1.. SPDX-License-Identifier: GPL-2.0 2 3====================================== 4EROFS - Enhanced Read-Only File System 5====================================== 6 7Overview 8======== 9 10EROFS filesystem stands for Enhanced Read-Only File System. It aims to form a 11generic read-only filesystem solution for various read-only use cases instead 12of just focusing on storage space saving without considering any side effects 13of runtime performance. 14 15It is designed to meet the needs of flexibility, feature extendability and user 16payload friendly, etc. Apart from those, it is still kept as a simple 17random-access friendly high-performance filesystem to get rid of unneeded I/O 18amplification and memory-resident overhead compared to similar approaches. 19 20It is implemented to be a better choice for the following scenarios: 21 22 - read-only storage media or 23 24 - part of a fully trusted read-only solution, which means it needs to be 25 immutable and bit-for-bit identical to the official golden image for 26 their releases due to security or other considerations and 27 28 - hope to minimize extra storage space with guaranteed end-to-end performance 29 by using compact layout, transparent file compression and direct access, 30 especially for those embedded devices with limited memory and high-density 31 hosts with numerous containers. 32 33Here is the main features of EROFS: 34 35 - Little endian on-disk design; 36 37 - Block-based distribution and file-based distribution over fscache are 38 supported; 39 40 - Support multiple devices to refer to external blobs, which can be used 41 for container images; 42 43 - 32-bit block addresses for each device, therefore 16TiB address space at 44 most with 4KiB block size for now; 45 46 - Two inode layouts for different requirements: 47 48 ===================== ============ ====================================== 49 compact (v1) extended (v2) 50 ===================== ============ ====================================== 51 Inode metadata size 32 bytes 64 bytes 52 Max file size 4 GiB 16 EiB (also limited by max. vol size) 53 Max uids/gids 65536 4294967296 54 Per-inode timestamp no yes (64 + 32-bit timestamp) 55 Max hardlinks 65536 4294967296 56 Metadata reserved 8 bytes 18 bytes 57 ===================== ============ ====================================== 58 59 - Metadata and data could be mixed as an option; 60 61 - Support extended attributes (xattrs) as an option; 62 63 - Support tailpacking data and xattr inline compared to byte-addressed 64 unaligned metadata or smaller block size alternatives; 65 66 - Support POSIX.1e ACLs by using xattrs; 67 68 - Support transparent data compression as an option: 69 LZ4 and MicroLZMA algorithms can be used on a per-file basis; In addition, 70 inplace decompression is also supported to avoid bounce compressed buffers 71 and page cache thrashing. 72 73 - Support direct I/O on uncompressed files to avoid double caching for loop 74 devices; 75 76 - Support FSDAX on uncompressed images for secure containers and ramdisks in 77 order to get rid of unnecessary page cache. 78 79 - Support multiple devices for multi blob container images; 80 81 - Support file-based on-demand loading with the Fscache infrastructure. 82 83The following git tree provides the file system user-space tools under 84development, such as a formatting tool (mkfs.erofs), an on-disk consistency & 85compatibility checking tool (fsck.erofs), and a debugging tool (dump.erofs): 86 87- git://git.kernel.org/pub/scm/linux/kernel/git/xiang/erofs-utils.git 88 89Bugs and patches are welcome, please kindly help us and send to the following 90linux-erofs mailing list: 91 92- linux-erofs mailing list <linux-erofs@lists.ozlabs.org> 93 94Mount options 95============= 96 97=================== ========================================================= 98(no)user_xattr Setup Extended User Attributes. Note: xattr is enabled 99 by default if CONFIG_EROFS_FS_XATTR is selected. 100(no)acl Setup POSIX Access Control List. Note: acl is enabled 101 by default if CONFIG_EROFS_FS_POSIX_ACL is selected. 102cache_strategy=%s Select a strategy for cached decompression from now on: 103 104 ========== ============================================= 105 disabled In-place I/O decompression only; 106 readahead Cache the last incomplete compressed physical 107 cluster for further reading. It still does 108 in-place I/O decompression for the rest 109 compressed physical clusters; 110 readaround Cache the both ends of incomplete compressed 111 physical clusters for further reading. 112 It still does in-place I/O decompression 113 for the rest compressed physical clusters. 114 ========== ============================================= 115dax={always,never} Use direct access (no page cache). See 116 Documentation/filesystems/dax.rst. 117dax A legacy option which is an alias for ``dax=always``. 118device=%s Specify a path to an extra device to be used together. 119fsid=%s Specify a filesystem image ID for Fscache back-end. 120=================== ========================================================= 121 122Sysfs Entries 123============= 124 125Information about mounted erofs file systems can be found in /sys/fs/erofs. 126Each mounted filesystem will have a directory in /sys/fs/erofs based on its 127device name (i.e., /sys/fs/erofs/sda). 128(see also Documentation/ABI/testing/sysfs-fs-erofs) 129 130On-disk details 131=============== 132 133Summary 134------- 135Different from other read-only file systems, an EROFS volume is designed 136to be as simple as possible:: 137 138 |-> aligned with the block size 139 ____________________________________________________________ 140 | |SB| | ... | Metadata | ... | Data | Metadata | ... | Data | 141 |_|__|_|_____|__________|_____|______|__________|_____|______| 142 0 +1K 143 144All data areas should be aligned with the block size, but metadata areas 145may not. All metadatas can be now observed in two different spaces (views): 146 147 1. Inode metadata space 148 149 Each valid inode should be aligned with an inode slot, which is a fixed 150 value (32 bytes) and designed to be kept in line with compact inode size. 151 152 Each inode can be directly found with the following formula: 153 inode offset = meta_blkaddr * block_size + 32 * nid 154 155 :: 156 157 |-> aligned with 8B 158 |-> followed closely 159 + meta_blkaddr blocks |-> another slot 160 _____________________________________________________________________ 161 | ... | inode | xattrs | extents | data inline | ... | inode ... 162 |________|_______|(optional)|(optional)|__(optional)_|_____|__________ 163 |-> aligned with the inode slot size 164 . . 165 . . 166 . . 167 . . 168 . . 169 . . 170 .____________________________________________________|-> aligned with 4B 171 | xattr_ibody_header | shared xattrs | inline xattrs | 172 |____________________|_______________|_______________| 173 |-> 12 bytes <-|->x * 4 bytes<-| . 174 . . . 175 . . . 176 . . . 177 ._______________________________.______________________. 178 | id | id | id | id | ... | id | ent | ... | ent| ... | 179 |____|____|____|____|______|____|_____|_____|____|_____| 180 |-> aligned with 4B 181 |-> aligned with 4B 182 183 Inode could be 32 or 64 bytes, which can be distinguished from a common 184 field which all inode versions have -- i_format:: 185 186 __________________ __________________ 187 | i_format | | i_format | 188 |__________________| |__________________| 189 | ... | | ... | 190 | | | | 191 |__________________| 32 bytes | | 192 | | 193 |__________________| 64 bytes 194 195 Xattrs, extents, data inline are followed by the corresponding inode with 196 proper alignment, and they could be optional for different data mappings. 197 _currently_ total 5 data layouts are supported: 198 199 == ==================================================================== 200 0 flat file data without data inline (no extent); 201 1 fixed-sized output data compression (with non-compacted indexes); 202 2 flat file data with tail packing data inline (no extent); 203 3 fixed-sized output data compression (with compacted indexes, v5.3+); 204 4 chunk-based file (v5.15+). 205 == ==================================================================== 206 207 The size of the optional xattrs is indicated by i_xattr_count in inode 208 header. Large xattrs or xattrs shared by many different files can be 209 stored in shared xattrs metadata rather than inlined right after inode. 210 211 2. Shared xattrs metadata space 212 213 Shared xattrs space is similar to the above inode space, started with 214 a specific block indicated by xattr_blkaddr, organized one by one with 215 proper align. 216 217 Each share xattr can also be directly found by the following formula: 218 xattr offset = xattr_blkaddr * block_size + 4 * xattr_id 219 220:: 221 222 |-> aligned by 4 bytes 223 + xattr_blkaddr blocks |-> aligned with 4 bytes 224 _________________________________________________________________________ 225 | ... | xattr_entry | xattr data | ... | xattr_entry | xattr data ... 226 |________|_____________|_____________|_____|______________|_______________ 227 228Directories 229----------- 230All directories are now organized in a compact on-disk format. Note that 231each directory block is divided into index and name areas in order to support 232random file lookup, and all directory entries are _strictly_ recorded in 233alphabetical order in order to support improved prefix binary search 234algorithm (could refer to the related source code). 235 236:: 237 238 ___________________________ 239 / | 240 / ______________|________________ 241 / / | nameoff1 | nameoffN-1 242 ____________.______________._______________v________________v__________ 243 | dirent | dirent | ... | dirent | filename | filename | ... | filename | 244 |___.0___|____1___|_____|___N-1__|____0_____|____1_____|_____|___N-1____| 245 \ ^ 246 \ | * could have 247 \ | trailing '\0' 248 \________________________| nameoff0 249 Directory block 250 251Note that apart from the offset of the first filename, nameoff0 also indicates 252the total number of directory entries in this block since it is no need to 253introduce another on-disk field at all. 254 255Chunk-based files 256----------------- 257In order to support chunk-based data deduplication, a new inode data layout has 258been supported since Linux v5.15: Files are split in equal-sized data chunks 259with ``extents`` area of the inode metadata indicating how to get the chunk 260data: these can be simply as a 4-byte block address array or in the 8-byte 261chunk index form (see struct erofs_inode_chunk_index in erofs_fs.h for more 262details.) 263 264By the way, chunk-based files are all uncompressed for now. 265 266Data compression 267---------------- 268EROFS implements LZ4 fixed-sized output compression which generates fixed-sized 269compressed data blocks from variable-sized input in contrast to other existing 270fixed-sized input solutions. Relatively higher compression ratios can be gotten 271by using fixed-sized output compression since nowadays popular data compression 272algorithms are mostly LZ77-based and such fixed-sized output approach can be 273benefited from the historical dictionary (aka. sliding window). 274 275In details, original (uncompressed) data is turned into several variable-sized 276extents and in the meanwhile, compressed into physical clusters (pclusters). 277In order to record each variable-sized extent, logical clusters (lclusters) are 278introduced as the basic unit of compress indexes to indicate whether a new 279extent is generated within the range (HEAD) or not (NONHEAD). Lclusters are now 280fixed in block size, as illustrated below:: 281 282 |<- variable-sized extent ->|<- VLE ->| 283 clusterofs clusterofs clusterofs 284 | | | 285 _________v_________________________________v_______________________v________ 286 ... | . | | . | | . ... 287 ____|____._________|______________|________.___ _|______________|__.________ 288 |-> lcluster <-|-> lcluster <-|-> lcluster <-|-> lcluster <-| 289 (HEAD) (NONHEAD) (HEAD) (NONHEAD) . 290 . CBLKCNT . . 291 . . . 292 . . . 293 _______._____________________________.______________._________________ 294 ... | | | | ... 295 _______|______________|______________|______________|_________________ 296 |-> big pcluster <-|-> pcluster <-| 297 298A physical cluster can be seen as a container of physical compressed blocks 299which contains compressed data. Previously, only lcluster-sized (4KB) pclusters 300were supported. After big pcluster feature is introduced (available since 301Linux v5.13), pcluster can be a multiple of lcluster size. 302 303For each HEAD lcluster, clusterofs is recorded to indicate where a new extent 304starts and blkaddr is used to seek the compressed data. For each NONHEAD 305lcluster, delta0 and delta1 are available instead of blkaddr to indicate the 306distance to its HEAD lcluster and the next HEAD lcluster. A PLAIN lcluster is 307also a HEAD lcluster except that its data is uncompressed. See the comments 308around "struct z_erofs_vle_decompressed_index" in erofs_fs.h for more details. 309 310If big pcluster is enabled, pcluster size in lclusters needs to be recorded as 311well. Let the delta0 of the first NONHEAD lcluster store the compressed block 312count with a special flag as a new called CBLKCNT NONHEAD lcluster. It's easy 313to understand its delta0 is constantly 1, as illustrated below:: 314 315 __________________________________________________________ 316 | HEAD | NONHEAD | NONHEAD | ... | NONHEAD | HEAD | HEAD | 317 |__:___|_(CBLKCNT)_|_________|_____|_________|__:___|____:_| 318 |<----- a big pcluster (with CBLKCNT) ------>|<-- -->| 319 a lcluster-sized pcluster (without CBLKCNT) ^ 320 321If another HEAD follows a HEAD lcluster, there is no room to record CBLKCNT, 322but it's easy to know the size of such pcluster is 1 lcluster as well. 323