| blob | a65baffb4ebf8cf17bafe3d8567547a2ec644502 |
1 .. SPDX-License-Identifier: GPL-2.0
3 Index Nodes
4 -----------
6 In a regular UNIX filesystem, the inode stores all the metadata
7 pertaining to the file (time stamps, block maps, extended attributes,
8 etc), not the directory entry. To find the information associated with a
9 file, one must traverse the directory files to find the directory entry
10 associated with a file, then load the inode to find the metadata for
11 that file. ext4 appears to cheat (for performance reasons) a little bit
12 by storing a copy of the file type (normally stored in the inode) in the
13 directory entry. (Compare all this to FAT, which stores all the file
14 information directly in the directory entry, but does not support hard
15 links and is in general more seek-happy than ext4 due to its simpler
16 block allocator and extensive use of linked lists.)
18 The inode table is a linear array of ``struct ext4_inode``. The table is
19 sized to have enough blocks to store at least
20 ``sb.s_inode_size * sb.s_inodes_per_group`` bytes. The number of the
21 block group containing an inode can be calculated as
22 ``(inode_number - 1) / sb.s_inodes_per_group``, and the offset into the
23 group's table is ``(inode_number - 1) % sb.s_inodes_per_group``. There
24 is no inode 0.
26 The inode checksum is calculated against the FS UUID, the inode number,
27 and the inode structure itself.
29 The inode table entry is laid out in ``struct ext4_inode``.
31 .. list-table::
32 :widths: 8 8 24 40
33 :header-rows: 1
34 :class: longtable
36 * - Offset
37 - Size
38 - Name
39 - Description
40 * - 0x0
41 - \_\_le16
42 - i\_mode
43 - File mode. See the table i_mode_ below.
44 * - 0x2
45 - \_\_le16
46 - i\_uid
47 - Lower 16-bits of Owner UID.
48 * - 0x4
49 - \_\_le32
50 - i\_size\_lo
51 - Lower 32-bits of size in bytes.
52 * - 0x8
53 - \_\_le32
54 - i\_atime
55 - Last access time, in seconds since the epoch. However, if the EA\_INODE
56 inode flag is set, this inode stores an extended attribute value and
57 this field contains the checksum of the value.
58 * - 0xC
59 - \_\_le32
60 - i\_ctime
61 - Last inode change time, in seconds since the epoch. However, if the
62 EA\_INODE inode flag is set, this inode stores an extended attribute
63 value and this field contains the lower 32 bits of the attribute value's
64 reference count.
65 * - 0x10
66 - \_\_le32
67 - i\_mtime
68 - Last data modification time, in seconds since the epoch. However, if the
69 EA\_INODE inode flag is set, this inode stores an extended attribute
70 value and this field contains the number of the inode that owns the
71 extended attribute.
72 * - 0x14
73 - \_\_le32
74 - i\_dtime
75 - Deletion Time, in seconds since the epoch.
76 * - 0x18
77 - \_\_le16
78 - i\_gid
79 - Lower 16-bits of GID.
80 * - 0x1A
81 - \_\_le16
82 - i\_links\_count
83 - Hard link count. Normally, ext4 does not permit an inode to have more
84 than 65,000 hard links. This applies to files as well as directories,
85 which means that there cannot be more than 64,998 subdirectories in a
86 directory (each subdirectory's '..' entry counts as a hard link, as does
87 the '.' entry in the directory itself). With the DIR\_NLINK feature
88 enabled, ext4 supports more than 64,998 subdirectories by setting this
89 field to 1 to indicate that the number of hard links is not known.
90 * - 0x1C
91 - \_\_le32
92 - i\_blocks\_lo
93 - Lower 32-bits of “block” count. If the huge\_file feature flag is not
94 set on the filesystem, the file consumes ``i_blocks_lo`` 512-byte blocks
95 on disk. If huge\_file is set and EXT4\_HUGE\_FILE\_FL is NOT set in
96 ``inode.i_flags``, then the file consumes ``i_blocks_lo + (i_blocks_hi
97 << 32)`` 512-byte blocks on disk. If huge\_file is set and
98 EXT4\_HUGE\_FILE\_FL IS set in ``inode.i_flags``, then this file
99 consumes (``i_blocks_lo + i_blocks_hi`` << 32) filesystem blocks on
100 disk.
101 * - 0x20
102 - \_\_le32
103 - i\_flags
104 - Inode flags. See the table i_flags_ below.
105 * - 0x24
106 - 4 bytes
107 - i\_osd1
108 - See the table i_osd1_ for more details.
109 * - 0x28
110 - 60 bytes
111 - i\_block[EXT4\_N\_BLOCKS=15]
112 - Block map or extent tree. See the section “The Contents of inode.i\_block”.
113 * - 0x64
114 - \_\_le32
115 - i\_generation
116 - File version (for NFS).
117 * - 0x68
118 - \_\_le32
119 - i\_file\_acl\_lo
120 - Lower 32-bits of extended attribute block. ACLs are of course one of
121 many possible extended attributes; I think the name of this field is a
122 result of the first use of extended attributes being for ACLs.
123 * - 0x6C
124 - \_\_le32
125 - i\_size\_high / i\_dir\_acl
126 - Upper 32-bits of file/directory size. In ext2/3 this field was named
127 i\_dir\_acl, though it was usually set to zero and never used.
128 * - 0x70
129 - \_\_le32
130 - i\_obso\_faddr
131 - (Obsolete) fragment address.
132 * - 0x74
133 - 12 bytes
134 - i\_osd2
135 - See the table i_osd2_ for more details.
136 * - 0x80
137 - \_\_le16
138 - i\_extra\_isize
139 - Size of this inode - 128. Alternately, the size of the extended inode
140 fields beyond the original ext2 inode, including this field.
141 * - 0x82
142 - \_\_le16
143 - i\_checksum\_hi
144 - Upper 16-bits of the inode checksum.
145 * - 0x84
146 - \_\_le32
147 - i\_ctime\_extra
148 - Extra change time bits. This provides sub-second precision. See Inode
149 Timestamps section.
150 * - 0x88
151 - \_\_le32
152 - i\_mtime\_extra
153 - Extra modification time bits. This provides sub-second precision.
154 * - 0x8C
155 - \_\_le32
156 - i\_atime\_extra
157 - Extra access time bits. This provides sub-second precision.
158 * - 0x90
159 - \_\_le32
160 - i\_crtime
161 - File creation time, in seconds since the epoch.
162 * - 0x94
163 - \_\_le32
164 - i\_crtime\_extra
165 - Extra file creation time bits. This provides sub-second precision.
166 * - 0x98
167 - \_\_le32
168 - i\_version\_hi
169 - Upper 32-bits for version number.
170 * - 0x9C
171 - \_\_le32
172 - i\_projid
173 - Project ID.
175 .. _i_mode:
177 The ``i_mode`` value is a combination of the following flags:
179 .. list-table::
180 :widths: 16 64
181 :header-rows: 1
183 * - Value
184 - Description
185 * - 0x1
186 - S\_IXOTH (Others may execute)
187 * - 0x2
188 - S\_IWOTH (Others may write)
189 * - 0x4
190 - S\_IROTH (Others may read)
191 * - 0x8
192 - S\_IXGRP (Group members may execute)
193 * - 0x10
194 - S\_IWGRP (Group members may write)
195 * - 0x20
196 - S\_IRGRP (Group members may read)
197 * - 0x40
198 - S\_IXUSR (Owner may execute)
199 * - 0x80
200 - S\_IWUSR (Owner may write)
201 * - 0x100
202 - S\_IRUSR (Owner may read)
203 * - 0x200
204 - S\_ISVTX (Sticky bit)
205 * - 0x400
206 - S\_ISGID (Set GID)
207 * - 0x800
208 - S\_ISUID (Set UID)
209 * -
210 - These are mutually-exclusive file types:
211 * - 0x1000
212 - S\_IFIFO (FIFO)
213 * - 0x2000
214 - S\_IFCHR (Character device)
215 * - 0x4000
216 - S\_IFDIR (Directory)
217 * - 0x6000
218 - S\_IFBLK (Block device)
219 * - 0x8000
220 - S\_IFREG (Regular file)
221 * - 0xA000
222 - S\_IFLNK (Symbolic link)
223 * - 0xC000
224 - S\_IFSOCK (Socket)
226 .. _i_flags:
228 The ``i_flags`` field is a combination of these values:
230 .. list-table::
231 :widths: 16 64
232 :header-rows: 1
234 * - Value
235 - Description
236 * - 0x1
237 - This file requires secure deletion (EXT4\_SECRM\_FL). (not implemented)
238 * - 0x2
239 - This file should be preserved, should undeletion be desired
240 (EXT4\_UNRM\_FL). (not implemented)
241 * - 0x4
242 - File is compressed (EXT4\_COMPR\_FL). (not really implemented)
243 * - 0x8
244 - All writes to the file must be synchronous (EXT4\_SYNC\_FL).
245 * - 0x10
246 - File is immutable (EXT4\_IMMUTABLE\_FL).
247 * - 0x20
248 - File can only be appended (EXT4\_APPEND\_FL).
249 * - 0x40
250 - The dump(1) utility should not dump this file (EXT4\_NODUMP\_FL).
251 * - 0x80
252 - Do not update access time (EXT4\_NOATIME\_FL).
253 * - 0x100
254 - Dirty compressed file (EXT4\_DIRTY\_FL). (not used)
255 * - 0x200
256 - File has one or more compressed clusters (EXT4\_COMPRBLK\_FL). (not used)
257 * - 0x400
258 - Do not compress file (EXT4\_NOCOMPR\_FL). (not used)
259 * - 0x800
260 - Encrypted inode (EXT4\_ENCRYPT\_FL). This bit value previously was
261 EXT4\_ECOMPR\_FL (compression error), which was never used.
262 * - 0x1000
263 - Directory has hashed indexes (EXT4\_INDEX\_FL).
264 * - 0x2000
265 - AFS magic directory (EXT4\_IMAGIC\_FL).
266 * - 0x4000
267 - File data must always be written through the journal
268 (EXT4\_JOURNAL\_DATA\_FL).
269 * - 0x8000
270 - File tail should not be merged (EXT4\_NOTAIL\_FL). (not used by ext4)
271 * - 0x10000
272 - All directory entry data should be written synchronously (see
273 ``dirsync``) (EXT4\_DIRSYNC\_FL).
274 * - 0x20000
275 - Top of directory hierarchy (EXT4\_TOPDIR\_FL).
276 * - 0x40000
277 - This is a huge file (EXT4\_HUGE\_FILE\_FL).
278 * - 0x80000
279 - Inode uses extents (EXT4\_EXTENTS\_FL).
280 * - 0x100000
281 - Verity protected file (EXT4\_VERITY\_FL).
282 * - 0x200000
283 - Inode stores a large extended attribute value in its data blocks
284 (EXT4\_EA\_INODE\_FL).
285 * - 0x400000
286 - This file has blocks allocated past EOF (EXT4\_EOFBLOCKS\_FL).
287 (deprecated)
288 * - 0x01000000
289 - Inode is a snapshot (``EXT4_SNAPFILE_FL``). (not in mainline)
290 * - 0x04000000
291 - Snapshot is being deleted (``EXT4_SNAPFILE_DELETED_FL``). (not in
292 mainline)
293 * - 0x08000000
294 - Snapshot shrink has completed (``EXT4_SNAPFILE_SHRUNK_FL``). (not in
295 mainline)
296 * - 0x10000000
297 - Inode has inline data (EXT4\_INLINE\_DATA\_FL).
298 * - 0x20000000
299 - Create children with the same project ID (EXT4\_PROJINHERIT\_FL).
300 * - 0x80000000
301 - Reserved for ext4 library (EXT4\_RESERVED\_FL).
302 * -
303 - Aggregate flags:
304 * - 0x705BDFFF
305 - User-visible flags.
306 * - 0x604BC0FF
307 - User-modifiable flags. Note that while EXT4\_JOURNAL\_DATA\_FL and
308 EXT4\_EXTENTS\_FL can be set with setattr, they are not in the kernel's
309 EXT4\_FL\_USER\_MODIFIABLE mask, since it needs to handle the setting of
310 these flags in a special manner and they are masked out of the set of
311 flags that are saved directly to i\_flags.
313 .. _i_osd1:
315 The ``osd1`` field has multiple meanings depending on the creator:
317 Linux:
319 .. list-table::
320 :widths: 8 8 24 40
321 :header-rows: 1
323 * - Offset
324 - Size
325 - Name
326 - Description
327 * - 0x0
328 - \_\_le32
329 - l\_i\_version
330 - Inode version. However, if the EA\_INODE inode flag is set, this inode
331 stores an extended attribute value and this field contains the upper 32
332 bits of the attribute value's reference count.
334 Hurd:
336 .. list-table::
337 :widths: 8 8 24 40
338 :header-rows: 1
340 * - Offset
341 - Size
342 - Name
343 - Description
344 * - 0x0
345 - \_\_le32
346 - h\_i\_translator
347 - ??
349 Masix:
351 .. list-table::
352 :widths: 8 8 24 40
353 :header-rows: 1
355 * - Offset
356 - Size
357 - Name
358 - Description
359 * - 0x0
360 - \_\_le32
361 - m\_i\_reserved
362 - ??
364 .. _i_osd2:
366 The ``osd2`` field has multiple meanings depending on the filesystem creator:
368 Linux:
370 .. list-table::
371 :widths: 8 8 24 40
372 :header-rows: 1
374 * - Offset
375 - Size
376 - Name
377 - Description
378 * - 0x0
379 - \_\_le16
380 - l\_i\_blocks\_high
381 - Upper 16-bits of the block count. Please see the note attached to
382 i\_blocks\_lo.
383 * - 0x2
384 - \_\_le16
385 - l\_i\_file\_acl\_high
386 - Upper 16-bits of the extended attribute block (historically, the file
387 ACL location). See the Extended Attributes section below.
388 * - 0x4
389 - \_\_le16
390 - l\_i\_uid\_high
391 - Upper 16-bits of the Owner UID.
392 * - 0x6
393 - \_\_le16
394 - l\_i\_gid\_high
395 - Upper 16-bits of the GID.
396 * - 0x8
397 - \_\_le16
398 - l\_i\_checksum\_lo
399 - Lower 16-bits of the inode checksum.
400 * - 0xA
401 - \_\_le16
402 - l\_i\_reserved
403 - Unused.
405 Hurd:
407 .. list-table::
408 :widths: 8 8 24 40
409 :header-rows: 1
411 * - Offset
412 - Size
413 - Name
414 - Description
415 * - 0x0
416 - \_\_le16
417 - h\_i\_reserved1
418 - ??
419 * - 0x2
420 - \_\_u16
421 - h\_i\_mode\_high
422 - Upper 16-bits of the file mode.
423 * - 0x4
424 - \_\_le16
425 - h\_i\_uid\_high
426 - Upper 16-bits of the Owner UID.
427 * - 0x6
428 - \_\_le16
429 - h\_i\_gid\_high
430 - Upper 16-bits of the GID.
431 * - 0x8
432 - \_\_u32
433 - h\_i\_author
434 - Author code?
436 Masix:
438 .. list-table::
439 :widths: 8 8 24 40
440 :header-rows: 1
442 * - Offset
443 - Size
444 - Name
445 - Description
446 * - 0x0
447 - \_\_le16
448 - h\_i\_reserved1
449 - ??
450 * - 0x2
451 - \_\_u16
452 - m\_i\_file\_acl\_high
453 - Upper 16-bits of the extended attribute block (historically, the file
454 ACL location).
455 * - 0x4
456 - \_\_u32
457 - m\_i\_reserved2[2]
458 - ??
460 Inode Size
461 ~~~~~~~~~~
463 In ext2 and ext3, the inode structure size was fixed at 128 bytes
464 (``EXT2_GOOD_OLD_INODE_SIZE``) and each inode had a disk record size of
465 128 bytes. Starting with ext4, it is possible to allocate a larger
466 on-disk inode at format time for all inodes in the filesystem to provide
467 space beyond the end of the original ext2 inode. The on-disk inode
468 record size is recorded in the superblock as ``s_inode_size``. The
469 number of bytes actually used by struct ext4\_inode beyond the original
470 128-byte ext2 inode is recorded in the ``i_extra_isize`` field for each
471 inode, which allows struct ext4\_inode to grow for a new kernel without
472 having to upgrade all of the on-disk inodes. Access to fields beyond
473 EXT2\_GOOD\_OLD\_INODE\_SIZE should be verified to be within
474 ``i_extra_isize``. By default, ext4 inode records are 256 bytes, and (as
475 of August 2019) the inode structure is 160 bytes
476 (``i_extra_isize = 32``). The extra space between the end of the inode
477 structure and the end of the inode record can be used to store extended
478 attributes. Each inode record can be as large as the filesystem block
479 size, though this is not terribly efficient.
481 Finding an Inode
482 ~~~~~~~~~~~~~~~~
484 Each block group contains ``sb->s_inodes_per_group`` inodes. Because
485 inode 0 is defined not to exist, this formula can be used to find the
486 block group that an inode lives in:
487 ``bg = (inode_num - 1) / sb->s_inodes_per_group``. The particular inode
488 can be found within the block group's inode table at
489 ``index = (inode_num - 1) % sb->s_inodes_per_group``. To get the byte
490 address within the inode table, use
491 ``offset = index * sb->s_inode_size``.
493 Inode Timestamps
494 ~~~~~~~~~~~~~~~~
496 Four timestamps are recorded in the lower 128 bytes of the inode
497 structure -- inode change time (ctime), access time (atime), data
498 modification time (mtime), and deletion time (dtime). The four fields
499 are 32-bit signed integers that represent seconds since the Unix epoch
500 (1970-01-01 00:00:00 GMT), which means that the fields will overflow in
501 January 2038. For inodes that are not linked from any directory but are
502 still open (orphan inodes), the dtime field is overloaded for use with
503 the orphan list. The superblock field ``s_last_orphan`` points to the
504 first inode in the orphan list; dtime is then the number of the next
505 orphaned inode, or zero if there are no more orphans.
507 If the inode structure size ``sb->s_inode_size`` is larger than 128
508 bytes and the ``i_inode_extra`` field is large enough to encompass the
509 respective ``i_[cma]time_extra`` field, the ctime, atime, and mtime
510 inode fields are widened to 64 bits. Within this “extra” 32-bit field,
511 the lower two bits are used to extend the 32-bit seconds field to be 34
512 bit wide; the upper 30 bits are used to provide nanosecond timestamp
513 accuracy. Therefore, timestamps should not overflow until May 2446.
514 dtime was not widened. There is also a fifth timestamp to record inode
515 creation time (crtime); this field is 64-bits wide and decoded in the
516 same manner as 64-bit [cma]time. Neither crtime nor dtime are accessible
517 through the regular stat() interface, though debugfs will report them.
519 We use the 32-bit signed time value plus (2^32 \* (extra epoch bits)).
520 In other words:
522 .. list-table::
523 :widths: 20 20 20 20 20
524 :header-rows: 1
526 * - Extra epoch bits
527 - MSB of 32-bit time
528 - Adjustment for signed 32-bit to 64-bit tv\_sec
529 - Decoded 64-bit tv\_sec
530 - valid time range
531 * - 0 0
532 - 1
533 - 0
534 - ``-0x80000000 - -0x00000001``
535 - 1901-12-13 to 1969-12-31
536 * - 0 0
537 - 0
538 - 0
539 - ``0x000000000 - 0x07fffffff``
540 - 1970-01-01 to 2038-01-19
541 * - 0 1
542 - 1
543 - 0x100000000
544 - ``0x080000000 - 0x0ffffffff``
545 - 2038-01-19 to 2106-02-07
546 * - 0 1
547 - 0
548 - 0x100000000
549 - ``0x100000000 - 0x17fffffff``
550 - 2106-02-07 to 2174-02-25
551 * - 1 0
552 - 1
553 - 0x200000000
554 - ``0x180000000 - 0x1ffffffff``
555 - 2174-02-25 to 2242-03-16
556 * - 1 0
557 - 0
558 - 0x200000000
559 - ``0x200000000 - 0x27fffffff``
560 - 2242-03-16 to 2310-04-04
561 * - 1 1
562 - 1
563 - 0x300000000
564 - ``0x280000000 - 0x2ffffffff``
565 - 2310-04-04 to 2378-04-22
566 * - 1 1
567 - 0
568 - 0x300000000
569 - ``0x300000000 - 0x37fffffff``
570 - 2378-04-22 to 2446-05-10
572 This is a somewhat odd encoding since there are effectively seven times
573 as many positive values as negative values. There have also been
574 long-standing bugs decoding and encoding dates beyond 2038, which don't
575 seem to be fixed as of kernel 3.12 and e2fsprogs 1.42.8. 64-bit kernels
576 incorrectly use the extra epoch bits 1,1 for dates between 1901 and
577 1970. At some point the kernel will be fixed and e2fsck will fix this
578 situation, assuming that it is run before 2310.