Hello Junjiro, first of all, many thanks for your terrific support.
On Thursday 30 June 2011, 14:06:01 sf...@users.sourceforge.net wrote: > Hello Pete, > > sf...@users.sourceforge.net: > > Thanks for testing. > > I think I could find a bug in the patch. Its first hunk is bad. > > I will try another approach. Just give me some time. > > Here are two patches for a new approach. > - a.patch is for upstream kernel (opensuse-11.1-31.14) > - b.patch is for latest aufs2.1-standalone-31 > They are passed my first level local tests. > > A new configuration CONFIG_AUFS_PROC_MAP is introduced, but it may be > unnecessary for you. I'd suggest you to try these steps. > > 1. apply b.patch, disable CONFIG_AUFS_PROC_MAP, build and run. > > I hope the known problem will be gone by b.patch. But I am not sure > since I don't have an exact reproducible test method. At this point, I tried hard to trick aufs into the issue again, but failed so far ;) > If everything is fine, then following steps will be unnecessary for you. > Some utilities like lsof(1) may produce incorrect result. If you never > allow such behaviour, then try next. Well, non functional proc map and mmap support is what brought be to aufs1 (away from unionfs way back in time as they caused issues with system initialization and printing, IIRC), therefore I continued.. > 2. apply a.patch, disable CONFIG_AUFS_PROC_MAP, build and run. > > This is just a test to make sure that a.patch is good. So the problem > around lsof won't be solve. Then last step, > > 3. enable CONFIG_AUFS_PROC_MAP, build and run. > > This new configuration should solve the problem around lsof. First tests seem to confirm, that the bug is still solved with CONFIG_AUFS_PROC_MAP enabled (checked that twice), however lsof output looks strange. This is from aufs2 (old config): COMMAND PID USER FD TYPE DEVICE SIZE/OFF NODE NAME init 1 root cwd DIR 0,21 4096 2 / init 1 root rtd DIR 0,21 4096 2 / init 1 root txt REG 0,21 794912 12 /sbin/init init 1 root 10u FIFO 0,16 0t0 5711 /dev/initctl This is aufs2.1 with both patches and CONFIG_AUFS_PROC_MAP enabled: COMMAND PID USER FD TYPE DEVICE SIZE/OFF NODE NAME init 1 root cwd DIR 0,21 4096 2 / init 1 root rtd DIR 0,21 4096 2 / init 1 root txt REG 0,21 794912 12 /sbin/init init 1 root mem REG 0,19 794912 10478715 /read-only/sbin/init (172.16.23.110:/work/dlbase/11.1) init 1 root 10u FIFO 0,16 0t0 5841 /dev/initctl Interestingly, lsof has grown a new mem fd entry, that points to the read only aufs branch on NFS. Is that to be expected? My kernel and aufs2.1 builds are still available on build service for you to check: https://build.opensuse.org/project/monitor?project=home%3Afrispete%3Akernel-2.6.31 BTW, this patch obsoleted the adjustments of vm_operations_struct from the beginning of this thread. Is that correct? > Now my local tests continue. Same here.. Looks good so far. Great. Many thanks again, Pete ------------------------------------------------------------------------------ All of the data generated in your IT infrastructure is seriously valuable. Why? It contains a definitive record of application performance, security threats, fraudulent activity, and more. Splunk takes this data and makes sense of it. IT sense. And common sense. http://p.sf.net/sfu/splunk-d2d-c2
