Bug#678696: Event based block device handling (fixes USB and nested devices problem)
On Wed, Mar 11, 2015 at 06:09:50PM +, Ben Hutchings wrote: On Tue, 2015-03-10 at 12:47 +0100, Goswin von Brederlow wrote: On Wed, Mar 04, 2015 at 09:30:24PM +, Ben Hutchings wrote: Thanks for your work on this bug. I ended up with a somewhat different implementation as I don't think it's necessary to duplicate the information that udev provides, and as we may now need to mount more than one filesystem. But I should have credited you in the changelog for prototyping this, and I forgot to do so. Ben. The idea with the udev rule was to wait up to ROOTDELAY seconds without a new device apearing before giving up. Now you wait ROOTDELAY seconds in total, which then can depend on the number of devices. It's now max(rootdelay, 30) because the rootdelay kernel parameter wasn't meant for this purpose at all. Add new disk and you have to increase the ROOTDELAY. I hope not! On one system the PSU isn't big enough to spin up all disks at once. So the SCSI controler is set to not start them on power on. Instead they come up sequentially. So one disk takes 5s, 2 disks 10s, 3 disks 15s, ... accordingly you have to set the delay. Add another disk and the total time goes up. Also it was ment so that block scripts could specifically target the new devices instead of having to scan all devices every time. For example say you have a crypt device and you forgot the password. Now I think you will be asked 30 times for the password before the initramfs gives up. True, but so far as I could see you didn't send scripts that made use of that. We could still implement that later, I think. And now that we potentially have to mount /usr (and possibly other filesystems), not just root and swap, lvm2's script needed to be told which device we're looking for, not which devices appeared. Ben. That isn't realy new. Debian already had root and swap. Adding a third doesn't realy change anything. LVM should already have known what devices to activate for root and swap. The problem I see there is detecting what devices to bring up. The /usr filesystem might be in a ZPOOL that uses a crypt device on a LVM logical volume on a raid. Or any other complex nesting. Could be any number of devices that are needed for /usr to become available. Again nothing new for /usr, / and swap already have that problem. Note: LVM has persistent metadata that tell it wether to bring up a device or not. So I'm not sure it makes much sense to guess what devices are needed and limiting lvm to only start those. The metadata already has this functionality and is more reliable than guessing what devices may be needed. MfG Goswin -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#678696: Event based block device handling (fixes USB and nested devices problem)
On Tue, 2015-03-10 at 12:47 +0100, Goswin von Brederlow wrote: On Wed, Mar 04, 2015 at 09:30:24PM +, Ben Hutchings wrote: Thanks for your work on this bug. I ended up with a somewhat different implementation as I don't think it's necessary to duplicate the information that udev provides, and as we may now need to mount more than one filesystem. But I should have credited you in the changelog for prototyping this, and I forgot to do so. Ben. The idea with the udev rule was to wait up to ROOTDELAY seconds without a new device apearing before giving up. Now you wait ROOTDELAY seconds in total, which then can depend on the number of devices. It's now max(rootdelay, 30) because the rootdelay kernel parameter wasn't meant for this purpose at all. Add new disk and you have to increase the ROOTDELAY. I hope not! Also it was ment so that block scripts could specifically target the new devices instead of having to scan all devices every time. For example say you have a crypt device and you forgot the password. Now I think you will be asked 30 times for the password before the initramfs gives up. True, but so far as I could see you didn't send scripts that made use of that. We could still implement that later, I think. And now that we potentially have to mount /usr (and possibly other filesystems), not just root and swap, lvm2's script needed to be told which device we're looking for, not which devices appeared. Ben. -- Ben Hutchings Any smoothly functioning technology is indistinguishable from a rigged demo. signature.asc Description: This is a digitally signed message part
Bug#678696: Event based block device handling (fixes USB and nested devices problem)
On Wed, Mar 04, 2015 at 09:30:24PM +, Ben Hutchings wrote: Thanks for your work on this bug. I ended up with a somewhat different implementation as I don't think it's necessary to duplicate the information that udev provides, and as we may now need to mount more than one filesystem. But I should have credited you in the changelog for prototyping this, and I forgot to do so. Ben. The idea with the udev rule was to wait up to ROOTDELAY seconds without a new device apearing before giving up. Now you wait ROOTDELAY seconds in total, which then can depend on the number of devices. Add a new disk and you have to increase the ROOTDELAY. Also it was ment so that block scripts could specifically target the new devices instead of having to scan all devices every time. For example say you have a crypt device and you forgot the password. Now I think you will be asked 30 times for the password before the initramfs gives up. MfG Goswin -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#678696: Event based block device handling (fixes USB and nested devices problem)
Thanks for your work on this bug. I ended up with a somewhat different implementation as I don't think it's necessary to duplicate the information that udev provides, and as we may now need to mount more than one filesystem. But I should have credited you in the changelog for prototyping this, and I forgot to do so. Ben. -- Ben Hutchings Horngren's Observation: Among economists, the real world is often a special case. signature.asc Description: This is a digitally signed message part
Bug#678696: Event based block device handling (fixes USB and nested devices problem)
On Sun, Aug 31, 2014 at 10:14:18AM +0200, Michael Prokop wrote: Hi, f'up to our recent discussion we had on IRC * Goswin von Brederlow [Sat Jun 23, 2012 at 09:25:28PM +0200]: the attached patch adds an event based loop for block devices to the init script. New blockdevices are recorded in /run/initramfs/block-events by an udev rule as they appear. The init script repeadately waits for that and then calls /scripts/local-block/* with a list of new devices storedin NEWDEVS until $ROOT and $resume (if set) exists or a timeout is reached. This fixes the problem that USB devices take too long to be discovered and crypto, raid, lvm or multipath can't be started on them. It also adds support for arbitrary nestings of them, e.g. raid5 over raid1. [...] First of all thanks again for the patch and your helpful feedback on IRC. I've tested your patch based on top of current i-t git master (v0.116) with a setup like: BOOT_IMAGE=/boot/vmlinuz-3.14-2-amd64 root=/dev/mapper/vg0-root ro quiet but it sadly fails to work as intended (it boots but doesn't find the block devices until the timeout is kicking in). I didn't investigate closer yet, but AFAICS it seems to be related to the fact that /dev/mapper/vg0-root doesn't exist at that time yet. If it boots after the timeout then the device must exist. So either the test for it is wrong or the device only gets created after the timeout. If you or someones else finds time to try and possibly further investigate I'd very much welcome and appreciate that. regards, -mika- Questions: Does your system boot without the patch? If not then you also need the lvm patch to provide the hook script that scans for a VG when new block devices are detected. Do you see any messages about vg0 being activated before the timeout or after? MfG Goswin -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#678696: Event based block device handling (fixes USB and nested devices problem)
Hi, f'up to our recent discussion we had on IRC * Goswin von Brederlow [Sat Jun 23, 2012 at 09:25:28PM +0200]: the attached patch adds an event based loop for block devices to the init script. New blockdevices are recorded in /run/initramfs/block-events by an udev rule as they appear. The init script repeadately waits for that and then calls /scripts/local-block/* with a list of new devices storedin NEWDEVS until $ROOT and $resume (if set) exists or a timeout is reached. This fixes the problem that USB devices take too long to be discovered and crypto, raid, lvm or multipath can't be started on them. It also adds support for arbitrary nestings of them, e.g. raid5 over raid1. [...] First of all thanks again for the patch and your helpful feedback on IRC. I've tested your patch based on top of current i-t git master (v0.116) with a setup like: BOOT_IMAGE=/boot/vmlinuz-3.14-2-amd64 root=/dev/mapper/vg0-root ro quiet but it sadly fails to work as intended (it boots but doesn't find the block devices until the timeout is kicking in). I didn't investigate closer yet, but AFAICS it seems to be related to the fact that /dev/mapper/vg0-root doesn't exist at that time yet. If you or someones else finds time to try and possibly further investigate I'd very much welcome and appreciate that. regards, -mika- signature.asc Description: Digital signature
Bug#678696: Event based block device handling (fixes USB and nested devices problem)
Package: initramfs-tools
Version: 0.106
Severity: wishlist
Tags: patch
Hi,
the attached patch adds an event based loop for block devices to the
init script. New blockdevices are recorded in
/run/initramfs/block-events by an udev rule as they appear. The init
script repeadately waits for that and then calls
/scripts/local-block/* with a list of new devices storedin NEWDEVS
until $ROOT and $resume (if set) exists or a timeout is reached.
This fixes the problem that USB devices take too long to be
discovered and crypto, raid, lvm or multipath can't be started on
them. It also adds support for arbitrary nestings of them, e.g. raid5
over raid1.
For the event loop I needed the $resume device translated to the
device node earlier. This was previously done in
scripts/local-premount/resume. It was essentially the same code used
to translate the $ROOT so I created a new function translate_device()
in script/functions and used that for both $ROOT and $resume.
Note: hooks/block-event and scripts/block-event are executable, not
represented in the patch.
MfG
Goswin
-- System Information:
Debian Release: wheezy/sid
APT prefers unstable
APT policy: (500, 'unstable')
Architecture: amd64 (x86_64)
Kernel: Linux 3.2.0-2-amd64 (SMP w/4 CPU cores)
Locale: LANG=C, LC_CTYPE=de_DE (charmap=ISO-8859-1)
Shell: /bin/sh linked to /bin/dash
Versions of packages initramfs-tools depends on:
ii cpio 2.11-7
ii klibc-utils2.0~rc3-1
ii module-init-tools 3.16-1
ii udev 175-3.1
Versions of packages initramfs-tools recommends:
ii busybox 1:1.19.3-5
Versions of packages initramfs-tools suggests:
ii bash-completion 1:1.99-3
-- no debconf information
Description: event based loop
This patch adds an event based loop for block devices to the init
script. New blockdevices are recorded in /run/initramfs/block-events
by an udev rule as they appear. The init script repeadately waits for
that and then calls /scripts/local-block/* with a list of new devices
storedin NEWDEVS until $ROOT and $resume (if set) exists or a timeout
is reached.
.
This fixes the problem that USB devices take too long to be
discovered and crypto, raid, lvm or multipath can't be started on
them. It also adds support for arbitrary nestings of them, e.g. raid5
over raid1.
.
For the event loop I needed the $resume device translated to the
device node earlier. This was previously done in
scripts/local-premount/resume. It was essentially the same code used
to translate the $ROOT so I created a new function translate_device()
in script/functions and used that for both $ROOT and $resume.
.
Note: hooks/block-event and scripts/block-event are executable, not
represented in the patch.
Author: Goswin von Brederlow
Last-Update: 2012-06-23
--
diff -Nru initramfs-tools-0.106/hooks/block-event initramfs-tools-0.106a0mrvn1/hooks/block-event
--- initramfs-tools-0.106/hooks/block-event 1970-01-01 01:00:00.0 +0100
+++ initramfs-tools-0.106a0mrvn1/hooks/block-event 2012-06-23 18:52:54.0 +0200
@@ -0,0 +1,13 @@
+#!/bin/sh
+
+mkdir -p $DESTDIR/etc/udev/rules.d
+
+cat $DESTDIR/etc/udev/rules.d/99-block-event.rules EOF
+# we are only interested in add and change actions for block devices
+ACTION==remove, GOTO=block_event_end
+SUBSYSTEM!=block, GOTO=block_event_end
+
+RUN+=/scripts/block-event $name
+
+LABEL=block_event_end
+EOF
diff -Nru initramfs-tools-0.106/init initramfs-tools-0.106a0mrvn1/init
--- initramfs-tools-0.106/init 2012-06-06 15:04:52.0 +0200
+++ initramfs-tools-0.106a0mrvn1/init 2012-06-23 21:14:40.0 +0200
@@ -69,41 +69,10 @@
init=${x#init=}
;;
root=*)
- ROOT=${x#root=}
- case $ROOT in
- LABEL=*)
- ROOT=${ROOT#LABEL=}
-
- # support any / in LABEL= path (escape to \x2f)
- case ${ROOT} in
- */*)
- if command -v sed /dev/null 21; then
-ROOT=$(echo ${ROOT} | sed 's,/,\\x2f,g')
- else
-if [ ${ROOT} != ${ROOT#/} ]; then
- ROOT=\x2f${ROOT#/}
-fi
-if [ ${ROOT} != ${ROOT%/} ]; then
- ROOT=${ROOT%/}\x2f
-fi
-IFS='/'
-newroot=
-for s in $ROOT; do
- newroot=${newroot:+${newroot}\\x2f}${s}
-done
-unset IFS
-ROOT=${newroot}
- fi
- esac
- ROOT=/dev/disk/by-label/${ROOT}
- ;;
- UUID=*)
- ROOT=/dev/disk/by-uuid/${ROOT#UUID=}
- ;;
- /dev/nfs)
+ ROOT=$(translate_device ${x#root=})
+ if [ $ROOT = /dev/nfs ]; then
[ -z ${BOOT} ] BOOT=nfs
- ;;
- esac
+ fi
;;
rootflags=*)
ROOTFLAGS=-o ${x#rootflags=}
@@ -190,7 +159,7 @@
export noresume
unset resume
else
- resume=${RESUME:-}
+ resume=$(translate_device ${RESUME:-})
fi
maybe_break top
@@ -205,6 +174,39 @@
[ -n ${netconsole} ] modprobe netconsole netconsole=${netconsole}
+maybe_break events
+TIMEOUT=20
+COUNT=$TIMEOUT
+echo Waiting for $ROOT
+if [ -n $resume ]; then
+ echo Waiting for $resume
+fi
+# run at least once
+touch /dev/.initramfs/block-events
+while [ $COUNT -gt 0 ];
Bug#678696: Event based block device handling (fixes USB and nested devices problem)
Hi,
I filed bugs for cryptsetup, lvm, mdadm and multipath-tools to support
this solution (i.e. provide scripts/local-block/ sniplets) [#678688
#678691 #678692 #678693].
If anyone wants to test this prior to those packages being fixed the
scriplets below can be used.
MfG
Goswin
== scripts/local-block/cryptoroot ==
#!/bin/sh
PREREQ=
prereqs()
{
echo $PREREQ
}
case $1 in
# get pre-requisites
prereqs)
prereqs
exit 0
;;
esac
if [ -x /scripts/local-top/cryptoroot ]; then
exec /scripts/local-top/cryptoroot
fi
== scripts/local-block/lvm2 ==
#!/bin/sh
PREREQ=
prereqs()
{
echo $PREREQ
}
case $1 in
# get pre-requisites
prereqs)
prereqs
exit 0
;;
esac
if [ -x /scripts/local-top/lvm2 ]; then
exec /scripts/local-top/lvm2
fi
== scripts/local-block/mdadm ==
#!/bin/sh
PREREQ=
prereqs()
{
echo $PREREQ
}
case $1 in
# get pre-requisites
prereqs)
prereqs
exit 0
;;
esac
if [ -x /scripts/local-top/mdadm ]; then
exec /scripts/local-top/mdadm
fi
--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
