**** BEGIN LOGGING AT Thu Mar 22 02:59:58 2012 Mar 22 11:48:38 is it normal for Ubuntu ARM builds to depend on having busybox in the initramfs? I disabled it and then I got a kernel panic on "init: grep: not found" Mar 22 13:58:28 highvoltage: Yes, it is. All of the premount init scripts need it to run. If you don't want that, don't use initrd. Mar 22 13:59:24 erm, *all* ubuntu initrds depend on busybox Mar 22 13:59:36 its the default shell in the initrd Mar 22 13:59:53 nothing arch specific with that Mar 22 14:16:50 ogra_: I usually disable busybox in my initrds on ubuntu machines and they work just fine Mar 22 14:16:54 (at least they appear to) Mar 22 14:22:38 well, how exactly do you disable it ? Mar 22 14:23:04 /init in the initrd is a shellscript Mar 22 14:23:17 and the only shell shipped in initrd is busybox Mar 22 14:24:04 (also all local-top/bottom/premount scripts are shell) Mar 22 14:24:47 you probably found an arm specific bug Mar 22 14:26:04 (i.e. that it is actually disableable while that option is ignored on other arches or some such) Mar 22 15:16:48 highvoltage, have you switched to ARM lately :) ? Mar 22 15:17:07 are you on a quest for lowpower thin clients? Mar 22 15:17:18 janimo`: I don't really switch to anything. I just assimilate, like the borg :) Mar 22 15:17:54 borg needs arms too Mar 22 15:18:10 indeed. Mar 22 16:36:06 ogra_: Can you review my change to lp:ubuntu/flash-kernel (lp:~gruemaster/flash-kernel/bug-961174)? Thanks. Mar 22 17:17:26 GrueMaster, hmm, that somewhat defeats the purpose of Env.txt etc, shouldnt they override boot.scr ? Mar 22 17:18:34 That's the problem. Apparently, Orchestra (or some provisioning tool) uses those, but our images don't. Mar 22 17:19:25 So if you use that tool to tell a system to do a netboot install, the boot partition never gets overwritten. Mar 22 17:19:47 Maybe it should go in f-k-i? Mar 22 17:20:13 well, its still wrong imho Mar 22 17:20:28 if some tool puts a broken file in the way it should also move it away Mar 22 17:20:55 if a user puts such a file there it should do what its supposed to do Mar 22 17:21:07 (override the system default) Mar 22 17:21:53 f-k-i would surely be a better place but my complaint still stands even there Mar 22 17:22:09 The other issue (as I understand it) is if it exists and bricks the system (i.e. won't boot), you can't recover via usb boot. For remote deployments, this will be a problem. Mar 22 17:23:34 right, still, the questions are a) why does it exist at all (why isnt boot.scr used) and b) why is its content broken Mar 22 17:24:25 if we would rm the file or mv it like you do we prevent the user from using a documented override mechanism (and imho *introduce* a bug) Mar 22 17:24:54 GrueMaster, can you ask rbasak why that file exists at all, why its broken in the bug ? Mar 22 17:25:14 removing it isnt the proper solution Mar 22 17:25:17 ogra_: because it's been created by a previous tenant Mar 22 17:25:30 since that would make the overriding impossible Mar 22 17:26:16 the purpose of this file is to give the user an easy way to modify boot settings ... Mar 22 17:26:36 if your tool uses that on a system level, fix the tool to use the system file please Mar 22 17:26:54 breaking behavior for the use cant be a solution to this bug Mar 22 17:27:19 ogra_: When I install Ubuntu on an Intel machine, I expect the installer to override whatever the boot configuration was previously. If the MBR was set up to boot Windows, after running the installer and confirming that I want the boot configuration changed, I expect the system to boot Ubuntu. Mar 22 17:27:45 ogra_: if you want to override this, you change the boot configuration _after_ running the installer but before booting into the new system Mar 22 17:27:56 if you install on an intel system you have various possibilities to override the bootloader settings Mar 22 17:28:05 ogra_: I don't see why the behaviour on a Pandaboard should be any different Mar 22 17:28:12 so that's not entirely true, you can add info. to grub for boot specifics which I don't think you can with u-boot/boot.scr at the moment Mar 22 17:28:24 If you want to override the boot, then tell the installer that you don't want to run flash-kernel-installer Mar 22 17:28:26 if you use u-boot the only way to i.e. ebnable a serial console in the kernel cmdline is to put an uEnv.txt in place Mar 22 17:28:45 or to fiddle with the provided boot.scr ... Mar 22 17:28:56 righjt Mar 22 17:28:57 right Mar 22 17:29:09 the .txt files were enabled for easier overriding through user action Mar 22 17:29:22 if you have any tool using that machanism, make it better use boot.scr Mar 22 17:29:31 What you have in there at the moment is a hack which is preventing the installer from doing its job properly. What I want is for the installer to work in all cases. If you want to add a hack later, then please don't break my use case! Mar 22 17:29:50 i dont want to add any hacks at all Mar 22 17:30:10 It is necessary for the installer to be able to erase any configuration left by a previous tenant Mar 22 17:30:25 if your use case usues a user definable file to set system settings your use case does it wrong Mar 22 17:30:44 thats like having a package put a used rule into /etc/udev/rules.d Mar 22 17:30:53 instead of using /lib/udev/rules.d Mar 22 17:31:07 My use case requires a machine to be able to be reinstalled regardless of what was there before Mar 22 17:31:07 you wouldnt use /etc in that case either Mar 22 17:31:13 Orchestra needs this too. Mar 22 17:31:39 still you make use of a user override mechanism on a system level Mar 22 17:31:49 I'm not making use of anything Mar 22 17:31:49 use the system mechanism for this Mar 22 17:31:59 The user leaving a uEnv.txt is a DoS Mar 22 17:32:06 I'm not using it Mar 22 17:32:14 well, who puts the .txt file in place ? Mar 22 17:32:34 the one that got in your way i mean Mar 22 17:32:35 Somebody who has root on the machine before I remotely decide to wipe and reinstall it to give it to someone else who has root on the machine. Mar 22 17:32:55 DoSing yourself isn't a DoS. Mar 22 17:34:02 rbasak, if there is a way that we can know we are in such an environment, i'm happy to add a hack for that specific env Mar 22 17:34:07 Maybe it would be nice if d-i detected this odd situation and asked about it, but this isn't "just like ovewriting a windows MBR", and it would be best to not pretend it was. Mar 22 17:34:16 I think the root of the problem is that partman currently doesn't reformat the boot partition during netinstall. Mar 22 17:34:30 breaking documented override behavior for everyone cant be the solution for a corner usecase Mar 22 17:34:31 ogra_: how about a preseed option to say "override any user overrides"? Mar 22 17:34:35 Known bug. Mar 22 17:34:50 rbasak, that sounds saner at least Mar 22 17:34:50 ogra_: if it sees that then it will either delete uEnv.txt or *. Mar 22 17:35:44 infinity: it's not "yourself", it's a previous tenant DoSing a future tenant. Imagine if your EC2 instance didn't come up because a previous user left a file somewhere. Mar 22 17:36:03 i'm not opposed to hacks (if you ask others they might call me the master of bad hacks even) ... but i'm opposed to break user abilities for providing their own defined overrides at install time Mar 22 17:36:30 ogra_: I think you've got the sense inverted. If the user wants to define an override at install time, he should say so using debconf! Mar 22 17:36:49 thats not possible on a u-boot level Mar 22 17:37:14 rbasak: Root is you. Calling yourself a "tenant" doesn't change things. There are a number of ways I can mess up an Intel system for the next person I sell it to as well. Mar 22 17:37:39 rbasak: And the "boot partition" on firmwareless SD system (ie: a Panda) is more akin to the BIOS than to a disk MBR. Mar 22 19:23:24 dannf: f-k highbank changes sponsored. Mar 22 19:24:06 dannf: Do you have a sane way to test this once it's in the archive? Mar 22 19:25:04 GrueMaster, bug 848782 looks suspiciously like a duplicate of bug 921137 Mar 22 19:25:05 Launchpad bug 848782 in flash-kernel "Serial console not enabled for passphrase prompt when using LUKS" [Medium,In progress] https://launchpad.net/bugs/848782 Mar 22 19:25:06 Launchpad bug 921137 in flash-kernel "Flash-kernel-installer doesn't support d-i debian-installer/add-kernel-opts in preseed " [Undecided,Confirmed] https://launchpad.net/bugs/921137 Mar 22 19:25:36 Similar, but slightly different. Mar 22 19:25:58 They can both be fixed in the same way, I think. I had a local branch for that and got sidetracked... Mar 22 19:26:07 well, i dont see any solution for 848782 apart from using add-kernel-opt Mar 22 19:26:08 s Mar 22 19:26:36 * infinity wonders where that went. Mar 22 19:26:45 that you use LUKS doesnt reallys imply that you use a serial console, that needs to be specified anyway Mar 22 19:27:37 (luks could indeed preseed that value if it knows there is only serial) Mar 22 19:27:58 Nah, it should just do the same thing grub-installer does here. Mar 22 19:28:24 And, like I said, I was working on that. Tobin and I discussed it a while ago. I just got sidetracked with other stuff and let it slip. :/ Mar 22 19:28:31 ah, k Mar 22 19:28:44 the discussion above made me look at the f-k buglist :) Mar 22 19:29:49 ogra_: So, I fixed all the *sums generation for all the images... Except for ac100.bootimg. Were you doing something abnormally hackish in how you deal with that? Mar 22 19:30:05 not that i can think of Mar 22 19:30:18 Hrm. Mar 22 19:30:26 I'll have to investigate that further. Mar 22 19:30:37 But at least the big images are all right now, which was the major concern. Mar 22 19:30:46 look at the debian-cd code, i just added the .bootimg type to the file types Mar 22 19:31:14 it isnt generated as .raw like the others though, since it comes directly out of live-build iirc Mar 22 19:31:23 probably that gets in our way Mar 22 19:31:31 feel free to leave that issue to me though Mar 22 19:31:47 (not today anymore, but I can look at it before B2) Mar 22 19:32:09 Yeah, it comes from live-build, same as manifest and other files. Mar 22 19:32:46 It could just be exposing some weird assumption in cdimage, given than this is the only image that produces two files that need summing. Mar 22 19:33:26 yeah Mar 22 19:33:51 I could probably paper over it by making debian-cd sum the bootimg files, but I'd like to figure out why cdimage is doing it wrong. Mar 22 19:35:30 iirc cdimage does it while renaming them from .raw to their actual filetype Mar 22 19:35:41 or do i remember wrong ? Mar 22 19:35:46 No, cdimage does it after the rename. Mar 22 19:35:53 ah Mar 22 19:36:07 But then maps certain extensions back to .raw, so it can match against the *SUMS in scratch. Mar 22 19:36:11 * ogra_ through its the same function in the makefile Mar 22 19:36:18 bootimg doesn't get mapped though (which is correct), so... Mar 22 19:37:13 I wonder how much of this mess I need to replicate to iterate through some local testing. Mar 22 19:37:39 pfft, just do it on nusakan and revert it if it breaks Mar 22 19:38:02 thats why we have it all in bzr :) Mar 22 19:38:29 infinity: cool - yeah, just qemu though Mar 22 19:39:47 ogra_: :P Mar 22 19:40:13 ogra_: Not doing it in production for the obvious "don't work in production" reasons, but mostly, I want to reduce this to a small enough test that I can do it in seconds, not hours. :P Mar 22 19:40:33 ah, yeah Mar 22 19:41:02 still, relpicating cdimage loacally cries for pain Mar 22 19:41:11 Oh, I've done that before. Mar 22 19:41:21 But in this case, it's reducing, not replicating. Mar 22 19:41:25 * ogra_ too, thats why i said that **** ENDING LOGGING AT Fri Mar 23 02:59:58 2012