gfp_mask-from-fs-io.rst 2.9 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566
  1. =================================
  2. GFP masks used from FS/IO context
  3. =================================
  4. :Date: May, 2018
  5. :Author: Michal Hocko <mhocko@kernel.org>
  6. Introduction
  7. ============
  8. Code paths in the filesystem and IO stacks must be careful when
  9. allocating memory to prevent recursion deadlocks caused by direct
  10. memory reclaim calling back into the FS or IO paths and blocking on
  11. already held resources (e.g. locks - most commonly those used for the
  12. transaction context).
  13. The traditional way to avoid this deadlock problem is to clear __GFP_FS
  14. respectively __GFP_IO (note the latter implies clearing the first as well) in
  15. the gfp mask when calling an allocator. GFP_NOFS respectively GFP_NOIO can be
  16. used as shortcut. It turned out though that above approach has led to
  17. abuses when the restricted gfp mask is used "just in case" without a
  18. deeper consideration which leads to problems because an excessive use
  19. of GFP_NOFS/GFP_NOIO can lead to memory over-reclaim or other memory
  20. reclaim issues.
  21. New API
  22. ========
  23. Since 4.12 we do have a generic scope API for both NOFS and NOIO context
  24. ``memalloc_nofs_save``, ``memalloc_nofs_restore`` respectively ``memalloc_noio_save``,
  25. ``memalloc_noio_restore`` which allow to mark a scope to be a critical
  26. section from a filesystem or I/O point of view. Any allocation from that
  27. scope will inherently drop __GFP_FS respectively __GFP_IO from the given
  28. mask so no memory allocation can recurse back in the FS/IO.
  29. .. kernel-doc:: include/linux/sched/mm.h
  30. :functions: memalloc_nofs_save memalloc_nofs_restore
  31. .. kernel-doc:: include/linux/sched/mm.h
  32. :functions: memalloc_noio_save memalloc_noio_restore
  33. FS/IO code then simply calls the appropriate save function before
  34. any critical section with respect to the reclaim is started - e.g.
  35. lock shared with the reclaim context or when a transaction context
  36. nesting would be possible via reclaim. The restore function should be
  37. called when the critical section ends. All that ideally along with an
  38. explanation what is the reclaim context for easier maintenance.
  39. Please note that the proper pairing of save/restore functions
  40. allows nesting so it is safe to call ``memalloc_noio_save`` or
  41. ``memalloc_noio_restore`` respectively from an existing NOIO or NOFS
  42. scope.
  43. What about __vmalloc(GFP_NOFS)
  44. ==============================
  45. vmalloc doesn't support GFP_NOFS semantic because there are hardcoded
  46. GFP_KERNEL allocations deep inside the allocator which are quite non-trivial
  47. to fix up. That means that calling ``vmalloc`` with GFP_NOFS/GFP_NOIO is
  48. almost always a bug. The good news is that the NOFS/NOIO semantic can be
  49. achieved by the scope API.
  50. In the ideal world, upper layers should already mark dangerous contexts
  51. and so no special care is required and vmalloc should be called without
  52. any problems. Sometimes if the context is not really clear or there are
  53. layering violations then the recommended way around that is to wrap ``vmalloc``
  54. by the scope API with a comment explaining the problem.