Bug 697144

It's not clear to me what bug you see; can you provide your mkfs.xfs commandline, its output, and the resulting xfs_info /mount/point output once it's mounted?

Thanks,
-Eric

mkfs.xfs sunit=128,swidth=1408


mkfs_info:

meta-data=/dev/mapper/xyratex_ld0-LD0 isize=256    agcount=32, agsize=166194432 blks
         =                       sectsz=512   attr=2
data     =                       bsize=4096   blocks=5318217728, imaxpct=5
         =                       sunit=32     swidth=352 blks
naming   =version 2              bsize=4096   ascii-ci=0
log      =internal               bsize=4096   blocks=521728, version=2
         =                       sectsz=512   sunit=8 blks, lazy-count=1
realtime =none                   extsz=4096   blocks=0, rtextents=0


any other values used for sunit & swidth and the results are always, I can see, smaller as if division by eight took place.

mkfs.xfs -d sunit=1024,swidth=11264 /dev/mapper/xyratex_ld1-LD1 -f
log stripe unit (524288 bytes) is too large (maximum is 256KiB)
log stripe unit adjusted to 32KiB
meta-data=/dev/mapper/xyratex_ld1-LD1 isize=256    agcount=20, agsize=268435328 blks
         =                       sectsz=512   attr=2
data     =                       bsize=4096   blocks=5318217728, imaxpct=5
         =                       sunit=128    swidth=1408 blks
naming   =version 2              bsize=4096   ascii-ci=0
log      =internal log           bsize=4096   blocks=521728, version=2
         =                       sectsz=512   sunit=8 blks, lazy-count=1
realtime =none                   extsz=4096   blocks=0, rtextents=0

but let say I factor this in and while create I use values respectively larger, then if I leave something like below in fstab:
sunit=256,swidth=2816

next time xfs_info again shows:
sunit=32     swidth=352 blks

The units used here are horrible and confusing.  at mkfs.xfs time, sunit/swidth are specified in multiples of 512 units.  So if you want 16k, you say 32.  IOW, you specify number of sectors.

But it's reported back to you in fs block units.  Hence the factor of 8 (4096/512).

It's all consistent, if you know that one ugly fact.

Here I say sunit=128, which means 128 sectors, or 16 4k blocks:

#  mkfs.xfs -dfile,name=fsfile,size=258m,sunit=128,swidth=1408
meta-data=fsfile                 isize=256    agcount=4, agsize=16512 blks
         =                       sectsz=512   attr=2
data     =                       bsize=4096   blocks=66048, imaxpct=25
         =                       sunit=16     swidth=176 blks
naming   =version 2              bsize=4096   ascii-ci=0
log      =internal log           bsize=4096   blocks=1200, version=2
         =                       sectsz=512   sunit=16 blks, lazy-count=1
realtime =none                   extsz=4096   blocks=0, rtextents=0

# mount -o loop fsfile  mnt/

# xfs_info mnt/
meta-data=/dev/loop0             isize=256    agcount=4, agsize=16512 blks
         =                       sectsz=512   attr=2
data     =                       bsize=4096   blocks=66048, imaxpct=25
         =                       sunit=16     swidth=176 blks
naming   =version 2              bsize=4096   ascii-ci=0
log      =internal               bsize=4096   blocks=1200, version=2
         =                       sectsz=512   sunit=16 blks, lazy-count=1
realtime =none                   extsz=4096   blocks=0, rtextents=0

in both cases I get back "16" for sunit... with that "blks" unit at the end of the line.

# rpm -q xfsprogs
xfsprogs-3.1.2-1.fc14.x86_64

The output is icky, but it is doing what was asked.

sunit/swidth take sectors; mkfs & xfs_info output blocks.

Using su/sw is sometimes more intuitive, as su simply takes bytes, and sw is a multiplier of su.

-Eric

ugh! it is totally misleading, either in the tool's output or in manuals it would certainly be GREAT if this was stressed out. yes, I've overlooked it, but if one ain't, somehow knowing beforehand, looking for it then easily overlooks it.

so, do we relay on mkfs.xfs and mount to auto detect fs params, or we should give it all at mount time or into fstab in order to have most optimal fs?

many! thanks Eric for explaining it all

Well, the mkfs.xfs manual is quite explicit in what units it takes.  Though I do have to look it up every time :)

The output does have the "blks" unit appended but that's not so obvious.

No need for mount-time options, if you've set it at mkfs time it's persistent.  Mounting with new options will change what's stored in the superblock, if geometry changes.

As for whether you should specify it at all or let mkfs sort it out, that depends somewhat on your storage.  For software raid, mkfs.xfs should just do the right thing.  For hardware, it depends on whether the hardware exports geometry.

-Eric

Celsiuses are Celsiuses, Fahrenheits are Fahrenheits, I believe you know what I mean.
Thanks again, I presume we close it as not-a-bug?

(In reply to comment #8)

> Thanks again, I presume we close it as not-a-bug?

Yes, I'll do that, thanks.