This is a large change for adding compression on reading and writing,
both for inline and regular extents.  It does some fairly large
surgery to the writeback paths.

Compression is off by default and enabled by mount -o compress.  Even
when the -o compress mount option is not used, it is possible to read
compressed extents off the disk.

If compression for a given set of pages fails to make them smaller, the
file is flagged to avoid future compression attempts later.

* While finding delalloc extents, the pages are locked before being sent down
to the delalloc handler.  This allows the delalloc handler to do complex things
such as cleaning the pages, marking them writeback and starting IO on their
behalf.

* Inline extents are inserted at delalloc time now.  This allows us to compress
the data before inserting the inline extent, and it allows us to insert
an inline extent that spans multiple pages.

* All of the in-memory extent representations (extent_map.c, ordered-data.c etc)
are changed to record both an in-memory size and an on disk size, as well
as a flag for compression.

From a disk format point of view, the extent pointers in the file are changed
to record the on disk size of a given extent and some encoding flags.
Space in the disk format is allocated for compression encoding, as well
as encryption and a generic 'other' field.  Neither the encryption or the
'other' field are currently used.

In order to limit the amount of data read for a single random read in the
file, the size of a compressed extent is limited to 128k.  This is a
software only limit, the disk format supports u64 sized compressed extents.

In order to limit the ram consumed while processing extents, the uncompressed
size of a compressed extent is limited to 256k.  This is a software only limit
and will be subject to tuning later.

Checksumming is still done on compressed extents, and it is done on the
uncompressed version of the data.  This way additional encodings can be
layered on without having to figure out which encoding to checksum.

Compression happens at delalloc time, which is basically singled threaded because
it is usually done by a single pdflush thread.  This makes it tricky to
spread the compression load across all the cpus on the box.  We'll have to
look at parallel pdflush walks of dirty inodes at a later time.

Decompression is hooked into readpages and it does spread across CPUs nicely.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

This improves the comments at the top of many functions.  It didn't
dive into the guts of functions because I was trying to
avoid merging problems with the new allocator and back reference work.

extent-tree.c and volumes.c were both skipped, and there is definitely
more work todo in cleaning and commenting the code.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

btrfs-vol -a /dev/xxx will zero the first and last two MB of the device.
The kernel code needs to wait for this IO to finish before it adds
the device.

btrfs metadata IO does not happen through the block device inode.  A
separate address space is used, allowing the zero filled buffer heads in
the block device inode to be written to disk after FS metadata starts
going down to the disk via the btrfs metadata inode.

The end result is zero filled metadata blocks after adding new devices
into the filesystem.

The fix is a simple filemap_write_and_wait on the block device inode
before actually inserting it into the pool of available devices.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

This patch updates the space balancing code to utilize the new
backref format.  Before, btrfs-vol -b would break any COW links
on data blocks or metadata.  This was slow and caused the amount
of space used to explode if a large number of snapshots were present.

The new code can keeps the sharing of all data extents and
most of the tree blocks.

To maintain the sharing of data extents, the space balance code uses
a seperate inode hold data extent pointers, then updates the references
to point to the new location.

To maintain the sharing of tree blocks, the space balance code uses
reloc trees to relocate tree blocks in reference counted roots.
There is one reloc tree for each subvol, and all reloc trees share
same root key objectid. Reloc trees are snapshots of the latest
committed roots of subvols (root->commit_root).

To relocate a tree block referenced by a subvol, there are two steps.
COW the block through subvol's reloc tree, then update block pointer in
the subvol to point to the new block. Since all reloc trees share
same root key objectid, doing special handing for tree blocks
owned by them is easy. Once a tree block has been COWed in one
reloc tree, we can use the resulting new block directly when the
same block is required to COW again through other reloc trees.
In this way, relocated tree blocks are shared between reloc trees,
so they are also shared between subvols.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Btrfs has a cache of reference counts in leaves, allowing it to
avoid reading tree leaves while deleting snapshots.  To reduce
contention with multiple subvolumes, this cache is private to each
subvolume.

This patch adds shared reference cache support. The new space
balancing code plays with multiple subvols at the same time, So
the old per-subvol reference cache is not well suited.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Btrfs metadata writeback is fairly expensive.  Once a tree block is written
it must be cowed before it can be changed again.  The btree writepages
code has a threshold based on a count of dirty btree bytes which is
updated as IO is sent out.

This changes btree_writepages to skip the writeout if there are less
than 32MB of dirty bytes from the btrees, improving performance
across many workloads.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Btrfs had compatibility code for kernels back to 2.6.18.  These have
been removed, and will be maintained in a separate backport
git tree from now on.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

This patch makes the back reference system to explicit record the
location of parent node for all types of extents. The location of
parent node is placed into the offset field of backref key. Every
time a tree block is balanced, the back references for the affected
lower level extents are updated.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Tree blocks were using async bio submission, but the sum was still
being done directly during writepage.  This moves the checksumming
into the worker thread.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

1) replace the per fs_info extent_io_tree that tracked free space with two
rb-trees per block group to track free space areas via offset and size.  The
reason to do this is because most allocations come with a hint byte where to
start, so we can usually find a chunk of free space at that hint byte to satisfy
the allocation and get good space packing.  If we cannot find free space at or
after the given offset we fall back on looking for a chunk of the given size as
close to that given offset as possible.  When we fall back on the size search we
also try to find a slot as close to the size we want as possible, to avoid
breaking small chunks off of huge areas if possible.

2) remove the extent_io_tree that tracked the block group cache from fs_info and
replaced it with an rb-tree thats tracks block group cache via offset.  also
added a per space_info list that tracks the block group cache for the particular
space so we can lookup related block groups easily.

3) cleaned up the allocation code to make it a little easier to read and a
little less complicated.  Basically there are 3 steps, first look from our
provided hint.  If we couldn't find from that given hint, start back at our
original search start and look for space from there.  If that fails try to
allocate space if we can and start looking again.  If not we're screwed and need
to start over again.

4) small fixes.  there were some issues in volumes.c where we wouldn't allocate
the rest of the disk.  fixed cow_file_range to actually pass the alloc_hint,
which has helped a good bit in making the fs_mark test I run have semi-normal
results as we run out of space.  Generally with data allocations we don't track
where we last allocated from, so everytime we did a data allocation we'd search
through every block group that we have looking for free space.  Now searching a
block group with no free space isn't terribly time consuming, it was causing a
slight degradation as we got more data block groups.  The alloc_hint has fixed
this slight degredation and made things semi-normal.

There is still one nagging problem I'm working on where we will get ENOSPC when
there is definitely plenty of space.  This only happens with metadata
allocations, and only when we are almost full.  So you generally hit the 85%
mark first, but sometimes you'll hit the BUG before you hit the 85% wall.  I'm
still tracking it down, but until then this seems to be pretty stable and make a
significant performance gain.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

I had incorrectly disabled the check for the block number being correct
in the header block.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

This is the same way the transaction code makes sure that all the
other tree blocks are safely on disk.  There's an extent_io tree
for each root, and any blocks allocated to the tree logs are
recorded in that tree.

At tree-log sync, the extent_io tree is walked to flush down the
dirty pages and wait for them.

The main benefit is less time spent walking the tree log and skipping
clean pages, and getting sequential IO down to the drive.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Since tree log blocks get freed every transaction, they never really
need to be written to disk.  This skips the step where we update
metadata to record they were allocated.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

* Pin down data blocks to prevent them from being reallocated like so:

trans 1: allocate file extent
trans 2: free file extent
trans 3: free file extent during old snapshot deletion
trans 3: allocate file extent to new file
trans 3: fsync new file

Before the tree logging code, this was legal because the fsync
would commit the transation that did the final data extent free
and the transaction that allocated the extent to the new file
at the same time.

With the tree logging code, the tree log subtransaction can commit
before the transaction that freed the extent.  If we crash,
we're left with two different files using the extent.

* Don't wait in start_transaction if log replay is going on.  This
avoids deadlocks from iput while we're cleaning up link counts in the
replay code.

* Don't deadlock in replay_one_name by trying to read an inode off
the disk while holding paths for the directory

* Hold the buffer lock while we mark a buffer as written.  This
closes a race where someone is changing a buffer while we write it.
They are supposed to mark it dirty again after they change it, but
this violates the cow rules.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

File syncs and directory syncs are optimized by copying their
items into a special (copy-on-write) log tree.  There is one log tree per
subvolume and the btrfs super block points to a tree of log tree roots.

After a crash, items are copied out of the log tree and back into the
subvolume.  See tree-log.c for all the details.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

The current code waits for the count of async bio submits to get below
a given threshold if it is too high right after adding the latest bio
to the work queue.  This isn't optimal because the caller may have
sequential adjacent bios pending they are waiting to send down the pipe.

This changeset requires the caller to wait on the async bio count,
and changes the async checksumming submits to wait for async bios any
time they self throttle.

The end result is much higher sequential throughput.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Before, the btrfs bdi congestion function was used to test for too many
async bios.  This keeps that check to throttle pdflush, but also
adds a check while queuing bios.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

This takes the csum mutex deeper in the call chain and releases it
more often.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Before this change, btrfs would use a bdi congestion function to make
sure there weren't too many pending async checksum work items.

This change makes the process creating async work items wait instead,
leading to fewer congestion returns from the bdi.  This improves
pdflush background_writeout scanning.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

After writing out all the remaining btree blocks in the transaction,
the commit code would use filemap_fdatawait to make sure it was all
on disk.  This means it would wait for blocks written by other procs
as well.

The new code walks the list of blocks for this transaction again
and waits only for those required by this transaction.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

* Make walk_down_tree wake up throttled tasks more often
* Make walk_down_tree call cond_resched during long loops
* As the size of the ref cache grows, wait longer in throttle
* Get rid of the reada code in walk_down_tree, the leaves don't get
  read anymore, thanks to the ref cache.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Large streaming reads make for large bios, which means each entry on the
list async work queues represents a large amount of data.  IO
congestion throttling on the device was kicking in before the async
worker threads decided a single thread was busy and needed some help.

The end result was that a streaming read would result in a single CPU
running at 100% instead of balancing the work off to other CPUs.

This patch also changes the pre-IO checksum lookup done by reads to
work on a per-bio basis instead of a per-page.  This results in many
extra btree lookups on large streaming reads.  Doing the checksum lookup
right before bio submit allows us to reuse searches while processing
adjacent offsets.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

The memory reclaiming issue happens when snapshot exists. In that
case, some cache entries may not be used during old snapshot dropping,
so they will remain in the cache until umount.

The patch adds a field to struct btrfs_leaf_ref to record create time. Besides,
the patch makes all dead roots of a given snapshot linked together in order of
create time. After a old snapshot was completely dropped, we check the dead
root list and remove all cache entries created before the oldest dead root in
the list.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

It was incorrectly clearing the up to date flag on the buffer even
when the buffer properly verified.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

A large reference cache is directly related to a lot of work pending
for the cleaner thread.  This throttles back new operations based on
the size of the reference cache so the cleaner thread will be able to keep
up.

Overall, this actually makes the FS faster because the cleaner thread will
be more likely to find things in cache.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

This changes the reference cache to make a single cache per root
instead of one cache per transaction, and to key by the byte number
of the disk block instead of the keys inside.

This makes it much less likely to have cache misses if a snapshot
or something has an extra reference on a higher node or a leaf while
the first transaction that added the leaf into the cache is dropping.

Some throttling is added to functions that free blocks heavily so they
wait for old transactions to drop.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Much of the IO done while dropping snapshots is done looking up
leaves in the filesystem trees to see if they point to any extents and
to drop the references on any extents found.

This creates a cache so that IO isn't required.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

Before setting an extent to delalloc, the code needs to wait for
pending ordered extents.

Also, the relocation code needs to wait for ordered IO before scanning
the block group again.  This is because the extents are not removed
until the IO for the new extents is finished
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Checksum items are not inserted into the tree until all of the io from a
given extent is complete.  This means one dirty page from an extent may
be written, freed, and then read again before the entire extent is on disk
and the checksum item is inserted.

The checksums themselves are stored in the ordered extent so they can
be inserted in bulk when IO is complete.  On read, if a checksum item isn't
found, the ordered extents were being searched for a checksum record.

This all worked most of the time, but the checksum insertion code tries
to reduce the number of tree operations by pre-inserting checksum items
based on i_size and a few other factors.  This means the read code might
find a checksum item that hasn't yet really been filled in.

This commit changes things to check the ordered extents first and only
dive into the btree if nothing was found.  This removes the need for
extra locking and is more reliable.
Signed-off-by: NChris Mason <chris.mason@oracle.com>