Re: [Cooker] LVM2 in cooker (sort of)rc.sysinit lvm2 fix ?
Svetoslav Slavtchev wrote: Quoting Luca Berra [EMAIL PROTECTED]: smth like debian suse? i think they have smth like /etc/rc.boot where let say installing raidtools would drop a init script for activating the soft raids installing lvm-tools the VG's LV's this could be a good idea to try post 9.2, methinks. after i used my LVM1 version LV's with LVM2, i can no longer access them with MDK-9.1/ LVM1 tools update kernel -25mdk i have 4 xfs LV's and 1 reiserfs ... both times device 3a:00 under cooker with LVM2 everythings works OK i saw this bug once, but i tried rebooting with LVM1 today and did not have any problem. could it be that lvm2 does (or did in a previous version) some weird things with lvm metadata? it is pratically possible [EMAIL PROTECTED] ~ # ./vgdisplay /proc/lvm/VGs/vg00 exists: Is the original LVM driver using this volume group? Can't lock vg00: skipping and can you load dm-mod after that and activate different VG (without deactivating vg00)? i think this shouldn't be possible. dunno and actually don't want to try. just say we won't support mixing lvm1 and lvm2. well, i know you belive it's absolutely save to have / on lvm (lvm cooker ml posts) but i'm still not convinced, and the mdk team probably too, so initrd lilo are not important as the tools devmapper (just my thoughts) ok, but i still would like to be able to do it manually if i so wish. as i am able to do it with lvm1 (changes to mkinitrd are small, if we want to keep mkinitrd as it is) may be we should then drop the HB08... and fix x86_64 to use the devmapper anyone out there running cooker on a x86_64 ? well' i checked the sources for 2.6 kernel and they use extern __inline__ int generic_fls(int x) so we could drop fls from device-mapper patch instead. regards, L.
Re: [Cooker] LVM2 in cooker (sort of)rc.sysinit lvm2 fix ?
Quoting Luca Berra [EMAIL PROTECTED]: Svetoslav Slavtchev wrote: Quoting Luca Berra [EMAIL PROTECTED]: smth like debian suse? i think they have smth like /etc/rc.boot where let say installing raidtools would drop a init script for activating the soft raids installing lvm-tools the VG's LV's this could be a good idea to try post 9.2, methinks. metoo :) after i used my LVM1 version LV's with LVM2, i can no longer access them with MDK-9.1/ LVM1 tools update kernel -25mdk i have 4 xfs LV's and 1 reiserfs ... both times device 3a:00 under cooker with LVM2 everythings works OK i saw this bug once, but i tried rebooting with LVM1 today and did not have any problem. could it be that lvm2 does (or did in a previous version) some weird things with lvm metadata? probably yes, but that was with devmapper-1.00.02 LVM2-2.00.05 and the standard LVM1 tools in MDK-9.1 it is pratically possible [EMAIL PROTECTED] ~ # ./vgdisplay /proc/lvm/VGs/vg00 exists: Is the original LVM driver using this volume group? Can't lock vg00: skipping and can you load dm-mod after that and activate different VG (without deactivating vg00)? i think this shouldn't be possible. dunno and actually don't want to try. just say we won't support mixing lvm1 and lvm2. well, i know you belive it's absolutely save to have / on lvm (lvm cooker ml posts) but i'm still not convinced, and the mdk team probably too, so initrd lilo are not important as the tools devmapper (just my thoughts) ok, but i still would like to be able to do it manually if i so wish. as i am able to do it with lvm1 (changes to mkinitrd are small, if we want to keep mkinitrd as it is) OK, let's hope some one will merge them :) may be we should then drop the HB08... and fix x86_64 to use the devmapper anyone out there running cooker on a x86_64 ? well' i checked the sources for 2.6 kernel and they use extern __inline__ int generic_fls(int x) so we could drop fls from device-mapper patch instead. good news :-) so we have to patch dm-io.[ch] (and update to the new version) svetljo
Re: [Cooker] LVM2 in cooker (sort of)rc.sysinit lvm2 fix ?
On Tue, Aug 19, 2003 at 11:27:49PM +0200, Svetoslav Slavtchev wrote: Hi, the attached diff fixes for me running lvm1 format VG's/LV's with the lvm2 tools and devmapper both for 2.4 and 2.6 kernels. probably the fix is ok, but i would rather add a lvmrc script, like hp-ux does, rc.sysinit is too much bloated for this. besides even with lvm1 doing vgscan every time is redundant. is there any chance for LVM2 and devmapper in MDK-9.2, or is it too late? i'd like to see them also. i have experimental patches for 2.4 (see the ZY patches) http://varna.demon.co.uk/~svetlio/ruby-contrib/mdk-cook/probably_broken/ 10x but the devmapper and lvm2 rpms are tottaly brocken (from time to time i just use the web space until i solve my fs corruption problems) i have a decent devmapper rpm at http://percy.comedia.it/~bluca/cooker/lvm2 including a lilo with device mapper support. The lvm rpm there suck a little. I had to create an lvm1 and lvm2 rpm which conflict with the lvm in the distro and use alternatives to allow installation in parallel. but i would like much better if i could devise a scheme to do runtime lvm version detection. Maybe updating the alternatives in rc.sysinit could be ok. But i think i can also write a wrapper that tries to in order the device mapper proc file and the lvm1 proc file and then decides which one to run. The worst thing is that a luser could have some vg managed by dm and some by lvm1 at the same time, but i don't know if we want to support that. We also need a working initrd which supports lvm2 and lvm1 also. also could some one give me some hints about the last patch ZY18_dev-mapper_preffix_dm_fls.patch devmapper defines internally generic_fls which conflicts with one of the x86_64 patches (HB08_define_fls) , so i decided to prefix the devmapper ones with generic_dm_fls. is this the right solution? anyone having a better one? I believe the two peices of code are the same so the code in devmapper can be dropped safely. well, devmapper code uses unsigned for the parameter that i believe would be more correct than what we have. Atm i just removed the fls function from dm-io.c and renamed the only occurrence where it was called to use generic_fls32() which is the devmapper one. regards, L. -- Luca Berra -- [EMAIL PROTECTED] Communication Media Services S.r.l. /\ \ / ASCII RIBBON CAMPAIGN XAGAINST HTML MAIL / \
Re: [Cooker] LVM2 in cooker (sort of)rc.sysinit lvm2 fix ?
Quoting Luca Berra [EMAIL PROTECTED]: On Tue, Aug 19, 2003 at 11:27:49PM +0200, Svetoslav Slavtchev wrote: Hi, the attached diff fixes for me running lvm1 format VG's/LV's with the lvm2 tools and devmapper both for 2.4 and 2.6 kernels. probably the fix is ok, but i would rather add a lvmrc script, like hp-ux does, rc.sysinit is too much bloated for this. besides even with lvm1 doing vgscan every time is redundant. smth like debian suse? there was a thread horrible rc.sysinit a while ago, but it won't change for 9.2 (and someone mentioned it would break the compatibility to redhat, i don't know we need it thought) is there any chance for LVM2 and devmapper in MDK-9.2, or is it too late? i'd like to see them also. i have experimental patches for 2.4 (see the ZY patches) http://varna.demon.co.uk/~svetlio/ruby-contrib/mdk-cook/probably_broken/ 10x but the devmapper and lvm2 rpms are tottaly brocken (from time to time i just use the web space until i solve my fs corruption problems) i have a decent devmapper rpm at http://percy.comedia.it/~bluca/cooker/lvm2 including a lilo with device mapper support. The lvm rpm there suck a little. I had to create an lvm1 and lvm2 rpm which conflict with the lvm in the distro and use alternatives to allow installation in parallel. but i would like much better if i could devise a scheme to do runtime lvm version detection. Maybe updating the alternatives in rc.sysinit could be ok. why do you think we have to install lvm1 lvm2 in parallel? to support kernels without devmapper? in case the distribution kernel has devmapper we don't need lvm1(just my thoughts) But i think i can also write a wrapper that tries to in order the device mapper proc file and the lvm1 proc file and then decides which one to run. me - luser: where is the mapper proc file? i just got it running yesterday, did't have time to look arround, but couldn't find /proc/lvm :-) The worst thing is that a luser could have some vg managed by dm and some by lvm1 at the same time, but i don't know if we want to support that. How? hell no, even if its teoretically possible, its probably pretty dangerous We also need a working initrd which supports lvm2 and lvm1 also. do we need it :-) mandrake do not support / on lvm is there lvm1 support currently ? also could some one give me some hints about the last patch ZY18_dev-mapper_preffix_dm_fls.patch devmapper defines internally generic_fls which conflicts with one of the x86_64 patches (HB08_define_fls) , so i decided to prefix the devmapper ones with generic_dm_fls. is this the right solution? anyone having a better one? I believe the two peices of code are the same so the code in devmapper can be dropped safely. well, devmapper code uses unsigned for the parameter that i believe would be more correct than what we have. Atm i just removed the fls function from dm-io.c and renamed the only occurrence where it was called to use generic_fls32() which is the devmapper one. may be we should then drop the HB08... and fix x86_64 to use the devmapper one, but i don't have x86_64 to fix/verify it best, svetljo
Re: [Cooker] LVM2 in cooker (sort of)rc.sysinit lvm2 fix ?
there was a thread horrible rc.sysinit a while ago, but it won't change for 9.2 (and someone mentioned it would break the compatibility to redhat, i don't know we need it thought) IMHO current rc.sysinit is so patched that nothing will be broken by any change.
Re: [Cooker] LVM2 in cooker (sort of)rc.sysinit lvm2 fix ?
On Wed, Aug 20, 2003 at 09:56:17AM +0200, Svetoslav Slavtchev wrote: Quoting Luca Berra [EMAIL PROTECTED]: On Tue, Aug 19, 2003 at 11:27:49PM +0200, Svetoslav Slavtchev wrote: probably the fix is ok, but i would rather add a lvmrc script, like hp-ux does, rc.sysinit is too much bloated for this. besides even with lvm1 doing vgscan every time is redundant. smth like debian suse? have to check what debian and suse are doing nowadays, but for lvm at least i would like to have some control on which volumes should be activated at boot and which should not. and vgscan should not be done at boot at all! there was a thread horrible rc.sysinit a while ago, but it won't change for 9.2 (and someone mentioned it would break the compatibility to redhat, i don't know we need it thought) rc.sysinit has to change a 32k script is just too difficule, and a small syntax error can hose your system. redhat compatibility? you mean keeping rh bugs, no thanks. why do you think we have to install lvm1 lvm2 in parallel? to support kernels without devmapper? well i'd like to be able to switch back and forth, at least for a while. in case the distribution kernel has devmapper we don't need lvm1(just my thoughts) this would be easier to implement. but using alternatives does not hurt either. me - luser: where is the mapper proc file? /dev/mapper/control, sorry. The worst thing is that a luser could have some vg managed by dm and some by lvm1 at the same time, but i don't know if we want to support that. How? hell no, even if its teoretically possible, its probably pretty dangerous it is pratically possible [EMAIL PROTECTED] ~ # ./vgdisplay /proc/lvm/VGs/vg00 exists: Is the original LVM driver using this volume group? Can't lock vg00: skipping We also need a working initrd which supports lvm2 and lvm1 also. do we need it :-) mandrake do not support / on lvm is there lvm1 support currently ? what do you mean with support? mandrake installer could not create a lvm root last time i checked, but lilo and mkinitrd are ready to boot from lvm since 9.0 IIRC. may be we should then drop the HB08... and fix x86_64 to use the devmapper one, but i don't have x86_64 to fix/verify it me neither. -- Luca Berra -- [EMAIL PROTECTED] Communication Media Services S.r.l. /\ \ / ASCII RIBBON CAMPAIGN XAGAINST HTML MAIL / \
Re: [Cooker] LVM2 in cooker (sort of)rc.sysinit lvm2 fix ?
Quoting Luca Berra [EMAIL PROTECTED]: On Wed, Aug 20, 2003 at 09:56:17AM +0200, Svetoslav Slavtchev wrote: Quoting Luca Berra [EMAIL PROTECTED]: On Tue, Aug 19, 2003 at 11:27:49PM +0200, Svetoslav Slavtchev wrote: probably the fix is ok, but i would rather add a lvmrc script, like hp-ux does, rc.sysinit is too much bloated for this. besides even with lvm1 doing vgscan every time is redundant. smth like debian suse? have to check what debian and suse are doing nowadays, but for lvm at least i would like to have some control on which volumes should be activated at boot and which should not. and vgscan should not be done at boot at all! i think they have smth like /etc/rc.boot where let say installing raidtools would drop a init script for activating the soft raids installing lvm-tools the VG's LV's there was a thread horrible rc.sysinit a while ago, but it won't change for 9.2 (and someone mentioned it would break the compatibility to redhat, i don't know we need it thought) rc.sysinit has to change a 32k script is just too difficule, and a small syntax error can hose your system. redhat compatibility? you mean keeping rh bugs, no thanks. why do you think we have to install lvm1 lvm2 in parallel? to support kernels without devmapper? well i'd like to be able to switch back and forth, at least for a while. after i used my LVM1 version LV's with LVM2, i can no longer access them with MDK-9.1/ LVM1 tools update kernel -25mdk i have 4 xfs LV's and 1 reiserfs under MDK-9.1 with LVM1 all xfs volumes are somehow remapped to the reiserfs one ( whenever i try to mount one of them, i get the content of reiserfs LV, mounted as reiserfs) something which should probably partitialy explain it mount 1st xfs volume; umount it mount 2nd xfs volume; umount it dmesg: reiserfs: checking transaction log (device 3a:00) ... Using r5 hash to sort names ReiserFS version 3.6.25 reiserfs: checking transaction log (device 3a:00) ... Using r5 hash to sort names ReiserFS version 3.6.25 both times device 3a:00 under cooker with LVM2 everythings works OK The worst thing is that a luser could have some vg managed by dm and some by lvm1 at the same time, but i don't know if we want to support that. How? hell no, even if its teoretically possible, its probably pretty dangerous it is pratically possible [EMAIL PROTECTED] ~ # ./vgdisplay /proc/lvm/VGs/vg00 exists: Is the original LVM driver using this volume group? Can't lock vg00: skipping and can you load dm-mod after that and activate different VG (without deactivating vg00)? i think this shouldn't be possible. We also need a working initrd which supports lvm2 and lvm1 also. do we need it :-) mandrake do not support / on lvm is there lvm1 support currently ? what do you mean with support? mandrake installer could not create a lvm root last time i checked, but lilo and mkinitrd are ready to boot from lvm since 9.0 IIRC. well, i know you belive it's absolutely save to have / on lvm (lvm cooker ml posts) but i'm still not convinced, and the mdk team probably too, so initrd lilo are not important as the tools devmapper (just my thoughts) may be we should then drop the HB08... and fix x86_64 to use the devmapper one, but i don't have x86_64 to fix/verify it me neither. anyone out there running cooker on a x86_64 ? svetljo
