/srv/irclogs.ubuntu.com/2005/12/11/#ubuntu-boot.txt

Kamioninfinity: l-r-m was breaking d-i, so I've uploaded what I think should be a fix for that12:25
Kamionhard to test without building new images though12:25
Keybukheh12:25
KamionKeybuk: line 7 of 10-devfs.rules looks suspicious; missing %?12:28
Keybukyup12:29
KamionKeybuk: could you replace awk '{print $2}' with sed 's/^[^ ] * //; s/ .*//' in /lib/udev/scsi-devfs please? we don't have awk in d-i12:30
=== Kamion -> bed
Keybukcan you mail that one to me?12:31
=== dilinger [i=dilinger@mouth.voxel.net] has joined #ubuntu-boot
=== pitti [n=pitti@ubuntu/member/pitti] has joined #ubuntu-boot
=== HiddenWolf [n=HiddenWo@136.6.dynamic.phpg.net] has joined #ubuntu-boot
=== fabbione tests root=/dev/disk/by-uuid/*
fabbionecat /proc/cmdline10:44
fabbioneroot=/dev/disk/by-uuid/4d216a51-2317-47a4-872f-125a6d035d08 ro quiet splash10:44
fabbioneit seems to work10:44
fabbionebut there is another side effect we didn't think when writing the specs10:45
fabbioneroot is not enough10:45
fabbionewe need to possibly detect other partitions and swap10:45
fabbioneotherwise even if we get root, we might miss the rest of the system10:45
fabbionethat defeats the purpose of the spec10:45
fabbionebah Keybuck isn't here10:46
infinityBut other partitions should be in /etc/fstab...10:49
infinity(by uuid, perhaps, but they should be there anyway)10:49
fabbionethe problem is that swap doesn't have a uuid10:54
Kamionfstab supports the UUID= and LABEL= syntaxes10:54
Kamionyou can use LABEL= for swap10:54
fabbioneand the thing is that yes, the other partitions needs to have uuid too10:54
fabbionefail to do so, we are in trouble10:54
infinityBut multiple swap partitions can have the same label.10:54
Kamionmm10:54
fabbionelike a / on ext3 = ok, /home on xfs = nok10:54
infinityI thought uuid-on-swap was something we were fixing?10:55
=== infinity thought he heard something about this during the breezy cycle.
fabbioneuuid is a property of the fs10:55
infinityYes, and?10:55
fabbioneswap isn't exactly a fs10:55
infinitymkswap could do uuid magic.10:55
infinityswap isn't an FS, but it still has magic.10:55
infinityNot much, but some.10:55
infinityAnd losing another few bytes for a UUID isn't the end of the world.10:56
fabbioneno, but you would introduce tons of other problems10:57
fabbionethink of the resume from hibernate that clear the swap or so10:57
fabbioneyour approach would assume that swap isn't used by anyone else10:57
fabbionewhat if the user dual boot breezy/dapper10:57
fabbioneboth clean the swaps in 2 different ways10:58
fabbionebang10:58
infinityYeah, ew.10:58
infinityHrmph.10:58
infinityOf course, if this sort of breakage isn't done some day, it can never be done.  Catch-22.10:58
fabbioneyes i agree10:58
infinityI'd love SOME way to uniquely identify swa ppartitions.10:58
fabbionethere is :)10:58
fabbionepartition IDs10:59
infinityHar.10:59
fabbionebut still10:59
fabbioneyou don't know if that swap is the one you are allowed to use10:59
fabbioneso it still the suck10:59
fabbionea possible solution would be to generate a uuid10:59
fabbioneand assign it to the swap as LABEL11:00
fabbionethat might do11:00
infinityOh wel, for all the noise people make about the dangers of shared swap, using the wrong swap, etc, etc, it seems to be a reasonably uncommonly-tripped bug.11:00
infinityPeople THINK about it far more than they actually get bitten by it.11:00
fabbionei would rather avoid to be the one crossburned for that11:00
infinityAnd yes, we could generate a uuid-like label.  Would be ugly in fdisk output, but that's not the end of the world.11:00
fabbionei think actually.. that going for LABEL=uuid is a much better approach11:01
fabbionebecause it might allow us to do it for all kind of FS11:01
fabbionesince LABEL is at partition level11:01
fabbioneand not FS11:01
infinityDo we ge LABELs on all partition tables, or just MSDOS?11:02
infinityI get them on PowerPC.  Yay.11:02
fabbioneno idea11:02
infinityWell, MacPPC anyway.11:02
fabbionebut i guess we do11:02
fabbionewhat fs?11:03
infinity /dev/hda7         Apple_UNIX_SVR2 Debian             55398400 @ 2098368  ( 26.4G)  Linux native11:03
infinity /dev/hda8         Apple_UNIX_SVR2 swap                1136576 @ 57496768 (555.0M)  Linux swap11:03
infinityUhh, ext3, not that that matters.11:03
fabbioneno, it shouldn't11:04
fabbionethat's why i was cross checking ;)11:04
infinityNot sure if all partition tables have label-like metadata that gets translated, but at least the three release arches would be okay.11:04
infinityI don't recall if BSD disklabels and slices do it.11:05
fabbioneno idea11:05
infinity(That was one motivation for UUID, though, is that since it's not at the partition table level, we're not subject to weird architectures and their weird partitioning schemes)11:06
fabbionethat's why we can use uuid over label protocols11:06
fabbioneall of this assuming we can stick such a long label11:07
infinityLook for uuid, fall back to label, if we have neither, cry?11:07
infinityNo, we can't have labels that long they'd need to be a bit less unique.11:07
infinityDOS labels are... 12 chars?... 16?... I don't recall..  But short.11:07
fabbionecrap11:08
fabbionei suggest we postpone the spec for dapper+111:08
fabbionethis is getting already too hairy11:08
=== Keybuk [n=scott@descent.netsplit.com] has joined #ubuntu-boot
fabbionehey Keybuk11:16
fabbioneKeybuk: in about 10 minutes you want to check the irc log for this chan (last 30 minutes discussion)11:17
fabbioneabout probe for root fs11:17
Keybukok11:19
Kamioninfinity: DOS labels are 11 characters, according to 'head debian-installer/installer/build/config/x86.cfg'11:24
Keybuk*giggle*11:40
Keybukfor those following the "Dude?  Where's My DMA?" bug11:40
Keybukit cleared itself up for mdz all by itsself11:41
Keybukwith BenC saying the only change to piix was a tweak to the modular IDE driver patch that couldn't POSSIBLY have caused it11:41
Keybukupstream IDE subsystem maintainer has identified the patch causes problems for us11:41
Keybukit's the (you guessed it) modular IDE driver patch!11:41
KamionKeybuk: let me know whenabouts you're uploading udev with the changes I mentioned on IRC last night; I'll need to rebuild d-i with those to get the live CD working11:58
Keybukyup11:59
Keybukwill be this morning11:59
Keybukjust ploughing through the things everyone sent me overnight for my attention this morning <g>11:59
Kamionta11:59
fabbionei wonder if we have dpkg-query in d-i somewhere...12:03
=== fabbione adds another item in the things to check
Kamionno12:08
Kamionno plans to add it either12:08
Kamionyou can look through /var/lib/dpkg/status in d-i, but obviously that will only list udebs12:09
Kamionif you want to inspect /target, chroot to it12:09
Keybukinfinity: what's the general stance on useful-things-for-debugging in initramfs12:09
Keybuklike lsmod, udevmonitor, etc.12:09
fabbioneKamion: yeah the issue is that we can't trust what's in /target ;)12:09
fabbioneKamion: that's why i am looking at several different options12:10
Kamionfabbione: then you can't trust /target/var/lib/dpkg/status either ...12:10
fabbioneKamion: i know. that's why there will be 2 checks available. one that uses status and one that will gather pkgs info from the files installed on the system12:10
fabbionebut the latter is way slower than the former and prone to more false positives12:11
fabbioneas it stands now i am implementing all small shell functions do to almost atomic operations that requires less tools as possible12:11
fabbionediscarding one in favour of another it's easy12:12
Keybukfabbione: you asked me to check the irc log, I assume about the /dev/disk stuff?12:13
fabbioneKeybuk: yes12:16
KeybukI'm not sure I understand the problem ... the root filesystem knows where the other partitions are12:17
infinityKeybuk : I wouldn't be against a DEBUG=yes flag in initramfs.conf that optionally copied in some debug stuff, and could be seen by other hooks (like udevs) to copy debug stuff for them...12:17
fabbioneKeybuk: no it doesn't... it does if the disk doesn't change name..12:18
infinityKeybuk : Not sure we want debug bloat in all initramfs images, hence the flag springs to mind as sane.12:18
Keybukfabbione: the disk is named12:18
Keybukit has a serial number12:18
fabbioneKeybuk: but we are using uuid exactly because we want to be able to move it around12:18
Keybukyou could use /dev/disk/by-id/usb-<MODEL NO>_<SERIAL NO>-part212:18
Keybuk(inside the root filesystem)12:18
Keybukthen use root=/dev/disk/by-uuid or whatever to find the root filesystem itself12:19
fabbionethat won't work either..12:19
Keybukwhy not?12:19
fabbione0 lrwxrwxrwx 1 root root   9 2005-12-02 09:41 scsi-0ATA_WDC_WD1200JB-00CLinux_ATA-SCSI_simulator -> ../../sda12:19
fabbione <-12:19
fabbionethis is an IDE disk in a USB external box12:19
Keybukright?12:20
fabbionei could remove it from there12:20
fabbionestick it in my desktop12:20
fabbionethat would change to ATA12:20
fabbionesorry12:20
fabbioneata-*12:20
Keybukuse partition labels then12:20
Keybukand /dev/disk/by-label12:20
fabbioneKeybuk: label can only be 11 chars12:21
fabbionethere is a high possibility of clashing12:21
fabbionethe reason why we did chose uuid was exactly because of its meaning12:21
Keybukpretty low probabibility12:21
Kamionwhat's wrong with using /dev/disk/by-uuid/ for the other filesystems too (apart from swap)?12:21
fabbioneKamion: not all FS support uuid12:21
KamionKeybuk: quite high if you have two parallel installations done with the same installer12:21
Keybukmost the FS we care about support uuid12:21
Keybukfor those that don't generate a random label12:21
fabbioneroot on ext3 and home on fancy fs12:21
fabbioneyou are doomed12:21
Keybukthat gives you a 1 in 269561249468963094528 chance of clash12:22
Keybukwe don't support home on "fancy fs"12:22
fabbioneKeybuk: yes we do. we still provide the option for manual partitioning12:22
fabbioneand iirc not all the selectable FS have uuid12:22
Keybukso?  just tell them "uh, no, go back and fix this"12:22
Keybukor for those FS that don't support uuid, generate a randomly unique LABEL and force it upon them12:23
fabbioneso you would end up with a fstab that will look like: /dev/disk/by-uuid/$longnumber / etc...12:24
fabbioneand other lines with LABEL=12:24
fabbionethat will be quite inconsistent12:24
Keybukjust use /dev/disk/by-uuid/$uuid and /dev/disk/by-label/$label12:24
Keybukthat's not inconsistent12:24
fabbioneKeybuk: what happens if 2 lables are clashing?12:24
fabbionelabels even12:25
fabbioneat udev level.. does it go foobar? or it continues like if nothing happen12:25
Keybukjust makes a link for one of them12:26
Keybukand the changes of two labels clashing is incredibly remote12:26
Keybukwhen we have 250 billion billion installations, we can worry12:26
fabbionewith 11 chars is a bit less than that12:27
Keybukand that's just assuming we only stick to alphanumerics12:27
Keybukno, that's 11 alpanumeric chars12:27
fabbioneright12:27
KeybukDOS is 11 8-bit chars, so it's a number so high my computer can't calculate it12:27
Kamionin reality people do kinda want labels to be human-readable ;)12:28
Kamionor at least PRINTABLE12:28
Keybukindeed12:28
fabbionei sort of agree with Kamion12:28
Keybukthat's why I stuck to alphanumerics12:28
Keybukhell, if you just limit yourself to upper case and numbers, you're still over a billion billion possible labels12:29
fabbionei was wondering if the first 11 digit of md5sum something would do12:30
fabbionelike md5sum the output of the / uuid with the date in nanosecond12:30
fabbionethat would probably be almost unique12:30
Keybukwhy not just pack the / uuid with the partition number or date?12:31
Keybukthe / uuid is 37 hex characters12:31
Keybukactually, 33 and 4 -s12:32
fabbioneyeah12:32
Keybukyou could certainly use much of that12:32
Keybukand there are shorter uuid standards too12:32
infinityCan we convert base16 to base36? :)12:32
fabbionethe md5 was to scramble numbers completely12:32
Keybukanyway, this isn't a SKY FALLING problem12:33
fabbioneno12:33
Keybukthere's plenty of ways to identify the partition we want to put the swap on12:33
fabbioneKeybuk: other than LABEL no...12:34
Keybukthere are :p12:34
Keybukwe could even go as far as tricks12:34
Keybukwe could improve /dev/disk/by-path to create a tree of directories, and then figure out which one the root filesystem is on, then use ../part2 if we REEEEALLLY wanted12:35
Keybukbut that's a bit silly12:35
Keybukactually, we probably wouldn't need to improve it at all12:35
Keybukjust figure out which /dev/disk/by-path is the root fs12:35
fabbionethe problem isn't root12:35
Keybukthat assumes you want swap on the same disk, of course12:35
fabbionewe can't assume :(12:36
fabbionewe would be able to do so only in one condition12:36
fabbionegetting Kamion to kill manual partitioning12:36
fabbionetill users are able to do partitions as they like12:36
fabbionewe are kind of doomed12:37
fabbioneOR12:37
fabbionethe possibility is to add all this chunk of code only if they use our recipe's12:37
fabbionevia guided partitioning12:37
=== fabbione eyes the clock and ponders food
Keybukor just make the manual partitioning generate the random label for every new partition12:38
Keybukthen if the user changes it to something more readable, which might clash (like "SWAP") that's their problem12:38
Keybukand the bug can be closed with something like12:38
Keybuk"This appears to be your foot that you shot, with your gun"12:39
fabbioneLOL12:39
fabbioneno really.. we can do it only if they use our Guided stuff12:39
Keybukno, really, we can do it in manual12:39
fabbionewe can't ensure that the user doesn't have 2 disks12:39
fabbioneand stick half system on one12:39
fabbioneand swap on the other12:39
Keybukwe don't need to12:39
fabbioneok let me eat and think about it12:40
Keybukjust use by-uuid for everything that has it, and by-label for things that don't12:40
fabbioneotherwise i will soon be grumpy ;)12:40
=== jbailey [n=jbailey@modemcable139.249-203-24.mc.videotron.ca] has joined #ubuntu-boot
KamionKeybuk: could you put the /dev/discs/ support back into scsi-devfs, please? d-i needs it01:20
Kamionand the Debian 076 package has it01:21
Keybuk"back into" ?01:24
Keybukis it not there?01:24
KeybukI just used the scsi-devfs script from upstream01:24
fabbioneKamion: the hooks i need for that md5sum stuff are in the rescue-udeb right?01:27
KamionKeybuk: it's not there, no01:28
Kamionfabbione: rescue-mode01:28
fabbioneKamion: thanks01:28
Kamionfabbione: see yaboot-installer for (at present) the only example of another udeb using it01:28
Keybukmeh, I'll have to see what Debian patches they have for that01:28
Keybuk*mutters and grumbles about the installer using obsolete, unsupported and broken scripts* :p01:28
=== ..[topic/#ubuntu-boot:Kamion] : known: fb drivers loaded, oss drivers loaded, no network plugging, alsa rules not reloaded, mtab not updated, /dev/pts not mounted, scsi-devfs || fixed: notify-reboot, vio_type segfault, no modules loaded, hal/pcmcia rules not reloaded, nfs root fails, sata root fails, pnp devices not loaded, init stop/start, grepmap bitching
Kamionit's gradually getting fixed, but the live CD doesn't have parted_devices so disk-detect falls over01:29
=== ..[topic/#ubuntu-boot:Keybuk] : known: oss drivers loaded, no network plugging, alsa rules not reloaded, mtab not updated, /dev/pts not mounted, scsi-devfs || fixed: notify-reboot, vio_type segfault, no modules loaded, hal/pcmcia rules not reloaded, nfs root fails, sata root fails, pnp devices not loaded, init stop/start, grepmap bitching, fb drivers loaded
KamionI've got a patch heading into parted upstream (I hope) to avoid some more of the problems01:29
Keybuk*nods*01:30
Keybukgetting the installer to eventually search /sys for drives and cds would be much sweeter01:31
Kamionit'll just use parted to walk drives01:31
Kamionhmm, scsi-devfs.sh seems to be in extra/ and in the .diff.gz; does it come from upstream anyway?01:32
fabbioneKamion: it looks pretty simple the README looks clueful01:32
fabbione"and" the README..01:32
Keybukit comes from upstream, I guess Marco's copied it and modified it for his own nefarious purposes01:32
Kamionoh, meh, it's completely different01:32
Kamionit's been in Debian since 0.019-3; I imagine it predates upstream's01:33
Kamionfabbione: thanks, I wrote the README ;)01:33
fabbioneKamion: that might be why i was able to understand it :)01:34
KamionI'll just use a non-SATA/SCSI system for testing for now01:36
Keybuk*giggle* at kay03:11
Keybukafter spending most of yesterday evening arguing that there was no reason to run udevd in initramfs, but just place as /sbin/hotplug03:12
Keybukhe's now decided we were right after all03:12
=== HWolf [n=HiddenWo@136.77.dynamic.phpg.net] has joined #ubuntu-boot
jbaileyKeybuk: ^5!03:31
=== pitti [n=pitti@ubuntu/member/pitti] has joined #ubuntu-boot
=== pitti [n=pitti@195.227.105.180] has joined #ubuntu-boot
Keybukhmm04:05
Keybukit seems the ide subsystem could be going away :)04:05
Kamionhuh?04:05
HiddenWolfHuh?04:05
Keybuklibata has got parallel ide support now04:05
Keybukand that seems to be strongly favoured to replace much of the ide subsystem04:05
Keybukpretty much how sata is handled04:06
HiddenWolfs/away/replaced04:06
jbaileyKeybuk: I asked about that the other day, and I thought ben said ide-generic would still be needed for a while.04:12
=== pitti [n=pitti@ubuntu/member/pitti] has joined #ubuntu-boot
Keybukyou didn't dig enough <g>04:17
=== HWolf [n=HiddenWo@136.169.dynamic.phpg.net] has joined #ubuntu-boot
=== jbailey [n=jbailey@modemcable139.249-203-24.mc.videotron.ca] has joined #ubuntu-boot
=== pitti [n=pitti@ubuntu/member/pitti] has left #ubuntu-boot []
Keybukright, I think that's all the pile-o-udev bugs solved06:27
KeybukKamion: what installer issues do you have (currently) ?06:28
KamionKeybuk: scsi-devfs lack of /dev/discs, live filesystem isn't building, casper needs to be updated to do something initramfs-ish instead of pivot_root06:32
KamionI've been concentrating on the live CD rather than on the install CD so it's entirely possible that there are install CD bugs I don't know about yet06:32
KamionI'm rolling new install CDs now and will test them shortly06:33
KeybukI'm looking into the devfs thing now, I think I may just drop that stuff in wholesale from Debian06:33
makx Keybuk: could you take a look at debian #342057, BusLogic not loaded because !sysfs (seems needed for vmware testing of initramfs-tools).06:40
jbaileymakx: I think we fixed that in the Ubuntu kernel.06:40
makxjbailey: you added sysfs hooks? :)06:41
jbaileyBen or Fabio did, I think.06:41
jbaileyI try to stay on this side of userspace. =)06:41
makxwhy's that not upstream :-P06:41
jbailey    - [scsi/BusLogic]  Add MODULE_DEVICE_TABLE06:42
Keybukit probably is upstream by now06:42
jbailey -- Ben Collins <bcollins@ubuntu.com>  Sat, 12 Nov 2005 23:34:38 -050006:42
jbaileyFrom the 12th to now, it's hard to guess.06:42
jbaileyI know there's alot queued to go upstream.06:42
makxa quick grep in git-commits-head shows it's not yet landed for linus.06:44
makxjbailey: anyway thanks for the hint, good news. :)06:45
jbaileymakx: Or perhaps check with BenC in #ubuntu-kernel06:49
makxjbailey: for now i'll include it in d-kernel, but yes.06:50
=== jbailey_ [n=jbailey@modemcable139.249-203-24.mc.videotron.ca] has joined #ubuntu-boot
KeybukKamion: wow, I see what you mean about these scsi-devfs scripts being totally different07:19
Keybukthey aren't even related07:19
Kamionright, I don't think they share ancestry07:21
Keybukthe upstream one looks far more complete07:21
Kamionit does a lot of stuff I don't care about ;-)07:22
Keybukwhat do you actually care about/07:22
Keybukdo you need all the /dev/bus/*, /dev/scsi/*, etc.07:23
Keybukor do you just want /dev/cdroms and /dev/discs ?07:23
Kamion/dev/scsi/ is good, I don't especially care about /dev/bus/ personally07:23
KamionI don't think I actually NEED /dev/scsi/ though07:23
KeybukI'm trying to work out how to make this work07:24
Keybukwe plug all the disks at once, so get_next_number returns 0 for every single disk07:24
Keybukwhat I'm wondering is ...07:27
Keybukif I throw all of this away07:27
Keybukand instead replace it with a small "enumerate disc-like objects" helper that just makes /dev/discs and /dev/cdroms07:27
Keybukwould that be sufficient for the installer?07:27
Keybukor does it really need /dev/ide and /dev/scsi ?07:27
Kamionno, it doesn't need those07:30
KamionI use them occasionally but I can easily adjust07:30
Keybukso it really just relies on /dev/discs and /dev/cdroms ?07:30
Kamionyep07:33
Keybukhow about /dev/floppies?07:34
Kamion/dev/floppy07:50
Kamionfloppy-retriever looks for /dev/floppy/007:50
Keybukdoes it matter if there are gaps in the numbering, btw?07:54
Kamion/dev/floppy/0 has to be thus, for the rest probably not07:54
Keybukhow about for cdroms and discs?07:55
Kamion"for the rest probably not"07:56
Keybukah, sorry07:56
=== HiddenWolf [n=HiddenWo@136.169.dynamic.phpg.net] has joined #ubuntu-boot
dilingerKamion: any chance of seeing my parted /sys/block patches making it into parted07:58
dilinger?07:58
dilingerfor ubuntu's installer07:58
Kamiondilinger: have they gone upstream / to Debian?07:58
KamionI've got parted back into sync now, would be kind of a shame to branch again07:58
KamionI can if need be though, file me a bug07:59
dilingerKamion: they have gone upstream; I have no idea debian's status w/ them07:59
KamionDebian's tracking upstream pretty closely07:59
dilingerupstream seems to want me to assign copyright to the FSF, but doesn't seem to be able to put me in contact w/ whomever i need to talk to about that07:59
dilinger*shrug*07:59
Kamionfile a bug and I can keep an eye on it07:59
KamionI've done FSF copyright assignment; you can probably just mail their clerk07:59
dilingerKamion: file a bug in which bts?08:00
KamionUbuntu08:00
dilingerbugzilla?  malone?08:00
Kamionbugzilla08:01
Kamionhttp://www.fsf.org/licensing/assigning.html - the maintainer is supposed to send you the questionnaire though08:01
dilingerwell, it's svenl who's expressed interest in the patch, but i guess he wanted otavio to mail me or something08:03
=== dilinger shrugs
dilingerok, submitted08:19
Kamionthanks08:20
KeybukKamion: http://people.ubuntu.com/~scott/storage_enum.sh08:47
KamionI'll take your word for it on the implementation, but the intent looks plausible08:49
Keybukthat should actually fix our current problems too, as it should reliably never use the same number twice08:50
Kamion   * Update installer startup script to use "cp -au" like the init script,09:33
Kamion     instead of "cp -af".09:33
KamionKeybuk: please revert that - busybox cp doesn't have -u09:33
Keybukd'oh09:33
Keybuktsk09:33
Keybukbusybox sucks09:33
Keybukit has a -f that doesn't behave like the real one09:33
Keybukand then doesn't have -u09:33
Kamionand in any case I'm sure it was the "cannot be opened" trap I hit09:33
Keybukreverted that bit09:40
=== jbailey [n=jbailey@modemcable139.249-203-24.mc.videotron.ca] has joined #ubuntu-boot

Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!