1 .. SPDX-License-Identifier: GPL-2.0
3 ======================================
4 EROFS - Enhanced Read-Only File System
5 ======================================
10 EROFS filesystem stands for Enhanced Read-Only File System. It aims to form a
11 generic read-only filesystem solution for various read-only use cases instead
12 of just focusing on storage space saving without considering any side effects
13 of runtime performance.
15 It is designed to meet the needs of flexibility, feature extendability and user
16 payload friendly, etc. Apart from those, it is still kept as a simple
17 random-access friendly high-performance filesystem to get rid of unneeded I/O
18 amplification and memory-resident overhead compared to similar approaches.
20 It is implemented to be a better choice for the following scenarios:
22 - read-only storage media or
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
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.
33 Here are the main features of EROFS:
35 - Little endian on-disk design;
37 - Block-based distribution and file-based distribution over fscache are
40 - Support multiple devices to refer to external blobs, which can be used
43 - 32-bit block addresses for each device, therefore 16TiB address space at
44 most with 4KiB block size for now;
46 - Two inode layouts for different requirements:
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 ===================== ============ ======================================
59 - Support extended attributes as an option;
61 - Support POSIX.1e ACLs by using extended attributes;
63 - Support transparent data compression as an option:
64 LZ4 and MicroLZMA algorithms can be used on a per-file basis; In addition,
65 inplace decompression is also supported to avoid bounce compressed buffers
66 and page cache thrashing.
68 - Support chunk-based data deduplication and rolling-hash compressed data
71 - Support tailpacking inline compared to byte-addressed unaligned metadata
72 or smaller block size alternatives;
74 - Support merging tail-end data into a special inode as fragments.
76 - Support large folios for uncompressed files.
78 - Support direct I/O on uncompressed files to avoid double caching for loop
81 - Support FSDAX on uncompressed images for secure containers and ramdisks in
82 order to get rid of unnecessary page cache.
84 - Support file-based on-demand loading with the Fscache infrastructure.
86 The following git tree provides the file system user-space tools under
87 development, such as a formatting tool (mkfs.erofs), an on-disk consistency &
88 compatibility checking tool (fsck.erofs), and a debugging tool (dump.erofs):
90 - git://git.kernel.org/pub/scm/linux/kernel/git/xiang/erofs-utils.git
92 Bugs and patches are welcome, please kindly help us and send to the following
93 linux-erofs mailing list:
95 - linux-erofs mailing list <linux-erofs@lists.ozlabs.org>
100 =================== =========================================================
101 (no)user_xattr Setup Extended User Attributes. Note: xattr is enabled
102 by default if CONFIG_EROFS_FS_XATTR is selected.
103 (no)acl Setup POSIX Access Control List. Note: acl is enabled
104 by default if CONFIG_EROFS_FS_POSIX_ACL is selected.
105 cache_strategy=%s Select a strategy for cached decompression from now on:
107 ========== =============================================
108 disabled In-place I/O decompression only;
109 readahead Cache the last incomplete compressed physical
110 cluster for further reading. It still does
111 in-place I/O decompression for the rest
112 compressed physical clusters;
113 readaround Cache the both ends of incomplete compressed
114 physical clusters for further reading.
115 It still does in-place I/O decompression
116 for the rest compressed physical clusters.
117 ========== =============================================
118 dax={always,never} Use direct access (no page cache). See
119 Documentation/filesystems/dax.rst.
120 dax A legacy option which is an alias for ``dax=always``.
121 device=%s Specify a path to an extra device to be used together.
122 fsid=%s Specify a filesystem image ID for Fscache back-end.
123 domain_id=%s Specify a domain ID in fscache mode so that different images
124 with the same blobs under a given domain ID can share storage.
125 =================== =========================================================
130 Information about mounted erofs file systems can be found in /sys/fs/erofs.
131 Each mounted filesystem will have a directory in /sys/fs/erofs based on its
132 device name (i.e., /sys/fs/erofs/sda).
133 (see also Documentation/ABI/testing/sysfs-fs-erofs)
140 Different from other read-only file systems, an EROFS volume is designed
141 to be as simple as possible::
143 |-> aligned with the block size
144 ____________________________________________________________
145 | |SB| | ... | Metadata | ... | Data | Metadata | ... | Data |
146 |_|__|_|_____|__________|_____|______|__________|_____|______|
149 All data areas should be aligned with the block size, but metadata areas
150 may not. All metadatas can be now observed in two different spaces (views):
152 1. Inode metadata space
154 Each valid inode should be aligned with an inode slot, which is a fixed
155 value (32 bytes) and designed to be kept in line with compact inode size.
157 Each inode can be directly found with the following formula:
158 inode offset = meta_blkaddr * block_size + 32 * nid
164 + meta_blkaddr blocks |-> another slot
165 _____________________________________________________________________
166 | ... | inode | xattrs | extents | data inline | ... | inode ...
167 |________|_______|(optional)|(optional)|__(optional)_|_____|__________
168 |-> aligned with the inode slot size
175 .____________________________________________________|-> aligned with 4B
176 | xattr_ibody_header | shared xattrs | inline xattrs |
177 |____________________|_______________|_______________|
178 |-> 12 bytes <-|->x * 4 bytes<-| .
182 ._______________________________.______________________.
183 | id | id | id | id | ... | id | ent | ... | ent| ... |
184 |____|____|____|____|______|____|_____|_____|____|_____|
188 Inode could be 32 or 64 bytes, which can be distinguished from a common
189 field which all inode versions have -- i_format::
191 __________________ __________________
192 | i_format | | i_format |
193 |__________________| |__________________|
196 |__________________| 32 bytes | |
198 |__________________| 64 bytes
200 Xattrs, extents, data inline are followed by the corresponding inode with
201 proper alignment, and they could be optional for different data mappings.
202 _currently_ total 5 data layouts are supported:
204 == ====================================================================
205 0 flat file data without data inline (no extent);
206 1 fixed-sized output data compression (with non-compacted indexes);
207 2 flat file data with tail packing data inline (no extent);
208 3 fixed-sized output data compression (with compacted indexes, v5.3+);
209 4 chunk-based file (v5.15+).
210 == ====================================================================
212 The size of the optional xattrs is indicated by i_xattr_count in inode
213 header. Large xattrs or xattrs shared by many different files can be
214 stored in shared xattrs metadata rather than inlined right after inode.
216 2. Shared xattrs metadata space
218 Shared xattrs space is similar to the above inode space, started with
219 a specific block indicated by xattr_blkaddr, organized one by one with
222 Each share xattr can also be directly found by the following formula:
223 xattr offset = xattr_blkaddr * block_size + 4 * xattr_id
227 |-> aligned by 4 bytes
228 + xattr_blkaddr blocks |-> aligned with 4 bytes
229 _________________________________________________________________________
230 | ... | xattr_entry | xattr data | ... | xattr_entry | xattr data ...
231 |________|_____________|_____________|_____|______________|_______________
235 All directories are now organized in a compact on-disk format. Note that
236 each directory block is divided into index and name areas in order to support
237 random file lookup, and all directory entries are _strictly_ recorded in
238 alphabetical order in order to support improved prefix binary search
239 algorithm (could refer to the related source code).
243 ___________________________
245 / ______________|________________
246 / / | nameoff1 | nameoffN-1
247 ____________.______________._______________v________________v__________
248 | dirent | dirent | ... | dirent | filename | filename | ... | filename |
249 |___.0___|____1___|_____|___N-1__|____0_____|____1_____|_____|___N-1____|
253 \________________________| nameoff0
256 Note that apart from the offset of the first filename, nameoff0 also indicates
257 the total number of directory entries in this block since it is no need to
258 introduce another on-disk field at all.
262 In order to support chunk-based data deduplication, a new inode data layout has
263 been supported since Linux v5.15: Files are split in equal-sized data chunks
264 with ``extents`` area of the inode metadata indicating how to get the chunk
265 data: these can be simply as a 4-byte block address array or in the 8-byte
266 chunk index form (see struct erofs_inode_chunk_index in erofs_fs.h for more
269 By the way, chunk-based files are all uncompressed for now.
273 EROFS implements fixed-sized output compression which generates fixed-sized
274 compressed data blocks from variable-sized input in contrast to other existing
275 fixed-sized input solutions. Relatively higher compression ratios can be gotten
276 by using fixed-sized output compression since nowadays popular data compression
277 algorithms are mostly LZ77-based and such fixed-sized output approach can be
278 benefited from the historical dictionary (aka. sliding window).
280 In details, original (uncompressed) data is turned into several variable-sized
281 extents and in the meanwhile, compressed into physical clusters (pclusters).
282 In order to record each variable-sized extent, logical clusters (lclusters) are
283 introduced as the basic unit of compress indexes to indicate whether a new
284 extent is generated within the range (HEAD) or not (NONHEAD). Lclusters are now
285 fixed in block size, as illustrated below::
287 |<- variable-sized extent ->|<- VLE ->|
288 clusterofs clusterofs clusterofs
290 _________v_________________________________v_______________________v________
291 ... | . | | . | | . ...
292 ____|____._________|______________|________.___ _|______________|__.________
293 |-> lcluster <-|-> lcluster <-|-> lcluster <-|-> lcluster <-|
294 (HEAD) (NONHEAD) (HEAD) (NONHEAD) .
298 _______._____________________________.______________._________________
300 _______|______________|______________|______________|_________________
301 |-> big pcluster <-|-> pcluster <-|
303 A physical cluster can be seen as a container of physical compressed blocks
304 which contains compressed data. Previously, only lcluster-sized (4KB) pclusters
305 were supported. After big pcluster feature is introduced (available since
306 Linux v5.13), pcluster can be a multiple of lcluster size.
308 For each HEAD lcluster, clusterofs is recorded to indicate where a new extent
309 starts and blkaddr is used to seek the compressed data. For each NONHEAD
310 lcluster, delta0 and delta1 are available instead of blkaddr to indicate the
311 distance to its HEAD lcluster and the next HEAD lcluster. A PLAIN lcluster is
312 also a HEAD lcluster except that its data is uncompressed. See the comments
313 around "struct z_erofs_vle_decompressed_index" in erofs_fs.h for more details.
315 If big pcluster is enabled, pcluster size in lclusters needs to be recorded as
316 well. Let the delta0 of the first NONHEAD lcluster store the compressed block
317 count with a special flag as a new called CBLKCNT NONHEAD lcluster. It's easy
318 to understand its delta0 is constantly 1, as illustrated below::
320 __________________________________________________________
321 | HEAD | NONHEAD | NONHEAD | ... | NONHEAD | HEAD | HEAD |
322 |__:___|_(CBLKCNT)_|_________|_____|_________|__:___|____:_|
323 |<----- a big pcluster (with CBLKCNT) ------>|<-- -->|
324 a lcluster-sized pcluster (without CBLKCNT) ^
326 If another HEAD follows a HEAD lcluster, there is no room to record CBLKCNT,
327 but it's easy to know the size of such pcluster is 1 lcluster as well.
329 Since Linux v6.1, each pcluster can be used for multiple variable-sized extents,
330 therefore it can be used for compressed data deduplication.