1 /* SPDX-License-Identifier: MIT */
2 /*
3 * EROFS (Enhanced ROM File System) on-disk format definition
4 *
5 * Copyright (C) 2017-2018 HUAWEI, Inc.
6 * https://www.huawei.com/
7 * Copyright (C) 2021, Alibaba Cloud
8 */
9 #ifndef __EROFS_FS_H
10 #define __EROFS_FS_H
11
12 /* to allow for x86 boot sectors and other oddities. */
13 #define EROFS_SUPER_OFFSET 1024
14
15 #define EROFS_FEATURE_COMPAT_SB_CHKSUM 0x00000001
16 #define EROFS_FEATURE_COMPAT_MTIME 0x00000002
17 #define EROFS_FEATURE_COMPAT_XATTR_FILTER 0x00000004
18 #define EROFS_FEATURE_COMPAT_SHARED_EA_IN_METABOX 0x00000008
19 #define EROFS_FEATURE_COMPAT_PLAIN_XATTR_PFX 0x00000010
20 #define EROFS_FEATURE_COMPAT_ISHARE_XATTRS 0x00000020
21
22 /*
23 * Any bits that aren't in EROFS_ALL_FEATURE_INCOMPAT should
24 * be incompatible with this kernel version.
25 */
26 #define EROFS_FEATURE_INCOMPAT_LZ4_0PADDING 0x00000001
27 #define EROFS_FEATURE_INCOMPAT_COMPR_CFGS 0x00000002
28 #define EROFS_FEATURE_INCOMPAT_BIG_PCLUSTER 0x00000002
29 #define EROFS_FEATURE_INCOMPAT_CHUNKED_FILE 0x00000004
30 #define EROFS_FEATURE_INCOMPAT_DEVICE_TABLE 0x00000008
31 #define EROFS_FEATURE_INCOMPAT_COMPR_HEAD2 0x00000008
32 #define EROFS_FEATURE_INCOMPAT_ZTAILPACKING 0x00000010
33 #define EROFS_FEATURE_INCOMPAT_FRAGMENTS 0x00000020
34 #define EROFS_FEATURE_INCOMPAT_DEDUPE 0x00000020
35 #define EROFS_FEATURE_INCOMPAT_XATTR_PREFIXES 0x00000040
36 #define EROFS_FEATURE_INCOMPAT_48BIT 0x00000080
37 #define EROFS_FEATURE_INCOMPAT_METABOX 0x00000100
38 #define EROFS_ALL_FEATURE_INCOMPAT \
39 ((EROFS_FEATURE_INCOMPAT_METABOX << 1) - 1)
40
41 #define EROFS_SB_EXTSLOT_SIZE 16
42
43 struct erofs_deviceslot {
44 u8 tag[64]; /* digest(sha256), etc. */
45 __le32 blocks_lo; /* total blocks count of this device */
46 __le32 uniaddr_lo; /* unified starting block of this device */
47 __le32 blocks_hi; /* total blocks count MSB */
48 __le16 uniaddr_hi; /* unified starting block MSB */
49 u8 reserved[50];
50 };
51 #define EROFS_DEVT_SLOT_SIZE sizeof(struct erofs_deviceslot)
52
53 /* erofs on-disk super block (currently 144 bytes at maximum) */
54 struct erofs_super_block {
55 __le32 magic; /* file system magic number */
56 __le32 checksum; /* crc32c to avoid unexpected on-disk overlap */
57 __le32 feature_compat;
58 __u8 blkszbits; /* filesystem block size in bit shift */
59 __u8 sb_extslots; /* superblock size = 128 + sb_extslots * 16 */
60 union {
61 __le16 rootnid_2b; /* nid of root directory */
62 __le16 blocks_hi; /* (48BIT on) blocks count MSB */
63 } __packed rb;
64 __le64 inos; /* total valid ino # (== f_files - f_favail) */
65 __le64 epoch; /* base seconds used for compact inodes */
66 __le32 fixed_nsec; /* fixed nanoseconds for compact inodes */
67 __le32 blocks_lo; /* blocks count LSB */
68 __le32 meta_blkaddr; /* start block address of metadata area */
69 __le32 xattr_blkaddr; /* start block address of shared xattr area */
70 __u8 uuid[16]; /* 128-bit uuid for volume */
71 __u8 volume_name[16]; /* volume name */
72 __le32 feature_incompat;
73 union {
74 /* bitmap for available compression algorithms */
75 __le16 available_compr_algs;
76 /* customized sliding window size instead of 64k by default */
77 __le16 lz4_max_distance;
78 } __packed u1;
79 __le16 extra_devices; /* # of devices besides the primary device */
80 __le16 devt_slotoff; /* startoff = devt_slotoff * devt_slotsize */
81 __u8 dirblkbits; /* directory block size in bit shift */
82 __u8 xattr_prefix_count; /* # of long xattr name prefixes */
83 __le32 xattr_prefix_start; /* start of long xattr prefixes */
84 __le64 packed_nid; /* nid of the special packed inode */
85 __u8 xattr_filter_reserved; /* reserved for xattr name filter */
86 __u8 ishare_xattr_prefix_id;
87 __u8 reserved[2];
88 __le32 build_time; /* seconds added to epoch for mkfs time */
89 __le64 rootnid_8b; /* (48BIT on) nid of root directory */
90 __le64 reserved2;
91 __le64 metabox_nid; /* (METABOX on) nid of the metabox inode */
92 __le64 reserved3; /* [align to extslot 1] */
93 };
94
95 /*
96 * EROFS inode datalayout (i_format in on-disk inode):
97 * 0 - uncompressed flat inode without tail-packing inline data:
98 * 1 - compressed inode with non-compact indexes:
99 * 2 - uncompressed flat inode with tail-packing inline data:
100 * 3 - compressed inode with compact indexes:
101 * 4 - chunk-based inode with (optional) multi-device support:
102 * 5~7 - reserved
103 */
104 enum {
105 EROFS_INODE_FLAT_PLAIN = 0,
106 EROFS_INODE_COMPRESSED_FULL = 1,
107 EROFS_INODE_FLAT_INLINE = 2,
108 EROFS_INODE_COMPRESSED_COMPACT = 3,
109 EROFS_INODE_CHUNK_BASED = 4,
110 EROFS_INODE_DATALAYOUT_MAX
111 };
112
erofs_inode_is_data_compressed(unsigned int datamode)113 static inline bool erofs_inode_is_data_compressed(unsigned int datamode)
114 {
115 return datamode == EROFS_INODE_COMPRESSED_COMPACT ||
116 datamode == EROFS_INODE_COMPRESSED_FULL;
117 }
118
119 /* bit definitions of inode i_format */
120 #define EROFS_I_VERSION_MASK 0x01
121 #define EROFS_I_DATALAYOUT_MASK 0x07
122
123 #define EROFS_I_VERSION_BIT 0
124 #define EROFS_I_DATALAYOUT_BIT 1
125 #define EROFS_I_NLINK_1_BIT 4 /* non-directory compact inodes only */
126 #define EROFS_I_DOT_OMITTED_BIT 4 /* (directories) omit the `.` dirent */
127 #define EROFS_I_ALL ((1 << (EROFS_I_NLINK_1_BIT + 1)) - 1)
128
129 /* indicate chunk blkbits, thus 'chunksize = blocksize << chunk blkbits' */
130 #define EROFS_CHUNK_FORMAT_BLKBITS_MASK 0x001F
131 /* with chunk indexes or just a 4-byte block array */
132 #define EROFS_CHUNK_FORMAT_INDEXES 0x0020
133 #define EROFS_CHUNK_FORMAT_48BIT 0x0040
134
135 #define EROFS_CHUNK_FORMAT_ALL ((EROFS_CHUNK_FORMAT_48BIT << 1) - 1)
136
137 /* 32-byte on-disk inode */
138 #define EROFS_INODE_LAYOUT_COMPACT 0
139 /* 64-byte on-disk inode */
140 #define EROFS_INODE_LAYOUT_EXTENDED 1
141
142 struct erofs_inode_chunk_info {
143 __le16 format; /* chunk blkbits, etc. */
144 __le16 reserved;
145 };
146
147 union erofs_inode_i_u {
148 __le32 blocks_lo; /* total blocks count (if compressed inodes) */
149 __le32 startblk_lo; /* starting block number (if flat inodes) */
150 __le32 rdev; /* device ID (if special inodes) */
151 struct erofs_inode_chunk_info c;
152 };
153
154 union erofs_inode_i_nb {
155 __le16 nlink; /* if EROFS_I_NLINK_1_BIT is unset */
156 __le16 blocks_hi; /* total blocks count MSB */
157 __le16 startblk_hi; /* starting block number MSB */
158 } __packed;
159
160 /* 32-byte reduced form of an ondisk inode */
161 struct erofs_inode_compact {
162 __le16 i_format; /* inode format hints */
163 __le16 i_xattr_icount;
164 __le16 i_mode;
165 union erofs_inode_i_nb i_nb;
166 __le32 i_size;
167 __le32 i_mtime;
168 union erofs_inode_i_u i_u;
169
170 __le32 i_ino; /* only used for 32-bit stat compatibility */
171 __le16 i_uid;
172 __le16 i_gid;
173 __le32 i_reserved;
174 };
175
176 /* 64-byte complete form of an ondisk inode */
177 struct erofs_inode_extended {
178 __le16 i_format; /* inode format hints */
179 __le16 i_xattr_icount;
180 __le16 i_mode;
181 union erofs_inode_i_nb i_nb;
182 __le64 i_size;
183 union erofs_inode_i_u i_u;
184
185 __le32 i_ino; /* only used for 32-bit stat compatibility */
186 __le32 i_uid;
187 __le32 i_gid;
188 __le64 i_mtime;
189 __le32 i_mtime_nsec;
190 __le32 i_nlink;
191 __u8 i_reserved2[16];
192 };
193
194 /*
195 * inline xattrs (n == i_xattr_icount):
196 * erofs_xattr_ibody_header(1) + (n - 1) * 4 bytes
197 * 12 bytes / \
198 * / \
199 * /-----------------------\
200 * | erofs_xattr_entries+ |
201 * +-----------------------+
202 * inline xattrs must starts in erofs_xattr_ibody_header,
203 * for read-only fs, no need to introduce h_refcount
204 */
205 struct erofs_xattr_ibody_header {
206 __le32 h_name_filter; /* bit value 1 indicates not-present */
207 __u8 h_shared_count;
208 __u8 h_reserved2[7];
209 __le32 h_shared_xattrs[]; /* shared xattr id array */
210 };
211
212 /* Name indexes */
213 #define EROFS_XATTR_INDEX_USER 1
214 #define EROFS_XATTR_INDEX_POSIX_ACL_ACCESS 2
215 #define EROFS_XATTR_INDEX_POSIX_ACL_DEFAULT 3
216 #define EROFS_XATTR_INDEX_TRUSTED 4
217 #define EROFS_XATTR_INDEX_LUSTRE 5
218 #define EROFS_XATTR_INDEX_SECURITY 6
219
220 /*
221 * bit 7 of e_name_index is set when it refers to a long xattr name prefix,
222 * while the remained lower bits represent the index of the prefix.
223 */
224 #define EROFS_XATTR_LONG_PREFIX 0x80
225 #define EROFS_XATTR_LONG_PREFIX_MASK 0x7f
226
227 #define EROFS_XATTR_FILTER_BITS 32
228 #define EROFS_XATTR_FILTER_DEFAULT UINT32_MAX
229 #define EROFS_XATTR_FILTER_SEED 0x25BBE08F
230
231 /* xattr entry (for both inline & shared xattrs) */
232 struct erofs_xattr_entry {
233 __u8 e_name_len; /* length of name */
234 __u8 e_name_index; /* attribute name index */
235 __le16 e_value_size; /* size of attribute value */
236 /* followed by e_name and e_value */
237 char e_name[]; /* attribute name */
238 };
239
240 /* long xattr name prefix */
241 struct erofs_xattr_long_prefix {
242 __u8 base_index; /* short xattr name prefix index */
243 char infix[]; /* infix apart from short prefix */
244 };
245
erofs_xattr_ibody_size(__le16 i_xattr_icount)246 static inline unsigned int erofs_xattr_ibody_size(__le16 i_xattr_icount)
247 {
248 if (!i_xattr_icount)
249 return 0;
250
251 /* 1 header + n-1 * 4 bytes inline xattr to keep continuity */
252 return sizeof(struct erofs_xattr_ibody_header) +
253 sizeof(__u32) * (le16_to_cpu(i_xattr_icount) - 1);
254 }
255
256 #define EROFS_XATTR_ALIGN(size) round_up(size, sizeof(struct erofs_xattr_entry))
257
erofs_xattr_entry_size(struct erofs_xattr_entry * e)258 static inline unsigned int erofs_xattr_entry_size(struct erofs_xattr_entry *e)
259 {
260 return EROFS_XATTR_ALIGN(sizeof(struct erofs_xattr_entry) +
261 e->e_name_len + le16_to_cpu(e->e_value_size));
262 }
263
264 /* represent a zeroed chunk (hole) */
265 #define EROFS_NULL_ADDR -1
266
267 /* 4-byte block address array */
268 #define EROFS_BLOCK_MAP_ENTRY_SIZE sizeof(__le32)
269
270 /* 8-byte inode chunk index */
271 struct erofs_inode_chunk_index {
272 __le16 startblk_hi; /* starting block number MSB */
273 __le16 device_id; /* back-end storage id (with bits masked) */
274 __le32 startblk_lo; /* starting block number of this chunk */
275 };
276
277 #define EROFS_DIRENT_NID_METABOX_BIT 63
278 #define EROFS_DIRENT_NID_MASK (BIT_ULL(EROFS_DIRENT_NID_METABOX_BIT) - 1)
279
280 /* dirent sorts in alphabet order, thus we can do binary search */
281 struct erofs_dirent {
282 __le64 nid; /* node number */
283 __le16 nameoff; /* start offset of file name */
284 __u8 file_type; /* file type */
285 __u8 reserved; /* reserved */
286 } __packed;
287
288 /*
289 * EROFS file types should match generic FT_* types and
290 * it seems no need to add BUILD_BUG_ONs since potential
291 * unmatchness will break other fses as well...
292 */
293
294 #define EROFS_NAME_LEN 255
295
296 /* maximum supported encoded size of a physical compressed cluster */
297 #define Z_EROFS_PCLUSTER_MAX_SIZE (1024 * 1024)
298
299 /* maximum supported decoded size of a physical compressed cluster */
300 #define Z_EROFS_PCLUSTER_MAX_DSIZE (12 * 1024 * 1024)
301
302 /* available compression algorithm types (for h_algorithmtype) */
303 enum {
304 Z_EROFS_COMPRESSION_LZ4 = 0,
305 Z_EROFS_COMPRESSION_LZMA = 1,
306 Z_EROFS_COMPRESSION_DEFLATE = 2,
307 Z_EROFS_COMPRESSION_ZSTD = 3,
308 Z_EROFS_COMPRESSION_MAX
309 };
310 #define Z_EROFS_ALL_COMPR_ALGS ((1 << Z_EROFS_COMPRESSION_MAX) - 1)
311
312 /* 14 bytes (+ length field = 16 bytes) */
313 struct z_erofs_lz4_cfgs {
314 __le16 max_distance;
315 __le16 max_pclusterblks;
316 u8 reserved[10];
317 } __packed;
318
319 /* 14 bytes (+ length field = 16 bytes) */
320 struct z_erofs_lzma_cfgs {
321 __le32 dict_size;
322 __le16 format;
323 u8 reserved[8];
324 } __packed;
325
326 #define Z_EROFS_LZMA_MAX_DICT_SIZE (8 * Z_EROFS_PCLUSTER_MAX_SIZE)
327
328 /* 6 bytes (+ length field = 8 bytes) */
329 struct z_erofs_deflate_cfgs {
330 u8 windowbits; /* 8..15 for DEFLATE */
331 u8 reserved[5];
332 } __packed;
333
334 /* 6 bytes (+ length field = 8 bytes) */
335 struct z_erofs_zstd_cfgs {
336 u8 format;
337 u8 windowlog; /* windowLog - ZSTD_WINDOWLOG_ABSOLUTEMIN(10) */
338 u8 reserved[4];
339 } __packed;
340
341 #define Z_EROFS_ZSTD_MAX_DICT_SIZE Z_EROFS_PCLUSTER_MAX_SIZE
342
343 /*
344 * Enable COMPACTED_2B for EROFS_INODE_COMPRESSED_COMPACT inodes:
345 * 4B (disabled) vs 4B+2B+4B (enabled)
346 */
347 #define Z_EROFS_ADVISE_COMPACTED_2B 0x0001
348 /* Enable extent metadata for EROFS_INODE_COMPRESSED_FULL inodes */
349 #define Z_EROFS_ADVISE_EXTENTS 0x0001
350 #define Z_EROFS_ADVISE_BIG_PCLUSTER_1 0x0002
351 #define Z_EROFS_ADVISE_BIG_PCLUSTER_2 0x0004
352 #define Z_EROFS_ADVISE_INLINE_PCLUSTER 0x0008
353 #define Z_EROFS_ADVISE_INTERLACED_PCLUSTER 0x0010
354 #define Z_EROFS_ADVISE_FRAGMENT_PCLUSTER 0x0020
355 /* Indicate the record size for each extent if extent metadata is used */
356 #define Z_EROFS_ADVISE_EXTRECSZ_BIT 1
357 #define Z_EROFS_ADVISE_EXTRECSZ_MASK 0x3
358
359 #define Z_EROFS_FRAGMENT_INODE_BIT 7
360 struct z_erofs_map_header {
361 union {
362 /* fragment data offset in the packed inode */
363 __le32 h_fragmentoff;
364 struct {
365 __le16 h_reserved1;
366 /* indicates the encoded size of tailpacking data */
367 __le16 h_idata_size;
368 };
369 __le32 h_extents_lo; /* extent count LSB */
370 };
371 __le16 h_advise;
372 union {
373 struct {
374 /* algorithm type (bit 0-3: HEAD1; bit 4-7: HEAD2) */
375 __u8 h_algorithmtype;
376 /*
377 * bit 0-3 : logical cluster bits - blkszbits
378 * bit 4-6 : reserved
379 * bit 7 : pack the whole file into packed inode
380 */
381 __u8 h_clusterbits;
382 } __packed;
383 __le16 h_extents_hi; /* extent count MSB */
384 } __packed;
385 };
386
387 enum {
388 Z_EROFS_LCLUSTER_TYPE_PLAIN = 0,
389 Z_EROFS_LCLUSTER_TYPE_HEAD1 = 1,
390 Z_EROFS_LCLUSTER_TYPE_NONHEAD = 2,
391 Z_EROFS_LCLUSTER_TYPE_HEAD2 = 3,
392 Z_EROFS_LCLUSTER_TYPE_MAX
393 };
394
395 #define Z_EROFS_LI_LCLUSTER_TYPE_MASK (Z_EROFS_LCLUSTER_TYPE_MAX - 1)
396
397 /* (noncompact only, HEAD) This pcluster refers to partial decompressed data */
398 #define Z_EROFS_LI_PARTIAL_REF (1 << 15)
399
400 /* Set on 1st non-head lcluster to store compressed block counti (in blocks) */
401 #define Z_EROFS_LI_D0_CBLKCNT (1 << 11)
402
403 struct z_erofs_lcluster_index {
404 __le16 di_advise;
405 /* where to decompress in the head lcluster */
406 __le16 di_clusterofs;
407
408 union {
409 __le32 blkaddr; /* for the HEAD lclusters */
410 /*
411 * [0] - distance to its HEAD lcluster
412 * [1] - distance to the next HEAD lcluster
413 */
414 __le16 delta[2]; /* for the NONHEAD lclusters */
415 } di_u;
416 };
417
418 #define Z_EROFS_MAP_HEADER_END(end) \
419 (ALIGN(end, 8) + sizeof(struct z_erofs_map_header))
420 #define Z_EROFS_FULL_INDEX_START(end) (Z_EROFS_MAP_HEADER_END(end) + 8)
421
422 #define Z_EROFS_EXTENT_PLEN_PARTIAL BIT(27)
423 #define Z_EROFS_EXTENT_PLEN_FMT_BIT 28
424 #define Z_EROFS_EXTENT_PLEN_MASK ((Z_EROFS_PCLUSTER_MAX_SIZE << 1) - 1)
425 struct z_erofs_extent {
426 __le32 plen; /* encoded length */
427 __le32 pstart_lo; /* physical offset */
428 __le32 pstart_hi; /* physical offset MSB */
429 __le32 lstart_lo; /* logical offset */
430 __le32 lstart_hi; /* logical offset MSB (>= 4GiB inodes) */
431 __u8 reserved[12]; /* for future use */
432 };
433
z_erofs_extent_recsize(unsigned int advise)434 static inline int z_erofs_extent_recsize(unsigned int advise)
435 {
436 return 4 << ((advise >> Z_EROFS_ADVISE_EXTRECSZ_BIT) &
437 Z_EROFS_ADVISE_EXTRECSZ_MASK);
438 }
439
440 /* check the EROFS on-disk layout strictly at compile time */
erofs_check_ondisk_layout_definitions(void)441 static inline void erofs_check_ondisk_layout_definitions(void)
442 {
443 const __le64 fmh = *(__le64 *)&(struct z_erofs_map_header) {
444 .h_clusterbits = 1 << Z_EROFS_FRAGMENT_INODE_BIT
445 };
446
447 BUILD_BUG_ON(sizeof(struct erofs_super_block) != 144);
448 BUILD_BUG_ON(sizeof(struct erofs_inode_compact) != 32);
449 BUILD_BUG_ON(sizeof(struct erofs_inode_extended) != 64);
450 BUILD_BUG_ON(sizeof(struct erofs_xattr_ibody_header) != 12);
451 BUILD_BUG_ON(sizeof(struct erofs_xattr_entry) != 4);
452 BUILD_BUG_ON(sizeof(struct erofs_inode_chunk_info) != 4);
453 BUILD_BUG_ON(sizeof(struct erofs_inode_chunk_index) != 8);
454 BUILD_BUG_ON(sizeof(struct z_erofs_map_header) != 8);
455 BUILD_BUG_ON(sizeof(struct z_erofs_lcluster_index) != 8);
456 BUILD_BUG_ON(sizeof(struct erofs_dirent) != 12);
457 /* keep in sync between 2 index structures for better extendibility */
458 BUILD_BUG_ON(sizeof(struct erofs_inode_chunk_index) !=
459 sizeof(struct z_erofs_lcluster_index));
460 BUILD_BUG_ON(sizeof(struct erofs_deviceslot) != 128);
461
462 /* exclude old compiler versions like gcc 7.5.0 */
463 BUILD_BUG_ON(__builtin_constant_p(fmh) ?
464 fmh != cpu_to_le64(1ULL << 63) : 0);
465 }
466
467 #endif
468