doc/BACKUP.md: Document backup strategies for lightningd.

ChangeLog-Added: Document: `doc/BACKUP.md` describes how to back up your C-lightning node.

.github/CODEOWNERS

@@ -9,6 +9,7 @@ wallet/                 @cdecker
 
 wallet/invoices.*       @ZmnSCPxj
 plugins/multifundchannel.c @ZmnSCPxj
+doc/BACKUP.md           @ZmnSCPxj @cdecker
 
 common/param.*          @wythe
 common/json.*           @wythe

doc/BACKUP.md

@@ -0,0 +1,312 @@
+Backing Up Your C-Lightning Node
+================================
+
+Lightning Network channels get their scalability and privacy benefits
+from the very simple technique of *not telling anyone else about your
+in-channel activity*.
+This is in contrast to onchain payments, where you have to tell everyone
+about each and every payment and have it recorded on the blockchain,
+leading to scaling problems (you have to push data at everyone) and
+privacy problems (everyone knows every payment you were ever involved in).
+
+Unfortunately, this removes a property that onchain users are so used
+to, they react in surprise when learning about this removal.
+Your onchain activity is recorded in all archival fullnodes, so if you
+forget all your onchain activity because your storage got fried, you
+just go redownload the activity from the nearest archival fullnode.
+
+But in Lightning, since *you* are the only one storing all your
+financial information, you ***cannot*** recover this financial
+information anywhere else.
+
+This means that on Lightning, you **have to** responsibly back up your
+financial information yourself, using various processes and automation.
+
+The discussion below assumes that you know where you put your
+`$LIGHTNINGDIR`, and you know the directory structure within.
+By default your `$LIGHTNINGDIR` will be in `~/.lightning/${COIN}`.
+For example, if you are running `--mainnet`, it will be
+`~/.lightning/bitcoin`.
+
+`hsm_secret`
+------------
+
+You need a copy of the `hsm_secret` file regardless of whatever backup
+strategy you use.
+This is just 32 bytes, and you can do something like the below and
+write the hexadecimal digits a few times on a piece of paper:
+
+    cd $LIGHTNINGDIR
+    xxd hsm_secret
+
+You can re-enter the hexdump into a text file later and use `xxd` to
+convert it back to a binary `hsm_secret`:
+
+    cat > hsm_secret_hex.txt <<HEX
+    00: 30cc f221 94e1 7f01 cd54 d68c a1ba f124
+    10: e1f3 1d45 d904 823c 77b7 1e18 fd93 1676
+    HEX
+    xxd -r hsm_secret_hex.txt > hsm_secret
+    chmod 0400 hsm_secret
+
+Notice that you need to ensure that the `hsm_secret` is only readable by
+the user, and is not writable, as otherwise `lightningd` will refuse to
+start.
+Hence the `chmod 0400 hsm_secret` command.
+
+Alternately, if you are deploying a new node that has no funds and
+channels yet, you can generate BIP39 words using any process, and
+create the `hsm_secret` using the `hsmtool generatehsm` command.
+If you did `make install` then `hsmtool` is installed as
+`lightning-hsmtool`, else you can find it in the `tools/` directory
+of the build directory.
+
+    lightning-hsmtool generatehsm hsm_secret
+
+Then enter the BIP39 words, plus an optional passphrase.
+
+You can regenerate the same `hsm_secret` file using the same BIP39
+words, which again, you can back up on paper.
+
+Recovery of the `hsm_secret` is sufficient to recover any onchain
+funds.
+Recovery of the `hsm_secret` is necessary, but insufficient, to recover
+any in-channel funds.
+To recover in-channel funds, you need to use one or more of the other
+backup strategies below.
+
+Database File Backups
+---------------------
+
+This is the least desirable backup strategy, as it *can* lead to loss
+of all in-channel funds if you use it.
+However, having *no* backup strategy at all *will* lead to loss of all
+in-channel funds, so this is still better than nothing.
+
+While `lightningd` is not running, just copy the `lightningd.sqlite3` file
+in the `$LIGHTNINGDIR` on backup media somewhere.
+
+To recover, just copy the backed up `lightningd.sqlite3` into your new
+`$LIGHTNINGDIR` together with the `hsm_secret`.
+
+You can also use any automated backup system as long as it includes the
+`lightningd.sqlite3` file (and optionally `hsm_secret`, but note that
+as a secret key, hackers getting a copy of your backups may allow them
+to steal your funds, even in-channel funds).
+
+This backup method is undesirable, since it cannot recover the following
+channels:
+
+* Channels with peers that do not support `option_dataloss_protect`.
+  * Most nodes on the network already support `option_dataloss_protect`
+    as of November 2020.
+  * If the peer does not support this option, then the entire channel
+    funds will be revoked by the peer.
+  * Specially-coded peers can pretend to support this option, but end
+    up taking the entire channel funds (or otherwise more than what it
+    should have) when you try to recover.
+    Standard releases of Lightning Network node software do not do this
+    (they just support this option according to the standard) but note
+    that anyone can modify their node software, especially with our
+    well-documented open-source code.
+* Channels created *after* the copy was made are not recoverable.
+  * Data for those channels does not exist in the backup, so your node
+    cannot recover them.
+
+Because of the above, this strategy is discouraged: you *can* potentially
+lose all funds in open channels.
+
+However, again, note that a "no backups #reckless" strategy leads to
+*definite* loss of funds, so you should still prefer this strategy rather
+than not have **any** backups at all.
+
+Even if you have the better options below, you might still want to do
+this as a fallback, as long as you ensure:
+
+* Attempt to recover using the other backup options below first.
+  Any one of them will be better than this backup option.
+* Recovering by this method ***MUST*** always be the ***last*** resort.
+* Recover using the most recent backup you can find.
+  Take time to look for the most recent available backup.
+
+Again, this strategy can lead to only ***partial*** recovery of funds,
+or even to complete failure to recover, so use the other methods below
+first to recover!
+
+### Backing Up While `lightningd` Is Running
+
+Since `sqlite3` will be writing to the file while `lightningd` is running,
+`cp`ing the `lightningd.sqlite3` file while `lighningd` is running may
+result in the file not being copied properly if `sqlite3` happens to be
+committing database transactions at that time, potentially leading to a
+corrupted backup file that cannot be recovered from.
+
+Instead, the **proper** way to do this would be:
+
+* `echo "VACUUM INTO 'backup.sqlite3';" | sqlite3 lightningd.sqlite3`
+* `sync`
+* `mv backup.sqlite3 ${YOURBACKUPLOCATION}/lightningd.sqlite3`
+
+This creates a consistent snapshot of the database, sampled in a
+transaction, that is assured to be openable later by `sqlite3`.
+The operation of the `lightningd` process will be paused while `sqlite3`
+is creating the backup, as well, since the `VACUUM INTO` command will
+be in a transaction that prevents the `lightningd` process from starting
+its own database transactions.
+
+You can also directly give the path to the backup media in the `VACUUM
+INTO` command, just remember to `sync` afterwards, and if the backup
+media is removable, properly unmount it before removing it.
+
+An advantage of the `VACUUM INTO` command is that it also removes the
+space used by deleted rows, meaning the backup it creates is smaller.
+
+Filesystem Redundancy
+---------------------
+
+You can set up a RAID-1 with multiple storage devices, and point the
+`$LIGHTNINGDIR` to the RAID-1 setup.
+That way, failure of one storage device will still let you recover
+funds.
+
+On a Linux system, one of the simpler things you can do would be to use
+BTRFS RAID-1 setup between a partition on your primary storage and a USB
+flash disk.
+The below "should" work, but assumes you are comfortable with low-level
+Linux administration.
+If you are on a system that would make you cry if you break it, you **MUST**
+stop your Lightning node and back up all files before doing the below.
+
+* Install `btrfs-progs` or `btrfs-tools` or equivalent.
+* Get a 32Gb USB flash disk.
+* Stop your Lightning node and back up everything, do not be stupid.
+* Repartition your hard disk to have a 30Gb partition.
+  * This is risky and may lose your data, so this is best done with a
+    brand-new hard disk that contains no data.
+* Connect the USB flash disk.
+* Find the `/dev/sdXX` devices for the HDD 30Gb partition and the flash disk.
+  * `lsblk -o NAME,TYPE,SIZE,MODEL` should help.
+* Create a RAID-1 `btrfs` filesystem.
+  * `mkfs.btrfs -m raid1 -d raid1 /dev/${HDD30GB} /dev/${USB32GB}`
+  * You may need to add `-f` if the USB flash disk is already formatted.
+* Create a mountpoint for the `btrfs` filesystem.
+* Create a `/etc/fstab` entry.
+  * Use the `UUID` option instad of `/dev/sdXX` since the exact device letter
+    can change across boots.
+  * You can get the UUID by `lsblk -o NAME,UUID`.
+    Specifying *either* of the devices is sufficient.
+  * e.g. `UUID=${UUID} ${BTRFSMOUNTPOINT} btrfs defaults 0 0`
+* `mount -a` then `df` to confirm it got mounted.
+* Copy the contents of the `$LIGHTNINGDIR` to the BTRFS mount point.
+  * Copy the entire directory, then `chown -R` the copy to the user who will
+    run the `lightningd`.
+  * If you are paranoid, run `diff -R` on both copies to check.
+* Remove the existing `$LIGHTNINGDIR`.
+* `ln -s ${BTRFSMOUNTPOINT}/lightningdirname ${LIGHTNINGDIR}`.
+  * Make sure the `$LIGHTNINGDIR` has the same structure as what you
+    originally had.
+* Add `crontab` entries for `root` that perform regular `btrfs` maintenance
+  tasks.
+  * `0 0 * * * /usr/bin/btrfs balance start -dusage=50 -dlimit=2 -musage=50 -mlimit=4 ${BTRFSMOUNTPOINT}`
+    This prevents BTRFS from running out of blocks even if it has unused
+    space *within* blocks, and is run at midnight everyday.
+    You may need to change the path to the `btrfs` binary.
+  * `0 0 * * 0 /usr/bin/btrfs scrub start -B -c 2 -n 4 ${BTRFSMOUNTPOINT}`
+    This detects bit rot (i.e. bad sectors) and auto-heals the filesystem,
+    and is run on Sundays at midnight.
+* Restart your Lightning node.
+
+If one or the other device fails completely, shut down your computer, boot
+on a recovery disk or similar, then:
+
+* Connect the surviving device.
+* Mount the partition/USB flash disk in `degraded` mode:
+  * `mount -o degraded /dev/sdXX /mnt/point`
+* Copy the `lightningd.sqlite3` and `hsm_secret` to new media.
+  * Do **not** write to the degraded `btrfs` mount!
+* Start up a `lightningd` using the `hsm_secret` and `lightningd.sqlite3`
+  and close all channels and move all funds to onchain cold storage you
+  control, then set up a new Lightning node.
+
+If the device that fails is the USB flash disk, you can replace it using
+BTRFS commands.
+You should probably stop your Lightning node while doing this.
+
+* `btrfs replace start /dev/sdOLD /dev/sdNEW ${BTRFSMOUNTPOINT}`.
+* Monitor status with `btrfs replace status ${BTRFSMOUNTPOINT}`.
+
+More sophisticated setups with more than two devices are possible.
+Take note that "RAID 1" in `btrfs` means "data is copied on up to two
+devices", meaning only up to one device can fail.
+You may be interested in `raid1c3` and `raid1c4` modes if you have
+three or four storage devices.
+BTRFS would probably work better if you were purchasing an entire set
+of new storage devices to set up a new node.
+
+On BSD you can use a ZFS RAID-Z setup, which is probably better than BTRFS
+on Linux (in particular the equivalent to the `crontab` entries on Linux
+BTRFS are automatically run in the background by ZFS).
+ZFS *can* be installed and used on Linux but requires greater effort to
+install on a typical Linux system;
+try looking at [ZFSonLinux](https://zfsonlinux.org).
+
+Filesystem RAID-1 like the above protects against failure of a single storage
+device, but does not protect you in case of certain disasters, such as fire or
+computer confiscation.
+
+`backup` Plugin And Remote NFS Mount
+------------------------------------
+
+You can get the `backup` plugin here:
+https://github.com/lightningd/plugins/tree/master/backup
+
+The `backup` plugin requires Python 3.
+
+* `cd` into its directory and install requirements.
+  * `pip3 install -r requirements.txt`
+* Figure out where you will put the backup files.
+  * Ideally you have an NFS or other network-based mount on your system,
+    into which you will put the backup.
+* Stop your Lightning node.
+* `/path/to/backup-cli init ${LIGHTNINGDIR} file:///path/to/nfs/mount`.
+  This creates an initial copy of the database at the NFS mount.
+* Add these settings to your `lightningd` configuration:
+  * `important-plugin=/path/to/backup.py`
+  * `backup-destination=file:///path/to/nfs/mount`
+* Restart your Lightning node.
+
+It is recommended that you use a network-mounted filesystem for the backup
+destination.
+
+Alternately, you *could* put it in another storage device (e.g. USB flash
+disk) on the same physical location, but if so, you might want to use the
+above `btrfs` RAID-1 instead, as that tends to be more performant
+(though using this `backup` plugin does not require you to repartition an
+existing hard disk, which might matter to you if you are using an
+everyday-use laptop or desktop).
+
+To recover:
+
+* Re-download the `backup` plugin and install Python 3 and the
+  requirements of `backup`.
+* `/path/to/backup-cli restore file:///path/to/nfs/mount ${LIGHTNINGDIR}`
+
+If your backup destination is a network-mounted filesystem that is in a
+remote location, then even loss of all hardware in one location will allow
+you to still recover your Lightning funds.
+
+However, if instead you are just replicating the database on another
+storage device on a single location, you remain vulnerable to disasters
+like fire or computer confiscation.
+
+PostgreSQL Cluster
+------------------
+
+`lightningd` may also be compiled with PostgreSQL support.
+PostgreSQL is generally faster than SQLITE3, and also supports running a
+PostgreSQL cluster to be used by `lightningd`, with automatic replication
+and turnover in case an entire node of the PostgreSQL cluster fails.
+
+Setting this up, however, is more involved.
+
+TODO: @cdecker

doc/index.rst

@@ -8,6 +8,7 @@ c-lightning Documentation
    INSTALL.md
    TOR.md
    FAQ <FAQ.md>
+   Backups <BACKUP.md>
 
 .. toctree::
    :maxdepth: 2