1.. SPDX-License-Identifier: GPL-2.0 2 3====================================== 4Enhanced Read-Only File System - EROFS 5====================================== 6 7Overview 8======== 9 10EROFS file-system stands for Enhanced Read-Only File System. Different 11from other read-only file systems, it aims to be designed for flexibility, 12scalability, but be kept simple and high performance. 13 14It is designed as a better filesystem solution for the following scenarios: 15 16 - read-only storage media or 17 18 - part of a fully trusted read-only solution, which means it needs to be 19 immutable and bit-for-bit identical to the official golden image for 20 their releases due to security and other considerations and 21 22 - hope to minimize extra storage space with guaranteed end-to-end performance 23 by using compact layout, transparent file compression and direct access, 24 especially for those embedded devices with limited memory and high-density 25 hosts with numerous containers; 26 27Here is the main features of EROFS: 28 29 - Little endian on-disk design; 30 31 - Currently 4KB block size (nobh) and therefore maximum 16TB address space; 32 33 - Metadata & data could be mixed by design; 34 35 - 2 inode versions for different requirements: 36 37 ===================== ============ ===================================== 38 compact (v1) extended (v2) 39 ===================== ============ ===================================== 40 Inode metadata size 32 bytes 64 bytes 41 Max file size 4 GB 16 EB (also limited by max. vol size) 42 Max uids/gids 65536 4294967296 43 File change time no yes (64 + 32-bit timestamp) 44 Max hardlinks 65536 4294967296 45 Metadata reserved 4 bytes 14 bytes 46 ===================== ============ ===================================== 47 48 - Support extended attributes (xattrs) as an option; 49 50 - Support xattr inline and tail-end data inline for all files; 51 52 - Support POSIX.1e ACLs by using xattrs; 53 54 - Support transparent data compression as an option: 55 LZ4 algorithm with the fixed-sized output compression for high performance; 56 57 - Multiple device support for multi-layer container images. 58 59The following git tree provides the file system user-space tools under 60development (ex, formatting tool mkfs.erofs): 61 62- git://git.kernel.org/pub/scm/linux/kernel/git/xiang/erofs-utils.git 63 64Bugs and patches are welcome, please kindly help us and send to the following 65linux-erofs mailing list: 66 67- linux-erofs mailing list <linux-erofs@lists.ozlabs.org> 68 69Mount options 70============= 71 72=================== ========================================================= 73(no)user_xattr Setup Extended User Attributes. Note: xattr is enabled 74 by default if CONFIG_EROFS_FS_XATTR is selected. 75(no)acl Setup POSIX Access Control List. Note: acl is enabled 76 by default if CONFIG_EROFS_FS_POSIX_ACL is selected. 77cache_strategy=%s Select a strategy for cached decompression from now on: 78 79 ========== ============================================= 80 disabled In-place I/O decompression only; 81 readahead Cache the last incomplete compressed physical 82 cluster for further reading. It still does 83 in-place I/O decompression for the rest 84 compressed physical clusters; 85 readaround Cache the both ends of incomplete compressed 86 physical clusters for further reading. 87 It still does in-place I/O decompression 88 for the rest compressed physical clusters. 89 ========== ============================================= 90dax={always,never} Use direct access (no page cache). See 91 Documentation/filesystems/dax.rst. 92dax A legacy option which is an alias for ``dax=always``. 93device=%s Specify a path to an extra device to be used together. 94=================== ========================================================= 95 96On-disk details 97=============== 98 99Summary 100------- 101Different from other read-only file systems, an EROFS volume is designed 102to be as simple as possible:: 103 104 |-> aligned with the block size 105 ____________________________________________________________ 106 | |SB| | ... | Metadata | ... | Data | Metadata | ... | Data | 107 |_|__|_|_____|__________|_____|______|__________|_____|______| 108 0 +1K 109 110All data areas should be aligned with the block size, but metadata areas 111may not. All metadatas can be now observed in two different spaces (views): 112 113 1. Inode metadata space 114 115 Each valid inode should be aligned with an inode slot, which is a fixed 116 value (32 bytes) and designed to be kept in line with compact inode size. 117 118 Each inode can be directly found with the following formula: 119 inode offset = meta_blkaddr * block_size + 32 * nid 120 121 :: 122 123 |-> aligned with 8B 124 |-> followed closely 125 + meta_blkaddr blocks |-> another slot 126 _____________________________________________________________________ 127 | ... | inode | xattrs | extents | data inline | ... | inode ... 128 |________|_______|(optional)|(optional)|__(optional)_|_____|__________ 129 |-> aligned with the inode slot size 130 . . 131 . . 132 . . 133 . . 134 . . 135 . . 136 .____________________________________________________|-> aligned with 4B 137 | xattr_ibody_header | shared xattrs | inline xattrs | 138 |____________________|_______________|_______________| 139 |-> 12 bytes <-|->x * 4 bytes<-| . 140 . . . 141 . . . 142 . . . 143 ._______________________________.______________________. 144 | id | id | id | id | ... | id | ent | ... | ent| ... | 145 |____|____|____|____|______|____|_____|_____|____|_____| 146 |-> aligned with 4B 147 |-> aligned with 4B 148 149 Inode could be 32 or 64 bytes, which can be distinguished from a common 150 field which all inode versions have -- i_format:: 151 152 __________________ __________________ 153 | i_format | | i_format | 154 |__________________| |__________________| 155 | ... | | ... | 156 | | | | 157 |__________________| 32 bytes | | 158 | | 159 |__________________| 64 bytes 160 161 Xattrs, extents, data inline are followed by the corresponding inode with 162 proper alignment, and they could be optional for different data mappings. 163 _currently_ total 5 data layouts are supported: 164 165 == ==================================================================== 166 0 flat file data without data inline (no extent); 167 1 fixed-sized output data compression (with non-compacted indexes); 168 2 flat file data with tail packing data inline (no extent); 169 3 fixed-sized output data compression (with compacted indexes, v5.3+); 170 4 chunk-based file (v5.15+). 171 == ==================================================================== 172 173 The size of the optional xattrs is indicated by i_xattr_count in inode 174 header. Large xattrs or xattrs shared by many different files can be 175 stored in shared xattrs metadata rather than inlined right after inode. 176 177 2. Shared xattrs metadata space 178 179 Shared xattrs space is similar to the above inode space, started with 180 a specific block indicated by xattr_blkaddr, organized one by one with 181 proper align. 182 183 Each share xattr can also be directly found by the following formula: 184 xattr offset = xattr_blkaddr * block_size + 4 * xattr_id 185 186:: 187 188 |-> aligned by 4 bytes 189 + xattr_blkaddr blocks |-> aligned with 4 bytes 190 _________________________________________________________________________ 191 | ... | xattr_entry | xattr data | ... | xattr_entry | xattr data ... 192 |________|_____________|_____________|_____|______________|_______________ 193 194Directories 195----------- 196All directories are now organized in a compact on-disk format. Note that 197each directory block is divided into index and name areas in order to support 198random file lookup, and all directory entries are _strictly_ recorded in 199alphabetical order in order to support improved prefix binary search 200algorithm (could refer to the related source code). 201 202:: 203 204 ___________________________ 205 / | 206 / ______________|________________ 207 / / | nameoff1 | nameoffN-1 208 ____________.______________._______________v________________v__________ 209 | dirent | dirent | ... | dirent | filename | filename | ... | filename | 210 |___.0___|____1___|_____|___N-1__|____0_____|____1_____|_____|___N-1____| 211 \ ^ 212 \ | * could have 213 \ | trailing '\0' 214 \________________________| nameoff0 215 Directory block 216 217Note that apart from the offset of the first filename, nameoff0 also indicates 218the total number of directory entries in this block since it is no need to 219introduce another on-disk field at all. 220 221Chunk-based file 222---------------- 223In order to support chunk-based data deduplication, a new inode data layout has 224been supported since Linux v5.15: Files are split in equal-sized data chunks 225with ``extents`` area of the inode metadata indicating how to get the chunk 226data: these can be simply as a 4-byte block address array or in the 8-byte 227chunk index form (see struct erofs_inode_chunk_index in erofs_fs.h for more 228details.) 229 230By the way, chunk-based files are all uncompressed for now. 231 232Data compression 233---------------- 234EROFS implements LZ4 fixed-sized output compression which generates fixed-sized 235compressed data blocks from variable-sized input in contrast to other existing 236fixed-sized input solutions. Relatively higher compression ratios can be gotten 237by using fixed-sized output compression since nowadays popular data compression 238algorithms are mostly LZ77-based and such fixed-sized output approach can be 239benefited from the historical dictionary (aka. sliding window). 240 241In details, original (uncompressed) data is turned into several variable-sized 242extents and in the meanwhile, compressed into physical clusters (pclusters). 243In order to record each variable-sized extent, logical clusters (lclusters) are 244introduced as the basic unit of compress indexes to indicate whether a new 245extent is generated within the range (HEAD) or not (NONHEAD). Lclusters are now 246fixed in block size, as illustrated below:: 247 248 |<- variable-sized extent ->|<- VLE ->| 249 clusterofs clusterofs clusterofs 250 | | | 251 _________v_________________________________v_______________________v________ 252 ... | . | | . | | . ... 253 ____|____._________|______________|________.___ _|______________|__.________ 254 |-> lcluster <-|-> lcluster <-|-> lcluster <-|-> lcluster <-| 255 (HEAD) (NONHEAD) (HEAD) (NONHEAD) . 256 . CBLKCNT . . 257 . . . 258 . . . 259 _______._____________________________.______________._________________ 260 ... | | | | ... 261 _______|______________|______________|______________|_________________ 262 |-> big pcluster <-|-> pcluster <-| 263 264A physical cluster can be seen as a container of physical compressed blocks 265which contains compressed data. Previously, only lcluster-sized (4KB) pclusters 266were supported. After big pcluster feature is introduced (available since 267Linux v5.13), pcluster can be a multiple of lcluster size. 268 269For each HEAD lcluster, clusterofs is recorded to indicate where a new extent 270starts and blkaddr is used to seek the compressed data. For each NONHEAD 271lcluster, delta0 and delta1 are available instead of blkaddr to indicate the 272distance to its HEAD lcluster and the next HEAD lcluster. A PLAIN lcluster is 273also a HEAD lcluster except that its data is uncompressed. See the comments 274around "struct z_erofs_vle_decompressed_index" in erofs_fs.h for more details. 275 276If big pcluster is enabled, pcluster size in lclusters needs to be recorded as 277well. Let the delta0 of the first NONHEAD lcluster store the compressed block 278count with a special flag as a new called CBLKCNT NONHEAD lcluster. It's easy 279to understand its delta0 is constantly 1, as illustrated below:: 280 281 __________________________________________________________ 282 | HEAD | NONHEAD | NONHEAD | ... | NONHEAD | HEAD | HEAD | 283 |__:___|_(CBLKCNT)_|_________|_____|_________|__:___|____:_| 284 |<----- a big pcluster (with CBLKCNT) ------>|<-- -->| 285 a lcluster-sized pcluster (without CBLKCNT) ^ 286 287If another HEAD follows a HEAD lcluster, there is no room to record CBLKCNT, 288but it's easy to know the size of such pcluster is 1 lcluster as well. 289