xref: /original-bsd/sys/ufs/lfs/README (revision bacd16ee)
1#	@(#)README	8.1 (Berkeley) 06/11/93
2
3The file system is reasonably stable, but incomplete.  There are
4places where cleaning performance can be improved dramatically (see
5comments in lfs_syscalls.c).  For details on the implementation,
6performance and why garbage collection always wins, see Dr. Margo
7Seltzer's thesis available for anonymous ftp from toe.cs.berkeley.edu,
8in the directory pub/personal/margo/thesis.ps.Z, or the January 1993
9USENIX paper.
10
11Missing Functionality:
12	Multiple block sizes and/or fragments are not yet implemented.
13
14----------
15The disk is laid out in segments.  The first segment starts 8K into the
16disk (the first 8K is used for boot information).  Each segment is composed
17of the following:
18
19	An optional super block
20	One or more groups of:
21		segment summary
22		0 or more data blocks
23		0 or more inode blocks
24
25The segment summary and inode/data blocks start after the super block (if
26present), and grow toward the end of the segment.
27
28	_______________________________________________
29	|         |            |         |            |
30	| summary | data/inode | summary | data/inode |
31	|  block  |   blocks   |  block  |   blocks   | ...
32	|_________|____________|_________|____________|
33
34The data/inode blocks following a summary block are described by the
35summary block.  In order to permit the segment to be written in any order
36and in a forward direction only, a checksum is calculated across the
37blocks described by the summary.  Additionally, the summary is checksummed
38and timestamped.  Both of these are intended for recovery; the former is
39to make it easy to determine that it *is* a summary block and the latter
40is to make it easy to determine when recovery is finished for partially
41written segments.  These checksums are also used by the cleaner.
42
43	Summary block (detail)
44	________________
45	| sum cksum    |
46	| data cksum   |
47	| next segment |
48	| timestamp    |
49	| FINFO count  |
50	| inode count  |
51	| flags        |
52	|______________|
53	|   FINFO-1    | 0 or more file info structures, identifying the
54	|     .        | blocks in the segment.
55	|     .        |
56	|     .        |
57	|   FINFO-N    |
58	|   inode-N    |
59	|     .        |
60	|     .        |
61	|     .        | 0 or more inode daddr_t's, identifying the inode
62	|   inode-1    | blocks in the segment.
63	|______________|
64
65Inode blocks are blocks of on-disk inodes in the same format as those in
66the FFS.  However, spare[0] contains the inode number of the inode so we
67can find a particular inode on a page.  They are packed page_size /
68sizeof(inode) to a block.  Data blocks are exactly as in the FFS.  Both
69inodes and data blocks move around the file system at will.
70
71The file system is described by a super-block which is replicated and
72occurs as the first block of the first and other segments.  (The maximum
73number of super-blocks is MAXNUMSB).  Each super-block maintains a list
74of the disk addresses of all the super-blocks.  The super-block maintains
75a small amount of checkpoint information, essentially just enough to find
76the inode for the IFILE (fs->lfs_idaddr).
77
78The IFILE is visible in the file system, as inode number IFILE_INUM.  It
79contains information shared between the kernel and various user processes.
80
81	Ifile (detail)
82	________________
83	| cleaner info | Cleaner information per file system.  (Page
84	|              | granularity.)
85	|______________|
86	| segment      | Space available and last modified times per
87	| usage table  | segment.  (Page granularity.)
88	|______________|
89	|   IFILE-1    | Per inode status information: current version #,
90	|     .        | if currently allocated, last access time and
91	|     .        | current disk address of containing inode block.
92	|     .        | If current disk address is LFS_UNUSED_DADDR, the
93	|   IFILE-N    | inode is not in use, and it's on the free list.
94	|______________|
95
96
97First Segment at Creation Time:
98_____________________________________________________________
99|        |       |         |       |       |       |       |
100| 8K pad | Super | summary | inode | ifile | root  | l + f |
101|        | block |         | block |       | dir   | dir   |
102|________|_______|_________|_______|_______|_______|_______|
103	  ^
104           Segment starts here.
105
106Some differences from the Sprite LFS implementation.
107
1081. The LFS implementation placed the ifile metadata and the super block
109   at fixed locations.  This implementation replicates the super block
110   and puts each at a fixed location.  The checkpoint data is divided into
111   two parts -- just enough information to find the IFILE is stored in
112   two of the super blocks, although it is not toggled between them as in
113   the Sprite implementation.  (This was deliberate, to avoid a single
114   point of failure.)  The remaining checkpoint information is treated as
115   a regular file, which means that the cleaner info, the segment usage
116   table and the ifile meta-data are stored in normal log segments.
117   (Tastes great, less filling...)
118
1192. The segment layout is radically different in Sprite; this implementation
120   uses something a lot like network framing, where data/inode blocks are
121   written asynchronously, and a checksum is used to validate any set of
122   summary and data/inode blocks.  Sprite writes summary blocks synchronously
123   after the data/inode blocks have been written and the existence of the
124   summary block validates the data/inode blocks.  This permits us to write
125   everything contiguously, even partial segments and their summaries, whereas
126   Sprite is forced to seek (from the end of the data inode to the summary
127   which lives at the end of the segment).  Additionally, writing the summary
128   synchronously should cost about 1/2 a rotation per summary.
129
1303. Sprite LFS distinguishes between different types of blocks in the segment.
131   Other than inode blocks and data blocks, we don't.
132
1334. Sprite LFS traverses the IFILE looking for free blocks.  We maintain a
134   free list threaded through the IFILE entries.
135
1365. The cleaner runs in user space, as opposed to kernel space.  It shares
137   information with the kernel by reading/writing the IFILE and through
138   cleaner specific system calls.
139
140