8556395.925003] XFS (sdi): xfs_log_force: error -5 returned.[8556407.425047] INFO: task umount:19286 blocked for more than 1200 seconds.----这个默认是120，该环境上是被人为设置1200[8556407.425653] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.[8556407.426050] umount          D ffff881fffdd47c0     0 19286  19283 0x00000080------------------------19286是pid，19283是ppid[8556407.426056]  ffff883e39587d80 0000000000000082 ffff883e39587fd8 ffff883e39587fd8[8556407.426066]  ffff883e39587fd8 00000000000147c0 ffff881fd3260000 ffff881f860fec80[8556407.426072]  ffff883ed79e1a20 ffff881f860fecc0 ffff881f860fece8 ffff881f860fec90------------这12个64位的值为栈里的值[8556407.426078] Call Trace:[8556407.426092]  [
     
      ] schedule+0x29/0x70[8556407.426137]  [
      
       ] xfs_ail_push_all_sync+0xc1/0x110 [xfs][8556407.426143]  [
       
        ] ? wake_up_atomic_t+0x30/0x30[8556407.426165]  [
        
         ] xfs_unmountfs+0x68/0x190 [xfs][8556407.426186]  [
         
          ] ? xfs_mru_cache_destroy+0x6b/0x90 [xfs][8556407.426204]  [
          
           ] xfs_fs_put_super+0x32/0x90 [xfs][8556407.426211] [
           
            ] generic_shutdown_super+0x56/0xe0-------已经执行完fsnotify_umount_inodes[8556407.426215] [
            
             ] kill_block_super+0x27/0x70[8556407.426218] [
             
              ] deactivate_locked_super+0x49/0x60[8556407.426222] [
              
               ] deactivate_super+0x46/0x60[8556407.426229] [
               
                ] mntput_no_expire+0xc5/0x120[8556407.426234] [
                
                 ] SyS_umount+0x15b/0x5b0[8556407.426239] [
                 
                  ] system_call_fastpath+0x16/0x1b[8556407.426242] sending NMI to all CPUs:
                 
                
               
              
             
            
           
          
         
        
       
      
     

对应的函数为：

/* * Push out all items in the AIL immediately and wait until the AIL is empty. */voidxfs_ail_push_all_sync(    struct xfs_ail  *ailp){    struct xfs_log_item    *lip;    DEFINE_WAIT(wait);    spin_lock(&ailp->xa_lock);    while ((lip = xfs_ail_max(ailp)) != NULL) {        prepare_to_wait(&ailp->xa_empty, &wait, TASK_UNINTERRUPTIBLE);        ailp->xa_target = lip->li_lsn;        wake_up_process(ailp->xa_task);        spin_unlock(&ailp->xa_lock);        schedule();        spin_lock(&ailp->xa_lock);    }    spin_unlock(&ailp->xa_lock);    finish_wait(&ailp->xa_empty, &wait);}

 排查的时候走了弯路，因为看的是：

crash> dis -l xfs_ail_push_all_sync+0xc1/usr/src/debug/kernel-3.10.0-327.22.2.el7/linux-3.10.0-327.22.2.el7.x86_64/include/linux/spinlock.h: 2930xffffffffa07c6cd1 
     
      : mov    %r13,%rdi
     

所以想当然地认为在第一个 获取spinlock的地方，

spin_lock(&ailp->xa_lock);---------以为在这hung住    while ((lip = xfs_ail_max(ailp)) != NULL) {

然后查找这把锁：

crash>  struct super_block.s_fs_info ffff883f7fe67800  s_fs_info = 0xffff883f7fe67000crash> mod -s xfs     MODULE       NAME                   SIZE  OBJECT FILEffffffffa0803d40  xfs                  939714  /lib/modules/3.10.0-327.22.2.el7.x86_64/kernel/fs/xfs/xfs.kocrash> struct -xo  xfs_mount.m_ail 0xffff883f7fe67000struct xfs_mount {  [ffff883f7fe67010] struct xfs_ail *m_ail;} struct   xfs_mount.m_ail 0xffff883f7fe67000  m_ail = 0xffff881f860fec80crash> struct -xo  xfs_ail.xa_lock 0xffff881f860fec80struct xfs_ail {  [ffff881f860fecc0] spinlock_t xa_lock;}

而且这把锁的值是：

crash> struct   xfs_ail.xa_lock 0xffff881f860fec80  xa_lock = {    {      rlock = {        raw_lock = {          {            head_tail = 2406518640,            tickets = {              head = 36720,              tail = 36720            }          }        }      }    }  }

我一看到这个是非0值，想当然认为是被持有了，为了验证我的想法，我甚至随便找了一把锁查看它的值，点背的时候，喝凉水都塞牙，我随手选择的锁是：

crash> p gc_lockgc_lock = $1 = {  {    rlock = {      raw_lock = {        {          head_tail = 0,          tickets = {            head = 0,            tail = 0          }        }      }    }  }}

后来仔细看了在没有开启 CONFIG_QUEUED_SPINLOCKS 的情况下，锁的定义才是如此，否则的话，是如下：

#ifndef __GENKSYMS__typedef struct qspinlock {    atomic_t    val;} arch_spinlock_t;#elsetypedef u32 __ticketpair_t;typedef struct arch_spinlock {    union {        __ticketpair_t    head_tail;        struct __raw_tickets {            __ticket_t head, tail;        } tickets;    };} arch_spinlock_t;#endif

所以要么0，要么1的情况只适合于第一种定义，以前因为业务默认配置的config问题，查的锁相关的crash，大多是第一种定义，但是当时没注意到这个细节，只是简单扫了一眼，浪费我两个小时，所以下盘不稳总会摔。

查看running状态的都没有持有这把锁，（其实按照第二种定义，是正常的，因为此时本就没人持有这把锁，因为head和tail相等）比较奇怪，后来找同事讨论，同事认为是while循环没跳出，所以获取锁的那行是schedule下面那行，还没有执行。

想起来自己曾经写过的博客《》,应该是如此

voidxfs_ail_push_all_sync(    struct xfs_ail  *ailp){    struct xfs_log_item    *lip;    DEFINE_WAIT(wait);    spin_lock(&ailp->xa_lock);    while ((lip = xfs_ail_max(ailp)) != NULL) {        prepare_to_wait(&ailp->xa_empty, &wait, TASK_UNINTERRUPTIBLE);        ailp->xa_target = lip->li_lsn;        wake_up_process(ailp->xa_task);        spin_unlock(&ailp->xa_lock);        schedule();--------------------这个和while循环一直跳不出去，        spin_lock(&ailp->xa_lock);    }    spin_unlock(&ailp->xa_lock);    finish_wait(&ailp->xa_empty, &wait);}

跳不出去的原因跟io相关，也就是wake_up 相关的内核线程之后，内核线程还是没有完成对应的任务：

crash> struct xfs_ail.xa_task 0xffff881f860fec80  xa_task = 0xffff883f8901dc00crash> task_struct.pid 0xffff883f8901dc00  pid = 42726crash> set 42726    PID: 42726COMMAND: "xfsaild/sdi"   TASK: ffff883f8901dc00  [THREAD_INFO: ffff881f3d828000]    CPU: 17  STATE: TASK_INTERRUPTIBLEcrash> btPID: 42726  TASK: ffff883f8901dc00  CPU: 17  COMMAND: "xfsaild/sdi" #0 [ffff881f3d82bcd0] __schedule at ffffffff8163df9b #1 [ffff881f3d82bd38] schedule at ffffffff8163e879 #2 [ffff881f3d82bd48] schedule_timeout at ffffffff8163c295 #3 [ffff881f3d82bdf8] xfsaild at ffffffffa07c6a4f [xfs] #4 [ffff881f3d82bec8] kthread at ffffffff810a661f #5 [ffff881f3d82bf50] ret_from_fork at ffffffff81649858

xfsaild 无法完成的原因也是因为IO异常:

附部分相关io异常的message日志：

kernel: sd 5:0:8:0: [sdi] CDB: Read(16) 88 00 00 00 00 00 03 4c b0 48 00 00 00 08 00 00kernel: sd 5:0:8:0: [sdi] CDB: Read(16) 88 00 00 00 00 00 15 32 8a 50 00 00 03 30 00 00kernel: blk_update_request: I/O error, dev sdi, sector 355634560kernel: sd 5:0:8:0: rejecting I/O to offline devicekernel: XFS (sdi): metadata I/O error: block 0xee173de0 ("xfs_trans_read_buf_map") error 5 numblks 8kernel: XFS (sdi): metadata I/O error: block 0x15da29d90 ("xfs_trans_read_buf_map") error 5 numblks 32kernel: sd 5:0:8:0: rejecting I/O to offline devicekernel: XFS (sdi): xfs_imap_to_bp: xfs_trans_read_buf() returned error -5.kernel: XFS (sdi): metadata I/O error: block 0xeca8a480 ("xfs_trans_read_buf_map") error 5 numblks 8kernel: sd 5:0:8:0: rejecting I/O to offline devicekernel: sd 5:0:8:0: rejecting I/O to offline device

对应的bug解决可以参照：