Re: aufs max branches
Tomas M:
and then I add branches one by one, by using
mount -o remount,add:1:/mnt/images/image1 aufs /union
for example. This works OK and branches are added to the union, but
remounting the union ro causes the oops I sent you. Do you still agree
it's a problem of the lenght of mount arguments?
I think so.
Will you try this patch?
Junjiro Okajima
Index: fs/aufs/opts.c
===
RCS file: /cvsroot/aufs/aufs/fs/aufs/opts.c,v
retrieving revision 1.24
diff -u -p -r1.24 opts.c
--- fs/aufs/opts.c 12 Feb 2007 02:57:13 - 1.24
+++ fs/aufs/opts.c 13 Feb 2007 12:04:21 -
@@ -633,9 +633,12 @@ int parse_opts(struct super_block *sb, c
Err(unknown option %s\n, opt_str);
break;
}
+
if (!err !skipped) {
if (unlikely(++opt opt_tail)) {
err = -E2BIG;
+ opt--;
+ opt-type = Opt_tail;
break;
}
opt-type = Opt_tail;
-
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier.
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnkkid=120709bid=263057dat=121642
Re: aufs max branches
Tomas M: I tested the patch. No difference, remount,ro causes oops in kernel/exit.c:860 How about BUG: ... NULL pointer ...? Your gif file showed me it before BUG: at kernel/exit.c:860. Junjiro Okajima - Using Tomcat but need to do more? Need to support web services, security? Get stuff done quickly with pre-integrated technology to make your job easier. Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo http://sel.as-us.falkag.net/sel?cmd=lnkkid=120709bid=263057dat=121642
Re: aufs max branches
Tomas M: It was in Linux Live CD all the time, using 'mount' command from busybox. Now I tried it in a normal Linux (installed) and it doesn't segfault, remount simply exits with E2BIG (Argument list too long). Thi is the effect of the patch I sent. If you used the un-patched aufs, it would not return the error and cause a problem. I noticed the line in /etc/mtab (mount arguments) can't be longer than 4096 characters. This is 2^12 I guess. Yes, it is the limitation of mount and /proc/mounts. It is the size of a memory page. And aufs will try sysfs as you suggested. -- Now I decided to skip unionctl at all, and use mount -n -o remount,add:1:/new/branch=ro aufs /union ::: Remount segfaults the same in Live CD. BUT! I am unable to reproduce the problem in my installed Linux, even if using busybox's mount. So it it very strange :) Please let me make sure. - you did it on your hdd-installed Linux, instead of Live CD - the aufs patch was applied - mount.aufs was not installed - you executed busybox mount -n -o remount,add:1:/new/branch=ro aufs /union - the busybox aborted by segfault Which problem do you wrote as unable to reproduce? -- So here is how did I fix it: - I call: mount -o remount,ro aufs . Well, probably it is a busybox issue. But aufs should work correctly which includes the expected error. Do you think aufs has no problem anymore? I am confused a little... Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: OK, I give..Help?
Chris Furlough: What I'd like to be able to do, is to mount something over the whole / mount point. Then you did it. And you see everything under /root/jukebox_cddb/aufs, except /proc, /sys, /dev/pts and /dev/shm, don't you? Is it enough for you? Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: aufs on a 2.4 kernel ?
Flemming Madsen: Is anybody by any chance contemplating a backport of aufs to the 2.4 kernel. Would be much appreciated for use in embedded systems. Unfortunately, I don't have such machine and disk space. If anybody start backporting, I will try answering the questions. I forgot many things in 2.4 since it was years ago when I touched it last time. I guess it is possible to backport aufs if all the features are not necessary. i.e. the limited or basic features only. Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: AUFS Kernel module crashes
Marty Rosenberg: so it seems like aufs gets extremely unstable when running with my setup. After attempting to run a du operation or two, the kernel module crashes. I'm not sure if it occurs before the module crash, but other actions such as ls will hang. Other people using the system have said that they can still run ls in subdirectories, however I have not found this to be true. In case it helps, there are approximately 773,950 files total that are in all of the source directories. Hello Marty, Thank you for your report, including /proc/mounts and others. Since vdir.c:603 is checking an internal lock, aufs seems to be totally broken. I will review aufs code. Do you know what is the difference between you and other people using the system? How about the compiler's version? Are you using the same compiler with other people? Or kernel configuration? Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: OK, I give..Help?
Chris Furlough: Then you did it. And you see everything under /root/jukebox_cddb/aufs, except /proc, /sys, /dev/pts and /dev/shm, don't you? Is it enough for you? Either I don't understand aufs, or it's not working. After the above mount, I went to /root, and created a file, and threw some text into it. Now, in theory, if I understand correctly, this file should ONLY exist virtually, as in only in the AUFS. So, I reboot, and when it comes up, the file is there, and I can cat it normally. Is the / dir NOT read-only? Is this file existing in AUFS across boots? You went to /root, it is out of aufs. Your aufs is /root/jukebox_cddb/aufs, isn't it? Your newly created file is saved in /root. I am afraid you don't understand aufs or mount. Probably, you need to try LiveCD, for example SLAX. Basically, it doesn't save the created files in your hard drive. Also, I recommend you to read the manual of chroot(8). Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: hm weird behavior of unionctl
Patrick: Heh, will do . but is that behavior of unionctl supposed to be like that? :) Of course, no it isn't. And Tomas M suggested you one of the way to research the problem. Thank you Tomas. Additionally, there is a description about the time of remount. When you have lots of inodes and dentries cached, it may take a long time. It depends upon your operations before remount, and the amount of your system memory. In your case, in order to know your situation, I'd like to ask you questions. - How did you mount your aufs? - How is /proc/mounts just before you remount or unionctl? Will you show us? - Have you ever read the manual of mount(8), or some documentations about /etc/mtab and /proc/mounts? Since it must be fundamental knowledge when you use linux or unix. And, - it is recommended to use mount -o remount,... instead of unionctl. - mount.aufs is not supposed to be executed manually. use mount(8). - if you use -n option to your mount(8), you will never get the entry in your mount listing by mount(8). Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: nwkq=N
Tomas M: I still can't find it. In aufs.in.5 there is written: default value is . The manual page, aufs.5 is built by make. Please make it. Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: unionctl issue
Fernando A. P. Gomes: [EMAIL PROTECTED] aufs]# unionctl /mnt/aufs/filesystem --add --before /mnt/aufs/dir1 --mode ro /etc [EMAIL PROTECTED] aufs]# unionctl /mnt/aufs/filesystem --list /etc (r--) /mnt/aufs/.changes (rw-) /mnt/aufs/dir1 (r--) /etc goes two position above dir1, when should go only one. Thank you very much. I will include this patch in next Monday release. Junjiro Okajima -- Index: util/unionctl === RCS file: /cvsroot/aufs/aufs/util/unionctl,v retrieving revision 1.18 diff -u -p -r1.18 unionctl --- util/unionctl 19 Feb 2007 03:26:18 - 1.18 +++ util/unionctl 20 Feb 2007 13:12:20 - @@ -129,7 +129,6 @@ do_add() do case $1 in --before) bindex=`find_bindex $2` - bindex=`expr $bindex - 1` || : ;; --after)bindex=`find_bindex $2` bindex=`expr $bindex + 1` || : - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: aufs BUG
Hi, Fernando A. P. Gomes: What did your rc script do? Nothing anormal, it's in attach. I will wait for your next mail. :-) Are there something about mounting, module loading/unloading, chroot(2) and operating around /var/log/squid? Aufs module is builtin I can't loading/unloading. The only operation around /var/log/squid is /etc/init.d/squid start (open log files in /var/log/squid). How about another modules? I want to make sure that all necessary modules are loaded correctly, and there is no kernel messages. What you mean with chroot? Is I do a chroot and a pivot_root. What I want to know is whether you did chroot(2) in your rc script, or not. Do you mean that you did both? And after your rc script, is the aufs entry in /proc/mounts correct? Yes Will you show me your /proc/mounts? Is /cdrom/tmpdir/3 working correctly, how about ls(1) to it? Which type is its filesystem? It's a cramfs filesystem mountend looped and unionctl to aufs. I realize that the bug is in more locations that de /var/log, in /etc I've the same problem with some files ( ex: /etc/wvdial.conf ). Yes, I read your another mails. Your aufs seems to handling ghosts of inodes. They seems to be gone. But aufs believes they are still existing, and shows you almost NULL data or garbages. So you can ls(1) to your branches directly, and it succeeded, right? Then, how about ls to your aufs after direct ls? Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: hm weird behavior of unionctl
Patrick: Do you mean that 'mount -o remout,mod:/real_root=rw /your/aufs' returns an error? What did it say? ::: mount -o remount,mod:/real_root=rw / mount: / not mounted already, or bad option Are there any aufs messages left? I think that is logical since neither /real_root nor /union exists or is Currently I don't agree. Will you try strace, and show me the output? Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: hm weird behavior of unionctl
Patrick: logs etc. to ram .. but sometimes I need modifications to be done permanently, thus infact I would need that. But isnt there another way? e.g. copying / mount-moving/binding real_root to the union or something like that? I think it depends on your initramfs. As long as you use initramfs, you cannot see /real_root, can you? If aufs introduces a new option which enables changeing permission mode of a hidden branch, you cannot write to it directly. But here is an idea, which might be rather complex operation. - introduce some mount options, + idel:branch index removes a branch specified by index, instead of path. + imod:branch index:new permission modifies a permission mode of the branch specified by index, instead of path. + imove:old branch index:new branch index re-order the branches, from old to new. it might be dangerous. + coo copyup on open + nocoo disables coo - mount -o remount,imod:1:rw,imove:1:0,imod:1:ro,coo /your/aufs - head -1 /your/aufs/whatever/you/want/to/copyup - mount -o remount,imod:1:rw,imove:1:0,imod:1:ro,nocoo /your/aufs Is that what you need? Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: Can't compile aufs patched 2.6.20-rc1 kernel cleanly
Tapani_Rikknen: This must be my ignorance but compiling new kernel (in a debian way with = make-kpkg) with aufs patch gives me following error messages: ::: include/linux/aufs_type.h:40:2: error: #error unknown = CONFIG_AUFS_BRANCH_MAX value fs/aufs/module.c:292:5: warning: AUFS_BRANCH_MAX is not defined Assuming your 'aufs patch' means 'aufs CVS tree on sourceforge', how did you configure/compile it? Just 'make -f local.mk'? (didn't get any extra info with 'aufs debugging feature' enabled) Do you mean that you enabled CONFIG_AUFS_DEBUG? It is for run-time, not compile-time. My original problem is the same as someone has described in the message: http://lists.debian.org/debian-knoppix/2007/02/msg00014.html And in my own message: http://knoppix.net/forum/viewtopic.php?t=3D26812 Will you try ls(1) or something just after mounting aufs, in order to make sure, - aufs is loaded correctly - aufs is mounted correctly - you can access mounted aufs correctly - are there any messages left in your syslog or something? I read both of the URLs and know that you did 'make -f local.mk.' On my test environment, I use plain linux-2.6.20. I will try linux-2.6.21-rc1 in a few weeks. I can compile aufs module fine but it don't work in my remastered = Knoppix 5.1.1 (giving segmentation fault message during boot process or = Which process aborts with segmentation fault? I would prefer patching kernel tree directly with aufs patch than = compiling it afterwards (make -f local.mk), if it could be possible. Did you try 'make -f locak.mk kconfig'? -Unionfs is going to be included on vanilla 2.6.21 from kernel.org = (if I'm right); what about aufs? Currently I have no such plan. -Can aufs be used with realtime preemption kernel = (http://people.redhat.com/mingo/realtime-preempt/)? Should aufs work on = liveCD with -rt patched kernel (I never succeeded with unionfs)? Although I have never tried this patch, I will try someday, probably far future... Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: aufs BUG
Tomas M: But it seems to me like Fernando is using switch_root, which is used if you're using initramfs. Initramfs is something very different from ::: Fernando wrote that he uses initrd. I don't know he switched to initramfs... Thank you for your follow-up and pointing out. Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: aufs for disaster recovery
Hello Ramy, Ramy M. Hassan : Unfortunatly, I am not able to use the current version of aufs for this purpose, because export via NFS is not yet supported by NFS. I see it in the list of (current work). Junjiro, could you please let me know when we should expect this feature in aufs ? Currently I have finished implementing it and testing now. I sent a patch to an aufs user since he requested this feature strongly and propsed as tester. So I guess he and I are testing it now. Will you join this test? Of course, it is a test and I never recommend you to use it for production use. Is there any tool that allows merging the two branches in the upper read-write branch, so that we can safly remove the corrupted branch to be reformatted ? I know it should be easy to implement (maybe the snapmerge script that comes with unionfs can be used). Unfortunately, I don't have such tool currently. Also I think that it is worth to try the snapmerge script in unionfs. While I am not sure, here is an idea. # mount -t aufs -o br:/rw:/ro none /aufs (some works) # mount -o remount,prepend:/new_rw,xino=/new_rw/xino,mod:rw=ro /aufs (continue working) # mount -t aufs -o ro,br:/rw=ro:/ro none /aufs2 # cp -ru /aufs2 /aufs # umount /aufs2 Our application uses O_DIRECT (as a compile time flag) for performance reasons. We can disable O_DIRECT to be able to use aufs, although it would be very nice if aufs supported O_DIRECT (Junjiro was asking if that was neccessary). Ok. I will append it to my work-list. I am afraid reacing it will take a longer time since some aufs users have some troubles now. You might know it if you read this ML. If you have a simple test case for O_DIRECT which I can use testing aufs, please send me. Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: aufs BUG
Fernando A. P. Gomes: switchroot doesn't have exit code. switchroot replaces the current process (the init in initrd) and then executes the new init. Since you are still using initrd instead of initramfs, I must ask you again. How did you identify that switchroot succeeded? Your kernel may execute init process after your linuxrc/switchroot aborts. But it may depends upon your parameters. switchroot moves the specified diretory to /. What was before that directory can't be accessed anymore. Do you mean that you took your aufs apart from its branches. Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: Possible to 'invalidate cache'?
Jrgen_P._Tjern: So either there were other conditions to cause my BUG @ super.c:357, or = the newest CVS fixed the problem by itself. There is a condition to reproduce this bug. $ cd /your/aufs/subdir $ sudo mount -o remount,udba=inotify /your/aufs then you will meet BUG @ super.c:357 again. It means the subdir is in use, and aufs tries some processing at remount-time. If you didn't use any of subdir, you would not meet the bug. I guess the samba service daemon refered a subdir when the first time you met the bug. Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: BUG @ hinotify.c:123!
Jrgen_P._Tjern: [14664.834300] kernel BUG at /root/aufs.wcvs/aufs/fs/aufs/hinotify.c:123! This is very strange too. I cannot guess what is wrong. But I will try kernel 2.6.20.3 and smb exrporting in a few days. Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: aufs Tuesday release (Re: skip aufs Monday release)
Tomas M: I think the Tuesday's release can't be compiled. ::: /usr/src/a/fs/aufs/misc.h:150: error: dereferencing pointer to incomplete type Sorry. It was due to my debug environment. Please update your CVS tree and try again. I have just updated misc.h. Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: aufs Tuesday release (Re: skip aufs Monday release)
Marty Rosenberg: I'd like to apologize for not giving enough information in the past, I'll try to be a bit more verbose this time. Config: I've patched a vanilla 2.6.18.8 kernel with this tuesday's release. M Another unionfs --- These options are generated automatically for 2.6.18.8 [*] Use simplified (fake) nameidata Maximum number of branches (127) --- [*] Use sysfs/fs/aufs --- INOTIFY and AUFS_HINOTIFY are disabled [*] NFS-exportable aufs [ ] ksize.patch was applied or not [*] Debug aufs [ ] Compatibility with Unionfs (obsolete) Thank you. I can see no problem in your configuration. But version 20070327 had a bug and you could not compile it. Did you fix it by yourself? If not, please check your loaded or linked aufs version. the contents of /etc/exports is ::: /data_test 192.168.0.0/255.255.255.0(async,no_subtree_check,ro) To export aufs, you need to specify fsid option in your exports. Please read the manual. But I don't think this is the reason of mount: stargate:/data_test failed, reason given by server: Permission denied. So I am afraid that your loaded or linked aufs version is old. pesky inode bug: still there. Still I am trying reproduce your problem, but I never succeeded. Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: Stale NFS file handle (wtf?)
Nothing interesting in dmesg. Compiled using gcc 3.4, local.mk and = kernel v2.6.20.3. Not using latest daily, will recompile to that and = reload module later on. I'm guessing that will solve the problem = You have changed your environment, thank you, but you still had a problem and I cannot reproduce your problem. I think I am stucking about your problem. If anyone can reproduce Jorgen's problem, please let me know how. Jorgen, If you meet BUG at iinfo.c:62 again, please send me the stacktrace. Your new kernel configuration will show us the correct stacktrace. And send me your .config too. Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: aufs backport to kernel 2.6.15
Hi Hans, Hans-Peter Jansen: before switching my users to openSUSE 10.2 diskless, I tackled the job of backporting aufs to (at least) 2.6.11 for SuSE 9.3 (diff attached). ::: All kudos to you! Thank you very much. I believe you have not only good skills but also frontier spirit. I am very glad your success. But I am afraid you will meet some problems in the future. I'm overwhelmed by this level of success. Thank you very much. If you visit Germany one day, or I am Japan, I would be pleased to invite you for a nice dinner at large - please let me know, seriously! Haha, I decided now. Today I shall have a dinner with sausages, potatos, sour cabbage, Butter AUFS Brot, beer, mit singing Ein Prosit. :-) If you want to do me a(nother) favour, please have a look on my changes and give me a note if something stands out badly. I you consider applying something like that, let me know, and I will try to clean it up. - The macros LOCK_INODE() and UNLOCK_INODE() are not bad, but where do you use them? - I prefer defining ATTR_FILE as zero, instead of removing. - it is recommended to define CONFIG_AUFS_DEBUG. Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: .config for kernel with flaky inodes.
Marty Rosenberg: is attached. It's kinda large and now compressed. I also realized that I ::: Thank you. And I want you to try this patch. Unfortunately this is just a trial, not a solution. With this patch, I hope your kernel will tell us more information. Also, I want you to try xino option. (an old mail) -- From: [EMAIL PROTECTED] Subject: Re: AUFS incorrectly identifies files as directories. To: Marty Rosenberg [EMAIL PROTECTED], aufs-users@lists.sourceforge.net Date: Wed, 21 Mar 2007 11:50:44 +0900 Now your environment is, - vanilla 2.6.19 kernel - aufs 20070312 (20070319?) - seven xfs branches with 773,950 files total - ls(1) and find(1) show you crazy inode info - after crazy inode, aufs crashes by BUG at vdir.c:605 I will try it again on my local environment, if you have another thing which is related to this problem, please let me know. I tested again, and found no problem. One thing I can guess which might be related to your problem is xino files. Although I am no sure, will you try xino mount option specifying a non-xfs path? For example, # mount -t aufs -o xino=/non-xfs/new-file,br=... -- Junjiro Okajima --- marty.config2007-03-30 14:36:03.0 +0900 +++ .config 2007-03-30 15:23:37.0 +0900 @@ -1,7 +1,7 @@ # # Automatically generated make config: don't edit # Linux kernel version: 2.6.18.8 -# Tue Mar 27 12:10:08 2007 +# Fri Mar 30 15:23:37 2007 # CONFIG_X86_32=y CONFIG_GENERIC_TIME=y @@ -39,11 +39,12 @@ CONFIG_IKCONFIG_PROC=y # CONFIG_RELAY is not set CONFIG_INITRAMFS_SOURCE= -CONFIG_CC_OPTIMIZE_FOR_SIZE=y +# CONFIG_CC_OPTIMIZE_FOR_SIZE is not set # CONFIG_EMBEDDED is not set CONFIG_UID16=y CONFIG_SYSCTL=y CONFIG_KALLSYMS=y +# CONFIG_KALLSYMS_ALL is not set # CONFIG_KALLSYMS_EXTRA_PASS is not set CONFIG_HOTPLUG=y CONFIG_PRINTK=y @@ -65,8 +66,8 @@ # CONFIG_MODULES=y CONFIG_MODULE_UNLOAD=y -CONFIG_MODULE_FORCE_UNLOAD=y -# CONFIG_MODVERSIONS is not set +# CONFIG_MODULE_FORCE_UNLOAD is not set +CONFIG_MODVERSIONS=y # CONFIG_MODULE_SRCVERSION_ALL is not set # CONFIG_KMOD is not set @@ -253,6 +254,7 @@ CONFIG_PCI_MMCONFIG=y # CONFIG_PCIEPORTBUS is not set # CONFIG_PCI_MSI is not set +# CONFIG_PCI_DEBUG is not set CONFIG_ISA_DMA_API=y # CONFIG_ISA is not set # CONFIG_MCA is not set @@ -437,6 +439,7 @@ CONFIG_STANDALONE=y CONFIG_PREVENT_FIRMWARE_BUILD=y # CONFIG_FW_LOADER is not set +# CONFIG_DEBUG_DRIVER is not set # CONFIG_SYS_HYPERVISOR is not set # @@ -1509,13 +1489,40 @@ CONFIG_TRACE_IRQFLAGS_SUPPORT=y # CONFIG_PRINTK_TIME is not set CONFIG_MAGIC_SYSRQ=y -CONFIG_UNUSED_SYMBOLS=y -# CONFIG_DEBUG_KERNEL is not set +# CONFIG_UNUSED_SYMBOLS is not set +CONFIG_DEBUG_KERNEL=y CONFIG_LOG_BUF_SHIFT=14 +CONFIG_DETECT_SOFTLOCKUP=y +# CONFIG_SCHEDSTATS is not set +# CONFIG_DEBUG_SLAB is not set +# CONFIG_DEBUG_RT_MUTEXES is not set +# CONFIG_RT_MUTEX_TESTER is not set +# CONFIG_DEBUG_SPINLOCK is not set +# CONFIG_DEBUG_MUTEXES is not set +# CONFIG_DEBUG_RWSEMS is not set +# CONFIG_DEBUG_LOCK_ALLOC is not set +# CONFIG_PROVE_LOCKING is not set +# CONFIG_DEBUG_SPINLOCK_SLEEP is not set +# CONFIG_DEBUG_LOCKING_API_SELFTESTS is not set +# CONFIG_DEBUG_KOBJECT is not set CONFIG_DEBUG_BUGVERBOSE=y +# CONFIG_DEBUG_INFO is not set # CONFIG_DEBUG_FS is not set -# CONFIG_UNWIND_INFO is not set +# CONFIG_DEBUG_VM is not set +# CONFIG_FRAME_POINTER is not set +CONFIG_UNWIND_INFO=y +CONFIG_STACK_UNWIND=y +CONFIG_FORCED_INLINING=y +# CONFIG_RCU_TORTURE_TEST is not set CONFIG_EARLY_PRINTK=y +CONFIG_DEBUG_STACKOVERFLOW=y +# CONFIG_DEBUG_STACK_USAGE is not set + +# +# Page alloc debug is incompatible with Software Suspend on i386 +# +# CONFIG_DEBUG_RODATA is not set +# CONFIG_4KSTACKS is not set CONFIG_X86_FIND_SMP_CONFIG=y CONFIG_X86_MPPARSE=y CONFIG_DOUBLEFAULT=y - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
aufs Monday release
o bugfix - bugfix: forgot dput in a error case. o news - introduce wkq_nowait(). - refine udba=inotify, still testing. - extract some functions from sbinfo.c, create a new file plink.c. Junjiro Okajima -- Index: fs/aufs/Makefile Index: fs/aufs/sbinfo.c - extract some functions from sbinfo.c, create a new file plink.c. Index: fs/aufs/dentry.c Index: fs/aufs/hinotify.c Index: fs/aufs/i_op_del.c Index: fs/aufs/inode.c Index: fs/aufs/opts.c - refine udba=inotify, still testing. Index: fs/aufs/export.c - bugfix: forgot dput in a error case. Index: fs/aufs/inode.h Index: fs/aufs/super.h - move some declarations from super.h to inode.h. Index: fs/aufs/super.c - refine show_options() simply. - refine udba=inotify, still testing. Index: fs/aufs/whout.c Index: fs/aufs/wkq.c - introduce wkq_nowait(). Index: README Index: fs/aufs/i_op_ren.c Index: fs/aufs/vfsub.c Index: fs/aufs/vfsub.h Index: include/linux/aufs_type.h - less important changes. - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: aufs on 64 bit nodes: warnings on compilation
Hello Vasily,
Vasily Tarasov:
Hello, I've just compiled aufs on my 64bit node and there were a lot of
warnings caused by 64bit specifics. The tiny patch below fixes most of them.
However there are still a bit warnings, but it depends greatly on the author
how to fix it, so I just report about them:
Thank you.
Won't you try this patch?
- apply cast operator to size_t and ssize_t.
- refine inline au_is_mmapped().
- rename static fread() and fwrite().
- refine vfsub_read/write() which is unrelated to this issue.
Junjiro Okajima
--
Index: fs/aufs/debug.h
===
RCS file: /cvsroot/aufs/aufs/fs/aufs/debug.h,v
retrieving revision 1.25
diff -u -p -r1.25 debug.h
--- fs/aufs/debug.h 27 Mar 2007 12:45:59 - 1.25
+++ fs/aufs/debug.h 3 Apr 2007 04:00:02 -
@@ -60,7 +60,7 @@ extern atomic_t aufs_cond;
}while(0)
#define TraceErrPtr(p) do { \
if (IS_ERR(p)) \
- TraceErr(PTR_ERR(p)); \
+ LKTRTrace(err %ld\n, PTR_ERR(p)); \
}while(0)
#define TraceEnter() LKTRLabel(enter)
Index: fs/aufs/f_op.c
===
RCS file: /cvsroot/aufs/aufs/fs/aufs/f_op.c,v
retrieving revision 1.21
diff -u -p -r1.21 f_op.c
--- fs/aufs/f_op.c 27 Mar 2007 12:49:00 - 1.21
+++ fs/aufs/f_op.c 3 Apr 2007 04:00:02 -
@@ -130,7 +130,8 @@ static ssize_t aufs_read(struct file *fi
int dlgt;
dentry = file-f_dentry;
- LKTRTrace(%.*s, cnt %d, pos %Ld\n, DLNPair(dentry), count, *ppos);
+ LKTRTrace(%.*s, cnt %lu, pos %Ld\n,
+ DLNPair(dentry), (unsigned long)count, *ppos);
sb = dentry-d_sb;
si_read_lock(sb);
@@ -177,7 +178,8 @@ static ssize_t aufs_write(struct file *f
int dlgt;
dentry = file-f_dentry;
- LKTRTrace(%.*s, cnt %d, pos %Ld\n, DLNPair(dentry), count, *ppos);
+ LKTRTrace(%.*s, cnt %lu, pos %Ld\n,
+ DLNPair(dentry), (unsigned long)count, *ppos);
inode = dentry-d_inode;
i_lock(inode);
@@ -467,8 +469,8 @@ static ssize_t aufs_sendfile(struct file
struct super_block *sb;
dentry = file-f_dentry;
- LKTRTrace(%.*s, pos %Ld, cnt %d, loopback %d\n,
- DLNPair(dentry), *ppos, count, loopback);
+ LKTRTrace(%.*s, pos %Ld, cnt %lu, loopback %d\n,
+ DLNPair(dentry), *ppos, (unsigned long)count, loopback);
sb = dentry-d_sb;
si_read_lock(sb);
Index: fs/aufs/file.h
===
RCS file: /cvsroot/aufs/aufs/fs/aufs/file.h,v
retrieving revision 1.21
diff -u -p -r1.21 file.h
--- fs/aufs/file.h 27 Mar 2007 12:46:34 - 1.21
+++ fs/aufs/file.h 3 Apr 2007 04:00:02 -
@@ -104,7 +104,7 @@ static inline int au_figen(struct file *
static inline int au_is_mmapped(struct file *f)
{
- return (int)ftofi(f)-fi_h_vm_ops;
+ return !!ftofi(f)-fi_h_vm_ops;
}
/* -- */
Index: fs/aufs/sysaufs.c
===
RCS file: /cvsroot/aufs/aufs/fs/aufs/sysaufs.c,v
retrieving revision 1.1
diff -u -p -r1.1 sysaufs.c
--- fs/aufs/sysaufs.c 27 Mar 2007 12:52:39 - 1.1
+++ fs/aufs/sysaufs.c 3 Apr 2007 04:00:02 -
@@ -374,7 +374,7 @@ static ssize_t sysaufs_read(struct kobje
loff_t len;
struct dentry *d;
- LKTRTrace(%d, offset %Ld, sz %u\n, index, offset, sz);
+ LKTRTrace(%d, offset %Ld, sz %lu\n, index, offset, (unsigned long)sz);
if (unlikely(!sz))
return 0;
Index: fs/aufs/vfsub.c
===
RCS file: /cvsroot/aufs/aufs/fs/aufs/vfsub.c,v
retrieving revision 1.3
diff -u -p -r1.3 vfsub.c
--- fs/aufs/vfsub.c 2 Apr 2007 01:19:29 - 1.3
+++ fs/aufs/vfsub.c 3 Apr 2007 04:00:02 -
@@ -294,17 +294,15 @@ struct read_args {
static void call_read_k(void *args)
{
struct read_args *a = args;
- LKTRTrace(%.*s, cnt %u, pos %Ld\n,
- DLNPair(a-file-f_dentry), a-count, *a-ppos);
+ LKTRTrace(%.*s, cnt %lu, pos %Ld\n,
+ DLNPair(a-file-f_dentry), (unsigned long)a-count,
+ *a-ppos);
*a-errp = do_vfsub_read_k(a-file, a-kbuf, a-count, a-ppos);
}
ssize_t vfsub_read_u(struct file *file, char __user *ubuf, size_t count,
loff_t *ppos, int dlgt)
{
- if (unlikely(!count))
- return 0;
-
if (!dlgt)
return do_vfsub_read_u(file, ubuf, count, ppos);
else {
@@ -316,12 +314,16 @@ ssize_t vfsub_read_u(struct file *file,
.ppos = ppos
};
+ if (unlikely(!count))
+
Re: Strange nodes whose names start with .wh..wh.
Hello Sandino, Sandino Flores Moreno: I'm using aufs to rapidly create file systems for embedded devices, and after creating the new file systems, but before doing any other thing, a file and directory are always created inside the changes_dir .wh..wh.aufs .wh..wh.plink/ I'm mounting with noplink option and using aufs 20070402 with kernel vanilla 2.6.20.4 What are they for? Is it possible to avoid them? .wh..wh.aufs is the source of all whiteouts. When your branch is marked as '=rw', it will be always created. (from the aufs manual) All whiteouts are hardlinked, including 'writable branch top dir/.wh..wh.aufs.' The hardlink on an ordinary (disk based) filesystem does not consume inode resource newly. But in linux tmpfs, the number of free inodes will be decremented by link(2). It is recommended to specify nr_inodes option to your tmpfs if you meet ENOSPC. Use this option after checking by 'df -i'. The options I'm using look like this: rw,dirs=changes_dir=rw:master_dir=ro,xino=/tmp/xino,noplink You can avoid .wh..wh.plink, if you write 'noplink' before 'dirs='. The mount options are interpreted from left to right basically. It is described in the aufs manual. But I have a plan to change this rule a little bit. - at remount-time, the options are interpreted from left to right - at mount-time, br: or dirs= is interpreted last. this new rule will not confuse you. Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: dmes logs ' different uid/gid/permission' but the branches have the same permissions/uid/gid
Sandino Flores Moreno: And the directory /data/EVM_filesystems/common/masters/CSSD_Linux_12.3RC1.master/target and the other branch directory have the same permissions/gid/uid Are you sure? If it is true, it must be aufs bug. Please show me your branches by,,, $ ls -ld branch1 branch2 ... What is this warning due to? Is it a signal of something bad? (from the aufs manual) -- .B warn_perm nowarn_perm Enables or disables the warning about uid/gid/permission at adding a branch. The default is warn_perm (cf. DIAGNOSTICS). ::: .SH DIAGNOSTICS When you add an branch to your union, aufs may warn you about the privilege or security of the branch, which is the permission bits, owner and group of the top directory of the branch. If your upper writable branch has a world writable top directory, a malicious user can create any files on the writable branch directly, like copyup and modify manually. I am afraid it can be a security issue. -- By the way, this is not the first time people asked me about this warning. Is it so strange? Is the description in the manual too poor? Or people doesn't read the manual, then where and how should I describe? Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
aufs Monday release
o news - introduce new branch attribute nolwh. - refine parsing options at mount/remount time, reported by Sandino Flores Moreno. + remove whiteout-base and whplink-dir in nopllink and br_mod(), but br_del(). + increment sbinfo generation in a new function aufs_remount_fs(). + support multiple xino/noxino options in another way. + force ro when unknown branch permission is specified. + at mount-time, interpret options in this order. * simple flags, except xino/noxino, udba=inotify and dlgt. * branches. * xino/noxino. * udba=inotify. * dlgt. + at remount-time, * interpret options in the given order left to right, except dlgt which is disabled in interpreting. * create whiteout-base and whplink-dir if neccessary. * dlgt if neccessary. + new functions, clean_wh(), clean_plink(), test_linkable(), plink_dir(). - support 64bit environtment, reported by Vasily Tarasov. - introduce a sample for kprobe. o bugfix - bugfix: do not traverse the sub-mounts under aufs at refreshing cached dirs after remount. - bugfix: simultaneous read in 'dlgt' mode, reported and fixed by Ulrich Holeschak. - bugfix: LINUX_VERSION_CODE for linux-2.6.16. o misc - still testing inotify. - support lockdep in nfsd, still testing. - minor optimization for dput(). - new test functions br_writable(), br_whable() and br_linkable_wh(), replacing au_is_whable(). - do write_lock() in reinit_br_wh(). - refine updating /sys/fs/aufs/stat. - wkq_nowait() calls schedule_work(). Junjiro Okajima -- Index: fs/aufs/branch.c - minor optimization for dput(). - new test functions br_writable(), br_whable() and br_linkable_wh(), replacing au_is_whable(). - introduce new branch attribute nolwh. - refine parsing options at mount/remount time, reported by Sandino Flores Moreno. cf. opts.c. Index: fs/aufs/branch.h - new test functions br_writable(), br_whable() and br_linkable_wh(), replacing au_is_whable(). - introduce new branch attribute nolwh. Index: fs/aufs/debug.h Index: fs/aufs/f_op.c Index: fs/aufs/file.h Index: fs/aufs/sysaufs.c - support 64bit environtment, reported by Vasily Tarasov. Index: fs/aufs/dir.c Index: fs/aufs/export.c Index: fs/aufs/super.h - support lockdep in nfsd, still testing. Index: fs/aufs/file.c - bugfix: LINUX_VERSION_CODE for linux-2.6.16. Index: fs/aufs/i_op.c - new test functions br_writable(), br_whable() and br_linkable_wh(), replacing au_is_whable(). Index: fs/aufs/i_op_add.c Index: fs/aufs/i_op_del.c Index: fs/aufs/i_op_ren.c - minor optimization for dput(). Index: fs/aufs/opts.c Index: util/aufs.in.5 - introduce new branch attribute nolwh. - refine parsing options at mount/remount time, reported by Sandino Flores Moreno. + remove whiteout-base and whplink-dir in nopllink and br_mod(), but br_del(). + increment sbinfo generation in a new function aufs_remount_fs(). + support multiple xino/noxino options in another way. + force ro when unknown branch permission is specified. + at mount-time, interpret options in this order. * simple flags, except xino/noxino, udba=inotify and dlgt. * branches. * xino/noxino. * udba=inotify. * dlgt. + at remount-time, * interpret options in the given order left to right, except dlgt which is disabled in interpreting. * create whiteout-base and whplink-dir if neccessary. * dlgt if neccessary. Index: fs/aufs/opts.h Index: fs/aufs/whout.h - refine parsing options at mount/remount time, reported by Sandino Flores Moreno. cf. opts.c. Index: fs/aufs/super.c - refine parsing options at mount/remount time, reported by Sandino Flores Moreno. cf. opts.c. - bugfix: do not traverse the sub-mounts under aufs at refreshing cached dirs after remount. Index: fs/aufs/debug.c Index: fs/aufs/hinotify.c Index: fs/aufs/module.c Index: include/linux/aufs_type.h Index: sample/diskless/comount.sh Index: util/c2tmac.c - less important changes. Index: fs/aufs/vfsub.c - support 64bit environtment, reported by Vasily Tarasov. - bugfix: simultaneous read in 'dlgt' mode, reported and fixed by Ulrich Holeschak. Index: fs/aufs/whout.c - refine parsing options at mount/remount time, reported by Sandino Flores Moreno. cf. opts.c. + new functions, clean_wh(), clean_plink(), test_linkable(), plink_dir(), - introduce new branch attribute nolwh. - do write_lock() in reinit_br_wh(). Index: fs/aufs/wkq.c - refine updating /sys/fs/aufs/stat. - wkq_nowait() calls schedule_work(). Index: fs/aufs/xino.c - support 64bit environtment, reported by Vasily Tarasov. - new
Re: Strange nodes whose names start with .wh..wh.
Sandino Flores Moreno: I modified my script, so now the noplink option is before the dirs= option, but the .wh* nodes are still created. I think you already realized this. Please try the latest aufs CVS, I have updated it a few minutes before. And try the new branch attribute 'rw+lwh', for example 'dirs=changes_dir=rw+nolwh:master_dir=ro,xino=/tmp/xino,noplink' You will not see both of .wh..wh.plink and .wh..wh.aufs. But, usually, it is recommended to use 'plink' and '=rw.' Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: aufs Tuesday release (Re: skip aufs Monday release)
Marty Rosenberg: To export aufs, you need to specify fsid option in your exports. Please read the manual. Sorry, this seems rather trivial, but I'm still unsure of how to get the fsid of a given file system. You can set it whatever you like. But it must be unique number. more or less the same config. I'll have to try this with another box. I think part of the problem might be that the system that I'm attempting to mount it on is on a 64 bit system. I remember that there were some issues with the number of bits for the handles, but I'm not sure. So your nfs server is 32bit system, and your nfs client is 64bit system. Actually I have not tried such case. Can somebody else try this case? Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: aufs bug, my first real one
Hello, Tomas M: mount -o remount,add:1:/mnt/live/memory/images/some_dir Then I start kbuildsycoca to refresh KDE menu. Sometimes it is OK but after adding few branches this way dmesg shows the error attached, after kbuildsycoca. my configuration: CONFIG_AUFS=m CONFIG_AUFS_FAKE_DM=y CONFIG_AUFS_BRANCH_MAX_1023=y CONFIG_AUFS_SYSAUFS=y CONFIG_AUFS_KSIZE_PATCH=y there is approximately 10 branches in union. I could not reproduce this problem, and I can't guess the cause. Won't you enable CONFIG_AUFS_DEBUG and reproduce it? I hope BUG_ON macro in somewhere can detect the problem. But I am not confident. Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
Re: aufs bug, my first real one
Tomas M: So I enabled debug and here is the result (attached) ::: kernel BUG at /usr/src/a/aufs/fs/aufs/iinfo.c:64! ::: Call Trace: [f8cec6be] au_new_inode+0x45e/0x510 [aufs] [f8ced1f9] aufs_permission+0x149/0x260 [aufs] [f8cecd1d] aufs_lookup+0x7d/0xd0 [aufs] [c0165660] do_lookup+0x130/0x170 Thnak you. It is much easier to debug. But this Call Trace is strange because of aufs_permission() between aufs_lookup() and au_new_inode(). - aufs_lookup() never call aufs_permission() - aufs_permission() never calls au_new_inode() I guess that you didn't enable CONFIG_FRAME_POINTER, so I want you to enable these configurations. - DEBUG_KERNEL - FRAME_POINTER - DEBUG_STACKOVERFLOW And please show me the output of this command (for the current aufs module) The result may depend on your compiler. $ objdump -d /your/current/aufs.ko | perl /your/kernel/src/scripts/checkstack.pl i386 Junjiro Okajima - Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.phpp=sourceforgeCID=DEVDEV
aufs Monday release
o bugfix - bugfix: introduce dcsub.[ch] to handle dentry cache. - bugfix: use readlink(1) to get cwd, in auplink, mount.aufs and unionctl, reported by Torsten Luettgert. Actually, to use readlink was suggested by Tomas M last year. o news - support initramfs. - verbose warning message for 'different uid/gid/perm.' - still testing udba=inotify. o misc - support GFP flags in au_kzrealloc(). - simplified lock/unlock functions definition by a new macro SimpleRWSem... Junjiro Okajima -- Index: fs/aufs/Makefile Index: fs/aufs/aufs.h - bugfix: introduce dcsub.[ch] to handle dentry cache. Index: fs/aufs/branch.c - support GFP flags in au_kzrealloc(). - verbose warning message for 'different uid/gid/perm.' - bugfix: introduce dcsub.[ch] to handle dentry cache. Index: fs/aufs/branch.h Index: fs/aufs/file.h Index: fs/aufs/inode.h Index: fs/aufs/super.h - simplified lock/unlock functions definition by a new macro SimpleRWSem... Index: fs/aufs/cpup.c Index: fs/aufs/debug.c Index: fs/aufs/debug.h Index: fs/aufs/export.c Index: fs/aufs/i_op_add.c Index: fs/aufs/i_op_del.c Index: fs/aufs/i_op_ren.c Index: fs/aufs/sysaufs.c Index: fs/aufs/wkq.c Index: include/linux/aufs_type.h Index: util/aufs.in.5 - less important changes Index: fs/aufs/dentry.c - support GFP flags in au_kzrealloc(). - still testing udba=inotify. Index: fs/aufs/f_op.c - bugfix: prevent from a race between file_list in remount. (this fix was removed temporary, since I could not test it well) Index: fs/aufs/file.c Index: fs/aufs/inode.c Index: fs/aufs/misc.c Index: fs/aufs/vdir.c - support GFP flags in au_kzrealloc(). Index: fs/aufs/hinotify.c - still testing udba=inotify. Index: fs/aufs/misc.h - simplified lock/unlock functions definition by a new macro SimpleRWSem... - support GFP flags in au_kzrealloc(). Index: fs/aufs/opts.c - revised simply. Index: fs/aufs/super.c - bugfix: introduce dcsub.[ch] to handle dentry cache. - revised simply. Index: sample/diskless/Makefile Index: sample/diskless/comount.sh Index: sample/diskless/diskless.6aufs Index: sample/diskless/initrd.mk Index: sample/diskless/linuxrc Index: sample/diskless/pxe.conf Index: README - support initramfs. Index: util/auplink Index: util/mount.aufs Index: util/unionctl - bugfix: use readlink(1) to get cwd, in auplink, mount.aufs and unionctl, reported by Torsten Luettgert. Actually, to use readlink was suggested by Tomas M last year. - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: aufs bug, my first real one
Tomas M: OK I modified kernel config to enabled this debug and I recompiled the kernel and aufs.ko as well. Then I reproduced the bug again, here are the results: - config: kernel config used (from /proc/config.gz) - stack.txt: objdump output - dmesg.log: bug report Thank you very much. Honestly speaking, I am still unsure what was wrong. But I guess your problem may be related to the today's bugfix which is for handling dcache. Please try the latest aufs. And I think this stack usage is too large, I will reduce it in next Monday release. 0x00010b96 aufs_rename [aufs]:508 0x000113fb aufs_rename [aufs]:508 Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
aufs Monday release
o bugfix - bugfix: file revalidation after rename(tgt). - strict inode lock in fsync() before 2.6.17. o news - restore rdacahe option, which was dropped by mistake. - print the depth of generic workqueue in sysfs/aufs/stat. - testing initramfs. - still testing udba=inotify. - simplified uid/gid/perm message. - reduce stack space. - make aufs_cond available even if CONFIG_LKTR is disabled. o misc - revise au_fin_finfo() and si lock. - simplified cache function definitions. - merge both of wait/nowait works into a single enqueue function. - rename wkq_wait() to au_wkq_wait(). - rename some name hash functions, and two new functions nash_new() and nhash_del(). + init_nhash() -- nhash_init() + move_nhash() -- nhash_move() + free_nhash() -- nhash_fin() - move au_is_kthread() to wkq.h, and rename to is_aufsd(). - rename au_init_wkq() to au_wkq_init(). - rename au_fin_wkq() to au_wkq_fin(). - rename au_inotify_exit() to au_inotify_fin(). - rename dbg_dlgt_exit() to dbg_dlgt_fin(). Junjiro Okajima -- Index: fs/aufs/branch.c - simplified uid/gid/perm message. Index: fs/aufs/cpup.c Index: fs/aufs/dentry.c Index: fs/aufs/i_op.c Index: fs/aufs/vfsub.c - rename wkq_wait() to au_wkq_wait(). Index: fs/aufs/debug.c Index: fs/aufs/debug.h - make aufs_cond available even if CONFIG_LKTR is disabled. Index: fs/aufs/dir.c - bugfix: file revalidation after rename(tgt). - revise au_fin_finfo() and si lock. - rename wkq_wait() to au_wkq_wait(). - reduce stack space. Index: fs/aufs/dir.h Index: fs/aufs/vdir.c - rename some name hash functions, and two new functions nash_new() and nhash_del(). + init_nhash() -- nhash_init() + move_nhash() -- nhash_move() + free_nhash() -- nhash_fin() Index: fs/aufs/f_op.c - revise au_fin_finfo() and si lock. - strict inode lock in fsync() before 2.6.17. Index: fs/aufs/file.c - rename wkq_wait() to au_wkq_wait(). - bugfix: file revalidation after rename(tgt). Index: fs/aufs/finfo.c - revise au_fin_finfo() and si lock. Index: fs/aufs/hinotify.c - still testing udba=inotify. - rename wkq_wait() to au_wkq_wait(). - rename au_inotify_exit() to au_inotify_fin(). Index: fs/aufs/i_op_del.c Index: fs/aufs/i_op_ren.c Index: fs/aufs/misc.c - reduce stack space. Index: fs/aufs/inode.h - rename au_inotify_exit() to au_inotify_fin(). Index: fs/aufs/misc.h - move au_is_kthread() to wkq.h, and rename to is_aufsd(). Index: fs/aufs/module.c - rename dbg_dlgt_exit() to dbg_dlgt_fin(). - rename au_init_wkq() to au_wkq_init(). - rename au_fin_wkq() to au_wkq_fin(). - rename au_inotify_exit() to au_inotify_fin(). Index: fs/aufs/module.h - move au_wkq to wkq.h. - simplified cache function definitions. Index: fs/aufs/opts.c - restore rdacahe option, which was dropped by mistake. - reduce stack space. Index: fs/aufs/plink.c - rename au_is_kthread() to is_aufsd(). - rename wkq_wait() to au_wkq_wait(). Index: fs/aufs/super.c - revised simply. - testing initramfs. Index: fs/aufs/sysaufs.c - print the depth of generic workqueue in sysfs/aufs/stat. Index: fs/aufs/whout.c - rename wkq_wait() to au_wkq_wait(). - rename some name hash functions. Index: fs/aufs/wkq.c - merge both of wait/nowait works into a single enqueue function. - print the depth of generic workqueue in sysfs/aufs/stat. - rename au_fin_wkq() to au_wkq_fin(). - rename au_init_wkq() to au_wkq_init(). - move au_is_kthread() to wkq.h, and rename to is_aufsd(). Index: fs/aufs/xino.c - revised simply. Index: util/aufs.in.5 - print the depth of generic workqueue in sysfs/aufs/stat. Index: README Index: fs/aufs/Makefile Index: fs/aufs/export.c Index: fs/aufs/iinfo.c Index: fs/aufs/inode.c Index: fs/aufs/vfsub.h Index: include/linux/aufs_type.h Index: sample/diskless/Makefile Index: sample/diskless/comount.sh Index: sample/diskless/def.mk Index: sample/diskless/initrd.mk Index: sample/diskless/linuxrc Index: sample/diskless/nfscd.mk - less important changes. - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: Want hack to prevent .wh.__dir_opaque creation
I guess your script which will be executed periodically will be such
like this (I didn't test it),
mount -o remount,udba=inotify /aufs
mv /rw/* /ro/*
mount -o remount,udba=reval /aufs
This sample was rough or rude.
You may need to care about whiteout, and the possibility of another
process may modify aufs.
Here is a modified sample.
You still need to decide how to handle whiteouts.
--
tmp=/rw/.wh..wh.$$
mount -o remount,udba=inotify /aufs
find /rw ! -name '.wh.*' \( -type d -fprintf $tmp.dir '%P\n' -o -fprintf
$tmp.nondir '%P\n' \)
cd /ro
xargs -r mkdir $tmp.dir
xargs -ri mv /rw/\{\} \{\} $tmp.nondir
rm $tmp.*
cd $OLDPWD
mount -o remount,udba=reval /aufs
--
Ideally, this script should be executed with locking-out all other
processes.
Junjiro Okajima
-
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
bugfix: stop consuming inode number rapidly.
As I announced in last Moday release, a minor bug is fixed which is the end of a revalidating problem. - bugfix: stop consuming inode number rapidly. Junjiro Okajima -- Index: fs/aufs/i_op_del.c Index: fs/aufs/i_op_ren.c - bugfix: stop consuming inode number rapidly. - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: aufs Monday release
Sandino Flores Moreno: ... on linux 2.6.21.1, I got this error: Sorry, I may be confused 2.6.21 with 2.6.22. Please try this patch. If it succeeds, I will update CVS. Junjiro Okajima -- Index: fs/aufs/opts.h === RCS file: /cvsroot/aufs/aufs/fs/aufs/opts.h,v retrieving revision 1.12 diff -u -p -r1.12 opts.h --- fs/aufs/opts.h 14 May 2007 03:40:27 - 1.12 +++ fs/aufs/opts.h 14 May 2007 06:12:52 - @@ -28,7 +28,7 @@ #include linux/version.h #include linux/aufs_type.h -#if LINUX_VERSION_CODE = KERNEL_VERSION(2,6,21) +#if LINUX_VERSION_CODE = KERNEL_VERSION(2,6,22) typedef const char* au_parser_pattern_t; #else typedef char* au_parser_pattern_t; Index: fs/aufs/sysaufs.h === RCS file: /cvsroot/aufs/aufs/fs/aufs/sysaufs.h,v retrieving revision 1.2 diff -u -p -r1.2 sysaufs.h --- fs/aufs/sysaufs.h 14 May 2007 03:40:13 - 1.2 +++ fs/aufs/sysaufs.h 14 May 2007 06:12:52 - @@ -27,7 +27,7 @@ #include linux/sysfs.h #include linux/version.h -#if LINUX_VERSION_CODE = KERNEL_VERSION(2,6,21) +#if LINUX_VERSION_CODE = KERNEL_VERSION(2,6,22) typedef struct kset au_subsys_t; #define au_subsys_to_kset(subsys) (subsys) #else - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: unionctl -- unionfs compatibility
Hi Ryan, Ryan Jud Hughes: It seems that aufs's unionctl wants a command like: unionctl /UNION --mode /foo ro whereas unionfs's wanted a command like: unionctl /UNION --mode ro /foo ::: Is there any reason not to do this? Because unionctl.c in unionfs-1.4.tar.gz shows us, fprintf(stderr, \tunionctl UNION --mode BRANCH (rw|ro|nfsro)\n); Unionfs-unionctl may accept the both cases. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: aufs working on 2.6.21.1 (x86-64)?
Michael Creel: Everything seems to go well, but when the CD is booted (actually, I use VMware to boot the image), there is a failure at the point that the aufs filesystem is mounted, leading to a failure bad /proc/mounts 1 message. ::: I am trying this using VMware, and I can't grab screen output to paste into a message, so I've made two screenshots showing what happens when booting. They are at http://pareto.uab.es/mcreel/pk64-aufs.png and http://pareto.uab.es/mcreel/pk64-aufs2.png In the first, you see that According to pk64-aufs2.png, the message failure bad /proc/mounts 1 is unrelated to your problem. Can you capture the entire message at kernel panic? Linux Tux is hiding them. :-) I want to know whether the aufs debug message was printed or not. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: Changing permissions in writable branch won't work sometimes
Hi, Wolfgang Barth: - Calling ntlm_auth (for AD authentication) from squid needs drwxr-x--- 2 root proxy so I only changed the group ownership and restarted all relevant processes. Do you mean that you changed it on the writable branch directly? If so, you need to use udba=inotify mount option. Why it works somtimes, because of your inode cache. It is recommended to change winbindd_privileged on aufs, instead of the writable branch directly. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
aufs Monday release
o misc - flush all the scheduled/nowait tasks at umouning and remounting. - refine the nowait task queuing. - support the enqueue error in workqueue. Junjiro Okajima -- Index: fs/aufs/cpup.c Index: fs/aufs/dentry.c Index: fs/aufs/dir.c Index: fs/aufs/file.c Index: fs/aufs/i_op.c Index: fs/aufs/plink.c Index: fs/aufs/vfsub.c Index: fs/aufs/xino.c - support the enqueue error in workqueue. Index: fs/aufs/hinotify.c Index: fs/aufs/whout.c - refine the nowait task queuing. - support the enqueue error in workqueue. Index: fs/aufs/super.c - flush all the scheduled/nowait tasks at umouning and remounting. Index: fs/aufs/wkq.c Index: fs/aufs/wkq.h - flush all the scheduled/nowait tasks at umouning and remounting. - refine the nowait task queuing. - support the enqueue error in workqueue. Index: fs/aufs/i_op_del.c Index: fs/aufs/i_op_ren.c Index: fs/aufs/module.c Index: fs/aufs/sbinfo.c Index: fs/aufs/super.h Index: fs/aufs/whout.h Index: include/linux/aufs_type.h - less important changes. - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: Changing permissions in writable branch won't work sometimes
Wolfgang Barth: No, I changed the permissions on aufs, as you recommended, not directly. If you can see the dir group id was changed expectedly and all the relevant processes were restarted correctly, then it must be an aufs bug. Will you show me the simple way to reproduce it? For example, how about to try strace to find out which process and which systemcall are returning the error? Thanks in advance. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: aufs bug
Hi, Wolfgang Barth: I can now reproduce my winbind problem in an easy case. Thank you. now is: % ls -ld /z-var/lib/xxx /DISK/var/lib/xxx /var/lib/xxx drwxr-x--- 2 root root 2048 May 21 11:16 /z-var/lib/xxx/ drwxr-x--- 2 root proxy 4096 May 21 13:37 /DISK/var/lib/xxx/ drwxr-x--- 2 root proxy 4096 May 21 13:37 /var/lib/xxx/ ::: [EMAIL PROTECTED]:~$ cd /var/lib/xxx -su: cd: /var/lib/xxx: Permission denied How about the permission bits of all parents? Please show me the results of, # ls -ld /var /var/lib /var/lib/xxx Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: aufs bug
Wolfgang Barth: % ls -ld /var /var/lib /var/lib/xxx drwxr-xr-x 26 root root 4096 May 21 11:56 /var/ drwxr-xr-x 32 root root 4096 May 21 13:48 /var/lib/ drwxr-x--- 2 root proxy 4096 May 21 13:37 /var/lib/xxx/ Hmm,,, it's a mystery. What is the shell of user proxy's? I want to take a look its source code. If you know where can I get the source files, please tell me too. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
aufs Monday release
o pre-announce As I had announced a few months ago, the unionctl script will not be supported. The script the description about it in the aufs manual will be moved from aufs_cvs/aufs/util to aufs_cvs/aufs/sample. And the symlink under aufs_cvs/aufs/ will not be created by removing it from local.mk. o news - support unmounting after 'mount --bind', update the internal vfsmount pointer when unmounting after linux-2.6.18. o misc - refine overlap-test at adding a branch. - refine dentry-d_parent with dget_parent() simply. - testing linux-vserver, new abstract function do_vfsub_unlink(). - rename DEBUG_ON macro to AuDebugOn. - add a note about the mode of a directory on the lower branch, suggested by Wolfgang Barth. - execute mount(8) in a simple way, it depsends on its version. Junjiro Okajima -- Index: fs/aufs/branch.c - rename DEBUG_ON macro to AuDebugOn. - refine overlap-test at adding a branch. Index: fs/aufs/cpup.c Index: fs/aufs/export.c Index: fs/aufs/f_op.c Index: fs/aufs/file.c Index: fs/aufs/i_op.c Index: fs/aufs/i_op_add.c Index: fs/aufs/i_op_del.c Index: fs/aufs/whout.c Index: fs/aufs/xino.c - rename DEBUG_ON macro to AuDebugOn. - refine dentry-d_parent with dget_parent() simply. Index: fs/aufs/dcsub.c - rename DEBUG_ON macro to AuDebugOn. - new function ac_dcsub_pages_rev(). Index: fs/aufs/dcsub.h - new function ac_dcsub_pages_rev(). Index: fs/aufs/debug.h Index: fs/aufs/dentry.c Index: fs/aufs/dinfo.c Index: fs/aufs/dir.c Index: fs/aufs/finfo.c Index: fs/aufs/hinotify.c Index: fs/aufs/i_op_ren.c Index: fs/aufs/iinfo.c Index: fs/aufs/inode.c Index: fs/aufs/inode.h Index: fs/aufs/misc.c Index: fs/aufs/misc.h Index: fs/aufs/module.c Index: fs/aufs/opts.c Index: fs/aufs/plink.c Index: fs/aufs/sbinfo.c Index: fs/aufs/sysaufs.c Index: fs/aufs/vdir.c Index: fs/aufs/wkq.c - rename DEBUG_ON macro to AuDebugOn. Index: fs/aufs/super.c - rename DEBUG_ON macro to AuDebugOn. - support unmounting after 'mount --bind', update the internal vfsmount pointer when unmounting after linux-2.6.18. Index: fs/aufs/vfsub.c Index: fs/aufs/vfsub.h - testing linux-vserver, new abstract function do_vfsub_unlink(). Index: util/aufs.in.5 - add a note about the mode of a directory on the lower branch, suggested by Wolfgang Barth Index: util/unionctl - execute mount(8) in a simple way, it depsends on its version. Index: Kconfig.in Index: README Index: fs/aufs/super.h Index: include/linux/aufs_type.h Index: sample/diskless/diskless.6aufs - less important changes. - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: aufs Monday release
Hi Wilhelm, Wilhelm Meier: FYI, aufs works without problems on linux-vserver with the patches I sent to the list some times ago. And it would be very good to see a clean solution for linux-vserver incorporated into aufs. I can guess supporting the linux-vserver interface (or function signature) is easy. But I am still considering about VSERVER_COWBL Enable COW Immutable Link Breaking configuration. Are you using with this enabled? Your patch added a new local macro VSERVER. I am replacing it by CONFIG_VSERVER which is introduced by linux-vserver. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
aufs Monday release
o news - the unionctl script is moved from util/ to sample/, and will not be supported any more. o bugfix - bugfix: support the moved root branch. o misc - testing is_subdir() kernel internal function. Junjiro Okajima -- Index: local.mk - testing is_subdir() kernel internal function. - the unionctl script is moved from util/ to sample/, and will not be supported any more. Index: fs/aufs/aufs.h Index: fs/aufs/branch.c Index: fs/aufs/dcsub.c Index: fs/aufs/dcsub.h Index: fs/aufs/dinfo.c Index: fs/aufs/sysaufs.c - testing is_subdir() kernel internal function. Index: fs/aufs/hinotify.c - bugfix: support the moved root branch. Index: fs/aufs/cpup.c Index: fs/aufs/debug.c Index: fs/aufs/debug.h Index: fs/aufs/dentry.c Index: fs/aufs/dir.c Index: fs/aufs/file.c Index: fs/aufs/finfo.c Index: fs/aufs/i_op.c Index: fs/aufs/i_op_add.c Index: fs/aufs/i_op_del.c Index: fs/aufs/i_op_ren.c Index: fs/aufs/iinfo.c Index: fs/aufs/inode.c Index: fs/aufs/inode.h Index: fs/aufs/misc.c Index: fs/aufs/misc.h Index: fs/aufs/module.c Index: fs/aufs/opts.c Index: fs/aufs/sbinfo.c Index: fs/aufs/super.c Index: fs/aufs/super.h Index: fs/aufs/sysaufs.h Index: fs/aufs/vdir.c Index: fs/aufs/vfsub.c Index: fs/aufs/whout.c Index: fs/aufs/wkq.c Index: fs/aufs/wkq.h Index: include/linux/aufs_type.h Index: util/aufs.in.5 - less important changes. - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: What to enable to corner a problem I have?
Hi, Sandino Flores Moreno: I'm using aufs to compile. ::: The embedded file systems lives in another aufs (so I can have all the produced or installed files in within the changes branch for that system). Do you mean you have two aufs mounted, like this? # mount -t aufs -o br:/rwA:/roA none /aufsA # mount -t aufs -o br:/rwB:/roB none /aufsB And you cound not see a file under /aufsA, but the file must exist there. If so, can you see the file directly on the branch filesystem? If the file is created directly (bypassing aufs) and you didn't use udba=inotify option, then you may not see the file. Refer to the aufs manual to know what UDBA is. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: Possible bug (?) - hard to reproduce
Jrgen_P._Tjern:
The thing that's acting up is a dir handle. I opendir() it, and I can
read fine from it. A bit later in the programs execution, and readdir()
returns NULL (right after I rewinddir()), and errno is not set.
So your application calls,
{
opendir();
readdir();
/* do something */
rewinddir();
readdir(); // returns NULL illegally
}
I'm trying to figure out what's wrong, and I'm starting to consider two
viable options: memory error or aufs. Could this be the doing of aufs?
Any way to debug it on a bit lower level? Any suggestions?
First, you need to check whether the second readdir(3) issues
getdents(2) or not. Try strace. Also check it arguments, especially the
i/o position in the dir.
If getdents(2) is called and the potision is 0 (ie. rewinded correctly),
then try aufs debug output such like this.
You need to enable CONFIG_AUFS_DEBUG. It will print debug messages to
syslog.
fs/aufs/dir.c:
static int aufs_readdir(struct file *file, void *dirent, filldir_t filldir)
{
au_debug_on();
;;;
au_debug_off();
}
Junjiro Okajima
-
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
Re: Possible bug (?) - hard to reproduce
You need to enable CONFIG_AUFS_DEBUG. It will print debug messages to syslog. One more note, You need to configure /etc/syslog.conf and /proc/sys/kernel/printk to receive kernel debug message. For example, # echo 8 /proc/sys/kernel/printk (/etc/syslog.conf) kern.debug /var/log/debug.log # restart your_syslog_daemon Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: System hangs after mounting with aufs
Hello Bertrand,
Bertrand D:
Working on an embedded projet, i planned on using aufs. The linux kernel I
use is based on a 2.6.20.7 (it was cross-compiled). It uses the ppc branch
and not the powerpc one. The kernel works great with different kinds of
filesystems. No stability issues. Except when i use aufs. Compiled as a
module or built-in the kernel, everytime i try to mount anything via aufs the
system hangs. No callstack appears on the console, no oops, nothing. I tried
with two different mount components (busybox and coreutils). I activated the
CONFIG_AUFS_DEBUG option, used sometimes the mount.aufs script, tried with or
without a mtab file. Nothing changed. It's stuck (and so am i). I tried the
20070604 version.
If you can use MagicSysRq key, try these keys and send me the output. It
will show where aufs stopped.
Or try this patch which will produce some debug logs at mount-time.
Also you need to receive kernel debug log.
MagicSysRq key
'd' - Shows all locks that are held.
'q' - Will dump a list of all running timers.
't' - Will dump a list of current tasks and their information to your
console.
'w' - Dumps tasks that are in uninterruptable (blocked) state.
--
Index: fs/aufs/super.c
===
RCS file: /cvsroot/aufs/aufs/fs/aufs/super.c,v
retrieving revision 1.53
diff -u -p -r1.53 super.c
--- fs/aufs/super.c 4 Jun 2007 02:17:35 - 1.53
+++ fs/aufs/super.c 13 Jun 2007 06:44:41 -
@@ -677,7 +677,7 @@ static int aufs_fill_super(struct super_
struct opts opts;
char *arg = raw_data;
- //au_debug_on();
+ au_debug_on();
if (unlikely(!arg || !*arg)) {
err = -EINVAL;
Err(no arg\n);
@@ -763,7 +763,7 @@ static int aufs_fill_super(struct super_
TraceErr(err);
err = cvt_err(err);
TraceErr(err);
- //au_debug_off();
+ au_debug_off();
return err;
}
--
To receive debug log
# echo 8 /proc/sys/kernel/printk
(/etc/syslog.conf)
kern.debug /var/log/debug.log
# restart your_syslog_daemon
Junjiro Okajima
-
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
Re: Permission bug?
Igor Karasynskyi: I'm use aufs as root filesystem in which I chrooted during boot and after this system run from there, my last branch is RW (index = 0) and all other is RO, on previous branch (index = 1) exist file /tmp/test (/tmp permissions is 777), owner of this file is common user, when I delete this file and try to execute shell command touch /tmp/test from user the command failed with this error: touch setting times of '/tmp/test': No such file or directory So your situation is, / = /rw + /ro $ ls /rw/tmp - no 'test' file $ ls /ro/tmp - 'test' file exists, 0777, normal user/group $ rm /tmp/test - succeeds - whiteout for 'test' is created. $ touch /tmp/test - unexpected error, No such file or directory Will you check these things? - after rm, check the whiteout by 'ls -l /rw/tmp/.wh.test' - check the dir permission by 'ls -ld / /rw /ro /rw/tmp /ro/tmp' - find the systemcall which returned the error by 'strace touch /tmp/test' If there is another user on your system and the branch filesystems (/rw and /ro in above example) are not hidden, someone else might change them directly (bypassing aufs). In this case, you should use 'udba=inotify' aufs mount option. Otherwise, aufs may behave differently as you expect. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: Permission bug?
Hi Igor,
Igor Karasynskyi:
Do you mean,,,
# mount -t aufs -o br:/rw:/ro none /aufs
# mount -o remount,mod:/rw=ro /aufs
and crashed?
I can't find information about bug in /var/log/*, so I try do
screenshot of this problem, on screenshot also steps that I do after
system boot.
So what you did is,
# mount -t aufs -o br:/rw1:/ro1:/ro2 none /aufs
# mount -o remount,prepend:/rw0 /aufs
# mount -o remount,mod:/rw1=ro /aufs
and crashed.
Won't you try this patch which produce more detailed debug information
at remount time, and send me the debug log?
You meight need to change these things to receive kernel debug log.
# echo 8 /proc/sys/kernel/printk
(/etc/syslog.conf)
kern.debug /var/log/debug.log
# restart your_syslog_daemon
serial-console, netconsole or remote-syslog are recommended too.
Junjiro Okajima
--
--- branch.c2007/06/14 11:49:17 1.56
+++ branch.c2007/06/15 00:49:11
@@ -798,13 +798,16 @@ int br_mod(struct super_block *sb, struc
/* no need file_list_lock() since sbinfo is locked */
//file_list_lock();
list_for_each_entry(file, sb-s_files, f_u.fu_list) {
+ au_debug_on();
LKTRTrace(%.*s\n, DLNPair(file-f_dentry));
+ DbgFile(file);
fi_read_lock(file);
if (!S_ISREG(file-f_dentry-d_inode-i_mode)
|| !(file-f_mode FMODE_WRITE)
|| fbstart(file) != bindex) {
FiMustNoWaiters(file);
fi_read_unlock(file);
+ au_debug_off();
continue;
}
@@ -812,8 +815,10 @@ int br_mod(struct super_block *sb, struc
hf = au_h_fptr(file);
hf-f_flags = au_file_roflags(hf-f_flags);
hf-f_mode = ~FMODE_WRITE;
+ //put_write_access(hf-f_dentry-d_inode);
FiMustNoWaiters(file);
fi_read_unlock(file);
+ au_debug_off();
}
//file_list_unlock();
--
-
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
Re: Permission bug?
Igor Karasynskyi: Yes you are right :) my first RO branch has 0777 and branches after that has 1777, this patch fixed this problem, thank you very much :) God news! About this problem with remount branch as RO: 1) I have image of Gentoo distributive (with 2.4 kernel) 2) after that I mounted branch in which I'm upgraded this image to 2.6 and other system software So as you see my system boot with 2.6 kernel with aufs and mount this braches, then chroot to them and start init. If I run system with upgraded software (to 2.6) all work's fine, but if I just boot system without (mounting branch with 2.6 upgrade) I receive error with remount, so problem occurred only in this situation. Do you mean that you are running aufs in 2.4 kernel? Aufs doesn't support linux-2.4. -- I apply your patch and repeat situation, netconsole log attached. Thank you for your test. There were sevral files are opened when you modify the branch permission. libc-2.3.4.so, libuuid.so.1.2, libblkid.so.1.0, ld-2.3.4.so, mount, libnss_files-2.3.4.so, libnss_nis-2.3.4.so, libnsl-2.3.4.so, libnss_compat-2.3.4.so, libc-2.3.4.so, libdl-2.3.4.so, ld-2.3.4.so, bash, vcsa2 The trigger was last 'vcsa2.' Is this file executable or something? And, most importantly, Is this file inside aufs? All other files existed in aufs and on the branch which was indexed as 2. Does vcsa2 exist on the same branch? (branches are indexed from 0. you can check them the line in /proc/mounts or /sys/fs/aufs/brs, br:br0:br1:br2:...) Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: Permission bug?
Igor Karasynskyi:
Won't you try this patch?
Didn't help, log attached
Sorry, I should write more.
Please remove the last debug patch which includes au_debug_on().
I make mistake, error return not mkswap but swapon
:::
stat64(/var/tmp/swap, {st_mode=S_IFREG|0644, st_size=10240, ...}) = 0
swapon(/var/tmp/swap) = -1 EINVAL (Invalid argument)
write(2, swapon: /var/tmp/swap: Invalid a..., 40swapon:
/var/tmp/swap: Invalid argument
) = 40
exit_group(-1) = ?
Currently it is an expected behaviour since I am doubtful about swap in
aufs. Is it really needed? As you wrote, you can mkswap on a branch
filesystem directly and the swap file is no need to be stackable.
Junjiro Okajima
-
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
Re: Permission bug?
Igor Karasynskyi: No it not really needed, it much more clear when swap located outside of aufs OK. Then I hope you had already read this description, too. (from the aufs manual) When you use aufs as root filesystem, it is recommended to consider to exclude some directories. For example, /tmp and /var/log are not need to stack in many cases. They don't usually need to copyup or to whiteout. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
aufs Monday release
o bugfix - bugfix: declare 'signed char', a portability problem reported by Bertrand D. - bugfix: strict branch index check at the first adding, reported by Bertrand D. - bugfix: unlink a whiteout under a dir who has a sticky bit, reported by Igor Karasynskyi. o misc - warn about overlapped branches when the top-dir is moved. - remove unnecessary dget_parent(), restoring a part of last ci. - testing 'imod' mount option. - add a note about an empty file in cramfs. - add a note about swapfile. Junjiro Okajima -- Index: fs/aufs/branch.c - bugfix: strict branch index check at the first adding time, reported by Bertrand D. Index: fs/aufs/hinotify.c - warn about overlapped branches when the top-dir is moved. Index: fs/aufs/i_op.c Index: fs/aufs/i_op_add.c Index: fs/aufs/i_op_del.c - remove unnecessary dget_parent(), restoring a part of last ci. Index: fs/aufs/opts.c - testing 'imod' mount option. Index: fs/aufs/vfsub.c - bugfix: unlink a whiteout under a dir who has a sticky bit, reported by Igor Karasynskyi. Index: fs/aufs/whout.c - remove unnecessary dget_parent(), restoring a part of last ci. - bugfix: unlink a whiteout under a dir who has a sticky bit, reported by Igor Karasynskyi. Index: include/linux/aufs_type.h - bugfix: declare 'signed char', a portability problem reported by Bertrand D. Index: util/aufs.in.5 - add a note about an empty file in cramfs. - add a note about swapfile. Index: fs/aufs/aufs.h Index: fs/aufs/branch.h Index: fs/aufs/cpup.c Index: fs/aufs/cpup.h Index: fs/aufs/debug.c Index: fs/aufs/debug.h Index: fs/aufs/dentry.c Index: fs/aufs/dentry.h Index: fs/aufs/dinfo.c Index: fs/aufs/dir.c Index: fs/aufs/dir.h Index: fs/aufs/export.c Index: fs/aufs/f_op.c Index: fs/aufs/file.c Index: fs/aufs/file.h Index: fs/aufs/i_op_ren.c Index: fs/aufs/iinfo.c Index: fs/aufs/inode.c Index: fs/aufs/inode.h Index: fs/aufs/misc.h Index: fs/aufs/module.c Index: fs/aufs/opts.h Index: fs/aufs/plink.c Index: fs/aufs/super.c Index: fs/aufs/super.h Index: fs/aufs/sysaufs.c Index: fs/aufs/sysaufs.h Index: fs/aufs/vdir.c Index: fs/aufs/vfsub.h Index: fs/aufs/wkq.c Index: fs/aufs/wkq.h Index: fs/aufs/xino.c Index: sample/watchguard/probe/probe.c Index: sample/watchguard/probe/probe.h - less important changes. - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: au_new_inode: broken ino
Jrgen_P._Tjern:
Branch 0 would be the first branch on the /proc/mounts-line? If so,
that's /vault/disk3, which is xfs. How do I find largest inode number?
Please send me the output of this script.
Set $path correctly before you execute it.
Junjiro Okajima
--
#!/bin/sh
path=/tmp
tmp=/tmp/$$
set -x
df -i $path
for i in 1 2
do
sudo find $path \
\( \( -name 3.Lbs \
-o -name libxtst6_1.0.1-3build1_i386.deb \
-o -name samba-common_3.0.22-1ubuntu4.2_i386.deb \) \
-fls $tmp.${i}a \) \
-o -printf '%i\n' |
sort -n $tmp.$i
done
diff -u $tmp.[12]a
diff -u $tmp.[12]
cat $tmp.1a
tail $tmp.1
egrep -w '(805306524|134217865|134217862)' $tmp.[12]
rm -f $tmp.*
-
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
Re: umount busy aufs
Hi, Tomas M: My goal is to properly save all filesystem modifications in the writable branch. So first I tried to cleanly unmount the union, but it's busy and can't be unmounted. Your aufs is the root filesystem, isn't it? Generally speaking, the root filesystem cannot be unmounted. It is remounted as readonly at shutdown/reboot time. How about this procedure instead of unmounting aufs? mount -no remount,ro /aufs for i in $writable_branches do mount -no remount,ro $i done exit If your xino files were on a hard drive, it is better to specify 'noxino' too. So, if the union can't be unmounted, I thought to remove the writable branch from it, using 'mount -o remount,del:/mnt/changes aufs /union'. Unfortunately this doesn't work either and there is no info in dmesg why that failed. That is strange. When you executed remount, was your syslogd up and running? The major reasons to reject deleting a branch are, - there is only one branch left in aufs. - a file or something is in use on the branch. - a directory is in use (which lsof may not show), and the directory doesn't exist on any other branch. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: au_new_inode: broken ino
Jrgen_P._Tjern: Branch 0 would be the first branch on the /proc/mounts-line? If so, that's /vault/disk3, which is xfs. How do I find largest inode number? Sorry, please add one line. #!/bin/sh path=/tmp tmp=/tmp/$$ set -x df -i $path for i in 1 2 do sudo mount -o remount $path sudo find $path \ \( \( -name 3.Lbs \ ::: Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: time delay before changes become visible
Hello Michael, Michael Towers: - I made a live cd using your latest larch scripts. - I booted into the live cd and did a pacman -Sl and pacman lists all packages in the repos, about 3925 packages. - I did a pacman -Syu and all my packages were up to date. Immediately after that I do a pacman -Sl again and nothing is listed anymore. - I wait about 1 minute and do a pacman -Sl and the 3925 packages are listed again. Thank you for your report. Are there any message left in your syslog? Does your packman access only aufs, or a branch filesystem directly? And please send me the output of 'strace pacman -Sl' which didn't show the list. I need to know which systemcall returned what error. aufs version 20070611 (but also other versions, I think an earlier one was worse) What was worse? Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: Hidden file bug
Hi, Igor Karasynskyi: after reboot and mounting aufs with new branch 4: br:4=rw:3=ro:2=ro:1=ro I don't see this file - OK, but if I do ls /install or cat /install then file exist. Try 3=ro+wh, instead of 3=ro. By the way, how about /dev/vcsa2 problem at remount,mod? I hope you have succeeded if you remove the debug patch and apply the last patch. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: umount busy aufs
Tomas M: I was just wondering if there is a way to see what resources (files/directories/etc) are used by aufs (perhaps through /proc or /sys). This would help very much. So please consider this as a feature request for aufs sometime in the future :) I have a plan to show the opened aufs files via sysfs. In your case, the cause can be the current working directory of a process. It is just referenced instead of opened. This debug patch prints 'busy' dentries which represents filenames when you invoke remount,del. If you don't get any filenames, then it means you hit other minor reason. Junjiro Okajima -- diff -x CVS -x RCS -x .pdiff -rup ../../aufs.anon/aufs/fs/aufs/branch.c ./fs/aufs/branch.c --- ../../aufs.anon/aufs/fs/aufs/branch.c 2007-06-18 11:43:34.0 +0900 +++ ./fs/aufs/branch.c 2007-06-18 22:07:56.069294432 +0900 @@ -592,7 +592,8 @@ static int test_children_busy(struct den err = -EBUSY; di_read_unlock(d, AUFS_I_RLOCK); if (err) - LKTRTrace(%.*s\n, DLNPair(d)); + //LKTRTrace(%.*s\n, DLNPair(d)); + Dbg(%.*s\n, DLNPair(d)); } } di_write_lock_child(root); /* aufs_write_lock() calls ..._child() */ - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: au_new_inode: broken ino
Jrgen_P._Tjern: I removed 3.Lbs as it's not on the fs any more. I ran your script with the changes, produced a lot of output. See the attached file. :-) Sorry, I should write more correctly. The path you set is /storage and it is aufs mount point, isn't it? It should be branch path. In your case, probably /vault/disk3 or something. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: au_new_inode: broken ino
Jrgen_P._Tjern:
I picked the most recent entry in my dmesg, and added that hi and fname
to the commands. (added 536871063 to the egrep)
Entry was:
[1081101.429555] aufs au_new_inode:325:find[31722]: broken ino, b0,
linux-meta/linux-image-generic_2.6.17.11_i386.deb, hi536871063, i50306.
Try udba=inotify.
Check attached file for output. :-)
I didn't know that xfs assigns such inode numbers.
Thank you for your tests.
Here is a debug patch to print more information when the message 'broken
ino' was produced. Please reproduce the message and send me the debug
log.
Junjiro Okajima
--
Index: fs/aufs/inode.c
===
RCS file: /cvsroot/aufs/aufs/fs/aufs/inode.c,v
retrieving revision 1.25
diff -u -p -r1.25 inode.c
--- fs/aufs/inode.c 18 Jun 2007 01:47:16 - 1.25
+++ fs/aufs/inode.c 19 Jun 2007 02:30:16 -
@@ -320,6 +320,20 @@ struct inode *au_new_inode(struct dentry
Warn1(broken ino, b%d, %s, %.*s, hi%lu, i%lu. Try udba=inotify.\n,
bstart, au_sbtype(h_dentry-d_sb), DLNPair(dentry), h_ino,
xino.ino);
+ {
+ static unsigned char c;
+ if (!c++) {
+ struct dentry *d = d_find_alias(inode);
+ au_debug_on();
+ DbgDentry(dentry);
+ DbgInode(inode);
+ if (d) {
+ DbgDentry(d);
+ dput(d);
+ }
+ au_debug_off();
+ }
+ }
xino.ino = 0;
err = xino_write0(sb, bstart, h_ino);
if (!err) {
-
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
Re: remount,del maybe doesn't free resources
Hello, Tomas M: But then, if I try to umount the branch (because the branch was a loop-mounted filesystem), sometimes I can see in the dmesg: VFS: Busy inodes after unmount of loop3, Self=destruct in 5 seconds. Have a nice day... ::: So my question is, does aufs completely free all the resources associated with the branch after 'remount,del' is issued? Or does it still use something? Basically aufs frees all the resources associated with the removed branch. If you didn't install /sbin/mount.aufs (and other scripts) and you didn't specify 'noplink' mount option, then unflushed pseudo-linked inode may still alive. If you know how to reproduce this problem, try first one of these, - install aufs scripts to /sbin - specify noplink Anyway aufs has something to fix. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: remount,del maybe doesn't free resources
Tomas M: I modified initrd in order to mount the entire aufs using 'noplink' mount option from the beginning. It doesn't help, there is still the same problem. What you did is, - mount -o remount,del:/squashfs /aufs - umount /squashfs - some operation - aufs crashes Please tell me how to reproduce this problem. I want to reproduce your environment on my pc. I tried the patch you sent me, but it does nothing, or I can't see any output in dmesg after remount,del, I'm sorry. And EBUSY error returned? Then it means, - one or more files were opened on that branch. or - there is only one branch left in aufs. Or can you receive the kernel debug log? Try, # echo 8 /proc/sys/kernel/printk (/etc/syslog.conf) kern.debug /var/log/debug.log # restart your_syslog_daemon Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: remount,del maybe doesn't free resources
Tomas M:
The 'umount /squashfs' part sometimes causes the error message
VFS: Busy inodes after unmount loop* ... etc, so I _think_ the loop
:::
About this problem, please test this patch.
Hm it's hard :) My problem is that I usually can't reproduce it in a
normal Linux environment, I'm still finding these things in Slax, which
is a chrooted OS and makes things hard to debug.
Nevertheless, you can download Slax from here
http://www.slax.org/dl/slax6broken.iso
I will try after your uploading completes.
I tried the patch you sent me, but it does nothing, or I can't see any
output in dmesg after remount,del, I'm sorry.
And EBUSY error returned?
Then it means,
- one or more files were opened on that branch.
This is possible but i'm unable to say, I don't know.
I hope the feature to see this info through /sys
will be implemented soon :)
Basically, /sys/fs/aufs//opened_files will be equivalent to lsof.
And then EBUSY is not depends upon only the opened files.
You wrote that lsof reported /dev/initctl was opened.
Are your sure that the device file is NOT inside aufs?
If a process (init or something) invoked before you over-mount /dev, the
process still refers the device file.
Junjiro Okajima
--
diff -x CVS -x RCS -x .pdiff -rup ../../aufs.anon/aufs/fs/aufs/opts.c
./fs/aufs/opts.c
--- ../../aufs.anon/aufs/fs/aufs/opts.c 2007-06-18 10:51:10.0 +0900
+++ ./fs/aufs/opts.c2007-06-19 15:24:33.679360152 +0900
@@ -764,7 +764,7 @@ static int au_do_opt_simple(struct super
* minus: error
*/
static int au_do_opt_br(struct super_block *sb, struct opt *opt, int remount,
- int *do_refresh)
+ int *do_refresh, int *deleted)
{
int err;
@@ -790,7 +790,7 @@ static int au_do_opt_br(struct super_blo
case Opt_idel:
err = br_del(sb, opt-del, remount);
if (!err)
- *do_refresh = err = 1;
+ *deleted = *do_refresh = err = 1;
break;
case Opt_mod:
@@ -906,7 +906,7 @@ static int verify_opts(struct super_bloc
int au_do_opts_mount(struct super_block *sb, struct opts *opts)
{
- int err, do_refresh;
+ int err, do_refresh, deleted;
struct inode *dir;
struct opt *opt;
unsigned int flags, given;
@@ -935,10 +935,11 @@ int au_do_opts_mount(struct super_block
au_flag_clr(sb, AuFlag_XINO | AuFlag_DLGT);
udba_set(sb, AuFlag_UDBA_REVAL);
- do_refresh = 0;
+ deleted = do_refresh = 0;
opt = opts-opt;
while (err = 0 opt-type != Opt_tail)
- err = au_do_opt_br(sb, opt++, /*remount*/0, do_refresh);
+ err = au_do_opt_br(sb, opt++, /*remount*/0, do_refresh,
+ deleted);
if (err 0)
err = 0;
else if (unlikely(err 0))
@@ -996,7 +997,7 @@ int au_do_opts_mount(struct super_block
}
int au_do_opts_remount(struct super_block *sb, struct opts *opts,
- int *do_refresh, unsigned int *given)
+ int *do_refresh, unsigned int *given, int *deleted)
{
int err, rerr;
struct inode *dir;
@@ -1012,7 +1013,7 @@ int au_do_opts_remount(struct super_bloc
//AuDebugOn(au_flag_test(sb, AuFlag_UDBA_INOTIFY));
err = 0;
- *do_refresh = 0;
+ *deleted = *do_refresh = 0;
*given = 0;
dlgt = au_flag_test(sb, AuFlag_DLGT);
opt_xino = NULL;
@@ -1025,7 +1026,8 @@ int au_do_opts_remount(struct super_bloc
au_flag_clr(sb, AuFlag_DLGT);
if (!err)
- err = au_do_opt_br(sb, opt, /*remount*/1, do_refresh);
+ err = au_do_opt_br(sb, opt, /*remount*/1, do_refresh,
+ deleted);
if (!err)
err = au_do_opt_xino(sb, opt, /*remount*/1, opt_xino);
diff -x CVS -x RCS -x .pdiff -rup ../../aufs.anon/aufs/fs/aufs/opts.h
./fs/aufs/opts.h
--- ../../aufs.anon/aufs/fs/aufs/opts.h 2007-06-18 10:51:10.0 +0900
+++ ./fs/aufs/opts.h2007-06-19 15:38:47.967488616 +0900
@@ -90,7 +90,7 @@ void au_free_opts(struct opts *opts);
int au_parse_opts(struct super_block *sb, char *str, struct opts *opts);
int au_do_opts_mount(struct super_block *sb, struct opts *opts);
int au_do_opts_remount(struct super_block *sb, struct opts *opts,
- int *do_refresh, unsigned int *given);
+ int *do_refresh, unsigned int *given, int *deleted);
#endif /* __KERNEL__ */
#endif /* __AUFS_OPTS_H__ */
diff -x CVS -x RCS -x .pdiff -rup ../../aufs.anon/aufs/fs/aufs/super.c
./fs/aufs/super.c
--- ../../aufs.anon/aufs/fs/aufs/super.c2007-06-18 10:53:21.0
+0900
+++ ./fs/aufs/super.c 2007-06-19 15:40:42.172126872 +0900
@@ -432,6 +432,66 @@
Re: remount,del maybe doesn't free resources
Tomas M: This was a different problem. Please consider our conversation about /dev/initctl resolved. I am working with two problems which are not related together: I am afraid you are confusing my two patches. :-) One is the problem of unmounting union (which probably doesn't work because init process is started from it). (please consider this resolved) The last debug patch is for this problem, which include Dbg(...) line. It just reports dentries in use. Second problem is I'm trying to remove branches while the system is running KDE, and here I get the problem with busy inodes on loop. The last patch which you are going to try is for this problem. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: au_new_inode: broken ino
Jrgen_P._Tjern: Here is the dmesg, it has only occured once after I applied the patch. I also updated to the newest aufs. :-) The machine isn't being used as regularly during summer, so use-pattern is a bit different now. :-) I hope this helps! Thank you very much for your tests. Currently I am considering about the bug in aufs inotify handler. Do you know whether this file 'Simon and Garfunkel ... .mp3' has ever been renamed or not? Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: au_new_inode: broken ino
Hello Alexey, Alexey Bazhin: I'm having same trouble, but i'm using udba=reval and have no modifications to branches or aufs at all... and i'm also using xfs... Thank you for your report. Then a bug probably lives outside inotify. If you know how to reproduce this problem, let me know. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
aufs Monday release
o current problem - isolated inode survived deleting a branch (still testing). - getdents(2) returns nothing. - cpup_wh_file() failure. - setting inode number which was previously assigned (broken ino msg). - hang at reboot/shutdown (it may not be an aufs problem). o bugfix - bugfix: skip non-aufs files at chaging the branch permission, reported by Igor Karasynskyi. - bugfix: test isolated inodes too at deleting branch, reported by Tomas M. - bugfix: call iput() for the isolated inode on the deleted branch, reported by Tomas M. - bugfix: detect newly appeared whiteout which exists on a inserted branch. - bugfix: refresh all dentries and inodes when adding/modifying a whiteout-able branch, and deleting a branch. o news - temporary workaround for the inode number 1 on cramfs. - note for discontigous inode numbers. o misc - refine option handling and some function arguments. - stop firing inotify when rename_whtmp(). - restore iigen after fired inotify when rename dir. - set 0 to bindex explicitly when prepending a branch. - testing cpup, xino, MagicSysRq key. Junjiro Okajima -- Index: fs/aufs/branch.c - bugfix: test isolated inodes too at deleting branch, reported by Tomas M. - bugfix: skip non-aufs files at chaging the branch permission, reported by Igor Karasynskyi. Index: fs/aufs/cpup.c - testing cpup bug. Index: fs/aufs/dentry.c - bugfix: detect newly appeared whiteout which exists on a inserted branch. Index: fs/aufs/i_op_del.c - stop firing inotify when rename_whtmp(). Index: fs/aufs/i_op_ren.c - restore iigen after fired inotify when rename dir. Index: fs/aufs/inode.c - bugfix: call iput() for the isolated inode on the deleted branch, reported by Tomas M. Index: fs/aufs/inode.h - bugfix: call iput() for the isolated inode on the deleted branch, reported by Tomas M. - restore iigen after fired inotify when rename dir. Index: fs/aufs/Makefile Index: fs/aufs/aufs.h Index: fs/aufs/module.c Index: fs/aufs/module.h Index: fs/aufs/sysaufs.c Index: fs/aufs/sysaufs.h - testing MagicSysRq key. Index: fs/aufs/opts.c - testing xino bitmap. - set 0 to bindex explicitly when prepending a branch. - refine option handling and some function arguments. - bugfix: refresh all dentries and inodes when adding/modifying a whiteout-able branch, and deleting a branch. Index: fs/aufs/opts.h - testing xino bitmap. - refine option handling and some function arguments. Index: fs/aufs/sbinfo.c - testing xino bitmap. Index: fs/aufs/super.c - testing xino bitmap. - bugfix: call iput() for the isolated inode on the deleted branch, reported by Tomas M. - bugfix: refresh all dentries and inodes when adding/modifying a whiteout-able branch, and deleting a branch. Index: fs/aufs/super.h - testing xino bitmap. - temporary workaround for the inode number 1 on cramfs. Index: fs/aufs/vdir.c - testing readdir. Index: fs/aufs/xino.c - testing xino bitmap. Index: util/aufs.in.5 - temporary workaround for the inode number 1 on cramfs. - note for discontigous inode numbers. Index: fs/aufs/file.c Index: fs/aufs/iinfo.c Index: include/linux/aufs_type.h Index: sample/diskless/comount.sh Index: sample/diskless/def.mk - less important changes. - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: Not all files are deleted
Hello Philippe, Philippe Malinge: I wrote a script (Cf. end of mail) creating 16384 files in a dummy directory (aufs directory), named dummy.3312, without problem. ::: It seems that each round deletes only the half of number files. If I do the same on standard NFS directory, no problem : all files and directory are removed at #1 round. Let me make sure, - on nfs client, you have mounted the nfs with 'proto=tcp' option. - it was nfs client where you executed the script and rm -fr. - nobody (both on nfs-server and nfs-client) was accessing the files. - when you export /diskless/cow/node2, the client succeeds rm -fr. Am I right? Just after you failed rm -fr, can you see the files named '.nfsXXX...?' Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: Not all files are deleted
Hi, Philippe Malinge: - it was nfs client where you executed the script and rm -fr. Yes, nfs client run : rm -fr /mnt/dummy.3312 ::: Just after you failed rm -fr, can you see the files named '.nfsXXX...?' No, but dummy.3312 was created on a read-only fs, and I try to remove it from /mnt, I can see /diskless/cow/node2/dummy.3312/.wh. files. The number of those files equals the number of removed files but not all as expected (rm -fr /diskless/cow/node2/dummy.3312). I think I am confused. Do you mean you executed the script on nfs server? to the RO branch? Or 'read-only fs' was just a typo? Were the '.nfsXXX...' files left under client:/mnt/dummy.3312? And do you mean that after rm -fr failure there left the whiteouts (dummy.3312/.wh.) and the regular files (dummy.3312/)? Won't you try strace to see the return value of unlink(2) to the children? Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: Not all files are deleted
Hi Philippe, Philippe Malinge: I guess below information will clarify the issue. Now we have common recognition that the target dir in on writable branch filesystem. And I need to ask you the same question again, And do you mean that after rm -fr failure there left the whiteouts (dummy.3312/.wh.) and the regular files (dummy.3312/)? Won't you try strace to see the return value of unlink(2) to the children? Shall we make these things clear by the output of strace? - rm(1) got all the entries under the dir. - rm(1) issued unlink(2) to all the gotten entries. - all unlink(2) succeeded. On my test environment, aufs on nfs-server can remove the dir who has 16384 children. Of course 'rm -fr' is executed on nfs-client. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: Not all files are deleted
Hi Philippe, Philippe Malinge: please find attached strace files for each round of rm -fr dummy.3312 executed on nfs-client side. strace files are named using the round number (1, 2, 3, 4, 5) Thank you for your test. I believe it is a known problem which was reported recently by a few people. Aufs readdir(3)/getdents(2) doesn't return the entry. There seems to be a time-lag until all the entries visible from aufs. Now I am preparing to dive into this problem. Please wait for a while. In your case, 'mount -o remount /your/aufs' just before 'rm -fr' may help you. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: RFC patch for porting aufs to rt kernel (2.6.21.5-rt17)
Tapani_Rikknen:
Andrew Burgess kirjoitti:
To get aufs to cpmpile under 2.6.21.5-rt17
I changed fs/aufs/misc.h from:
I just compiled 2.6.21.5 with rt18 patch =
(http://people.redhat.com/mingo/realtime-preempt/) WITHOUT CHANGING =
ANYTHING and was surpised that it compiled fine! I have never succeeded =
with -rt patch and unionfs or aufs before (perhaps leaving ISA support =
out helped).
But the read-write semaphore in RT patch doesn't support downgrade which
means changing write-lock to read-lock.
I'd suggest RT users to begin with inserting this to aufs.h. (I have
never tried RT patch)
This macros make aufs to use (standard) compat_rw_semaphore, instead of
rt_rw_semaphore.
#ifdef CONFIG_PREEMPT_RT
#define rw_semaphorecompat_rw_semaphore
#define init_rwsem compat_init_rwsem
#endif
CC [M] fs/aufs/branch.o
fs/aufs/branch.c: In function =91br_mod=92:
fs/aufs/branch.c:884: warning: initialization from incompatible pointer =
type
fs/aufs/branch.c:884: warning: comparison of distinct pointer types =
lacks a cast
Is your line 884 this?
list_for_each_entry(file, sb-s_files, f_u.fu_list) {
Junjiro Okajima
-
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
Re: remount,del maybe doesn't free resources
Tomas M: I didn't test the patch you sent me last time yet because I'm packing some stuff and I'm leaving to holidays for 10 days. So no need to hurry, take your time :) Hi Tomas, When you return from your vacation, please test the latest aufs (last Monday release). Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: au_new_inode: broken ino
Jrgen P. Tjern: [2297447.631266] aufs 20070702 [2297509.899071] aufs au_new_inode:347:smbd[27416]: Un-notified UDBA or directly renamed dir, b0, xfs, Simon - Garfunkel - The Best Of Simon - Garfunkel - Song For The Asking.mp3, hi1610612893, i16. It's not an access directly to the underlying fs, so how come this happens? Is this anything like hte broken_ino? It is equivalent to the old msg 'broken ino.' And as I wrote, it is harmless basically. I think I found the root cause. Probably current aufs cannot handle 'too large inode number', in your case 1610612893 on xfs. The aufs xino file is indexed by the inode number on branch fs. If the inode number is too large, it may exceed the file size limit. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: au_new_inode: broken ino
Jrgen P. Tjern: [2297447.631266] aufs 20070702 [2297509.899071] aufs au_new_inode:347:smbd[27416]: Un-notified UDBA or directly renamed dir, b0, xfs, Simon - Garfunkel - The Best Of Simon - Garfunkel - Song For The Asking.mp3, hi1610612893, i16. Please show me your /sys/fs/aufs/*/xino file, if it exists. And please try this patch. Junjiro Okajima -- Index: fs/aufs/xino.c === RCS file: /cvsroot/aufs/aufs/fs/aufs/xino.c,v retrieving revision 1.32 diff -u -p -r1.32 xino.c --- fs/aufs/xino.c 2 Jul 2007 05:11:13 - 1.32 +++ fs/aufs/xino.c 2 Jul 2007 12:58:23 - @@ -151,7 +151,8 @@ int xino_write(struct super_block *sb, a br = stobr(sb, bindex); AuDebugOn(!br || !br-br_xino); - pos = h_ino * sizeof(*xino); + pos = h_ino; + pos *= sizeof(*xino); sz = xino_fwrite(stosi(sb)-si_xwrite, br-br_xino, xino, sizeof(*xino), pos); //if (LktrCond) sz = 1; @@ -358,7 +359,8 @@ int xino_read(struct super_block *sb, au br = stobr(sb, bindex); file = br-br_xino; AuDebugOn(!file); - pos = h_ino * sizeof(*xino); + pos = h_ino; + pos *= sizeof(*xino); if (i_size_read(file-f_dentry-d_inode) pos + sizeof(*xino)) return 0; /* no ino */ - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: Not all files are deleted
Hello, Philippe Malinge: please find attached strace files for each round of rm -fr dummy.3312 executed on nfs-client side. strace files are named using the round number (1, 2, 3, 4, 5) Thank you for your test. I believe it is a known problem which was reported recently by a few people. I could reproduce your problem. If your 'rm' command version is less than 'rm (GNU coreutils) 6.9+', please try the latest 'rm.' I will try aufs will work on older rm, but it may be hard. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
aufs Monday release
o current problems
- setting inode number which was previously assigned (broken ino
msg) (testing).
- getdents(2) returns nothing. this may involve two problems. (it was
one. testing)
- cpup_wh_file() failure. (testing)
o bugfix
- bugfix: force rewind at re-initializing vdir, reported by Michael
Towers, Jorgen P. Tjerno and Sandino Flores Moreno.
- bugfix: copyup unlinked file.
+ bug #1, the file may or may not be opened, aufs needs to hold the
copied-up temporary dentry, reported by Andreas Baetz.
+ bug #2, the file may be opened in multiple times.
+ stop dropping in cpup_regular().
+ revert AuDebugOn() from temporary WARN_ON().
+ move cpup_wh_file() in file.c to cpup.c as generic cpup_wh().
+ print wh_dentry in iinfo.
+ set wh_dentry in iifno.
+ call cpup_wh() when the file was removed.
- bugfix: overflow in multiply, reported by Jorgen P. Tjerno and Alexey
Bazhin.
- bugfix: possible overflow in multiply.
o misc
- si_read/write_lock() supports AuLock_FLUSH which flushes the
remaining hinotify hanlder.
- maintain the number of 'nowait' tasks.
- testing where si_read_trylock() should be called.
- testing CONFIG_AUFS_ROBR
- ksize.patch will be unnecessary after linux-2.6.22, kzrealloc() calls
krealloc().
- extract a part of inode.h as a new file hinode.h.
- support lockdep at removing nfs branch.
- use hi_lock/unlock() for a hidden inodes, instead of i_lock/unlock().
- rename AUFS_[ID]_[RW]LOCK to AuLock_[ID][RW].
- rename i_lock/unlock() to vfsub_i_lock/unlock().
- rename filldir_ino_t to au_filldir_ino_t.
- rename AUFS_{DEBLK,NHASH}_SIZE to AuSize_{DEBLK,NHASH}.
- refine the message string.
- move au_hinotifyq to wkq.h as au_nowait_tasks, new inline functions
au_nwt_{init,inc,dec,flush}().
- move read/writef_t to super.h.
- rename find_sb_locked() to find_sb_lock().
- extract a part of this file as vserver.patch.
- consolidate arguments of do_diropq() as flags.
- consolidate arguments of au_wkq_run() as flags.
- testing FaunOS.
Junjiro Okajima
--
Index: Kconfig.in
Index: fs/aufs/module.c
- ksize.patch will be unnecessary after linux-2.6.22.
Index: vserver.patch
- forgot to check-in last week.
Index: fs/aufs/aufs.h
- introduce a new header file hinode.h.
Index: fs/aufs/branch.c
- support lockdep at removing nfs branch.
- use hi_lock/unlock() for a hidden inodes, instead of i_lock/unlock().
- testing CONFIG_AUFS_ROBR
- rename AUFS_[ID]_[RW]LOCK to AuLock_[ID][RW].
Index: fs/aufs/cpup.c
- bugfix: copyup unlinked file.
+ bug #1, the file may or may not be opened, aufs needs to hold the
copied-up temporary dentry, reported by Andreas Baetz.
+ bug #2, the file may be opened in multiple times.
+ stop dropping in cpup_regular().
+ revert AuDebugOn() from temporary WARN_ON().
+ move cpup_wh_file() in file.c to cpup.c as generic cpup_wh().
- use hi_lock/unlock() for a hidden inodes, instead of i_lock/unlock().
- rename AUFS_[ID]_[RW]LOCK to AuLock_[ID][RW].
Index: fs/aufs/cpup.h
- bugfix: copyup unlinked file.
+ move cpup_wh_file() in file.c to cpup.c as generic cpup_wh().
Index: fs/aufs/debug.c
- rename AUFS_NHASH_SIZE to AuSize_NHASH.
- bugfix: copyup unlinked file.
+ print wh_dentry in iinfo.
Index: fs/aufs/dentry.c
- testing CONFIG_AUFS_ROBR
- use hi_lock/unlock() for a hidden inodes, instead of i_lock/unlock().
- rename AUFS_[ID]_[RW]LOCK to AuLock_[ID][RW].
- si_read/write_lock() supports AuLock_FLUSH which flushes the
remaining hinotify hanlder.
Index: fs/aufs/dinfo.c
Index: fs/aufs/i_op_del.c
- rename AUFS_[ID]_[RW]LOCK to AuLock_[ID][RW].
Index: fs/aufs/dir.c
- si_read/write_lock() supports AuLock_FLUSH which flushes the
remaining hinotify hanlder.
- rename i_lock/unlock() to vfsub_i_lock/unlock().
- rename AUFS_[ID]_[RW]LOCK to AuLock_[ID][RW].
- rename filldir_ino_t to au_filldir_ino_t.
Index: fs/aufs/dir.h
- use typedef instead of macro.
- rename AUFS_{DEBLK,NHASH}_SIZE to AuSize_{DEBLK,NHASH}.
Index: fs/aufs/export.c
- rename filldir_ino_t to au_filldir_ino_t.
- rename i_lock/unlock() to vfsub_i_lock/unlock().
- rename AUFS_[ID]_[RW]LOCK to AuLock_[ID][RW].
- si_read/write_lock() supports AuLock_FLUSH which flushes the
remaining hinotify hanlder.
- testing CONFIG_AUFS_ROBR
Index: fs/aufs/f_op.c
- si_read/write_lock() supports AuLock_FLUSH which flushes the
remaining hinotify hanlder.
- rename AUFS_[ID]_[RW]LOCK to AuLock_[ID][RW].
- rename i_lock/unlock() to vfsub_i_lock/unlock().
- testing CONFIG_AUFS_ROBR
- use hi_lock/unlock() for a hidden inodes, instead of i_lock/unlock().
Index: fs/aufs/file.c
- use hi_lock/unlock() for a hidden inodes, instead of i_lock/unlock().
- si_read/write_lock() supports AuLock_FLUSH which flushes the
remaining hinotify hanlder.
- rename AUFS_[ID]_[RW]LOCK to AuLock_[ID][RW].
- bugfix: copyup unlinked file.
+ move cpup_wh_file() in file.c to cpup.c as generic cpup_wh().
+ set wh_dentry in iifno.
Index:
Re: umount busy aufs
Hello Tomas, Do you remember this mail? Tomas M: I was just wondering if there is a way to see what resources (files/directories/etc) are used by aufs (perhaps through /proc or /sys). This would help very much. So please consider this as a feature request for aufs sometime in the future :) Currently I am planning to print these things on memory. Do you think more things? - mounted superblock - root dentry - root inode - inode without dentry The last thing in the above list was the trigger of our previous problem, the inode which was set inotify by KDE. Generally, lsof command shows the opend files. So I don't think they are necessary here. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: bug in kernel 2.6.22 with removing branches
Hello Chris, chris rogers: I try removing a branch and it crash the kernel 2.6.22. So the kernel 2.6.22 should not be supported yet. Add branches have no problems. It works like normal. So i think its cause of the new mount option trunc_xib and notrunc_xib that this is happen. But if its only in kernel 2.6.22 as far as know. I will test aufs with my old kernel 2.6.21.5 to see if happens again. Thank you for your reports. I will build 2.6.22 on my pc, but it will take a time since my harddrive is almost full. Will you send me the kernel stack trace when it crashed? I hope this will get fix in the next week. I'll try, but I am not sure I can fix it. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: mount aufs test_add
Tomas M: As I can read aufs/branch.c, line 377, there is a test if inode-i_nlink ... nevertheless 'stat /mnt' outputs Links: 4. I have no idea if this information is relevant though :) Should the posixovl filesystem be fixed, or should aufs be fixed? I tried adding a branch which is mounted by FUSE_CVS/fuse/example/fusexmp.c, and succeeded. If you try fusexmp and it succeeds, then the posixovl should be fixed. Just a guess, executing 'stat /mnt' before mounting aufs may affect the result. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: mount aufs test_add
Tomas M: I tried adding a branch which is mounted by FUSE_CVS/fuse/example/fusexmp.c, and succeeded. If you try fusexmp and it succeeds, then the posixovl should be fixed. Thank you for the answer. I noticed that you may be misunderstanding what I wrote. It may depends upon the fuse version, kernel version or something else. If you try fusexmp or another fuse and it succeeds, then you can check the difference between the one and the posixovl. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: posixovl+aufs again
Tomas M: #!/bin/bash mkdir dir1 touch dir1/something mkdir union mount.posixovl dir1 mount -t aufs -o br:dir1=rw aufs union I guess you can succeed if you use dir1 directly without mount.posixovl. If we run 'mount.posixovl' with -d parameter (debug), we can see that nothing access the mounted directory dir1 at all during the aufs mount, GETATTR is not called, so it seems aufs can not know nlink properly. Getattr is totally different from lookup. Every filesystem should return correct inode as a result of lookup (when the target file exists). You should not confuse lookup with getattr. If you want to change the value of nlink, I guess what you should override is lookup instead of getattr. Why don't you use fuse_lowlevel_ops.lookup instead of fuse_operations? If we stat the dir1 and after then we mount aufs, it's OK. If customizing lookup is hard for you, to issue stat(2) before using posixovl as an aufs branch is the best solution. Jan suggests the following: Try replacing aufs/opts.c, near line 422: err = path_lookup(add-path, lkup_dirflags, add-nd); use lkup_dirflags | LOOKUP_REVAL. (One would have to add it to more places like branch deletion) Would you please consider Jan's suggestion? Do you mean LOOKUP_REVAL calls fuse getattr? I don't think it is correct. Did you try it actually by yourself? Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
aufs Monday release
o current problems - setting inode number which was previously assigned (broken ino msg) (fixed). - getdents(2) returns nothing. (fixed) - cpup_wh_file() failure. (fixed) o bugfix - bugfix: cpup whiteout which was called #2 in last ci. o news - begin supporting linux-2.6.22 + introduce lhash patch for 2.6.22. + no more ksize.patch. - introduce new branch permission 'rr.' + rename au_is_ro() to au_is_rr(). - introduce new (re)mount option, 'verbose' family, currenlty for deleting a branch only. - support MagicSysRq key for debugging. o misc - testing CONFIG_AUFS_ROBR. - testing new mount option for 'refreshing opened file' + a generation to branch. + typedef for every generation data. + stop packing branch id and si generation. + return of 'coo' option. - discard pre-allocated block for xino bitmap. - add a note about unmouting ore remounting root aufs at shutdown time. - minor optimization for handling udba=inotify. - rename hi_lock_XXX() to vfsub_i_lock_nested(). - rename hi_unlock() to vfsub_i_unlock(). - print inotify info for inode. - move lock subclasses from hinode.h to vfsub.h. Junjiro Okajima -- Index: Kconfig.in - testing CONFIG_AUFS_ROBR. - begin supporting linux-2.6.22 + introduce lhash patch for 2.6.22. Index: fs/aufs/aufs.h - begin supporting linux-2.6.22 + no more ksize.patch. Index: fs/aufs/branch.c - introduce new branch permission 'rr.' + rename au_is_ro() to au_is_rr(). - rename hi_lock_XXX() to vfsub_i_lock_nested(). - rename hi_unlock() to vfsub_i_unlock(). - testing new mount option for 'refreshing opened file' + a generation to branch. + typedef for every generation data. - introduce new (re)mount option, 'verbose' family, currenlty for deleting a branch only. Index: fs/aufs/branch.h - begin supporting linux-2.6.22 + introduce lhash patch for 2.6.22. - testing new mount option for 'refreshing opened file' + a generation to branch. - introduce new branch permission 'rr.' + rename au_is_ro() to au_is_rr(). Index: fs/aufs/cpup.c Index: fs/aufs/f_op.c Index: fs/aufs/i_op_ren.c - minor optimization for handling udba=inotify. - rename hi_lock_XXX() to vfsub_i_lock_nested(). - rename hi_unlock() to vfsub_i_unlock(). Index: fs/aufs/debug.c - support MagicSysRq key for debugging. - begin supporting linux-2.6.22 + no more ksize.patch. - print inotify info for inode. Index: fs/aufs/debug.h - support MagicSysRq key for debugging. - introduce new (re)mount option, 'verbose' family, currenlty for deleting a branch only. Index: fs/aufs/dentry.c - rename hi_lock_XXX() to vfsub_i_lock_nested(). - rename hi_unlock() to vfsub_i_unlock(). - minor optimization for handling udba=inotify. - testing new mount option for 'refreshing opened file' + typedef for every generation data. Index: fs/aufs/dentry.h Index: fs/aufs/file.h Index: fs/aufs/inode.h Index: fs/aufs/misc.h Index: fs/aufs/sbinfo.c - testing new mount option for 'refreshing opened file' + typedef for every generation data. Index: fs/aufs/dir.c Index: fs/aufs/i_op_add.c Index: fs/aufs/plink.c - rename hi_lock_XXX() to vfsub_i_lock_nested(). - rename hi_unlock() to vfsub_i_unlock(). Index: fs/aufs/export.c - testing new mount option for 'refreshing opened file' + typedef for every generation data. + stop packing branch id and si generation. Index: fs/aufs/file.c - minor optimization for handling udba=inotify. - testing new mount option for 'refreshing opened file' + return of 'coo' option. + typedef for every generation data. - rename hi_lock_XXX() to vfsub_i_lock_nested(). - rename hi_unlock() to vfsub_i_unlock(). - bugfix: cpup whiteout which was called #2 in last ci. Index: fs/aufs/hinode.h - move lock subclasses to vfsub.h. - rename hi_lock_XXX() to vfsub_i_lock_nested(). - rename hi_unlock() to vfsub_i_unlock(). Index: fs/aufs/i_op.c - rename hi_lock_XXX() to vfsub_i_lock_nested(). - rename hi_unlock() to vfsub_i_unlock(). - minor optimization for handling udba=inotify. - bugfix: cpup whiteout which was called #2 in last ci. Index: fs/aufs/i_op_del.c Index: fs/aufs/iinfo.c - minor optimization for handling udba=inotify. Index: fs/aufs/misc.c - begin supporting linux-2.6.22 + no more ksize.patch. - rename hi_lock_XXX() to vfsub_i_lock_nested(). - rename hi_unlock() to vfsub_i_unlock(). Index: fs/aufs/module.c - support MagicSysRq key for debugging. - begin supporting linux-2.6.22 + no more ksize.patch. Index: fs/aufs/module.h Index: fs/aufs/sysaufs.h - support MagicSysRq key for debugging. Index: fs/aufs/opts.c - introduce new (re)mount option, 'verbose' family, currenlty for deleting a branch only. - testing new mount option for 'refreshing opened file' + return of 'coo' option. - introduce new branch permission 'rr.' - minor optimization for handling udba=inotify. Index: fs/aufs/opts.h - testing new mount option for 'refreshing opened file' + return of 'coo' option. Index:
Re: posixovl+aufs again
I guess fuse lookup will be called if you override it. That's why I suggest it. Additionally, mount.posixovl needs to initialize its root inode, since the root inode is always in cache and lookup for it will not be called. Note: - 'the root inode' here means the root of posixovl fs. - in your case, at least, intializing root inode will solve the problem. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: posixovl+aufs again
Tomas M: Just for the case somebody is interested, here is the patch which should fix the problem in kernel: Interesting fixing. Will find(1) on this fuse complain about the dir nlink? Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: Inclusion in the kernel?
Hello Russell, Russell Harmon: I was just wondering, has anyone tried to get aufs included in the kernel? If not, why not? Currently, I don't have a plan to ask kernel people to include aufs. But someday I will try it. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: fuse+aufs
Miklos Szeredi: Basically yes. Accessing the file type in inode-i_mode is OK, but for all other fields getattr() needs to be called. If it is true, current linux lookup routines are totally broken since they check inode.i_mode so often without vfs_getattr(). They only check the file type. When you create/remove something, lookup operation involves accessing inode-i_uid, i_gid, permission bits in i_mode, i_ino and i_flags. For example, - lookup involves permission check which uses i_flags. i_flags is handled by ioctl generally. in this case, ioctl(GETFLAGS) or something is necessary too? - may_open() checks inode-i_uid. - unlink(2) and rmdir(2) check the sticky bit and inode-i_uid. - notify_change() which is called by chmod family is checking i_[ug]id. - find_inode_number() searches the cached inode by its name and returns i_ino. - vfs_quota_on() calls path_lookup() and then write back the gotten inode by vfs_quota_on_inode() and write_inode_now(). It involves referenccing and updating inode members. - while I don't know whether fuse supports NFS-export or not, nfsd depends upon inode-i_ino in many cases (before NFS GETATTR operation). Are they all wrong too? Or i_nlink is the exception? It wouldn't help. The problem is that fuse doesn't (and cannot) initialize the fields of the root inode at mount time. What I meant is, for example, to issue stat(2) just after fuse_main(). Is it unavialbe too? Note, that relying on cached attribute values is wrong anyway, in network filesystems the attributes can become stale and need to be refreshed before being used. Yes, that's why d_revalidate and FS_REVAL_DOT exists. But revalidating or refreshing doesn't involve getattr I think. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: fuse+aufs
Miklos Szeredi: But I could understand that you are still asserting getattr is necessary even in the cases of may_open() or something, and that is a VFS lookup bug. Am I right? Yes :) It is very hard for me that they are VFS lookup bug. If they are really VFS lookup bug, you or Tomas should ask for kernel people to fix them. That would just be papering over the real problem, a race would still be there, and the use of stale meta-data for network filesystems (fuse or non-fuse) would still be there. The network filesystems should have their dentry/inode caches and reuse them within the specified period as their lifetime. After the period expired, the dentry should be revalidated and lookup (and refresh its attribute) if needed. Within the period, the staleness will be allowed since they are remote. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: fuse+aufs
Miklos Szeredi: It is very hard for me that they are VFS lookup bug. Why? What has the sticky check or the O_NOATIME check to do with aufs? If it was VFS bug, it must be a generic problem, not specific to aufs. While you have mentioned about the race problem, the permission check in lookup is done without i_mutex/i_lock held. Do you think it is a bug too? But I gave up discussing about lookup and getattr with you. Now FUSE becomes very special thing to me and aufs. For users, aufs will call vfs_getattr in case of the branch is FUSE. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: fuse+aufs
Tomas M: I prefer to just mention this problem in aufs documentation, so people will 'stat' a fuse-based filesystem's mountpoint before adding it as aufs branch. It is not enough since Miklos thinks getattr is necessary for every VFS lookup. Of course, I don't agree. I don't understand whether FUSE does the right think or AUFS does it the right way, nevertheless I prefer to keep aufs code clean, that means, don't implement any workarounds for special filesystems. I am feeling it is time to handle FUSE specially. It will be rough or imcomplete and tiny handling. Because according to him, the bug is in VFS. It means a lower filesystem hardly fix it. Junjiro Okajima - This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/
Re: [fuse-devel] fuse+aufs
Miklos Szeredi: You misunderstand. What I think is: To be able to correctly perform permission checks based on cached inode attributes, those attributes may need to be refreshed before making the permission checks. You seem to be replacing the problem. The problem is more generic, not only permission check. What you wrote is, - in order to refer any inode attributes except file type in i_mode, the caller have to update the inode. - the caller must not rely on the inode attributes in a positive dentry which is returned by path_lookup(). isn't it? The two cases you mentioned (sticky bit and may_open()) is a part of the list I wrote as example where VFS accesses inode attributes. And I don't have a paticular opinion about judging those are bugs. I already decided to add a special handling for fuse into aufs. Because you wrote, - you are the one who expected to fix VFS about this issue. - you have another thing more important. Let's take a concrete example: ::: Should that rmdir succeed? Surely not, the sticky check would fail. But if the stale attributes are used, than it _will_ succeed. I thought we have already agreed about the staleness of cached nfs inode including REVAL_DOT. Junjiro Okajima - This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2005. http://clk.atdmt.com/MRT/go/vse012070mrt/direct/01/
Re: fuse+aufs
Tomas M: Now FUSE becomes very special thing to me and aufs. For users, aufs will call vfs_getattr in case of the branch is FUSE. I prefer to just mention this problem in aufs documentation, so people will 'stat' a fuse-based filesystem's mountpoint before adding it as aufs branch. Tomas, I have implemented a workaround for fuse in today's aufs release. While it is still testing on my pc, will you try enabling CONFIG_AUFS_WORKAROUND_FUSE and add your fuse-fs as an aufs branch? Thanx in advance. Junjiro Okajima - This SF.net email is sponsored by: Splunk Inc. Still grepping through log files to find problems? Stop. Now Search log events and configuration files using AJAX and a browser. Download your FREE copy of Splunk now http://get.splunk.com/
Re: Makefile?
It is better to post it to the ML since I think who needs this makefile is aufs users instead of me. Oh, I didn't realize that his mail was sent to the ML. Sorry. Junjiro Okajima - This SF.net email is sponsored by: Splunk Inc. Still grepping through log files to find problems? Stop. Now Search log events and configuration files using AJAX and a browser. Download your FREE copy of Splunk now http://get.splunk.com/
