While running some snashot aware defrag tests I noticed I was panicing every
once and a while in key_search.  This is because of the optimization that says
if we find a key at slot 0 it will be at slot 0 all the way down the rest of the
tree.  This isn't the case for btrfs_search_old_slot since it will likely replay
changes to a buffer if something has changed since we took our sequence number.
So short circuit this optimization by setting prev_cmp to -1 every time we call
key_search so we will do our normal binary search.  With this patch I am no
longer seeing the panics I was seeing before.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

Remove unused eb parameter from btrfs_item_nr
Signed-off-by: NRoss Kirk <ross.kirk@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>

If we failed to actually allocate the correct size of the extent to relocate we
will end up in an infinite loop because we won't return an error, we'll just
move on to the next extent.  So fix this up by returning an error, and then fix
all the callers to return an error up the stack rather than BUG_ON()'ing.
Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

When the binary search returns 0 (exact match), the target key
will necessarily be at slot 0 of all nodes below the current one,
so in this case the binary search is not needed because it will
always return 0, and we waste time doing it, holding node locks
for longer than necessary, etc.

Below follow histograms with the times spent on the current approach of
doing a binary search when the previous binary search returned 0, and
times for the new approach, which directly picks the first item/child
node in the leaf/node.

Current approach:

Count: 6682
Range: 35.000 - 8370.000; Mean: 85.837; Median: 75.000; Stddev: 106.429
Percentiles:  90th: 124.000; 95th: 145.000; 99th: 206.000
  35.000 -   61.080:  1235 ################
  61.080 -  106.053:  4207 #####################################################
 106.053 -  183.606:  1122 ##############
 183.606 -  317.341:   111 #
 317.341 -  547.959:     6 |
 547.959 - 8370.000:     1 |

Approach proposed by this patch:

Count: 6682
Range:  6.000 - 135.000; Mean: 16.690; Median: 16.000; Stddev:  7.160
Percentiles:  90th: 23.000; 95th: 27.000; 99th: 40.000
   6.000 -    8.418:    58 #
   8.418 -   11.670:  1149 #########################
  11.670 -   16.046:  2418 #####################################################
  16.046 -   21.934:  2098 ##############################################
  21.934 -   29.854:   744 ################
  29.854 -   40.511:   154 ###
  40.511 -   54.848:    41 #
  54.848 -   74.136:     5 |
  74.136 -  100.087:     9 |
 100.087 -  135.000:     6 |

These samples were captured during a run of the btrfs tests 001, 002 and
004 in the xfstests, with a leaf/node size of 4Kb.
Signed-off-by: NFilipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

Internally, btrfs_header_chunk_tree_uuid() calculates an unsigned long, but
casts it to a pointer, while all callers cast it to unsigned long again.
Signed-off-by: NGeert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

Internally, btrfs_header_fsid() calculates an unsigned long, but casts
it to a pointer, while all callers cast it to unsigned long again.
Signed-off-by: NGeert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

u64 is "unsigned long long" on all architectures now, so there's no need to
cast it when formatting it using the "ll" length modifier.
Signed-off-by: NGeert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

make C=2 fs/btrfs/ CF=-D__CHECK_ENDIAN__

I tried to filter out the warnings for which patches have already
been sent to the mailing list, pending for inclusion in btrfs-next.

All these changes should be obviously safe.
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>

If you are sending a snapshot and specifying a parent snapshot we will walk the
trees and figure out where they differ and send the differences only.  The way
we check for differences are if the leaves aren't the same and if the keys are
not the same within the leaves.  So if neither leaf is the same (ie the leaf has
been cow'ed from the parent snapshot) we walk each item in the send root and
check it against the parent root.  If the items match exactly then we don't do
anything.  This doesn't quite work for inode refs, since they will just have the
name and the parent objectid.  If you move the file from a directory and then
remove that directory and re-create a directory with the same inode number as
the old directory and then move that file back into that directory we will
assume that nothing changed and you will get errors when you try to receive.

In order to fix this we need to do extra checking to see if the inode ref really
is the same or not.  So do this by passing down BTRFS_COMPARE_TREE_SAME if the
items match.  Then if the key type is an inode ref we can do some extra
checking, otherwise we just keep processing.  The extra checking is to look up
the generation of the directory in the parent volume and compare it to the
generation of the send volume.  If they match then they are the same directory
and we are good to go.  If they don't we have to add them to the changed refs
list.

This means we have to track the generation of the ref we're trying to lookup
when we iterate all the refs for a particular inode.  So in the case of looking
for new refs we have to get the generation from the parent volume, and in the
case of looking for deleted refs we have to get the generation from the send
volume to compare with.

There was also the issue of using a ulist to keep track of the directories we
needed to check.  Because we can get a deleted ref and a new ref for the same
inode number the ulist won't work since it indexes based on the value.  So
instead just dup any directory ref we find and add it to a local list, and then
process that list as normal and do away with using a ulist for this altogether.

Before we would fail all of the tests in the far-progs that related to moving
directories (test group 32).  With this patch we now pass these tests, and all
of the tests in the far-progs send testing suite.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

There is no reason we can't just set the path to blocking and then do normal
GFP_NOFS allocations for these extent buffers.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

We can get ENOMEM trying to allocate dummy bufs for the rewind operation of the
tree mod log.  Instead of BUG_ON()'ing in this case pass up ENOMEM.  I looked
back through the callers and I'm pretty sure I got everybody who did BUG_ON(ret)
in this path.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

Previously we held the tree mod lock when adding stuff because we use it to
check and see if we truly do want to track tree modifications.  This is
admirable, but GFP_ATOMIC in a critical area that is going to get hit pretty
hard and often is not nice.  So instead do our basic checks to see if we don't
need to track modifications, and if those pass then do our allocation, and then
when we go to insert the new modification check if we still care, and if we
don't just free up our mod and return.  Otherwise we're good to go and we can
carry on.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

commit 47fb091f(Btrfs: fix unlock after free on rewinded tree blocks)
takes an extra increment on the reference of allocated dummy extent buffer, so now we
cannot free this dummy one, and end up with extent buffer leak.
Signed-off-by: NLiu Bo <bo.li.liu@oracle.com>
Reviewed-by: NJan Schmidt <list.btrfs@jan-o-sch.net>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

There is another bug in the tree mod log stuff in that we're calling
tree_mod_log_free_eb every single time a block is cow'ed.  The problem with this
is that if this block is shared by multiple snapshots we will call this multiple
times per block, so if we go to rewind the mod log for this block we'll BUG_ON()
in __tree_mod_log_rewind because we try to rewind a free twice.  We only want to
call tree_mod_log_free_eb if we are actually freeing the block.  With this patch
I no longer hit the panic in __tree_mod_log_rewind.  Thanks,

Cc: stable@vger.kernel.org
Reviewed-by: NJan Schmidt <list.btrfs@jan-o-sch.net>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

We need to hold the tree mod log lock in __tree_mod_log_rewind since we walk
forward in the tree mod entries, otherwise we'll end up with random entries and
trip the BUG_ON() at the front of __tree_mod_log_rewind.  This fixes the panics
people were seeing when running

find /whatever -type f -exec btrfs fi defrag {} \;

Thansk,

Cc: stable@vger.kernel.org
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

This patch does two things.  First we no longer explicitly read in the blocks
we're trying to readahead.  For things like balance_level we may never actually
use the blocks so this just adds uneeded latency, and balance_level and
split_node will both read in the blocks they care about explicitly so if the
blocks need to be waited on it will be done there.  Secondly we no longer drop
the path if we do readahead, we just set the path blocking before we call
reada_for_balance() and then we're good to go.  Hopefully this will cut down on
the number of re-searches.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

This patch does two things, first it only does one call to
btrfs_buffer_uptodate() with the gen specified instead of once with 0 and then
again with gen specified.  The other thing is to call btrfs_read_buffer() on the
buffer we've found instead of dropping it and then calling read_tree_block().
This will keep us from doing yet another radix tree lookup for a buffer we've
already found.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

During splitting a leaf, pushing items around to hopefully get some space only
works when we have a parent, ie. we have at least one sibling leaf.
Signed-off-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

As for splitting a leaf, root is just the leaf, and tree mod log does not apply
on leaf, so in this case, we don't do log_removal.

As for splitting a node, the old root is kept as a normal node and we have nicely
put records in tree mod log for moving keys and items, so in this case we don't do
that either.

As above, insert_new_root can get rid of log_removal.
Signed-off-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

The size parameter to btrfs_extend_item() is the number of bytes
to add to the item, not the size of the item after the operation
(like it is for btrfs_truncate_item(), there the size parameter
is not the number of bytes to take away, but the total size of
the item after truncation).
Fix it in the comment.
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

The size parameter to btrfs_extend_item() is the number of bytes
to add to the item, not the size of the item after the operation
(like it is for btrfs_truncate_item(), there the size parameter
is not the number of bytes to take away, but the total size of
the item after truncation).
Fix it in the comment.
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NJiri Kosina <jkosina@suse.cz>

Chris hit a bug where we weren't finding extent records when running extent ops.
This is because we use the delayed_ref_head when running the extent op, which
means we can't use the ->type checks to see if we are metadata.  We also lose
the level of the metadata we are working on.  So to fix this we can just check
the ->is_data section of the extent_op, and we can store the level of the buffer
we were modifying in the extent_op.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

Big patch, but all it does is add statics to functions which
are in fact static, then remove the associated dead-code fallout.

removed functions:

btrfs_iref_to_path()
__btrfs_lookup_delayed_deletion_item()
__btrfs_search_delayed_insertion_item()
__btrfs_search_delayed_deletion_item()
find_eb_for_page()
btrfs_find_block_group()
range_straddles_pages()
extent_range_uptodate()
btrfs_file_extent_length()
btrfs_scrub_cancel_devid()
btrfs_start_transaction_lflush()

btrfs_print_tree() is left because it is used for debugging.
btrfs_start_transaction_lflush() and btrfs_reada_detach() are
left for symmetry.

ulist.c functions are left, another patch will take care of those.
Signed-off-by: NEric Sandeen <sandeen@redhat.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

Sequence numbers for delayed refs have been introduced in the first version
of the qgroup patch set. To solve the problem of find_all_roots on a busy
file system, the tree mod log was introduced. The sequence numbers for that
were simply shared between those two users.

However, at one point in qgroup's quota accounting, there's a statement
accessing the previous sequence number, that's still just doing (seq - 1)
just as it would have to in the very first version.

To satisfy that requirement, this patch makes the sequence number counter 64
bit and splits it into a major part (used for qgroup sequence number
counting) and a minor part (incremented for each tree modification in the
log). This enables us to go exactly one major step backwards, as required
for qgroups, while still incrementing the sequence counter for tree mod log
insertions to keep track of their order. Keeping them in a single variable
means there's no need to change all the code dealing with comparisons of two
sequence numbers.

The sequence number is reset to 0 on commit (not new in this patch), which
ensures we won't overflow the two 32 bit counters.

Without this fix, the qgroup tracking can occasionally go wrong and WARN_ONs
from the tree mod log code may happen.
Signed-off-by: NJan Schmidt <list.btrfs@jan-o-sch.net>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

We kept leaking extent buffers when mounting a broken file system and it turns
out it's because not everybody uses read_tree_block properly.  You need to check
and make sure the extent_buffer is uptodate before you use it.  This patch fixes
everybody who calls read_tree_block directly to make sure they check that it is
uptodate and free it and return an error if it is not.  With this we no longer
leak EB's when things go horribly wrong.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

Argument 'trans' is not used in btrfs_extend_item().
Signed-off-by: NTsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

If argument 'trans' is unnecessary in the function where
fixup_low_keys() is called, 'trans' is deleted.
Signed-off-by: NTsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

Argument 'trans' is not used in fixup_low_keys(). So, remove it.
Signed-off-by: NTsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

When tree_mod_log_rewind decides to make a copy of the current tree buffer
for its modifications, it subsequently freed the buffer before unlocking it.
Obviously, those operations are required in reverse order.
Signed-off-by: NJan Schmidt <list.btrfs@jan-o-sch.net>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

The tree mod log functions were accessing root->node->... directly, without
use of btrfs_root_node() or explicit rcu locking. This could lead to an
extent buffer reference being leaked and another reference being freed too
early when preemtion was enabled.
Signed-off-by: NJan Schmidt <list.btrfs@jan-o-sch.net>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

Commit d9abbf1c changed tree mod log locking around ROOT_REPLACE operations.
When a tree root is split, however, we were logging removal of all elements
from the root node before logging removal of half of the elements for the
split operation. This leads to a BUG_ON when rewinding.

This commit removes the erroneous logging of removal of all elements.
Signed-off-by: NJan Schmidt <list.btrfs@jan-o-sch.net>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

A user sent me a btrfs-image of a file system that was panicing on mount during
the log recovery.  I had originally thought these problems were from a bug in
the free space cache code, but that was just a symptom of the problem.  The
problem is if your application does something like this

[prealloc][prealloc][prealloc]

the internal extent maps will merge those all together into one extent map, even
though on disk they are 3 separate extents.  So if you go to write into one of
these ranges the extent map will be right since we use the physical extent when
doing the write, but when we log the extents they will use the wrong sizes for
the remainder prealloc space.  If this doesn't happen to trip up the free space
cache (which it won't in a lot of cases) then you will get bogus entries in your
extent tree which will screw stuff up later.  The data and such will still work,
but everything else is broken.  This patch fixes this by not allowing extents
that are on the modified list to be merged.  This has the side effect that we
are no longer adding everything to the modified list all the time, which means
we now have to call btrfs_drop_extents every time we log an extent into the
tree.  So this allows me to drop all this speciality code I was using to get
around calling btrfs_drop_extents.  With this patch the testcase I've created no
longer creates a bogus file system after replaying the log.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

We currently store the first key of the tree block inside the reference for the
tree block in the extent tree.  This takes up quite a bit of space.  Make a new
key type for metadata which holds the level as the offset and completely removes
storing the btrfs_tree_block_info inside the extent ref.  This reduces the size
from 51 bytes to 33 bytes per extent reference for each tree block.  In practice
this results in a 30-35% decrease in the size of our extent tree, which means we
COW less and can keep more of the extent tree in memory which makes our heavy
metadata operations go much faster.  This is not an automatic format change, you
must enable it at mkfs time or with btrfstune.  This patch deals with having
metadata stored as either the old format or the new format so it is easy to
convert.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

To resolve backrefs, ROOT_REPLACE operations in the tree mod log are
required to be tied to at least one KEY_REMOVE_WHILE_FREEING operation.
Therefore, those operations must be enclosed by tree_mod_log_write_lock()
and tree_mod_log_write_unlock() calls.

Those calls are private to the tree_mod_log_* functions, which means that
removal of the elements of an old root node must be logged from
tree_mod_log_insert_root. This partly reverts and corrects commit ba1bfbd5
(Btrfs: fix a tree mod logging issue for root replacement operations).

This fixes the brand-new version of xfstest 276 as of commit cfe73f71.

Cc: stable@vger.kernel.org
Signed-off-by: NJan Schmidt <list.btrfs@jan-o-sch.net>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

The entry point at the defrag ioctl always sets "cache only" to 0;
the codepaths haven't run for a long time as far as I can
tell.  Chris says they're dead code, so remove them.
Signed-off-by: NEric Sandeen <sandeen@redhat.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

This keeps static checkers happy.
Signed-off-by: NEric Sandeen <sandeen@redhat.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

When replaying a log tree with qgroups enabled, tree_mod_log_rewind does a
sanity-check of the number of items against the maximum possible number.
It calculates that number with the nodesize of fs_root. Unfortunately
fs_root is not yet set at this stage. So instead use the nodesize from
tree_root, which is already initialized.
Signed-off-by: NArne Jansen <sensille@gmx.net>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

This reverts commit 6a7a665d.

This was bug was fixed differently in 3.6, so this commit
isn't needed.

Conflicts:
	fs/btrfs/ctree.c
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

This reverts commit 95c80bb1.

The bug addressed by this commit was fixed differently back in 3.6
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

I noticed while doing fsync tests that we were always dropping the path and
re-searching when we first cow the log root even though we've already gotten
the write lock on the root.  That's because we don't take into account that
there might not be a parent node, so fix the check to make sure there is
actually a parent node before we undo all of this work for nothing.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>