If btrfs_del_root fails in btrfs_drop_snapshot, we'll pick up the
error but then return 0 anyway due to mixing err and ret.

Fixes: 79787eaa ("btrfs: replace many BUG_ONs with proper error handling")
Cc: <stable@vger.kernel.org> # v3.4+
Signed-off-by: NJeff Mahoney <jeffm@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We discovered a box that had double allocations, and suspected the space
cache may be to blame.  While auditing the write out path I noticed that
if we've already setup the space cache we will just carry on.  This
means that any error we hit after cache_save_setup before we go to
actually write the cache out we won't reset the inode generation, so
whatever was already written will be considered correct, except it'll be
stale.  Fix this by _always_ resetting the generation on the block group
inode, this way we only ever have valid or invalid cache.

With this patch I was no longer able to reproduce cache corruption with
dm-log-writes and my bpf error injection tool.

Cc: stable@vger.kernel.org
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Make wait_on_atomic_t() pass the TASK_* mode onto its action function as an
extra argument and make it 'unsigned int throughout.

Also, consolidate a bunch of identical action functions into a default
function that can do the appropriate thing for the mode.

Also, change the argument name in the bit_wait*() function declarations to
reflect the fact that it's the mode and not the bit number.

[Peter Z gives this a grudging ACK, but thinks that the whole atomic_t wait
should be done differently, though he's not immediately sure as to how]
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Acked-by: NPeter Zijlstra <peterz@infradead.org>
cc: Ingo Molnar <mingo@kernel.org>

If we get a significant amount of delayed refs for a single block (think
modifying multiple snapshots) we can end up spending an ungodly amount
of time looping through all of the entries trying to see if they can be
merged.  This is because we only add them to a list, so we have O(2n)
for every ref head.  This doesn't make any sense as we likely have refs
for different roots, and so they cannot be merged.  Tracking in a tree
will allow us to break as soon as we hit an entry that doesn't match,
making our worst case O(n).

With this we can also merge entries more easily.  Before we had to hope
that matching refs were on the ends of our list, but with the tree we
can search down to exact matches and merge them at insert time.
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The way we handle delalloc metadata reservations has gotten
progressively more complicated over the years.  There is so much cruft
and weirdness around keeping the reserved count and outstanding counters
consistent and handling the error cases that it's impossible to
understand.

Fix this by making the delalloc block rsv per-inode.  This way we can
calculate the actual size of the outstanding metadata reservations every
time we make a change, and then reserve the delta based on that amount.
This greatly simplifies the code everywhere, and makes the error
handling in btrfs_delalloc_reserve_metadata far less terrifying.
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Right now we do a lot of weird hoops around outstanding_extents in order
to keep the extent count consistent.  This is because we logically
transfer the outstanding_extent count from the initial reservation
through the set_delalloc_bits.  This makes it pretty difficult to get a
handle on how and when we need to mess with outstanding_extents.

Fix this by revamping the rules of how we deal with outstanding_extents.
Now instead everybody that is holding on to a delalloc extent is
required to increase the outstanding extents count for itself.  This
means we'll have something like this

btrfs_delalloc_reserve_metadata	- outstanding_extents = 1
 btrfs_set_extent_delalloc	- outstanding_extents = 2
btrfs_release_delalloc_extents	- outstanding_extents = 1

for an initial file write.  Now take the append write where we extend an
existing delalloc range but still under the maximum extent size

btrfs_delalloc_reserve_metadata - outstanding_extents = 2
  btrfs_set_extent_delalloc
    btrfs_set_bit_hook		- outstanding_extents = 3
    btrfs_merge_extent_hook	- outstanding_extents = 2
btrfs_delalloc_release_extents	- outstanding_extnets = 1

In order to make the ordered extent transition we of course must now
make ordered extents carry their own outstanding_extent reservation, so
for cow_file_range we end up with

btrfs_add_ordered_extent	- outstanding_extents = 2
clear_extent_bit		- outstanding_extents = 1
btrfs_remove_ordered_extent	- outstanding_extents = 0

This makes all manipulations of outstanding_extents much more explicit.
Every successful call to btrfs_delalloc_reserve_metadata _must_ now be
combined with btrfs_release_delalloc_extents, even in the error case, as
that is the only function that actually modifies the
outstanding_extents counter.

The drawback to this is now we are much more likely to have transient
cases where outstanding_extents is much larger than it actually should
be.  This could happen before as we manipulated the delalloc bits, but
now it happens basically at every write.  This may put more pressure on
the ENOSPC flushing code, but I think making this code simpler is worth
the cost.  I have another change coming to mitigate this side-effect
somewhat.

I also added trace points for the counter manipulation.  These were used
by a bpf script I wrote to help track down leak issues.
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

These are useful for debugging problems where we mess with
trans->block_rsv to make sure we're not screwing something up.
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

This is just excessive information in the ref_head, and makes the code
complicated.  It is a relic from when we had the heads and the refs in
the same tree, which is no longer the case.  With this removal I've
cleaned up a bunch of the cruft around this old assumption as well.
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We do a couple different cleanup operations on the ref head.  We adjust
counters, we'll free any reserved space if we didn't end up using the
ref, and we clear the pending csum bytes.  Move all these disparate
things into cleanup_ref_head and clean up the logic in
__btrfs_run_delayed_refs so that it handles the !ref case a lot cleaner,
as well as making run_one_delayed_ref() only deal with real refs and not
the ref head.
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We only use this logic if our ref isn't a ref_head, so move it up into
the if (ref) case since we know that this is a normal ref and not a
delayed ref head.
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Move this code out to a helper function to further simplivy
__btrfs_run_delayed_refs.
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Move the extent_op cleanup for an empty head ref to a helper function to
help simplify __btrfs_run_delayed_refs.
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Simplify the error handling in __btrfs_run_delayed_refs by breaking out
the code used to return a head back to the delayed_refs tree for
processing into a helper function.
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We were having corruption issues that were tied back to problems with
the extent tree.  In order to track them down I built this tool to try
and find the culprit, which was pretty successful.  If you compile with
this tool on it will live verify every ref update that the fs makes and
make sure it is consistent and valid.  I've run this through with
xfstests and haven't gotten any false positives.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
[ update error messages, add fixup from Dan Carpenter to handle errors
  of read_tree_block ]
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We need the actual root for the ref verifier tool to work, so change
these functions to pass the root around instead.  This will be used in
a subsequent patch.
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Bool initializations should use true and false. Bool tests don't need
comparisons.
Signed-off-by: NThomas Meyer <thomas@m3y3r.de>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Nikolay reported that generic/273 was failing currently with ENOSPC.
Turns out this is because we get to the point where the outstanding
reservations are greater than the pinned space on the fs.  This is a
mistake, previously we used the current reservation amount in
may_commit_transaction, not the entire outstanding reservation amount.
Fix this to find the minimum byte size needed to make progress in
flushing, and pass that into may_commit_transaction.  From there we can
make a smarter decision on whether to commit the transaction or not.
This fixes the failure in generic/273.

From Nikolai, IOW: when we go to the final stage of deciding whether to
do trans commit, instead of passing all the reservations from all
tickets we just pass the reservation for the current ticket. Otherwise,
in case all reservations exceed pinned space, then we don't commit
transaction and fail prematurely. Before we passed num_bytes from
flush_space, where num_bytes was the sum of all pending reserverations,
but now all we do is take the first ticket and commit the trans if we
can satisfy that.

Fixes: 957780eb ("Btrfs: introduce ticketed enospc infrastructure")
Cc: stable@vger.kernel.org # 4.8
Reported-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Tested-by: NNikolay Borisov <nborisov@suse.com>
[ added Nikolai's comment ]
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Signed-off-by: NSatoru Takeuchi <satoru.takeuchi@gmail.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

__link_block_group is called from only 2 places and at each call site the
space_info being passed is the same as the space info assigned to the passed
cache struct. Let's remove the redundant argument and make the function
reference the space_info from the passed block_group_cache. No functional
changes
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NJosef Bacik <jbacik@fb.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
[ renamed to link_block_group ]
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Some static functions are needlessly forward declared. Let's remove those
declarations since they add no value.
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

btrfs_del_roots always uses the tree_root.  Let's pass fs_info instead.
Signed-off-by: NJeff Mahoney <jeffm@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Every shared ref has a parent tree block, which can be get from
btrfs_extent_inline_ref_offset().  And the tree block must be aligned
to the nodesize, so we'd know this inline ref is not valid if this
block's bytenr is not aligned to the nodesize, in which case, most
likely the ref type has been misused.

This adds the above mentioned check and also updates
print_extent_item() called by btrfs_print_leaf() to point out the
invalid ref while printing the tree structure.
Signed-off-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Since we have a helper which can do sanity check, this converts all
btrfs_extent_inline_ref_type to it.
Signed-off-by: NLiu Bo <bo.li.liu@oracle.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

An invalid value of extent inline ref type may be read from a
malicious image which may force btrfs to crash.

This adds a helper which does sanity check for the ref type, so we can
know if it's sane, return he type, otherwise return an error.
Signed-off-by: NLiu Bo <bo.li.liu@oracle.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
[ minimal tweak const types, causing warnings due to other cleanup patches ]
Signed-off-by: NDavid Sterba <dsterba@suse.com>

btrfs_make_block_group is always called with chunk_objectid set to
BTRFS_FIRST_CHUNK_TREE_OBJECTID. There's no reason why this behavior will
change anytime soon, so let's remove the argument and decrease the cognitive
load when reading the code path. No functional change
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

    This patch provides a band aid to improve the 'out of the box'
behaviour of btrfs for disks that are detected as being an ssd.  In a
general purpose mixed workload scenario, the current ssd mode causes
overallocation of available raw disk space for data, while leaving
behind increasing amounts of unused fragmented free space. This
situation leads to early ENOSPC problems which are harming user
experience and adoption of btrfs as a general purpose filesystem.

This patch modifies the data extent allocation behaviour of the ssd mode
to make it behave identical to nossd mode.  The metadata behaviour and
additional ssd_spread option stay untouched so far.

Recommendations for future development are to reconsider the current
oversimplified nossd / ssd distinction and the broken detection
mechanism based on the rotational attribute in sysfs and provide
experienced users with a more flexible way to choose allocator behaviour
for data and metadata, optimized for certain use cases, while keeping
sane 'out of the box' default settings.  The internals of the current
btrfs code have more potential than what currently gets exposed to the
user to choose from.

    The SSD story...

    In the first year of btrfs development, around early 2008, btrfs
gained a mount option which enables specific functionality for
filesystems on solid state devices. The first occurance of this
functionality is in commit e18e4809, labeled "Add mount -o ssd, which
includes optimizations for seek free storage".

The effect on allocating free space for doing (data) writes is to
'cluster' writes together, writing them out in contiguous space, as
opposed to a 'tetris' way of putting all separate writes into any free
space fragment that fits (which is what the -o nossd behaviour does).

A somewhat simplified explanation of what happens is that, when for
example, the 'cluster' size is set to 2MiB, when we do some writes, the
data allocator will search for a free space block that is 2MiB big, and
put the writes in there. The ssd mode itself might allow a 2MiB cluster
to be composed of multiple free space extents with some existing data in
between, while the additional ssd_spread mount option kills off this
option and requires fully free space.

The idea behind this is (commit 536ac8ae): "The [...] clusters make it
more likely a given IO will completely overwrite the ssd block, so it
doesn't have to do an internal rwm cycle."; ssd block meaning nand erase
block. So, effectively this means applying a "locality based algorithm"
and trying to outsmart the actual ssd.

Since then, various changes have been made to the involved code, but the
basic idea is still present, and gets activated whenever the ssd mount
option is active. This also happens by default, when the rotational flag
as seen at /sys/block/<device>/queue/rotational is set to 0.

    However, there's a number of problems with this approach.

    First, what the optimization is trying to do is outsmart the ssd by
assuming there is a relation between the physical address space of the
block device as seen by btrfs and the actual physical storage of the
ssd, and then adjusting data placement. However, since the introduction
of the Flash Translation Layer (FTL) which is a part of the internal
controller of an ssd, these attempts are futile. The use of good quality
FTL in consumer ssd products might have been limited in 2008, but this
situation has changed drastically soon after that time. Today, even the
flash memory in your automatic cat feeding machine or your grandma's
wheelchair has a full featured one.

Second, the behaviour as described above results in the filesystem being
filled up with badly fragmented free space extents because of relatively
small pieces of space that are freed up by deletes, but not selected
again as part of a 'cluster'. Since the algorithm prefers allocating a
new chunk over going back to tetris mode, the end result is a filesystem
in which all raw space is allocated, but which is composed of
underutilized chunks with a 'shotgun blast' pattern of fragmented free
space. Usually, the next problematic thing that happens is the
filesystem wanting to allocate new space for metadata, which causes the
filesystem to fail in spectacular ways.

Third, the default mount options you get for an ssd ('ssd' mode enabled,
'discard' not enabled), in combination with spreading out writes over
the full address space and ignoring freed up space leads to worst case
behaviour in providing information to the ssd itself, since it will
never learn that all the free space left behind is actually free.  There
are two ways to let an ssd know previously written data does not have to
be preserved, which are sending explicit signals using discard or
fstrim, or by simply overwriting the space with new data.  The worst
case behaviour is the btrfs ssd_spread mount option in combination with
not having discard enabled. It has a side effect of minimizing the reuse
of free space previously written in.

Fourth, the rotational flag in /sys/ does not reliably indicate if the
device is a locally attached ssd. For example, iSCSI or NBD displays as
non-rotational, while a loop device on an ssd shows up as rotational.

The combination of the second and third problem effectively means that
despite all the good intentions, the btrfs ssd mode reliably causes the
ssd hardware and the filesystem structures and performance to be choked
to death. The clickbait version of the title of this story would have
been "Btrfs ssd optimizations considered harmful for ssds".

The current nossd 'tetris' mode (even still without discard) allows a
pattern of overwriting much more previously used space, causing many
more implicit discards to happen because of the overwrite information
the ssd gets. The actual location in the physical address space, as seen
from the point of view of btrfs is irrelevant, because the actual writes
to the low level flash are reordered anyway thanks to the FTL.

    Changes made in the code

1. Make ssd mode data allocation identical to tetris mode, like nossd.
2. Adjust and clean up filesystem mount messages so that we can easily
identify if a kernel has this patch applied or not, when providing
support to end users. Also, make better use of the *_and_info helpers to
only trigger messages on actual state changes.

    Backporting notes

Notes for whoever wants to backport this patch to their 4.9 LTS kernel:
* First apply commit 951e7966 "btrfs: drop the nossd flag when
  remounting with -o ssd", or fixup the differences manually.
* The rest of the conflicts are because of the fs_info refactoring. So,
  for example, instead of using fs_info, it's root->fs_info in
  extent-tree.c
Signed-off-by: NHans van Kranenburg <hans.van.kranenburg@mendix.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We have a WARN_ON(!var) inside an if branch which is executed (among
others) only when var is true.
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Currently should_alloc_chunk uses ->total_bytes - ->bytes_readonly to
signify the total amount of bytes in this space info. However, given
Jeff's patch which adds bytes_pinned and bytes_may_use to the calculation
of num_allocated it becomes a lot more clear to just eliminate num_bytes
altogether and add the bytes_readonly to the amount of used space. That
way we don't change the results of the following statements. In the
process also start using btrfs_space_info_used.
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NJeff Mahoney <jeffm@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

In a heavy write scenario, we can end up with a large number of pinned bytes.
This can translate into (very) premature ENOSPC because pinned bytes
must be accounted for when allowing a reservation but aren't accounted for
when deciding whether to create a new chunk.

This patch adds the accounting to should_alloc_chunk so that we can
create the chunk.
Signed-off-by: NJeff Mahoney <jeffm@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Signed-off-by: NDavid Sterba <dsterba@suse.com>

The return value of flush_space was used to have significance in the
early days when the code was first introduced and before the ticketed
enospc rework. Since the latter got introduced the return value lost any
significance whatsoever to its callers. So let's remove it. While at it
also remove the unused ticket variable in
btrfs_async_reclaim_metadata_space. It was used in the initial version
of the ticketed ENOSPC work, however Wang Xiaoguang detected a problem
with this and fixed it in ce129655 ("btrfs: introduce tickets_id to
determine whether asynchronous metadata reclaim work makes progress").
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
[ add comment ]
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Signed-off-by: NDavid Sterba <dsterba@suse.com>

This also adjusts the respective callers in other files. Those were
found with -Wunused-parameter.

btrfs_full_stripe_len's mapping_tree - introduced by 53b381b3
("Btrfs: RAID5 and RAID6") but it was never really used even in that
commit

btrfs_is_parity_mirror's mirror_num - same as above

chunk_drange_filter's chunk_offset - introduced by 94e60d5a ("Btrfs:
devid subset filter") and never used.
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The helpers append "\n" so we can keep the actual strings shorter. The
extra newline will print an empty line.  Some messages have been
slightly modified to be more consistent with the rest (lowercase first
letter).
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Many commits ago the data space_info in alloc_data_chunk_ondemand used to be
acquired from the inode. At that point commit
33b4d47f ("Btrfs: deal with NULL space info") got introduced to deal with
spurios cases where the space info could be null, following a rebalance.
Nowadays, however, the space info is referenced directly from the btrfs_fs_info
struct which is initialised at filesystem mount time. This makes the null
checks redundant, so remove them.
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

All callers of flush_space pass the same number for orig/num_bytes
arguments. Let's remove one of the numbers and also modify the trace
point to show only a single number - bytes requested.

Seems that last point where the two parameters were treated differently
is before the ticketed enospc rework.
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The sectorsize member of btrfs_block_group_cache is unused. So remove it, this
reduces the number of holes in the struct.

With patch:
/* size: 856, cachelines: 14, members: 40 */
/* sum members: 837, holes: 4, sum holes: 19 */
/* bit holes: 1, sum bit holes: 29 bits */
/* last cacheline: 24 bytes */

Without patch:
/* size: 864, cachelines: 14, members: 41 */
/* sum members: 841, holes: 5, sum holes: 23 */
/* bit holes: 1, sum bit holes: 29 bits */
/* last cacheline: 32 bytes */
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

If a lot of metadata is reserved for outstanding delayed allocations, we
rely on shrink_delalloc() to reclaim metadata space in order to fulfill
reservation tickets. However, shrink_delalloc() has a shortcut where if
it determines that space can be overcommitted, it will stop early. This
made sense before the ticketed enospc system, but now it means that
shrink_delalloc() will often not reclaim enough space to fulfill any
tickets, leading to an early ENOSPC. (Reservation tickets don't care
about being able to overcommit, they need every byte accounted for.)

Fix it by getting rid of the shortcut so that shrink_delalloc() reclaims
all of the metadata it is supposed to. This fixes early ENOSPCs we were
seeing when doing a btrfs receive to populate a new filesystem, as well
as early ENOSPCs Christoph saw when doing a big cp -r onto Btrfs.

Fixes: 957780eb ("Btrfs: introduce ticketed enospc infrastructure")
Tested-by: NChristoph Anton Mitterer <mail@christoph.anton.mitterer.name>
Cc: stable@vger.kernel.org
Reviewed-by: NJosef Bacik <jbacik@fb.com>
Signed-off-by: NOmar Sandoval <osandov@fb.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

If we have a block group that is all of the following:
1) uncached in memory
2) is read-only
3) has a disk cache state that indicates we need to recreate the cache

AND the file system has enough free space fragmentation such that the
request for an extent of a given size can't be honored;

AND have a single CPU core;

AND it's the block group with the highest starting offset such that
there are no opportunities (like reading from disk) for the loop to
yield the CPU;

We can end up with a lockup.

The root cause is simple.  Once we're in the position that we've read in
all of the other block groups directly and none of those block groups
can honor the request, there are no more opportunities to sleep.  We end
up trying to start a caching thread which never gets run if we only have
one core.  This *should* present as a hung task waiting on the caching
thread to make some progress, but it doesn't.  Instead, it degrades into
a busy loop because of the placement of the read-only check.

During the first pass through the loop, block_group->cached will be set
to BTRFS_CACHE_STARTED and have_caching_bg will be set.  Then we hit the
read-only check and short circuit the loop.  We're not yet in
LOOP_CACHING_WAIT, so we skip that loop back before going through the
loop again for other raid groups.

Then we move to LOOP_CACHING_WAIT state.

During the this pass through the loop, ->cached will still be
BTRFS_CACHE_STARTED, which means it's not cached, so we'll enter
cache_block_group, do a lot of nothing, and return, and also set
have_caching_bg again.  Then we hit the read-only check and short circuit
the loop.  The same thing happens as before except now we DO trigger
the LOOP_CACHING_WAIT && have_caching_bg check and loop back up to the
top.  We do this forever.

There are two fixes in this patch since they address the same underlying
bug.

The first is to add a cond_resched to the end of the loop to ensure
that the caching thread always has an opportunity to run.  This will
fix the soft lockup issue, but find_free_extent will still loop doing
nothing until the thread has completed.

The second is to move the read-only check to the top of the loop.  We're
never going to return an allocation within a read-only block group so
we may as well skip it early.  The check for ->cached == BTRFS_CACHE_ERROR
would cause the same problem except that BTRFS_CACHE_ERROR is considered
a "done" state and we won't re-set have_caching_bg again.

Many thanks to Stephan Kulow <coolo@suse.de> for his excellent help in
the testing process.
Signed-off-by: NJeff Mahoney <jeffm@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Dave Jones hit a WARN_ON(nr < 0) in btrfs_wait_ordered_roots() with
v4.12-rc6.  This was because commit 70e7af24 made it possible for
calc_reclaim_items_nr() to return a negative number.  It's not really a
bug in that commit, it just didn't go far enough down the stack to find
all the possible 64->32 bit overflows.

This switches calc_reclaim_items_nr() to return a u64 and changes everyone
that uses the results of that math to u64 as well.
Reported-by: NDave Jones <davej@codemonkey.org.uk>
Fixes: 70e7af24 ("Btrfs: fix delalloc accounting leak caused by u32 overflow")
Signed-off-by: NChris Mason <clm@fb.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>