[00:02] uhm [00:02] which is easier for you? [00:02] i think fixed ml only right now [00:02] too much other stuff to test together [00:02] okay, agreed ... [00:03] i mean, it looks like the dl should work fine as that's minimal changes [00:03] but ml/O1 plus the new tools need to be tested together [00:05] I like split patches [00:06] fs/fs.o(.text+0x45cf2): undefined reference to `IS_IMMUTABLE' [00:06] arse [00:06] well they ease rediffing a _lot_ ... [00:07] hmm, strange, how should this happen, unless you took very old splitups? [00:07] ahh, no probably xfs/jfs ... [00:08] It's fs/nfs/dir.c [00:08] wow, they have immutable files? [00:08] no, the kernel nfs server actually honours the immutable bit on the -ac kernel ;-) [00:08] anyone tested jfs under the later vserver patches? [00:09] not yet, but I think this should be the future fs for vserver ... (or was it the fs of the future ;) [00:09] I'll use it as soon as they can do stable online resizing [00:10] hmm, you cant do that with ext3 either, at least not without patches, or am I wrong? [00:10] i'll have to test that [00:10] i'm also adding another disk to a server and am pondering jfs [00:11] yes, I tried the patches and they're not stable [00:11] -t reiserfs -o notail for me for now [00:11] yay, my kernel linked [00:11] jfs in general isn't stable? [00:11] congrats ... [00:11] I don't use unifying or anything [00:12] jfs seems pretty stable to me ... [00:12] It's the online filesystem resizing, I couldn't find the userland JFS util [00:12] i don't have to resize... hrm, i'll have to test this [00:12] Bertl: you use it? [00:13] All my servers have a small root partition, some swap and the rest is LVM. Means that decisions on partition sizes need not be final [00:13] well, I used to use it for some time, about a year ago, then switched to ext3 because of the compatibility with ext2, but as the ext3 failures/silent degradation continues ... [00:14] Hmm, I'm wondering if I shouldn't give this new kernel a whorl. Hmm, where's my cracked copy of VMware ... [00:15] argh, miles away. Looks like it's a reboot offence [00:15] you use unlicensed software? [00:15] mugwump: uml :) [00:15] I prefer qemu at the moment ... [00:16] neptune:/usr/src# apt-get install qemu [00:16] Reading Package Lists... Done [00:16] Building Dependency Tree... Done [00:16] E: Couldn't find package qemu [00:16] neptune:/usr/src# [00:17] qemu sucks, it isn't even in Debian [00:17] gah [00:17] no variable hertz patch for 2.4.23 [00:18] ah, 2.4.233 patch works... it's the c17g patch that conflicts with it [00:19] just ignore me, simple patch to fix :) [00:21] matta: heh [00:24] Got a quick formula for running qemu Herb? [00:24] I've got it compiled, do I just set it loose on the bzImage ? [00:25] yeah, you just grab the patches from my experimental ... [00:25] apply the 2.4 version ... select QEMU, recompile ... [00:25] and you can simply start it with the vl tool ... [00:26] hmm, they are under stuff, sorry ... [00:26] http://vserver.13thfloor.at/Stuff/qemu-enable-2.4.x.diff [00:27] where do I select QEMU ? [00:27] Ah, CONFIG_QEMU [00:27] you do make oldconfig ... [00:28] apparently jfs doesn't have quota support.. [00:28] I have quota support, it's called resizable LVM partitions :-) [00:29] yeah, but this makes your ILI stuff unuseable ;) [00:29] hmm, jfs no quota, hows that? [00:30] alekibango (~john@b59.brno.mistral.cz) joined #vserver. [00:30] I only use ILI over the root filesystem (ie, /usr, /sbin, /lib, etc - where all the libraries & executables are), achieved through bind mounts [00:31] Bertl: that's what it says on the website [00:31] and this is in the jfs code: [00:31] super.c: /* Silently ignore the quota options */ [00:31] super.c: } else if (!strcmp(this_char, "grpquota") [00:31] super.c: || !strcmp(this_char, "noquota") [00:31] super.c: || !strcmp(this_char, "quota") [00:31] super.c: || !strcmp(this_char, "usrquota")) [00:32] Bertl: i'm going to try everything minus ml right now [00:32] well, ext2/ext3 does this too .. if you look at the code ... [00:32] oh [00:32] those are userspace options ... [00:32] JFS TODO list: [00:32] Plans for our near term development items [00:32] - increase the FS performance [00:32] - add quota support [00:32] from their website [00:32] says the same in Documentation/filesystems/jfs.txt [00:32] that _is_ a hint ... [00:32] maybe they never updated it [00:33] IIRC, I never tried to use quota on jfs ... so it is possible ... [00:34] In file included from /usr/src/linux-2.4.23pre7-home/include/linux/mm.h:4, [00:34] from /usr/src/linux-2.4.23pre7-home/include/linux/slab.h:14, [00:34] from /usr/src/linux-2.4.23pre7-home/include/linux/proc_fs.h:5, [00:34] from init/main.c:15: [00:34] usr/src/linux-2.4.23pre7-home/include/linux/sched.h:134: warning: type defaults to `int' in declaration of `structtask_structtask_t' [00:34] sounds bad, what patches atm? [00:34] hrm, maybe that is from adding variable hz [00:35] oh no, my kernel just crashes in qemu [00:35] i'm gonna try building just c17g2-O1 [00:35] doesn't even say "Linux version ..." [00:35] then add disk stuff... [00:35] then the extras [00:36] I think I'm going to try it on the bare metal. BRB. [00:36] mugwump (~sv@62.253.119.28) left irc: Quit: leaving [00:37] fails with same error with just patch-2.4.23-pre7-O1c17g2.diff [00:37] lots of errors... couple pages [00:39] serving (~serving@213.186.191.62) joined #vserver. [00:39] mugwump (~sv@62.253.119.28) joined #vserver. [00:39] Hmm, well qemu failed miserably to predict that this kernel would boot successfully :-) [00:40] well, good for you, probably you have hardware defined, which qemu can't handle ... [00:41] That'd do it, good point [00:41] I can give you a config file, which has a good chance to work ... if you want ... [00:41] ooo goody [00:42] http://vserver.13thfloor.at/Stuff/QEMU_c17f.config [00:42] RH (~john877@24.171.20.14) left irc: Ping timeout: 513 seconds [00:43] cheers [00:44] heh, util-vserver-0.23.6 is trying to call setup() to chcontext [00:45] mugwump: i think you need to use 23-90 [00:45] er, 23.90 [00:46] IIRC, there should be a 92'er version on savanna ... [00:46] does it work though? :) [00:48] Bertl: does patch-2.4.23-pre7-O1c17g2.diff compile for you? [00:50] haven't tried yet ... let me see ... [00:50] but good news, the VM stuff seems to account VM exacter in 2.4.23-pre7 [00:56] really? [00:56] that's odd. [00:56] i wonder what changed... [00:57] they added some checks for copied/used pages ... [00:58] ok, got the new version, and what do you know ... it seems to work [00:58] Now to test the syscall :-) [00:58] array.c:353:2: #endif without #if [00:59] that is what I get with 2.4.23-pre7-O1c17g2 [01:01] argh ... okay I know what happened ... will take a few minutes ... [01:02] mhepp (~mhepp@r72s22p13.home.nbox.cz) left irc: Remote host closed the connection [01:04] Oh no, what's my syscall number :-) [01:04] Oh, wait, it's 273 innit [01:04] Action: Bertl smiles ... [01:08] serving (~serving@213.186.191.62) left irc: Ping timeout: 492 seconds [01:12] Is sysctl() how you call a generic system call in C ? [01:13] hmm, nope have a look at enricos tools ... [01:16] is it an ioctl? [01:16] basicall you define it like this ... [01:16] _syscall2(int, pivot_root, const char *, new_root, const char *, put_old); [01:16] then you call it like this .. [01:16] r = pivot_root(new_root, put_old); [01:17] mugwump: are you making a syscall for scheduler tunables? [01:19] I'm just trying to test it from Perl :-) [01:19] yes he is ... [01:19] eek perl? [01:20] Sure, why not [01:20] well every server I installed perl on, ended up as CPAN archive sooner or later ... [01:21] ... and the problem with that is ? [01:21] well I want a server not a CPAN archive ... [01:22] serving what though? [01:23] I tend to only install CPAN packages through rpm/dpkg, keeps the servers cleaner. [01:24] did you know you could rm -rf /root/.cpan/* ? [01:24] :) [01:24] http://vserver.13thfloor.at/Experimental/patch-2.4.23-pre7-O1c17g2.diff.bz2 [01:24] okay please reload the patch ... [01:24] k [01:25] I hate CPAN.pm, it offends the SysAdmin in me [01:25] same error... [01:26] lemme delete the tree again [01:26] all I did was take stock 2.4.22, add 2.4.23pre7, and then 2.4.23-pre7-O1c17g2 [01:28] if the issue remains, I'll need your .config file ... [01:30] http://66.103.140.20/config-broken [01:30] Damn, I'm getting -ENOSYS [01:31] cc -D__KERNEL__ -I/usr/src/linux-2.4.23pre7-home/include -Wall -Wstrict-prototypes -Wno-trigraphs -O2 -fno-strict-aliasing -fno-common -fomit-frame-pointer -pipe -mpreferred-stack-boundary=2 -march=i686 -DKBUILD_BASENAME=main -c -o init/main.o init/main.c [01:31] In file included from /usr/src/linux-2.4.23pre7-home/include/linux/mm.h:4, [01:31] from /usr/src/linux-2.4.23pre7-home/include/linux/slab.h:14, [01:31] from /usr/src/linux-2.4.23pre7-home/include/linux/proc_fs.h:5, [01:31] from init/main.c:15: [01:31] /usr/src/linux-2.4.23pre7-home/include/linux/sched.h:134: warning: type defaults to `int' in declaration of `structtask_structtask_t' [01:31] couple more pages of stuff like that... [01:32] everything applied clean [01:32] i must go for a little bit, will you be around? [01:32] if not just post patches and I will try them.. [01:35] serving (~serving@213.186.191.62) joined #vserver. [01:35] HI all [01:36] shuri: are you here ? [01:36] hi serving! [01:37] Hi Bertl :) wassup ? [01:37] @sam did you add your command to the syscall switch? [01:38] @matt I'm looking into it ... [01:38] @serving nothing new ... [01:38] ... in the last 4 minutes ;) [01:40] bertl: yes, I just can't figure out how to call it :-} [01:40] @matt what platform is this ... [01:40] I can call the VERSION ioctl fine, but even the new_s_context call is returning -EFAULT [01:41] @sam well I gave an example, the exact version is in enricos tools ... [01:41] @sam the version syscall (not ioctl) returns what? [01:41] 65536 [01:41] that means the switch is working ... [01:42] let me have a look at your virtual.[hc] please ... [01:43] I've updated the patches on my site, but I'll copy up the two files there too [01:43] http://vilain.net/linux/ctx/virtual.[hc] [01:45] okay I assume your vc_set_sched() doesn't return ENOSYS, right? [01:45] That's what it's returning, which means that the input isn't right [01:45] oops, EFAULT it was ... [01:46] or what exactly do you get? [01:46] no, new_s_context was returning -EFAULT. It's certainly my problem with getting Perl to do a syscall [01:46] maybe your arguments are wrong for new_s_context ... [01:47] if the syscall switch fails ... you get -EINVAL or -ENOSYS ... [01:47] -EINVAL if syscall is reached but command not present ... -ENOSYS otherwise ... [01:47] I can see the difference, eg if I set the wrong version number [01:48] ooh wait [01:48] shmget(1057030145, 13, IPC_NOWAIT|0x814e088|0200upeek: ptrace(PTRACE_PEEKUSER, ... ): No such process [01:48] :-> [01:48] that is normal for new_s_context ... [01:49] Which is good, it means that my syscall works in Perl [01:49] I mean, it means I have figured out how to do a new_s_context in Perl [01:49] seems so ... [01:49] JonB (~jon@129.142.112.33) left irc: Quit: zzzzzzz [01:50] neptune:/home/sv# grep ctxsched /proc/self/status [01:50] ctxsched: 808/2000 = (FR: 2 per 16; ago 2) [01:50] neptune:/home/sv# [01:51] my ($context) = `cat /proc/self/status` =~ m{^s_context: (\d+)}m; [01:51] my $data = pack("I*", $context, 0, 2, 16, 500, 2000 ); [01:51] print "context is $context\n"; [01:51] syscall (273, 1023475713, $context, $data) or die $!; [01:51] It dies with "Illegal Seek" (errno 0!) [01:51] I would too ;) [01:57] I'm returning the wrong values I think [01:58] @matt when do you get those errors? [02:05] Do you think it's bad style for a syscall to return 0 on success? [02:06] nope, zero is okay as long as it's -0 ... [02:06] -0 ? Normal 0 not good enough ? :) [02:07] Gotta keep those non twos-complement integer format machines in mind I guess [02:07] well you know ERROR: SUCCESS ... ;) [02:07] My syscall wasn't failing, my perl fragment should have tested for <0 rather than just "or die $!" [02:16] hi [02:16] Bertl: really? [02:16] gcc version 3.2 20020903 (Red Hat Linux 8.0 3.2-7) [02:16] no problem ... everything works fine .. I get a dozen warnings about drm, but that's it ... [02:17] weird [02:17] 2.95 of course ... on my side ... [02:17] anyway, I'll check with the diff again ... [02:17] make bzImage ? [02:18] init/main.c: At top level: [02:18] init/main.c:431: conflicting types for `child_reaper' [02:18] init/main.c:431: invalid use of undefined type `union task_union' [02:18] init/main.c: In function `init': [02:18] init/main.c:569: dereferencing pointer to incomplete type [02:18] make: *** [init/main.o] Error 1 [02:18] make bzImage modules .. [02:18] hmmm [02:18] what platform is this? [02:18] i'll try on a diff server [02:18] redhat 8 [02:18] i can try on 7.3 [02:18] nope i386/ppc ... [02:18] i386 [02:19] okay check on another server ... [02:22] hmm, wait ... just a moment ... [02:24] argh! [02:24] ? [02:24] okay, my fault ... should be fixed in 20 seconds ... [02:24] yes, same error on different server :) [02:24] I have a script to cleanup patches ... and I changed the sed version ... but didn't check the script ... [02:25] so i never got the new one? [02:26] http://vserver.13thfloor.at/Experimental/patch-2.4.23-pre7-O1c17g2.diff.bz2 [02:26] oh yes, you got the new one, but with whitespaces removed at arbitrary positions :( [02:26] oh [02:29] that works :) [02:29] ml is almost ready ... just some checks ... [02:30] gotcha [02:40] okay final testrun ... then we have some patches ... [02:41] i'm testing compile with other patches now [02:41] okay ... [02:52] NZ|Egor (~sgarner@apollo.quattro.net.nz) left irc: Quit: so long, and thanks for all the fish [02:53] Simon (~sgarner@apollo.quattro.net.nz) joined #vserver. [02:53] well that boots and start vservers [02:53] ctxsched: 4938/5000 = (FR: 1 per 4; ago 11) [02:53] hi simon, new name? [02:53] that that from a higher hz ? [02:53] before it was 1000 [02:53] probably ... [02:57] http://vserver.13thfloor.at/Experimental/patch-2.4.23-pre7-O1-c17g2-ml0.04.diff.bz2 [02:58] hope the change in vm accounting doesn't affect the usability ... [03:00] Gah, I'm such a muppet ... foo &= bit doesn't set a bit ... [03:03] only in languages where & is a unary operator ;) [03:04] mugwump: do you know why the max tokens would go from 1000 to 5000? HZ? [03:04] hz went from 100 to 500 [03:04] i guess that makes sense [03:04] Yeah, the defaults are defined in terms of HZ [03:11] heh, had to rediff fakemem [03:13] oh my god? [03:14] where is that patch by the way, maybe we include it .. [03:14] i'm testing now then i'll make a diff [03:20] you said pass "rootflags=tagctx" for quota hash on root ? [03:20] yes, you still have to add it ... [03:30] seems to work as before [03:30] i wouldn't say accounting is any better or worse than before [03:30] limits definitely work [03:30] reporting seems accurate [03:32] hey folks ... http://vserver.13thfloor.at/Experimental/patch-2.4.23-pre7-c17g2-pi0.01.diff [03:33] pi ? [03:33] had this on the todo list for some time ... [03:33] pivot root? [03:33] no, proc info ... [03:33] oh? [03:33] grr [03:34] that's awsome [03:34] will that work against the O1 patch ? [03:34] well it's not really useful yet ... but it's a start to work on ... [03:34] oh [03:34] i'll look at the diff [03:34] we can adapt it to O(1) and we could move the scheduler info to that proc place ... [03:35] actually it applies cleanly to O(1) 8-) [03:36] vm accounting is a little worse if anything [03:36] total used free shared buffers cached [03:36] Mem: 262144 178196 83948 0 0 42468 [03:36] mugwump (~sv@62.253.119.28) left irc: Quit: Must ... sleep ... [03:36] that's after sendmail, proftpd, apache, cron, anacron, atd [03:37] hmm, looks good ... [03:37] yeah [03:37] it's on target [03:38] kill apache and it takes it down to 46MB [03:38] what I did is on my default build I change httpd.conf to only start 3 servers by default [03:38] sure, the server threads count ... [03:38] save people memory [03:39] we have to get the RSS limits working .. this is much more useful for a customer limitation ... [03:39] yeah [03:39] I'm thinking of a grant and limit value ... [03:40] this way we can have both, a guaranteed RSS and a maximum ... [03:41] below the guarantee (if done properly) the vserver won't swap out, above the maximum, it will fail ... [03:42] yeah [03:43] i just have to verify disk limits, but I think everything works. [03:43] and of course scheduling is better than before [03:47] I would be happy if you could write down a short howto regarding the scheduler, pointing out what the advantage is, and how you tested ... [03:48] oh yeah, i did promise you some docs... [03:48] maybe we can make the O(1) default ... [03:49] that'd be a good idea, it's extremely well suited for a vserver envrionment [03:49] the preemtion stuff isn't in yet, correct? [03:50] nope [03:50] 1 or 2 failures during patching [03:50] everything looks fine [03:50] okay ... I'll have a look at it ... [03:50] i'm gonna do a few more different tests [03:50] but I think I will upgrade a server to this tomorrow evening [03:50] write them down, if it makes sense ... [03:50] as it's going down for memory/hard disk upgrades anyway [03:51] did you test on a SMP system yet? [03:56] no. [03:56] i don't have a test SMP server [03:56] just because scheduler stuff often fires back on SMP ... [03:56] hrm... [03:57] i'm not really sure how to test [03:57] i don't have a test SMP machine [03:57] do you? [03:57] that is the problem, vmware/qemu/etc isn't an option ... [03:58] bochs has some basic SMP support ... but that is dog slow ... [04:00] everything seems functional like before [04:00] ml accounts, reports, and limits properly [04:00] dl accounts, reports, and limits properly [04:02] I would recommend to have a non-O(1) kernel compiled when you do the upgrade ... [04:03] well, i'll my old kernel [04:03] server needs more ram so that's getting done no matter what [04:03] if the scheduler fails on SMP, it will do loudly ... [04:17] alekibango (~john@b59.brno.mistral.cz) left irc: Read error: Connection reset by peer [04:19] hrm [04:19] we need to ensc to add CAP_QUOTACTL support to his tools [04:19] now that it's in c17g [04:29] we will ask him to add the other capabilities too ... [04:30] IIRC some CAPS are not available with the tools ... [04:30] preempt is non trivial with O(1) scheduler ... [04:31] not a big deal [04:31] pre-empt is nice, but more for realtime/desktop use anyway [04:32] it gave a good deal of responsiveness on vserver tough ... [04:34] this is the patch list/order I am using in case you care :) [04:34] patch-2.4.23-pre7 [04:34] patch-2.4.22-1030-vh [04:34] patch-2.4.22-1030-vh-O1c17g2-fix.diff [04:34] patch-2.4.23-pre7-O1c17g2.diff [04:34] patch-2.4.22-c17e-mq0.11.diff [04:34] patch-2.4.23-pre7-O1-c17g2-mq0.11-cx0.06.diff [04:34] patch-2.4.22-c17e-mq0.11-cx0.06-cq0.11.diff [04:34] patch-2.4.22-c17e-mq0.11-cx0.06-cq0.11-dl0.05.diff [04:34] patch-2.4.22-ctx17a-vr0.13.diff [04:34] patch-2.4.23-pre7-O1-c17g2-ml0.04.diff [04:34] patch-2.4.23-pre7-O1-c17g2-ml0.04-fakemem.diff [04:34] patch-c17e-signal-ctx1.diff [04:34] no-proc-mounts.diff [04:35] hmm, vh == variable herz ... [04:35] yes [04:35] hertz of course ... [04:35] what does the patch-2.4.22-1030-vh-O1c17g2-fix.diff do? [04:36] makes c17g2 apply cleanly [04:36] root@vps2 testing]# cat patch-2.4.22-1030-vh-O1c17g2-fix.diff [04:36] --- linux-2.4.23pre7.orig/kernel/signal.c 2003-10-19 20:08:57.000000000 -0400 [04:36] +++ linux-2.4.23pre7/kernel/signal.c 2003-10-19 20:09:25.000000000 -0400 [04:36] @@ -13,8 +13,9 @@ [04:36] #include [04:36] #include [04:36] #include [04:36] -#include [04:36] + [04:36] #include [04:36] +#include [04:36] nothing special.. [04:36] ahh okay ... [04:37] you should add this to the vh, or apply it later, then ... [04:37] looks good, where can I get patch-2.4.23-pre7-O1-c17g2-ml0.04-fakemem.diff ? [04:38] http://66.103.140.20/patch-2.4.23-pre7-O1-c17g2-ml0.04-fakemem.diff [04:40] and did you get a 'Could not get ctx_si structure' yet? [04:41] hrm.. nope! [04:41] hmm anyway, you can get races with this code ... [04:41] how so ? [04:42] s_info is released when the last 'user' calls sys_release_... [04:42] i don't know of a better way [04:42] while you check ctx_si (twice) for being not null, you do not hold it or lock ... [04:43] the window for a race is small ... let me see how I do this, by the way ... [04:43]