[00:13:51] --- dragos.tatulea has become available [00:13:55] gm [01:14:26] --- Brandon Allbery has left [01:15:40] --- Brandon Allbery has become available [01:20:34] --- Brandon Allbery has left [01:22:47] --- Brandon Allbery has become available [02:32:34] --- dragos.tatulea has left [02:34:21] --- dragos.tatulea has become available [02:35:39] --- dragos.tatulea has left [02:53:18] --- Simon Wilkinson has become available [03:04:12] morning [03:15:23] --- Simon Wilkinson has left [03:32:07] --- haba has become available [03:36:47] --- Simon Wilkinson has become available [03:39:08] good morning everyone, hope you have less of a cold than i have *sneeze*, brain filled with cotton. [03:48:59] --- Simon Wilkinson has left [04:17:53] --- dragos.tatulea has become available [04:21:31] --- dragos.tatulea has left [04:55:31] --- Brandon Allbery has left [04:57:23] --- Brandon Allbery has become available [05:45:28] --- Simon Wilkinson has become available [06:06:01] --- matt has become available [06:10:28] --- Simon Wilkinson has left [06:35:14] --- Simon Wilkinson has become available [06:44:51] --- Brandon Allbery has left [06:48:55] --- Brandon Allbery has become available [06:49:11] --- Brandon Allbery has left [06:54:55] --- Simon Wilkinson has left [07:26:50] --- dragos.tatulea has become available [07:27:25] --- Brandon Allbery has become available [07:32:42] shadow@gmail.com/owlE16C8B4F: Do you remember the EINVAL bug when doing "cat /afs/andrew.cmu.edu/usr/shadow/* > dev/null"? [07:40:37] --- Simon Wilkinson has become available [07:41:13] hey Simon [07:45:56] --- Brandon Allbery has left [07:47:40] --- Brandon Allbery has become available [07:53:11] --- Brandon Allbery has left [07:57:39] --- Brandon Allbery has become available [08:12:18] my client doesn't do beeps. i'm just shadow [08:12:26] > Do you remember the EINVAL bug when doing > "cat /afs/andrew.cmu.edu/usr/shadow/* > dev/null"? [08:12:27] yes [08:17:54] I'm trying to reproduce it now...but, again, cat'ing the dir takes forever (~30 minutes or more) [08:18:07] And the contents don't seem to be saved in the cache. [08:18:44] Get a network trace, see if you can see what it's doing, and what's taking so long. [08:19:00] Well, I'm seeing a steady flow of packets. [08:19:11] Providing you've fs setcrypt off, wireshark does a pretty good job of decoding AFS packet traces. [08:19:18] Yes, but the interesting thing is what those packets are. [08:19:22] Are they retransmits? [08:19:30] Are we bulkstating when we should? [08:19:31] you want to interrupt during cat. that's the point [08:19:40] no [08:19:46] and we are never bulkstating [08:19:55] which we should fix [08:20:04] i have notes from zephyr about that [08:20:10] Are we never bulkstating? My n810 test was seeing bulkstats? [08:20:42] >you want to interrupt during cat. that's the point [08:20:45] Why? [08:21:39] oh, i have a mac. no bulkstat [08:21:45] sorry [08:21:50] /interrupt/disconnect [08:23:42] Yes. [08:24:09] Derrick's problem was that he fs discon'd during a cat, and got some locks being held when they shouldn't have been. [08:25:01] oh yeah [08:25:50] I forgot [08:27:24] --- Simon Wilkinson has left [08:27:57] Well....it doesn't happen with the latest 1_5_x. [08:29:31] well, it didn't happen reliably for me either [08:29:40] but it happened "sometimes" [08:29:46] i will need to test more, soon [08:30:11] let me try doing some extras [08:44:03] --- Simon Wilkinson has become available [08:44:11] --- Brandon Allbery has left [08:46:16] --- Brandon Allbery has become available [09:01:01] --- Brandon Allbery has left [09:01:23] --- Simon Wilkinson has left [09:02:52] --- Brandon Allbery has become available [09:04:52] --- summatusmentis has become available [09:10:00] Does it seem feasible to add a GetVCache function? I am looking around the code now to verify how doable this is. [09:10:14] --- summatusmentis has left [09:19:07] --- summatusmentis has become available [09:30:07] --- Brandon Allbery has left [09:32:02] --- Brandon Allbery has become available [09:33:51] --- summatusmentis has left [09:44:19] --- Brandon Allbery has left [09:45:11] --- Brandon Allbery has become available [09:51:04] --- Brandon Allbery has left [09:52:04] add a GetVCache function?? [09:57:18] What configure options for irix build? [09:59:41] --enable-transarc-paths [10:00:34] PATH=/usr/bin/X11:/usr/bin:/bin:/usr/sbin:/sbin ./configure --enable-transarc-paths we will see. [10:01:25] make... Ok, it complains that we don't have a correct license installed any more -> it compiles. [10:03:08] (The MIPSpro C compiles is "complainware" but it continues to work) [10:04:33] Nikke (from Umeå) said yesterday he made an AIX 6 partition for you. [10:20:30] UX:make does not like something in the Makefile [10:21:10] make dest [10:21:28] he did. i need to build it, and to ask to keep my account [10:21:39] oh, use gmake? [10:21:46] jost [10:21:52] just typed gmake :) [10:21:57] we used some variable which pisses off irix make [10:22:14] it's the pickiest damn thing around. just like their cc [10:31:51] I am able to test the irix-client when I can reboot the machine, but not really this week. I doubt I can test the irix-server. [10:32:34] I wonder who still has Irix around. Ours will live one more year because of DMF. [10:33:43] ncac.gwu.edu or somesuch [10:33:53] also i bet nrl does too [10:34:10] Then ours will RIP. What about End-Of-Life of OpenAFS for IRIX? [10:35:43] (I'm every time astonished that it still works. Last time was 2007-03-09 (1.4.3)) [10:44:56] We have now the 1.4.8-almostpre3 client on centos 5 in production, since 2008-10-29, so I'd say it kinda works. [10:45:53] --- matt has left [10:49:34] --- Simon Wilkinson has become available [10:51:41] --- matt has become available [10:51:56] dragos.tatulea: What do you mean by "add a GetVCache function", last time I checked we already had one. [10:54:55] dragos necessarily knows that function, so he must have mistyped... [10:55:22] Indeed. Or, he means something on top of it. Let's see if he reappears. [11:05:25] > Then ours will RIP. What about End-Of-Life of OpenAFS for IRIX? [11:05:33] no pressing need: it still works [11:06:15] server is boring, won't have broken [11:06:40] Speaking of that. The change I've got to make a 'file' structure inside the vcache really could do with having someone build it on IRIX and AIX, and let me know what all the errors are. Any volunteers? [11:07:20] i can do aix 6 later. [11:07:43] Cool. I'll speak to you once 1.4.8 is out the door. [11:14:23] 1.4.8 builds through lkm, though I find BSD_KERNEL_BUILD doesn't subst by default (works if given the advertised "default" explicitly), and lkm loads; I'm will to do some server and client testing later on (do I have a day)? [11:15:46] Well, changes requiring respins will make packagers sad. So let's hope it all works! [11:16:09] ? [11:16:30] If testing reveals problems, and fixing those problems requires code fixes, then we'll need to respin all of the packages. [11:16:41] I'm making the packages? [11:16:47] er, what? [11:16:59] I thought Matt was suggesting he needs to do testing on FreeBSD. [11:17:02] impedence mismatch somewhere, or a mix, or i missed something [11:17:09] I think I'm just confused. [11:17:24] Ignore moe. [11:17:26] me, even. [11:18:17] I was just asking about Derrick's release timing [11:19:28] Okay. Well, I think we've got a few hours more of RPM building to go. [11:20:04] check, thanks Simon [11:20:09] the problem with release timing is since i am not doing all the builds myself i never know what the timing is [11:20:31] it's: -here's a tarball [11:20:36] (time passes) [11:20:40] -announce a release [11:21:00] think of step 2 as the underpants gnomes ? step [11:23:37] > Indeed. Or, he means something on top of it. Let's see if he reappears: Yeah...but GetVCache gets a fetches a vcache from wherever and PutVcache does just a refcount decrement. I was thinking about a global function that does the refcount increment. osi_vnhold is supposed to do that but there are several usecases that do more than that. [11:24:00] Ah, so you want an abstraction for osi_vnhold ? [11:24:09] Something like that, yes. [11:24:20] I think the problem is that there are cases where all you need to do is the osi_vnhold, and other cases where more magic is required. [11:24:30] exactly [11:24:31] As far as I can tell 'more magic' is usually restricted to Darwin. [11:24:36] exactly [11:25:00] I don't know enough about Darwin to tell whether we have the correct sprinkling of magic, or whether there are places without magic that really should have it (if you see what I mean). [11:25:23] I suspect, for that reason, that an additional abstraction would have to be carefully designed and considered. [11:25:39] That's why I asked. [11:26:26] Why not propose something specific? What would the abstraction be, and what code would it replace. Bug Derrick, or someone else who knows Darwin internals about what the correct Darwin version of the abstraction is. [11:26:47] My suspicion is that there's two ways of holding a vnode - one that says that you're going to play with the metadata, and the other which says you're going to do IO on it. [11:27:08] Good tip. [11:27:12] Matt might also be able to help, as I believe the Darwin VFS layer is at least slightly similar to the BSD one. [11:27:50] shadow, matt: Could you provide us with details on this? [11:28:31] RefVCache [11:28:51] i'll have to look and see which magic you want on darin [11:29:15] chaskiel wrote the darwin stuff. i know what the point is but i need to reread to figure out which is which case now [11:29:23] > as I believe the Darwin VFS layer [11:29:28] not anymore [11:29:39] refcount versus iocount [11:29:41] now [11:29:50] whereas bsd has only the usual refcount [11:30:02] Ah, so iocount is a purely Darwin thing? [11:30:12] i believe so [11:30:23] it came in with the revised 10.4 vfs interface [11:30:26] That's kind of them. And its so well documented, too. [11:31:57] cc-1164 cc: ERROR File = test_pam.c, Line = 137 Argument of type "int" is incompatible with parameter of type "const char *". r->resp = strdup(getpassphrase(m->msg)); [11:32:32] Well, if it's only this...then having an even more abstract function for that would be kinda' useless, if not worse [11:32:32] --disable-pam [11:33:38] http://lists.apple.com/archives/darwin-kernel/2005/Jun/msg00042.html [11:33:56] notice the lack of answers. at the time it was undocumented. [11:34:57] lol [11:35:01] some code sample eventually explained it. let me see if i have it around [11:35:21] no. but http://developer.apple.com/samplecode/EmptyFS/listing4.html seems to be the sameish [11:36:10] note the date of chaskiel's message, and then of that document's first rev [11:37:08] yeah, one year [11:37:33] they needed time to figure it out for themselves :P [11:38:28] unlikely. the people coding don't likely write end-user docs tho [11:38:43] So how do we remember the build flags for "next time"? [11:39:25] shadow, Simon Wilkinson: Well, thanks for the help anyway. At least I cleared my usecase. [11:43:02] harald, i should start shipping build-openafs.sh if i didn't already [11:43:28] --- Russ has become available [11:43:31] i already have a script designed to be the institutionalized knowldge [11:43:44] in fact [11:43:47] *-irix*) MAKE=gmake ./configure --enable-transarc-paths --disable-pam MAKE=gmake make dest exit 0 ;; [11:44:22] * Russ tries to figure out what to do about 1.4.8 for Debian. It's too late for lenny to put in a full release, and experimental is already 1.5. Hm. [11:44:28] The Linux RPMs are now regenerating with the new CellServDB. [11:45:09] I really should look further into RPMs of the 1.5.x series for Fedora ... maybe next week ... [11:45:20] As soon as lenny releases, I'll upload it to unstable and backport it to backports.org, but it's not particularly clear what to do before then. [11:45:54] PATH=/usr/bin/X11:/usr/bin:/bin:/usr/sbin:/sbin ./configure --enable-transarc-paths --disable-pam ; MAKE=/usr/gnu/bin/gmake PATH=/usr/bin/X11:/usr/bin:/bin:/usr/sbin:/sbin $MAKE dest [11:46:20] => /afs/stacken.kth.se/home/haba/src/openafs-1.4.8-irix/sgi_65/dest [11:47:25] $ uname -a IRIX64 pynchon 6.5 07202013 IP35 [11:59:40] maybe run a chown over my dest tree unless you want to dist as UID=18118 :) [12:00:23] Hm, if we generate the release tarballs with a script, that script could use tar --owner=0 --group=0 to handle that sort of thing automatically. [12:01:02] Well, if you have GNU tar. [12:04:30] can you also tar it up for me? [12:04:34] (and gzip) [12:04:42] so i don't have to copy it uncompressed to here? [12:08:50] well well, let me see, somewhere gnu tar around. [12:12:43] I'm not quite sure last releases 1.4.7/irix-6.5/sgi_65.tar.gz does contain the right stuff. [12:16:57] My gtar on pynchon was too old. Next host.... [12:22:04] Ok, have a look at sgi_65.tar.gz sgi_65.tar.gz.md5 in the same dir [12:24:43] It contains the dest tree and not what the last releases irix tar contained (sgi_65/src/WINNT/... for example) [12:32:47] --- dev-zero@jabber.org has left: Replaced by new connection [12:32:47] --- dev-zero@jabber.org has become available [12:35:28] simon (hold vnode): how did we get from GetPutVCache to osi_vnhold? [12:35:40] --- tkeiser@sinenomine.net/owl has left [12:36:53] What Dragos was looking for was an inverse of PutVCache [12:37:16] That is, a function that you can call with a vcache structure that will increment the reference count. [12:37:38] At the moment, there are multiple different ways of doing that scattered throughout the code, without much consistency. [12:37:40] the (open|free|net)bsd all differ. generally, they agree that locking the vnode holds it; net still has a vn_hold(), but its highly special [12:38:31] I think the issue is those places where you want to hold a reference, but release the lock, [12:38:44] vref [12:39:00] Yup. But we've got no consistent way of doing that at present. [12:41:56] Id need to prep a bit before getting deep into it, but I think that things have been going towards uniformly calling osi_v(hold|ref|something) functions or macros, which can cover the vfs differences [12:53:27] Is making anything that uses the 'auth' module also require 'util' likely to be a problem? [13:42:50] I doubt it [13:45:26] It builds fine, anyway. I think I'm going to use afs_snprintf and be done with it ... [13:46:54] --- Simon Wilkinson has left [13:50:20] --- matt has left [14:21:32] that sounds like a fine idea. it avoids all of the "what is the platform specific snprintf return behavior" issues [14:47:58] --- edgester has become available [14:56:32] --- SecureEndpoints has left [15:14:51] --- tkeiser@sinenomine.net/owl has become available [15:34:58] --- tkeiser@sinenomine.net/owl has left [16:30:18] --- SecureEndpoints has become available [16:42:53] belated: grabbing it now, haba [16:47:57] > Id need to prep a bit before getting deep into it, but I think that > things have been going towards uniformly calling yes [19:55:50] --- edgester has left [22:27:55] --- summatusmentis has become available [22:31:53] --- summatusmentis has left [23:13:32] --- Russ has left: Disconnected