mirror of
https://mirrors.bfsu.edu.cn/git/linux.git
synced 2024-11-19 02:04:19 +08:00
a8cd4561ea
s/seperate/separate Signed-off-by: Anand Gadiyar <gadiyar@ti.com> Signed-off-by: Jiri Kosina <jkosina@suse.cz>
242 lines
9.6 KiB
Plaintext
242 lines
9.6 KiB
Plaintext
|
|
The LogFS Flash Filesystem
|
|
==========================
|
|
|
|
Specification
|
|
=============
|
|
|
|
Superblocks
|
|
-----------
|
|
|
|
Two superblocks exist at the beginning and end of the filesystem.
|
|
Each superblock is 256 Bytes large, with another 3840 Bytes reserved
|
|
for future purposes, making a total of 4096 Bytes.
|
|
|
|
Superblock locations may differ for MTD and block devices. On MTD the
|
|
first non-bad block contains a superblock in the first 4096 Bytes and
|
|
the last non-bad block contains a superblock in the last 4096 Bytes.
|
|
On block devices, the first 4096 Bytes of the device contain the first
|
|
superblock and the last aligned 4096 Byte-block contains the second
|
|
superblock.
|
|
|
|
For the most part, the superblocks can be considered read-only. They
|
|
are written only to correct errors detected within the superblocks,
|
|
move the journal and change the filesystem parameters through tunefs.
|
|
As a result, the superblock does not contain any fields that require
|
|
constant updates, like the amount of free space, etc.
|
|
|
|
Segments
|
|
--------
|
|
|
|
The space in the device is split up into equal-sized segments.
|
|
Segments are the primary write unit of LogFS. Within each segments,
|
|
writes happen from front (low addresses) to back (high addresses. If
|
|
only a partial segment has been written, the segment number, the
|
|
current position within and optionally a write buffer are stored in
|
|
the journal.
|
|
|
|
Segments are erased as a whole. Therefore Garbage Collection may be
|
|
required to completely free a segment before doing so.
|
|
|
|
Journal
|
|
--------
|
|
|
|
The journal contains all global information about the filesystem that
|
|
is subject to frequent change. At mount time, it has to be scanned
|
|
for the most recent commit entry, which contains a list of pointers to
|
|
all currently valid entries.
|
|
|
|
Object Store
|
|
------------
|
|
|
|
All space except for the superblocks and journal is part of the object
|
|
store. Each segment contains a segment header and a number of
|
|
objects, each consisting of the object header and the payload.
|
|
Objects are either inodes, directory entries (dentries), file data
|
|
blocks or indirect blocks.
|
|
|
|
Levels
|
|
------
|
|
|
|
Garbage collection (GC) may fail if all data is written
|
|
indiscriminately. One requirement of GC is that data is separated
|
|
roughly according to the distance between the tree root and the data.
|
|
Effectively that means all file data is on level 0, indirect blocks
|
|
are on levels 1, 2, 3 4 or 5 for 1x, 2x, 3x, 4x or 5x indirect blocks,
|
|
respectively. Inode file data is on level 6 for the inodes and 7-11
|
|
for indirect blocks.
|
|
|
|
Each segment contains objects of a single level only. As a result,
|
|
each level requires its own separate segment to be open for writing.
|
|
|
|
Inode File
|
|
----------
|
|
|
|
All inodes are stored in a special file, the inode file. Single
|
|
exception is the inode file's inode (master inode) which for obvious
|
|
reasons is stored in the journal instead. Instead of data blocks, the
|
|
leaf nodes of the inode files are inodes.
|
|
|
|
Aliases
|
|
-------
|
|
|
|
Writes in LogFS are done by means of a wandering tree. A naïve
|
|
implementation would require that for each write or a block, all
|
|
parent blocks are written as well, since the block pointers have
|
|
changed. Such an implementation would not be very efficient.
|
|
|
|
In LogFS, the block pointer changes are cached in the journal by means
|
|
of alias entries. Each alias consists of its logical address - inode
|
|
number, block index, level and child number (index into block) - and
|
|
the changed data. Any 8-byte word can be changes in this manner.
|
|
|
|
Currently aliases are used for block pointers, file size, file used
|
|
bytes and the height of an inodes indirect tree.
|
|
|
|
Segment Aliases
|
|
---------------
|
|
|
|
Related to regular aliases, these are used to handle bad blocks.
|
|
Initially, bad blocks are handled by moving the affected segment
|
|
content to a spare segment and noting this move in the journal with a
|
|
segment alias, a simple (to, from) tupel. GC will later empty this
|
|
segment and the alias can be removed again. This is used on MTD only.
|
|
|
|
Vim
|
|
---
|
|
|
|
By cleverly predicting the life time of data, it is possible to
|
|
separate long-living data from short-living data and thereby reduce
|
|
the GC overhead later. Each type of distinc life expectency (vim) can
|
|
have a separate segment open for writing. Each (level, vim) tupel can
|
|
be open just once. If an open segment with unknown vim is encountered
|
|
at mount time, it is closed and ignored henceforth.
|
|
|
|
Indirect Tree
|
|
-------------
|
|
|
|
Inodes in LogFS are similar to FFS-style filesystems with direct and
|
|
indirect block pointers. One difference is that LogFS uses a single
|
|
indirect pointer that can be either a 1x, 2x, etc. indirect pointer.
|
|
A height field in the inode defines the height of the indirect tree
|
|
and thereby the indirection of the pointer.
|
|
|
|
Another difference is the addressing of indirect blocks. In LogFS,
|
|
the first 16 pointers in the first indirect block are left empty,
|
|
corresponding to the 16 direct pointers in the inode. In ext2 (maybe
|
|
others as well) the first pointer in the first indirect block
|
|
corresponds to logical block 12, skipping the 12 direct pointers.
|
|
So where ext2 is using arithmetic to better utilize space, LogFS keeps
|
|
arithmetic simple and uses compression to save space.
|
|
|
|
Compression
|
|
-----------
|
|
|
|
Both file data and metadata can be compressed. Compression for file
|
|
data can be enabled with chattr +c and disabled with chattr -c. Doing
|
|
so has no effect on existing data, but new data will be stored
|
|
accordingly. New inodes will inherit the compression flag of the
|
|
parent directory.
|
|
|
|
Metadata is always compressed. However, the space accounting ignores
|
|
this and charges for the uncompressed size. Failing to do so could
|
|
result in GC failures when, after moving some data, indirect blocks
|
|
compress worse than previously. Even on a 100% full medium, GC may
|
|
not consume any extra space, so the compression gains are lost space
|
|
to the user.
|
|
|
|
However, they are not lost space to the filesystem internals. By
|
|
cheating the user for those bytes, the filesystem gained some slack
|
|
space and GC will run less often and faster.
|
|
|
|
Garbage Collection and Wear Leveling
|
|
------------------------------------
|
|
|
|
Garbage collection is invoked whenever the number of free segments
|
|
falls below a threshold. The best (known) candidate is picked based
|
|
on the least amount of valid data contained in the segment. All
|
|
remaining valid data is copied elsewhere, thereby invalidating it.
|
|
|
|
The GC code also checks for aliases and writes then back if their
|
|
number gets too large.
|
|
|
|
Wear leveling is done by occasionally picking a suboptimal segment for
|
|
garbage collection. If a stale segments erase count is significantly
|
|
lower than the active segments' erase counts, it will be picked. Wear
|
|
leveling is rate limited, so it will never monopolize the device for
|
|
more than one segment worth at a time.
|
|
|
|
Values for "occasionally", "significantly lower" are compile time
|
|
constants.
|
|
|
|
Hashed directories
|
|
------------------
|
|
|
|
To satisfy efficient lookup(), directory entries are hashed and
|
|
located based on the hash. In order to both support large directories
|
|
and not be overly inefficient for small directories, several hash
|
|
tables of increasing size are used. For each table, the hash value
|
|
modulo the table size gives the table index.
|
|
|
|
Tables sizes are chosen to limit the number of indirect blocks with a
|
|
fully populated table to 0, 1, 2 or 3 respectively. So the first
|
|
table contains 16 entries, the second 512-16, etc.
|
|
|
|
The last table is special in several ways. First its size depends on
|
|
the effective 32bit limit on telldir/seekdir cookies. Since logfs
|
|
uses the upper half of the address space for indirect blocks, the size
|
|
is limited to 2^31. Secondly the table contains hash buckets with 16
|
|
entries each.
|
|
|
|
Using single-entry buckets would result in birthday "attacks". At
|
|
just 2^16 used entries, hash collisions would be likely (P >= 0.5).
|
|
My math skills are insufficient to do the combinatorics for the 17x
|
|
collisions necessary to overflow a bucket, but testing showed that in
|
|
10,000 runs the lowest directory fill before a bucket overflow was
|
|
188,057,130 entries with an average of 315,149,915 entries. So for
|
|
directory sizes of up to a million, bucket overflows should be
|
|
virtually impossible under normal circumstances.
|
|
|
|
With carefully chosen filenames, it is obviously possible to cause an
|
|
overflow with just 21 entries (4 higher tables + 16 entries + 1). So
|
|
there may be a security concern if a malicious user has write access
|
|
to a directory.
|
|
|
|
Open For Discussion
|
|
===================
|
|
|
|
Device Address Space
|
|
--------------------
|
|
|
|
A device address space is used for caching. Both block devices and
|
|
MTD provide functions to either read a single page or write a segment.
|
|
Partial segments may be written for data integrity, but where possible
|
|
complete segments are written for performance on simple block device
|
|
flash media.
|
|
|
|
Meta Inodes
|
|
-----------
|
|
|
|
Inodes are stored in the inode file, which is just a regular file for
|
|
most purposes. At umount time, however, the inode file needs to
|
|
remain open until all dirty inodes are written. So
|
|
generic_shutdown_super() may not close this inode, but shouldn't
|
|
complain about remaining inodes due to the inode file either. Same
|
|
goes for mapping inode of the device address space.
|
|
|
|
Currently logfs uses a hack that essentially copies part of fs/inode.c
|
|
code over. A general solution would be preferred.
|
|
|
|
Indirect block mapping
|
|
----------------------
|
|
|
|
With compression, the block device (or mapping inode) cannot be used
|
|
to cache indirect blocks. Some other place is required. Currently
|
|
logfs uses the top half of each inode's address space. The low 8TB
|
|
(on 32bit) are filled with file data, the high 8TB are used for
|
|
indirect blocks.
|
|
|
|
One problem is that 16TB files created on 64bit systems actually have
|
|
data in the top 8TB. But files >16TB would cause problems anyway, so
|
|
only the limit has changed.
|