1141430432 M * Bertl it compiled and booted here, which was quite expected :) 1141430456 M * Bertl the real testing is on the userspace part ... 1141430571 M * bonbons Yep :) 1141431065 Q * gerrit Ping timeout: 480 seconds 1141431768 M * bonbons Bertl: should the init process be started with or without STATE_SETUP flag on context? 1141431924 Q * lilalinux Remote host closed the connection 1141432196 J * lilalinux ~plasma@80.69.35.186 1141432382 Q * lilalinux Remote host closed the connection 1141432417 M * Bertl bonbons: doesn't matter, but if, then you have to migrate ot from within the setup 1141432488 M * bonbons Bertl: I tried with and without, process never gets PID=1... trying with the INFO_INIT and STATE_INIT flags 1141432531 M * bonbons ok, INFO_INIT does the trick 1141432683 M * Bertl yes, info init is required 1141432697 M * Bertl it is basically the 'virtualizer' 1141432712 M * Bertl (i.e. the mapping initpid <-> 1) 1141432855 M * bonbons Ok, now the guests starts up correct, just have to check the namespace (but that's unrelated) 1141433248 M * Bertl I'm so glad we decided to use this versioned interface for the syscall commands, it makes such changes really simple 1141433342 M * bonbons Yep, it's very easy! I'm just wondering how an init-clash is handled (migrate with VXM_SET_INIT to a context that already has an init) 1141433389 M * Bertl that is simple to answer 1141433431 M * Bertl vx_set_init() does: 1141433432 M * Bertl vxi->vx_initpid = p->tgid 1141433454 M * Bertl so the 'old' init process will suddenly get a different pid 1141433463 M * Bertl and the 'new' one will be 'init' 1141433465 M * bonbons so it just throws the old init out of it's seat :) 1141433475 M * Bertl for the reaper it's a little more difficult 1141433505 M * Bertl the task struct for the reaper is locked via a reference 1141433523 M * Bertl but this is no big deal as the reaper change also happens when the reaper dies 1141433857 M * bonbons Hollow: Here is the svn diff with addition of new wat to set guest's init: http://homepage.internet.lu/brunop/vserver/libvserver_setinit.diff 1141433896 M * Bertl bonbons: okay, so I take it that it works as expected and should be included in 2.1.1, right? 1141433986 M * bonbons I classify it as 'Works for me™' 1141434434 M * bonbons Bertl: Thanks for the new feature and the time spent on debugging "old way", I would say it can go to 2.1.1. maybe I try out util-vserver on it tomorrow 1141434566 M * Bertl okay, great! will include it ... 1141434578 M * bonbons now it's really time to get to bed! Good night 1141434679 Q * bonbons Quit: Leaving 1141434774 Q * matta Ping timeout: 480 seconds 1141435178 J * gerrit ~gerrit@c-24-22-19-208.hsd1.or.comcast.net 1141435624 Q * doener Quit: leaving 1141435661 Q * brc Quit: No windows for this server 1141436610 J * dizzle dizz@c-71-198-222-254.hsd1.ca.comcast.net 1141440973 M * Bertl okay, I'm off to bed .. back tomorrow! 1141440979 M * Bertl night everyone, cya! 1141440983 N * Bertl Bertl_zZ 1141441044 Q * gerrit Ping timeout: 480 seconds 1141441288 J * matta ~matta@71.224.125.126 1141441523 J * nortonisterrible ~nortonist@CPE-24-166-155-173.new.res.rr.com 1141441570 Q * nortonisterrible Read error: Connection reset by peer 1141441583 J * gerrit ~gerrit@c-67-160-146-170.hsd1.or.comcast.net 1141443194 Q * dizzle Quit: 1141444953 Q * ESL|caseme Quit: 1141445880 Q * wally Quit: Miranda IM! Smaller, Faster, Easier. http://miranda-im.org 1141445951 Q * matta Read error: Connection reset by peer 1141452107 Q * FireEgl Ping timeout: 480 seconds 1141452965 J * FireEgl Atlantica@Atlantica.Tcldrop.Org 1141458230 Q * Hollow Quit: SIGTERM 1141458361 J * Hollow ~hollow@home.xnull.de 1141460317 M * Hollow Bertl_zZ: the new migrate switch is awesome, thx! 1141464202 J * Viper0482 ~Viper0482@p54975370.dip.t-dialin.net 1141464857 J * bonbons ~bonbons@83.222.39.180 1141464868 J * dothebart ~willi@xdsl-81-173-175-206.netcologne.de 1141465077 M * Hollow bonbons: morning! 1141465092 M * bonbons morning Hollow! 1141465100 M * Hollow the new migrate switch works like a charm :) 1141465126 M * bonbons Yep, makes thing just very easy :) 1141465166 M * Hollow just the way i like it 1141465176 M * Hollow have commited the fixed start command + libvserver to svn 1141465182 M * dev what the migrate switch? :) 1141465196 M * dev where migrate? 1141465199 M * Hollow dev: new syscall version for the migrate syscall 1141465204 M * Hollow process migration 1141465212 M * dev and where does it migrate a process? 1141465224 M * Hollow from context 0 (host) to the guest context 1141465233 M * dev ahhh.... 1141465262 M * Hollow but there is special case uglyness for init, because it wants to be pid=1 ;) 1141465264 M * dev I see... what are the applications for this? 1141465274 M * Hollow well, enter the vps e.g. 1141465275 M * dev and how are resources reaccounted? 1141465289 Q * tudenbart Ping timeout: 480 seconds 1141465296 M * dev AFAIK, enter the vps was already implemented, no? 1141465304 M * Hollow for resource accounting you have to wait for bertl, no idea how it is done 1141465309 M * bonbons Hollow: guestt I will soon commit my big changes to 1.0.4 branch :) 1141465320 M * Hollow dev: we're talking about the new, from-scratch implementation for the utils 1141465346 M * Hollow and bertl implemented a new migrate syscall for that 1141465354 M * Hollow to set some flags etc 1141465373 M * Hollow bonbons: ok, go ahead :) 1141465388 M * dev is it possible to migrate 2 process from httpd? :))) and leave other httpd threads in host? :) 1141465398 M * Hollow sure 1141465429 M * dev and how about isolation? they require communication between themselfs 1141465465 M * Hollow well, as long as you don't close file descriptors it should work, the question rather is, why you would do that? 1141465493 M * Hollow you can easily break out of the chroot if you have an open fd to the host namesapce 1141465525 M * dev exactly! it is insecure and makes me believe the whole concept of containers/namespaces is used as a toy in this case. but maybe there some practical value I can't see, who knows 1141465559 M * Hollow well, normally you don't have open fds to the host 1141465590 M * bonbons Hollow: did you look at the \0 issue in vu_printf() ? 1141465600 M * Hollow no, not yet 1141465611 M * dev hmm... but you can have mmap'ed files. imagine that you migrated SUID application to the VPS. in this case VPS owner will be able to access and maybe even modify executable... 1141465614 M * Hollow bonbons: you have a simple test case probably? 1141465648 M * dev Hollow: have you started comparison with OpenVZ already? 1141465658 M * Hollow well, in theory the only program the should migrate is the vps start process which calls init in the end, any other use of migration is insecure 1141465707 M * Hollow dev: no, not yet sorry 1141465716 M * dev then, this operation should be called VPS init or create, not migrate :/ 1141465739 M * dev ok. feel free to ask me for the details when ready. I have a lot of features in mind :) 1141465749 M * Hollow well, you are still able to migrate.. and people do it nevertheless, for administrative purposes... how does vzctl enter handle this? 1141465770 M * bonbons "nidof | hexdump -C" -> 0x00 at the end (in 1.0.4 branch) 1141465830 M * dev it does migrate for years already :) but it actually I don't like migrate name, as nothing is migrated. Just context is changed. And vzctl does a lot of measures to make it in secure way, including creation of terminal. Because, having terminal from host is insecure. 1141465873 M * Hollow dev: yep, i just send a patch to the ML with the terminal proxy backported to util-vserver 1141465895 M * Hollow bonbons: but \0 in the end is ok, no? 1141465900 M * dev I wish you helped OpenVZ that way :)))) 1141465908 M * bonbons even a simple int main() {vuprintf("Hello World"); return 0; } should show it and output will be "Hello World\n\0" 1141465908 M * Hollow heh 1141465947 M * dev is there something you prefer more in vserver? I'd be glad to fix it :) 1141466097 M * Hollow bonbons: http://phpfi.com/105497 1141466134 M * bonbons Hollow: it's at end of each vu_printf(), I was previously mis-leeded because I had lots of printfs, and did not look so much at start/end of output 1141466167 M * Hollow so, what's the problem with \0 in the end? 1141466175 M * Hollow i need it to terminate the string, no? 1141466184 M * bonbons yep, your test-case shows it (the \n I added was wrong, bad println() habits!) 1141466220 M * Hollow ok, so wheres the error? ;) 1141466249 M * bonbons yep, but the terminating \0 should never be written to output, nor reported in the length of vu_snprintf() 1141466264 M * Hollow hm 1141466277 A * Hollow takes a look 1141466303 M * bonbons Hollow assume you do int main() { vu_printf("Hello"); vu_printf(" "); vu_printf("World\n"); return 0} 1141466321 M * bonbons output would be "Hello\0 \0World\n\0" 1141466385 M * Hollow indeed 1141466513 M * bonbons just decrementing the returned length in vsnprintf should do the job I think 1141466524 M * Hollow bonbons: in vu_vdprintf change "len = write(fd, buf, buflen + 1);" to "len = write(fd, buf, buflen);" 1141466866 M * bonbons yep that fixed it (my line above was wrong, is it too early in the morning?) 1141467301 M * Hollow bonbons: that also fixed the bug in vcd, thanks! 1141467650 M * bonbons my big changes got commited to 1.0.4, but still a few issues pending... 1141468081 M * Hollow seems like there are still many debug vflags -G's ;) 1141468305 M * bonbons yep, did not remove them from lib/vps.sh 1141470570 J * Smutje_ ~Smutje@xdsl-87-78-18-204.netcologne.de 1141470679 Q * Smutje Ping timeout: 480 seconds 1141470679 N * Smutje_ Smutje 1141470776 Q * mnemoc Quit: leaving 1141471240 M * bonbons Hollow: is there somewhere a 'testme' script to execute from INSIDE a guest, to test things like chroot breakout and others that vserver should be preventing? 1141471259 M * Hollow not that i know of 1141472214 M * daniel_hozac Hollow: oo, vlogin, nice. 1141472299 M * daniel_hozac hmm. why does it need the nid? 1141472423 M * Hollow because it nigrates to the network context 1141472485 M * daniel_hozac oh, i see. no giant chaining. 1141472527 M * Hollow indeed 1141472545 M * daniel_hozac i guess Enrico won't like that. :) 1141472555 M * Hollow yup, i don't care :) 1141472587 M * daniel_hozac well, no love from Enrico == no merging into util-vserver ;) 1141472593 M * Hollow maybe the nid thing has to be adapted to chbind because util-vserver doesn't use static nids, right? 1141472599 M * daniel_hozac it does. 1141472602 M * daniel_hozac since 0.30.210. 1141472612 M * daniel_hozac and nid is not necessarily equal to xid. 1141472613 M * Hollow ok, so than we can keep it the way it is 1141472616 M * Hollow hm 1141472624 M * Hollow bad thing! 1141472641 M * daniel_hozac i made it support .../ncontext for those wishing to use the same network context for several xids. 1141472662 M * daniel_hozac i can't see the use-cases for it, but it was so easy to add. 1141472684 M * Hollow yup, so the vserver.login stuff has to be fixed i guess 1141472720 M * Hollow and well... it doesn't work with chaining, because you have to migrate at very specific places during the whole pseudo creation 1141472902 M * daniel_hozac hmm, why? 1141472965 M * Hollow hm, after looking at the code again, it maybe working with chaining too 1141472967 M * daniel_hozac openpty ought to work from inside the chroot as well, no? 1141473009 M * Hollow yeah 1141473054 M * Hollow crappy daemon stuff 1141473069 M * Hollow does fork() inherit signals 1141473070 M * Hollow ? 1141473107 M * Hollow the man page only talkes about pending signals, not about signal handlers 1141473275 M * daniel_hozac i think so. 1141474053 J * mnemoc ~amery@user4-2.tutopia-dialup.ifxnw.cl 1141474095 M * bonbons Hollow: for vlogin, if target shell binary does not exist the fork goes to 100% CPU inside guest. I get that fixed by setting the signal handlers after fork, and letting the forked instance sleep(0) 1141474177 M * Hollow hm, the signls are set after fork.. 1141474251 M * bonbons Yep, but I guess the forked process terminates BEFORE the original gets time to set it's signal handlers 1141474279 M * Hollow something like that yeah... todays cpus are simply to fast.. 1141474280 M * Hollow ;) 1141474341 M * Hollow but can't we just set them before fork()? 1141474350 M * Hollow if they are inherited all should be fine 1141474457 M * daniel_hozac does the child need them at all? 1141474497 M * Hollow daniel_hozac: yeah, because of window size change signal 1141474506 M * Hollow but it probably only needs that one 1141474513 M * bonbons no, setting the before does not work 1141474529 M * daniel_hozac signal handlers get inherited over execve? 1141476194 M * bonbons Hollow: commited output cleanup, vlogin completition and reenabled vmount's fstab parsing (1.0.4 branch) 1141476886 M * Hollow ok, great! 1141477060 M * Hollow bonbons: btw, i added stacked configs a while back as you suggested, great idea! 1141477130 M * Hollow http://phpfi.com/104948 1141477975 M * derjohn Anyone experiences with openldap in a guest? 1141478049 M * daniel_hozac i run my slapd in a guest, and use {nss,pam}_ldap in three. 1141478088 M * derjohn sound good, so no capability stuff to care for? 1141478110 M * daniel_hozac no. 1141478122 M * derjohn i dare to ask a little off-topic: how to I migrate from ldbm to bdb backend? Is there a more convenient way than go via ldif? Is there a ready-made script for doing so? 1141478151 M * daniel_hozac slapcat | slapadd isn't really hard, is it? 1141478236 M * Hollow glibc manual sez: The child process inherits its mask of blocked signals and signal actions from the parent process. 1141478262 M * Hollow so, daniel_hozac, you were right 1141478325 A * derjohn looks how to tell the slapadd to be bdb suddenly 1141478454 M * daniel_hozac derjohn: -b? -f? 1141478610 M * derjohn daniel_hozac, yes ;) But i think I'll go via a file .. I have to change the slapd.conf anyway, so I wont save much time (hope the host will be down < 2 mins ....) 1141478653 M * daniel_hozac right, cp slapd.conf{,.new}; vi slapd.conf.new; slapcat | slapadd -f slapd.conf.new 1141478969 J * bubulak_ ~bubulak@cicka.wnet.sk 1141478970 Q * bubulak Read error: Connection reset by peer 1141479904 J * fwl ~fwl@83-215-237-1.seek.stat.salzburg-online.at 1141480078 J * doener ~doener@i5387C20F.versanet.de 1141482825 J * coocoon ~coocoon@p54A061B3.dip.t-dialin.net 1141482846 M * coocoon hello to all 1141483236 M * daniel_hozac hi 1141483240 Q * gerrit Ping timeout: 480 seconds 1141484514 Q * coocoon Ping timeout: 480 seconds 1141485193 J * coocoon ~coocoon@p54A05924.dip.t-dialin.net 1141486338 M * derjohn daniel_hozac, my ldap migration was successful. interestingly it said that openldap 2.3 fixes vaious issues, while it's sad (without i) that Debian has only a 2.2 at the time (but things seem to change ATM ...) 1141486362 M * derjohn daniel_hozac, do you run safepoints or special tuning parameters? 1141486387 M * daniel_hozac i'm not even sure what that means :) 1141486423 M * daniel_hozac i just hacked it together to share users among the vservers and clients. 1141486468 M * derjohn daniel_hozac, with bdb backend you can set things like cachesize and safepoints (when is the RAM flushed to disk etc.) in slapd.conf 1141486498 M * derjohn I expected you to have very geekish values in yout slapd.conf !!! :) 1141486502 M * derjohn *your 1141486537 M * daniel_hozac i've barely touched it :) 1141486577 M * daniel_hozac just set up the TLS stuff, but that's about it. 1141486596 M * derjohn maybe you are lucky and fedora has 2.3 by default (or the fedora directory at all???) 1141486620 M * daniel_hozac not FC4, unfortunately. 1141486627 M * daniel_hozac 2.2.29 1141486645 M * derjohn derjohn: read at least all 2.3.x changelogs to be convinced. 2.2. is out of date. 1141486657 M * daniel_hozac FC5 will have 2.3.19. 1141486783 M * daniel_hozac upgrading openldap from 2.2 to 2.3 post-release seems way too risky to be worth it. 1141486856 M * derjohn daniel_hozac, yes :) but you could try e.g. in a vserver guest if the migration works ;) 1141486888 M * daniel_hozac well, 2.2 WORKSFORME :) 1141486916 M * daniel_hozac (and lots of things depend on openldap) 1141488751 J * [-^q^-] ~DogSamura@gb.jb.102.172.revip.asianet.co.th 1141488956 Q * fwl Quit: Leaving 1141488961 N * Bertl_zZ Bertl 1141488975 M * Bertl greetings folks! 1141489010 M * coocoon hello bertl 1141489026 A * FaUl spams a bit: to all german people: please sign http://itc.napier.ac.uk/e-Petition/bundestag/view_petition.asp?PetitionID=60 1141489035 J * fwl ~fwl@83-215-237-1.seek.stat.salzburg-online.at 1141489570 J * rs ~rs@vol75-7-82-229-177-124.fbx.proxad.net 1141489582 Q * rs Quit: 1141489619 J * rs ~rs@vol75-7-82-229-177-124.fbx.proxad.net 1141489727 Q * Viper0482 Quit: bin raus, 1141489867 Q * [-^q^-] Quit: My Computer is Protected from Trojans, Worms and Bots for FREE by Swatit http://www.lockdowncorp.com/bots/downloadswatit.html 1141489937 Q * michal` Ping timeout: 480 seconds 1141490184 J * michal` ~michal@www.rsbac.org 1141490562 M * derjohn FaUl, the directive from EU level is sadly as goog as vaild 1141490566 M * derjohn *valid 1141490581 M * derjohn FaUl, want some info? then gimme your e-mail addy 1141490672 M * FaUl derjohn: well - i'm not so much into that political things 1141490698 M * derjohn alles scheisse (tm) 1141490705 M * Bertl okay, off for dinner, back shortly ... 1141490712 N * Bertl Bertl_oO 1141490713 M * FaUl but i guess there are some possibilitys for the local gouvernment to get around it 1141490748 M * FaUl as thats the usual tactic from politicans :-) 1141492238 J * bubulak ~bubulak@cicka.wnet.sk 1141492314 Q * bubulak_ Read error: Connection reset by peer 1141492350 J * Viper0482 ~Viper0482@p54975370.dip.t-dialin.net 1141492907 N * Bertl_oO Bertl 1141492911 M * Bertl back now ... 1141493143 M * phreak`` welcome Bertl! :) 1141493166 J * bubulak_ ~bubulak@cicka.wnet.sk 1141493167 Q * bubulak Read error: Connection reset by peer 1141493521 M * fwl hallo phreak, your username sounds familiar to me !? 1141493554 J * restill ~restill@c-24-11-171-10.hsd1.mi.comcast.net 1141493583 M * fwl but maybe a coincident 1141493682 M * fwl yes, just coincidently the same name as a collegue of mine. ;) (looked at your user data) 1141493929 M * Bertl hey phreak``! 1141494212 M * phreak`` heh, fwl no problem :) 1141494218 M * phreak`` hey Bertl :) 1141494617 M * bonbons Bertl: I get "vxW: !!! limit: f264804c[VM,9] = -27 on exit." or similar in kernel log when my guest exits, what does it mean? 1141494653 M * daniel_hozac bonbons: that there's an accounting problem somewhere. 1141494689 M * Bertl that means that your guest exits with -27 VM pages 1141494768 M * bonbons I had other values as well: "...[,20] = 3...", "...[VM,9] = -486..." 1141494786 M * Bertl bonbons: on which kernels/patches do you observe that? 1141494795 M * mnemoc join #dosemu 1141494798 M * mnemoc sorry 1141494803 M * Bertl *G* 1141494809 M * mnemoc :( 1141494821 M * bonbons I'm observing that on 2.6.16rc5-vs2.1.1rc10 + migrate patch 1141494835 M * Bertl arch? 1141494896 M * bonbons x86 (pentium-m) 1141494932 M * Bertl hum ... does testme.sh -L cause similar? 1141494988 M * Bertl bonbons: it happens with your tools, I presume? 1141495073 M * bonbons yep, not checked if I can rperoduce with util-vserver, though the tools are not used for terminating the guest, from inside I just do a reboot (REBOOT_KILL flag set) 1141495130 M * Bertl okay, would be great if you could have a look at the testme.sh -L cases, and maybe adjust some of them for your tools 1141495156 M * Bertl they are there to trigger such cases 1141495171 M * bonbons I think I know in what direction to look, vlogin has something to do with it: only happens after at least one vlogin connection to guest 1141495207 M * Bertl if we have a moderately simply way to reproduce it, we can easily nail it down in the kernel 1141495223 P * JimmyGulp Leaving 1141495244 P * glen_ 1141495282 M * bonbons Will try out on an older kernel, with util-vserver than vlogin, just to confirm 1141495866 M * bonbons Did not show up on that quite older box (2.6.15-vs2.1.0 + util-vserver 0.30.209-r1) 1141495961 M * bonbons on the box where it happens it's easy to reproduce, works every time 1141496112 M * bonbons Bertl, time to eat here, will do more testing when back (in about 40 min), but don't hesitate to already provide first guidelines 1141496198 J * lilalinux ~plasma@h1-gw.of.net-lab.net 1141496213 M * bonbons Bertl: vlogin's source is here: http://homepage.internet.lu/brunop/vserver/vlogin.c 1141496297 M * Bertl how do you trigger it? 1141496382 M * Bertl welcome lilalinux! 1141496721 Q * fwl Quit: Leaving 1141496723 P * lilalinux Leaving 1141496736 J * lilalinux ~plasma@h1-gw.of.net-lab.net 1141497053 M * Bertl ahh, another nap attack! back later ... 1141497060 N * Bertl Bertl_zZ 1141497098 J * liquid3649_ ~Viper0482@p54975370.dip.t-dialin.net 1141497375 Q * Viper0482 Ping timeout: 480 seconds 1141498924 M * bonbons Bertl: triggering is as easy as start guest, vlogin to it (cd $guestroot; vlogin -N -x $xid -n $xid), exit vlogin session, stop the guest 1141498984 M * bonbons vkill -x $xid ==> 2 lines "...[,20] = 3..." and "...[,21] = 3..." (constant result for that schenario it seems) 1141499721 M * bonbons Bertl_zZ: I can even reproduce it there with this basic sequence: vcontext -C -x 12 -f PERSISTANT; vlogin -x 12 (and exit session in guest); vflags -S -x 12 -f ~PERSISTANT 1141500631 M * bonbons Bertl_zZ: Tested on 2.6.16rc4-vs2.1.1-rc9, same happens there! vs2.1.0.4 and vs2.1.0.0 on 2.6.15 to not show up the error (same generic-x86 binaries [vcontext,vlogin,vflags] as on rc9) 1141500792 M * daniel_hozac bonbons: how many pages are (un|re)accounted for? 1141500835 M * bonbons depends on the case, always +3 on vkill, varying negative number on REBOOT_KILL 1141500864 M * daniel_hozac how about with the basic sequence? 1141500876 M * daniel_hozac cat /proc/virtual/12/limit before unsetting persistant. 1141500878 Q * derjohn Quit: by(t)e 1141501027 M * bonbons VM: current=-28, min=0, max=-6, soft=-1, hard=-1, hits=0 1141501112 M * daniel_hozac is that vlogin using the new migrate? 1141501131 M * bonbons 1 exec of vlogin reduces current by 27 (-28 => -55 => -82) 1141501164 M * bonbons I tried it with, and I'm now on 2.1.1rc9 kernel without 1141501271 M * daniel_hozac and VM is the only one with odd values? 1141501301 M * bonbons I uploaded the 3 binaries I use to test here (i386): http://homepage.internet.lu/brunop/vserver/vxW_limit.tar, for vlogin source, vlogin.c in same dir 1141501730 Q * liquid3649_ Quit: bin raus, 1141501733 M * bonbons The rest is nearly all to zero except for some max values (RSS[max]=437, Files[max]=21, PROC[max]=2), soft/hard is always -1 1141501742 J * Viper0482 ~Viper0482@p54975370.dip.t-dialin.net 1141501849 Q * restill Quit: Leaving 1141501945 M * bonbons daniel_hozac, will test a few older vs2.1.1rc kernels I have around to see if the apearance of it can be tracked down to a given version 1141502023 M * bonbons WM[current] limit also goes negative on 2.1.0.0, just no complain in kernel log 1141502178 M * daniel_hozac are you sure you have no extra patches or anything like that? 1141502272 M * bonbons For the 2.1.0 it's gentoo emerge, the 2.1.1rcs are vanilla kernel (2.6.16rc4/5)+"vanilla" vserver patches 1141502358 M * daniel_hozac i guess it has to be something that vlogin does that is so different. 1141502493 M * bonbons Yep, just don't know what... Except the pty-related code there is not so much special in there 1141502564 M * bonbons network context or namespace are also skipped in my test-case 1141502633 M * daniel_hozac and 2.6.15 doesn't show it? 1141502781 M * bonbons they all show the negative VM current number, only the recent vserver patches (tested rc9, rc10) complain in kernel log 1141502898 M * bonbons 2.6.15 + vs2.1.0/vs2.1.0.4 (installed from Gentoo) 1141502985 M * cehteh so! encrypted packet-writing to dvd .. nice backup :) 1141503006 M * cehteh (mobile backup, not for office) 1141503772 Q * coocoon Quit: KVIrc 3.2.0 'Realia' 1141504201 M * daniel_hozac bonbons: oh, really? 1141504280 M * bonbons Yep, I tested on 3 different boxes (2.6.16rc4-vs2.1.1rc9, 2.6.16rc4-vs2.1.1rc10; 2.6.15-gentoo-vs2.1.0-r1; 2.6.15-gentoo-vs2.1.0.4) 1141504303 M * daniel_hozac does it appear on stable at all? 1141504313 M * bonbons the two on 2.6.16 do the printk(), the two others just show the negative numbers in /proc/virtual//limit 1141504362 M * bonbons I have no stable vserver kernel around 1141504518 M * daniel_hozac i guess using vserver-utils is really out of the question. 1141504733 M * bonbons That's what I guess too, just tried a vlogin to a guest created by util-vserver, vlogin and exit decremented it's VM current count by 27 1141504850 M * bonbons entering increases it, leaving decrements more than entering incremented it 1141505048 M * daniel_hozac "Failed to execute default shell: No such file or directory" 1141505058 M * daniel_hozac but it still decremented by 28. 1141505123 M * bonbons you must execute vlogin from the root of the guest's filesystem. vlogin first migrates and chroots, then it sets up the pty and finally it tries to execv() 1141505380 M * bonbons daniel_hozac: doing a vexec just comes back to original values. Unless I miss something vlogin is doing a fork and pty setup in addition to what vexec is doing 1141505466 M * daniel_hozac hmm. 1141505474 M * daniel_hozac the values decrease by 28 pages here. 1141505485 M * daniel_hozac vlogin uses 29 pages VM. 1141505563 M * daniel_hozac could you try removing the signal() calls? 1141505600 M * bonbons I can try, all of them? 1141505613 M * daniel_hozac hmm, question, when should vlogin die? 1141505648 M * bonbons the master process should die just after the forked child dies 1141505665 M * daniel_hozac oh, the signal handler does it. 1141505709 J * restill ~restill@c-24-11-171-10.hsd1.mi.comcast.net 1141505776 M * daniel_hozac why does the signal handler install itself for SIGWINCH? 1141505812 M * daniel_hozac is the signal handler for SIGWINCH reset each time it's called? 1141505812 M * bonbons daniel_hozac, I don't know, that you must ask Hollow 1141505853 Q * restill Quit: 1141505936 M * bonbons I removed all signal(), except for SIGCHLD, not difference in VM page consumption 1141506331 N * Bertl_zZ Bertl 1141506338 M * Bertl much better ... 1141506360 M * bonbons Bertl, what's much better? 1141506378 M * Bertl my condition after the short nap :) 1141506397 M * bonbons ah, ok... you will need it ;) 1141506432 M * Bertl thought so ... 1141506453 M * Bertl so we are losing a few pages on vlogin exit? 1141506479 M * Bertl (or to be more precise, we miss a few pages on vlogin enter?) 1141506517 M * bonbons around 25-30 each time.... 1141506560 M * Bertl okay, and this happens independant of the context creation, I take it? 1141506566 M * bonbons count seems depending on the box 1141506576 M * Bertl i.e. doing the following should expose it: 1141506598 M * Bertl chcontext --xid 42 -- sleep 10 & 1141506615 M * Bertl vlogin ... 42 / exit 1141506619 M * bonbons exactly, just a vlogin to a util-vserver managed guest does the same as a vserver-utils created one; all vs2.1.** pached kernels I tried out have the issue 1141506648 M * Bertl okay, that sounds good then, let me get my conenctions up and running and check that 1141506666 M * Bertl could you upload an up-to-date vlogin for me (binary)? 1141506742 M * bonbons ther's one already for a couple of hours ;) http://homepage.internet.lu/brunop/vserver/vxW_limit.tar (also contains vcontext and vflags) 1141506753 M * Bertl okay, great! 1141506761 M * bonbons that version uses the "old" migrate call 1141506790 M * Bertl okay, not that it would matter 1141506804 M * bonbons and is compiled for generic x86 1141506823 M * Bertl which might matter :) 1141506871 M * Bertl you should start doing version numbers ... 1141506888 M * bonbons only the lib part may matter a bit (athlon-xp, but that's still quite generic) 1141506912 M * bonbons yep, will need to get the svn revisions into the binaries! 1141506969 M * bonbons just don't know how to do that... 1141507794 Q * samuel_ Ping timeout: 480 seconds 1141508169 M * Bertl daniel_hozac: ping? 1141508195 Q * lilalinux Remote host closed the connection 1141508334 J * tudenbart ~willi@xdsl-81-173-172-250.netcologne.de 1141508708 M * bonbons Bertl, already found something out? 1141508738 M * Bertl reading mm code .. but I guess I have an idea what happens 1141508779 Q * dothebart Ping timeout: 480 seconds 1141508780 M * daniel_hozac bonbons: $Revision$ doesn't work for svn? 1141508843 M * bonbons daniel_hozac, don't know... $log$ is said not supported in subversion's FAQ 1141508959 M * daniel_hozac $Rev$ seems to work here. 1141508997 M * bonbons daniel_hozac, fine, will try 1141509013 M * daniel_hozac you'll need to set keywords to Rev for the file too, i think. 1141509130 Q * nokoya Ping timeout: 480 seconds 1141509165 J * nokoya young@hi-230-82.tm.net.org.my 1141509900 M * daniel_hozac Bertl: so what's happening? 1141510003 M * Bertl http://vserver.13thfloor.at/Stuff/BUGHUNT/bonbons-0001/ 1141510104 M * Bertl I find this one very interesting: 1141510105 M * Bertl [ 38.772623] vxD: vx_add_cres[ -1,VM, 9]: 0 += -27 (8f987860) @mm/mmap.c:1955 1141510124 M * daniel_hozac indeed. 1141510140 M * Bertl and here again for the vlogin 1141510141 M * Bertl [ 50.011310] vxD: vx_add_cres[ 256,VM, 9]: 707 += -27 (8f987060) @mm/mmap.c:1955 1141510163 M * Bertl but! 1141510168 M * Bertl [ 49.947768] vxD: vx_add_cres[ -1,VM, 9]: 0 += 27 (00000000) @kernel/fork.c:966 1141510175 M * daniel_hozac so mm->total_vm is -27 at that point? 1141510199 M * daniel_hozac 27, i mean. 1141510215 M * Bertl yup, but it doesn't get accounted on migrate as it seems 1141510247 M * Bertl or to say it rpecisely, it is accounted to the wrong (-1) context 1141510275 M * Bertl which IMHO means that the mm is not in the context at this time 1141510305 M * Bertl (which sounds quite right as the mm was not moved in) 1141510308 M * Hollow pure chaos! 1141510320 M * Bertl lol 1141510352 M * Hollow i just came back from the airport after a 5 hour traffic jam to drive back because the flight was canceld 1141510359 M * Bertl daniel_hozac: so I wonder, shall we move the mm into the context too, or change the dup to _not_ account for the context? 1141510367 M * Hollow please, winter, go 1141510370 M * Hollow go away :) 1141510374 M * Bertl Hollow: second that! 1141510402 Q * michal` Ping timeout: 480 seconds 1141510449 M * daniel_hozac Bertl: i guess moving the mm into the context would be most appropriate, no? 1141510458 M * Bertl daniel_hozac: maybe we should just change the p->mm->mm_vx_info in fork.c:966 to the task vx_info? 1141510517 M * Bertl and the same for the cleanup_vm? 1141510562 M * Bertl bonbons: you vlogin uses clone() right? 1141510568 M * daniel_hozac no, fork(). 1141510584 M * Bertl ah, okay :) 1141510627 M * Bertl okay, let's try if that 'trick' works ... 1141510694 M * Hollow which issue are you currently investigating? 1141510704 M * daniel_hozac vlogin messes with the accounting :) 1141510704 M * Bertl http://vserver.13thfloor.at/Stuff/BUGHUNT/bonbons-0001 1141510748 M * Hollow ic.. i noticed that too 1141510758 M * Hollow didn't know vlogin is causing it.. 1141510770 M * daniel_hozac yep. vlogin is evil! 1141510771 M * daniel_hozac ;) 1141510787 M * Hollow migration is evil per se ;) 1141510810 M * Bertl Hollow: second that again! 1141510832 J * gerrit ~gerrit@c-67-160-146-170.hsd1.or.comcast.net 1141510840 M * Bertl welcome gerrit! 1141510898 M * Bertl looks good, at least fixes the issue with vlogin 1141510905 M * Bertl now testing a few other cases 1141511205 M * Bertl jup, to speak with bonbons' words: works for me(tm) 1141511383 M * Bertl http://vserver.13thfloor.at/Experimental/delta-vmfork-fix01.diff 1141511398 M * bonbons Bertl: fine, let's see how it behaves on my box :), and hope there are no other such accounting issues 1141511481 M * Bertl bonbons: yeah, but if you find them, we'll fix em' :) 1141511586 M * bonbons will keep an eye on those fines under /proc/virt*// :) 1141511604 M * Bertl yup, not all of them can be correct at exit 1141511604 M * bonbons s/fines/files/ 1141511631 M * Bertl some accounting is corrected _after_ the context exits (e.g. sockets) 1141511658 Q * shedi Quit: Leaving 1141511662 M * Bertl but I'm working on some improved debug messages to monitor those too 1141511682 M * Bertl bonbons: in any case, negative values _are_ a good sign for a bug 1141511716 M * bonbons that's what I understood, except for -1 which can mean infinite ;) 1141511732 M * Bertl only for the limits, not for the accounted values 1141511747 M * Bertl i.e. accounted, min/max always should be positive 1141511771 A * Bertl .o( maybe we should add some logging for negative values there too ) 1141511807 M * Bertl daniel_hozac: what do you think about splitting the limit in acc and limit (debug side)? 1141511831 M * Bertl so that we can enable accounting bits independant from the limit checks? 1141511914 J * shedi ~siggi@inferno.lhi.is 1141511948 M * Bertl wb shedi! 1141511960 M * shedi hello Bertl 1141512512 M * daniel_hozac Bertl: makes sense. 1141513444 Q * lonewolff Quit: Reconnecting 1141513451 J * lonewolff lonewolff@adleman.lonewolff.info 1141513485 M * bonbons Bertl: the fix works for me too :) 1141513557 M * Bertl good! 1141513637 Q * lonewolff Quit: 1141513644 Q * malaiwah Quit: No windows for this server 1141513651 J * lonewolff lonewolff@adleman.lonewolff.info 1141513775 Q * lonewolff Quit: 1141513836 J * lonewolff lonewolff@adleman.lonewolff.info 1141514212 M * gerrit reHi Bertl, btw 1141514503 Q * Viper0482 Quit: bin raus, 1141514666 J * Smutje_ ~Smutje@xdsl-84-44-185-91.netcologne.de 1141514774 Q * Smutje Ping timeout: 480 seconds 1141514774 N * Smutje_ Smutje