summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--docs/BLOCK_DEVICE_LOCKING.md19
-rw-r--r--man/udevadm.xml116
2 files changed, 135 insertions, 0 deletions
diff --git a/docs/BLOCK_DEVICE_LOCKING.md b/docs/BLOCK_DEVICE_LOCKING.md
index 13ae3f6e04..40fc61c671 100644
--- a/docs/BLOCK_DEVICE_LOCKING.md
+++ b/docs/BLOCK_DEVICE_LOCKING.md
@@ -75,6 +75,25 @@ And please keep in mind: BSD file locks (`flock()`) and POSIX file locks
orthogonal. The scheme discussed above uses the former and not the latter,
because these types of locks more closely match the required semantics.
+If multiple devices are to be locked at the same time (for example in order to
+format a RAID file system), the devices should be locked in the order of the
+the device nodes' major numbers (primary ordering key, ascending) and minor
+numbers (secondary ordering key, ditto), in order to avoid ABBA locking issues
+between subsystems.
+
+Note that the locks should only be taken while the device is repartitioned,
+file systems formatted or `dd`'ed in, and similar cases that
+apply/remove/change superblocks/partition information. It should not be held
+during normal operation, i.e. while file systems on it are mounted for
+application use.
+
+The [`udevadm
+lock`](https://www.freedesktop.org/software/systemd/man/udevadm.html) command
+is provided to lock block devices following this scheme from the command line,
+for the use in scripts and similar. (Note though that it's typically preferable
+to use native support for block device locking in tools where that's
+available.)
+
Summarizing: it is recommended to take `LOCK_EX` BSD file locks when
manipulating block devices in all tools that change file system block devices
(`mkfs`, `fsck`, …) or partition tables (`fdisk`, `parted`, …), right after
diff --git a/man/udevadm.xml b/man/udevadm.xml
index e299a75879..3248cfd256 100644
--- a/man/udevadm.xml
+++ b/man/udevadm.xml
@@ -51,6 +51,9 @@
<cmdsynopsis>
<command>udevadm wait <optional>options</optional> <replaceable>device|syspath</replaceable></command>
</cmdsynopsis>
+ <cmdsynopsis>
+ <command>udevadm lock <optional>options</optional> <replaceable>command</replaceable></command>
+ </cmdsynopsis>
</refsynopsisdiv>
<refsect1><title>Description</title>
@@ -747,6 +750,87 @@
<xi:include href="standard-options.xml" xpointer="help" />
</variablelist>
</refsect2>
+
+ <refsect2>
+ <title>udevadm lock
+ <arg choice="opt"><replaceable>options</replaceable></arg>
+ <arg choice="opt"><replaceable>command</replaceable></arg>
+ …
+ </title>
+
+ <para><command>udevadm lock</command> takes an (advisory) exclusive lock(s) on a block device (or
+ multiple therof), as per <ulink url="https://systemd.io/BLOCK_DEVICE_LOCKING">Locking Block Device
+ Access</ulink> and invokes a program with the lock(s) taken. When the invoked program exits the lock(s)
+ are automatically released.</para>
+
+ <para>This tool is in particular useful to ensure that
+ <citerefentry><refentrytitle>systemd-udevd.service</refentrytitle><manvolnum>8</manvolnum></citerefentry>
+ does not probe a block device while changes are made to it, for example partitions created or file
+ systems formatted. Note that many tools that interface with block devices natively support taking
+ relevant locks, see for example
+ <citerefentry><refentrytitle>sfdisk</refentrytitle><manvolnum>8</manvolnum></citerefentry>'s
+ <option>--lock</option> switch.</para>
+
+ <para>The command expects at least one block device specified via <option>--device=</option> or
+ <option>--backing=</option>, and a command line to execute as arguments.</para>
+
+ <variablelist>
+ <varlistentry>
+ <term><option>--device=<replaceable>DEVICE</replaceable></option></term>
+ <term><option>-d <replaceable>DEVICE</replaceable></option></term>
+
+ <listitem><para>Takes a path to a device node of the device to lock. This switch may be used
+ multiple times (and in combination with <option>--backing=</option>) in order to lock multiple
+ devices. If a partition block device node is specified the containing "whole" block device is
+ automatically determined and used for the lock, as per the specification. If multiple devices are
+ specified, they are deduplicated, sorted by the major/minor of their device nodes and then locked
+ in order.</para>
+
+ <para>This switch must be used at least once, to specify at least one device to
+ lock. (Alternatively, use <option>--backing=</option>, see below.)</para></listitem>
+ </varlistentry>
+
+ <varlistentry>
+ <term><option>--backing=<replaceable>PATH</replaceable></option></term>
+ <term><option>-b <replaceable>PATH</replaceable></option></term>
+
+ <listitem><para>If a path to a device node is specified, identical to
+ <option>--device=</option>. However, this switch alternatively accepts a path to a regular file or
+ directory, in which case the block device of the file system the file/directory resides on is
+ automatically determined and used as if it was specified with
+ <option>--device=</option>.</para></listitem>
+ </varlistentry>
+
+ <varlistentry>
+ <term><option>--timeout=<replaceable>SECS</replaceable></option></term>
+ <term><option>-t <replaceable>SECS</replaceable></option></term>
+
+ <listitem><para>Specifies how long to wait at most until all locks can be taken. Takes a value in
+ seconds, or in the usual supported time units, see
+ <citerefentry><refentrytitle>systemd.time</refentrytitle><manvolnum>7</manvolnum></citerefentry>. If
+ specified as zero the lock is attempted and if not successful the invocation will immediately
+ fail. If passed as <literal>infinity</literal> (the default) the invocation will wait indefinitely
+ until the lock can be acquired. If the lock cannot be taken in the specified time the specified
+ command will not be executed and the invocation will fail.</para></listitem>
+ </varlistentry>
+
+ <varlistentry>
+ <term><option>--print</option></term>
+ <term><option>-p</option></term>
+
+ <listitem><para>Instead of locking the specified device(s) and executing a command, just print the
+ device path(s) that would be locked, and execute no command. This command is useful to determine
+ the "whole" block device in case a partition block device is specified. The devices will be sorted
+ by their device node major number as primary ordering key and the minor number as secondary
+ ordering key (i.e. they are shown in the order they'd be locked). Note that the number of lines
+ printed here can be less than the the number of <option>--device=</option> and
+ <option>--backing=</option> switches specified in case these resolve to the same "whole"
+ devices.</para></listitem>
+ </varlistentry>
+
+ <xi:include href="standard-options.xml" xpointer="help" />
+ </variablelist>
+ </refsect2>
</refsect1>
<refsect1>
@@ -761,6 +845,38 @@
</refsect1>
<refsect1>
+ <title>Example</title>
+
+ <example>
+ <title>Format a File System</title>
+
+ <para>Take a lock on the backing block device while creating a file system, to ensure that
+ <command>systemd-udevd</command> doesn't probe or announce the new superblock before it is
+ comprehensively written:</para>
+
+ <programlisting># udevadm lock --device=/dev/sda1 mkfs.ext4 /dev/sda1</programlisting>
+ </example>
+
+ <example>
+ <title>Format a RAID File System</title>
+
+ <para>Similar, but take locks on multiple devices at once:</para>
+
+ <programlisting># udevadm lock --device=/dev/sda1 --device=/dev/sdb1 mkfs.btrfs /dev/sda1 /dev/sdb1</programlisting>
+ </example>
+
+ <example>
+ <title>Copy in a File System</title>
+
+ <para>Take a lock on the backing block device while copying in a prepared file system image, to ensure
+ that <command>systemd-udevd</command> doesn't probe or announce the new superblock before it is fully
+ written:</para>
+
+ <programlisting># udevadm lock -d /dev/sda1 dd if=fs.raw of=/dev/sda1</programlisting>
+ </example>
+ </refsect1>
+
+ <refsect1>
<title>See Also</title>
<para><citerefentry>
<refentrytitle>udev</refentrytitle><manvolnum>7</manvolnum>