[dm-devel] Patchset 2.4.20-dm-10

[Joe Thornber]

http://people.sistina.com/~thornber/patches/2.4-stable/2.4.20/2.4.20-dm-10.tar.bz2


Changes since 2.4.20-dm-9
-------------------------

Revision 50:
  Revert patches 44, 45 and 47 - will be re-integrated later.
  [Alasdair Kergon]

Revision 51:
  Support for more architectures (parisc, mips64, ppc64, s390x, sparc64,
  x86_64) [Alasdair Kergon]

Revision 52:
  In persistent_prepare(), the "enough-room-left" check only verifies that the
  start of the next-free chunk is within the cow device. If the cow device
  is not an even multiple of the chunk-size, the snapshot code will eventually
  try to write off the end of the cow device.
  
  Check that the entire next-free chunk is within the cow device. 
  [Kevin Corry]

Revision 53:
  In dm_create_persistent(), when setting up an existing snapshot, if the valid
  bit from the snapshot header is not set, refuse to activate the snapshot.
  [Kevin Corry]

Revision 54:
  In __map_buffer(), the call to dm_table_find_target() will never return a
  NULL pointer (provided every DM device table has a valid "targets" array,
  which should be the case). It may, however, return a pointer to a blank target
  record (since there are always a KEYS_PER_NODE-multiple of target records
  in the B-tree, but not all of these target records are necessarily used).
  This will occur when DM receives an I/O request past the end of a device.
  (You'd think this shouldn't happen, but it does - generic_make_request()
  will allow an I/O one sector past the end of a device with an even number
  of sectors.)
  
  Instead of checking for a NULL return from dm_table_find_target(), check
  for a valid "type" field in the returned target.
  [Kevin Corry]

Revision 55:
  Whitespace tweaking.  [Alasdair Kergon]

Revision 56:
  Tidy some DMWARNs. [Alasdair Kergon]

Revision 57:
  Update version number.  [Alasdair Kergon]