changelog

linux 内核源代码

ToDo/Notes:
	- Find and fix bugs.
	- The only places in the kernel where a file is resized are
	  ntfs_file_write*() and ntfs_truncate() for both of which i_mutex is
	  held.  Just have to be careful in read-/writepage and other helpers
	  not running under i_mutex that we play nice.  Also need to be careful
	  with initialized_size extension in ntfs_file_write*() and writepage.
	  UPDATE: The only things that need to be checked are the compressed
	  write and the other attribute resize/write cases like index
	  attributes, etc.  For now none of these are implemented so are safe.
	- Implement filling in of holes in aops.c::ntfs_writepage() and its
	  helpers.
	- Implement mft.c::sync_mft_mirror_umount().  We currently will just
	  leave the volume dirty on umount if the final iput(vol->mft_ino)
	  causes a write of any mirrored mft records due to the mft mirror
	  inode having been discarded already.  Whether this can actually ever
	  happen is unclear however so it is worth waiting until someone hits
	  the problem.

2.1.29 - Fix a deadlock at mount time.

	- During mount the VFS holds s_umount lock on the superblock.  So when
	  we try to empty the journal $LogFile contents by calling
	  ntfs_attr_set() when the machine does not have much memory and the
	  journal is large ntfs_attr_set() results in the VM trying to balance
	  dirty pages which in turn tries to that the s_umount lock and thus we
	  get a deadlock.  The solution is to not use ntfs_attr_set() and
	  instead do the zeroing by hand at the block level rather than page
	  cache level.
	- Fix sparse warnings.

2.1.28 - Fix a deadlock.

	- Fix deadlock in fs/ntfs/inode.c::ntfs_put_inode().  Thanks to Sergey
	  Vlasov for the report and detailed analysis of the deadlock.  The fix
	  involved getting rid of ntfs_put_inode() altogether and hence NTFS no
	  longer has a ->put_inode super operation.

2.1.27 - Various bug fixes and cleanups.

	- Fix two compiler warnings on Alpha.  Thanks to Andrew Morton for
	  reporting them.
	- Fix an (innocent) off-by-one error in the runlist code.
	- Fix a buggette in an "should be impossible" case handling where we
	  continued the attribute lookup loop instead of aborting it.
	- Use buffer_migrate_page() for the ->migratepage function of all ntfs
	  address space operations.
	- Fix comparison of $MFT and $MFTMirr to not bail out when there are
	  unused, invalid mft records which are the same in both $MFT and
	  $MFTMirr.
	- Add support for sparse files which have a compression unit of 0.
	- Remove all the make_bad_inode() calls.  This should only be called
	  from read inode and new inode code paths.
	- Limit name length in fs/ntfs/unistr.c::ntfs_nlstoucs() to maximum
	  allowed by NTFS, i.e. 255 Unicode characters, not including the
	  terminating NULL (which is not stored on disk).
	- Improve comments on file attribute flags in fs/ntfs/layout.h.
	- Fix a bug in fs/ntfs/inode.c::ntfs_read_locked_index_inode() where we
	  forgot to update a temporary variable so loading index inodes which
	  have an index allocation attribute failed.
	- Add a missing call to flush_dcache_mft_record_page() in
	  fs/ntfs/inode.c::ntfs_write_inode().
	- Handle the recently introduced -ENAMETOOLONG return value from
	  fs/ntfs/unistr.c::ntfs_nlstoucs() in fs/ntfs/namei.c::ntfs_lookup().
	- Semaphore to mutex conversion.  (Ingo Molnar)

2.1.26 - Minor bug fixes and updates.

	- Fix a potential overflow in file.c where a cast to s64 was missing in
	  a left shift of a page index.
	- The struct inode has had its i_sem semaphore changed to a mutex named
	  i_mutex.
	- We have struct kmem_cache now so use it instead of the typedef
	  kmem_cache_t.  (Pekka Enberg)
	- Implement support for sector sizes above 512 bytes (up to the maximum
	  supported by NTFS which is 4096 bytes).
	- Do more detailed reporting of why we cannot mount read-write by
	  special casing the VOLUME_MODIFIED_BY_CHKDSK flag.
	- Miscellaneous updates to layout.h.
	- Cope with attribute list attribute having invalid flags.  Windows
	  copes with this and even chkdsk does not detect or fix this so we
	  have to cope with it, too.  Thanks to Pawel Kot for reporting the
	  problem.

2.1.25 - (Almost) fully implement write(2) and truncate(2).

	- Change ntfs_map_runlist_nolock(), ntfs_attr_find_vcn_nolock() and
	  {__,}ntfs_cluster_free() to also take an optional attribute search
	  context as argument.  This allows calling these functions with the
	  mft record mapped.  Update all callers.
	- Fix potential deadlock in ntfs_mft_data_extend_allocation_nolock()
	  error handling by passing in the active search context when calling
	  ntfs_cluster_free().
	- Change ntfs_cluster_alloc() to take an extra boolean parameter
	  specifying whether the cluster are being allocated to extend an
	  attribute or to fill a hole.
	- Change ntfs_attr_make_non_resident() to call ntfs_cluster_alloc()
	  with @is_extension set to TRUE and remove the runlist terminator
	  fixup code as this is now done by ntfs_cluster_alloc().
	- Change ntfs_attr_make_non_resident to take the attribute value size
	  as an extra parameter.  This is needed since we need to know the size
	  before we can map the mft record and our callers always know it.  The
	  reason we cannot simply read the size from the vfs inode i_size is
	  that this is not necessarily uptodate.  This happens when
	  ntfs_attr_make_non_resident() is called in the ->truncate call path.
	- Fix ntfs_attr_make_non_resident() to update the vfs inode i_blocks
	  which is zero for a resident attribute but should no longer be zero
	  once the attribute is non-resident as it then has real clusters
	  allocated.
	- Add fs/ntfs/attrib.[hc]::ntfs_attr_extend_allocation(), a function to
	  extend the allocation of an attributes.  Optionally, the data size,
	  but not the initialized size can be extended, too.
	- Implement fs/ntfs/inode.[hc]::ntfs_truncate().  It only supports
	  uncompressed and unencrypted files and it never creates sparse files
	  at least for the moment (making a file sparse requires us to modify
	  its directory entries and we do not support directory operations at
	  the moment).  Also, support for highly fragmented files, i.e. ones
	  whose data attribute is split across multiple extents, is severly
	  limited.  When such a case is encountered, EOPNOTSUPP is returned.
	- Enable ATTR_SIZE attribute changes in ntfs_setattr().  This completes
	  the initial implementation of file truncation.  Now both open(2)ing
	  a file with the O_TRUNC flag and the {,f}truncate(2) system calls
	  will resize a file appropriately.  The limitations are that only
	  uncompressed and unencrypted files are supported.  Also, there is
	  only very limited support for highly fragmented files (the ones whose
	  $DATA attribute is split into multiple attribute extents).
	- In attrib.c::ntfs_attr_set() call balance_dirty_pages_ratelimited()
	  and cond_resched() in the main loop as we could be dirtying a lot of
	  pages and this ensures we play nice with the VM and the system as a
	  whole.
	- Implement file operations ->write, ->aio_write, ->writev for regular
	  files.  This replaces the old use of generic_file_write(), et al and
	  the address space operations ->prepare_write and ->commit_write.
	  This means that both sparse and non-sparse (unencrypted and
	  uncompressed) files can now be extended using the normal write(2)
	  code path.  There are two limitations at present and these are that
	  we never create sparse files and that we only have limited support
	  for highly fragmented files, i.e. ones whose data attribute is split
	  across multiple extents.   When such a case is encountered,
	  EOPNOTSUPP is returned.
	- $EA attributes can be both resident and non-resident.
	- Use %z for size_t to fix compilation warnings.  (Andrew Morton)
	- Fix compilation warnings with gcc-4.0.2 on SUSE 10.0.
	- Document extended attribute ($EA) NEED_EA flag.  (Based on libntfs
	  patch by Yura Pakhuchiy.)

2.1.24 - Lots of bug fixes and support more clean journal states.

	- Support journals ($LogFile) which have been modified by chkdsk.  This
	  means users can boot into Windows after we marked the volume dirty.
	  The Windows boot will run chkdsk and then reboot.  The user can then
	  immediately boot into Linux rather than having to do a full Windows
	  boot first before rebooting into Linux and we will recognize such a
	  journal and empty it as it is clean by definition.  Note, this only
	  works if chkdsk left the journal in an obviously clean state.
	- Support journals ($LogFile) with only one restart page as well as
	  journals with two different restart pages.  We sanity check both and
	  either use the only sane one or the more recent one of the two in the
	  case that both are valid.
	- Add fs/ntfs/malloc.h::ntfs_malloc_nofs_nofail() which is analogous to
	  ntfs_malloc_nofs() but it performs allocations with __GFP_NOFAIL and
	  hence cannot fail.
	- Use ntfs_malloc_nofs_nofail() in the two critical regions in
	  fs/ntfs/runlist.c::ntfs_runlists_merge().  This means we no longer
	  need to panic() if the allocation fails as it now cannot fail.
	- Fix two nasty runlist merging bugs that had gone unnoticed so far.
	  Thanks to Stefano Picerno for the bug report.
	- Remove two bogus BUG_ON()s from fs/ntfs/mft.c.
	- Fix handling of valid but empty mapping pairs array in
	  fs/ntfs/runlist.c::ntfs_mapping_pairs_decompress().
	- Report unrepresentable inodes during ntfs_readdir() as KERN_WARNING
	  messages and include the inode number.  Thanks to Yura Pakhuchiy for
	  pointing this out.
	- Change ntfs_rl_truncate_nolock() to throw away the runlist if the new
	  length is zero.
	- Add runlist.[hc]::ntfs_rl_punch_nolock() which punches a caller
	  specified hole into a runlist.
	- Fix a bug in fs/ntfs/index.c::ntfs_index_lookup().  When the returned
	  index entry is in the index root, we forgot to set the @ir pointer in
	  the index context.  Thanks to Yura Pakhuchiy for finding this bug.
	- Remove bogus setting of PageError in ntfs_read_compressed_block().
	- Add fs/ntfs/attrib.[hc]::ntfs_resident_attr_value_resize().
	- Fix a bug in ntfs_map_runlist_nolock() where we forgot to protect
	  access to the allocated size in the ntfs inode with the size lock.
	- Fix ntfs_attr_vcn_to_lcn_nolock() and ntfs_attr_find_vcn_nolock() to
	  return LCN_ENOENT when there is no runlist and the allocated size is
	  zero.
	- Fix load_attribute_list() to handle the case of a NULL runlist.
	- Fix handling of sparse attributes in ntfs_attr_make_non_resident().
	- Add BUG() checks to ntfs_attr_make_non_resident() and ntfs_attr_set()
	  to ensure that these functions are never called for compressed or
	  encrypted attributes.
	- Fix cluster (de)allocators to work when the runlist is NULL and more
	  importantly to take a locked runlist rather than them locking it
	  which leads to lock reversal.
	- Truncate {a,c,m}time to the ntfs supported time granularity when
	  updating the times in the inode in ntfs_setattr().
	- Fixup handling of sparse, compressed, and encrypted attributes in
	  fs/ntfs/inode.c::ntfs_read_locked_{,attr_,index_}inode(),
	  fs/ntfs/aops.c::ntfs_{read,write}page().
	- Make ntfs_write_block() not instantiate sparse blocks if they contain
	  only zeroes.
	- Optimize fs/ntfs/aops.c::ntfs_write_block() by extending the page
	  lock protection over the buffer submission for i/o which allows the
	  removal of the get_bh()/put_bh() pairs for each buffer.
	- Fix fs/ntfs/aops.c::ntfs_{read,write}_block() to handle the case
	  where a concurrent truncate has truncated the runlist under our feet.
	- Fix page_has_buffers()/page_buffers() handling in fs/ntfs/aops.c.
	- In fs/ntfs/aops.c::ntfs_end_buffer_async_read(), use a bit spin lock
	  in the first buffer head instead of a driver global spin lock to
	  improve scalability.
	- Minor fix to error handling and error message display in
	  fs/ntfs/aops.c::ntfs_prepare_nonresident_write().
	- Change the mount options {u,f,d}mask to always parse the number as
	  an octal number to conform to how chmod(1) works, too.  Thanks to
	  Giuseppe Bilotta and Horst von Brand for pointing out the errors of
	  my ways.
	- Fix various bugs in the runlist merging code.  (Based on libntfs
	  changes by Richard Russon.)
	- Fix sparse warnings that have crept in over time.
	- Change ntfs_cluster_free() to require a write locked runlist on entry
	  since we otherwise get into a lock reversal deadlock if a read locked
	  runlist is passed in. In the process also change it to take an ntfs
	  inode instead of a vfs inode as parameter.
	- Fix the definition of the CHKD ntfs record magic.  It had an off by
	  two error causing it to be CHKB instead of CHKD.
	- Fix a stupid bug in __ntfs_bitmap_set_bits_in_run() which caused the
	  count to become negative and hence we had a wild memset() scribbling
	  all over the system's ram.