New service that waits on zvol links to be created

The zfs-volume-wait.service scans existing zvols and waits for their
links under /dev to be created. Any service that depends on zvol
links to be there should add a dependency on zfs-volumes.target.
By default, this target is not enabled.

Reviewed-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
Reviewed-by: Antonio Russo <antonio.e.russo@gmail.com>
Reviewed-by: Richard Laager <rlaager@wiktel.com>
Reviewed-by: loli10K <ezomori.nozomu@gmail.com>
Reviewed-by: John Gallagher <john.gallagher@delphix.com>
Reviewed-by: George Wilson <gwilson@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Pavel Zakharov <pzakharov@delphix.com>
Closes openzfs#8975

cmd/Makefile.am

@@ -5,4 +5,4 @@ if USING_PYTHON
 SUBDIRS += arcstat arc_summary dbufstat
 endif
 
-SUBDIRS += mount_zfs zed zvol_id
+SUBDIRS += mount_zfs zed zvol_id zvol_wait

cmd/zvol_wait/Makefile.am

@@ -0,0 +1 @@
+dist_bin_SCRIPTS = zvol_wait

cmd/zvol_wait/zvol_wait

@@ -0,0 +1,93 @@
+#!/bin/sh
+
+count_zvols() {
+	if [ -z "$zvols" ]; then
+		echo 0
+	else
+		echo "$zvols" | wc -l
+	fi
+}
+
+filter_out_zvols_with_links() {
+	while read -r zvol; do
+		if [ ! -L "/dev/zvol/$zvol" ]; then
+			echo "$zvol"
+		fi
+	done
+}
+
+filter_out_deleted_zvols() {
+	while read -r zvol; do
+		if zfs list "$zvol" >/dev/null 2>&1; then
+			echo "$zvol"
+		fi
+	done
+}
+
+list_zvols() {
+	zfs list -t volume -H -o name,volmode | while read -r zvol_line; do
+		name=$(echo "$zvol_line" | awk '{print $1}')
+		volmode=$(echo "$zvol_line" | awk '{print $2}')
+		# /dev links are not created for zvols with volmode = "none".
+		[ "$volmode" = "none" ] || echo "$name"
+	done
+}
+
+zvols=$(list_zvols)
+zvols_count=$(count_zvols)
+if [ "$zvols_count" -eq 0 ]; then
+	echo "No zvols found, nothing to do."
+	exit 0
+fi
+
+echo "Testing $zvols_count zvol links"
+
+outer_loop=0
+while [ "$outer_loop" -lt 20 ]; do
+	outer_loop=$((outer_loop + 1))
+
+	old_zvols_count=$(count_zvols)
+
+	inner_loop=0
+	while [ "$inner_loop" -lt 30 ]; do
+		inner_loop=$((inner_loop + 1))
+
+		zvols="$(echo "$zvols" | filter_out_zvols_with_links)"
+
+		zvols_count=$(count_zvols)
+		if [ "$zvols_count" -eq 0 ]; then
+			echo "All zvol links are now present."
+			exit 0
+		fi
+		sleep 1
+	done
+
+	echo "Still waiting on $zvols_count zvol links ..."
+	#
+	# Although zvols should normally not be deleted at boot time,
+	# if that is the case then their links will be missing and
+	# we would stall.
+	#
+	if [ "$old_zvols_count" -eq "$zvols_count" ]; then
+		echo "No progress since last loop."
+		echo "Checking if any zvols were deleted."
+
+		zvols=$(echo "$zvols" | filter_out_deleted_zvols)
+		zvols_count=$(count_zvols)
+
+		if [ "$old_zvols_count" -ne "$zvols_count" ]; then
+			echo "$((old_zvols_count - zvols_count)) zvol(s) deleted."
+		fi
+
+		if [ "$zvols_count" -ne 0 ]; then
+			echo "Remaining zvols:"
+			echo "$zvols"
+		else
+			echo "All zvol links are now present."
+			exit 0
+		fi
+	fi
+done
+
+echo "Timed out waiting on zvol links"
+exit 1

configure.ac

@@ -123,6 +123,7 @@ AC_CONFIG_FILES([
 	cmd/zed/zed.d/Makefile
 	cmd/raidz_test/Makefile
 	cmd/zgenhostid/Makefile
+	cmd/zvol_wait/Makefile
 	contrib/Makefile
 	contrib/bash_completion.d/Makefile
 	contrib/dracut/Makefile

etc/systemd/system/50-zfs.preset.in

@@ -5,4 +5,5 @@ enable zfs-import.target
 enable zfs-mount.service
 enable zfs-share.service
 enable zfs-zed.service
+enable zfs-volume-wait.service
 enable zfs.target

etc/systemd/system/Makefile.am

@@ -7,7 +7,9 @@ systemdunit_DATA = \
 	zfs-import-scan.service \
 	zfs-mount.service \
 	zfs-share.service \
+	zfs-volume-wait.service \
 	zfs-import.target \
+	zfs-volumes.target \
 	zfs.target
 
 EXTRA_DIST = \
@@ -17,6 +19,8 @@ EXTRA_DIST = \
 	$(top_srcdir)/etc/systemd/system/zfs-mount.service.in \
 	$(top_srcdir)/etc/systemd/system/zfs-share.service.in \
 	$(top_srcdir)/etc/systemd/system/zfs-import.target.in \
+	$(top_srcdir)/etc/systemd/system/zfs-volume-wait.service.in \
+	$(top_srcdir)/etc/systemd/system/zfs-volumes.target.in \
 	$(top_srcdir)/etc/systemd/system/zfs.target.in \
 	$(top_srcdir)/etc/systemd/system/50-zfs.preset.in
 

etc/systemd/system/zfs-volume-wait.service.in

@@ -0,0 +1,13 @@
+[Unit]
+Description=Wait for ZFS Volume (zvol) links in /dev
+DefaultDependencies=no
+After=systemd-udev-settle.service
+After=zfs-import.target
+
+[Service]
+Type=oneshot
+RemainAfterExit=yes
+ExecStart=@bindir@/zvol_wait
+
+[Install]
+WantedBy=zfs-volumes.target

etc/systemd/system/zfs-volumes.target.in

@@ -0,0 +1,7 @@
+[Unit]
+Description=ZFS volumes are ready
+After=zfs-volume-wait.service
+Requires=zfs-volume-wait.service
+
+[Install]
+WantedBy=zfs.target

man/man1/Makefile.am

@@ -1,4 +1,4 @@
-dist_man_MANS = zhack.1 ztest.1 raidz_test.1
+dist_man_MANS = zhack.1 ztest.1 raidz_test.1 zvol_wait.1
 EXTRA_DIST = cstyle.1
 
 install-data-local:

man/man1/zvol_wait.1

@@ -0,0 +1,21 @@
+.Dd July 5, 2019
+.Dt ZVOL_WAIT 1 SMM
+.Os Linux
+.Sh NAME
+.Nm zvol_wait
+.Nd Wait for ZFS volume links in
+.Em /dev
+to be created.
+.Sh SYNOPSIS
+.Nm
+.Sh DESCRIPTION
+When a ZFS pool is imported, ZFS will register each ZFS volume
+(zvol) as a disk device with the system. As the disks are registered,
+.Xr \fBudev 7\fR
+will asynchronously create symlinks under
+.Em /dev/zvol
+using the zvol's name.
+.Nm
+will wait for all those symlinks to be created before returning.
+.Sh SEE ALSO
+.Xr \fBudev 7\fR

rpm/generic/zfs.spec.in

@@ -322,7 +322,7 @@ image which is ZFS aware.
 
 %if 0%{?_systemd}
     %define systemd --enable-systemd --with-systemdunitdir=%{_unitdir} --with-systemdpresetdir=%{_presetdir} --with-systemdmodulesloaddir=%{_modulesloaddir} --with-systemdgeneratordir=%{_systemdgeneratordir} --disable-sysvinit
-    %define systemd_svcs zfs-import-cache.service zfs-import-scan.service zfs-mount.service zfs-share.service zfs-zed.service zfs.target zfs-import.target
+    %define systemd_svcs zfs-import-cache.service zfs-import-scan.service zfs-mount.service zfs-share.service zfs-zed.service zfs.target zfs-import.target zfs-volume-wait.service zfs-volumes.target
 %else
     %define systemd --enable-sysvinit --disable-systemd
 %endif
@@ -419,6 +419,7 @@ systemctl --system daemon-reload >/dev/null || true
 %{_sbindir}/*
 %{_bindir}/raidz_test
 %{_bindir}/zgenhostid
+%{_bindir}/zvol_wait
 # Optional Python 2/3 scripts
 %{_bindir}/arc_summary
 %{_bindir}/arcstat