Day 3 - LVM and Swap: PV·VG·LV Layers · From pvcreate to lvextend·resize2fs · mkswap·swapon

📌 Key Takeaways

• LVM inserts an abstraction layer between disks and volumes, separating logical usage from physical constraints.
• Three layers: PV (disk) → VG (pool) → LV (unit of use). Creation goes pvcreate → vgcreate → lvcreate.
• Extension is two steps — lvextend (the container) + resize2fs/xfs_growfs (the contents). If the VG is short on space, vgextend comes first.
• xfs cannot be shrunk, and xfs_growfs takes a mount point. ext4's resize2fs can grow online; shrinking requires unmounting.
• Swap is not a filesystem — initialize it with mkswap and turn it on with swapon; in fstab it is none swap sw 0 0.

The Three Layers of LVM: PV → VG → LV

With traditional partitions, once you fix the size, growing it is tricky. Even if the neighboring partition has spare room, you cannot borrow it unless it is physically adjacent. LVM breaks this rigidity by bundling physical disks into a single storage pool.

• Flow: turn a disk/partition into a PV with pvcreate → bundle several PVs into a VG with vgcreate → cut LVs from the VG with lvcreate → format the LV with mkfs and mount it for use.

# Full creation flow (example using disks sdb, sdc)
pvcreate /dev/sdb1 /dev/sdc1        # 1) Make the two partitions PVs
vgcreate myvg /dev/sdb1 /dev/sdc1   # 2) Bundle them into a VG named myvg
lvcreate -L 10G -n mylv myvg        # 3) Create a 10G LV 'mylv' from myvg
mkfs.ext4 /dev/myvg/mylv            # 4) Format
mount /dev/myvg/mylv /mnt/data      # 5) Mount

An LV's device path appears in two forms — /dev/<VG name>/<LV name> (e.g., /dev/myvg/mylv) and /dev/mapper/<VG>-<LV> (e.g., /dev/mapper/myvg-mylv). They are aliases for the same thing. You can use either in fstab, but an LV also has its own unique UUID that can be checked with blkid.

A VG manages space by slicing it into fixed-size chunks called PE (Physical Extent) (default 4MB). An LV's size is in fact an integer multiple of PEs. Using lowercase -l as in lvcreate -l 100 myvg specifies "number of PEs", while uppercase -L 400M specifies "capacity". Thanks to the PE concept, an LV looks like one contiguous volume even when scattered across several PVs.

The power of LVM is that it handles capacity "without turning the disk off". When /var on a running server fills up, you add a new disk as a PV to grow the VG (vgextend), extend the LV (lvextend), and then extend the filesystem — and space grows with no service interruption. This flexibility, impossible with traditional partitions, is why LVM became the standard on servers.

Extending an LV: lvextend and resize2fs/xfs_growfs

The representative scenario is extending an LV. The key is that it splits into two steps.

1. Grow the LV (the container) itself — lvextend
2. Grow the filesystem on top of it — resize2fs (ext) or xfs_growfs (xfs)

# Extend an LV by another 5G when the VG has spare room
lvextend -L +5G /dev/myvg/mylv     # Add +5G (absolute value is -L 15G)
resize2fs /dev/myvg/mylv           # Fit the ext4 filesystem to the LV size
# For xfs:
xfs_growfs /mnt/data               # (specify the mount point, online growth)
 
# Both steps at once (ext only, -r = also resize the fs):
lvextend -r -L +5G /dev/myvg/mylv

Even if you grow the LV, the filesystem on it does not grow automatically. lvextend only enlarges the "container"; the "contents (filesystem)" inside must be grown separately with resize2fs/xfs_growfs. The trap is skipping this second step and ending up with "the LV grew but df shows the same capacity". Exam: "What command reflects the capacity in ext4 after lvextend?" → resize2fs.

⚠️ Pitfall: resize2fs (ext4) can grow online (while mounted), but shrinking works only after unmounting. xfs, on the other hand, cannot be shrunk at all (xfs_growfs is grow-only). So "How do you shrink an xfs LV?" → The only way is to back up and recreate. The grow arguments also differ — resize2fs takes the device path, while xfs_growfs takes the mount point.

If the VG has no free space, you must grow the VG before extending the LV. After making a new disk a PV with pvcreate, incorporate it into the VG with vgextend myvg /dev/sdd1 and the VG's available PEs increase. Check the free room with "Free PE / Size" in vgdisplay myvg. Overall order: pvcreate → vgextend → lvextend → resize2fs.

# Try it yourself — three ways to inspect LVM state
pvs / pvdisplay        # PV list / details
vgs / vgdisplay        # VG list / details (check Free PE)
lvs / lvdisplay        # LV list / details
lsblk                  # Disk-PV-LV relationship tree

LVM Commands at a Glance

Major commands per layer in a table — for short-answer prep.

⚠️ Pitfall: Do not confuse -L (uppercase, capacity: -L 10G) and -l (lowercase, PE count/percentage: -l 100 or -l 100%FREE) in lvcreate. "Create an LV using all the remaining space of the VG" is lvcreate -l 100%FREE -n mylv myvg. And -n is the option that specifies the LV name.

An LVM snapshot (lvcreate -s) "freezes" an LV at a point in time to guarantee consistency during a backup. It stores only the changes separately with CoW (Copy-on-Write), so it uses almost no space at first. Even if the original changes during the backup, the snapshot preserves that moment.

Swap — Turning Disk into Virtual Memory

• Swap is an area that, when physical memory (RAM) runs short, temporarily moves memory pages that are not in use to disk (swap out) to free up RAM.
• Swap is not a filesystem — you do not mount it; you initialize it with a dedicated format (mkswap) and register it with the kernel (swapon) to use it.
• Swap partition: a dedicated partition with partition type 82
• Swap file: using a regular file as swap (flexible but slightly slower)

# Creating a swap partition
mkswap /dev/sdb2           # Initialize the partition as a swap area
swapon /dev/sdb2           # Activate
swapoff /dev/sdb2          # Deactivate
 
# Creating a swap file (2GB)
dd if=/dev/zero of=/swapfile bs=1M count=2048   # or fallocate -l 2G /swapfile
chmod 600 /swapfile        # 600 recommended for security
mkswap /swapfile
swapon /swapfile
 
# Check status
swapon -s          # or cat /proc/swaps
free -h            # Check total/used in the Swap row

mkswap does not create a filesystem; it is a separate operation that writes a swap signature and header. That is why swap is turned on with swapon, not mount. "What is the command to initialize a swap area?" → mkswap, "To activate it?" → swapon are the standard short answers.

To activate swap automatically at boot, register it in fstab. But the format differs from a regular filesystem — none (or swap) in the mount point slot, swap for the type, sw for the options, and pass and dump are 0:

/dev/sdb2  none  swap  sw  0  0
/swapfile  none  swap  sw  0  0

After registering, activate all swaps in fstab in bulk with swapon -a.

How aggressively the kernel uses swap is controlled by vm.swappiness (0 to 100, default 60). A low value uses RAM as much as possible and spares swap, while a high value pushes to swap early. Change it temporarily with sysctl vm.swappiness=10, and make it permanent by writing it in /etc/sysctl.conf. Database servers often set swappiness low because performance plunges once swapping occurs.

Swap size was once orthodoxly "2x RAM", but that has lost meaning on modern systems with tens of GB of RAM. However, to use hibernation (hibernate, saving all of RAM to disk), swap must be at least as large as RAM — because it has to hold the entire RAM contents. This exception is worth knowing.

# Try it yourself — full swap flow
free -h                    # Current memory/swap
swapon -s                  # List of active swap
fallocate -l 1G /swapfile && chmod 600 /swapfile
mkswap /swapfile && swapon /swapfile
swapon -s                  # Was /swapfile added?
swapoff /swapfile          # Clean up

Next, Day 4 covers parity by RAID level and mdadm, disk quotas, and the inspection tools fsck·df·du·lsblk·blkid.

📖 Glossary

• LVM : A logical volume manager that inserts an abstraction layer between disks and volumes to enable non-disruptive growth.
• PV (Physical Volume) : A disk or partition incorporated into LVM. Corresponds to the bottles poured into the pool.
• VG (Volume Group) : A storage pool bundling PVs. Space is carved from it to make LVs.
• LV (Logical Volume) : The actual unit of use cut from a VG. You format it and mount it to use.
• PE (Physical Extent) : The fixed-size chunk (default 4MB) in which a VG manages space. An LV's size is an integer multiple of PEs.
• -L vs -l : Specify by capacity (-L 10G) versus by PE count or percentage (-l 100%FREE).
• /dev/mapper/VG-LV : Another path notation pointing to the same target as /dev/VG/LV.
• lvextend + resize2fs : The step that grows the container and the step that grows the filesystem inside it. You need both for df to reflect it.
• xfs_growfs : The xfs-only grow command. Takes the mount point as an argument, and shrinking is impossible.
• vgextend : The command that adds a new PV to a VG to grow the pool. If there is no spare room before extending an LV, this comes first.
• LVM snapshot : A feature that freezes a point in time with lvcreate -s. Being CoW, only the changes take up space.
• Swap : A disk area that evacuates unused pages when RAM is short. Not a filesystem.
• mkswap / swapon / swapoff : Initialize / activate / deactivate a swap area.
• vm.swappiness : A value (0 to 100, default 60) that decides how aggressively the kernel uses swap.
• Hibernation : A feature that saves all RAM contents to disk. That is why swap must be at least as large as RAM.