Initial WIP new user's manual

This commit is contained in:
Kent Overstreet 2020-07-25 16:20:56 -04:00
parent 20a7160f9a
commit 21ade39653
2 changed files with 113 additions and 0 deletions

View File

@ -114,6 +114,9 @@ endif
# Rebuild the 'version' command any time the version string changes
cmd_version.o : .version
doc/bcachefs.5: doc/bcachefs.5.txt
a2x -f manpage doc/bcachefs.5.txt
.PHONY: install
install: INITRAMFS_HOOK=$(INITRAMFS_DIR)/hooks/bcachefs
install: INITRAMFS_SCRIPT=$(INITRAMFS_DIR)/scripts/local-premount/bcachefs

110
doc/bcachefs.5.txt Normal file
View File

@ -0,0 +1,110 @@
BCACHEFS(5)
===========
NAME
----
bcachefs - bcachefs overview, user's manual and configuration
DESCRIPTION
-----------
Bcachefs is a multi device copy on write filesystem that supports
Checksumming
Compression
Encryption
Reflink
Caching
Replication
Erasure coding (reed-solomon)
And more. This document is intended to be an overview of the various features
and use cases.
Configuration
-------------
Most configuration is done via filesystem options that can be set at format
time, mount time (as mount -o parameters), or changed at runtime via sysfs (via
the /sys/fs/bcachefs/<UUID>/options/ directory).
Many of those options (particularly those that control the IO path) can also be
set on individual files and directories, via the bcachefs setattr command (which
internally mostly works via the extended attribute interface, but the setattr
command takes care to propagate options to children correctly).
* TODO: include master list of options from opts.h
#include "opts.mdwn"
Device management
-----------------
Devices can be added, removed, and resized at will, at runtime. There is no
fixed topology or data layout, as with hardware RAID or ZFS, and devices need
not be the same size - the allocator will stripe across multiple disks,
preferring to allocate from disks with more free space so that disks all fill up
at the same time.
We generally avoid per-device options, preferring instead options that can be
overridden on files or directories, but there are some:
*durability*
Device labels, targets
----------------------
Configuration options that point to targets (i.e. a disk or group of disks) may
be passed either a device (i.e. /dev/sda), or a label. Labels are assigned to
disks (and need not be unique), and these labels form a nested heirarchy: this
allows disks to be grouped together and referred to either individually or as a
group.
For example, given disks formatted with these labels:
bcachefs format -g controller1.hdd.hdd1 /dev/sda \
-g controller1.hdd.hdd2 /dev/sdb \
-g controller1.ssd.ssd1 /dev/sdc \
-g controller1.ssd.ssd1 /dev/sdd \
-g controller2.hdd1 /dev/sde \
-g controller2.hdd2 /dev/sdf
Configuration options such as foreground_target may then refer to controller1,
or controller1.hdd, or controller1.hdd1 - or to /dev/sda directly.
Data placement, caching
-----------------------
The following options control which disks data is written to:
* foreground_target
* background_target
* promote_target
The foreground_target option is used to direct writes from applications. The
background_target option, if set, will cause data to be moved to that target in
the background by the rebalance thread some time after it has been initially
written - leaving behind the original copy, but marking it cached so that it can
be discarded by the allocator. The promote_target will will cause reads to write
a cached copy of the data being read to that target, if it doesn't exist.
Together, these options can be used for writeback caching, like so:
foregroud_target=ssd
background_target=hdd
promote_target=ssd
Writethrough caching requires telling bcachefs not to trust the cache device,
which does require a per-device option and thus can't completely be done with
per-file options. This is done by setting the device's durability to 0.
These options can all be set on individual files or directories. They can also
be used to pin a specific file or directory to a specific device or target:
foreground_target=ssd
background_target=
promote_target=
Note that if the target specified is full, the write will spill over to the rest
of the filesystem.
Data protection
---------------
foo