dax.txt 4.1 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697
  1. Direct Access for files
  2. -----------------------
  3. Motivation
  4. ----------
  5. The page cache is usually used to buffer reads and writes to files.
  6. It is also used to provide the pages which are mapped into userspace
  7. by a call to mmap.
  8. For block devices that are memory-like, the page cache pages would be
  9. unnecessary copies of the original storage. The DAX code removes the
  10. extra copy by performing reads and writes directly to the storage device.
  11. For file mappings, the storage device is mapped directly into userspace.
  12. Usage
  13. -----
  14. If you have a block device which supports DAX, you can make a filesystem
  15. on it as usual. The DAX code currently only supports files with a block
  16. size equal to your kernel's PAGE_SIZE, so you may need to specify a block
  17. size when creating the filesystem. When mounting it, use the "-o dax"
  18. option on the command line or add 'dax' to the options in /etc/fstab.
  19. Implementation Tips for Block Driver Writers
  20. --------------------------------------------
  21. To support DAX in your block driver, implement the 'direct_access'
  22. block device operation. It is used to translate the sector number
  23. (expressed in units of 512-byte sectors) to a page frame number (pfn)
  24. that identifies the physical page for the memory. It also returns a
  25. kernel virtual address that can be used to access the memory.
  26. The direct_access method takes a 'size' parameter that indicates the
  27. number of bytes being requested. The function should return the number
  28. of bytes that can be contiguously accessed at that offset. It may also
  29. return a negative errno if an error occurs.
  30. In order to support this method, the storage must be byte-accessible by
  31. the CPU at all times. If your device uses paging techniques to expose
  32. a large amount of memory through a smaller window, then you cannot
  33. implement direct_access. Equally, if your device can occasionally
  34. stall the CPU for an extended period, you should also not attempt to
  35. implement direct_access.
  36. These block devices may be used for inspiration:
  37. - axonram: Axon DDR2 device driver
  38. - brd: RAM backed block device driver
  39. - dcssblk: s390 dcss block device driver
  40. Implementation Tips for Filesystem Writers
  41. ------------------------------------------
  42. Filesystem support consists of
  43. - adding support to mark inodes as being DAX by setting the S_DAX flag in
  44. i_flags
  45. - implementing the direct_IO address space operation, and calling
  46. dax_do_io() instead of blockdev_direct_IO() if S_DAX is set
  47. - implementing an mmap file operation for DAX files which sets the
  48. VM_MIXEDMAP and VM_HUGEPAGE flags on the VMA, and setting the vm_ops to
  49. include handlers for fault, pmd_fault and page_mkwrite (which should
  50. probably call dax_fault(), dax_pmd_fault() and dax_mkwrite(), passing the
  51. appropriate get_block() callback)
  52. - calling dax_truncate_page() instead of block_truncate_page() for DAX files
  53. - calling dax_zero_page_range() instead of zero_user() for DAX files
  54. - ensuring that there is sufficient locking between reads, writes,
  55. truncates and page faults
  56. The get_block() callback passed to the DAX functions may return
  57. uninitialised extents. If it does, it must ensure that simultaneous
  58. calls to get_block() (for example by a page-fault racing with a read()
  59. or a write()) work correctly.
  60. These filesystems may be used for inspiration:
  61. - ext2: the second extended filesystem, see Documentation/filesystems/ext2.txt
  62. - ext4: the fourth extended filesystem, see Documentation/filesystems/ext4.txt
  63. Shortcomings
  64. ------------
  65. Even if the kernel or its modules are stored on a filesystem that supports
  66. DAX on a block device that supports DAX, they will still be copied into RAM.
  67. The DAX code does not work correctly on architectures which have virtually
  68. mapped caches such as ARM, MIPS and SPARC.
  69. Calling get_user_pages() on a range of user memory that has been mmaped
  70. from a DAX file will fail as there are no 'struct page' to describe
  71. those pages. This problem is being worked on. That means that O_DIRECT
  72. reads/writes to those memory ranges from a non-DAX file will fail (note
  73. that O_DIRECT reads/writes _of a DAX file_ do work, it is the memory
  74. that is being accessed that is key here). Other things that will not
  75. work include RDMA, sendfile() and splice().