IJ02296: FINE-GRAIN DIRECTORY LOCKING STACK OVERFLOW IN RECURSION

APAR status

• Closed as program error.

Error description

• During fetch of directory block for readdir using
  fine-grain
  directory locking (FGDL), it is possible for stack
  overflow
  as result of recursion used to fetch all blocks covered
  by
  FGDL token range. This could cause very large directory
  to
  get compacted after most of the files were deleted.
  
  Reported in:
  Spectrum Scale 4.1.1.16 on AIX 6.1
  
  Known Impact:
  Server sees a stack overflow.
  

Local fix

• Delete large directories that have very few files in them
  and recreate them.
  

Problem summary

• Stack overflow as result of excessive recursion could lead to
  either GPFS daemon crash with Signal 11 or hang in kernel.
  This could happen when readdir operation is operating under
  FGDL mode.
  

Problem conclusion

• Restructure FGDL directory fetching code to use loop instead of
  recursion.  Also added code to guard against been stuck in a
  loop.
  

Temporary fix

Comments

•                                          411
  

APAR Information

• APAR is sysrouted FROM one or more of the following:

  IJ01070

• APAR is sysrouted TO one or more of the following:

Modules/Macros

• MO411 S/MACROS:   None
  

Fix information

• Fixed component name

  SPECTRUM SCALE

• Fixed component ID

  5725Q01AP

Applicable component levels

• R411 PSY U884675

     19/04/30 I 1000