bcache-tools: changes to make the udev rules work

bcache-tools: changes to make the udev rules work

[Rolf Fokkens]

Hi!

When making bcache tools run on Fedora I had to make some small tweaks 
on the udev rules file:

- first trying with blkid doesn't work, because it doesn't know about bcache
- full pathnames are needed, also for bcache-register because udev 
doesn't include /sbin and /usr/sbin in PATH.

This is the diff:

--- bcache-tools-20130820/61-bcache.rules.rules    2013-08-20 
22:03:46.000000000 +0200
+++ bcache-tools-20130820/61-bcache.rules    2013-08-25 
15:53:32.690312711 +0200
@@ -5,14 +5,14 @@
  ACTION=="remove", GOTO="bcache_end"

  # Backing devices: scan, symlink, register
-IMPORT{program}="/sbin/blkid -o udev $tempnode"
+#IMPORT{program}="/usr/sbin/blkid -o udev $tempnode"
  # blkid and probe-bcache can disagree, in which case don't register
-ENV{ID_FS_TYPE}=="?*", ENV{ID_FS_TYPE}!="bcache", GOTO="bcache_backing_end"
+#ENV{ID_FS_TYPE}=="?*", ENV{ID_FS_TYPE}!="bcache", 
GOTO="bcache_backing_end"

-IMPORT{program}="/sbin/probe-bcache -o udev $tempnode"
+IMPORT{program}="/usr/sbin/probe-bcache -o udev $tempnode"
  ENV{ID_FS_UUID_ENC}=="?*", SYMLINK+="disk/by-uuid/$env{ID_FS_UUID_ENC}"
  SUBSYSTEM=="block", ACTION=="add|change", ENV{ID_FS_TYPE}=="bcache", \
-        RUN+="bcache-register $tempnode"
+        RUN+="/usr/lib/udev/bcache-register $tempnode"
  LABEL="bcache_backing_end"

  # Cached devices: symlink

Any feedback is welcome,

Rolf

Re: bcache-tools: changes to make the udev rules work

[Gabriel de Perthuis]

On Mon, 26 Aug 2013 11:23:32 +0200, Rolf Fokkens wrote:
> Hi!

Hello!

> When making bcache tools run on Fedora I had to make some small tweaks 
> on the udev rules file:
> 
> - first trying with blkid doesn't work, because it doesn't know about bcache
> - full pathnames are needed, also for bcache-register because udev 
> doesn't include /sbin and /usr/sbin in PATH.

The blkid check is to make sure that there isn't a non-bcache superblock.
blkid doesn't need to understand bcache for this.

Make path changes if you really need, but udev is supposed to find unqualified
paths like bcache-register in $udevlibdir ( /lib/udev on debian, possibly
/usr/lib/udev in fedora).

> This is the diff:
> 
> --- bcache-tools-20130820/61-bcache.rules.rules    2013-08-20 
> 22:03:46.000000000 +0200
> +++ bcache-tools-20130820/61-bcache.rules    2013-08-25 
> 15:53:32.690312711 +0200
> @@ -5,14 +5,14 @@
>   ACTION=="remove", GOTO="bcache_end"
> 
>   # Backing devices: scan, symlink, register
> -IMPORT{program}="/sbin/blkid -o udev $tempnode"
> +#IMPORT{program}="/usr/sbin/blkid -o udev $tempnode"
>   # blkid and probe-bcache can disagree, in which case don't register
> -ENV{ID_FS_TYPE}=="?*", ENV{ID_FS_TYPE}!="bcache", GOTO="bcache_backing_end"
> +#ENV{ID_FS_TYPE}=="?*", ENV{ID_FS_TYPE}!="bcache", 
> GOTO="bcache_backing_end"

Those first two changes should be reverted.

> -IMPORT{program}="/sbin/probe-bcache -o udev $tempnode"
> +IMPORT{program}="/usr/sbin/probe-bcache -o udev $tempnode"

This change isn't broken, though you could make use of any /sbin
compat symlinks and not carry an upstream delta.

>   ENV{ID_FS_UUID_ENC}=="?*", SYMLINK+="disk/by-uuid/$env{ID_FS_UUID_ENC}"
>   SUBSYSTEM=="block", ACTION=="add|change", ENV{ID_FS_TYPE}=="bcache", \
> -        RUN+="bcache-register $tempnode"
> +        RUN+="/usr/lib/udev/bcache-register $tempnode"
>   LABEL="bcache_backing_end"

Unless there's been a change in udev, this change is unnecessary and
introduces a distro-specific path.

Re: bcache-tools: changes to make the udev rules work

[Rolf Fokkens]

On 08/26/2013 05:27 PM, Gabriel de Perthuis wrote:
> Make path changes if you really need, but udev is supposed to find unqualified
> paths like bcache-register in $udevlibdir ( /lib/udev on debian, possibly
> /usr/lib/udev in fedora).
You're right; works fine without the full path.
>
>> This is the diff:
>>
>> --- bcache-tools-20130820/61-bcache.rules.rules    2013-08-20
>> 22:03:46.000000000 +0200
>> +++ bcache-tools-20130820/61-bcache.rules    2013-08-25
>> 15:53:32.690312711 +0200
>> @@ -5,14 +5,14 @@
>>    ACTION=="remove", GOTO="bcache_end"
>>
>>    # Backing devices: scan, symlink, register
>> -IMPORT{program}="/sbin/blkid -o udev $tempnode"
>> +#IMPORT{program}="/usr/sbin/blkid -o udev $tempnode"
>>    # blkid and probe-bcache can disagree, in which case don't register
>> -ENV{ID_FS_TYPE}=="?*", ENV{ID_FS_TYPE}!="bcache", GOTO="bcache_backing_end"
>> +#ENV{ID_FS_TYPE}=="?*", ENV{ID_FS_TYPE}!="bcache",
>> GOTO="bcache_backing_end"
> Those first two changes should be reverted.
I tried, but with the blkid check first my /dev/bcache* devices are not 
reliably created. I have 2 in total (/dev/bcache0 and /dev/bcach1) and 
with the blkid there's only one after any reboot. Although the name 
varies (bcache0 or bcache1) it actually is always the /dev/bcache* 
device for the 'second' bcache device that is created. second one means: 
the first one is the combi of /dev/sda3 & /dev/sdb1, the second one is 
the combi of /dev/sdc1 and /dev/sdd1. I'm not sure how I can figure out 
what's actually happening during boot.
>
>> -IMPORT{program}="/sbin/probe-bcache -o udev $tempnode"
>> +IMPORT{program}="/usr/sbin/probe-bcache -o udev $tempnode"
> This change isn't broken, though you could make use of any /sbin
> compat symlinks and not carry an upstream delta.
Moved back to using the symlinks, though later on ...

Re: bcache-tools: changes to make the udev rules work

[Gabriel de Perthuis]

> I tried, but with the blkid check first my /dev/bcache* devices are
> not reliably created. I have 2 in total (/dev/bcache0 and
> /dev/bcach1) and with the blkid there's only one after any reboot.
> Although the name varies (bcache0 or bcache1) it actually is always
> the /dev/bcache* device for the 'second' bcache device that is
> created. second one means: the first one is the combi of /dev/sda3 &
> /dev/sdb1, the second one is the combi of /dev/sdc1 and /dev/sdd1.
> I'm not sure how I can figure out what's actually happening during
> boot.

Possibly /dev/sdc1 or /dev/sdd1 has a stray non-bcache superblock.

You can check with wipefs, and remove it with wipefs -a.

Re: bcache-tools: changes to make the udev rules work

[Gabriel de Perthuis]

>> I tried, but with the blkid check first my /dev/bcache* devices are
>> not reliably created. I have 2 in total (/dev/bcache0 and
>> /dev/bcach1) and with the blkid there's only one after any reboot.
>> Although the name varies (bcache0 or bcache1) it actually is always
>> the /dev/bcache* device for the 'second' bcache device that is
>> created. second one means: the first one is the combi of /dev/sda3 &
>> /dev/sdb1, the second one is the combi of /dev/sdc1 and /dev/sdd1.
>> I'm not sure how I can figure out what's actually happening during
>> boot.
>
> Possibly /dev/sdc1 or /dev/sdd1 has a stray non-bcache superblock.

(I mean the other two, your "first" pair)

> You can check with wipefs, and remove it with wipefs -a.

Re: bcache-tools: changes to make the udev rules work

[Rolf Fokkens]

On 08/26/2013 11:27 PM, Gabriel de Perthuis wrote:
> Possibly /dev/sdc1 or /dev/sdd1 has a stray non-bcache superblock.
You're right, see below. That was an ugly one.

Do you think having blkid identify bcache itself would have helped, or 
is there always a chance of contradictive signatures which may confuse  
blkid?
>
> You can check with wipefs, and remove it with wipefs -a.
The evidence:

[root@localhost ~]# wipefs /dev/sda3
[root@localhost ~]# wipefs /dev/sdb1
offset               type
----------------------------------------------------------------
0x218                LVM2_member   [raid]
                      UUID:  ZSSjZv-FU0H-Zeaa-92GU-37Zs-KyLS-JMF03Y

[root@localhost ~]# wipefs /dev/sdc1
[root@localhost ~]# wipefs /dev/sdd1
[root@localhost ~]# blkid /dev/sdb1
/dev/sdb1: UUID="ZSSjZv-FU0H-Zeaa-92GU-37Zs-KyLS-JMF03Y" TYPE="LVM2_member"
[root@localhost ~]#

Re: bcache-tools: changes to make the udev rules work

[Gabriel de Perthuis]

> You're right, see below. That was an ugly one.
> 
> Do you think having blkid identify bcache itself would have helped, 
> or is there always a chance of contradictive signatures which may 
> confuse  blkid?

make-bcache should use libblkid to detect existing superblocks,
and bail and recommend wipefs.

Support in the other direction (having libblkid support bcache) is just a
nice-to-have, and will maybe allow simplifying a few things in bcache-progs.
The udev rule is written not to break if libblkid gets bcache support, but
it can't really be simplified, we still want to detect stray superblocks
even if the latest tools can't create them.

Re: bcache-tools: changes to make the udev rules work

[matthew patton]

>The rationale is here:
>
>
>http://fedoraproject.org/wiki/Features/UsrMove


<quote>But, advanced features in today's systems can not really bootup with an empty /usr anymore. More and more fails in subtle ways in such setups.</quote>


What a crock of (#$@. "We broke the rules and now we're going to perpetuate the mistake because we can't be bothered to go fix our past screwups." Damn kids!! No one says it has to boot to network-aware multi user (init=3 or higher) without /usr but it damn well better post to single-user and be able to mount all "local" (ie non-network dependent) block devices without /usr. I swear, the work product out of RedHat gets worse every year.

>Might be very Fedora specific, so I can make a non-Fedora diff. And even on Fedora /sbin is a symlink to /usr/sbin so I might not have to assume the tools are  in /usr



Definitely do NOT assume they put stuff in /usr/sbin. Always use /sbin and if RedHat forgot something, make them fix their symlink abomination (what is this, Fedora aspires to be fscking Slowaris with /etc/init -> /sbin/init and holy hell breaks if you don't have the /etc/init link?).