=== danjared_ is now known as danjared [15:49] cyphermox, horum =( [15:49] i dasd formated the drives, formated the drives / partitions on them. [15:50] and now i'm trying to exit the lvm menu and that fails. [15:50] d-i is simply stuck in "Starting up the partitioner" [15:50] Scanning disks... [15:50] 45% [15:50] i wonder how i can troubleshoot this, or get it going again. [15:59] sswitch to another vt? [16:00] are there multiple vts on s390x? [16:00] no vts... i'm installing via ssh [16:00] there is syslog. [16:00] well syslog might give you a clue maybe [16:00] i'm blaming 30parted init.d script (part of partman init) [16:00] ERROR: unsupported sector size 4096 on /dev/dm-0. [16:00] horum. [16:01] why do you still have a dm-0? [16:03] I mean, if you want to run dasdfmt outside of parted generally, as a separate step that says something like "Reinitialize the drives", then that step probably should kick lvm hard enough that it forgets everything [16:03] and then probably udev-reload or whatever the command was [16:04] update-dev --settle >/dev/null I think [16:06] right, and maybe not even because in theory when partman starts, partman-lvm is what will pvscan and vgscan, and then vgchange -a [16:07] which means dd if=/dev/zero some beginning part of the drive might be enough to convince it that there isn't anything [16:09] xnox: have you started a debian vm after all, I'm curious about the udev rules that might be there [16:10] digging into this [16:10] it seems like open_dialog DUMP is called by 35dump script [16:10] and it tries to "dump" the dasd drive... and gets stuck [16:10] ah [16:10] is that command send to ... parted? [16:10] yes [16:11] how do i do that just from shell without e.g. all of the d-i and partman in the way? [16:11] parted -> not found. [16:11] horum [16:11] anna-install parted-udeb [16:11] partman uses libparted to do magic. [16:12] ah, yeah, parted must barf at dumping the partition table somehow [16:12] there ought to be some other message in syslog before that [16:13] or it's just the dm device that makes things ugly [16:15] hm, DUMP is not a parted command, now is it.... [16:16] parted can print all the devices just fine. [16:17] aha [16:17] xnox: partman-command may help you [16:18] I got bored of debugging that stuff being hard a while back, and wrote a helper [16:18] hehe [16:18] http://paste.ubuntu.com/14867242/ [16:18] xnox: parted_server.c in partman-base is what implements those commands [16:18] cjwatson, cyphermox ^ that can't be good, right? [16:18] eh, maybe? depends if that's actually what it's hitting [16:18] check /var/log/partman to see if you see the same thing there [16:19] # partman-command DUMP [16:19] /bin/partman-command: line 42: can't open /var/lib/partman/outfifo: no such file [16:20] parted_server has to be actually running at the time [16:20] ah [16:20] ok [16:20] if it's crashed then check /var/log/partman first [16:22] cyphermox, well, i purged lvms, now trying to install with lvms =) [16:26] today is not my day [16:26] http://paste.ubuntu.com/14867293/ [16:26] that's probably the same thing [16:26] what does /var/log/partman say? [16:28] not much. [16:28] http://paste.ubuntu.com/14867315/ [16:28] as if life is good. [16:29] OK, so that probably is the same thing [16:29] horum. [16:29] * xnox ponders how to reproduce this bug somewhere where i have gdb [16:30] hmmm.... http://comments.gmane.org/gmane.comp.gnu.parted.devel/4367 [16:31] you could brutally shove gdb and the relevant bits and pieces of libraries and debug symbols into the installer env [16:31] wouldn't be the first time I've done that :) [16:32] xnox: pretty sure I already cherry-picked descendants of those patches [16:33] anyway, there's enough code that getting gdb in place is probably easier than guessing [16:35] yeah [16:36] hahaha, shove gdb into installer environment =) [16:37] anna-install gdb-udeb did not work. I shall ask doko to fix that =) [16:38] don't do that :) [16:38] anna-install openssh-client-udeb, scp the stuff you need from somewhere [16:38] i have ssh, yes. [16:51] launchpad timing out filing a bug... [16:51] (Error ID: OOPS-ffb6abd15e291af596278d1a8b6a07ac) [16:54] cjwatson, cyphermox - well that's a first for me =) [16:54] divide by zero! [16:55] cjwatson: ah, that's what you meant by *brutally* shoving things in the installer environment :) [16:56] https://bugs.launchpad.net/ubuntu/+source/parted/+bug/1541510 [17:01] well, I'd start by applying the patches you pointed out earlier [17:01] at least one hunk of [PATCH 2/2] isn't there. [17:02] or you know, whatever ended up being committed from those emails.. [17:08] well, i can clone the parted trunk, rebuild it and retest with that. [17:09] cause if parted master is also broke, there is nothing to cherrypick [17:09] looking at the code. [17:09] it looks as if there is a simple ioctl performed that should return number of heads and doesnt [17:16] right, and the last few commits change what APIs are used to find out disk geometry. [17:20] heh, nevermind, we indeed do have at least some of this in the parted package [17:21] cyphermox: isn't there> you mean the bit that makes ped_device_like_dasd part of the API? That was explicitly rejected and isn't necessary [17:22] the LP OOPS above is a not uncommon but transient problem [17:25] cjwatson: nah, I was looking at libparted/labels/dasd.c and failed to notice the patches weren't applied. [17:28] I think you're going to see the same thing upstream [17:30] The more generic _device_probe_geometry has explicit checks for getting non-zero sectors and heads back from HDIO_GETGEO [17:30] and if it doesn't get them it uses defaults [17:30] But fdasd_get_geometry doesn't have quite the same checks around its use of HDIO_GETGEO [17:31] You ought to compare with the relevant kernel code, but my suspicion is that you don't get a sensible geometry back for an LV [17:32] This is all inside dasd_probe, and an LV isn't really a DASD [17:33] So one thing I'd try is making that situation be an error from fdasd_get_geometry and seeing if that behaves reasonably [17:34] Something like http://paste.ubuntu.com/14868640/ maybe? [17:35] Ew is that file really tabstop=4? Anyway [18:03] xnox: ^- suggestions above for you BTW [18:05] horum, ok. [18:07] cjwatson, i was like throwing SIGFPE is not nice for a library =) http://paste.ubuntu.com/14869013/ [18:08] also library functions that have "void" return signature are also not nice. [18:11] xnox@devac03:~$ sudo parted /dev/mapper/new-lvol0 [18:11] GNU Parted 3.2 [18:11] Using /dev/mapper/new-lvol0 [18:11] Welcome to GNU Parted! Type 'help' to view a list of commands. [18:11] (parted) print [18:11] BUG: trk is zero in vtoc_set_freespace call [18:11] BUG: trk is zero in vtoc_set_freespace call [18:11] Model: Linux device-mapper (linear) (dm) [18:11] Disk /dev/mapper/new-lvol0: 2097MB [18:11] Sector size (logical/physical): 4096B/4096B [18:11] Partition Table: dasd [18:11] Disk Flags: [18:11] Number Start End Size File system Flags [18:11] nicer.... [18:12] although, people really should not be trying to partition a logical volume, but meh, libvirt and openstack have other ideas.... [18:35] xnox: I think that patch is a mistake. [18:36] xnox: It should be using some different probing method that isn't DASD in this case. [18:36] xnox: Can you try my patch above? [18:36] sure. [18:36] An LV is not a DASD and so the attempt to probe it as dasd ought to fail [18:36] IMO anyway [18:37] i thought that all the probing patches committed parted were specifically to do that, e.g. regardless of upper layers keep probing until one gets down to dasd.... [18:37] or so i read the mailing list thread. [18:38] Maybe it needs to be adjusted at some other level (as well?) then [18:44] cjwatson, with just your patch, without mine, i get: [18:44] Error: FPE_INTDIV (Integer: divide by zero) [18:44] let me rebuild without stripping all the debug symbols, and run it in gdb to check where it exploades now. [18:51] same places... same bug... [18:51] i'm guessing fdasd_error() doesn't really do anything, and e.g. a goto error is needed too to go and do something else. [20:05] xnox: Yeah, maybe details need to be tweaked. I still think this basic approach is closer to what you want than trying to make dasd struggle on in this mode [20:06] ok. [20:06] this thing kind of derailed my goal to install three zvms with lvm2 and like do a juju environment on them. [20:06] *sigh* [20:24] cjwatson, actually i agree with you. that code is about e.g. finding dasd exposed via virtio. not about lvm. these are not the drives parted is looking for. === zequence_ is now known as zequence [22:55] HI [22:55] any one active now? [22:55] :) [22:59] Ruby_Rocks_007: there's mostly always someone around, you should just ask your question and stay around, waiting for someone to answer :) [22:59] (as per topic) [23:00] Sure, [23:01] My question is: how can I write a shell script, and ask preseed to call it? [23:01] my problem is: I have a SSD drive detected behind mechanical disk, [23:02] but my OS needs to be installed onto SSD drive (detected as sdm instead of sda), [23:02] how can I notify preseed to put OS on sdm, or sdy (if other 24 mechanical disks iinstalled), instead of sda? [23:04] there's a preseed you can set to pick the right drive, except you'd tell it /dev/sdm, or /dev/sdy, and it will use that drive, whatever it is [23:04] if you need more logic, then I would look into https://www.debian.org/releases/stable/i386/apbs05.html.en [23:05] it shows how you can preseed things as an early_command, or how to include another preseed file, or even run a script to set the debconf values. [23:06] I see your point, this company have close to 10 diff hardware combinations, [23:06] it’ll be great if preseed can call my shell to dynamically pick up the SSD drive. :) [23:07] I can not use a fixed one, say, sda, sdm, sdy, blah, that will be tedious and error-prone, I assume.