Home / ubuntu / Questions / 1361461
Accepted
FalcoGer
Asked: 2021-09-02 18:42:54 +0800 CST

Running cachefilesd cache on ramdisk (tmpfs) for PXE NFS boot

  • 772

I am running ubuntu 20.04 LTS (kernel 5.4.0-81-generic) from PXE and I wish to add a local cache. Because I don't want to have a partition on all my devices for caching, I wish to use a ramdisk for that purpose.

I have managed to enable the fsc option during boot with the help of this answer.

However when I run the cachefilesd service, it failes with -95: Operation not supported. This is probably because tmpfs doesn't support user_xattr, which is required.

I have read here that you can compile/patch the kernel to enable user_xattr, though this is a security risk. Besides being a maintenance nightmare, I do not wish to go that far and fiddle around with automating that with every kernel update.

I have read here (german) and this and it's response.

It indicates that you can make a tmpfs, then make a loopback device, format that to ext4 and then run your fs-cache on there.

However, since I'm using PXE and the NFS mount point is /, how do I do that during boot, preferably before cachedfilesd.service is started?

My bootloader (vesamenu.c32)

# [...]
label                   ubuntunfs
        menu label      ^Ubuntu NFS
        text help
Run ubuntu from Network File System.
        endtext
        kernel          ubuntu/boot/vmlinuz
        initrd          ubuntu/boot/initrd.img
        append          root=/dev/nfs nfsroot=192.168.1.5:/srv/tftp/ubuntu,rw,fsc,relatime ip=dhcp rw
# [...]

My fstab

/proc    /proc    proc    defaults   0 0
/sys     /sys     sysfs   defaults   0 0
/dev/nfs /        nfs     defaults,relatime,fsc   1 1

# ramdisk for fs-cache for nfs
tmpfs    /media/ramdisk    tmpfs    defaults,size=5%      0       0

none            /tmp            tmpfs   defaults        0       0
none            /var/run        tmpfs   defaults        0       0
none            /var/lock       tmpfs   defaults        0       0
none            /var/tmp        tmpfs   defaults        0       0

my /etc/cachefilesd.conf

dir /media/ramdisk
tag mycache
brun 90%
bcull 70%
bstop 30%
frun 90%
fcull 70%
fstop 30%

culltable 14
boot

1 Answers

  1. Best Answer

    I did it with a systemd service.

    [Unit]
Description=Cache creator
Before=cachefilesd.service

[Service]
ExecStart=/bin/bash -c "dd if=/dev/zero of=/media/ramdisk/cache.ext4 bs=1k count=`df /media/ramdisk | tail -n 1 | sed -e 's/ \+/ /g' | cut -d ' ' -f 2` && mke2fs /media/ramdisk/cache.ext4 && mount -o rw,noatime,nodiratime,user_xattr /media/ramdisk/cache.ext4 /media/cache/"
Restart=no
Type=oneshot

[Install]
WantedBy=multi-user.target

    The script is taking the 1k block count (from df) of the ramdisk as the input for dd's input file (if=) argument, otherwise dd will return an error code because the disk is full and && will not work in that situation. Obviously this only works if ramdisk is empty.

    • 0