[00:05] JonB (~jonbendts@129.142.112.33.ip.tele2adsl.dk) joined #vserver. [00:38] Sh[a]de (shade@cpe109.bb101.cablesurf.de) joined #vserver. [00:38] hello back everybody [00:41] no Bertl yet... [00:50] moin [00:52] argh [00:52] ok.. today i'm gonna get this junk workin.. [00:52] re [00:53] re [00:53] any1 in here who can tell me how bertl wants the 1.23 to be stressed? [00:54] got a Hyperthreading or dual cpu system? [00:54] i think that was the issue :) [00:57] so i shouldn't have a problem on a 2.4.24 kernel right? [00:58] real smp [00:58] dual athlon mp 2800+ [00:58] johnny: ? [00:59] robjs (rob@212.69.243.51) left irc: Remote host closed the connection [01:01] kestrelw (~athomas@o2rosock0a.optus.net.au) joined #vserver. [01:06] jut wondering if anybody had any troubles with a 2.4.24 kernel [01:09] upgraded a host-system to from 2.4.22-vs1.20 (maybe it was 1.21?) to 2.4.24-vs1.22 4 days ago, the machine is stable as never before [01:10] s/to from/from/ [01:10] word [01:10] that's the only patch you have installed? [01:10] yes [01:11] Nick change: Bertl_oO -> Bertl [01:11] hi folks! [01:11] hi bertl! [01:11] < Doener> upgraded a host-system to from 2.4.22-vs1.20 (maybe it was [01:11] 1.21?) to 2.4.24-vs1.22 4 days ago, the machine is stable as [01:12] never before [01:12] then I guess you are going to like 1.23 ;) [01:14] Bertl, i'm doing a plan vs1.23 patch now [01:14] no grsec [01:14] just 2.4.24 [01:14] good idea, for a start ... [01:18] to all: any suggestions which direction linux-vserver should take in the near future? wishes? issues? [01:19] never crash? [01:19] :) [01:19] i'd say the tools need worked on [01:19] more than the patch [01:20] but i haven't checked out util-vserver yet.. [01:20] so will revise opinion in a bit [01:20] ok Bertl how can i make grub try this kernel once ? [01:21] enter the grub shell, and specify 'savedefault --once --default=' [01:22] hmm, what distro? [01:30] morning [01:30] morning, how is your vserver today? [01:30] gentoo [01:30] herbert: it is most excellent :) [01:32] actually, i have a user-space tool which I might forward on to enrico and jacques. it's a shell replacement that you give a user on your real server which logs you in to a vserver using the GECOS field: athomas:x:1004:100:athomas@swapoff:/:/sbin/vslogin [01:33] updates utmp, wtmp and changes and restores ownership of the pty to the correct virtual user [01:33] johnny: hmm, check first of the sequence is correct, some grub patches need a specific sequence of --once and --default to get it right, have a look at your grub info page ... [01:34] kestrelw: hmm interesting, you think of such cases where the user managed somehow to disable sshd, right? [01:35] well, actually it came about because the virtual servers are running on rfc1918 networks on dummy0, due to IP address allocation constraints [01:36] kestrelw: I was thinking of modifying pam_chroot to do the same thing [01:36] so, here i stand [01:36] but, as you say, it would also work if the user didn't want sshd running on their vserver [01:36] Linux red1 2.4.24-vs1.23 #1 SMP Tue Jan 13 18:09:36 UTC 2004 i686 unknown [01:36] waiting for orders :) [01:37] mrbawb: not a bad idea...unfortunately that is not an option for me as i am running slackware, which does not use pam [01:37] maharaja: well try to break it, try to bring it down, really hard! [01:38] bertl: my problem is, that i do not know how to *g* but ill take a look [01:38] shall i do stuff inside a vserver, or on the host system? [01:39] well, best both ... [01:39] Bertl: here's a feature request for vserver: http://www.chiark.greenend.org.uk/~sgtatham/putty/wishlist/moon-on-stick.html :) [01:39] let me see, I have some killers lurking around, you compile the thing with 2.95/2.96 did you? [01:39] doh [01:39] hehe [01:40] 3.2 or 3.3 [01:40] whats the main difference regarding vservers? [01:41] lol [01:41] MrBawb: that is a must-have! 8-) [01:42] where is --once .. i can't seem to find a reference to it at all [01:42] well 3.0, 3.1 and some of 3.2 seem to produce 'instable' code ... [01:43] mhm, i guess i test it with the current setup then :) [01:43] if its stable, it's good to know [01:44] okay ... [01:45] http://vserver.13thfloor.at/Experimental/killer-01.c [01:45] http://vserver.13thfloor.at/Experimental/killer-02.c [01:45] http://vserver.13thfloor.at/Experimental/killer-nathan-01.c [01:45] http://vserver.13thfloor.at/Experimental/killer-nathan-02.c [01:46] crashed with killer-01.c [01:46] together with some loops cat-ing the /proc//status files [01:46] maharaja: you must be kidding?! [01:46] or maybe it's only not responding to ping/... ;) [01:46] what does that d o? [01:46] ill have to switch monitors [01:47] well it's a fork bomb with context creation ... [01:49] that's not good .. [01:50] well... i haven't seen a reference to once yet [01:50] hmm, what do you mean by reference? [01:51] in grub info [01:51] i did search for it too [01:51] ah, maybe your grub version doesn't include that ... it's distro dependant I guess ... [01:52] if that is the case, I'm sure there are some patches around, which add it for gentoo too ... [01:52] sys-boot/grub-0.93.20030118 [01:52] tis from last year .. almost an entire year [01:53] http://www.mail-archive.com/bug-grub@gnu.org/msg07805.html [01:53] maja|ipv6 (maharaja@is.the.one.who.rules.at) joined #vserver. [01:53] re [01:53] http://www.pixelbeat.org/patches/grub-0.93-once.diff [01:54] Nick change: maja|ipv6 -> maja [01:54] hi maja! [01:54] do you want me to log something? [01:54] well is there something to log? [01:55] Nick change: maja -> maharaja_ [01:55] it generates a lot of numbers on the screen [01:55] way to fast to see what it is [01:55] well that is okay, if it dies, it stops printing those numbers ... [01:55] and replaces them with a big ugly panic ;) [01:55] ill test on the machine, without network and stuff [01:56] you can adjust the whole thing by using killer-02.c which doesn't explode in your face, or at least stops after 100000 forks ... [01:57] maharaja_ (maharaja@is.the.one.who.rules.at) left irc: Client Quit [01:57] IIRC nathan managed to run about 3 killers and 4-6 proc loops on his machine ... [01:58] maharaja_ (maharaja@is.the.one.who.rules.at) joined #vserver. [01:58] 23:59 < Bertl> IIRC nathan managed to run about 3 killers and 4-6 proc loops on his machine ... [01:58] so back, and inside of a screen ;) [01:58] ,h, [01:58] mhm [01:58] well, i don't *g* [01:58] i wrote down the last number: 52589 [01:59] but i guess that does not help [01:59] hmm, okay, but it didn't panic, right? [01:59] well, honestly i do not exactly know how a panic looks like [02:00] is the panic if linux has no root drive to mount the same? [02:00] like blinkin leds and stuff [02:00] yup [02:00] ok, so no panic [02:00] there was no response at all [02:00] no numlock [02:00] no interaction [02:00] i ran the killer on the host [02:00] hmm, that is a little odd ... [02:01] as a unprivileged user [02:01] running it on the host is okay .. [02:01] s/a/an/ [02:01] I mean that is the idea behind it ... [02:02] well, it's hard to tell what happened ... if you do not get a response it could 'just' be very busy ... [02:02] nathan_ (~nathan@h0004e24bdf09.ne.client2.attbi.com) joined #vserver. [02:02] ahh, the crash master appeared! [02:02] heya [02:02] nathan: i'm crashier than you :) [02:02] thats fine by me :) [02:02] i like it when other people are crashing too, i get lonely [02:03] Bertl, just got back from working on a local dual, might be able to schedule some time on it if we get into major problems again. [02:03] okay, so how is vs1.23 doing for you, nathan? [02:03] v1.23 is fine so far [02:03] havent stressed it, just playing with the vproc [02:03] could you give maharaja a few hints how to test? [02:04] hmm well what would we like to test? shall we still hit up the proc stuff to get a second opinion? [02:04] he lost his machine with killer-01 ... [02:04] really? [02:04] on 1.23? [02:04] yep [02:04] 2.9{5,6}? [02:04] gcc [02:04] well, 1.23 gcc 3.2 but no panic [02:05] our basic testing revealed that gcc 3.0 was generating unstable code though, i thought? [02:05] we never pinpointed any buggy code relating to the crash [02:05] yes, that is my opinion too ... [02:06] ok, so i shall use 2.95 and recompile my kernel? [02:06] yes [02:06] k [02:06] keep the 3.0 binary around though [02:06] just so we can test again to confirm [02:06] but I guess without some little tricks of yours, he won't be able to start all the stuff you did ... [02:06] Tamama (Pluh@a62-216-20-152.adsl.cistron.nl) left irc: Quit: one little two little three little piggies OINK! OINK! OINK! [02:07] maharaja SMP right? [02:07] yes, AMD this time, IIRC [02:07] ah [02:08] nathan: dual athlon mp 2800+, tyan board with amd chipset, 2x512mb ddr ecc reg [02:08] JonB (~jonbendts@129.142.112.33.ip.tele2adsl.dk) left #vserver. [02:08] at least, i stress the raid system too :) [02:08] rebuild is still in progress [02:08] -j 10 option to make? [02:09] ? [02:09] -j should help you build your kernels a little faster on multiproc [02:10] mhm [02:12] whats the best method in using 2.95 [02:13] do you have it installed? [02:13] simply changing the link /usr/bin/gcc to point to /usr/bin/gcc-2.95? [02:13] yes [02:13] doe sit work if you call gcc-2.96 [02:13] no [02:13] edit Makefile [02:13] ah okay 2.95 ... [02:13] just edit the Makefile and change gcc to gcc-2.95 [02:13] there will be two places, HOST_GCC and KERNEL_GCC iirc [02:13] actuall KERNEL_GCC is the important one ... [02:13] ok, and i shall add -j 10 to the cflags [02:14] and add -g to the KERNEL_CFLAGS [02:14] -j is a make option [02:14] so like: make -j 10 bzImage [02:15] CFLAGS_KERNEL it is [02:15] ok [02:18] what does "make dep" actually do? [02:18] creates dependency information accross the kernel [02:18] across? heh [02:19] i'm just wondering, why you hit "make dep bzImagee" ifor if "make bzImage " would suffice [02:20] make dep should be at least run once when first building a kernel [02:20] and then make bzImage could be used and hope that makefile rebuilds everything needed [02:20] mhm [02:20] Last message repeated 1 time(s). [02:20] laggy [02:21] Bertl, hows the ipv4root locking code? good? [02:21] i dont remember if we ever tested that on 1.23 [02:21] basically you need make dep on a 2.4 kernel whenever the dependancies changed, means you included a new file or something like this ... [02:22] nathan_: well it should be safe, but as you said, we never tested, IIRC [02:22] Action: nathan_ notes that the vproc code is rather clean and elegant [02:22] kernel vproc code i mean [02:23] well, I have to admit, it's stolen, but as this is a classical approach ... and even legal on GPL ;) [02:24] where was it pulled from? [02:24] the idea I mean ... base.c the rewrites done on the procfs ... [02:25] but there are not too many ways to implement this anyway ... [02:26] hmm i ponder if i should just rip it and apply it to my 1.3.4 that i want to get running or start messing with 1.3.5 [02:26] I did a little testing regarding the pid proc buffer, which is set to 20 entries per request ... and has quadric behaviour ... [02:27] proc pid buffer? [02:27] I guess this is something where we should try to 'improve' the 2.4 kernel for vserver ... [02:27] ok, i'll install and reboot [02:27] brb [02:30] does the proc pid buffer cache proc_readdir/proc_lookup or what? [02:32] mhm, ok, somehow the v1.23 patch slipped out [02:32] recompiling [02:34] hehe ... sorry ;) [02:34] nathan_: yes, the actual pid information is retrieved in pack to 20 .. [02:35] and each time the entire tasklist is walked to reach the correct position ... [02:35] so first request : 20 walks [02:35] second request : 40 walks [02:35] third request 60 walks ... [02:35] with 1000-2000 pids, this _is_ slow ... [02:36] if you want to take a look at my .config: http://ipax.tk/~raoul/config [02:36] 2min50 for my kernel [02:36] mhm, the load did not go over 1.... [02:36] anyways, rebooting ;) [02:36] hmm you should check if highmem really pays off on that machine ... [02:38] yeah, just removed last bit of perl code from util-vserver... [02:40] (new and enhanced) implementations for lsxid,chxid,showattr,setattr are in CVS too [02:40] ahhh great ... where can I get it?! [02:41] and most parts are compiling with gcc296 ;) [02:41] even better, you are a genious ;) thank you! [02:42] Bertl: either directly from savannah CVS, or I will make a tarball [02:42] that would be nice, I like those tarballs ... [02:43] they are so juicy and ... brb, getting something to eat ;) [02:44] re [02:44] ok, killer.c is running fine [02:44] load avg of 35 [02:44] load avg of 35.77 [02:45] mahara, so last boot you did the same thing, but with gcc3 correct? [02:46] guess so, we should post an advisory to use 2.95/2.96 for kernel compiling ... [02:46] yea sounds it [02:47] and maybe start testing with 3.3.2 or so ... [02:48] ok [02:48] Last message repeated 1 time(s). [02:48] i crashed [02:48] same symtoms [02:48] hmmm [02:48] really? [02:48] but a lot later [02:49] Action: nathan_ scratches head [02:49] define crashed? [02:49] nothing worked anymore [02:49] i saw irssi on the screen [02:49] hmm locked, right? [02:49] yebamaybe [02:49] what do you mean by locked? [02:50] well, no cpu, no disk activity ... nothing ... [02:50] as far as i can tell, yes [02:50] it's just standing there and waiting ... [02:50] its pretty noisy in here [02:50] do you have a disk led? [02:51] mhm [02:51] somehow i'm again messing up [02:52] ill wipe everything off my homedir and start with a fresh tar xzf linux-2.4.24.tar.bz2 [02:52] hmm ... [02:52] sry about that [02:52] did you do a clean tar for gcc change? or just changed Makefile on the last build? [02:52] i guess it was 3.3 again [02:53] i got several src dirs prepared for a monster kernel make stress test [02:53] and i guess i used the one where i did not modify the Makefile [02:55] phhh ... maja, you should stress the machine, not our nerves ... 8-) [02:55] sorry about that [02:55] don't worry ... [02:55] any testing is useful and appreciated ... [02:56] nathan_: what type of server have you got? [02:56] maharaja, dual 2ghz xeon [02:57] ad gcc changes to gcc-2.95 [02:57] i only find HOSTCC in the makefile [02:57] a little further down ... [02:57] and CFLAGS_KERNEL where i added -g [02:58] or do you mean CC = $((CROSS_COMPILE))gcc [02:58] CC = $(CROSS_COMPILE)gcc [02:58] ok [02:58] so i simply have to append -2.95 to these two lines [02:58] yup [02:59] at least, i see that the resync of my raid is working :) [02:59] after those crashes [03:00] Bertl, what do you use to generate your diffs? [03:00] with the prototype in the diff headers [03:00] huh? [03:01] what are you asking me? [03:01] 1197 --- linux-2.4.23/fs/exec.c Fri Nov 28 19:26:21 2003$ [03:01] 1198 +++ linux-2.4.23-vs1.3.0/fs/exec.c Wed Dec 10 01:39:55 2003$ [03:01] 1199 @@ -729,7 +729,7 @@ void compute_creds(struct linux_binprm *$ [03:01] for instance [03:01] oh wait actually [03:02] diff -NurpP --minimal linux-2.4.24-vs1.3.4/fs/ioctl.c linux-2.4.24-vs1.3.5/fs/io [03:02] ;) [03:02] Action: nathan_ tries [03:03] ah -p :) [03:04] well you can make it -NPurp (if you remmeber that better ;) [03:04] or -PurpN ... [03:05] i like -PurpN [03:08] Bertl: http://www-user.tu-chemnitz.de/~ensc/util-vserver/alpha/ [03:10] vconfig is needed by util-vserver-0.27.193-0 [03:10] iproute is needed by util-vserver-0.27.193-0 [03:10] dietlibc >= 0:0.22 is needed by util-vserver-0.27.193-0 [03:10] huh?! [03:11] iproute I can understand, what is vconfig? [03:11] first two just for correct paths [03:11] vconfig is the VLAN stuff [03:11] ahh okay, I remember ... [03:12] I do not know if the rpm works [03:12] and dietlibc, what about that? I'm on no diet ;) [03:12] it's recommended, but it builds without it also [03:13] ah okay, checking ... [03:15] src/vps.c: In function `readOutput': [03:15] src/vps.c:133: parse error before `char' [03:15] src/vps.c:139: `offset' undeclared (first use in this function) [03:17] ah ok, I tested gcc296 compatibility before implementing vps... [03:18] <_shuri> hi [03:19] re [03:20] ok [03:20] it too crashes with 2.95 [03:20] well, that is bad ... it hangs? or any panic? [03:20] s/crashes/locks/ [03:20] no panics at all [03:21] okay, could you try to enable kernel debuging in the make menuconfig, and select spinlock and magic sysreq? [03:22] fyi, i will put my config/makefiles to ipax.tk/~raoul/vserver [03:22] oki ... [03:24] and i do not know if it makes any differences, but i use the make-kpkg from debian to build my kernels [03:24] hmm, well I don't know what this does with the kernel ... you? [03:24] yo [03:25] what was the purpose of that message Bertl? [03:25] it should do nothing harmfull [03:26] johnny: which one? [03:26] vconfig is needed by util-vserver-0.27.193-0 [03:26] iproute is needed by util-vserver-0.27.193-0 [03:26] dietlibc >= 0:0.22 is needed by util-vserver-0.27.193-0 [03:27] ahh that was for enrico ... he explained that to me ... [03:28] hmm... wonder if i should really try to patch grub... [03:28] Bertl: gcc296 fault is caused by a semicolon after a #define; replace '#define HUNK_SIZE 0x40;' with '#define HUNK_SIZE 0x4000' in src/vps.c [03:29] ok, i compiled the kernel with the debugging stuff [03:29] hmm... who still uses gcc 2 ? [03:29] reboot? [03:29] we do ;) [03:29] ensc: hah, actually a coding error? [03:29] debian? [03:29] i guess that they use gcc2 for woody [03:29] Bertl: not really; it was still valid C99 ;) [03:29] i thought debian was switching over to gcc 2.95 soon.. [03:29] err to 3 [03:29] from 2.96 [03:30] bertl: shall i reboot with the new kernel? [03:30] yes, it now has -g spinlock debug and magic sysreq? [03:30] ensc: seems to compile fine now ... [03:30] yes [03:31] good, reboot ... [03:31] Bertl: gcc296 faults on 'vunify' are known... [03:32] well that is g++296 ;) [03:32] Bertl: no, vunify is/will be C [03:33] hmm, okay what is known then? please explain? [03:33] but some (boring) parts are missing in its implementation [03:33] ah okay, so you actually replaced the vunify by something sane in C? [03:33] src/vunify.c:198: warning: ISO C89 forbids specifying structure member to initialize [03:33] src/vunify.c:198: initializer element is not computable at load time [03:33] src/vunify.c:198: (near initialization for `d_path.d') [03:33] done [03:33] well, warnings, you get used to them after a hundred kernel compiles ,) [03:34] Bertl: I had to replace 'vunify' in any case since it is tied very much to the old cfg-scheme. [03:34] bertl: standing by ;) [03:34] maharaja_: okay now you crash it again but first try to find the magic sysrq ... [03:34] bertl: how do i do that? [03:35] I always forget the key combo .. (because I actually never do it on the keyboard) let me look for it ... [03:36] On x86 - You press the key combo 'ALT-SysRq-'. Note - Some [03:36] keyboards may not have a key labeled 'SysRq'. The 'SysRq' key is [03:36] also known as the 'Print Screen' key. [03:36] try with ALT-SysRq-h first ... [03:36] this should give you a help ... [03:36] ensc, does your toolset work better with other non rpm based distros ? [03:38] mhm, every key key except h gives me the help screen [03:38] johnny: vunify is unifying arbitrary paths now (with powerful excludelists which can use information from pkg-management), so it is not tied to rpm/dpkg anymore. But the vserver build methods are for rpm/apt-get only [03:39] maharaja_: okay, so you get a 2 line help there? [03:39] yes [03:39] loglevel0-8 .... Unmount [03:39] for how long ensc? [03:40] is it somewhat abstracted [03:40] maharaja_: okay that looks good ... [03:40] actually, do you ahve a serial null-modem cable available? [03:41] no [03:41] ill have a look but i dont think so [03:41] johnny: yes, it is abstracted and I have a 'copy' build-method in mind. I hope that I can release a beta version in this month [03:41] johnny: for other build-methods I do not know enough distributions [03:42] Bertl: no such thing in reach [03:42] well i can help you with gentoo... [03:42] although [03:42] perhaps zeroinstall and autopackage would be good options too [03:43] maharaja_: well, it would save you a lot of typing ... [03:43] okay, lets see where whe get ... try to crash it again ... [03:43] but i could give you a ssh shell if you [03:43] hmm.... i wonder ifi should just restart this .. [03:43] what do you think Bertl... [03:43] if that helps [03:43] maharaja_: that doesn't help .. [03:43]