Description of problem: multipathd doesn't reconfigure map when path deletion event occurs if selinux is enabled. Version-Release number of selected component: device-mapper-multipath-0.4.7-5.2 selinux-policy-2.4.3-8.el5 kernel-2.6.18-1.2747.el5 device-mapper-1.02.12-2.el5 How reproducible: Always Steps to Reproduce: 1. Enable selinux 2. Create multipath map # multipath 3. Start multipathd # /etc/init.d/multipathd start 4. Delete one path from the multipath map (Assume that the multipath map consists of /dev/sdg and /dev/sdh.) # echo 1 > /sys/block/sdh/device/delete Actual results: /dev/sdh isn't removed from the multipath map. ------------------------------------------------------------------------ [root@nec-tx7-1 ~]# multipath create: disk2 (1NEC_iStorage_2000_000000092680024600002) NEC,iStorage 2000 [size=67G][features=0][hwhandler=0] \_ round-robin 0 [prio=1][undef] \_ 4:0:0:2 sdg 8:96 [undef][ready] \_ round-robin 0 [prio=1][undef] \_ 5:0:0:2 sdh 8:112 [undef][ready] [root@nec-tx7-1 ~]# /etc/init.d/multipathd start Starting multipathd daemon: [ OK ] [root@nec-tx7-1 ~]# echo 1 > /sys/block/sdh/device/delete [root@nec-tx7-1 ~]# dmsetup table disk2: 0 139841536 multipath 0 0 2 1 round-robin 0 1 1 8:96 1000 round-robin 0 1 1 8:112 1000 [root@nec-tx7-1 ~]# dmsetup status disk2: 0 139841536 multipath 1 0 0 2 1 A 0 1 0 8:96 A 0 E 0 1 0 8:112 F 1 [root@nec-tx7-1 ~]# ------------------------------------------------------------------------ Expected results: multipathd should remove /dev/sdh from the multipath map. Additional info: "multipathd -v3" log messages when the path deletion event occurs are below. multipathd seems not to receive uevent. When selinux is disabled, map reconfigure to delete the path works fine. ------------------------------------------------------------------------ Nov 16 14:28:22 nec-tx7-1 multipathd: sdg: mask = 0x8 Nov 16 14:28:22 nec-tx7-1 multipathd: sdg: prio = 1 Nov 16 14:28:22 nec-tx7-1 multipathd: sdh: readsector0 checker reports path is down Nov 16 14:28:22 nec-tx7-1 multipathd: checker failed path 8:112 in map disk2 Nov 16 14:28:22 nec-tx7-1 kernel: scsi 5:0:0:2: rejecting I/O to dead device Nov 16 14:28:22 nec-tx7-1 kernel: device-mapper: multipath: Failing path 8:112. Nov 16 14:28:22 nec-tx7-1 multipathd: disk2: remaining active paths: 1 Nov 16 14:28:22 nec-tx7-1 multipathd: disk2: devmap event #2 Nov 16 14:28:22 nec-tx7-1 multipathd: disk2: discover Nov 16 14:28:22 nec-tx7-1 multipathd: disk2: rr_weight = 1 (internal default) Nov 16 14:28:22 nec-tx7-1 multipathd: disk2: pgfailover = -1 (internal default) Nov 16 14:28:22 nec-tx7-1 multipathd: disk2: no_path_retry = NONE (internal default) Nov 16 14:28:22 nec-tx7-1 multipathd: pg_timeout = NONE (internal default) Nov 16 14:28:27 nec-tx7-1 kernel: scsi 5:0:0:2: rejecting I/O to dead device Nov 16 14:28:27 nec-tx7-1 multipathd: sdh: readsector0 checker reports path is down Nov 16 14:28:27 nec-tx7-1 multipathd: sdh: mask = 0x8 Nov 16 14:28:32 nec-tx7-1 kernel: scsi 5:0:0:2: rejecting I/O to dead device Nov 16 14:28:32 nec-tx7-1 multipathd: sdh: readsector0 checker reports path is down Nov 16 14:28:32 nec-tx7-1 multipathd: sdh: mask = 0x8 ------------------------------------------------------------------------
I'm not sure if it was the fixes for 215973, or some other changes to the SELinux policy, but with the latest selinux policies, I cannot see this. Can you check as see if this is resolved? You may need to run restorecon /var/lib/multipath if the context for it is incorrect. It should be: [root@cypher-01 SPECS]# ls -Z /var/lib/ | grep multipath drwxr-xr-x root root system_u:object_r:lvm_var_lib_t multipath My fix for 215973 should keep people from having to do this.
Ben, with the following procedure, the problem seems disappeared. I guess if the new selinux-policy is included in the distribution, 'rm /var/run/multipathd.sock' isn't needed, right? I applied selinux-policy update. # rpm -Uvh /root/selinux-policy-*-2.4.5-4.el5.noarch.rpm # ls -Zd /var/lib/multipath drwx------ root root system_u:object_r:lvm_var_lib_t /var/lib/multipath Then, remove existing /var/run/multipathd.sock. # ls -Z /var/run/multipathd.sock srwxrwxrwx root root root:object_r:var_run_t /var/run/multipathd.sock Start the daemon. # /etc/init.d/multipathd start Starting multipathd daemon: [ OK ] # ls -Z /var/run/multipathd.sock srwxrwxrwx root root root:object_r:lvm_var_run_t /var/run/multipathd.sock # multipath -l disk2 disk2 (1NEC_iStorage_2000_000000092680024600002) dm-2 NEC,iStorage 2000 [size=67G][features=1 queue_if_no_path][hwhandler=0] \_ round-robin 0 [prio=0][active] \_ 5:0:0:2 sdg 8:96 [active][undef] \_ 4:0:0:2 sdh 8:112 [active][undef] # echo 1 > /sys/block/sdh/device/delete [root@nec-tx7-1 ~]# multipath -l disk2 disk2 (1NEC_iStorage_2000_000000092680024600002) dm-2 NEC,iStorage 2000 [size=67G][features=1 queue_if_no_path][hwhandler=0] \_ round-robin 0 [prio=0][enabled] \_ 5:0:0:2 sdg 8:96 [active][undef] Before the multipathd.sock wasn't removed, multipathd was killed after start up with the following syslog message: Dec 1 11:19:14 nec-tx7-1 setroubleshoot: SELinux is preventing /sbin/multipathd (lvm_t) "unlink" to multipathd.sock (var_run_t). See audit.log for complete SELinux messages. id = 002b8fa3-dbea-492a-a12d-358defb9d83d and the map was not updated after the path failure.
Yes. since the policy is included from the start, multipathd should always create multipathd.sock with the correct context.