gfp_mask-from-fs-io.rst 3.0 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768
  1. .. _gfp_mask_from_fs_io:
  2. =================================
  3. GFP masks used from FS/IO context
  4. =================================
  5. :Date: May, 2018
  6. :Author: Michal Hocko <mhocko@kernel.org>
  7. Introduction
  8. ============
  9. Code paths in the filesystem and IO stacks must be careful when
  10. allocating memory to prevent recursion deadlocks caused by direct
  11. memory reclaim calling back into the FS or IO paths and blocking on
  12. already held resources (e.g. locks - most commonly those used for the
  13. transaction context).
  14. The traditional way to avoid this deadlock problem is to clear __GFP_FS
  15. respectively __GFP_IO (note the latter implies clearing the first as well) in
  16. the gfp mask when calling an allocator. GFP_NOFS respectively GFP_NOIO can be
  17. used as shortcut. It turned out though that above approach has led to
  18. abuses when the restricted gfp mask is used "just in case" without a
  19. deeper consideration which leads to problems because an excessive use
  20. of GFP_NOFS/GFP_NOIO can lead to memory over-reclaim or other memory
  21. reclaim issues.
  22. New API
  23. ========
  24. Since 4.12 we do have a generic scope API for both NOFS and NOIO context
  25. ``memalloc_nofs_save``, ``memalloc_nofs_restore`` respectively ``memalloc_noio_save``,
  26. ``memalloc_noio_restore`` which allow to mark a scope to be a critical
  27. section from a filesystem or I/O point of view. Any allocation from that
  28. scope will inherently drop __GFP_FS respectively __GFP_IO from the given
  29. mask so no memory allocation can recurse back in the FS/IO.
  30. .. kernel-doc:: include/linux/sched/mm.h
  31. :functions: memalloc_nofs_save memalloc_nofs_restore
  32. .. kernel-doc:: include/linux/sched/mm.h
  33. :functions: memalloc_noio_save memalloc_noio_restore
  34. FS/IO code then simply calls the appropriate save function before
  35. any critical section with respect to the reclaim is started - e.g.
  36. lock shared with the reclaim context or when a transaction context
  37. nesting would be possible via reclaim. The restore function should be
  38. called when the critical section ends. All that ideally along with an
  39. explanation what is the reclaim context for easier maintenance.
  40. Please note that the proper pairing of save/restore functions
  41. allows nesting so it is safe to call ``memalloc_noio_save`` or
  42. ``memalloc_noio_restore`` respectively from an existing NOIO or NOFS
  43. scope.
  44. What about __vmalloc(GFP_NOFS)
  45. ==============================
  46. vmalloc doesn't support GFP_NOFS semantic because there are hardcoded
  47. GFP_KERNEL allocations deep inside the allocator which are quite non-trivial
  48. to fix up. That means that calling ``vmalloc`` with GFP_NOFS/GFP_NOIO is
  49. almost always a bug. The good news is that the NOFS/NOIO semantic can be
  50. achieved by the scope API.
  51. In the ideal world, upper layers should already mark dangerous contexts
  52. and so no special care is required and vmalloc should be called without
  53. any problems. Sometimes if the context is not really clear or there are
  54. layering violations then the recommended way around that is to wrap ``vmalloc``
  55. by the scope API with a comment explaining the problem.