The btrfs snapshotting code requires that once a root has been
snapshotted, we don't change it during a commit.

But there are two cases to lead to tree corruptions:

1) multi-thread snapshots can commit serveral snapshots in a transaction,
   and this may change the src root when processing the following pending
   snapshots, which lead to the former snapshots corruptions;

2) the free inode cache was changing the roots when it root the cache,
   which lead to corruptions.

This fixes things by making sure we force COW the block after we create a
snapshot during commiting a transaction, then any changes to the roots
will result in COW, and we get all the fs roots and snapshot roots to be
consistent.
Signed-off-by: NLiu Bo <liubo2009@cn.fujitsu.com>
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

We can not do flushable reservation for the relocation when we create snapshot,
because it may make the transaction commit task and the flush task wait for
each other and the deadlock happens.
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

While we're allocating ram for a new transaction, we drop our spinlock.
When we get the lock back, we do check to see if a transaction started
while we slept, but we don't check to make sure it isn't blocked
because a commit has already started.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Failure testing was tripping up over stale PageError bits in
metadata pages.  If we have an io error on a block, and later on
end up reusing it, nobody ever clears PageError on those pages.

During commit, we'll find PageError and think we had trouble writing
the block, which will lead to aborts and other problems.

This changes clean_tree_block and the btrfs writepage code to
clear the PageError bit.  In both cases we're either completely
done with the page or the page has good stuff and the error bit
is no longer valid.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

fs_info has now ~9kb, more than fits into one page. This will cause
mount failure when memory is too fragmented. Top space consumers are
super block structures super_copy and super_for_commit, ~2.8kb each.
Allocate them dynamically. fs_info will be ~3.5kb. (measured on x86_64)

Add a wrapper for freeing fs_info and all of it's dynamically allocated
members.
Signed-off-by: NDavid Sterba <dsterba@suse.cz>

Currently btrfs_block_rsv_check does 2 things, it will either refill a block
reserve like in the truncate or refill case, or it will check to see if there is
enough space in the global reserve and possibly refill it.  However because of
overcommit we could be well overcommitting ourselves just to try and refill the
global reserve, when really we should just be committing the transaction.  So
breack this out into btrfs_block_rsv_refill and btrfs_block_rsv_check.  Refill
will try to reserve more metadata if it can and btrfs_block_rsv_check will not,
it will only tell you if the factor of the total space is still reserved.
Thanks,
Signed-off-by: NJosef Bacik <josef@redhat.com>

We started setting trans->block_rsv = NULL to allow the delayed refs flushing
stuff to use the right block_rsv and then just made
btrfs_trans_release_metadata() unconditionally use the trans block rsv.  The
problem with this is we need to reserve some space in the transaction and then
migrate it to the global block rsv, so we need to be able to free that out
properly.  So instead just move btrfs_trans_release_metadata() before the
delayed ref flushing and use trans->block_rsv for the freeing.  Thanks,
Signed-off-by: NJosef Bacik <josef@redhat.com>

Some users have requested this and I've found I needed a way to disable cache
loading without actually clearing the cache, so introduce the no_space_cache
option.  Before we check the super blocks cache generation field and if it was
populated we always turned space caching on.  Now we check this and set the
space cache option on, and then parse the mount options so that if we want it
off it get's turned off.  Then we check the mount option all the places we do
the caching work instead of checking the super's cache generation.  This makes
things more consistent and lets us turn space caching off.  Thanks,
Signed-off-by: NJosef Bacik <josef@redhat.com>

While looking for a performance regression a user was complaining about, I
noticed that we had a regression with the varmail test of filebench.  This was
introduced by

0d10ee2e

which keeps us from calling writepages in writepage.  This is a correct change,
however it happens to help the varmail test because we write out in larger
chunks.  This is largly to do with how we write out dirty pages for each
transaction.  If you run filebench with

load varmail
set $dir=/mnt/btrfs-test
run 60

prior to this patch you would get ~1420 ops/second, but with the patch you get
~1200 ops/second.  This is a 16% decrease.  So since we know the range of dirty
pages we want to write out, don't write out in one page chunks, write out in
ranges.  So to do this we call filemap_fdatawrite_range() on the range of bytes.
Then we convert the DIRTY extents to NEED_WAIT extents.  When we then call
btrfs_wait_marked_extents() we only have to filemap_fdatawait_range() on that
range and clear the NEED_WAIT extents.  This doesn't get us back to our original
speeds, but I've been seeing ~1380 ops/second, which is a <5% regression as
opposed to a >15% regression.  That is acceptable given that the original commit
greatly reduces our latency to begin with.  Thanks,
Signed-off-by: NJosef Bacik <josef@redhat.com>

Checksums are charged in 2 different ways.  The first case is when we're writing
to the disk, we account for the new checksums with the delalloc block rsv.  In
order for this to work we check if we're allocating a block for the csum root
and if trans->block_rsv == the delalloc block rsv.  But when we're deleting the
csums because of cow, this is charged to the global block rsv, and is done when
we run the delayed refs.  So we need to make sure that trans->block_rsv == NULL
when running the delayed refs.  So set it to NULL and reset it in
should_end_transaction, and set it to NULL in commit_transaction.  This got rid
of the ridiculous amount of warnings I was seeing when trying to do a balance.
Thanks,
Signed-off-by: NJosef Bacik <josef@redhat.com>

The only thing that we need to have a trans handle for is in
reserve_metadata_bytes and thats to know how much flushing we can do.  So
instead of passing it around, just check current->journal_info for a
trans_handle so we know if we can commit a transaction to try and free up space
or not.  Thanks,
Signed-off-by: NJosef Bacik <josef@redhat.com>

The alloc warnings everybody has been seeing is because we have been reserving
space for csums, but we weren't actually using that space.  So make
get_block_rsv() return the trans->block_rsv if we're modifying the csum root.
Also set the trans->block_rsv to NULL so that if we modify the csum root when
running delayed ref's that comes out of the global reserve like it's supposed
to.  With this patch I'm not seeing those alloc warnings anymore.  Thanks,
Signed-off-by: NJosef Bacik <josef@redhat.com>

If you run xfstest 224 it you will get lots of messages about not being able to
delete inodes and that they will be cleaned up next mount.  This is because
btrfs_block_rsv_check was not calling reserve_metadata_bytes with the ability to
flush, so if there was not enough space, it simply failed.  But in truncate and
evict case we could easily flush space to try and get enough space to do our
work, so make btrfs_block_rsv_check take a flush argument to pass down to
reserve_metadata_bytes.  Now xfstests 224 runs fine without all those
complaints.  Thanks,
Signed-off-by: NJosef Bacik <josef@redhat.com>

This patch kills off the calculation for the amount of space needed for the
orphan operations during a snapshot.  The thing is we only do snapshots on
commit, so any space that is in the block_rsv->freed[] isn't going to be in the
new snapshot anyway, so there isn't any reason to require that space to be
reserved for the snapshot to occur.  Thanks,
Signed-off-by: NJosef Bacik <josef@redhat.com>

At the beginning of create_pending_snapshot, trans->block_rsv is set
to pending->block_rsv and is used for snapshot things, however, when
it is done, we do not recover it as will.
Signed-off-by: NLiu Bo <liubo2009@cn.fujitsu.com>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

wait_for_commit() always returns 0.
Signed-off-by: NLi Zefan <lizf@cn.fujitsu.com>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Use wait_event() when possible to avoid code duplication.
Signed-off-by: NLi Zefan <lizf@cn.fujitsu.com>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Hit this nice little deadlock.  What happens is this

__btrfs_end_transaction with throttle set, --use_count so it equals 0
  btrfs_commit_transaction
    <somebody else actually manages to start the commit>
    btrfs_end_transaction --use_count so now its -1 <== BAD
      we just return and wait on the transaction

This is bad because we just return after our use_count is -1 and don't let go
of our num_writer count on the transaction, so the guy committing the
transaction just sits there forever.  Fix this by inc'ing our use_count if we're
going to call commit_transaction so that if we call btrfs_end_transaction it's
valid.  Thanks,
Signed-off-by: NJosef Bacik <josef@redhat.com>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

We have to do weird things when handling enospc in the transaction joining code.
Because we've already joined the transaction we cannot commit the transaction
within the reservation code since it will deadlock, so we have to return EAGAIN
and then make sure we don't retry too many times.  Instead of doing this, just
do the reservation the normal way before we join the transaction, that way we
can do whatever we want to try and reclaim space, and then if it fails we know
for sure we are out of space and we can return ENOSPC.  Thanks,
Signed-off-by: NJosef Bacik <josef@redhat.com>

Snapshot creation has two phases.  One is the initial snapshot setup,
and the second is done during commit, while nobody is allowed to modify
the root we are snapshotting.

The delayed metadata insertion code can break that rule, it does a
delayed inode update on the inode of the parent of the snapshot,
and delayed directory item insertion.

This makes sure to run the pending delayed operations before we
record the snapshot root, which avoids corruptions.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

The recent commit to get rid of our trans_mutex introduced
some races with block group relocation.  The problem is that relocation
needs to do some record keeping about each root, and it was relying
on the transaction mutex to coordinate things in subtle ways.

This fix adds a mutex just for the relocation code and makes sure
it doesn't have a big impact on normal operations.  The race is
really fixed in btrfs_record_root_in_trans, which is where we
step back and wait for the relocation code to finish accounting
setup.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

We can lockup if we try to allow new writers join the transaction and we have
flushoncommit set or have a pending snapshot.  This is because we set
no_trans_join and then loop around and try to wait for ordered extents again.
The problem is the ordered endio stuff needs to join the transaction, which it
can't do because no_trans_join is set.  So instead wait until after this loop to
set no_trans_join and then make sure to wait for num_writers == 1 in case
anybody got started in between us exiting the loop and setting no_trans_join.
This could easily be reproduced by mounting -o flushoncommit and running xfstest
13.  It cannot be reproduced with this patch.  Thanks,
Reported-by: NJim Schutt <jaschut@sandia.gov>
Signed-off-by: NJosef Bacik <josef@redhat.com>

Normally current->jouranl_info is cleared by commit_transaction.  For an
async snap or subvol creation, though, it runs in a work queue.  Clear
it in btrfs_commit_transaction_async() to avoid leaking a non-NULL
journal_info when we return to userspace.  When the actual commit runs in
the other thread it won't care that it's current->journal_info is already
NULL.
Signed-off-by: NSage Weil <sage@newdream.net>
Tested-by: NJim Schutt <jaschut@sandia.gov>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

In btrfs_wait_for_commit if we came upon a transaction that had committed we
just exited, but that's bad since we are holding the trans_lock.  So break
instead so that the lock is dropped.  Thanks,
Reported-by: NDavid Sterba <dsterba@suse.cz>
Signed-off-by: NJosef Bacik <josef@redhat.com>

wrap checking of filesystem 'closing' flag and fix a few missing memory
barriers.
Signed-off-by: NDavid Sterba <dsterba@suse.cz>

Originally this was going to be used as a way to give hints to the allocator,
but frankly we can get much better hints elsewhere and it's not even used at all
for anything usefull.  In addition to be completely useless, when we initialize
an inode we try and find a freeish block group to set as the inodes block group,
and with a completely full 40gb fs this takes _forever_, so I imagine with say
1tb fs this is just unbearable.  So just axe the thing altoghether, we don't
need it and it saves us 8 bytes in the inode and saves us 500 microseconds per
inode lookup in my testcase.  Thanks,
Signed-off-by: NJosef Bacik <josef@redhat.com>

We use trans_mutex for lots of things, here's a basic list

1) To serialize trans_handles joining the currently running transaction
2) To make sure that no new trans handles are started while we are committing
3) To protect the dead_roots list and the transaction lists

Really the serializing trans_handles joining is not too hard, and can really get
bogged down in acquiring a reference to the transaction.  So replace the
trans_mutex with a trans_lock spinlock and use it to do the following

1) Protect fs_info->running_transaction.  All trans handles have to do is check
this, and then take a reference of the transaction and keep on going.
2) Protect the fs_info->trans_list.  This doesn't get used too much, basically
it just holds the current transactions, which will usually just be the currently
committing transaction and the currently running transaction at most.
3) Protect the dead roots list.  This is only ever processed by splicing the
list so this is relatively simple.
4) Protect the fs_info->reloc_ctl stuff.  This is very lightweight and was using
the trans_mutex before, so this is a pretty straightforward change.
5) Protect fs_info->no_trans_join.  Because we don't hold the trans_lock over
the entirety of the commit we need to have a way to block new people from
creating a new transaction while we're doing our work.  So we set no_trans_join
and in join_transaction we test to see if that is set, and if it is we do a
wait_on_commit.
6) Make the transaction use count atomic so we don't need to take locks to
modify it when we're dropping references.
7) Add a commit_lock to the transaction to make sure multiple people trying to
commit the same transaction don't race and commit at the same time.
8) Make open_ioctl_trans an atomic so we don't have to take any locks for ioctl
trans.

I have tested this with xfstests, but obviously it is a pretty hairy change so
lots of testing is greatly appreciated.  Thanks,
Signed-off-by: NJosef Bacik <josef@redhat.com>

We currently track trans handles in current->journal_info, but we don't actually
use it.  This patch fixes it.  This will cover the case where we have multiple
people starting transactions down the call chain.  This keeps us from having to
allocate a new handle and all of that, we just increase the use count of the
current handle, save the old block_rsv, and return.  I tested this with xfstests
and it worked out fine.  Thanks,
Signed-off-by: NJosef Bacik <josef@redhat.com>

I keep forgetting that btrfs_join_transaction() just ignores the num_items
argument, which leads me to sending pointless patches and looking stupid :).  So
just kill the num_items argument from btrfs_join_transaction and
btrfs_start_ioctl_transaction, since neither of them use it.  Thanks,
Signed-off-by: NJosef Bacik <josef@redhat.com>

Changelog V5 -> V6:
- Fix oom when the memory load is high, by storing the delayed nodes into the
  root's radix tree, and letting btrfs inodes go.

Changelog V4 -> V5:
- Fix the race on adding the delayed node to the inode, which is spotted by
  Chris Mason.
- Merge Chris Mason's incremental patch into this patch.
- Fix deadlock between readdir() and memory fault, which is reported by
  Itaru Kitayama.

Changelog V3 -> V4:
- Fix nested lock, which is reported by Itaru Kitayama, by updating space cache
  inode in time.

Changelog V2 -> V3:
- Fix the race between the delayed worker and the task which does delayed items
  balance, which is reported by Tsutomu Itoh.
- Modify the patch address David Sterba's comment.
- Fix the bug of the cpu recursion spinlock, reported by Chris Mason

Changelog V1 -> V2:
- break up the global rb-tree, use a list to manage the delayed nodes,
  which is created for every directory and file, and used to manage the
  delayed directory name index items and the delayed inode item.
- introduce a worker to deal with the delayed nodes.

Compare with Ext3/4, the performance of file creation and deletion on btrfs
is very poor. the reason is that btrfs must do a lot of b+ tree insertions,
such as inode item, directory name item, directory name index and so on.

If we can do some delayed b+ tree insertion or deletion, we can improve the
performance, so we made this patch which implemented delayed directory name
index insertion/deletion and delayed inode update.

Implementation:
- introduce a delayed root object into the filesystem, that use two lists to
  manage the delayed nodes which are created for every file/directory.
  One is used to manage all the delayed nodes that have delayed items. And the
  other is used to manage the delayed nodes which is waiting to be dealt with
  by the work thread.
- Every delayed node has two rb-tree, one is used to manage the directory name
  index which is going to be inserted into b+ tree, and the other is used to
  manage the directory name index which is going to be deleted from b+ tree.
- introduce a worker to deal with the delayed operation. This worker is used
  to deal with the works of the delayed directory name index items insertion
  and deletion and the delayed inode update.
  When the delayed items is beyond the lower limit, we create works for some
  delayed nodes and insert them into the work queue of the worker, and then
  go back.
  When the delayed items is beyond the upper bound, we create works for all
  the delayed nodes that haven't been dealt with, and insert them into the work
  queue of the worker, and then wait for that the untreated items is below some
  threshold value.
- When we want to insert a directory name index into b+ tree, we just add the
  information into the delayed inserting rb-tree.
  And then we check the number of the delayed items and do delayed items
  balance. (The balance policy is above.)
- When we want to delete a directory name index from the b+ tree, we search it
  in the inserting rb-tree at first. If we look it up, just drop it. If not,
  add the key of it into the delayed deleting rb-tree.
  Similar to the delayed inserting rb-tree, we also check the number of the
  delayed items and do delayed items balance.
  (The same to inserting manipulation)
- When we want to update the metadata of some inode, we cached the data of the
  inode into the delayed node. the worker will flush it into the b+ tree after
  dealing with the delayed insertion and deletion.
- We will move the delayed node to the tail of the list after we access the
  delayed node, By this way, we can cache more delayed items and merge more
  inode updates.
- If we want to commit transaction, we will deal with all the delayed node.
- the delayed node will be freed when we free the btrfs inode.
- Before we log the inode items, we commit all the directory name index items
  and the delayed inode update.

I did a quick test by the benchmark tool[1] and found we can improve the
performance of file creation by ~15%, and file deletion by ~20%.

Before applying this patch:
Create files:
        Total files: 50000
        Total time: 1.096108
        Average time: 0.000022
Delete files:
        Total files: 50000
        Total time: 1.510403
        Average time: 0.000030

After applying this patch:
Create files:
        Total files: 50000
        Total time: 0.932899
        Average time: 0.000019
Delete files:
        Total files: 50000
        Total time: 1.215732
        Average time: 0.000024

[1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3

Many thanks for Kitayama-san's help!
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Reviewed-by: NDavid Sterba <dave@jikos.cz>
Tested-by: NTsutomu Itoh <t-itoh@jp.fujitsu.com>
Tested-by: NItaru Kitayama <kitayama@cl.bb4u.ne.jp>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

This adds an initial implementation for scrub. It works quite
straightforward. The usermode issues an ioctl for each device in the
fs. For each device, it enumerates the allocated device chunks. For
each chunk, the contained extents are enumerated and the data checksums
fetched. The extents are read sequentially and the checksums verified.
If an error occurs (checksum or EIO), a good copy is searched for. If
one is found, the bad copy will be rewritten.
All enumerations happen from the commit roots. During a transaction
commit, the scrubs get paused and afterwards continue from the new
roots.

This commit is based on the series originally posted to linux-btrfs
with some improvements that resulted from comments from David Sterba,
Ilya Dryomov and Jan Schmidt.
Signed-off-by: NArne Jansen <sensille@gmx.net>

Remove code which has been #if0-ed out for a very long time and does not
seem to be related to current codebase anymore.
Signed-off-by: NDavid Sterba <dsterba@suse.cz>

all callers pass GFP_NOFS, but the GFP mask argument is not used in the
function; GFP_ATOMIC is passed to radix tree initialization and it's the
only correct one, since we're using the preload/insert mechanism of
radix tree.
Let's drop the gfp mask from btrfs function, this will not change
behaviour.
Signed-off-by: NDavid Sterba <dsterba@suse.cz>

This is similar to block group caching.

We dedicate a special inode in fs tree to save free ino cache.

At the very first time we create/delete a file after mount, the free ino
cache will be loaded from disk into memory. When the fs tree is commited,
the cache will be written back to disk.

To keep compatibility, we check the root generation against the generation
of the special inode when loading the cache, so the loading will fail
if the btrfs filesystem was mounted in an older kernel before.
Signed-off-by: NLi Zefan <lizf@cn.fujitsu.com>

There's a potential problem in 32bit system when we exhaust 32bit inode
numbers and start to allocate big inode numbers, because btrfs uses
inode->i_ino in many places.

So here we always use BTRFS_I(inode)->location.objectid, which is an
u64 variable.

There are 2 exceptions that BTRFS_I(inode)->location.objectid !=
inode->i_ino: the btree inode (0 vs 1) and empty subvol dirs (256 vs 2),
and inode->i_ino will be used in those cases.

Another reason to make this change is I'm going to use a special inode
to save free ino cache, and the inode number must be > (u64)-256.
Signed-off-by: NLi Zefan <lizf@cn.fujitsu.com>

Currently btrfs stores the highest objectid of the fs tree, and it always
returns (highest+1) inode number when we create a file, so inode numbers
won't be reclaimed when we delete files, so we'll run out of inode numbers
as we keep create/delete files in 32bits machines.

This fixes it, and it works similarly to how we cache free space in block
cgroups.

We start a kernel thread to read the file tree. By scanning inode items,
we know which chunks of inode numbers are free, and we cache them in
an rb-tree.

Because we are searching the commit root, we have to carefully handle the
cross-transaction case.

The rb-tree is a hybrid extent+bitmap tree, so if we have too many small
chunks of inode numbers, we'll use bitmaps. Initially we allow 16K ram
of extents, and a bitmap will be used if we exceed this threshold. The
extents threshold is adjusted in runtime.
Signed-off-by: NLi Zefan <lizf@cn.fujitsu.com>

I've been working on making our O_DIRECT latency not suck and I noticed we were
taking the trans_mutex in btrfs_end_transaction.  So to do this we convert
num_writers and use_count to atomic_t's and just decrement them in
btrfs_end_transaction.  Instead of deleting the transaction from the trans list
in put_transaction we do that in btrfs_commit_transaction() since that's the
only time it actually needs to be removed from the list.  Thanks,
Signed-off-by: NJosef Bacik <josef@redhat.com>

I saw a lockup where we kept getting into this start transaction->commit
transaction loop because of enospce.  The fact is if we fail to make our
reservation, we've tried _everything_ several times, so we only need to try and
commit the transaction once, and if that doesn't work then we really are out of
space and need to just exit.  Thanks,
Signed-off-by: NJosef Bacik <josef@redhat.com>

root_item->flags and root_item->byte_limit are not initialized when
a subvolume is created. This bug is not revealed until we added
readonly snapshot support - now you mount a btrfs filesystem and you
may find the subvolumes in it are readonly.

To work around this problem, we steal a bit from root_item->inode_item->flags,
and use it to indicate if those fields have been properly initialized.
When we read a tree root from disk, we check if the bit is set, and if
not we'll set the flag and initialize the two fields of the root item.
Reported-by: NAndreas Philipp <philipp.andreas@gmail.com>
Signed-off-by: NLi Zefan <lizf@cn.fujitsu.com>
Tested-by: NAndreas Philipp <philipp.andreas@gmail.com>
cc: stable@kernel.org
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Free btrfs_trans_handle when join_transaction() fails
in start_transaction()
Signed-off-by: NYoshinori Sano <yoshinori.sano@gmail.com>
Signed-off-by: NChris Mason <chris.mason@oracle.com>