Move BCACHEFS_STATFS_MAGIC value to UAPI <linux/magic.h> under
BCACHEFS_SUPER_MAGIC definition (use common approach for name) and reuse the
definition in bcachefs_format.h BCACHEFS_STATFS_MAGIC.
There are other bcachefs magic definitions: BCACHE_MAGIC, BCHFS_MAGIC,
which use UUID_INIT() and are used only in libbcachefs. Therefore move
only BCACHEFS_STATFS_MAGIC value, which can be used outside of
libbcachefs for f_type field in struct statfs in statfs() or fstatfs().
Suggested-by: Su Yue <glass.su@suse.com>
Signed-off-by: Petr Vorel <pvorel@suse.cz>
Acked-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
On recovery from clean shutdown we don't typically read the journal, but
we still want to avoid overwriting existing entries in the journal for
list_journal debugging.
Thus, add some fields to the member info section so we can remember
where we left off.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
The bucket_gens array is a single array allocation (one byte per
bucket), and kernel allocations are still limited to INT_MAX.
Check this limit to avoid failing the bucket_gens array allocation.
Reported-by: syzbot+b29f436493184ea42e2b@syzkaller.appspotmail.com
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
KEY_TYPE_error is left behind when we have to delete all pointers in an
extent in fsck; it allows errors to be correctly returned by reads
later.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This adds a small (64 bit) per-device bitmap that tracks ranges that
have btree nodes, for accelerating btree node scan if it is ever needed.
- New helpers, bch2_dev_btree_bitmap_marked() and
bch2_dev_bitmap_mark(), for checking and updating the bitmap
- Interior btree update path updates the bitmaps when required
- The check_allocations pass has a new fsck_err check,
btree_bitmap_not_marked
- New on disk format version, mi_btree_mitmap, which indicates the new
bitmap is present
- Upgrade table lists the required recovery pass and expected fsck error
- Btree node scan uses the bitmap to skip ranges if we're on the new
version
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This is needed for building Rust bindings on big endian architectures
like s390x. Currently this is only done in userspace, but it might
happen in-kernel in the future. When creating a Rust binding for struct
bkey, the "packed" attribute is needed to get a type with the correct
member offsets in the big endian case. However, rustc does not allow
types to have both a "packed" and "align" attribute. Thus, in order to
get a Rust type compatible with the C type, we must omit the "aligned"
attribute in C.
This does not affect the struct's size or member offsets, only its
toplevel alignment, which should be an acceptable impact.
The little endian version can have the "align" attribute because the
"packed" attr is redundant, and rust-bindgen will omit the "packed" attr
when an "align" attr is present and it can do so without changing a
type's layout
Signed-off-by: Thomas Bertschinger <tahbertschinger@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Add a btree to record a parent -> child subvolume relationships,
according to the filesystem heirarchy.
The subvolume_children btree is a bitset btree: if a bit is set at pos
p, that means p.offset is a child of subvolume p.inode.
This will be used for efficiently listing subvolumes, as well as
recursive deletion.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This gives us a way to record the date and time every journal entry was
written - useful for debugging.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Add a field to bch_snapshot for creation time; this will be important
when we start exposing the snapshot tree to userspace.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Add new fields for split brain detection:
- bch_member->seq, which tracks the sequence number of the last superblock
write that happened to each member device
- bch_sb->write_time, which tracks the time of the last superblock write,
to allow detection of when two members have diverged but had the same
number of superblock writes.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Previosuly, the transaction commit path would have to add keys to the
btree write buffer as a separate operation, requiring additional global
synchronization.
This patch introduces a new journal entry type, which indicates that the
keys need to be copied into the btree write buffer prior to being
written out. We switch the journal entry type back to
JSET_ENTRY_btree_keys prior to write, so this is not an on disk format
change.
Flushing the btree write buffer may require pulling keys out of journal
entries yet to be written, and quiescing outstanding journal
reservations; we previously added journal->buf_lock for synchronization
with the journal write path.
We also can't put strict bounds on the number of keys in the journal
destined for the write buffer, which means we might overflow the size of
the preallocated buffer and have to reallocate - this introduces a
potentially fatal memory allocation failure. This is something we'll
have to watch for, if it becomes an issue in practice we can do
additional mitigation.
The transaction commit path no longer has to explicitly check if the
write buffer is full and wait on flushing; this is another performance
optimization. Instead, when the btree write buffer is close to full we
change the journal watermark, so that only reservations for journal
reclaim are allowed.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
- add a tracepoint for write_buffer_flush_sync; this is expensive
- fix the write_buffer_flush_slowpath tracepoint
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This counter is redundant; it's simply the sum of BCH_DATA_stripe and
BCH_DATA_parity buckets.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Add a new superblock section that contains a list of
{ minor version, recovery passes, errors_to_fix }
that is - a list of recovery passes that must be run when downgrading
past a given version, and a list of errors to silently fix.
The upcoming disk accounting rewrite is not going to be fully
compatible: we're going to have to regenerate accounting both when
upgrading to the new version, and also from downgrading from the new
version, since the new method of doing disk space accounting is a
completely different architecture based on deltas, and synchronizing
them for every jounal entry write to maintain compatibility is going to
be too expensive and impractical.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Add two new superblock fields. Since the main section of the superblock
is now fully, we have to add a new variable length section for them -
bch_sb_field_ext.
- recovery_passes_requried: recovery passes that must be run on the
next mount
- errors_silent: errors that will be silently fixed
These are to improve upgrading and dwongrading: these fields won't be
cleared until after recovery successfully completes, so there won't be
any issues with crashing partway through an upgrade or a downgrade.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Renamed from trace_move_extent_alloc_mem_fail, because there are other
reasons we colud fail (disk space allocation failure).
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
bkey embeds a bpos that is misaligned on big endian; this is so that
bch2_bkey_swab() works correctly without having to differentiate between
packed and non-packed keys (a debatable design decision).
This means it can't have the __aligned() tag on big endian.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
rebalance_work entries may refer to entries in the extents btree, which
is a snapshots btree, or they may also refer to entries in the reflink
btree, which is not.
Hence rebalance_work keys may use the snapshot field but it's not
required to be nonzero - add a new btree flag to reflect this.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
gcc 10 seems to complain about array bounds in situations where gcc 11
does not - curious.
This unfortunately requires adding some casts for now; we may
investigate getting rid of our __u64 _data[] VLA in a future patch so
that our start[0] members can be VLAs.
Reported-by: John Stoffel <john@stoffel.org>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Add a new superblock section to keep counts of errors seen since
filesystem creation: we'll be addingcounters for every distinct fsck
error.
The new superblock section has entries of the for [ id, count,
time_of_last_error ]; this is intended to let us see what errors are
occuring - and getting fixed - via show-super output.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
We now track IO errors per device since filesystem creation.
IO error counts can be viewed in sysfs, or with the 'bcachefs
show-super' command.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This adds a new btree, rebalance_work, to eliminate scanning required
for finding extents that need work done on them in the background - i.e.
for the background_target and background_compression options.
rebalance_work is a bitset btree, where a KEY_TYPE_set corresponds to an
extent in the extents or reflink btree at the same pos.
A new extent field is added, bch_extent_rebalance, which indicates that
this extent has work that needs to be done in the background - and which
options to use. This allows per-inode options to be propagated to
indirect extents - at least in some circumstances. In this patch,
changing IO options on a file will not propagate the new options to
indirect extents pointed to by that file.
Updating (setting/clearing) the rebalance_work btree is done by the
extent trigger, which looks at the bch_extent_rebalance field.
Scanning is still requrired after changing IO path options - either just
for a given inode, or for the whole filesystem. We indicate that
scanning is required by adding a KEY_TYPE_cookie key to the
rebalance_work btree: the cookie counter is so that we can detect that
scanning is still required when an option has been flipped mid-way
through an existing scan.
Future possible work:
- Propagate options to indirect extents when being changed
- Add other IO path options - nr_replicas, ec, to rebalance_work so
they can be applied in the background when they change
- Add a counter, for bcachefs fs usage output, showing the pending
amount of rebalance work: we'll probably want to do this after the
disk space accounting rewrite (moving it to a new btree)
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
members_v2 has dynamically resizable entries so that we can extend
bch_member. The members can no longer be accessed with simple array
indexing Instead members_v2_get is used to find a member's exact
location within the array and returns a copy of that member.
Alternatively member_v2_get_mut retrieves a mutable point to a member.
Signed-off-by: Hunter Shaffer <huntershaffer182456@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Now that we have the logged operations btree, we can make
finsert/fcollapse atomic w.r.t. unclean shutdown as well.
This adds bch_logged_op_finsert to represent the state of an finsert or
fcollapse, which is a bit more complicated than truncate since we need
to track our position in the "shift extents" operation.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Previously, we guaranteed atomicity of truncate after unclean shutdown
with the BCH_INODE_I_SIZE_DIRTY flag - which required a full scan of the
inodes btree.
Recently the deleted inodes btree was added so that we no longer have to
scan for deleted inodes, but truncate was unfinished and that change
left it broken.
This patch uses the new logged operations btree to fix truncate
atomicity; we now log an operation that can be replayed at the start of
a truncate.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Add a new btree for long running logged operations - i.e. for logging
operations that we can't do within a single btree transaction, so that
they can be resumed if we crash.
Keys in the logged operations btree will represent operations in
progress, with the state of the operation stored in the value.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
It's no longer legal to use a zero size array as a flexible array
member - this causes UBSAN to complain.
This patch switches our zero size arrays to normal flexible array
members when possible, and inserts casts in other places (e.g. where we
use the zero size array as a marker partway through an array).
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
After deleteing snapshots, we may be left with a snapshot tree where
some nodes only have one child, and we have a linear chain.
Interior snapshot nodes are never used directly (i.e. they never have
subvolumes that point to them), they are only referered to by child
snapshot nodes - hence, they are redundant.
The existing code talks about redundant snapshot nodes as forming and
equivalence class; i.e. nodes for which snapshot_t->equiv is equal. In a
given equivalence class, we only ever need a single key at a given
position - i.e. multiple versions with different snapshot fields are
redundant.
The existing snapshot cleanup code deletes these redundant keys, but not
redundant nodes. It turns out this is buggy, because we assume that
after snapshot deletion finishes we should only have a single key per
equivalence class, but the btree update path doesn't preserve this -
overwriting keys in old snapshots doesn't check for the equivalence
class being equal, and thus we can end up with duplicate keys in the
same equivalence class and fsck complaining about snapshot deletion not
having run correctly.
The equivalence class notion has been leaking out of the core snapshots
code and into too much other code, i.e. fsck, so this patch takes a
different approach: snapshot deletion now moves keys to the node in an
equivalence class being kept (the leafiest node) and then deletes the
redundant nodes in the equivalance class.
Some work has to be done to correctly delete interior snapshot nodes;
snapshot node depth and skiplist fields for descendent nodes have to be
fixed.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
subvolume.c has gotten a bit large, this splits out a separate file just
for managing snapshot trees - BTREE_ID_snapshots.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
To ensure we aren't shooting ourselves in the foot after merge for
potentially doing future revisions for dirent or for storing multiple
names for casefolding, limit this to 512 for now.
Previously this define was linked to the max size a d_name in
bch_dirent could be.
Signed-off-by: Joshua Ashton <joshua@froggi.es>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Add a new bitset btree for inodes pending deletion; this means we no
longer have to scan the full inodes btree after an unclean shutdown.
Specifically, this adds:
- a trigger to update the deleted_inodes btree based on changes to the
inodes btree
- a new recovery pass
- and check_inodes is now only a fsck pass.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
clang had a few more warnings about enum conversion, and also didn't
like the opts.c initializer.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This refactoring centralizes defining per-btree properties.
bch2_key_types_allowed was also about to overflow a u32, so expand that
to a u64.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This extents KEY_TYPE_snapshot to include some new fields:
- depth, to indicate depth of this particular node from the root
- skip[3], skiplist entries for quickly walking back up to the root
These are to improve bch2_snapshot_is_ancestor(), making it O(ln(n))
instead of O(n) in the snapshot tree depth.
Skiplist nodes are picked at random from the set of ancestor nodes, not
some fixed fraction.
This introduces bcachefs_metadata_version 1.1, snapshot_skiplists.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Now that we've got forward compatibility sorted out, we should be doing
more frequent version upgrades in the future.
To avoid having to run a full fsck for every version upgrade, this
improves the BCH_METADATA_VERSIONS() table to explicitly specify a
bitmask of recovery passes to run when upgrading to or past a given
version.
This means we can also delete PASS_UPGRADE().
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This introduces major/minor versioning to the superblock version number.
Major version number changes indicate incompatible releases; we can move
forward to a new major version number, but not backwards. Minor version
numbers indicate compatible changes - these add features, but can still
be mounted and used by old versions.
With the recent patches that make it possible to roll out new btrees and
key types without breaking compatibility, we should be able to roll out
most new features without incompatible changes.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
The version_upgrade parameter is now an enum, not a bool, and it's
persistent in the superblock:
- compatible (default): upgrade to the latest compatible version
- incompatible: upgrade to latest incompatible version
- none
Currently all upgrades are incompatible upgrades, but the next release
will introduce major:minor versions.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Version upgrades are not atomic operations: when we do a version upgrade
we need to update the superblock before we start using new features, and
then when the upgrade completes we need to update the superblock again.
This adds a new superblock field so we can detect and handle incomplete
version upgrades.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
- endianness fixes
- mark some things static
- fix a few __percpu annotations
- fix silent enum conversions
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This adds a new helper for checking if an on-disk version is compatible
with the running version of bcachefs - prep work for introducing
major:minor version numbers.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This adds the extent entry for extents that rebalance needs to do
something with.
We're adding this ahead of the main rebalance_work patchset, because
adding new extent entries can't be done in a forwards-compatible way.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
We now have 20 bits for the btree ID in the on disk format - sufficient
for 1 million distinct btrees.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Add two new fields to bch_subvolume:
- otime: creation time
- parent: For snapshots, this is the id of the subvolume the snapshot
was created from
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This adds a new btree which gets us a persistent per-snapshot-tree
identifier.
- BTREE_ID_snapshot_trees
- KEY_TYPE_snapshot_tree
- bch_snapshot now has a field that points to a snapshot_tree
This is going to be used to designate one snapshot ID/subvolume out of a
given tree of snapshots as the "main" subvolume, so that we can do quota
accounting in that subvolume and not the rest.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
We don't store backpointers in alloc keys anymore, since we gained the
btree write buffer.
This patch drops support for backpointers in alloc keys, and revs the on
disk format version so that we know a fsck is required.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Rust bindgen doesn't cope well with anonymous structs and unions. This
patch drops the fancy anonymous structs & unions in bkey_i that let us
use the same helpers for bkey_i and bkey_packed; since bkey_packed is an
internal type that's never exposed to outside code, it's only a minor
inconvenienc.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Rust bindgen doesn't do anonymous structs very nicely: BKEY_PADDED()
only needs the anonymous struct when it's used on the stack, to
guarantee layout, not when it's embedded in another struct.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Now that we have much more efficient updates to the LRU btree, this
patch adds a new LRU that indexes buckets by fragmentation.
This means copygc no longer has to scan every bucket to find buckets
that need to be evacuated.
Changes:
- A new field in bch_alloc_v4, fragmentation_lru - this corresponds to
the bucket's position in the fragmentation LRU. We add a new field
for this instead of calculating it as needed because we may make the
fragmentation LRU optional; this field indicates whether a bucket is
on the fragmentation LRU.
Also, zoned devices will introduce variable bucket sizes; explicitly
recording the LRU position will be safer for them.
- A new copygc path for using the fragmentation LRU instead of
scanning every bucket and building up an in-memory heap.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This patch changes how the LRU index works:
Instead of using KEY_TYPE_lru where the bucket the lru entry points to
is part of the value, this switches to KEY_TYPE_set and encoding the
bucket we refer to in the low bits of the key.
This means that we no longer have to check for collisions when inserting
LRU entries. We'll be making using of this in the next patch, which adds
a btree write buffer - a pure write buffer for btree updates, where
updates are appended to a simple array and then periodically sorted and
batch inserted.
This is a new on disk format version, and a forced upgrade.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
To improve mount times, add a btree for just bucket gens, 256 of them
per key: this means we'll have to scan drastically less metadata at
startup.
This adds
- trigger for keeping it in sync with the all btree
- initialization code, for filesystems from previous versions
- new path for reading bucket gens
- new fsck code
And a new on disk format version.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This adds support for nocow mode, where we do writes in-place when
possible. Patch components:
- New boolean filesystem and inode option, nocow: note that when nocow
is enabled, data checksumming and compression are implicitly disabled
- To prevent in-place writes from racing with data moves
(data_update.c) or bucket reuse (i.e. a bucket being reused and
re-allocated while a nocow write is in flight, we have a new locking
mechanism.
Buckets can be locked for either data update or data move, using a
fixed size hash table of two_state_shared locks. We don't have any
chaining, meaning updates and moves to different buckets that hash to
the same lock will wait unnecessarily - we'll want to watch for this
becoming an issue.
- The allocator path also needs to check for in-place writes in flight
to a given bucket before giving it out: thus we add another counter
to bucket_alloc_state so we can track this.
- Fsync now may need to issue cache flushes to block devices instead of
flushing the journal. We add a device bitmask to bch_inode_info,
ei_devs_need_flush, which tracks devices that need to have flushes
issued - note that this will lead to unnecessary flushes when other
codepaths have already issued flushes, we may want to replace this with
a sequence number.
- New nocow write path: look up extents, and if they're writable write
to them - otherwise fall back to the normal COW write path.
XXX: switch to sequence numbers instead of bitmask for devs needing
journal flush
XXX: ei_quota_lock being a mutex means bch2_nocow_write_done() needs to
run in process context - see if we can improve this
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
- bch2_extent_merge checks unwritten bit
- read path returns 0s for unwritten extents without actually reading
- reflink path skips over unwritten extents
- bch2_bkey_ptrs_invalid() checks for extents with both written and
unwritten extents, and non-normal extents (stripes, btree ptrs) with
unwritten ptrs
- fiemap checks for unwritten extents and returns
FIEMAP_EXTENT_UNWRITTEN
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Move bi_size and bi_sectors into the non-varint portion of the inode, so
that the write path can update them without going through the relatively
expensive unpack/pack operations.
Other changes:
- Add a field for the offset of the varint section, so we can add new
non-varint fields without needing a new inode type, like alloc_v3
- Move bi_mode into the flags field, so that the varint section can be
u64 aligned
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This patch adds backpointers: we now have a reverse index from device
and offset on that device (specifically, offset within a bucket) back to
btree nodes and (non cached) data extents.
The first 40 backpointers within a bucket are stored in the alloc key;
after that backpointers spill over to the next backpointers btree. This
is to help avoid performance regressions from additional btree updates
on large streaming workloads.
This patch adds all the code for creating, checking and repairing
backpointers. The next patch in the series is going to use backpointers
for copygc - finally getting rid of the need to scan all extents to do
copygc.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This adds a new method of doing btree updates - a straight write buffer,
implemented as a flat fixed size array.
This is only useful when we don't need to read from the btree in order
to do the update, and when reading is infrequent - perfect for the LRU
btree.
This will make LRU btree updates fast enough that we'll be able to use
it for persistently indexing buckets by fragmentation, which will be a
massive boost to copygc performance.
Changes:
- A new btree_insert_type enum, for btree_insert_entries. Specifies
btree, btree key cache, or btree write buffer.
- bch2_trans_update_buffered(): updates via the btree write buffer
don't need a btree path, so we need a new update path.
- Transaction commit path changes:
The update to the btree write buffer both mutates global, and can
fail if there isn't currently room. Therefore we do all write buffer
updates in the transaction all at once, and also if it fails we have
to revert filesystem usage counter changes.
If there isn't room we flush the write buffer in the transaction
commit error path and retry.
- A new persistent option, for specifying the number of entries in the
write buffer.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This separates out the slowpath into a separate function, and inlines
bch2_alloc_v4_mut into bch2_trans_start_alloc_update(), the main place
it's called.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Add a new bcachefs-specific magic number for the superblock, instead of
continuing to use the old bcache magic number3
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
checkpatch.pl gives lots of warnings that we don't want - suggested
ignore list:
ASSIGN_IN_IF
UNSPECIFIED_INT - bcachefs coding style prefers single token type names
NEW_TYPEDEFS - typedefs are occasionally good
FUNCTION_ARGUMENTS - we prefer to look at functions in .c files
(hopefully with docbook documentation), not .h
file prototypes
MULTISTATEMENT_MACRO_USE_DO_WHILE
- we have _many_ x-macros and other macros where
we can't do this
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
We've outgrown our own deadlock avoidance strategy.
The btree iterator API provides an interface where the user doesn't need
to concern themselves with lock ordering - different btree iterators can
be traversed in any order. Without special care, this will lead to
deadlocks.
Our previous strategy was to define a lock ordering internally, and
whenever we attempt to take a lock and trylock() fails, we'd check if
the current btree transaction is holding any locks that cause a lock
ordering violation. If so, we'd issue a transaction restart, and then
bch2_trans_begin() would re-traverse all previously used iterators, but
in the correct order.
That approach had some issues, though.
- Sometimes we'd issue transaction restarts unnecessarily, when no
deadlock would have actually occured. Lock ordering restarts have
become our primary cause of transaction restarts, on some workloads
totally 20% of actual transaction commits.
- To avoid deadlock or livelock, we'd often have to take intent locks
when we only wanted a read lock: with the lock ordering approach, it
is actually illegal to hold _any_ read lock while blocking on an intent
lock, and this has been causing us unnecessary lock contention.
- It was getting fragile - the various lock ordering rules are not
trivial, and we'd been seeing occasional livelock issues related to
this machinery.
So, since bcachefs is already a relational database masquerading as a
filesystem, we're stealing the next traditional database technique and
switching to a cycle detector for avoiding deadlocks.
When we block taking a btree lock, after adding ourself to the waitlist
but before sleeping, we do a DFS of btree transactions waiting on other
btree transactions, starting with the current transaction and walking
our held locks, and transactions blocking on our held locks.
If we find a cycle, we emit a transaction restart. Occasionally (e.g.
the btree split path) we can not allow the lock() operation to fail, so
if necessary we'll tell another transaction that it has to fail.
Result: trans_restart_would_deadlock events are reduced by a factor of
10 to 100, and we'll be able to delete a whole bunch of grotty, fragile
code.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Also, do some reorganizing/renaming, convert atomic counters in bch_fs
to persistent counters, and add a few missing counters.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Like the previous patch for bucket invalidates, add another counter for
a core allocator path.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
