1 # $NetBSD: README,v 1.3 1999/03/15 00:46:47 perseant Exp $
3 # @(#)README 8.1 (Berkeley) 6/11/93
5 The file system is reasonably stable...I think.
7 For details on the implementation, performance and why garbage
8 collection always wins, see Dr. Margo Seltzer's thesis available for
9 anonymous ftp from toe.cs.berkeley.edu, in the directory
10 pub/personal/margo/thesis.ps.Z, or the January 1993 USENIX paper.
13 The disk is laid out in segments. The first segment starts 8K into the
14 disk (the first 8K is used for boot information). Each segment is composed
17 An optional super block
18 One or more groups of:
21 0 or more inode blocks
23 The segment summary and inode/data blocks start after the super block (if
24 present), and grow toward the end of the segment.
26 _______________________________________________
28 | summary | data/inode | summary | data/inode |
29 | block | blocks | block | blocks | ...
30 |_________|____________|_________|____________|
32 The data/inode blocks following a summary block are described by the
33 summary block. In order to permit the segment to be written in any order
34 and in a forward direction only, a checksum is calculated across the
35 blocks described by the summary. Additionally, the summary is checksummed
36 and timestamped. Both of these are intended for recovery; the former is
37 to make it easy to determine that it *is* a summary block and the latter
38 is to make it easy to determine when recovery is finished for partially
39 written segments. These checksums are also used by the cleaner.
41 Summary block (detail)
51 | FINFO-1 | 0 or more file info structures, identifying the
52 | . | blocks in the segment.
59 | . | 0 or more inode daddr_t's, identifying the inode
60 | inode-1 | blocks in the segment.
63 Inode blocks are blocks of on-disk inodes in the same format as those in
64 the FFS. However, spare[0] contains the inode number of the inode so we
65 can find a particular inode on a page. They are packed page_size /
66 sizeof(inode) to a block. Data blocks are exactly as in the FFS. Both
67 inodes and data blocks move around the file system at will.
69 The file system is described by a super-block which is replicated and
70 occurs as the first block of the first and other segments. (The maximum
71 number of super-blocks is MAXNUMSB). Each super-block maintains a list
72 of the disk addresses of all the super-blocks. The super-block maintains
73 a small amount of checkpoint information, essentially just enough to find
74 the inode for the IFILE (fs->lfs_idaddr).
76 The IFILE is visible in the file system, as inode number IFILE_INUM. It
77 contains information shared between the kernel and various user processes.
81 | cleaner info | Cleaner information per file system. (Page
84 | segment | Space available and last modified times per
85 | usage table | segment. (Page granularity.)
87 | IFILE-1 | Per inode status information: current version #,
88 | . | if currently allocated, last access time and
89 | . | current disk address of containing inode block.
90 | . | If current disk address is LFS_UNUSED_DADDR, the
91 | IFILE-N | inode is not in use, and it's on the free list.
95 First Segment at Creation Time:
96 _____________________________________________________________
98 | 8K pad | Super | summary | inode | ifile | root | l + f |
99 | | block | | block | | dir | dir |
100 |________|_______|_________|_______|_______|_______|_______|
104 Some differences from the Sprite LFS implementation.
106 1. The LFS implementation placed the ifile metadata and the super block
107 at fixed locations. This implementation replicates the super block
108 and puts each at a fixed location. The checkpoint data is divided into
109 two parts -- just enough information to find the IFILE is stored in
110 two of the super blocks, although it is not toggled between them as in
111 the Sprite implementation. (This was deliberate, to avoid a single
112 point of failure.) The remaining checkpoint information is treated as
113 a regular file, which means that the cleaner info, the segment usage
114 table and the ifile meta-data are stored in normal log segments.
115 (Tastes great, less filling...)
117 2. The segment layout is radically different in Sprite; this implementation
118 uses something a lot like network framing, where data/inode blocks are
119 written asynchronously, and a checksum is used to validate any set of
120 summary and data/inode blocks. Sprite writes summary blocks synchronously
121 after the data/inode blocks have been written and the existence of the
122 summary block validates the data/inode blocks. This permits us to write
123 everything contiguously, even partial segments and their summaries, whereas
124 Sprite is forced to seek (from the end of the data inode to the summary
125 which lives at the end of the segment). Additionally, writing the summary
126 synchronously should cost about 1/2 a rotation per summary.
128 3. Sprite LFS distinguishes between different types of blocks in the segment.
129 Other than inode blocks and data blocks, we don't.
131 4. Sprite LFS traverses the IFILE looking for free blocks. We maintain a
132 free list threaded through the IFILE entries.
134 5. The cleaner runs in user space, as opposed to kernel space. It shares
135 information with the kernel by reading/writing the IFILE and through
136 cleaner specific system calls.