[v7,0/8] parallels: Add duplication check, repair at open, fix bugs

Message

Alexander Ivanov July 1, 2023, 10:07 a.m. UTC

Fix incorrect data end calculation in parallels_open().

Check if data_end greater than the file size.

Add change_info argument to parallels_check_leak().

Add checking and repairing duplicate offsets in BAT

Image repairing in parallels_open().

v7:
3: Renamed "change_info" argument to "explicit", move info printing under
   explicit condition.
4: Different patch. Add data_start field to BDRVParallelsState for future
   host_cluster_index() function.
5: Prevously was 4th. Used s->data_start instead of s->header->data_off.
   Add bitmap size increasing if file length is not cluster aligned. Revert
   a couple conditions for better readability. Changed sectors calculation
   for better code transparency. Replaced sector variable by host_sector and
   guest_sector. Renamed off to host_off. Moved out_repare_bat: below return
   to get jumps on error path only.
6: Prevously was 5th.
7: New patch. Replace bdrv_getlength() by bdrv_co_getlength() in
   parallels_check_outside_image() because it is used in coroutine context.
8: New patch. Add data_off check.

v6:
2: Different patch. Refused to split image leak handling. Instead there is a
   patch with a data_end check.
3: Different patch. There is a patch with change_info argument.
4: Removed changing fprintf by qemu_log from this patchset. Previously 3rd
   patch became 4th. Replaced qemu_memalign() by qemu_blockalign(). Got
   rid of iovecs, replaced bdrv_co_pwritev() by bdrv_co_pwrite(). Added
   assert(cluster_index < bitmap_size). Now BAT changes are reverted if
   there was an error in the cluster copying process. Simplified a sector
   calculation.
5: Moved header magic check to the appropriate place. Added a
   migrate_del_blocker() call and s->bat_dirty_bmap freeing on error.

v5:
3: Fixed a byteorder bug, fixed zero-length image handling and fixed uint32
   truncation.

v4:
2,5: Rebased.

v3:
2: Added (size >= res->image_end_offset) assert and changed the comment in
   parallels_get_leak_size(). Changed error printing and leaks fixing order.
3: Removed highest_offset() helper, instead image_end_offset field is used.
5: Moved highest_offset() code to parallels_open() - now it is used only in
   this function. Fixed data_end update condition. Fixed a leak of
   s->migration_blocker.

v2:
2: Moved outsude parallels_check_leak() 2 helpers:
   parallels_get_leak_size() and parallels_fix_leak().
   
3: Used highest_offset() helper in parallels_check_leak(). Fixed a typo.
   Added comments. Replaced g_malloc() call by qemu_memalign(). Replaced
   bdrv_pread() call by bdrv_co_pread(). Got rid of keeping bytes and
   sectors in the same variable. Added setting the bitmap of the used
   clusters for a new allocated cluster if it isn't out of the bitmap.
   Moved the leak fix to the end of all the checks. Removed a dependence
   on image format for the duplicate check.
   
4 (old): Merged this patch to the previous.
4 (former 5): Fixed formatting.
5 (former 6): Fixed comments. Added O_INACTIVE check in the condition.
              Replaced inuse detection by header_unclean checking.
              Replaced playing with corutines by bdrv_check() usage.


Alexander Ivanov (8):
  parallels: Incorrect data end calculation in parallels_open()
  parallels: Check if data_end greater than the file size
  parallels: Add "explicit" argument to parallels_check_leak()
  parallels: Add data_start field to BDRVParallelsState
  parallels: Add checking and repairing duplicate offsets in BAT
  parallels: Image repairing in parallels_open()
  parallels: Use bdrv_co_getlength() in parallels_check_outside_image()
  parallels: Add data_off check

 block/parallels.c | 317 +++++++++++++++++++++++++++++++++++++++-------
 block/parallels.h |   1 +
 2 files changed, 274 insertions(+), 44 deletions(-)