XFS uses the internal tmpfile() infrastructure for the whiteout inode
used for RENAME_WHITEOUT operations. For tmpfile inodes, XFS allocates
the inode, drops di_nlink, adds the inode to the agi unlinked list,
calls d_tmpfile() which correspondingly drops i_nlink of the vfs inode,
and then finishes the common inode setup (e.g., clear I_NEW and unlock).

The d_tmpfile() call was originally made inxfs_create_tmpfile(), but was
pulled up out of that function as part of the following commit to
resolve a deadlock issue:

	330033d6 xfs: fix tmpfile/selinux deadlock and initialize security

As a result, callers of xfs_create_tmpfile() are responsible for either
calling d_tmpfile() or fixing up i_nlink appropriately. The whiteout
tmpfile allocation helper does neither. As a result, the vfs ->i_nlink
becomes inconsistent with the on-disk ->di_nlink once xfs_rename() links
it back into the source dentry and calls xfs_bumplink().

Update the assert in xfs_rename() to help detect this problem in the
future and update xfs_rename_alloc_whiteout() to decrement the link
count as part of the manual tmpfile inode setup.
Signed-off-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

It was missed when we converted everything in XFs to use negative error
numbers, so fix it now. Bug introduced in 3.17 by commit 2451337d ("xfs: global
error sign conversion"), and should go back to stable kernels.

Thanks to Brian Foster for noticing it.

cc: <stable@vger.kernel.org> # 3.17, 3.18, 3.19, 4.0
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

xfs_attr_inactive() is supposed to clean up the attribute fork when
the inode is being freed. While it removes attribute fork extents,
it completely ignores attributes in local format, which means that
there can still be active attributes on the inode after
xfs_attr_inactive() has run.

This leads to problems with concurrent inode writeback - the in-core
inode attribute fork is removed without locking on the assumption
that nothing will be attempting to access the attribute fork after a
call to xfs_attr_inactive() because it isn't supposed to exist on
disk any more.

To fix this, make xfs_attr_inactive() completely remove all traces
of the attribute fork from the inode, regardless of it's state.
Further, also remove the in-core attribute fork structure safely so
that there is nothing further that needs to be done by callers to
clean up the attribute fork. This means we can remove the in-core
and on-disk attribute forks atomically.

Also, on error simply remove the in-memory attribute fork. There's
nothing that can be done with it once we have failed to remove the
on-disk attribute fork, so we may as well just blow it away here
anyway.

cc: <stable@vger.kernel.org> # 3.12 to 4.0
Reported-by: NWaiman Long <waiman.long@hp.com>
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

This results in BMBT corruption, as seen by this test:

# mkfs.xfs -f -d size=40051712b,agcount=4 /dev/vdc
....
# mount /dev/vdc /mnt/scratch
# xfs_io -ft -c "extsize 16m" -c "falloc 0 30g" -c "bmap -vp" /mnt/scratch/foo

which results in this failure on a debug kernel:

XFS: Assertion failed: (blockcount & xfs_mask64hi(64-BMBT_BLOCKCOUNT_BITLEN)) == 0, file: fs/xfs/libxfs/xfs_bmap_btree.c, line: 211
....
Call Trace:
 [<ffffffff814cf0ff>] xfs_bmbt_set_allf+0x8f/0x100
 [<ffffffff814cf18d>] xfs_bmbt_set_all+0x1d/0x20
 [<ffffffff814f2efe>] xfs_iext_insert+0x9e/0x120
 [<ffffffff814c7956>] ? xfs_bmap_add_extent_hole_real+0x1c6/0xc70
 [<ffffffff814c7956>] xfs_bmap_add_extent_hole_real+0x1c6/0xc70
 [<ffffffff814caaab>] xfs_bmapi_write+0x72b/0xed0
 [<ffffffff811c72ac>] ? kmem_cache_alloc+0x15c/0x170
 [<ffffffff814fe070>] xfs_alloc_file_space+0x160/0x400
 [<ffffffff81ddcc29>] ? down_write+0x29/0x60
 [<ffffffff815063eb>] xfs_file_fallocate+0x29b/0x310
 [<ffffffff811d2bc8>] ? __sb_start_write+0x58/0x120
 [<ffffffff811e3e18>] ? do_vfs_ioctl+0x318/0x570
 [<ffffffff811cd680>] vfs_fallocate+0x140/0x260
 [<ffffffff811ce6f8>] SyS_fallocate+0x48/0x80
 [<ffffffff81ddec09>] system_call_fastpath+0x12/0x17

The tracepoint that indicates the extent that triggered the assert
failure is:

xfs_iext_insert:   idx 0 offset 0 block 16777224 count 2097152 flag 1

Clearly indicating that the extent length is greater than MAXEXTLEN,
which is 2097151. A prior trace point shows the allocation was an
exact size match and that a length greater than MAXEXTLEN was asked
for:

xfs_alloc_size_done:  agno 1 agbno 8 minlen 2097152 maxlen 2097152
					    ^^^^^^^        ^^^^^^^

We don't see this problem with extent size hints through the IO path
because we can't do single IOs large enough to trigger MAXEXTLEN
allocation. fallocate(), OTOH, is not limited in it's allocation
sizes and so needs help here.

The issue is that the extent size hint alignment is rounding up the
extent size past MAXEXTLEN, because xfs_bmapi_write() is not taking
into account extent size hints when calculating the maximum extent
length to allocate. xfs_bmapi_reserve_delalloc() is already doing
this, but direct extent allocation is not.

Unfortunately, the calculation in xfs_bmapi_reserve_delalloc() is
wrong, and it works only because delayed allocation extents are not
limited in size to MAXEXTLEN in the in-core extent tree. hence this
calculation does not work for direct allocation, and the delalloc
code needs fixing. This may, in fact be the underlying bug that
occassionally causes transaction overruns in delayed allocation
extent conversion, so now we know it's wrong we should fix it, too.
Many thanks to Brian Foster for finding this problem during review
of this patch.

Hence the fix, after much code reading, is to allow
xfs_bmap_extsize_align() to align partial extents when full
alignment would extend the alignment past MAXEXTLEN. We can safely
do this because all callers have higher layer allocation loops that
already handle short allocations, and so will simply run another
allocation to cover the remainder of the requested allocation range
that we ignored during alignment. The advantage of this approach is
that it also removes the need for callers to do anything other than
limit their requests to MAXEXTLEN - they don't really need to be
aware of extent size hints at all.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Because the counters use a custom batch size, the comparison
functions need to be aware of that batch size otherwise the
comparison does not work correctly. This leads to ASSERT failures
on generic/027 like this:

 XFS: Assertion failed: 0, file: fs/xfs/xfs_mount.c, line: 1099
 ------------[ cut here ]------------
....
 Call Trace:
  [<ffffffff81522a39>] xfs_mod_icount+0x99/0xc0
  [<ffffffff815285cb>] xfs_trans_unreserve_and_mod_sb+0x28b/0x5b0
  [<ffffffff8152f941>] xfs_log_commit_cil+0x321/0x580
  [<ffffffff81528e17>] xfs_trans_commit+0xb7/0x260
  [<ffffffff81503d4d>] xfs_bmap_finish+0xcd/0x1b0
  [<ffffffff8151da41>] xfs_inactive_ifree+0x1e1/0x250
  [<ffffffff8151dbe0>] xfs_inactive+0x130/0x200
  [<ffffffff81523a21>] xfs_fs_evict_inode+0x91/0xf0
  [<ffffffff811f3958>] evict+0xb8/0x190
  [<ffffffff811f433b>] iput+0x18b/0x1f0
  [<ffffffff811e8853>] do_unlinkat+0x1f3/0x320
  [<ffffffff811d548a>] ? filp_close+0x5a/0x80
  [<ffffffff811e999b>] SyS_unlinkat+0x1b/0x40
  [<ffffffff81e0892e>] system_call_fastpath+0x12/0x71

This is a regression introduced by commit 501ab323 ("xfs: use generic
percpu counters for inode counter").

This patch fixes the same problem for both the inode counter and the
free block counter in the superblocks.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Function percpu_counter_read just return the current counter, which can be
negative. This will cause the checking of "allocated inode
counts <= m_maxicount" false positive. Use percpu_counter_read_positive can
solve this problem, and be consistent with the purpose to introduce percpu
mechanism to xfs.
Signed-off-by: NGeorge Wang <xuw2015@gmail.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

generic_file_direct_write() does all sorts of things to make DIO
work "sorta ok" with mixed buffered IO workloads. We already do
most of this work in xfs_file_aio_dio_write() because of the locking
requirements, so there's only a couple of things it does for us.

The first thing is that it does a page cache invalidation after the
->direct_IO callout. This can easily be added to the XFS code.

The second thing it does is that if data was written, it updates the
iov_iter structure to reflect the data written, and then does EOF
size updates if necessary. For XFS, these EOF size updates are now
not necessary, as we do them safely and race-free in IO completion
context. That leaves just the iov_iter update, and that's also moved
to the XFS code.

Therefore we don't need to call generic_file_direct_write() and in
doing so remove redundant buffered writeback and page cache
invalidation calls from the DIO submission path. We also remove a
racy EOF size update, and make the DIO submission code in XFS much
easier to follow. Wins all round, really.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

When we are doing AIO DIO writes, the IOLOCK only provides an IO
submission barrier. When we need to do EOF zeroing, we need to ensure
that no other IO is in progress and all pending in-core EOF updates
have been completed. This requires us to wait for all outstanding
AIO DIO writes to the inode to complete and, if necessary, run their
EOF updates.

Once all the EOF updates are complete, we can then restart
xfs_file_aio_write_checks() while holding the IOLOCK_EXCL, knowing
that EOF is up to date and we have exclusive IO access to the file
so we can run EOF block zeroing if we need to without interference.
This gives EOF zeroing the same exclusivity against other IO as we
provide truncate operations.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

xfs_end_io_direct_write() can race with other IO completions when
updating the in-core inode size. The IO completion processing is not
serialised for direct IO - they are done either under the
IOLOCK_SHARED for non-AIO DIO, and without any IOLOCK held at all
during AIO DIO completion. Hence the non-atomic test-and-set update
of the in-core inode size is racy and can result in the in-core
inode size going backwards if the race if hit just right.

If the inode size goes backwards, this can trigger the EOF zeroing
code to run incorrectly on the next IO, which then will zero data
that has successfully been written to disk by a previous DIO.

To fix this bug, we need to serialise the test/set updates of the
in-core inode size. This first patch introduces locking around the
relevant updates and checks in the DIO path. Because we now have an
ioend in xfs_end_io_direct_write(), we know exactly then we are
doing an IO that requires an in-core EOF update, and we know that
they are not running in interrupt context. As such, we do not need to
use irqsave() spinlock variants to protect against interrupts while
the lock is held.

Hence we can use an existing spinlock in the inode to do this
serialisation and so not need to grow the struct xfs_inode just to
work around this problem.

This patch does not address the test/set EOF update in
generic_file_write_direct() for various reasons - that will be done
as a followup with separate explanation.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

DIO writes that lie entirely within EOF have nothing to do in IO
completion. In this case, we don't need no steekin' ioend, and so we
can avoid allocating an ioend until we have a mapping that spans
EOF.

This means that IO completion has two contexts - deferred completion
to the dio workqueue that uses an ioend, and interrupt completion
that does nothing because there is nothing that can be done in this
context.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Currently a DIO overwrite that extends the EOF (e.g sub-block IO or
write into allocated blocks beyond EOF) requires a transaction for
the EOF update. Thi is done in IO completion context, but we aren't
explicitly handling this situation properly and so it can run in
interrupt context. Ensure that we defer IO that spans EOF correctly
to the DIO completion workqueue, and now that we have an ioend in IO
completion we can use the common ioend completion path to do all the
work.

Note: we do not preallocate the append transaction as we can have
multiple mapping and allocation calls per direct IO. hence
preallocating can still leave us with nested transactions by
attempting to map and allocate more blocks after we've preallocated
an append transaction.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Currently we can only tell DIO completion that an IO requires
unwritten extent completion. This is done by a hacky non-null
private pointer passed to Io completion, but the private pointer
does not actually contain any information that is used.

We also need to pass to IO completion the fact that the IO may be
beyond EOF and so a size update transaction needs to be done. This
is currently determined by checks in the io completion, but we need
to determine if this is necessary at block mapping time as we need
to defer the size update transactions to a completion workqueue,
just like unwritten extent conversion.

To do this, first we need to allocate and pass an ioend to to IO
completion. Add this for unwritten extent conversion; we'll do the
EOF updates in the next commit.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

The mapping size calculation is done last in __xfs_get_blocks(), but
we are going to need the actual mapping size we will use to map the
direct IO correctly in xfs_map_direct(). Factor out the calculation
for code clarity, and move the call to be the first operation in
mapping the extent to the returned buffer.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Clarify and separate the buffer mapping logic so that the direct IO mapping is
not tangled up in propagating the extent status to teh mapping buffer. This
makes it easier to extend the direct IO mapping to use an ioend in future.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

that's the bulk of filesystem drivers dealing with inodes of their own
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

We want to drop all I/O path locks when recalling layouts, and that includes
i_mutex for the write path.  Without this we get stuck processe when recalls
take too long.

[dchinner: fix build with !CONFIG_PNFS]
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

xfs_attr3_leaf_remove() removes an attribute from an attr leaf block. If
the attribute nameval data happens to be at the start of the nameval
region, a new start offset (firstused) for the region is calculated
(since the region grows from the tail of the block to the start). Once
the new firstused is calculated, it is checked for zero in an apparent
overflow check.

Now that the in-core firstused is 32-bit, overflow is not possible and
this check can be removed. Since the purpose for this check is not
documented and appears to exist since the port to Linux, be conservative
and replace it with an assert.
Signed-off-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

The on-disk xfs_attr3_leaf_hdr structure firstused field is 16-bit and
subject to overflow when fs block size is 64k. The field is typically
initialized to block size when an attr leaf block is initialized. This
problem is demonstrated by assert failures when running xfstests
generic/117 on an fs with 64k blocks.

To support the existing attr leaf block algorithms for insertion,
rebalance and entry movement, increase the size of the in-core firstused
field to 32-bit and handle the potential overflow on conversion to/from
the on-disk structure. If the overflow condition occurs, set a special
value in the firstused field that is translated back on header read. The
special value is only required in the case of an empty 64k attr block. A
value of zero is used because firstused is initialized to the block size
and grows backwards from there. Furthermore, the attribute block header
occupies the first bytes of the block. Thus, a value of zero has no
other legitimate meaning for this structure. Two new conversion helpers
are created to manage the conversion of firstused to and from disk.
Signed-off-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

The firstused field of the xfs_attr3_leaf_hdr structure is subject to an
overflow when fs blocksize is 64k. In preparation to handle this
overflow in the header conversion functions, pass the attribute geometry
to the functions that convert the in-core structure to and from the
on-disk structure.
Signed-off-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

There's a bit of a loophole in norecovery mount handling right
now: an initial mount must be readonly, but nothing prevents
a mount -o remount,rw from producing a writable, unrecovered
xfs filesystem.

It might be possible to try to perform a log recovery when this
is requested, but I'm not sure it's worth the effort.  For now,
simply disallow this sort of transition.
Signed-off-by: NEric Sandeen <sandeen@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Signed-off-by: NFengguang Wu <fengguang.wu@intel.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

... avoiding write_iter/fcntl races.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

... returning -E... upon error and amount of data left in iter after
(possible) truncation upon success.  Note, that normal case gives
a non-zero (positive) return value, so any tests for != 0 _must_ be
updated.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Conflicts:
	fs/ext4/file.c

Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

all remaining callers are passing 0; some just obscure that fact.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Now that no one is using rw, remove it completely.
Signed-off-by: NOmar Sandoval <osandov@osandov.com>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

The rw parameter to direct_IO is redundant with iov_iter->type, and
treated slightly differently just about everywhere it's used: some users
do rw & WRITE, and others do rw == WRITE where they should be doing a
bitwise check. Simplify this with the new iov_iter_rw() helper, which
always returns either READ or WRITE.
Signed-off-by: NOmar Sandoval <osandov@osandov.com>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Most filesystems call through to these at some point, so we'll start
here.
Signed-off-by: NOmar Sandoval <osandov@osandov.com>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

All places outside of core VFS that checked ->read and ->write for being NULL or
called the methods directly are gone now, so NULL {read,write} with non-NULL
{read,write}_iter will do the right thing in all cases.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

struct kiocb now is a generic I/O container, so move it to fs.h.
Also do a #include diet for aio.h while we're at it.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

This patch implements fallocate's FALLOC_FL_INSERT_RANGE for XFS.

1) Make sure that both offset and len are block size aligned.
2) Update the i_size of inode by len bytes.
3) Compute the file's logical block number against offset. If the computed
   block number is not the starting block of the extent, split the extent
   such that the block number is the starting block of the extent.
4) Shift all the extents which are lying bewteen [offset, last allocated extent]
   towards right by len bytes. This step will make a hole of len bytes
   at offset.
Signed-off-by: NNamjae Jeon <namjae.jeon@samsung.com>
Signed-off-by: NAshish Sangwan <a.sangwan@samsung.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

added a positive error return value.

This value filters up through the return layers and should be
negative as the other return values are in the same function.
Signed-off-by: NJoe Perches <joe@perches.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

xfs_mru_cache_insert() can be called from within transaction context
during block allocation like so:

write(2)
  ....
    xfs_get_blocks
      xfs_iomap_write_direct
        start transaction
        xfs_bmapi_write
          xfs_bmapi_allocate
            xfs_bmap_btalloc
              xfs_bmap_btalloc_filestreams
                xfs_filestream_new_ag
                  xfs_filestream_pick_ag
                    xfs_mru_cache_insert
                      radix_tree_preload(GFP_KERNEL)

In this case, GFP_KERNEL is incorrect and can potentially lead to
deadlocks in memory reclaim. It should use GFP_NOFS allocations to
avoid lock recursion problems.

[dchinner: rewrote commit message]
Signed-off-by: NByoungyoung Lee <blee@gatech.edu>
Signed-off-by: NSanidhya Kashyap <sanidhya.gatech@gmail.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Use %pS for actual addresses, otherwise you'll get bad output
on arches like ppc64 where %pF expects a function descriptor.
Signed-off-by: NScott Wood <scottwood@freescale.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Use icnodehdr for struct xfs_da3_icnode_hdr instead of nodehdr
(already declared above).
Signed-off-by: NFabian Frederick <fabf@skynet.be>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

new_parent and src_is_directory are only used in 0/1 context.
Signed-off-by: NFabian Frederick <fabf@skynet.be>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

If xfs_filestream_get_parent() fails, we have a null pip,
goto out, and attempt to IRELE(NULL).  This causes a null
pointer dereference and BUG().

Fix this by directly returning NULLAGNUMBER in this case.
Reported-by: NAdrien Nader <adrien@notk.org>
Signed-off-by: NEric Sandeen <sandeen@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

This code is redundant now that we have verifiers that sanity check
the buffers as they are read from disk.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NCarlos Maiolino <cmaiolino@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Whiteouts are used by overlayfs -  it has a crazy convention that a
whiteout is a character device inode with a major:minor of 0:0.
Because it's not documented anywhere, here's an example of what
RENAME_WHITEOUT does on ext4:

# echo foo > /mnt/scratch/foo
# echo bar > /mnt/scratch/bar
# ls -l /mnt/scratch
total 24
-rw-r--r-- 1 root root     4 Feb 11 20:22 bar
-rw-r--r-- 1 root root     4 Feb 11 20:22 foo
drwx------ 2 root root 16384 Feb 11 20:18 lost+found
# src/renameat2 -w /mnt/scratch/foo /mnt/scratch/bar
# ls -l /mnt/scratch
total 20
-rw-r--r-- 1 root root     4 Feb 11 20:22 bar
c--------- 1 root root  0, 0 Feb 11 20:23 foo
drwx------ 2 root root 16384 Feb 11 20:18 lost+found
# cat /mnt/scratch/bar
foo
#

In XFS rename terms, the operation that has been done is that source
(foo) has been moved to the target (bar), which is like a nomal
rename operation, but rather than the source being removed, it have
been replaced with a whiteout.

We can't allocate whiteout inodes within the rename transaction due
to allocation being a multi-commit transaction: rename needs to
be a single, atomic commit. Hence we have several options here, form
most efficient to least efficient:

    - use DT_WHT in the target dirent and do no whiteout inode
      allocation.  The main issue with this approach is that we need
      hooks in lookup to create a virtual chardev inode to present
      to userspace and in places where we might need to modify the
      dirent e.g. unlink.  Overlayfs also needs to be taught about
      DT_WHT. Most invasive change, lowest overhead.

    - create a special whiteout inode in the root directory (e.g. a
      ".wino" dirent) and then hardlink every new whiteout to it.
      This means we only need to create a single whiteout inode, and
      rename simply creates a hardlink to it. We can use DT_WHT for
      these, though using DT_CHR means we won't have to modify
      overlayfs, nor anything in userspace. Downside is we have to
      look up the whiteout inode on every operation and create it if
      it doesn't exist.

    - copy ext4: create a special whiteout chardev inode for every
      whiteout.  This is more complex than the above options because
      of the lack of atomicity between inode creation and the rename
      operation, requiring us to create a tmpfile inode and then
      linking it into the directory structure during the rename. At
      least with a tmpfile inode crashes between the create and
      rename doesn't leave unreferenced inodes or directory
      pollution around.

By far the simplest thing to do in the short term is to copy ext4.
While it is the most inefficient way of supporting whiteouts, but as
an initial implementation we can simply reuse existing functions and
add a small amount of extra code the the rename operation.

When we get full whiteout support in the VFS (via the dentry cache)
we can then look to supporting DT_WHT method outlined as the first
method of supporting whiteouts. But until then, we'll stick with
what overlayfs expects us to be: dumb and stupid.
Signed-off-by: NDave Chinner <dchinner@redhat.com>

Now that xfs_finish_rename() exists, there is no reason for
xfs_cross_rename() to return to xfs_rename() to finish off the
rename transaction. Drive the completion code into
xfs_cross_rename() and handle all errors there so as to simplify
the xfs_rename() code.

Further, push the rename exchange target_ip check to early in the
rename code so as to make the error handling easy and obviously
correct.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NEric Sandeen <sandeen@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>