The "&fs_info->dev_replace.rwsem" and "&dev_replace->rwsem" refer to
the same lock but Smatch is not clever enough to figure that out so it
leads to static checker warnings.  It's better to use it consistently
anyway.
Signed-off-by: NDan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The backup_root_index member stores the index at which the backup root
should be saved upon next transaction commit. However, there is a
small deviation from this behavior in the form of a check in
backup_super_roots which checks if current root generation equals to the
generation of the previous root. This can trigger in the following
scenario:

slot0: gen-2
slot1: gen-1
slot2: gen
slot3: unused

Now suppose slot3 (which is also the root specified in the super block)
is corrupted hence init_tree_roots chooses to use the backup root at
slot2, meaning read_backup_root will read slot2 and assign the
superblock generation to gen-1. Despite this backup_root_index will
point at slot3 because its init happens in init_backup_root_slot, long
before any parsing of the backup roots occur. Then on next transaction
start, gen-1 will be incremented by 1 making the root's generation
equal gen. Subsequently, on transaction commit the following check
triggers:

  if (btrfs_backup_tree_root_gen(root_backup) ==
           btrfs_header_generation(info->tree_root->node))

This causes the 'next_backup', which is the index at which the backup is
going to be written to, to set to last_backup, which will be slot2.

All of this is a very confusing way of expressing the following
invariant:

 Always write a backup root at the index following the last used backup
 root.

This commit streamlines this logic by setting backup_root_index to the
next index after the one used for mount.
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The old name name was an awful misnomer because it didn't really find
the oldest super backup per-se but rather its slot. For example if we
have:

slot0: gen - 2
slot1: gen - 1
slot2: gen
slot3: empty

init_backup_root_slot will return slot3 and not slot0.

The new name is more appropriate since the function doesn't care whether
there is a valid backup in the returned slot or not.
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

This function has been superseded by previous commits and is no longer
used so just remove it.
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Since the filesystem is not well formed and no trees are loaded it's
pointless holding the objectid_mutex. Just remove its usage.
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The code responsible for reading and initializing tree roots is
scattered in open_ctree among 2 labels, emulating a loop. This is rather
confusing to reason about. Instead, factor the code to a new function,
init_tree_roots which implements the same logical flow.

There are a couple of notable differences, namely:

* Instead of using next_backup_root it's using the newly introduced
  read_backup_root.

* If read_backup_root returns an error init_tree_roots propagates the
  error and there is no special handling of that case e.g. the code jumps
  straight to 'fail_tree_roots' label. The old code, however, was
  (erroneously) jumping to 'fail_block_groups' label if next_backup_root
  did fail, this was unnecessary since the tree roots init logic doesn't
  modify the state of block groups.
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

This function will replace next_root_backup with a much saner/cleaner
interface.
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

It's no longer needed following cleanups around find_newest_backup_root
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Backup roots are always written in a circular manner. By definition we
can only ever have 1 backup root whose generation equals to that of the
superblock. Hence, the 'if' in the for loop will trigger at most once.
This is sufficient to return the newest backup root.

Furthermore the newest_gen parameter is always set to the generation of
the superblock. This value can be obtained from the fs_info.

This patch removes the unnecessary code dealing with the wraparound
case and makes 'newest_gen' a local variable.
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The inode delalloc mutex was added a long time ago by commit f248679e
("Btrfs: add a delalloc mutex to inodes for delalloc reservations"), and
the reason for its introduction is not very clear from the change log. It
claims it solves bogus warnings from lockdep, however it lacks an example
report/warning from lockdep, or any explanation.

Since we have enough concurrentcy protection from the locks of the space
info and block reserve objects, and such lockdep warnings don't seem to
exist anymore (at least on a 5.3 kernel I couldn't get them with fstests,
ltp, fs_mark, etc), remove it, simplifying things a bit and decreasing
the size of the btrfs_inode structure. With some quick fio tests doing
direct IO and mmap writes I couldn't observe any significant performance
increase either (direct IO writes that don't increase the file's size
don't hold the inode's lock for their entire duration and mmap writes
don't hold the inode's lock at all), which are the only type of writes
that could see any performance gain due to less serialization.

Review feedback from Josef:

The problem was taking the i_mutex in mmap, which is how I was
protecting delalloc reservations originally.  The delalloc mutex didn't
come with all of the other dependencies.  That's what the lockdep
messages were about, removing the lock isn't going to make them appear
again.

We _had_ to lock around this because we used to do tricks to keep from
over-reserving, and if we didn't serialize delalloc reservations we'd
end up with ugly accounting problems when we tried to clean things up.

However with my recentish changes this isn't the case anymore.  Every
operation is responsible for reserving its space, and then adding it to
the inode.  Then cleaning up is straightforward and can't be mucked up
by other users.  So we no longer need the delalloc mutex to safe us from
ourselves.
Reviewed-by: NJosef Bacik <josef@toxicpanda.com>
Signed-off-by: NFilipe Manana <fdmanana@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

It is not used anymore since commit 957780eb ("Btrfs: introduce
ticketed enospc infrastructure"), so just remove it.
Signed-off-by: NFilipe Manana <fdmanana@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

In btrfs_statfs() we cache fs_info::space_info in a local variable only
to use it once in a list_for_each_rcu() statement.

Not only is the local variable unnecessary it even makes the code harder
to follow as it's not clear which list it is iterating.
Signed-off-by: NJohannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The on-disk format of block group item makes use of the key that stores
the offset and length. This is further used in the code, although this
makes thing harder to understand. The key is also packed so the
offset/length is not properly aligned as u64.

Add start (key.objectid) and length (key.offset) members to block group
and remove the embedded key.  When the item is searched or written, a
local variable for key is used.
Reviewed-by: NJohannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NQu Wenruo <wqu@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Accessors defined by BTRFS_SETGET_FUNCS take a raw extent buffer and
manipulate the items there, there's no special prefix required. The
block group accessors had _disk_ because previously the names were
occupied by the on-stack accessors. As this has been addressed in the
previous patch, we can now unify the naming.
Reviewed-by: NJohannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NQu Wenruo <wqu@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

All accessors defined by BTRFS_SETGET_STACK_FUNCS contain _stack_ in the
name, the block group ones were not following that scheme, so let's
switch them.
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The members ::used and ::flags are now in the block group cache
structure, the last one is chunk_objectid, but that's set to a fixed
value and otherwise unused. The item is constructed from a local
variable before write, so we can remove the embedded one from block
group.
Reviewed-by: NJohannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NQu Wenruo <wqu@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The flags are read from the item that's embedded to block group struct,
but the item will be removed. Use the ::flags after read and before
write.
Reviewed-by: NJohannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NQu Wenruo <wqu@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

For unknown reasons, the member 'used' in the block group struct is
stored in the b-tree item and accessed everywhere using the special
accessor helper. Let's unify it and make it a regular member and only
update the item before writing it to the tree.

The item is still being used for flags and chunk_objectid, there's some
duplication until the item is removed in following patches.
Reviewed-by: NJohannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NQu Wenruo <wqu@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The last user of btrfs_bio::flags was removed in commit 326e1dbb
("block: remove management of bi_remaining when restoring original
bi_end_io"), remove it.

(Tagged for stable as the structure is heavily used and space savings
are desirable.)

CC: stable@vger.kernel.org # 4.4+
Signed-off-by: NQu Wenruo <wqu@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Add blake2b (with 256 bit digest) to the list of possible checksumming
algorithms used by BTRFS.
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Currently all the checksum algorithms generate a fixed size digest size
and we use it.  The on-disk format can hold up to BTRFS_CSUM_SIZE bytes
and BLAKE2b produces digest of 512 bits by default. We can't do that and
will use the blake2b-256, this needs to be passed to the crypto API.

Separate that from the base algorithm name and add a member to request
specific driver, in this case with the digest size.

The only place that uses the driver name is the crypto API setup.
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Show the used driver for the checksum algorithm for the filesystem in
sysfs file /sys/fs/btrfs/UUID/features/checksum, eg.

  crc32c (crc32c-generic)
Signed-off-by: NJohannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Export supported checksum algorithms via sysfs in the list of static
features:

  /sys/fs/btrfs/features/supported_checksums

Space spearated list of checksum algorithm names.
Co-developed-by: NJohannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Add sha256 to the list of possible checksumming algorithms used by BTRFS.
Signed-off-by: NJohannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Add xxhash64 to the list of possible checksumming algorithms used by
BTRFS.
Signed-off-by: NJohannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

To remove use of extent_map::bdev we need to find a replacement, and the
latest_bdev is the only one we can use here, because inode::i_bdev and
superblock::s_bdev are NULL.

The DIO code uses bdev in two places:

* to read blocksize to perform alignment checks in
  do_blockdev_direct_IO, but we do them in btrfs code before any call to
  DIO

* in the following call chain:

  do_direct_IO
    get_more_blocks
     sdio->get_block() <-- this is btrfs_get_blocks_direct

  subsequently the map_bh->b_dev member is used in clean_bdev_aliases
  and dio_new_bio to set the bio's bdev to that of the buffer_head.
  However, because we have provided a submit function dio_bio_submit
  calls our submission function and ignores the bdev.

So it's safe to pass any valid bdev that's used within the filesystem.
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

This is a preparatory patch for removing extent_map::bdev. There's some
history behind the code so this is only precaution to catch if things
break before the actual removal happens.

Logically, comparing a raw low-level block device (bdev) does not make
sense for extent maps (high-level objects). This had no effect in
practice but was quite confusing in the code.  The lookup_map is set iff
EXTENT_FLAG_FS_MAPPING is set.

The two pointers were stored in the same bytes and used potentially in
two meanings. Now they're split, so the asserts are in place to check
that the condition will not change.

The lookup map pointer misused bdev, this has been changed in commit
95617d69 ("btrfs: cleanup, stop casting for extent_map->lookup
everywhere") to the explicit type. But the semantics hasn't changed and
bdev was not actually used to decide if maps are mergeable.
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Instead of checking if we've read a BTRFS_CHUNK_ITEM_KEY from disk and
then process it we could just bail out early if the read disk key wasn't
a BTRFS_CHUNK_ITEM_KEY.

This removes a level of indentation and makes the code nicer to read.
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NJohannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

In btrfs_may_alloc_data_chunk() we're checking if the chunk type is of
type BTRFS_BLOCK_GROUP_DATA and if it is we process it.

Instead of checking if the chunk type is a BTRFS_BLOCK_GROUP_DATA chunk
we can negate the check and bail out early if it isn't.

This makes the code a bit more readable.
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NJohannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

In lock_stripe_add() we're caching the bucket for the stripe hash table
just for a single call to dereference the stripe hash.

If we just directly call rbio_bucket() we can safe the pointless local
variable.

Also move the dereferencing of the stripe hash outside of the variable
declaration block to not break over the 80 characters limit.
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NJohannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

In lock_stripe_add() we're traversing the stripe hash list and check if
the current list element's raid_map equals is equal to the raid bio's
raid_map. If both are equal we continue processing.

If we'd check for inequality instead of equality we can reduce one level
of indentation.
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NJohannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We don't modify the data passed to tracepoints, some of the declarations
are already const, add it to the rest.
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Remove typecasts from trace printk, adjust types and move typecast to
the assignment if necessary. When assigning, the types are more obvious
compared to matching the variables to the format strings.
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Instead of using an input pointer parameter as the return value and have
an int as the return type of find_desired_extent, rework the function to
directly return the found offset. Doing that the 'ret' variable in
btrfs_llseek_file can be removed. Additional (subjective) benefit is
that btrfs' llseek function now resemebles those of the other major
filesystems.
Reviewed-by: NJohannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Handle SEEK_END/SEEK_CUR in a single 'default' case by directly
returning from generic_file_llseek. This makes the 'out' label
redundant.  Finally return directly the vale from vfs_setpos. No
semantic changes.
Reviewed-by: NJohannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Modifying the file position is done on a per-file basis. This renders
holding the inode lock for writing useless and makes the performance of
concurrent llseek's abysmal.

Fix this by holding the inode for read. This provides protection against
concurrent truncates and find_desired_extent already includes proper
extent locking for the range which ensures proper locking against
concurrent writes. SEEK_CUR and SEEK_END can be done lockessly.

The former is synchronized by file::f_lock spinlock. SEEK_END is not
synchronized but atomic, but that's OK since there is not guarantee that
SEEK_END will always be at the end of the file in the face of tail
modifications.

This change brings ~82% performance improvement when doing a lot of
parallel fseeks. The workload essentially does:

    for (d=0; d<num_seek_read; d++)
      {
	/* offset %= 16777216; */
	fseek (f, 256 * d % 16777216, SEEK_SET);
	fread (buffer, 64, 1, f);
      }

Without patch:

num workprocesses = 16
num fseek/fread = 8000000
step = 256
fork 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

real	0m41.412s
user	0m28.777s
sys	2m16.510s

With patch:

num workprocesses = 16
num fseek/fread = 8000000
step = 256
fork 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

real	0m11.479s
user	0m27.629s
sys	0m21.040s
Signed-off-by: NNikolay Borisov <nborisov@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We can infer the ops from the type that is now passed to all functions
that would need it, this makes workspace_manager::ops redundant and can
be removed.
Reviewed-by: NJohannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Replace indirect calls to free_workspace by switch and calls to the
specific callbacks. This is mainly to get rid of the indirection due to
spectre vulnerability mitigations.
Reviewed-by: NJohannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We can infer the workspace_manager from type and the type will be used
in the following patch to call a common helper for free_workspace.
Reviewed-by: NJohannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Replace indirect calls to alloc_workspace by switch and calls to the
specific callbacks. This is mainly to get rid of the indirection due to
spectre vulnerability mitigations.
Reviewed-by: NJohannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: NNikolay Borisov <nborisov@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>