[01:00:47] --- dev-zero@jabber.org has left [01:11:21] --- reuteras has left [01:41:07] --- sxw laptop has become available [02:28:08] --- Jeffrey Altman has left: Disconnected [02:28:24] --- jaltman has left: Disconnected [02:31:58] --- sxw laptop has left [02:41:28] --- dev-zero@jabber.org has become available [03:15:59] --- sxw laptop has become available [03:17:07] --- sxw laptop has left [03:23:19] --- haba has become available [05:15:57] --- manfred has become available [05:19:18] helo [05:19:25] hi [05:19:39] small question on makefile.in [05:20:00] i need to change the makefile for JAVA dir [05:20:29] because on the mac osx is a bit different the linking .. and other small thing [05:20:30] s [05:21:01] how can create a ifeq for osx platform [05:21:07] now i have : [05:21:35] we don't assume gmake, so ifeq isn't safe [05:21:46] ifeq (${SYS_NAME}, x86_darwin_100) JNI_INC=-I ${JAVA_HOME}/include LIBJAFS_NAME=libjafs.jnilib LIBJAFSADM_NAME=libjafsadm.jnilib else JNI_INC=-I ${JAVA_HOME}/include -I `dirname ${JAVA_HOME}/include/*/jni_md.h` LIBJAFS_NAME=libjafs.so LIBJAFSADM_NAME=libjafsadm.so endif [05:22:12] oh god no. use autoconf substitution. notice the @XXX@ variables? look at src/cf/osconf.m4 [05:22:37] ok never used .. i check on it [05:22:38] also, -I `dirname ${JAVA_HOME}/include/*/jni_md.h` is scary. let autoconf find it [05:22:58] the JNI_INC=-I ${JAVA_HOME}/include -I `dirname ${JAVA_HOME}/include/*/jni_md.h` is not mine .. [05:23:02] there's path detection logic for other files. you want the same [05:23:16] oh, you're right. we had it before. sigh. [05:23:18] is present .. probably from the stone age.. [05:23:28] it should die. but oh well. [05:23:40] when your patch is done i'll figure out how to fix it [05:23:46] i need the path and the linking [05:24:09] file names [05:24:10] ifeq (${SYS_NAME}, x86_darwin_100) SHARED_FLAGS = -dynamiclib else SHARED_FLAGS = -shared -Xlinker -Bsymbolic endif [05:24:14] --- stevenjenkins has left [05:24:18] no ifeq [05:24:21] seriously. [05:25:01] use e.g. SHARED_FLAGS = @SHARED_FLAGS@ and then set that from autoconf [05:25:08] tho we already do that [05:25:09] I know how to fix the compilation problem .. I don't know how to fix the autogen [05:25:23] er. fix the autogen how? [05:25:55] any reason you can't instead of using ${CC} ${SHARED_FLAGS} just use ${SHLIB_LINKER}? [05:26:11] which already uses -dynamiclib or -XlinkerX -whatever? [05:27:04] --- stevenjenkins has become available [05:29:03] and for the lib name ? LIBJAFS_NAME=libjafs.jnilib / LIBJAFS_NAME=libjafs.so [05:29:09] always m4 ? [05:29:54] --- sxw has become available [05:31:13] it does make more sense to use autoconf for those. you have no other reasonable choice. it's either autoconf variable substition, or a korn shell case statement (there are examples in the makefile) [05:34:01] --- sxw has left [05:38:04] i test now with SHLIB_SUFFIX [05:38:18] because the problem is on the extension [05:38:40] seems to work [05:39:30] cool [05:40:14] :-P [05:40:25] the make file .. [05:40:35] now the problem is on the symbols .. not found .. [05:41:14] the symbols? [05:50:00] Undefined symbols: "_rxkad_random_mutex", referenced from: _rxi_InitPthread in libafsrpc.a(rx.o) "_des_random_mutex", referenced from: _rxi_InitPthread in libafsrpc.a(rx.o) "_des_init_mutex", referenced from: _rxi_InitPthread in libafsrpc.a(rx.o) ld: symbol(s) not found collect2: ld returned 1 exit status Undefined symbols: "_des_init_mutex", referenced from: _des_init_mutex$non_lazy_ptr in libafsrpc.a(rx.o) "_rxkad_random_mutex", referenced from: _rxkad_random_mutex$non_lazy_ptr in libafsrpc.a(rx.o) "_des_random_mutex", referenced from: _des_random_mutex$non_lazy_ptr in libafsrpc.a(rx.o) [05:50:56] git fetch [05:51:04] git checkout origin/master [05:51:48] that's an issue with ranlib and we commited the fix yesterday [05:52:02] i guess wait.... don't checkout origin/master or you will lose your changes [05:52:13] unless you committed, in which case, you can cherry-pick it [05:52:54] --- Claudio Bisegni has become available [05:53:16] ok thanks .. [05:54:27] you will need to ./regen.sh [06:06:02] --- pod has left [07:04:21] last small problem [07:04:21] Undefined symbols: "_rxkad_random_mutex", referenced from: _rxi_InitPthread in libafsrpc.a(rx.o) "_des_random_mutex", referenced from: _rxi_InitPthread in libafsrpc.a(rx.o) "_des_init_mutex", referenced from: _rxi_InitPthread in libafsrpc.a(rx.o) ld: symbol(s) not found [07:07:51] still the same problem. still the same fix. [07:08:15] cheat for now: edit src/config/Makefile.config [07:08:29] make RANLIB expan to "ranlib -c" [07:11:26] --- deason has become available [07:11:29] --- jaltman has become available [07:34:52] ld: duplicate symbol _ubik_GetVersion in /Developer/workspace/AFSManager/openafs/lib/libubik.a(ubik.o) and ubik.o for architecture x86_64 [07:37:24] --- meffie has become available [07:37:35] what the hell are you building on? you appear to have both ranlib bugs [07:38:10] the way things seemed to have worked was you could *either* use ranlib -c or ranlib (wihtout -c) depending which bug you had [07:39:06] i can't fix it: it's apple's bug. you need to figure out which works, and use it all the way through for a make (after a make clean) [07:39:09] :-) [07:39:16] mac osx 10.6 [07:39:35] on 10.6, RANLIB=ranlib [07:39:42] on 10.5, RANLIB=ranlib -c [07:39:55] on 10.4, RANLIB=ranlib [07:41:29] the openafs it is ok with ranlib .. [07:42:18] ok. well, use that. then... the link error earlier came from what? [07:43:15] from java dir [07:43:23] or better .. [07:43:24] sorry [07:43:44] i try to compile with ranlib -c ... and I received a duplicate error [07:44:08] than is correct [07:44:31] now a try to compile with ranlib also the java [07:54:50] - openafs compilation ok [07:55:09] - JAVA libjafs ok [07:55:38] - Java libjafsadm ko [08:00:41] --- Claudio Bisegni has left [08:38:18] in some part of the website, the last modified is malformed ... Last modified: $Date$ [08:47:24] > the last modified is malformed Blame Russ, who converted the web site to a version control system that doesn't do keyword expansion. [08:52:55] :-) [08:58:30] derrick I read now your mail .. [10:07:33] --- haba has left [10:40:43] --- manfred has left [11:04:03] --- sxw has become available [11:24:34] --- sxw has left [11:34:39] --- dev-zero@jabber.org has left [12:30:49] --- dev-zero@jabber.org has become available [12:36:32] --- manfred has become available [13:02:27] --- jaltman has left: Disconnected [13:25:36] --- haba has become available [13:56:39] --- jaltman has become available [14:02:48] --- manfred has left [14:06:57] --- deason has left [14:06:57] --- deason has become available [15:32:26] So, it turns out writepage() is even more of a mess than I first thought. You can end up reading/writing with the credentials of any user of AFS on the system you happen to be on. [15:33:25] i wish i could tell you i was shocked [15:34:55] Basically, if you start running low on pages during a write() operation, the kernel may decide to start flushing pages - _any_ pages. And in some kernels, it does that in the context of the original write(), which might be by any user and (in earlier versions) to any fs. [15:35:23] we *tried* to exclude those, istr. badly, almost certainly [15:35:26] Thankfully, we do access control on open(), so none of this actually lets you do things with the wrong credentials. It's most likely to lead to EIO. [15:36:41] I have a patch which does the same thing as we do on IRIX - stash the creds of the last user to open the file in the vnode, and then use those creds to perform writeback. [15:37:18] Still not ideal - what happens if the last opener's creds expire, but an earlier users would still be valid - but more likely to work than what we've got at the moment ... [15:38:05] we have to defer the writes until we can write... or lose them [15:38:27] we should probably do it more than just on irix; i bet most of the systems have this problem [15:38:33] I guess we don't actually care if DoPartialWrite fails. [15:38:53] (returning it's error code to the caller of writepage() is probably the wrong thing to do) [15:39:41] And afs_close had better have the right credentials, or you're boned in the same way as normal. [15:56:02] --- deason has left [15:56:09] --- Jeffrey Altman has become available [16:00:34] 687, 685 and 674 are the current set of changes from trying to track down Marc's mmap issues. The irony that none of them have actually solved it is not lost on me. In fact, fixing the deadlocks just exposes the real problem more! [16:01:40] ok, lemme fire up a linux vm [16:03:57] Only 685 is Linux specifc. [16:04:22] The other two are issues from the fetchstore refactoring (674 is already pushed, thanks!) [16:04:26] might as well do it anyway [16:05:57] I have another pending change that fixes our radix tree usage, which is also broken. The symptom of that is that we'll get asked to write the same page multiple times. [16:06:18] ... but I need to figure out how redirtying fits with AOP_WRITEPAGE_ACTIVATE [16:12:02] i wonder why my VMs are hurting me so badly memory wise. i have 4gb [16:13:37] Lovely, 125519 is another HashOutDCache crash [16:14:50] now i remember why i keep trying to not look at rt [16:40:20] looks like it works. looks like [17:41:27] --- jaltman has left: Replaced by new connection [17:41:27] --- jaltman has become available [17:50:34] --- deason has become available [20:42:13] --- deason has left [22:16:28] --- reuteras has become available [23:15:31] --- Simon Wilkinson has left [23:59:41] --- manfred has become available