• Tejun Heo's avatar
    blkcg: relocate root_blkg setting and clearing · 577cee1e
    Tejun Heo authored
    Hello, Jens.
    The original thread can be read from
    While it leads to oops, given that it only triggers under specific
    configurations which aren't common.  I don't think it's necessary to
    backport it through -stable and merging it during the coming merge
    window should be enough.
    ----- 8< -----
    Currently, q->root_blkg and q->root_rl.blkg are set from
    blkcg_activate_policy() and cleared from blkg_destroy_all().  This
    doesn't necessarily coincide with the lifetime of the root blkcg_gq
    leading to the following oops when blkcg is enabled but no policy is
    activated because __blk_queue_next_rl() malfunctions expecting the
    root_blkg pointers to be set.
      BUG: unable to handle kernel NULL pointer dereference at           (null)
      IP: [<ffffffff810c58cb>] __wake_up_common+0x2b/0x90
      PGD 60f7a9067 PUD 60f4c9067 PMD 0
      Oops: 0000 [#1] SMP DEBUG_PAGEALLOC
      gsmi: Log Shutdown Reason 0x03
      Modules linked in: act_mirred cls_tcindex cls_prioshift sch_dsmark xt_multiport iptable_mangle sata_mv elephant elephant_dev_num cdc_acm uhci_hcd ehci_hcd i2c_d
      CPU: 9 PID: 41382 Comm: iSCSI-write- Not tainted 3.11.0-dbg-DEV #19
      Hardware name: Intel XXX
      task: ffff88060d16eec0 ti: ffff88060d170000 task.ti: ffff88060d170000
      RIP: 0010:[<ffffffff810c58cb>] [<ffffffff810c58cb>] __wake_up_common+0x2b/0x90
      RSP: 0000:ffff88060d171818  EFLAGS: 00010096
      RAX: 0000000000000082 RBX: ffff880baa3dee60 RCX: 0000000000000000
      RDX: 0000000000000000 RSI: 0000000000000003 RDI: ffff880baa3dee60
      RBP: ffff88060d171858 R08: 0000000000000000 R09: 0000000000000000
      R10: 0000000000000000 R11: 0000000000000002 R12: ffff880baa3dee98
      R13: 0000000000000003 R14: 0000000000000000 R15: 0000000000000003
      FS:  00007f977cba6700(0000) GS:ffff880c79c60000(0000) knlGS:0000000000000000
      CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
      CR2: 0000000000000000 CR3: 000000060f7a5000 CR4: 00000000000007e0
       0000000000000082 0000000000000000 ffff88060d171858 ffff880baa3dee60
       0000000000000082 0000000000000003 0000000000000000 0000000000000000
       ffff88060d171898 ffffffff810c7848 ffff88060d171888 ffff880bde4bc4b8
      Call Trace:
       [<ffffffff810c7848>] __wake_up+0x48/0x70
       [<ffffffff8131da53>] __blk_drain_queue+0x123/0x190
       [<ffffffff8131dbb5>] blk_cleanup_queue+0xf5/0x210
       [<ffffffff8141877a>] __scsi_remove_device+0x5a/0xd0
       [<ffffffff81418824>] scsi_remove_device+0x34/0x50
       [<ffffffff814189cb>] scsi_remove_target+0x16b/0x220
       [<ffffffff814210f1>] __iscsi_unbind_session+0xd1/0x1b0
       [<ffffffff814212b2>] iscsi_remove_session+0xe2/0x1c0
       [<ffffffff814213a6>] iscsi_destroy_session+0x16/0x60
       [<ffffffff81423a59>] iscsi_session_teardown+0xd9/0x100
       [<ffffffff8142b75a>] iscsi_sw_tcp_session_destroy+0x5a/0xb0
       [<ffffffff81420948>] iscsi_if_rx+0x10e8/0x1560
       [<ffffffff81573335>] netlink_unicast+0x145/0x200
       [<ffffffff815736f3>] netlink_sendmsg+0x303/0x410
       [<ffffffff81528196>] sock_sendmsg+0xa6/0xd0
       [<ffffffff815294bc>] ___sys_sendmsg+0x38c/0x3a0
       [<ffffffff811ea840>] ? fget_light+0x40/0x160
       [<ffffffff811ea899>] ? fget_light+0x99/0x160
       [<ffffffff811ea840>] ? fget_light+0x40/0x160
       [<ffffffff8152bc79>] __sys_sendmsg+0x49/0x90
       [<ffffffff8152bcd2>] SyS_sendmsg+0x12/0x20
       [<ffffffff815fb642>] system_call_fastpath+0x16/0x1b
      Code: 66 66 66 66 90 55 48 89 e5 41 57 41 89 f7 41 56 41 89 ce 41 55 41 54 4c 8d 67 38 53 48 83 ec 18 89 55 c4 48 8b 57 38 4c 89 45 c8 <4c> 8b 2a 48 8d 42 e8 49
    Fix it by moving r->root_blkg and q->root_rl.blkg setting to
    blkg_create() and clearing to blkg_destroy() so that they area
    initialized when a root blkg is created and cleared when destroyed.
    Reported-and-tested-by: default avatarAnatol Pomozov <anatol.pomozov@gmail.com>
    Signed-off-by: default avatarTejun Heo <tj@kernel.org>
    Signed-off-by: default avatarJens Axboe <axboe@kernel.dk>