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 are 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 - Support extended attributes as an option;
60
61 - Support a bloom filter that speeds up negative extended attribute lookups;
62
63 - Support POSIX.1e ACLs by using extended attributes;
64
65 - Support transparent data compression as an option:
66   LZ4, MicroLZMA and DEFLATE algorithms can be used on a per-file basis; In
67   addition, inplace decompression is also supported to avoid bounce compressed
68   buffers and unnecessary page cache thrashing.
69
70 - Support chunk-based data deduplication and rolling-hash compressed data
71   deduplication;
72
73 - Support tailpacking inline compared to byte-addressed unaligned metadata
74   or smaller block size alternatives;
75
76 - Support merging tail-end data into a special inode as fragments.
77
78 - Support large folios to make use of THPs (Transparent Hugepages);
79
80 - Support direct I/O on uncompressed files to avoid double caching for loop
81   devices;
82
83 - Support FSDAX on uncompressed images for secure containers and ramdisks in
84   order to get rid of unnecessary page cache.
85
86 - Support file-based on-demand loading with the Fscache infrastructure.
87
88The following git tree provides the file system user-space tools under
89development, such as a formatting tool (mkfs.erofs), an on-disk consistency &
90compatibility checking tool (fsck.erofs), and a debugging tool (dump.erofs):
91
92- git://git.kernel.org/pub/scm/linux/kernel/git/xiang/erofs-utils.git
93
94For more information, please also refer to the documentation site:
95
96- https://erofs.docs.kernel.org
97
98Bugs and patches are welcome, please kindly help us and send to the following
99linux-erofs mailing list:
100
101- linux-erofs mailing list   <linux-erofs@lists.ozlabs.org>
102
103Mount options
104=============
105
106===================    =========================================================
107(no)user_xattr         Setup Extended User Attributes. Note: xattr is enabled
108                       by default if CONFIG_EROFS_FS_XATTR is selected.
109(no)acl                Setup POSIX Access Control List. Note: acl is enabled
110                       by default if CONFIG_EROFS_FS_POSIX_ACL is selected.
111cache_strategy=%s      Select a strategy for cached decompression from now on:
112
113		       ==========  =============================================
114                         disabled  In-place I/O decompression only;
115                        readahead  Cache the last incomplete compressed physical
116                                   cluster for further reading. It still does
117                                   in-place I/O decompression for the rest
118                                   compressed physical clusters;
119                       readaround  Cache the both ends of incomplete compressed
120                                   physical clusters for further reading.
121                                   It still does in-place I/O decompression
122                                   for the rest compressed physical clusters.
123		       ==========  =============================================
124dax={always,never}     Use direct access (no page cache).  See
125                       Documentation/filesystems/dax.rst.
126dax                    A legacy option which is an alias for ``dax=always``.
127device=%s              Specify a path to an extra device to be used together.
128fsid=%s                Specify a filesystem image ID for Fscache back-end.
129domain_id=%s           Specify a domain ID in fscache mode so that different images
130                       with the same blobs under a given domain ID can share storage.
131fsoffset=%llu          Specify block-aligned filesystem offset for the primary device.
132===================    =========================================================
133
134Sysfs Entries
135=============
136
137Information about mounted erofs file systems can be found in /sys/fs/erofs.
138Each mounted filesystem will have a directory in /sys/fs/erofs based on its
139device name (i.e., /sys/fs/erofs/sda).
140(see also Documentation/ABI/testing/sysfs-fs-erofs)
141
142On-disk details
143===============
144
145Summary
146-------
147Different from other read-only file systems, an EROFS volume is designed
148to be as simple as possible::
149
150                                |-> aligned with the block size
151   ____________________________________________________________
152  | |SB| | ... | Metadata | ... | Data | Metadata | ... | Data |
153  |_|__|_|_____|__________|_____|______|__________|_____|______|
154  0 +1K
155
156All data areas should be aligned with the block size, but metadata areas
157may not. All metadatas can be now observed in two different spaces (views):
158
159 1. Inode metadata space
160
161    Each valid inode should be aligned with an inode slot, which is a fixed
162    value (32 bytes) and designed to be kept in line with compact inode size.
163
164    Each inode can be directly found with the following formula:
165         inode offset = meta_blkaddr * block_size + 32 * nid
166
167    ::
168
169                                 |-> aligned with 8B
170                                            |-> followed closely
171     + meta_blkaddr blocks                                      |-> another slot
172       _____________________________________________________________________
173     |  ...   | inode |  xattrs  | extents  | data inline | ... | inode ...
174     |________|_______|(optional)|(optional)|__(optional)_|_____|__________
175              |-> aligned with the inode slot size
176                   .                   .
177                 .                         .
178               .                              .
179             .                                    .
180           .                                         .
181         .                                              .
182       .____________________________________________________|-> aligned with 4B
183       | xattr_ibody_header | shared xattrs | inline xattrs |
184       |____________________|_______________|_______________|
185       |->    12 bytes    <-|->x * 4 bytes<-|               .
186                           .                .                 .
187                     .                      .                   .
188                .                           .                     .
189            ._______________________________.______________________.
190            | id | id | id | id |  ... | id | ent | ... | ent| ... |
191            |____|____|____|____|______|____|_____|_____|____|_____|
192                                            |-> aligned with 4B
193                                                        |-> aligned with 4B
194
195    Inode could be 32 or 64 bytes, which can be distinguished from a common
196    field which all inode versions have -- i_format::
197
198        __________________               __________________
199       |     i_format     |             |     i_format     |
200       |__________________|             |__________________|
201       |        ...       |             |        ...       |
202       |                  |             |                  |
203       |__________________| 32 bytes    |                  |
204                                        |                  |
205                                        |__________________| 64 bytes
206
207    Xattrs, extents, data inline are placed after the corresponding inode with
208    proper alignment, and they could be optional for different data mappings.
209    _currently_ total 5 data layouts are supported:
210
211    ==  ====================================================================
212     0  flat file data without data inline (no extent);
213     1  fixed-sized output data compression (with non-compacted indexes);
214     2  flat file data with tail packing data inline (no extent);
215     3  fixed-sized output data compression (with compacted indexes, v5.3+);
216     4  chunk-based file (v5.15+).
217    ==  ====================================================================
218
219    The size of the optional xattrs is indicated by i_xattr_count in inode
220    header. Large xattrs or xattrs shared by many different files can be
221    stored in shared xattrs metadata rather than inlined right after inode.
222
223 2. Shared xattrs metadata space
224
225    Shared xattrs space is similar to the above inode space, started with
226    a specific block indicated by xattr_blkaddr, organized one by one with
227    proper align.
228
229    Each share xattr can also be directly found by the following formula:
230         xattr offset = xattr_blkaddr * block_size + 4 * xattr_id
231
232::
233
234                           |-> aligned by  4 bytes
235    + xattr_blkaddr blocks                     |-> aligned with 4 bytes
236     _________________________________________________________________________
237    |  ...   | xattr_entry |  xattr data | ... |  xattr_entry | xattr data  ...
238    |________|_____________|_____________|_____|______________|_______________
239
240Directories
241-----------
242All directories are now organized in a compact on-disk format. Note that
243each directory block is divided into index and name areas in order to support
244random file lookup, and all directory entries are _strictly_ recorded in
245alphabetical order in order to support improved prefix binary search
246algorithm (could refer to the related source code).
247
248::
249
250                  ___________________________
251                 /                           |
252                /              ______________|________________
253               /              /              | nameoff1       | nameoffN-1
254  ____________.______________._______________v________________v__________
255 | dirent | dirent | ... | dirent | filename | filename | ... | filename |
256 |___.0___|____1___|_____|___N-1__|____0_____|____1_____|_____|___N-1____|
257      \                           ^
258       \                          |                           * could have
259        \                         |                             trailing '\0'
260         \________________________| nameoff0
261                             Directory block
262
263Note that apart from the offset of the first filename, nameoff0 also indicates
264the total number of directory entries in this block since it is no need to
265introduce another on-disk field at all.
266
267Chunk-based files
268-----------------
269In order to support chunk-based data deduplication, a new inode data layout has
270been supported since Linux v5.15: Files are split in equal-sized data chunks
271with ``extents`` area of the inode metadata indicating how to get the chunk
272data: these can be simply as a 4-byte block address array or in the 8-byte
273chunk index form (see struct erofs_inode_chunk_index in erofs_fs.h for more
274details.)
275
276By the way, chunk-based files are all uncompressed for now.
277
278Long extended attribute name prefixes
279-------------------------------------
280There are use cases where extended attributes with different values can have
281only a few common prefixes (such as overlayfs xattrs).  The predefined prefixes
282work inefficiently in both image size and runtime performance in such cases.
283
284The long xattr name prefixes feature is introduced to address this issue.  The
285overall idea is that, apart from the existing predefined prefixes, the xattr
286entry could also refer to user-specified long xattr name prefixes, e.g.
287"trusted.overlay.".
288
289When referring to a long xattr name prefix, the highest bit (bit 7) of
290erofs_xattr_entry.e_name_index is set, while the lower bits (bit 0-6) as a whole
291represent the index of the referred long name prefix among all long name
292prefixes.  Therefore, only the trailing part of the name apart from the long
293xattr name prefix is stored in erofs_xattr_entry.e_name, which could be empty if
294the full xattr name matches exactly as its long xattr name prefix.
295
296All long xattr prefixes are stored one by one in the packed inode as long as
297the packed inode is valid, or in the meta inode otherwise.  The
298xattr_prefix_count (of the on-disk superblock) indicates the total number of
299long xattr name prefixes, while (xattr_prefix_start * 4) indicates the start
300offset of long name prefixes in the packed/meta inode.  Note that, long extended
301attribute name prefixes are disabled if xattr_prefix_count is 0.
302
303Each long name prefix is stored in the format: ALIGN({__le16 len, data}, 4),
304where len represents the total size of the data part.  The data part is actually
305represented by 'struct erofs_xattr_long_prefix', where base_index represents the
306index of the predefined xattr name prefix, e.g. EROFS_XATTR_INDEX_TRUSTED for
307"trusted.overlay." long name prefix, while the infix string keeps the string
308after stripping the short prefix, e.g. "overlay." for the example above.
309
310Data compression
311----------------
312EROFS implements fixed-sized output compression which generates fixed-sized
313compressed data blocks from variable-sized input in contrast to other existing
314fixed-sized input solutions. Relatively higher compression ratios can be gotten
315by using fixed-sized output compression since nowadays popular data compression
316algorithms are mostly LZ77-based and such fixed-sized output approach can be
317benefited from the historical dictionary (aka. sliding window).
318
319In details, original (uncompressed) data is turned into several variable-sized
320extents and in the meanwhile, compressed into physical clusters (pclusters).
321In order to record each variable-sized extent, logical clusters (lclusters) are
322introduced as the basic unit of compress indexes to indicate whether a new
323extent is generated within the range (HEAD) or not (NONHEAD). Lclusters are now
324fixed in block size, as illustrated below::
325
326          |<-    variable-sized extent    ->|<-       VLE         ->|
327        clusterofs                        clusterofs              clusterofs
328          |                                 |                       |
329 _________v_________________________________v_______________________v________
330 ... |    .         |              |        .     |              |  .   ...
331 ____|____._________|______________|________.___ _|______________|__.________
332     |-> lcluster <-|-> lcluster <-|-> lcluster <-|-> lcluster <-|
333          (HEAD)        (NONHEAD)       (HEAD)        (NONHEAD)    .
334           .             CBLKCNT            .                    .
335            .                               .                  .
336             .                              .                .
337       _______._____________________________.______________._________________
338          ... |              |              |              | ...
339       _______|______________|______________|______________|_________________
340              |->      big pcluster       <-|-> pcluster <-|
341
342A physical cluster can be seen as a container of physical compressed blocks
343which contains compressed data. Previously, only lcluster-sized (4KB) pclusters
344were supported. After big pcluster feature is introduced (available since
345Linux v5.13), pcluster can be a multiple of lcluster size.
346
347For each HEAD lcluster, clusterofs is recorded to indicate where a new extent
348starts and blkaddr is used to seek the compressed data. For each NONHEAD
349lcluster, delta0 and delta1 are available instead of blkaddr to indicate the
350distance to its HEAD lcluster and the next HEAD lcluster. A PLAIN lcluster is
351also a HEAD lcluster except that its data is uncompressed. See the comments
352around "struct z_erofs_vle_decompressed_index" in erofs_fs.h for more details.
353
354If big pcluster is enabled, pcluster size in lclusters needs to be recorded as
355well. Let the delta0 of the first NONHEAD lcluster store the compressed block
356count with a special flag as a new called CBLKCNT NONHEAD lcluster. It's easy
357to understand its delta0 is constantly 1, as illustrated below::
358
359   __________________________________________________________
360  | HEAD |  NONHEAD  | NONHEAD | ... | NONHEAD | HEAD | HEAD |
361  |__:___|_(CBLKCNT)_|_________|_____|_________|__:___|____:_|
362     |<----- a big pcluster (with CBLKCNT) ------>|<--  -->|
363           a lcluster-sized pcluster (without CBLKCNT) ^
364
365If another HEAD follows a HEAD lcluster, there is no room to record CBLKCNT,
366but it's easy to know the size of such pcluster is 1 lcluster as well.
367
368Since Linux v6.1, each pcluster can be used for multiple variable-sized extents,
369therefore it can be used for compressed data deduplication.
370