2 The LogFS Flash Filesystem
3 ==========================
11 Two superblocks exist at the beginning and end of the filesystem.
12 Each superblock is 256 Bytes large, with another 3840 Bytes reserved
13 for future purposes, making a total of 4096 Bytes.
15 Superblock locations may differ for MTD and block devices. On MTD the
16 first non-bad block contains a superblock in the first 4096 Bytes and
17 the last non-bad block contains a superblock in the last 4096 Bytes.
18 On block devices, the first 4096 Bytes of the device contain the first
19 superblock and the last aligned 4096 Byte-block contains the second
22 For the most part, the superblocks can be considered read-only. They
23 are written only to correct errors detected within the superblocks,
24 move the journal and change the filesystem parameters through tunefs.
25 As a result, the superblock does not contain any fields that require
26 constant updates, like the amount of free space, etc.
31 The space in the device is split up into equal-sized segments.
32 Segments are the primary write unit of LogFS. Within each segments,
33 writes happen from front (low addresses) to back (high addresses. If
34 only a partial segment has been written, the segment number, the
35 current position within and optionally a write buffer are stored in
38 Segments are erased as a whole. Therefore Garbage Collection may be
39 required to completely free a segment before doing so.
44 The journal contains all global information about the filesystem that
45 is subject to frequent change. At mount time, it has to be scanned
46 for the most recent commit entry, which contains a list of pointers to
47 all currently valid entries.
52 All space except for the superblocks and journal is part of the object
53 store. Each segment contains a segment header and a number of
54 objects, each consisting of the object header and the payload.
55 Objects are either inodes, directory entries (dentries), file data
56 blocks or indirect blocks.
61 Garbage collection (GC) may fail if all data is written
62 indiscriminately. One requirement of GC is that data is seperated
63 roughly according to the distance between the tree root and the data.
64 Effectively that means all file data is on level 0, indirect blocks
65 are on levels 1, 2, 3 4 or 5 for 1x, 2x, 3x, 4x or 5x indirect blocks,
66 respectively. Inode file data is on level 6 for the inodes and 7-11
69 Each segment contains objects of a single level only. As a result,
70 each level requires its own seperate segment to be open for writing.
75 All inodes are stored in a special file, the inode file. Single
76 exception is the inode file's inode (master inode) which for obvious
77 reasons is stored in the journal instead. Instead of data blocks, the
78 leaf nodes of the inode files are inodes.
83 Writes in LogFS are done by means of a wandering tree. A naïve
84 implementation would require that for each write or a block, all
85 parent blocks are written as well, since the block pointers have
86 changed. Such an implementation would not be very efficient.
88 In LogFS, the block pointer changes are cached in the journal by means
89 of alias entries. Each alias consists of its logical address - inode
90 number, block index, level and child number (index into block) - and
91 the changed data. Any 8-byte word can be changes in this manner.
93 Currently aliases are used for block pointers, file size, file used
94 bytes and the height of an inodes indirect tree.
99 Related to regular aliases, these are used to handle bad blocks.
100 Initially, bad blocks are handled by moving the affected segment
101 content to a spare segment and noting this move in the journal with a
102 segment alias, a simple (to, from) tupel. GC will later empty this
103 segment and the alias can be removed again. This is used on MTD only.
108 By cleverly predicting the life time of data, it is possible to
109 seperate long-living data from short-living data and thereby reduce
110 the GC overhead later. Each type of distinc life expectency (vim) can
111 have a seperate segment open for writing. Each (level, vim) tupel can
112 be open just once. If an open segment with unknown vim is encountered
113 at mount time, it is closed and ignored henceforth.
118 Inodes in LogFS are similar to FFS-style filesystems with direct and
119 indirect block pointers. One difference is that LogFS uses a single
120 indirect pointer that can be either a 1x, 2x, etc. indirect pointer.
121 A height field in the inode defines the height of the indirect tree
122 and thereby the indirection of the pointer.
124 Another difference is the addressing of indirect blocks. In LogFS,
125 the first 16 pointers in the first indirect block are left empty,
126 corresponding to the 16 direct pointers in the inode. In ext2 (maybe
127 others as well) the first pointer in the first indirect block
128 corresponds to logical block 12, skipping the 12 direct pointers.
129 So where ext2 is using arithmetic to better utilize space, LogFS keeps
130 arithmetic simple and uses compression to save space.
135 Both file data and metadata can be compressed. Compression for file
136 data can be enabled with chattr +c and disabled with chattr -c. Doing
137 so has no effect on existing data, but new data will be stored
138 accordingly. New inodes will inherit the compression flag of the
141 Metadata is always compressed. However, the space accounting ignores
142 this and charges for the uncompressed size. Failing to do so could
143 result in GC failures when, after moving some data, indirect blocks
144 compress worse than previously. Even on a 100% full medium, GC may
145 not consume any extra space, so the compression gains are lost space
148 However, they are not lost space to the filesystem internals. By
149 cheating the user for those bytes, the filesystem gained some slack
150 space and GC will run less often and faster.
152 Garbage Collection and Wear Leveling
153 ------------------------------------
155 Garbage collection is invoked whenever the number of free segments
156 falls below a threshold. The best (known) candidate is picked based
157 on the least amount of valid data contained in the segment. All
158 remaining valid data is copied elsewhere, thereby invalidating it.
160 The GC code also checks for aliases and writes then back if their
161 number gets too large.
163 Wear leveling is done by occasionally picking a suboptimal segment for
164 garbage collection. If a stale segments erase count is significantly
165 lower than the active segments' erase counts, it will be picked. Wear
166 leveling is rate limited, so it will never monopolize the device for
167 more than one segment worth at a time.
169 Values for "occasionally", "significantly lower" are compile time
175 To satisfy efficient lookup(), directory entries are hashed and
176 located based on the hash. In order to both support large directories
177 and not be overly inefficient for small directories, several hash
178 tables of increasing size are used. For each table, the hash value
179 modulo the table size gives the table index.
181 Tables sizes are chosen to limit the number of indirect blocks with a
182 fully populated table to 0, 1, 2 or 3 respectively. So the first
183 table contains 16 entries, the second 512-16, etc.
185 The last table is special in several ways. First its size depends on
186 the effective 32bit limit on telldir/seekdir cookies. Since logfs
187 uses the upper half of the address space for indirect blocks, the size
188 is limited to 2^31. Secondly the table contains hash buckets with 16
191 Using single-entry buckets would result in birthday "attacks". At
192 just 2^16 used entries, hash collisions would be likely (P >= 0.5).
193 My math skills are insufficient to do the combinatorics for the 17x
194 collisions necessary to overflow a bucket, but testing showed that in
195 10,000 runs the lowest directory fill before a bucket overflow was
196 188,057,130 entries with an average of 315,149,915 entries. So for
197 directory sizes of up to a million, bucket overflows should be
198 virtually impossible under normal circumstances.
200 With carefully chosen filenames, it is obviously possible to cause an
201 overflow with just 21 entries (4 higher tables + 16 entries + 1). So
202 there may be a security concern if a malicious user has write access
211 A device address space is used for caching. Both block devices and
212 MTD provide functions to either read a single page or write a segment.
213 Partial segments may be written for data integrity, but where possible
214 complete segments are written for performance on simple block device
220 Inodes are stored in the inode file, which is just a regular file for
221 most purposes. At umount time, however, the inode file needs to
222 remain open until all dirty inodes are written. So
223 generic_shutdown_super() may not close this inode, but shouldn't
224 complain about remaining inodes due to the inode file either. Same
225 goes for mapping inode of the device address space.
227 Currently logfs uses a hack that essentially copies part of fs/inode.c
228 code over. A general solution would be preferred.
230 Indirect block mapping
231 ----------------------
233 With compression, the block device (or mapping inode) cannot be used
234 to cache indirect blocks. Some other place is required. Currently
235 logfs uses the top half of each inode's address space. The low 8TB
236 (on 32bit) are filled with file data, the high 8TB are used for
239 One problem is that 16TB files created on 64bit systems actually have
240 data in the top 8TB. But files >16TB would cause problems anyway, so
241 only the limit has changed.