RHEL: XFS basic operations -

RHEL: XFS basic operations

# Tested on RHEL 6 & 7

# XFS filesystem supports up to 16 EiB filesystems and up to 8 EiB and directory structures
# holding tens of millions entries.
# On the other hand it is a single node filesystem. It comes integrated in RHEL 7, in order
# to use it on RHEL 6 we need a subscription to the Scalable File System Add-ON.

# Although XFS file system supports up to 16 EiB, Red Hat only supports filesystems up to
# 100 TiB

# Other limitations of XFS is that it's less suited for single threads creating and deleting
# a large number of small files; in addition it uses about twice CPU resources that ext4 so
# under those circumstances it is better to use ext4 filesystems.

# In general XFS is best suited for large systems with fast storage.

# As already indicated, to use XFS filesystems we need the "Scalable Filesystems" group /

# 'xfsprogs' package

# Create an xfs filesystem
# ------------------------------------------------------------------------------------------

mkfs.xfs /dev/sdd2

# If we know storage parameters, we can specify chunk size ('su', stripe unit) and/or
# stripe width ('sw') in order to improve xfs fs performance. For example, to create an xfs
# file system with a stripe-unit size of 32 KB and 4 units per stripe, we will specify:

mkfs.xfs -d su=32k,sw=4 /dev/sdd2
   meta-data=/dev/sdd2              isize=256    agcount=4, agsize=655352 blks
            =                       sectsz=512   attr=2
   data     =                       bsize=4096   blocks=2621408, imaxpct=25
            =                       sunit=8      swidth=32 blks
   naming   =version 2              bsize=4096   ascii-ci=0
   log      =internal log           bsize=4096   blocks=2560, version=2
            =                       sectsz=512   sunit=8 blks, lazy-count=1
   realtime =none                   extsz=4096   blocks=0, rtextents=0

# External journal
# ------------------------------------------------------------------------------------------

# By default xfs store journal internally. As synchronous metadata writes to the journal
# must complete successfully before any associated data such a layout can lead to disk
# contention. To improve performance we can consider placing journal on a separate physical
# device. To create an external journal, use the "-l logdev=device,size=size" option during

# xfs creation. If we omit the size parameter, mkfs.xfs selects a size based on the size
# of the file system.

mkfs.xfs -l logdev=/dev/sde2 /dev/sdd2

# Mount an xfs F.S. without/with external journal
# ------------------------------------------------------------------------------------------

# by editing /etc/fstab

   /dev/datavg/lv_xfsdata     /myxfs      xfs   defaults        0 0

# or

   /dev/datavg/lv_xfsdata     /myxfs      xfs   logdev=/dev/datavg/lvxfsjournal     0 0

# via 'mount' command

mount /dev/datavg/lv_xfsdata /myxfs

# or

mount -o logdev=/dev/datavg/lv_xfsjournal /dev/datavg/lv_xfsdata   /myxfs

# Grow an xfs filesystem
# ------------------------------------------------------------------------------------------

# 1.- We CANNOT grow an unmounted xfs
# 2.- An xfs filesystem CANNOT be shrunk

# We can increase the size of a XFS file system if there is enough space on the underlying
# device. If necessary, increase the size of the logical volume (or disk partition or LUN
# and make changes visible to the system).

df -h /myxfs
   Filesystem            Size Used Avail Use% Mounted on
   /dev/sdf1             2.0G   33M 2.0G   2% /myxfs

# To specify the final size of the xfs, we use the '-D' option. The size is expressed in

# filesystem blocks. With a default block size of 4096 bytes, to grow our xfs up to 3 GiB:

# 3 GiB = 3221225472 byte = 786432 blocks

xfs_growfs -D 786432 /myxfs
   meta-data=/dev/sdf1              isize=256    agcount=4, agsize=131530 blks
            =                       sectsz=512   attr=2
   data     =                       bsize=4096   blocks=526120, imaxpct=25
            =                       sunit=0      swidth=0 blks
   naming   =version 2              bsize=4096   ascii-ci=0
   log      =internal               bsize=4096   blocks=2560, version=2
            =                       sectsz=512   sunit=0 blks, lazy-count=1
   realtime =none                   extsz=4096   blocks=0, rtextents=0
   data blocks changed from 526120 to 786432

df -h /myxfs
   Filesystem            Size Used Avail Use% Mounted on
   /dev/sdf1             3.0G   33M 3.0G   2% /myxfs

# To grow the xfs to the largest size possible:

xfs_growfs -d /myxfs

df -h /myxfs
   Filesystem            Size Used Avail Use% Mounted on
   /dev/sdf1              10G   33M   10G   1% /myxfs

# Reduce an xfs filesystem
# ------------------------------------------------------------------------------------------

# Unfortunately, it is not possible to reduce an xfs filesystem

# Repair an xfs when file system is not cleanly unmounted, ...
# ------------------------------------------------------------------------------------------

xfs_repair [ -l <logdev> ] /dev/sdd2

# An xfs file system with a dirty log cannot be repaired. To clear out the log you can
# mount, then unmount. If this fails try '-L' option to xfs_repair to force clear the
# log (as a last resort as it may result in a data corruption)

# Display/modify label and UUID of an xfs
# ------------------------------------------------------------------------------------------

# Display existing label:

xfs_admin -l /dev/sde
   label = ""

# Set a new label (filesystem has to be unmounted):

xfs_admin -L "NewLabel" /dev/sde
   writing all SBs
   new label = "NewLabel"

# Display existing UUID:

xfs_admin -u /dev/sde
   UUID = 51b11165-c59d-44a6-8f4e-3616aaf79a4d

# Generate a new UUID (filesystem has to be unmounted):

xfs_admin -U generate /dev/sde
   Clearing log and setting UUID
   writing all SBs
   new UUID = a05ff818-dc74-4d59-aff8-92b360c2a2ed

# Clear the UUID (filesystem has to be unmounted):

xfs_admin -U nil /dev/sde
   Clearing log and setting UUID
   writing all SBs
   new UUID = 00000000-0000-0000-0000-000000000000

# Let's see what happens if we try to mount an xfs with a nil UUID:

mount /dev/sde /xfs01
   mount: wrong fs type, bad option, bad superblock on /dev/sde,
      missing codepage or helper program, or other error
      In some cases useful info is found in syslog - try
      dmesg | tail or so

dmesg | tail -1
   XFS (sde): Filesystem has nil UUID - can't mount

# Defragmenting an xfs
# ------------------------------------------------------------------------------------------

xfs_fsr /dev/sdd2

# To defragment a single file run xfs_fsr <path_to_file>

# If no option is given to 'xfs_fsr' command it will defragment all the xfs on the server.
# Since this can potentially be a very long running operation the 'xfs_fsr' tool will
# stop after a number of seconds specified with the '-t' option (by default 7200 seconds,
# this is 2 hours)