EROFS
Read-only file system for Linux
From Wikipedia, the free encyclopedia
EROFS (Enhanced Read-Only File System) is a lightweight read-only file system initially developed by Huawei, originally for the Linux kernel and now maintained by an open-source community.
| Developer(s) | Huawei originally, Alibaba Cloud, Bytedance, Coolpad, Google, OPPO, Red Hat, Shanghai Jiao Tong University, VIVO, Uniontech, South China University of Technology |
|---|---|
| Full name | Enhanced Read-Only File System |
| Introduced | November 24, 2019 with Linux 5.4 |
| Limits | |
| Max volume size | 1 EiB since 6.15 |
| Max file size | |
| Max no. of files | Depends on volume size |
| Max filename length | 255 bytes |
| Features | |
| Dates recorded | File change time (extended only)[1] |
| Date resolution | 1 ns |
| Attributes | POSIX, Extended file attributes |
| File system permissions | POSIX, ACL |
| Transparent compression | Yes (LZ4; LZMA since 5.16; DEFLATE since 6.6)[1] |
| Data deduplication | Yes (extent-based) |
| Other | |
| Supported operating systems | Linux |
EROFS aims to form a generic read-only file system solution for various read-only use cases (embedded devices, containers and more) instead of just focusing on storage space saving without considering any side effects of runtime performance.[1]
For example, it provides a solution to save storage space by using transparent compression as an option for scenarios that need high-performance read-only requirements on their devices with limited hardware resources, e.g. smartphones like Android and IoT operating systems such as HarmonyOS alongside its HarmonyOS NEXT core system iteration.[2][3] All of Huawei's new products shipped with EMUI 9.0.1 or later used EROFS,[4] and it was promoted as one of the key features of EMUI 9.1.[5] Oppo, Xiaomi and some Samsung products also use EROFS.[6][7]
Also, it provides a content-addressable chunk-based container image solution together with lazy pulling feature to accelerate container startup speed by using new file-based fscache backend since Linux kernel v5.19 and file-backed mounts since v6.12.[8][9]
The file system was formally merged into the mainline kernel with Linux kernel v5.4.[10]
Features
The file system has two different inode on-disk layouts. One is compact, and the other is extended.[1]
- Little-endian on-disk design[1]
- 48-bit block addressing, which currently limits the total possible capacity of an EROFS filesystem to 1 EiB of 4 KiB block size.
- Metadata and data can be mixed by design for on-disk flexibility together with tail-packing inline data technology[1]
- Support POSIX attributes and permissions, extended file attributes and ACL[1]
- Fixed-size output transparent compression with LZ4, MicroLZMA (since Linux 5.16[11]), zstd and/or DEFLATE (since Linux 6.6[12]) for relative higher compression ratios[1]
- In-place decompression for higher sequential read [13][14]
- Big pcluster feature allowing up to 1 MiB big pclusters for better compression ratios since Linux 5.13.[15][16]
- Direct I/O, Direct Access (DAX) support, chunk-based data deduplication for uncompressed files since Linux 5.15.[1][17]
- Multiple device support for multiple layer container images since Linux 5.16.[1]
- Ztailpacking support since Linux 5.17.[18]
- File-based Fscache backend support since Linux 5.19 with "on-demand mode".[19][20]
- Rolling hash deduplicated data compression and fragment support since Linux 6.1. [1]