Superblock (and its copies) is the only data structure in btrfs which has a
fixed location on a device. Since we cannot overwrite in a sequential write
required zone, we cannot place superblock in the zone. One easy solution
is limiting superblock and copies to be placed only in conventional zones.
However, this method has two downsides: one is reduced number of superblock
copies. The location of the second copy of superblock is 256GB, which is in
a sequential write required zone on typical devices in the market today.
So, the number of superblock and copies is limited to be two. Second
downside is that we cannot support devices which have no conventional zones
at all.
To solve these two problems, we employ superblock log writing. It uses two
adjacent zones as a circular buffer to write updated superblocks. Once the
first zone is filled up, start writing into the second one. Then, when
both zones are filled up and before starting to write to the first zone
again, reset the first zone.
We can determine the position of the latest superblock by reading write
pointer information from a device. One corner case is when both zones are
full. For this situation, we read out the last superblock of each zone, and
compare them to determine which zone is older.
The following zones are reserved as the circular buffer on ZONED btrfs.
- primary superblock: offset 0B (and the following zone)
- first copy: offset 512G (and the following zone)
- Second copy: offset 4T (4096G, and the following zone)
If these reserved zones are conventional, superblock is written fixed at
the start of the zone without logging.
Currently, superblock reading/writing is done by pread/pwrite. This
commit replace the call sites with sbread/sbwrite to wrap the functions.
For zoned btrfs, btrfs_sb_io which is called from sbread/sbwrite
reverses the IO position back to a mirror number, maps the mirror number
into the superblock logging position, and do the IO.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Run a zoned filesystem on non-zoned devices. This is done by "slicing
up" the block device into fixed-sized chunks and emulate a conventional
zone on each of them. The emulated zone size is determined from the size
of device extent.
This is mainly aimed at testing of zoned filesystems, i.e. the zoned
chunk allocator, on regular block devices.
Currently, we always use EMULATED_ZONE_SIZE (256MiB) for the emulated
zone size. In the future, this will be customized by mkfs option.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Placing both data and metadata in a block group is impossible in ZONED
mode. For data, we can allocate a space for it and write it immediately
after the allocation. For metadata, however, we cannot do that, because the
logical addresses are recorded in other metadata buffers to build up the
trees. As a result, a data buffer can be placed after a metadata buffer,
which is not written yet. Writing out the data buffer will break the
sequential write rule.
Check and disallow MIXED_BG with ZONED mode.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The zone append write command has a maximum IO size restriction it
accepts. This is because a zone append write command cannot be split, as
we ask the device to place the data into a specific target zone and the
device responds with the actual written location of the data.
Introduce max_zone_append_size to zone_info and fs_info to track the
value, so we can limit all I/O to a zoned block device that we want to
write using the zone append command to the device's limits.
Zone append command is mandatory for zoned btrfs. So, reject a device
with max_zone_append_size == 0.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Introduce function btrfs_check_zoned_mode() to check if ZONED flag is
enabled on the file system and if the file system consists of zoned
devices with equal zone size.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Get the zone information (number of zones and zone size) from all the
devices, if the volume contains a zoned block device. To avoid costly
run-time zone report commands to test the device zones type during block
allocation, it also records all the zone status (zone type, write
pointer position, etc.).
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
With the zoned feature enabled, a zoned block device-aware btrfs
allocates block groups aligned to the device zones and always written in
sequential zones at the zone write pointer position.
It also supports "emulated" zoned mode on a non-zoned device. In the
emulated mode, btrfs emulates conventional zones by slicing the device
into fixed-size zones.
We don't support conversion from the ext4 volume with the zoned feature
because we can't be sure all the converted block groups are aligned to
zone boundaries.
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Likewise in the kernel code, provide fs_info access from struct
btrfs_device. This will help to unify the code between the kernel and
the userland.
Since fs_info can be NULL at the time of btrfs_add_to_fsid(), let's use
btrfs_open_devices() to set fs_info to the devices.
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
alloc_chunk_ctl::calc_size is actually the stripe_size in the kernel
side code. Let's rename it to clarify what the "calc" is.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Chunk_bytes_by_type() takes type, calc_size, and ctl as arguments. But
the first two can be obtained from the ctl. Let's drop these arguments
for simplicity.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Since commit b9444efb66 ("btrfs-progs: don't pretend RAID56 has a
different stripe length"), alloc_chunk_ctl::stripe_len is always fixed
to BTRFS_STRIPE_LEN. Let's replace alloc_chunk_ctl::stripe_len with
BTRFS_STRIPE_LEN, like in the kernel code.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Several calculations in the chunk allocation process use this pattern.
x /= y;
x *= y;
Replace this pattern with round_down().
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
In the DUP profile, we can use only half of the space available in a
device extent. Fix the calculation of calc_size for it.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
btrfs_alloc_data_chunk() and create_chunk() have the most part in common.
Let's rewrite btrfs_alloc_data_chunk() using create_chunk().
There are two differences between btrfs_alloc_data_chunk() and
create_chunk(). create_chunk() uses find_next_chunk() to decide the
logical address of the chunk, and it uses btrfs_alloc_dev_extent() to
decide the physical address of a device extent. On the other hand,
btrfs_alloc_data_chunk() uses *start for both logical and physical
addresses.
To support the btrfs_alloc_data_chunk()'s use case, we use ctl->start
and ctl->dev_offset. If these values are set (non-zero), use the
specified values as the address. It is safe to use 0 to indicate the
value is not set here. Because both lower addresses of logical
(0..BTRFS_FIRST_CHUNK_TREE_OBJECT_ID) and physical
(0..BTRFS_BLOCK_RESERVED_1M_FOR_SUPER) are reserved.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Factor out create_chunk() from btrfs_alloc_chunk(). This new function
creates a chunk.
There is no functional changes.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Factor out decide_stripe_size() from btrfs_alloc_chunk(). This new
function calculates the actual stripe size to allocate and decides the
size of a stripe (ctl->calc_size).
This commit has no functional changes.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Move parameter initialization code for regular allocator to
init_alloc_chunk_ctl_policy_regular(). This will help adding another
allocator in the future.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Convert alloc_chunk_ctl::type to take the original type in
btrfs_alloc_chunk(). This will help refactoring in the following commits.
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Factor out the function dev_extent_search_start() from
find_free_dev_extent_start() to decide the starting position of a device
extent search.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Introduce chunk allocation policy for btrfs. This policy controls how
chunks and device extents are allocated from devices.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
In case the right buffer is emptied it's first set to NULL and
subsequently it's dereferenced to get its size to pass to root_sub_used.
This naturally leads to a NULL pointer dereference. The correct thing to
do is to pass the stashed right->len in "blocksize".
Issue: #296
Pull-request: #360
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
For passing authentication keys to the checksumming functions we need a
container for the key.
Pass in a btrfs_fs_info to btrfs_csum_data() so we can use the fs_info
as a container for the authentication key.
Note this is not always possible for all callers of btrfs_csum_data() so
we're just passing in NULL for now
Functions calling btrfs_csum_data() with a NULL fs_info argument are
currently not supported in the context of an authenticated file system.
Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: David Sterba <dsterba@suse.com>
Extending open_ctree with more parameters would be difficult, we'll need
to add more so factor out the parameters to a structure for easier
extension.
Signed-off-by: David Sterba <dsterba@suse.com>
Currently only single checksum byte is printed. Fix it so that
the whole checksum is printed, in the order as the bytes are stored in
the buffer. This matches what kernel does, though it might not
correspond to the cases of CRC32C and XXHASH as if they were stored in
integer variable and printed in the native format. For consistency we
need to print the same format.
Signed-off-by: Dāvis Mosāns <davispuh@gmail.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When the btrfs_read_fs_root() function is searching a ROOT_ITEM with
location key offset other than -1, it currently fails via BUG_ON.
The offset can have other value than -1, though. This can happen for
example if a subvolume is renamed:
$ btrfs subvolume create X && sync
Create subvolume './X'
$ btrfs inspect-internal dump-tree /dev/root | grep -B 2 'name: X$
location key (270 ROOT_ITEM 18446744073709551615) type DIR
transid 283 data_len 0 name_len 1
name: X
$ mv X Y && sync
$ btrfs inspect-internal dump-tree /dev/root | grep -B 2 'name: Y$
location key (270 ROOT_ITEM 0) type DIR
transid 285 data_len 0 name_len 1
name: Y
As can be seen the offset changed from -1ULL to 0.
Do not fail in this case.
Signed-off-by: Marek Behún <marek.behun@nic.cz>
CC: Qu Wenruo <wqu@suse.com>
CC: Tom Rini <trini@konsulko.com>
Signed-off-by: David Sterba <dsterba@suse.com>
While debugging a corruption problem I realized we don't spit out the
flags for nodes, which is needed when debugging relocation problems so
we know which nodes are the RELOC root items and which are the actual fs
tree's items. Fix this by unifying the header printing helper so both
leaf's and nodes get the same information printed out.
node 41070940160 level 1 items 34 free space 87 generation 7709536 owner ROOT_TREE
node 41070940160 flags 0x1(WRITTEN) backref revision 1
Same for leaves:
leaf 41070944256 items 12 free space 515 generation 7709536 owner ROOT_TREE
leaf 41070944256 flags 0x1(WRITTEN) backref revision 1
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Add a runtime feature (-R) flag for the free space tree. A filesystem
that is mkfs'd with -R free-space-tree then mounted with no options has
the same contents as one mkfs'd without the option, then mounted with
'-o space_cache=v2'.
The only tricky thing is in exactly how to call the tree creation code.
Using btrfs_create_free_space_tree as is did not quite work, because an
extra reference to the eb (root->commit_root) is leaked, which mkfs
complains about with a warning. I opted to follow how the uuid tree is
created by adding it to the dirty roots list for cleanup by
commit_tree_roots in commit_transaction. As a result,
btrfs_create_free_space_tree no longer exactly matches the version in
the kernel sources.
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Boris Burkov <boris@bur.io>
Signed-off-by: David Sterba <dsterba@suse.com>
This function exists in kernel side but using the _item suffix, and
objectid argument is placed before the name argument. Change the
function to reflect the kernel version.
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Marcos Paulo de Souza <mpdesouza@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
This would sync the code between kernel and btrfs-progs, and save at
least 1 byte for each btrfs_block_group_cache.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The implementation of ulist_* is same for kernel and userspace, without
dependencies, so we can keep it separately for code sync.
Signed-off-by: David Sterba <dsterba@suse.com>
