[09:33] <CIA-37> debian-installer: cjwatson * r1479 ubuntu/debian/changelog: releasing version 20101020ubuntu40
[19:50] <maxb> Hrm. It looks like grub-probe assumes the incorrect path for any md devices created during the natty installer, causing the grub-install call to fail
[20:32] <CarlFK> if my netboot installer append has BOOTIF=01-00-14-22-94-06-6a  should I also need netcfg/choose_inter
[20:32] <CarlFK> face=eth0 ?
[22:35] <v12> hi, i'm having trouble using the alternate installer from USB.  i have grub installed and launch the installer from USB via loopback.  it fails with "searching for Ubuntu installation", and more specifically "cdrom mount failed".  i do not have a boot=... parameter in the kernel line.  how do i tell the installer where its root is?
[22:36] <v12> or, put another way, what is the appropriate boot=... directory for the alternate installer?
[22:39] <v12> topic points to 404 webpage (~scott).  bad sign...
[22:55] <cjwatson> CarlFK: nope, the default (=auto) should do
[22:55] <cjwatson> hm, that link is definitely stale - it was for performance measurements
[22:56] <CarlFK> cjwatson: on a dell laptop whos wifi shows up as eth0, the default gave me "pick an interface"
[22:56] <cjwatson> CarlFK: oh, right, it won't be marked as seen.  use netcfg/choose_interface=auto then
[22:57] <cjwatson> v12: you should definitely never need to tell the installer where its root is by hand
[22:57] <CarlFK> cjwatson: ah, got it.  thanks.  was thinking there was a bug.  will try auto, that sounds sane.
[22:57] <cjwatson> v12: that's sufficiently so that I'm not even sure how one might tell it ...
[22:58] <cjwatson> I don't believe we pass root= normally
[22:58] <cjwatson> boot= is nonexistent AFAI
[22:58] <cjwatson> K
[22:58] <CarlFK> cjwatson: you mind looking at bug 728710 - this has been annoying me for a few releases.  just now got to reporting it but it seems to be slipping into a crack.
[22:58] <ubot2> Launchpad bug 728710 in debian-installer "d-i apt-setup/local0/repository not working" [Undecided,New] https://launchpad.net/bugs/728710
[22:59] <cjwatson> CarlFK: ask me when I haven't just got back from the pub :)
[22:59] <cjwatson> v12: what version of Ubuntu is this?
[22:59] <CarlFK> roger that :)
[23:00] <cjwatson> v12: if it's earlier than natty, cdrom-detect/try-usb=true might help
[23:00] <v12> cjwatson: it's Natty.  i omitted "auto"  i'll try that
[23:01] <v12> is there somewhere i can check for the "auto" logic?  i'm not sure where to browse the source
[23:01] <cjwatson> v12: the auto bit wasn't directed at you
[23:01] <cjwatson> and won't help you
[23:01] <v12> ahh
[23:02] <cjwatson> CarlFK: the snippets of syslog you've posted there aren't very useful - they just describe the bit where d-i is processing the preseed file into the debconf database, not the bit much later on when it actually tries to use them.  could you attach a full DEBCONF_DEBUG=developer syslog?
[23:02] <v12> does the iso need to be in a root directory?  mine is in /boot/isos/natty.iso
[23:02] <CarlFK> cjwatson: will do
[23:03] <cjwatson> v12: let me check
[23:03] <cjwatson> v12: oh, you're doing it that way?  you actually need something a bit different
[23:03]  * cjwatson hunts down the doc link
[23:05] <cjwatson> https://help.ubuntu.com/11.04/installation-guide/i386/boot-usb-files.html
[23:06] <lifeless> cjwatson: oh hai
[23:06] <cjwatson> you can set up a similar boot entry with grub of course, although that describes syslinux
[23:06] <lifeless> cjwatson: could I grab a couple of tips about grub2 & dmraid ?
[23:06] <cjwatson> I suspect the standard loopback.cfg bits may only work properly with the desktop cd
[23:06] <cjwatson> lifeless: maybe, see above comment about just being back from pub
[23:07] <lifeless> cjwatson: heh; I am doing nose following, I hopefully just need broad advice
[23:07] <lifeless> cjwatson: I've filed a bug 803658 about this. the tl;dr version is:
[23:07] <ubot2> Launchpad bug 803658 in grub2 "grub-install /dev/mapper/isw_$UUID_$NAME0 failing with ICH10R raid 1+0" [Undecided,New] https://launchpad.net/bugs/803658
[23:07] <lifeless> dmraid 1+0  (striped layered on mirrored)
[23:08] <lifeless> grub-probe errors out 'cannot find a grub drive for ...'
[23:08] <cjwatson> what's the device.map look like?
[23:08] <cjwatson> if you have one
[23:08] <lifeless> inside the target?
[23:08] <cjwatson> /boot/grub/device.map in whatever filesystem namespace grub-install is running in, yes
[23:09] <lifeless> no such file
[23:09] <cjwatson> ok, failure to autodetect is a bad sign, but two things
[23:10] <lifeless> cjwatson: I wanted to check my understanding, which is that grub will happily read the bios provided mapped array at boot, and that its just a logic failure figuring this out in the prober
[23:10] <cjwatson> (a) figure out what grub-probe command is being run (sh -x /usr/sbin/grub-install <args>, check somewhere near the end), run it with -vv and attach output to bug
[23:10] <v12> cjwatson: says nothing about how to tell the iso not to worry about the CD. i am already booting the kernel in /install.amd/vmlinuz
[23:11] <v12> whoops i mean /install/
[23:11] <cjwatson> (b) you may be able to work around this by creating /boot/grub/device.map (grub-mkdevicemap can create a skeleton); an entry in device.map constitutes a promise that the BIOS can read from that drive
[23:11] <cjwatson> v12: those directions instruct you to use a different build of the installer which has different logic in this area
[23:11] <v12> unfortunately i am trying to use efi
[23:12] <cjwatson> that shouldn't particularly matter
[23:12] <cjwatson> you'll have to translate the boot instructions into grub-speak, certainly, since syslinux won't work
[23:13] <cjwatson> lifeless: yes, at boot time grub should be able to read from anything that the BIOS INT 13h facilities provide
[23:13] <lifeless> cjwatson: attached
[23:14] <lifeless> cjwatson: perhaps grub2 has the same broken filtering that parted does for raid 1+0
[23:14] <cjwatson> it won't be the same as such, but may produce similar results; I agree it's a logic failure in grub-probe
[23:15] <cjwatson> in general we need to probe the root device in order to figure out which modules to build into the core image
[23:16] <lifeless> the probe seems to entirely ignore the array0 device (dm-2)
[23:17] <cjwatson> I *think* an explicit device.map entry should work around this; it doesn't particularly matter what you call it, I'd just use (hdN) where N is one greater than the last entry in what grub-mkdevicemap produces
[23:18] <v12> cjwatson: how does the alternate installer get used for USB?  has anyone done it before?
[23:19] <cjwatson> v12: follow the directions I provided, using the hd-media image
[23:19] <cjwatson> or else, it should be possible to boot the alternate installer by running isohybrid on the ISO and then writing it directly to the entire USB disk
[23:19] <cjwatson> (whether that will work on EFI I don't recall)
[23:19] <lifeless> cjwatson: it still errors on probe
[23:20] <lifeless> same error
[23:20] <v12> aha i'll look for the hd-media image...
[23:20] <cjwatson> lifeless: exactly what line did you put in device.map?
[23:21] <lifeless> I'll paste to the bug (not ircing from that machine)
[23:22] <lifeless> cjwatson: is there now
[23:22] <cjwatson> we certainly seem to need logic in find_partition_start for striped devices
[23:24] <cjwatson> lifeless: if you could attach the output of 'dmsetup table' as well, I can probably attack this tomorrow; I'm afraid it requires a greater level of alertness than I possess right now
[23:24] <lifeless> attached.
[23:25] <lifeless> I'll poke around a bit
[23:25] <lifeless> thanks!
[23:25] <cjwatson> this is partly arising from the need to map partitions onto their offsets from the start of the disk
[23:26] <lifeless> the partition table isn't accurate?
[23:26] <cjwatson> it is, but we do this in order to be able to map partition devices onto their containing disk devices
[23:26] <cjwatson> which is actually startlingly difficult to do in the general case
[23:26] <lifeless> cjwatson: I can imagine - in the striped case it seems impossible to me
[23:26] <cjwatson> shouldn't be, we can get all the links from device-mapper's table
[23:27] <lifeless> cjwatson: doesn't htat imply grub having to reimplement the stripe logic?
[23:27] <cjwatson> since the "containing disk" that we care about is the striped device, not either mirrored device
[23:27] <cjwatson> I don't think so, at least not much, but I'll have to look tomorrow
[23:27] <lifeless> cjwatson: I'm probably misunderstanding the direction of containing there ;)
[23:27] <lifeless> but its late for you
[23:28] <cjwatson> yeah, I don't think I'm being very clear
[23:28] <cjwatson> anyway, bed :)
[23:28] <lifeless> I'll see if I can wipe my disk out at least twice before lunch
[23:28] <cjwatson> sorry not to be able to solve immediately
[23:28] <lifeless> de nada
[23:28] <lifeless> now if I can just remember how to get debian/patches applied
[23:28] <cjwatson> I don't think I've seen the striped case before.  might need to set something up in a vm
[23:28] <lifeless> to build normally
[23:28] <cjwatson> debian/rules patch
[23:28] <cjwatson> in the case of grub2
[23:28] <lifeless> thanks!
[23:29] <cjwatson> (which 'debian/rules build' does for you, but 'debian/rules build/grub-pc' doesn't - that's a bug)
[23:30] <CarlFK> cjwatson: thanks for =auto.  worked as advertised
[23:37] <cjwatson> CarlFK: oh good