Add a new mechanic to cgroup-v2 and implement the control of page-cache. Our linux version is 4.15.0 and the directory linux-source-4.15.0 is the corresponding source code tree.
To implement the new pagecache-limit mechanic, I modified the following files.
- include/linux/memcontrol.h
- include/linux/page_count.h
- mm/filemap.c
- mm/page_count
- mm/memcontrol.c
- mm/vmscan.c If you use other versions of linux, please replace these files.
You can install the new kernel following the instructions below:
# click Kernel hacking -> Complie-time checks -> compile the kernel with debug info
# generate drarf4 debuginfo
# provide GDB scripts for kernel debugging
make menuconfig
make -j8
make install
make modules_install
Then you can boot the new kernel through GRUB. Before you boot the new kernel, you shold firstly disable the cgroup-v1 to allow cgroup-v2 using all sub-system:
vim /etc/default/grub
# append systemd.unified_cgroup_hierarchy=1 to GRUB_CMDLINE_LINUX
update-grup
After you boot the new kernel:
cd /sys/fs/cgroup/unified
echo '+memory' >cgroup.subtree_control
mkdir test & cd test & ls
You will see two new user interface files which is using to catch the usage of pagecache and the limitation of pagecache and you can use them like other user interface files.
memory.pagecache_current
memory.pagecache_limit
Run the scripts run_pagecache_test.sh. The script use dd to copy a file of 500M, and it limit the pagecache to 100M. After copying, you can use cat memory.pagecache_current
to see how many bytes the dd process used. And you can also use free -h
to see increments in buffer/cache usage.
At the same time, we provide a simple program to test, test.c, you can replace dd cmd with ./test
to do more tests.