xref: /netbsd-src/sys/ufs/lfs/README (revision ce0bb6e8d2e560ecacbe865a848624f94498063b)
1#	$NetBSD: README,v 1.2 1994/06/29 06:46:43 cgd Exp $
2
3#	@(#)README	8.1 (Berkeley) 6/11/93
4
5The file system is reasonably stable, but incomplete.  There are
6places where cleaning performance can be improved dramatically (see
7comments in lfs_syscalls.c).  For details on the implementation,
8performance and why garbage collection always wins, see Dr. Margo
9Seltzer's thesis available for anonymous ftp from toe.cs.berkeley.edu,
10in the directory pub/personal/margo/thesis.ps.Z, or the January 1993
11USENIX paper.
12
13Missing Functionality:
14	Multiple block sizes and/or fragments are not yet implemented.
15
16----------
17The disk is laid out in segments.  The first segment starts 8K into the
18disk (the first 8K is used for boot information).  Each segment is composed
19of the following:
20
21	An optional super block
22	One or more groups of:
23		segment summary
24		0 or more data blocks
25		0 or more inode blocks
26
27The segment summary and inode/data blocks start after the super block (if
28present), and grow toward the end of the segment.
29
30	_______________________________________________
31	|         |            |         |            |
32	| summary | data/inode | summary | data/inode |
33	|  block  |   blocks   |  block  |   blocks   | ...
34	|_________|____________|_________|____________|
35
36The data/inode blocks following a summary block are described by the
37summary block.  In order to permit the segment to be written in any order
38and in a forward direction only, a checksum is calculated across the
39blocks described by the summary.  Additionally, the summary is checksummed
40and timestamped.  Both of these are intended for recovery; the former is
41to make it easy to determine that it *is* a summary block and the latter
42is to make it easy to determine when recovery is finished for partially
43written segments.  These checksums are also used by the cleaner.
44
45	Summary block (detail)
46	________________
47	| sum cksum    |
48	| data cksum   |
49	| next segment |
50	| timestamp    |
51	| FINFO count  |
52	| inode count  |
53	| flags        |
54	|______________|
55	|   FINFO-1    | 0 or more file info structures, identifying the
56	|     .        | blocks in the segment.
57	|     .        |
58	|     .        |
59	|   FINFO-N    |
60	|   inode-N    |
61	|     .        |
62	|     .        |
63	|     .        | 0 or more inode daddr_t's, identifying the inode
64	|   inode-1    | blocks in the segment.
65	|______________|
66
67Inode blocks are blocks of on-disk inodes in the same format as those in
68the FFS.  However, spare[0] contains the inode number of the inode so we
69can find a particular inode on a page.  They are packed page_size /
70sizeof(inode) to a block.  Data blocks are exactly as in the FFS.  Both
71inodes and data blocks move around the file system at will.
72
73The file system is described by a super-block which is replicated and
74occurs as the first block of the first and other segments.  (The maximum
75number of super-blocks is MAXNUMSB).  Each super-block maintains a list
76of the disk addresses of all the super-blocks.  The super-block maintains
77a small amount of checkpoint information, essentially just enough to find
78the inode for the IFILE (fs->lfs_idaddr).
79
80The IFILE is visible in the file system, as inode number IFILE_INUM.  It
81contains information shared between the kernel and various user processes.
82
83	Ifile (detail)
84	________________
85	| cleaner info | Cleaner information per file system.  (Page
86	|              | granularity.)
87	|______________|
88	| segment      | Space available and last modified times per
89	| usage table  | segment.  (Page granularity.)
90	|______________|
91	|   IFILE-1    | Per inode status information: current version #,
92	|     .        | if currently allocated, last access time and
93	|     .        | current disk address of containing inode block.
94	|     .        | If current disk address is LFS_UNUSED_DADDR, the
95	|   IFILE-N    | inode is not in use, and it's on the free list.
96	|______________|
97
98
99First Segment at Creation Time:
100_____________________________________________________________
101|        |       |         |       |       |       |       |
102| 8K pad | Super | summary | inode | ifile | root  | l + f |
103|        | block |         | block |       | dir   | dir   |
104|________|_______|_________|_______|_______|_______|_______|
105	  ^
106           Segment starts here.
107
108Some differences from the Sprite LFS implementation.
109
1101. The LFS implementation placed the ifile metadata and the super block
111   at fixed locations.  This implementation replicates the super block
112   and puts each at a fixed location.  The checkpoint data is divided into
113   two parts -- just enough information to find the IFILE is stored in
114   two of the super blocks, although it is not toggled between them as in
115   the Sprite implementation.  (This was deliberate, to avoid a single
116   point of failure.)  The remaining checkpoint information is treated as
117   a regular file, which means that the cleaner info, the segment usage
118   table and the ifile meta-data are stored in normal log segments.
119   (Tastes great, less filling...)
120
1212. The segment layout is radically different in Sprite; this implementation
122   uses something a lot like network framing, where data/inode blocks are
123   written asynchronously, and a checksum is used to validate any set of
124   summary and data/inode blocks.  Sprite writes summary blocks synchronously
125   after the data/inode blocks have been written and the existence of the
126   summary block validates the data/inode blocks.  This permits us to write
127   everything contiguously, even partial segments and their summaries, whereas
128   Sprite is forced to seek (from the end of the data inode to the summary
129   which lives at the end of the segment).  Additionally, writing the summary
130   synchronously should cost about 1/2 a rotation per summary.
131
1323. Sprite LFS distinguishes between different types of blocks in the segment.
133   Other than inode blocks and data blocks, we don't.
134
1354. Sprite LFS traverses the IFILE looking for free blocks.  We maintain a
136   free list threaded through the IFILE entries.
137
1385. The cleaner runs in user space, as opposed to kernel space.  It shares
139   information with the kernel by reading/writing the IFILE and through
140   cleaner specific system calls.
141
142