Currently, sysfs directory removal is inconsistent in that it would
remove any files directly under it but wouldn't recurse into
directories.  Thanks to group subdirectories, this doesn't even match
with kobject boundaries.  sysfs is in the process of being separated
out so that it can be used by multiple subsystems and we want to have
a consistent behavior - either removal of a sysfs_dirent should remove
every descendant entries or none instead of something inbetween.

This patch implements proper recursive removal in
__sysfs_remove_dir().  The function now walks its subtree in a
post-order walk to remove all descendants.

This is a behavior change but kobject / driver layer, which currently
is the only consumer, has already been updated to handle duplicate
removal attempts, so nothing should be broken after this change.
Signed-off-by: NTejun Heo <tj@kernel.org>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>

sysfs currently has a rather weird behavior regarding removals.  A
directory removal would delete all files directly under it but
wouldn't recurse into subdirectories, which, while a bit inconsistent,
seems to make sense at the first glance as each directory is
supposedly associated with a kobject and each kobject can take care of
the directory deletion; however, this doesn't really hold as we have
groups which can be directories without a kobject associated with it
and require explicit deletions.

We're in the process of separating out sysfs from kboject / driver
core and want a consistent behavior.  A removal should delete either
only the specified node or everything under it.  I think it is helpful
to support recursive atomic removal and later patches will implement
it.

Such change means that a sysfs_dirent associated with kobject may be
deleted before the kobject itself is removed if one of its ancestor
gets removed before it.  As sysfs_remove_dir() puts the base ref, we
may end up with dangling pointer on descendants.  This can be solved
by holding an extra reference on the sd from kobject.

Acquire an extra reference on the associated sysfs_dirent on directory
creation and put it after removal.
Signed-off-by: NTejun Heo <tj@kernel.org>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>

sysfs_addrm_start/finish() enclose sysfs_dirent additions and
deletions and sysfs_addrm_cxt is used to record information necessary
to finish the operations.  Currently, sysfs_addrm_start() takes
@parent_sd, records it in sysfs_addrm_cxt, and assumes that all
operations in the block are performed under that @parent_sd.

This assumption has been fine until now but we want to make some
operations behave recursively and, while having @parent_sd recorded in
sysfs_addrm_cxt doesn't necessarily prevents that, it becomes
confusing.

This patch removes sysfs_addrm_cxt->parent_sd and makes
sysfs_add_one() take an explicit @parent_sd parameter.  Note that
sysfs_remove_one() doesn't need the extra argument as its parent is
always known from the target @sd.

While at it, add __acquires/releases() notations to
sysfs_addrm_start/finish() respectively.

This patch doesn't make any functional difference.
Signed-off-by: NTejun Heo <tj@kernel.org>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>

Some internal sysfs functions which take explicit namespace argument
are weird in that they place the optional @ns in front of @name which
is contrary to the established convention.  This is confusing and
error-prone especially as @ns and @name may be interchanged without
causing compilation warning.

Swap the positions of @name and @ns in the following internal
functions.

 sysfs_find_dirent()
 sysfs_rename()
 sysfs_hash_and_remove()
 sysfs_name_hash()
 sysfs_name_compare()
 create_dir()

This patch doesn't introduce any functional changes.
Signed-off-by: NTejun Heo <tj@kernel.org>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Kay Sievers <kay@vrfy.org>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>

The pre-existing sysfs interfaces which take explicit namespace
argument are weird in that they place the optional @ns in front of
@name which is contrary to the established convention.  For example,
we end up forcing vast majority of sysfs_get_dirent() users to do
sysfs_get_dirent(parent, NULL, name), which is silly and error-prone
especially as @ns and @name may be interchanged without causing
compilation warning.

This renames sysfs_get_dirent() to sysfs_get_dirent_ns() and swap the
positions of @name and @ns, and sysfs_get_dirent() is now a wrapper
around sysfs_get_dirent_ns().  This makes confusions a lot less
likely.

There are other interfaces which take @ns before @name.  They'll be
updated by following patches.

This patch doesn't introduce any functional changes.

v2: EXPORT_SYMBOL_GPL() wasn't updated leading to undefined symbol
    error on module builds.  Reported by build test robot.  Fixed.
Signed-off-by: NTejun Heo <tj@kernel.org>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Kay Sievers <kay@vrfy.org>
Cc: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>

The way namespace tags are implemented in sysfs is more complicated
than necessary.  As each tag is a pointer value and required to be
non-NULL under a namespace enabled parent, there's no need to record
separately what type each tag is or where namespace is enabled.

If multiple namespace types are needed, which currently aren't, we can
simply compare the tag to a set of allowed tags in the superblock
assuming that the tags, being pointers, won't have the same value
across multiple types.  Also, whether to filter by namespace tag or
not can be trivially determined by whether the node has any tagged
children or not.

This patch rips out kobj_ns_type handling from sysfs.  sysfs no longer
cares whether specific type of namespace is enabled or not.  If a
sysfs_dirent has a non-NULL tag, the parent is marked as needing
namespace filtering and the value is tested against the allowed set of
tags for the superblock (currently only one but increasing this number
isn't difficult) and the sysfs_dirent is ignored if it doesn't match.

This removes most kobject namespace knowledge from sysfs proper which
will enable proper separation and layering of sysfs.  The namespace
sanity checks in fs/sysfs/dir.c are replaced by the new sanity check
in kobject_namespace().  As this is the only place ktype->namespace()
is called for sysfs, this doesn't weaken the sanity check
significantly.  I omitted converting the sanity check in
sysfs_do_create_link_sd().  While the check can be shifted to upper
layer, mistakes there are well contained and should be easily visible
anyway.
Signed-off-by: NTejun Heo <tj@kernel.org>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Kay Sievers <kay@vrfy.org>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>

There's no reason for sysfs to be calling ktype->namespace().  It is
backwards, obfuscates what's going on and unnecessarily tangles two
separate layers.

There are two places where symlink code calls ktype->namespace().

* sysfs_do_create_link_sd() calls it to find out the namespace tag of
  the target directory.  Unless symlinking races with cross-namespace
  renaming, this equals @target_sd->s_ns.

* sysfs_rename_link() uses it to find out the new namespace to rename
  to and the new namespace can be different from the existing one.
  The function is renamed to sysfs_rename_link_ns() with an explicit
  @ns argument and the ktype->namespace() invocation is shifted to the
  device layer.

While this patch replaces ktype->namespace() invocation with the
recorded result in @target_sd, this shouldn't result in any behvior
difference.
Signed-off-by: NTejun Heo <tj@kernel.org>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Kay Sievers <kay@vrfy.org>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>

For some unrecognizable reason, namespace information is communicated
to sysfs through ktype->namespace() callback when there's *nothing*
which needs the use of a callback.  The whole sequence of operations
is completely synchronous and sysfs operations simply end up calling
back into the layer which just invoked it in order to find out the
namespace information, which is completely backwards, obfuscates
what's going on and unnecessarily tangles two separate layers.

This patch doesn't remove ktype->namespace() but shifts its handling
to kobject layer.  We probably want to get rid of the callback in the
long term.

This patch adds an explicit param to sysfs_{create|rename|move}_dir()
and renames them to sysfs_{create|rename|move}_dir_ns(), respectively.
ktype->namespace() invocations are moved to the calling sites of the
above functions.  A new helper kboject_namespace() is introduced which
directly tests kobj_ns_type_operations->type which should give the
same result as testing sysfs_fs_type(parent_sd) and returns @kobj's
namespace tag as necessary.  kobject_namespace() is extern as it will
be used from another file in the following patches.

This patch should be an equivalent conversion without any functional
difference.
Signed-off-by: NTejun Heo <tj@kernel.org>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Kay Sievers <kay@vrfy.org>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>

sysfs ns (namespace) implementation became more convoluted than
necessary while trying to hide ns information from visible interface.
The relatively recent attr ns support is a good example.

* attr ns tag is determined by sysfs_ops->namespace() callback while
  dir tag is determined by kobj_type->namespace().  The placement is
  arbitrary.

* Instead of performing operations with explicit ns tag, the namespace
  callback is routed through sysfs_attr_ns(), sysfs_ops->namespace(),
  class_attr_namespace(), class_attr->namespace().  It's not simpler
  in any sense.  The only thing this convolution does is traversing
  the whole stack backwards.

The namespace callbacks are unncessary because the operations involved
are inherently synchronous.  The information can be provided in in
straight-forward top-down direction and reversing that direction is
unnecessary and against basic design principles.

This backward interface is unnecessarily convoluted and hinders
properly separating out sysfs from driver model / kobject for proper
layering.  This patch updates attr ns support such that

* sysfs_ops->namespace() and class_attr->namespace() are dropped.

* sysfs_{create|remove}_file_ns(), which take explicit @ns param, are
  added and sysfs_{create|remove}_file() are now simple wrappers
  around the ns aware functions.

* ns handling is dropped from sysfs_chmod_file().  Nobody uses it at
  this point.  sysfs_chmod_file_ns() can be added later if necessary.

* Explicit @ns is propagated through class_{create|remove}_file_ns()
  and netdev_class_{create|remove}_file_ns().

* driver/net/bonding which is currently the only user of attr
  namespace is updated to use netdev_class_{create|remove}_file_ns()
  with @bh->net as the ns tag instead of using the namespace callback.

This patch should be an equivalent conversion without any functional
difference.  It makes the code easier to follow, reduces lines of code
a bit and helps proper separation and layering.
Signed-off-by: NTejun Heo <tj@kernel.org>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Kay Sievers <kay@vrfy.org>
Acked-by: NDavid S. Miller <davem@davemloft.net>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>

The expansion of to_sysfs_dirent() contains an unncessary trailing
semicolon making it impossible to use in the middle of statements.
Drop it.
Signed-off-by: NTejun Heo <tj@kernel.org>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>

Commit f5ea1100 cleans up the disk to host conversions for
node directory entries, but because a variable is reused in
xfs_node_toosmall() the next node is not correctly found.
If the original node is small enough (<= 3/8 of the node size),
this change may incorrectly cause a node collapse when it should
not. That will cause an assert in xfstest generic/319:

   Assertion failed: first <= last && last < BBTOB(bp->b_length),
   file: /root/newest/xfs/fs/xfs/xfs_trans_buf.c, line: 569

Keep the original node header to get the correct forward node.

(When a node is considered for a merge with a sibling, it overwrites the
 sibling pointers of the original incore nodehdr with the sibling's
 pointers.  This leads to loop considering the original node as a merge
 candidate with itself in the second pass, and so it incorrectly
 determines a merge should occur.)
Signed-off-by: NMark Tinguely <tinguely@sgi.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

[v3: added Dave Chinner's (slightly modified) suggestion to the commit header,
	cleaned up whitespace.  -bpm]

While printing 32-bit node numbers, an 8-byte string is not enough.
Increase the size of the string to 12 chars.

This got left out in commit 49fa8140 ("fs/ocfs2/super.c: Use bigger
nodestr to accomodate 32-bit node numbers").
Signed-off-by: NGoldwyn Rodrigues <rgoldwyn@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The memcpy() in bio_copy_data() was using the wrong offset vars, leading
to data corruption in weird unusual setups.
Signed-off-by: NKent Overstreet <kmo@daterainc.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: linux-stable <stable@vger.kernel.org> # >= v3.9
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

After a fair number of xfstests runs, xfs/182 started to fail
regularly with a corrupted directory - a directory read verifier was
failing after recovery because it found a block with a XARM magic
number (remote attribute block) rather than a directory data block.

The first time I saw this repeated failure I did /something/ and the
problem went away, so I was never able to find the underlying
problem. Test xfs/182 failed again today, and I found the root
cause before I did /something else/ that made it go away.

Tracing indicated that the block in question was being correctly
logged, the log was being flushed by sync, but the buffer was not
being written back before the shutdown occurred. Tracing also
indicated that log recovery was also reading the block, but then
never writing it before log recovery invalidated the cache,
indicating that it was not modified by log recovery.

More detailed analysis of the corpse indicated that the filesystem
had a uuid of "a4131074-1872-4cac-9323-2229adbcb886" but the XARM
block had a uuid of "8f32f043-c3c9-e7f8-f947-4e7f989c05d3", which
indicated it was a block from an older filesystem. The reason that
log recovery didn't replay it was that the LSN in the XARM block was
larger than the LSN of the transaction being replayed, and so the
block was not overwritten by log recovery.

Hence, log recovery cant blindly trust the magic number and LSN in
the block - it must verify that it belongs to the filesystem being
recovered before using the LSN. i.e. if the UUIDs don't match, we
need to unconditionally recovery the change held in the log.

This patch was first tested on a block device that was repeatedly
causing xfs/182 to fail with the same failure on the same block with
the same directory read corruption signature (i.e. XARM block). It
did not fail, and hasn't failed since.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

It uses a kernel internal structure in it's definition rather than
the user visible structure that is passed to the ioctl.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

When we free an inode, we do so via RCU. As an RCU lookup can occur
at any time before we free an inode, and that lookup takes the inode
flags lock, we cannot safely assert that the flags lock is not held
just before marking it dead and running call_rcu() to free the
inode.

We check on allocation of a new inode structre that the lock is not
held, so we still have protection against locks being leaked and
hence not correctly initialised when allocated out of the slab.
Hence just remove the assert...
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Regression introduced by commit 46f9d2eb ("xfs: aborted buf items can
be in the AIL") which fails to lock the AIL before removing the
item. Spinlock debugging throws a warning about this.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

There are two locks involved in managing the journal lists. The general
reiserfs_write_lock and the journal->j_flush_mutex.

While flush_journal_list is sleeping to acquire the j_flush_mutex or to
submit a block for write, it will drop the write lock. This allows
another thread to acquire the write lock and ultimately call
flush_used_journal_lists to traverse the list of journal lists and
select one for flushing. It can select the journal_list that has just
had flush_journal_list called on it in the original thread and call it
again with the same journal_list.

The second thread then drops the write lock to acquire j_flush_mutex and
the first thread reacquires it and continues execution and eventually
clears and frees the journal list before dropping j_flush_mutex and
returning.

The second thread acquires j_flush_mutex and ends up operating on a
journal_list that has already been released. If the memory hasn't
been reused, we'll soon after hit a BUG_ON because the transaction id
has already been cleared. If it's been reused, we'll crash in other
fun ways.

Since flush_journal_list will synchronize on j_flush_mutex, we can fix
the race by taking a proper reference in flush_used_journal_lists
and checking to see if it's still valid after the mutex is taken. It's
safe to iterate the list of journal lists and pick a list with
just the write lock as long as a reference is taken on the journal list
before we drop the lock. We already have code to handle whether a
transaction has been flushed already so we can use that to handle the
race and get rid of the trans_id BUG_ON.
Signed-off-by: NJeff Mahoney <jeffm@suse.com>
Signed-off-by: NJan Kara <jack@suse.cz>

Commit a3172027 introduced test_transaction as a requirement for
flushing old lists -- but it can never return 1 unless the transaction
has already been flushed.

As a result, we have a routine that iterates the j_realblocks list but
doesn't actually do anything. Since it's been this way since 2006 and
the latency numbers were what Chris expected, let's just rip it out.
Signed-off-by: NJeff Mahoney <jeffm@suse.com>
Signed-off-by: NJan Kara <jack@suse.cz>

A user has reported an oops in udf_statfs() that was caused by
numOfPartitions entry in LVID structure being corrupted. Fix the problem
by verifying whether numOfPartitions makes sense at least to the extent
that LVID fits into a single block as it should.
Reported-by: NJuergen Weigert <jw@suse.com>
Signed-off-by: NJan Kara <jack@suse.cz>

Users have been complaining of the uuid tree stuff warning that there is no uuid
root when trying to do snapshot operations.  This is because if you mount -o ro
we will not create the uuid tree.  But then if you mount -o rw,remount we will
still not create it and then any subsequent snapshot/subvol operations you try
to do will fail gloriously.  Fix this by creating the uuid_root on remount rw if
it was not already there.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

btrfs_ioctl_file_extent_same() uses __put_user_unaligned() to copy some data
back to it's argument struct. Unfortunately, not all architectures provide
__put_user_unaligned(), so compiles break on them if btrfs is selected.

Instead, just copy the whole struct in / out at the start and end of
operations, respectively.
Signed-off-by: NMark Fasheh <mfasheh@suse.de>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

Commit 2bc55652 (Btrfs: don't update atime on
RO subvolumes) ensures that the access time of an inode is not updated when
the inode lives in a read-only subvolume.
However, if a directory on a read-only subvolume is accessed, the atime is
updated. This results in a write operation to a read-only subvolume. I
believe that access times should never be updated on read-only subvolumes.

To reproduce:

 # mkfs.btrfs -f /dev/dm-3
 (...)
 # mount /dev/dm-3 /mnt
 # btrfs subvol create /mnt/sub
 	Create subvolume '/mnt/sub'
 # mkdir /mnt/sub/dir
 # echo "abc" > /mnt/sub/dir/file
 # btrfs subvol snapshot -r /mnt/sub /mnt/rosnap
 	Create a readonly snapshot of '/mnt/sub' in '/mnt/rosnap'
 # stat /mnt/rosnap/dir
 	File: `/mnt/rosnap/dir'
 	Size: 8         Blocks: 0          IO Block: 4096   directory
 Device: 16h/22d    Inode: 257         Links: 1
 Access: (0755/drwxr-xr-x)  Uid: (    0/    root)   Gid: (    0/    root)
 	Access: 2013-09-11 07:21:49.389157126 -0400
 	Modify: 2013-09-11 07:22:02.330156079 -0400
 	Change: 2013-09-11 07:22:02.330156079 -0400
 # ls /mnt/rosnap/dir
 	file
 # stat /mnt/rosnap/dir
 	File: `/mnt/rosnap/dir'
 	Size: 8         Blocks: 0          IO Block: 4096   directory
 Device: 16h/22d    Inode: 257         Links: 1
 Access: (0755/drwxr-xr-x)  Uid: (    0/    root)   Gid: (    0/    root)
 	Access: 2013-09-11 07:22:56.797151670 -0400
 	Modify: 2013-09-11 07:22:02.330156079 -0400
 	Change: 2013-09-11 07:22:02.330156079 -0400
Reported-by: NKoen De Wit <koen.de.wit@oracle.com>
Signed-off-by: NGuangyu Sun <guangyu.sun@oracle.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

The kernel log entries for device label %s and device fsid %pU
are missing the btrfs: prefix. Add those here.
Signed-off-by: NFrank Holton <fholton@gmail.com>
Reviewed-by: NDavid Sterba <dsterba@suse.cz>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

It's still possible to flip the filesystem into RW mode after it's
remounted RO due to an abort. There are lots of places that check for
the superblock error bit and will not write data, but we should not let
the filesystem appear read-write.
Signed-off-by: NDavid Sterba <dsterba@suse.cz>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

This patch makes it possible to set BTRFS_FS_TREE_OBJECTID as the default
subvolume by passing a subvolume id of 0.
Signed-off-by: Nchandan <chandan@linux.vnet.ibm.com>
Reviewed-by: NDavid Sterba <dsterba@suse.cz>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

In btrfs_sync_file(), if the call to btrfs_log_dentry_safe() returns
a negative error (for e.g. -ENOMEM via btrfs_log_inode()), we would
return without ending/freeing the transaction.
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

The BUG() was replaced by btrfs_error() and return -EIO with the
patch "get rid of one BUG() in write_all_supers()", but the missing
mutex_unlock() was overlooked.

The 0-DAY kernel build service from Intel reported the missing
unlock which was found by the coccinelle tool:

    fs/btrfs/disk-io.c:3422:2-8: preceding lock on line 3374
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

We don't do the iput when we fail to allocate our delayed delalloc work in
__start_delalloc_inodes, fix this.
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

This isn't used for anything anymore, just remove it.
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

This is a left over of how we used to wait for ordered extents, which was to
grab the inode and then run filemap flush on it.  However if we have an ordered
extent then we already are holding a ref on the inode, and we just use
btrfs_start_ordered_extent anyway, so there is no reason to have an extra ref on
the inode to start work on the ordered extent.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

Forever ago I made the worst case calculator say that we could potentially split
into 3 blocks for every level on the way down, which isn't right.  If we split
we're only going to get two new blocks, the one we originally cow'ed and the new
one we're going to split.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

This reverts commit 70afa399.  It is causing
performance issues and wasn't actually correct.  There were problems with the
way we flushed delalloc and that was the real cause of the early enospc.
Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

Various people have hit a deadlock when running btrfs/011.  This is because when
replacing nocow extents we will take the i_mutex to make sure nobody messes with
the file while we are replacing the extent.  The problem is we are already
holding a transaction open, which is a locking inversion, so instead we need to
save these inodes we find and then process them outside of the transaction.

Further we can't just lock the inode and assume we are good to go.  We need to
lock the extent range and then read back the extent cache for the inode to make
sure the extent really still points at the physical block we want.  If it
doesn't we don't have to copy it.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

So if we have dir_index items in the log that means we also have the inode item
as well, which means that the inode's i_size is correct.  However when we
process dir_index'es we call btrfs_add_link() which will increase the
directory's i_size for the new entry.  To fix this we need to just set the dir
items i_size to 0, and then as we find dir_index items we adjust the i_size.
btrfs_add_link() will do it for new entries, and if the entry already exists we
can just add the name_len to the i_size ourselves.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

A user reported a bug where his log would not replay because he was getting
-EEXIST back.  This was because he had a file moved into a directory that was
logged.  What happens is the file had a lower inode number, and so it is
processed first when replaying the log, and so we add the inode ref in for the
directory it was moved to.  But then we process the directories DIR_INDEX item
and try to add the inode ref for that inode and it fails because we already
added it when we replayed the inode.  To solve this problem we need to just
process any DIR_INDEX items we have in the log first so this all is taken care
of, and then we can replay the rest of the items.  With this patch my reproducer
can remount the file system properly instead of erroring out.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

Liu introduced a local copy of the last log commit for an inode to make sure we
actually log an inode even if a log commit has already taken place.  In order to
make sure we didn't relog the same inode multiple times he set this local copy
to the current trans when we log the inode, because usually we log the inode and
then sync the log.  The exception to this is during rename, we will relog an
inode if the name changed and it is already in the log.  The problem with this
is then we go to sync the inode, and our check to see if the inode has already
been logged is tripped and we don't sync the log.  To fix this we need to _also_
check against the roots last log commit, because it could be less than what is
in our local copy of the log commit.  This fixes a bug where we rename a file
into a directory and then fsync the directory and then on remount the directory
is no longer there.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

If you just create a directory and then fsync that directory and then pull the
power plug you will come back up and the directory will not be there.  That is
because we won't actually create directories if we've logged files inside of
them since they will be created on replay, but in this check we will set our
logged_trans of our current directory if it happens to be a directory, making us
think it doesn't need to be logged.  Fix the logic to only do this to parent
directories.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

So forever we have had this thing to limit the amount of delalloc pages we'll
setup to be written out to 128mb.  This is because we have to lock all the pages
in this range, so anything above this gets a bit unweildly, and also without a
limit we'll happily allocate gigantic chunks of disk space.  Turns out our check
for this wasn't quite right, we wouldn't actually limit the chunk we wanted to
write out, we'd just stop looking for more space after we went over the limit.
So if you do a giant 20gb dd on my box with lots of ram I could get 2gig
extents.  This is fine normally, except when you go to relocate these extents
and we can't find enough space to relocate these moster extents, since we have
to be able to allocate exactly the same sized extent to move it around.  So fix
this by actually enforcing the limit.  With this patch I'm no longer seeing
giant 1.5gb extents.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

By the current code, if the requested size is very large, and all the extents
in the free space cache are small, we will waste lots of the cpu time to cut
the requested size in half and search the cache again and again until it gets
down to the size the allocator can return. In fact, we can know the max extent
size in the cache after the first search, so we needn't cut the size in half
repeatedly, and just use the max extent size directly. This way can save
lots of cpu time and make the performance grow up when there are only fragments
in the free space cache.

According to my test, if there are only 4KB free space extents in the fs,
and the total size of those extents are 256MB, we can reduce the execute
time of the following test from 5.4s to 1.4s.
  dd if=/dev/zero of=<testfile> bs=1MB count=1 oflag=sync

Changelog v2 -> v3:
- fix the problem that we skip the block group with the space which is
  less than we need.

Changelog v1 -> v2:
- address the problem that we return a wrong start position when searching
  the free space in a bitmap.
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>