_Filesystem_Structure/allocation_groups.asciidoc > @@ -1099,7 +1099,7 @@ recs[1-85] = [startino,freecount,free] > Most of the inode chunks on this filesystem are totally full, since the > +free+ > value is zero. This means that we ought to expect inode 160 to be linked > somewhere in the directory structure. However, notice that > 0xff80000000000000 This patch seems to have been line wrapped. Would you mind resending it with the DCO tag I mentioned above, and without wrapping? Or as a text attachment if that's the only option? --D > -in record 85 -- this means that we would expect inode 5856 to be free. > Moving > +in record 85 -- this means that we would expect inode 5847 to be free. > Moving > on to the free inode B+tree, we see that this is indeed the case: > --- > > -- > Lukas Herbolt
