netroot: modular support for network root devices

modules.d/95nfs/nfsroot new file +78 -0

new file mode 100755
index 0000000..b8858bc
@@ -0,0 +1,125 @@
+#!/bin/sh
+
+. /lib/dracut-lib
+
+PATH=$PATH:/sbin:/usr/sbin
+
+# XXX needs error handling like ifup/dhclient-script
+
+# XXX need to lock our attempts if we're doing the mount here
+
+getarg rdnetdebug && {
+    exec > /tmp/nfsroot.$1.$$.out
+    exec 2>> /tmp/nfsroot.$1.$$.out
+    set -x
+}
+
+[ "$NFS_LOCKED" ] || {
+	NFS_LOCKED=true
+	export NFS_LOCKED
+	exec flock -xo /tmp/nfs.lock -c "$0 $*"
+	exit 1
+}
+
+[ -e /tmp/nfsdone ] && exit 0
+
+nfs_done() {
+    >/tmp/nfsdone
+    exit 0
+}
+
+root=$(getarg root)
+case $root in
+    nfs|/dev/nfs)	type=nfs  ;;
+    nfs4|/dev/nfs4)	type=nfs4 ;;
+    auto|'')		type=auto  ;;
+esac
+
+rootfstype=$(getarg rootfstype)
+case $rootfstype in
+    nfs|nfs4|auto)	type=$rootfstype  ;;
+esac
+
+# If we're not doing NFS at all, don't keep banging our head
+[ -n "$type" ] || nfs_done
+
+[ -e /tmp/net.$1.dhcpopts ] && . /tmp/net.$1.dhcpopts
+
+nfsroot=$(getarg nfsroot)
+[ -n "$nfsroot" ] || nfsroot="$new_root_path"
+[ -n "$nfsroot" ] || nfs_done
+
+# check for IP address at front, if there is none, use
+# new_dhcp_server_identifier
+#
+# XXX kernel nfsroot uses , to separate NFS options at end
+#
+nfsserver=${nfsroot%%:*}; nfsroot=${nfsroot#*:}
+nfspath=${nfsroot%%:*}
+flags=${nfsroot#*:}
+[ "$nfsserver" = "$nfspath" ] && nfsserver=$new_dhcp_server_identifier
+[ "$nfspath" = "$flags" ] && unset flags
+
+[ -n "$nfsserver" ] || no_nfs
+
+# look through the flags and see if any are overridden by the command line
+while [ -n "$flags" ]; do
+    f=${flags%%,*}; flags=${flags#*,}
+    [ "$f" = "nfs" -o "$f" = "nfs4" ] && {
+	[ "$type" = "auto" ] && type=$f
+	continue
+    }
+    [ "$f" = "ro" -o "$f" = "rw" ] && {
+	nfsrw=$f
+	continue
+    }
+    [ "$f" = "lock" -o "$f" = "nolock" ] && {
+	nfslock=$f
+	continue

modules.d/95nfs/nfsroot new file +24 -0

+
+# XXX really, want to retry in a loop I think, but not here...
+
+[ "$type" = "nfs4" -o "$type" = "auto" ] && {
+    # XXX really needed? Do we need non-root users before we start it in
+    # XXX the real root image?
+    [ -n "$(pidof rpc.idmapd)" ] || rpc.idmapd
+
+    # NFSv4 does locks internally
+    mount -t nfs4 -o${nfsflags}${nfslock+,$nfslock} \
+			$nfsserver:$nfspath /sysroot && nfs_done
+
+    # If we're specified to be NFSv4, then stop when we fail
+    # Don't mark us done, as this may be transient
+    [ "$type" = "nfs4" ] && exit 0
+}
+
+# we're NFSv{2,3} or auto and NFSv4 failed. We don't support using locks
+# on NFSv{2,3} because that requires a helper to transfer the rpcbind state
+# rpcbind to the new root
+[ -z "$nfslock" -o "$nfslock" = "lock" ] &&
+    echo "Locks unsupported on NFSv{2,3}, using nolock" 1>&2
+mount -t nfs -onolock,$nfsflags $nfsserver:$nfspath /sysroot && nfs_done
+exit 0

modules.d/95nfs/nfsroot-cleanup.sh new file +8 -0

new file mode 100644
index 0000000..b8f3a2f
@@ -0,0 +1,19 @@
+#!/bin/sh
+
+pid=$(pidof rpc.statd)
+[ -n "$pid" ] && kill $pid
+
+pid=$(pidof rpcbind)
+[ -n "$pid" ] && kill $pid
+

modules.d/95nfs/nfsroot line 88

Quoted 2 lines · David Dillow

modules.d/95nfs/nfsroot new file +2 -0

+# load our modules explicitly, so we can fail fast in the future
+modprobe nfs || nfs_done

Hmmm... isn't that a bit dangerous? I mean what if nfs support is compiled into the kernel?

Quoted 1 line

+[ -n "$(pidof rpcbind)" ] || rpcbind

I see a potential conflict here: Debian based distros use portmap not rpcbind.

Quoted 1 line

+[ -n "$(pidof rpc.statd)" ] || rpc.statd

Question: Do we really need statd?

modules.d/95nfs/nfsroot line 101

Quoted 3 lines

modules.d/95nfs/nfsroot new file +3 -0

+# XXX should I do rpc.idmapd here, or wait and start in the new root
+# XXX waiting assumes root can read everything it needs right up until
+# XXX we start it...

Dracut should only care about getting the root-fs. idmapd is really a system configuration, don't start that.

modules.d/95nfs/nfsroot-cleanup.sh line 18

Quoted 12 lines

modules.d/95nfs/nfsroot-cleanup.sh +10 -0

 
+if grep -q rpc_pipefs /proc/mounts; then 
+    # try to create the destination directory
+    [ -d $NEWROOT/var/lib/nfs/rpc_pipefs ] || mkdir -p $NEWROOT/var/lib/nfs/rpc_pipefs 2>/dev/null
+
+    if [ -d $NEWROOT/var/lib/nfs/rpc_pipefs ]; then
+	mount --move /var/lib/nfs/rpc_pipefs $NEWROOT/var/lib/nfs/rpc_pipefs
+    else
+	umount /var/lib/nfs/rpc_pipefs
+    fi
+fi

Question: Do we really have to do that? I mean nfsroot worked until now without considering rpc_pipefs.

Quoted 1 line

+#!/bin/bash # for highlighting

Please don't do this.

Which editor are you using that highlights bash but not sh correctly?

Quoted 1 line · Victor Lowther

Which editor are you using that highlights bash but not sh correctly?

vim on F10 has different rules for sh and bash -- things like $() and ${t%%:*} are lit in red for sh, but they are valid; they don't highlight that way if vim sees #!/bin/bash

Quoted 6 lines · David Dillow

Quoted 1 line · Victor Lowther

Which editor are you using that highlights bash but not sh correctly?

vim on F10 has different rules for sh and bash -- things like $() and ${t%%:*} are lit in red for sh, but they are valid; they don't highlight that way if vim sees #!/bin/bash

Ah, that explains it. I am a fan of the OS with an editor tacked onto the side.

Quoted 4 lines · Victor Lowther

Quoted 3 lines · David Dillow

Quoted 1 line · Victor Lowther

Which editor are you using that highlights bash but not sh correctly?

Quoted 6 lines

Quoted 3 lines

vim on F10 has different rules for sh and bash -- things like $() and ${t%%:*} are lit in red for sh, but they are valid; they don't highlight that way if vim sees #!/bin/bash

Ah, that explains it. I am a fan of the OS with an editor tacked onto the side.

So, ditch the whole shebang at the top? No #!/bin/sh or #!/bin/bash?

Quoted 9 lines · David Dillow

Quoted 8 lines · Victor Lowther

Quoted 6 lines · David Dillow

Quoted 1 line · Victor Lowther

Which editor are you using that highlights bash but not sh correctly?

vim on F10 has different rules for sh and bash -- things like $() and ${t%%:*} are lit in red for sh, but they are valid; they don't highlight that way if vim sees #!/bin/bash

Quoted 4 lines

Quoted 2 lines

Ah, that explains it. I am a fan of the OS with an editor tacked onto the side.

So, ditch the whole shebang at the top? No #!/bin/sh or #!/bin/bash?

All scripts on the initramfs should be POSIX -- #!/bin/sh is another way of declaring that for people who haven't R'ed TFM. #!/bin/bash is misleading.

vim on F10 has different rules for sh and bash -- things like $() and ${t%%:*} are lit in red for sh, but they are valid; they don't highlight that way if vim sees #!/bin/bash

Quoted 3 lines · David Dillow

vim on F10 has different rules for sh and bash -- things like $() and ${t%%:*} are lit in red for sh, but they are valid; they don't highlight that way if vim sees #!/bin/bash

Add "let g:is_posix = 1" to your .vimrc (see :help ft-sh-syntax for other possible settings).

I noticed when doing the block root= parser that resume didn't support labeled resume devices. Fix that up and get rid of a dependency on /bin/echo in the process.

Signed-off-by David Dillow <dave-i1Mk8JYDVaaSihdK6806/[email protected]>

---
 modules.d/95resume/install            |    1 -
 modules.d/95resume/parse-resume.sh    |    8 ++++++++
 modules.d/95resume/resume-genrules.sh |   14 ++++++++------
 3 files changed, 16 insertions(+), 7 deletions(-)

modules.d/95resume/install +0 -0

index 945611e..cecd2fc 100755

modules.d/95resume/install +0 -0

 inst_hook cmdline 10 "$moddir/parse-resume.sh"
 inst_hook pre-udev 30 "$moddir/resume-genrules.sh"
 inst_hook mount 90 "$moddir/resume.sh"

modules.d/95resume/parse-resume.sh +8 -0

index 447e151..94d7112 100755
@@ -5,3 +5,11 @@ if resume=$(getarg resume=) && ! getarg noresume; then
 else
     unset resume
 fi
+
+case "$resume" in
+    LABEL=*)
+	resume="$(echo $resume | sed 's,/,\\x2f,g')"
+	resume="/dev/disk/by-label/${resume#LABEL=}" ;;
+    UUID=*)
+	resume="/dev/disk/by-uuid/${resume#UUID=}" ;;
+esac

modules.d/95resume/resume-genrules.sh +0 -0

index 108c3f7..7c95715 100755

modules.d/95resume/resume-genrules.sh +8 -0

+if [ -n "$resume" ]; then
+    (
+    printf 'KERNEL=="%s", RUN+="/bin/echo %%M:%%m > /sys/power/resume"\n' \
+		${resume#/dev/}
+    printf 'SYMLINK=="%s", RUN+="/bin/echo %%M:%%m > /sys/power/resume"\n' \
+		${resume#/dev/}
+    ) >> /etc/udev/rules.d/99-resume.rules
+fi

modules.d/95resume/install line 2

Quoted 5 lines · David Dillow

modules.d/95resume/install +0 -1

@@ -1,5 +1,4 @@
 #!/bin/bash
-dracut_install echo

erm, not exactly.

modules.d/95resume/resume-genrules.sh line 6

Quoted 9 lines

modules.d/95resume/resume-genrules.sh +0 -6

@@ -1,6 +1,8 @@
-resume=$(getarg resume=) && ! getarg noresume  && {
-(
-	/bin/echo -e 'KERNEL=="'${resume#/dev/}'", RUN+="/bin/echo %M:%m > /sys/power/resume"'
-	/bin/echo -e 'SYMLINK=="'${resume#/dev/}'", RUN+="/bin/echo %M:%m > /sys/power/resume"'
-) >> /etc/udev/rules.d/99-resume.rules
-}

The real dependency is in the RUN+= part.

/sbin/netroot is a jumping off point to allow various network root devices to share infrastructure. It will loop over scriptlets in the netroot handler, looking for a handler to run for this type of netroot. Handlers can do choose to act based on command line options to the kernel, or via DHCP options received on this interface. They should massage root= into a form suitable for their handler.

Signed-off-by David Dillow <dave-i1Mk8JYDVaaSihdK6806/[email protected]>

---
 dracut                               |    2 +-
 modules.d/40network/60-net.rules     |    1 +
 modules.d/40network/dhcp-fallback.sh |   14 +++++++++
 modules.d/40network/ifup             |    8 ++--
 modules.d/40network/install          |    2 +
 modules.d/40network/netroot          |   52 ++++++++++++++++++++++++++++++++++
 modules.d/99base/init                |   11 ++++++-
 7 files changed, 84 insertions(+), 6 deletions(-)
 create mode 100755 modules.d/40network/dhcp-fallback.sh
 create mode 100755 modules.d/40network/netroot

dracut +1 -1

index 84c194e..e5e7ecd 100755
@@ -96,7 +96,7 @@ if [[ -f $outfile && ! $force ]]; then
     exit 1
 fi
 
-hookdirs="cmdline pre-udev pre-mount pre-pivot mount emergency"
+hookdirs="cmdline pre-udev netroot pre-mount pre-pivot mount emergency"
 
 readonly initdir=$(mktemp -d -t initramfs.XXXXXX)
 trap 'rm -rf "$initdir"' 0 # clean up after ourselves no matter how we die.

modules.d/40network/60-net.rules +1 -0

index 4b030c3..6c79508 100644
@@ -1 +1,2 @@
 ACTION=="add", SUBSYSTEM=="net", RUN+="/sbin/ifup $env{INTERFACE}"
+ACTION=="online", SUBSYSTEM=="net", RUN+="/sbin/netroot $env{INTERFACE}"

modules.d/40network/dhcp-fallback.sh new file +0 -0

new file mode 100755
index 0000000..1f4ec0c

modules.d/40network/dhcp-fallback.sh new file +9 -0

+
+if [ "$root" = "dhcp" -a -z "$netroot" ]; then
+    rootok=1
+    netroot=dhcp
+fi
+
+if [ "${netroot+set}" = "set" ]; then
+    eval "echo netroot='$netroot'" > /tmp/netroot.info
+fi

modules.d/40network/ifup +3 -1

index ee4bc4a..981a207 100755
@@ -12,8 +12,10 @@ getarg rdnetdebug && {
 
 netif=$1
 
-# bail immediatly if the interface is already up
+# bail immediately if the interface is already up
+# or we don't need the network
 [ -f "/tmp/net.$netif.up" ] && exit 0
+[ ! -f /tmp/netroot.info ] && exit 0
 
 # loopback is always handled the same way
 [ "$netif" = "lo" ] && {

modules.d/40network/ifup +1 -0

+if [ -z "$ip" ]; then
     do_dhcp;
 else
     # spin through the kernel command line, looking for ip= lines

modules.d/40network/install +2 -0

index d9ec7f0..3112494 100755
@@ -2,8 +2,10 @@
 dracut_install ip dhclient hostname
 instmods =net
 inst "$moddir/ifup" "/sbin/ifup"
+inst "$moddir/netroot" "/sbin/netroot"
 inst "$moddir/dhclient-script" "/sbin/dhclient-script"
 instmods =networking ecb arc4
 inst_rules "$moddir/60-net.rules"
+inst_hook cmdline 99 "$moddir/dhcp-fallback.sh"
 inst_hook pre-pivot 10 "$moddir/kill-dhclient.sh"
 mkdir -p "${initdir}/var/run"

modules.d/40network/netroot new file +42 -0

new file mode 100755
index 0000000..3d03e72
@@ -0,0 +1,52 @@
+#!/bin/sh
+
+. /lib/dracut-lib
+
+getarg rdnetdebug && {
+    exec >/tmp/netroot.$1.$$.out
+    exec 2>>/tmp/netroot.$1.$$.out
+    set -x
+}
+
+# Only try to configure from one network interface at a time
+#
+[ "$NETROOT_LOCKED" ] || {
+    NETROOT_LOCKED=true
+    export NETROOT_LOCKED
+    exec flock -xo /tmp/netroot.lock -c "$0 $*"
+    exit 1
+}
+
+netif=$1
+
+# If we've already found a root, or we don't have the info we need,
+# then no point in looking further
+#
+[ -e /tmp/netroot.done ] && exit 0
+[ -s /tmp/netroot.info -a -s /tmp/root.info ] || exit 0
+
+# Pick up our config from the command line; we may already know the
+# handler to run
+#
+. /tmp/root.info
+. /tmp/netroot.info
+[ -e /tmp/net.$netif.dhcpopts ] && . /tmp/net.$netif.dhcpopts
+
+# Now, let the installed network root handlers figure this out
+#
+source_all netroot
+
+# If we didn't get a handler set, then we're done
+#
+if [ -z "$handler" ]; then
+    # XXX informative error message?

modules.d/99base/init +3 -1

index 73e96fd..b4fce6a 100755
@@ -47,12 +47,21 @@ fi
 
 if [ -z "${root%%error:*}" ]; then
     case "${root%%:*}" in
-       '') echo "FATAL: no root= option specified" ;;
+       '') echo "FATAL: no root= option specified, and no network support" ;;
        error) echo "FATAL: ${root#error:}" ;;
     esac
     emergency_shell
 fi
 
+# Network root scripts may need updated root= options,
+# so deposit them where they can see them (udev purges the env)

modules.d/40network/dhcp-fallback.sh line 5

Quoted 3 lines · David Dillow

modules.d/40network/dhcp-fallback.sh new file +3 -0

+if [ -z "$root" ]; then
+    root=dhcp
+fi

Is this really necessary? I mean if no root option is supplied, we don't really know if dhcp is really desired or the default /dev/sda1 fallback should be used.

I'd suggest to only ever do netroot if the user explicitely requests it. This would include disabling all network related udev scripts if we don't want netroot.

Quoted 1 line

-if [ "$root" = "dhcp" -a -z "$ip" ]; then

Hmmm... Weird configuration: What do we do if root=dhcp and the ip options contain's non-dhcp related stuff? (Never thought of that before)

Quoted 4 lines

modules.d/40network/netroot new file +4 -0

+# Run the handler; don't store the root, it may change from device to device
+# XXX other variables to export?
+export NEWROOT
+if $handler $netif $root; then

modules.d/40network/netroot line 52

Quoted 3 lines

modules.d/40network/netroot new file +3 -0

+    >/tmp/netroot.done
+fi
+exit 0

Question 1: Is it really necessary to have each handler implements it's own mount-call? I don't know about iSCSI but isn't it usually just a matter of setting fstype, options and source? I think this could be generalized.

Question 2: How do plan to loop over mount-scripts? As you described if might be desirable to retry mounting if the server is overloaded. I see a few problems there if we loop inside netboot/handler-scripts: We could be trying to mount on the wrong interface.

This adds basic support for root on a network block device to the netroot framework.

Signed-off-by David Dillow <dave-i1Mk8JYDVaaSihdK6806/[email protected]>

---
 modules.d/95nbd/check            |   11 ++++
 modules.d/95nbd/install          |    7 +++
 modules.d/95nbd/nbd-netroot.sh   |   11 ++++
 modules.d/95nbd/nbdroot          |   96 ++++++++++++++++++++++++++++++++++++++
 modules.d/95nbd/parse-nbdroot.sh |   49 +++++++++++++++++++
 5 files changed, 174 insertions(+), 0 deletions(-)
 create mode 100755 modules.d/95nbd/check
 create mode 100755 modules.d/95nbd/install
 create mode 100755 modules.d/95nbd/nbd-netroot.sh
 create mode 100755 modules.d/95nbd/nbdroot
 create mode 100755 modules.d/95nbd/parse-nbdroot.sh

modules.d/95nbd/check new file +11 -0

new file mode 100755
index 0000000..f94d63b
@@ -0,0 +1,11 @@
+#!/bin/sh
+# We depend on network modules being loaded
+[ "$1" = "-d" ] && echo network
+
+# If hostonly was requested, fail the check if we are not actually
+# booting from root.
+[ "$1" = "-h" ] && ! egrep -q '/ /dev/nbd[0-9]*' /proc/mounts && exit 1
+
+# If our prerequisites are not met, fail anyways.
+which nbd-client >/dev/null 2>&1 || exit 1
+exit 0

modules.d/95nbd/install new file +7 -0

new file mode 100755
index 0000000..3a80da1
@@ -0,0 +1,7 @@
+#!/bin/bash
+
+inst nbd-client
+inst_hook cmdline 90 "$moddir/parse-nbdroot.sh"
+inst_hook netroot 90 "$moddir/nbd-netroot.sh"
+inst "$moddir/nbdroot" "/sbin/nbdroot"
+instmods nbd

modules.d/95nbd/nbd-netroot.sh new file +11 -0

new file mode 100755
index 0000000..74b3da8
@@ -0,0 +1,11 @@
+#!/bin/sh # for highlighting
+
+if [ "$root" = "dhcp" ]; then
+    if [ -n "$new_root_path" -a -z "${new_root_path%%nbd:*}" ]; then
+	root="$new_root_path"
+    fi
+fi
+
+if [ -z "${root%nbd:*}" ]; then
+    handler=/sbin/nbdroot
+fi

modules.d/95nbd/nbdroot new file +77 -0

new file mode 100755
index 0000000..51120c2
@@ -0,0 +1,96 @@
+#!/bin/sh
+
+. /lib/dracut-lib
+
+PATH=$PATH:/sbin:/usr/sbin
+
+# XXX needs error handling like ifup/dhclient-script
+
+if getarg rdnetdebug; then
+    exec > /tmp/nbdroot.$1.$$.out
+    exec 2>> /tmp/nbdroot.$1.$$.out
+    set -x
+fi
+
+# root is in the form root=nbd:server:port:fstype:fsopts:nbdopts
+netif="$1"
+root="$2"
+
+root=${root#nbd:}
+nbdserver=${root%%:*}; root=${root#*:}
+nbdport=${root%%:*}; root=${root#*:}
+nbdfstype=${root%%:*}; root=${root#*:}
+nbdflags=${root%%:*}
+nbdopts=${root#*:}
+
+if [ "$nbdopts" = "$nbdflags" ]; then
+    unset nbdopts
+fi
+if [ "$nbdflags" = "$nbdfstype" ]; then
+    unset nbdflags
+fi
+if [ "$nbdfstype" = "$nbdport" ]; then
+    unset nbdfstype
+fi
+if [ -z "$nbdfstype" ]; then
+    nbdfstype=auto
+fi
+
+# look through the NBD options and pull out the ones that need to
+# go before the host etc. Append a ',' so we know we terminate the loop
+nbdopts=${nbdopts},
+while [ -n "$nbdopts" ]; do
+    f=${nbdopts%%,*}
+    nbdopts=${nbdopts#*,}
+    if [ -z "$f" ]; then
+        break
+    fi
+    if [ -z "${f%bs=*}" -o -z "${f%timeout=*}" ]; then
+        preopts="$preopts $f"
+        continue
+    fi
+    opts="$opts $f"
+done
+
+# look through the flags and see if any are overridden by the command line
+nbdflags=${nbdflags},
+while [ -n "$nbdflags" ]; do
+    f=${nbdflags%%,*}
+    nbdflags=${nbdflags#*,}
+    if [ -z "$f" ]; then
+        break
+    fi
+    if [ "$f" = "ro" -o "$f" = "rw" ]; then
+        nbdrw=$f
+        continue
+    fi
+    fsopts=${fsopts+$fsopts,}$f
+done
+
+getarg ro && nbdrw=ro
+getarg rw && nbdrw=rw
+fsopts=${fsopts+$fsopts,}${nbdrw}
+
+modprobe nbd || exit 1
+
+# XXX better way to wait for the device to be made?
+i=0

modules.d/95nbd/nbdroot new file +1 -0

+exit 0

modules.d/95nbd/parse-nbdroot.sh new file +49 -0

new file mode 100755
index 0000000..4bedaab
@@ -0,0 +1,49 @@
+#!/bin/dash
+
+# It'd be nice if this could share rules with 99-block.sh, but since
+# the kernel side adds nbd{1..16} when the module is loaded -- before
+# they are associated with a server -- we cannot use the udev add rule
+# to find it
+#
+# XXX actually we could, if we move to root=XXX and netroot=XXX, then
+# you could do root=LABEL=/ nbdroot=XXX, or netroot=nbd:XXX
+#
+# However, we need to be 90-nbd.sh to catch root=/dev/nbd*
+#
+# Preferred format:
+#	root=nbd:srv:port[:fstype[:rootflags[:nbdopts]]]
+#
+# nbdopts is a comma seperated list of options to give to nbd-client
+#
+#
+# Legacy formats:
+#	nbdroot=srv,port
+#	nbdroot=srv:port[:fstype[:rootflags[:nbdopts]]]
+#	root=dhcp nbdroot=srv:port[:fstype[:rootflags[:nbdopts]]]
+#	root=nbd nbdroot=srv:port[:fstype[:rootflags[:nbdopts]]]
+#
+
+case "$root" in
+    nbd|dhcp|'')
+	if getarg nbdroot= > /dev/null; then
+	    root=nbd:$(getarg nbdroot=)
+	fi
+	;;
+esac
+
+# Convert the Debian style to our syntax, but avoid matches on fs arguments
+case "$root" in
+    nbd:*,*)
+	if check_occurances "$root" ',' 1 && check_occurances "$root" ':' 1;
+	then
+	    root=${root%*,}:${root#*,}
+	fi
+	;;
+esac
+
+if [ "${root%%:*}" = "nbd" ]; then
+    # XXX validate options here?
+    # XXX generate udev rules?
+    rootok=1
+    netroot=nbd
+fi

modules.d/95nbd/nbdroot line 87

Quoted 4 lines · David Dillow

modules.d/95nbd/nbdroot new file +4 -0

+# XXX netroot expects to have the handler mount things, but we should
+# XXX allow LVM, LUKS, etc over nbd
+
+nbd-client $preopts "$nbdserver" "$nbdport" /dev/nbd0 $opts || exit 1

Ah, but we are not really a netroot in the sense that NFS is -- all we are doing is attaching a block device. Once it is attached the usual root-on-block-device rules should kick in as soon as udev sees the block devices.

modules.d/95nbd/nbdroot line 94

Quoted 6 lines

modules.d/95nbd/nbdroot new file +6 -0

+if ! mount -t $nbdfstype -o$fsopts /dev/nbd0 $NEWROOT; then
+    # Mount failed, clean up after ourselves so if we try a different
+    # interface it can succeed
+    nbd-client -d /dev/nbd0
+    exit 1
+fi

why not just poke udev with udevadm trigger if there is any doubt?

Quoted 11 lines · Victor Lowther

modules.d/95nbd/nbdroot line 87

Quoted 4 lines · David Dillow

modules.d/95nbd/nbdroot new file +4 -0

+# XXX netroot expects to have the handler mount things, but we should
+# XXX allow LVM, LUKS, etc over nbd
+
+nbd-client $preopts "$nbdserver" "$nbdport" /dev/nbd0 $opts || exit 1

Ah, but we are not really a netroot in the sense that NFS is -- all we are doing is attaching a block device. Once it is attached the usual root-on-block-device rules should kick in as soon as udev sees the block devices.

Exactly, hence the comment -- and one elsewhere about wanting to move the mount to the network block devices via udev. Unfortunately, NBD does all of its device adds at module load time, so we cannot trigger a mount on those as there's nothing behind the device.

I was thinking about 'echo add /path/to/nbd0/uevent' here so we could force another loop through udev.

Quoted 8 lines

modules.d/95nbd/nbdroot line 94

Quoted 6 lines

modules.d/95nbd/nbdroot new file +6 -0

+if ! mount -t $nbdfstype -o$fsopts /dev/nbd0 $NEWROOT; then
+    # Mount failed, clean up after ourselves so if we try a different
+    # interface it can succeed
+    nbd-client -d /dev/nbd0
+    exit 1
+fi

why not just poke udev with udevadm trigger if there is any doubt?

Not sure what you mean by that in this context -- do you mean the root mount via udev as discussed above, or the cleanup right here?

I'm not sure how to do the cleanup when mounting via udev; I'd need to think.

This builds on the NFS root work to allow other network methods to hook into the parsing and network infrastructure. One network interface is configured at a time, but if we don't find a root there, it may not be unconfigured -- this should probably be added.