[00:09] mhepp (~mhepp@r72s22p13.home.nbox.cz) left irc: Remote host closed the connection [00:09] hmm oddly enough my kernel doesnt compile now with the vs1.26 patch. [00:09] fs/fs.o(.text+0x9d0): In function `sys_fchmod': [00:09] : undefined reference to `IS_IMMUTABLE' [00:09] fs/fs.o(.text+0xab4): In function `sys_chmod': [00:09] xfs? [00:10] hmm, wait, there is no IS_IMMUTABLE in sys_chmod ? [00:23] where is fs.o comming from? i dont see an fs.c in the fs dir. [00:24] fs.o is the result of combining all sources from fs/* [00:25] but I have an idea where the IS_UMMUTABLE comes from ... [00:25] only thign applied to this kernel is teh bme patch and the vs patch. [00:25] there is the IS_IUNLINK check in sys_chmod ... which uses that ... [00:26] the funny stuff is both are defined in the same file, so either the compile should fail, or IS_IMMUTABLE should not be referenced ... [00:26] definitely your build is somewhat messed up ... [00:26] could you provide the .config file? [00:27] sure [00:28] talon.home.cosmic-cow.net/.config [00:28] i applied teh bme patch aganst rc1 and then teh vs patch aganst rc2 [00:28] it started as a vanilla rc2 kernel. [00:29] and thats the same config file ive been using all along. [00:30] Khahan (~Filbert@D577451D.kabel.telenet.be) left irc: Quit: the world is full of apathy, but I don't care [00:35] patching file fs/open.c [00:35] Hunk #6 FAILED at 489. [00:35] Hunk #7 FAILED at 523. [00:35] Hunk #8 succeeded at 564 with fuzz 1 (offset 1 line). [00:35] 2 out of 8 hunks FAILED -- saving rejects to file fs/open.c.rej [00:35] and you did take care of that? [00:39] hmm i might have missed that message. [00:39] 2.4.25-rc2 + bme0.04 [00:39] and then the vs1.26 ontop of that, just fails ... [00:41] i will try it without the bme patch then. i know it all worked togeather when i was using rc1. [00:41] might be, I have no problem adapting the bme to vs, once it's finished [00:44] not feeling so great today, heads a bit fuzzy. [00:44] i hope im not comming down with anything. [00:45] get some vitamines, a hot tea or milk, maybe an aspirin and go to bed ... [00:45] JonB (~NoSuchUse@kg203.kollegiegaarden.dk) joined #vserver. [00:46] hot tea sounds like a good idea. [00:46] i will see if i can get the oops to happen with this fresh kernel and go to bed i guess. [00:47] make it so ... [00:47] you a star trek fan or just like the phrase? [00:48] sure I am ... [00:48] a trekkie [00:48] Action: talon idly wonders if you look anything like picard. [00:48] fire at will. oh no not again! ;) [01:09] infowolfe (~infowolfe@pcp04891550pcs.frnkmd01.md.comcast.net) joined #vserver. [01:09] good morning Bertl [01:09] or evening, whichever it is :-p [01:09] ok booting new kernel. [01:10] Bertl (~herbert@MAIL.13thfloor.at) left irc: Ping timeout: 501 seconds [01:15] just wondering, are there any nifty features in 1.3.7 that i should be enamored with or should i just stick with 1.26? [01:16] can anyone else see www.13thfloor.at? [01:17] not just you. [01:17] woudl explain why bertl pinged out. [01:17] Action: talon runs an mtr. [01:18] hmm interesting. [01:18] can ping it and traceroute gets all the way theere. [01:18] no response from teh web server though. [01:21] weird kernel hang maybe? [01:21] :-\ [01:21] who knows [01:22] thank god i've got another copy of vs1.26.patch :-p [01:22] same here. [01:23] lol [01:23] im doig a hwole lot more kernel compiles than i ever expected to do since running across this project. [01:23] why? [01:24] i've been running 1.3.4 since it came out :-p [01:24] so i'm "downgrading" to 1.27 in the hopes that some of my current problems will go away (maybe the box just needs a reboot :-p) who knows [01:24] infowolfe: because normally i just patch teh kernel, it just works and i get on with life. with this though i keep running up aganst all kinds of odd bugs. [01:24] talon, like? [01:24] mostly quota system stuff. [01:24] but thats been fixed mostly. [01:25] now im tracking down an issue with nfs. [01:25] i also ran into issues with teh quotatools package. while investigating per context quotas. [01:25] yah, quotas are a bit experimental [01:25] and i don't think anybody's really run nfs [01:25] i use mine for web/gameserving [01:26] i dont mind so much since every bug i find and verify goes towards making a better vserver which im certanly intrested in. [01:26] totally [01:26] im trying to base a distro around vserver. [01:26] so i want to verify every feature as working and put it into a kernel package. [01:26] i just wish we could have a 2.6 vserver with all the fun stuff the 2.4 has :-( [01:26] talon, what kinda distro? [01:26] deb, rpm, tgz? [01:26] well that takes time. and id ratehr htey take their time and do it right vs having it tomorrow. [01:27] infowolfe: its a slackware based distro with a package manager borrowed form archlinux. [01:27] its going to be bundeled with a server appliance. [01:27] ewwww, slackware :-p [01:27] lol [01:27] why not go with gentoo? [01:27] somewhat simialr to the cobalt but not targeted exclusivly towards idiots. [01:27] (other than having to compile stuff yourself) [01:28] by the time its doen it wont look exactly like gentoo or slackware. [01:28] if it's based in slack though, i dunno, i'm a gentoo zealot :-p [01:28] i love the package management system [01:28] gentoo is a lot like slack in my opinnnion. [01:29] only thign different is emerge and the new rc system. [01:29] except gentoo doesn't suck as much :-p [01:29] the rest is easy and streightforward with no extra bells and whistles like slack. [01:29] and named runlevels, etc etc etc [01:29] i didnt go with the slackware package system for obvious reasons. [01:29] lol [01:29] and i probably wont keep teh rc system as is. [01:30] i'd have suggested a binary-only system built on gentoo, personally [01:30] serving (~serving@213.186.188.205) left irc: Read error: Connection reset by peer [01:30] it'd probably be a lot easier for you [01:30] everythign including vservers will be ocnfigurable via a perl library though. [01:31] with web/curses/command line tools built around that library. [01:32] with everythign documented ofcourse. [01:33] that way you can extend teh tools that come with the system, write your own or script anything you want. [01:35] the idea is to provide a cobalt like appliance environment for people that dont know what they are doing, while not getting in the way of peole that want to do more advanced stuff. [01:36] that's going to be a pain in the ass :-p [01:36] and i honestly think it'd be easiest for you to use gentoo as a base [01:38] i already got most of packaging finished which was the main pain in the ass. the rest i have to do anyway and it doesnt matter what distro its on top of. (i even have plans to make the configuration/virtual server stuff work on solaris 10 with the new zones feature) [01:40] its a hell of a lot more interesting than my last project. [01:42] which was? [01:42] working on unix again after doing java for almost two years is relaxing. [01:42] i'd love to turn gentoo into a more vserver friendly project, but it's a pain in the ass [01:42] infowolfe: 3rd party add on objects for chilisoft ASP. [01:43] think IIS for unix. [01:43] not make it more cracked out than you can imagine. [01:43] add in a java to COM bridge thats buggy as hell. [01:44] Action: infowolfe pukes all over the floor [01:44] then curl up ina fetal posistion under your desk. [01:44] i learned ot hate java with a passion. [01:44] gosling is on ym death list. [01:45] on the other hand i got a few nice bits from the sun ISV program like free development software. [01:45] lol [01:46] java = evil incarnate [01:46] amen to that. [01:46] praise php! [01:46] it wouldnt be so bad if the API wasnt so hell bent on keeping you form doing anythign useful instead of tieing you up ina plush streight jacket. [01:46] and if sun woudl stop changing it every other day. [01:46] lol [01:47] y'know why i don't like java? [01:47] and EOLing the last version after than your head spins. [01:47] and the write one run anywhere is a fucking joke.... [01:47] unless you program for java 1.1 or 1.2 only. [01:47] which is impossible to get anythign done in. [01:47] i can do much better things web-based with php and dhtml :-p [01:47] i wouldnt een consider java for web based stuff. [01:47] if someone needs something other than that, i'll give them putty and an ncurses menu system :-p [01:47] jsp is a joke [01:48] i made a few buck selling those java objects though. [01:48] this wasnt jsp. [01:48] it was ASP using vbscript under linux/solaris and using my java objects as COM objects to make extra functionality available. [01:48] Bertl (~herbert@vs253.rosehosting.com) joined #vserver. [01:48] JSP is even worse than that. [01:48] hui ... [01:49] lmfao [01:49] Bertl! [01:49] we've missed you [01:49] well, not my fault ... [01:49] what happened? [01:49] ok. [01:49] provider has some issues ... [01:49] it loosk like its the quota patch. [01:49] :( sorry [01:50] that does the oops [01:50] let me grab teh oops [01:50] what was the program you wanted me to run the oops code through ? [01:50] Frank00polo (~franko@wbar7.dal1-4-13-067-211.dsl-verizon.net) joined #vserver. [01:51] talon: ksymoops I'd guess. [01:51] which bit of the message? [01:51] Unable to handle kernel NULL pointer dereference at virtual address 00000034 [01:51] printing eip: [01:51] c01438e2 [01:51] *pde = 00000000 [01:51] Oops: 0000 [01:51] CPU: 0 [01:51] EIP: 0010:[] Not tainted [01:51] EFLAGS: 00000246 [01:51] eax: 00000000 ebx: 00000000 ecx: c9ebd910 edx: 00000000 [01:51] esi: c7cb47c0 edi: c7cb484c ebp: c7c5d240 esp: c7c5ff50 [01:51] ds: 0018 es: 0018 ss: 0018 [01:51] Process rm (pid: 314, stackpage=c7c5f000) [01:51] Stack: c7cb47c0 c7c5d240 c7c5d240 00000000 00000000 c01421e2 c7c89ec0 c7c5ff98 [01:51] 00000000 c9e15000 c7c5d240 c7c5ff90 c7c5d240 c0143c23 c7cb47c0 c7c5d240 [01:51] c7c89ec0 c9eef440 c9e15005 00000003 0024db2a 00000010 00000000 00000004 [01:52] Call Trace: [] [] [] [01:52] Code: 66 83 78 34 00 74 16 8b 4c 24 20 b8 00 e0 ff ff 21 e0 8b 51 [01:52] talon: everything. [01:53] >>EIP; c01438e2 <===== [01:53] Trace; c01421e2 [01:53] Trace; c0143c23 [01:53] Trace; c010727f [01:53] Code; c01438e2 [01:53] 00000000 <_EIP>: [01:53] Code; c01438e2 <===== [01:54] 0: 66 83 78 34 00 cmpw $0x0,0x34(%eax) <===== [01:54] Code; c01438e7 [01:54] 5: 74 16 je 1d <_EIP+0x1d> [01:54] Code; c01438e9 [01:54] 7: 8b 4c 24 20 mov 0x20(%esp,1),%ecx [01:54] Code; c01438ed [01:54] b: b8 00 e0 ff ff mov $0xffffe000,%eax [01:54] Code; c01438f2 [01:54] 10: 21 e0 and %esp,%eax [01:54] Code; c01438f4 [01:54] 12: 8b 51 00 mov 0x0(%ecx),%edx [01:54] okay, you compiled the kernel with -g? [01:54] no.. [01:54] how do i configure it to use -g [01:54] hmm, unfortunate ... [01:54] is that in menuconfig? [01:54] i can recompile it. [01:55] talon: makefile [01:55] CFLAGS_KERNEL = [01:55] change that to [01:55] CFLAGS_KERNEL = -g [01:56] do not change anything else ... [01:56] ok. [01:56] building new kernel [01:56] and that is with 1.26 + q0.13 but not with 1.26? [01:57] yeah. [01:57] without q0.13 it doesnt happen. [01:57] but with it does. [01:57] no bme patch this time around. [01:57] good [01:58] heh and all i wsa trying to do was tar up some files to nfs and grab from another box :) [01:58] wasnt even expecting to have new bugs for you so soon. [01:59] not even using it in production unless you consider a dev environment production use. [02:02] hmm, you did try with quota enabled? [02:02] I mean in both cases? [02:05] no. didnt even touch quotas or start any contexts yet. [02:05] when the problrm started. [02:05] hmm, but in the kernel config I mean ... [02:05] rc.local maps sdb1 to vroot/0 though. [02:05] oh yeah. [02:05] my config has quotas enabled. [02:05] havent tried without them compiled in. [02:06] so both, the vs1.26 and the vs1.26+q0.13 had quota enabled ... [02:06] not even suer if thats a good idea with q0.13 applied. [02:06] yeah. [02:06] okay ... [02:06] that config i gae you is the config i use for all kernels. [02:06] under the vmware box. [02:07] okay, let's see what the compile gives ... [02:07] yeah, sitll compiling with -g. [02:07] thats going ot be one big kernel. [02:12] unless it splits teh debugging symbols out. [02:13] debug symbols are in the vmlinux, the bzImage is clean ... [02:15] ok so its similar ot building BSD kernels for debugging. [02:17] i guess im just different enough from teh avrage user to keep running into this sort of thing. [02:18] keep it this way, open source can benefit from that .... [02:18] oh yeah, re teh quotacheck comments last night. his fix appears to be correct as in teh changelog. but i think its countign files in filesystems mounted under the filesystem it checks. [02:19] is thre a good way aside form checking fstab to tell if your crossing a mount point ? [02:19] you mean 'over' ;) yeah probably ... [02:19] erm i mean mtab. [02:19] im pretty sure its stating files in /proc for example. [02:19] correct method would be to stop when a dir is on a different fs, like xdev in find or -x for du [02:20] actually the check is pretty easy ... but I don't know if the quota tools do it ... [02:20] dont think they do. [02:20] should be trivial to make it so though. [02:20] well, that is probably a good reason for another bug report ;) [02:20] and another patch :) [02:21] had fun poking around in quotatools last time. [02:22] good practice trying to figure out someone elses code. [02:22] was starting ot get a bit rusty with C. [02:22] thank god its not in C++. [02:22] did you add macro support to edquota? :) [02:23] heh no. i aded a way to force the old filesystem scan method despite the mtab saying ext2 or ext3 [02:23] talon.home.cosmic-cow.net/quotacheck.patch-nodio [02:24] also patched an accounting bug with antoehr patch that was just fixed in 3.11 [02:26] i only did teh setquota.sh script because i was a dumbass and didnt do a man -k and figured that somthign useful non interactivly like setquota would be listed in teh SEE ALSO section of the edquota manpage. [02:26] ok debugging kernel built [02:27] booting. [02:27] hmm, well, actually no need to do that .. but anyway ... [02:30] id figure teh symbol possiitions would change. [02:30] so i rebooted ot generate a new oops to feed in. [02:30] yeah, is probably better anyway ... [02:32] oh yeah i have a feature request for ensc. nothing major. just a change to chbind to emit a warning and execute the command anyway if it cant find the kernel interface. this would help with rc scripts that use chbind to keep them from interfering with vservers. when you boot a vanilla kernel for whatever reason. ofcourse i could jsut have an if in the shell script that runs teh daemon again. hmm never mind. [02:34] hmm, right! [02:35] well, a smart way would be to define a special runlevel for vserver operation ... [02:35] yeah but thats still a pain since you have to change initdefault. [02:35] when you want to boot a vanilla kernel. [02:36] this way you could detect the kernel vserver capabilities on startup, and switch to whatever level is useful when vserver support is there [02:36] hmm oh i see. that might be interesting. [02:37] ksymoops -k vmlinux -m System.map from the source dir right ? [02:37] ksymoops -v vmlinux -m System.map -K [02:40] >>EIP; c01438e2 <===== [02:40] Trace; c01421e2 [02:40] Trace; c0143c23 [02:40] Trace; c010727f [02:40] Code; c01438e2 [02:40] 00000000 <_EIP>: [02:40] Code; c01438e2 <===== [02:40] 0: 66 83 78 34 00 cmpw $0x0,0x34(%eax) <===== [02:40] Code; c01438e7 [02:40] 5: 74 16 je 1d <_EIP+0x1d> [02:40] Code; c01438e9 [02:40] 7: 8b 4c 24 20 mov 0x20(%esp,1),%ecx [02:40] Code; c01438ed [02:40] b: b8 00 e0 ff ff mov $0xffffe000,%eax [02:40] Code; c01438f2 [02:40] 10: 21 e0 and %esp,%eax [02:40] Code; c01438f4 [02:41] 12: 8b 51 00 mov 0x0(%ecx),%edx [02:41] okay, now do: 'addr2line -e vmlinux c01438e2' [02:41] from inside ksymoops or back to the shell? [02:42] shelll [02:42] /usr/src/vserver/linux-2.4.25-rc2vs/include/linux/quotaops.h:95 [02:43] hmm, actually means that the quota stuff is not guilty (q0.13) [02:43] or at least it's very unlikely that it changes the way inodes appear/disappear ... [02:44] all i know is it doesnt happen without q0.13 [02:44] yeah that is okay ... [02:44] didn't say it's not my fault ;) [02:45] the quoa stuff seems to keep sucking me in one way or another. [02:46] my best idea is that the nfs code somehow releases the inode before the vx_* is called ... [02:47] try the following : add a iget(dentry->d_inode) and iput(dentry->d_inode) at the folowing places: [02:47] fs/namei.c just after the kernel lock and before the kernel unlock in line 1510 [02:48] wait, that isn't iget ... [02:49] igrab() [02:50] so both an igrab and an iput in both places? [02:50] no actually it's __iget, if that is available there ... [02:50] which isn't available *sigh* [02:51] okay, we take the igrab [02:51] lock_kernel(); [02:51] igrab(dentry->d_inode); [02:51] error = dir->i_op->unlink(dir, dentry); [02:51] if (!error) [02:51] vx_dq_unlink(dentry->d_inode); [02:52] iput(dentry->d_inode); [02:52] unlock_kernel(); [02:52] ok added. [02:52] and just to make sure .. we also do ... [02:53] include/linux/quotaops.h [02:53] line 95 [02:53] if (!inode) BUG(); [02:54] just before if (inode->i_nlink == 0) ? [02:54] yes [02:54] whats BUG(); do? [02:54] thowing an oops ... [02:55] +r [02:55] ok [02:55] erading to biuld again then? [02:55] ready to i mean. [02:55] yeah, but only make bzImage no make clean please ;) [02:55] no make dep either ? [02:57] not required ... we did not change the dependencies ... [02:57] ok. [02:57] just doing bzImage [02:57] repeat after me: "in kbuild we trust ;)" [02:57] heh hard for me to repeat. i dont trust anything. [02:58] ah forgot that, sorry ... 8-) [02:58] find all sorts of interesting things swept under the carpet that way. [02:59] Action: Bertl just checked hist carpet, but found nothing under it ... [02:59] im afraid of loking under mine. not teling whats growing under it. [03:00] ok new image ready. [03:00] hmm, growing? wasn't that the fridge? [03:01] fridge gets cleaned out more often. [03:01] blah forgot to rerun lilo [03:01] hmm, time to switch to grub ;) [03:03] grub is a bit too heavy for me. id rather have a bootloader similar to teh BSD loader or silo. [03:04] would be nice if silo ran on x86. would work just like lilo without the need to rerun lilo [03:05] ok [03:05] still get an oops [03:05] well, wait until you test some boot options with lilo ... [03:05] okay, let's hear [03:07] >>EIP; c0143a0a <===== [03:07] Trace; c01415a5 [03:07] Trace; c01421e2 [03:07] Trace; c0143c73 [03:07] Trace; c010727f [03:07] Code; c0143a0a [03:07] 00000000 <_EIP>: [03:07] Code; c0143a0a <===== [03:07] 0: 0f 0b ud2a <===== [03:07] Code; c0143a0c [03:07] 2: 5f pop %edi [03:07] Code; c0143a0d [03:07] 3: 00 80 6f 2e c0 e9 add %al,0xe9c02e6f(%eax) [03:07] Code; c0143a13 [03:07] 9: e6 fe out %al,$0xfe [03:07] Code; c0143a15 [03:07] b: ff (bad) [03:07]