The next patch in the series is (finally!) going to change btree splits
(and interior updates in general) to not take intent locks all the way
up to the root - instead only locking the nodes they'll need to modify.

However, this will be introducing a race since if we're not holding a
write lock on a btree node it can be written out by another thread, and
then we might not have enough space for a new bset entry.

We can handle this by retrying - we just need to introduce a new error
path.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

In order to avoid locking all btree nodes up to the root for btree node
splits, we're going to have to introduce a new error path into
bch2_btree_insert_node(); this mean we can't have done any writes or
modified global state before that point.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

We'd like to prioritize aborting transactions that have done less work -
however, it appears breaking cycles by telling other threads to abort
may still be buggy, so disable that for now.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Some lock operations can't fail; a cycle of nofail locks is impossible
to recover from. So we want to get rid of these nofail locking
operations, but as this is tricky it'll be done incrementally.

If such a cycle happens, this patch prints out which codepaths are
involved so we know what to work on next.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Cached pointers are generally dropped, not moved: this led to an
assertion firing in the data update path when there were no new replicas
being written.

This path adds a data_options field for pointers to be dropped, and
tweaks move_extent() to check if we're only dropping pointers, not
writing new ones, before kicking off a data update operation.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

We should fix this, but for now this makes this more usable.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

When errors=panic, we want to make sure we print the error before
calling bch2_inconsistent_error().
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

btree_node_lock_nopath() is something we'd like to get rid of, it's
always prone to deadlocks if we accidentally are holding other locks,
because it doesn't mark the lock it's taking in a path: we'll want to
get rid of it in the future, but for now this patch works it by calling
bch2_trans_unlock().
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Useful debugging function.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

This changes bch2_check_for_deadlock() to print the longest chains it
finds - when we have a deadlock because the cycle detector isn't finding
something, this will let us see what it's missing.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

We were incorrectly returning -BCH_ERR_insufficient_devices when we'd
received a different error from bch2_bucket_alloc_trans(), which
(erronously) turns into -EROFS further up the call chain.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bch2_btree_delete_range_trans() was using btree_trans_too_many_iters()
to avoid path overflow, but this was buggy here (and also
btree_trans_too_many_iters() is suspect in general).

btree_trans_too_many_iters() only returns true when we're close to the
maximum number of paths - within 8 - but extent insert/delete assumes
that it can use more paths than that.

Instead, we need to call bch2_trans_begin() on every loop iteration.
Since we don't want to call bch2_trans_begin() (restarting the outer
transaction) if the call was a no-op - if we had no work to do - we have
to structure things a bit oddly.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

This refactoring puts our various allocation path counters into a
dedicated struct - the upcoming nocow patch is going to add another
counter.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

There was a rare bug when path->locks_want was nonzero, but not
BTREE_MAX_DEPTH, where we'd return on a valid node that wasn't locked -
oops.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Move slowpath code to a separate, non-inline function.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Prep work for further refactoring.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

This used to be needed more for buffered IO, but now the block layer has
writeback throttling - we can delete this now.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

It now includes more info - whether the bucket was for metadata or data
- and also call it in the same place as the bucket_alloc_fail
tracepoint.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

This function is fairly small and only used in two places: one very hot,
the other cold, so it should definitely be inlined.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

This moves the slowpath of check_pos_snapshot_overwritten() to a
separate function, and inlines the fast path - helping performance on
btrees that don't use snapshot and for users that aren't using
snapshots.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Previously, jset_validate() was formatting the initial part of an error
string for every entry it validating - expensive.

This moves that code to journal_entry_err_msg(), which is now only
called if there's an actual error.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

 - move slowpath code to a separate function, btree_path_overflow()
 - no need to use hweight64
 - copy nr_max_paths from btree_transaction_stats to btree_trans,
   avoiding a data dependency in the fast path
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Small performance optimization.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

We need counters to be initialized before initializing shrinkers - the
shrinker callbacks will update those counters. This fixes a segfault in
userspace.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

We've seen long error messages get truncated here, so convert to the new
bch2_print_string_as_lines().
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

 - factor out fsck_err_get()
 - if the "bcachefs (%s):" prefix has already been applied, don't
   duplicate it
 - convert to printbufs instead of static char arrays
 - tidy up control flow a bit
 - use bch2_print_string_as_lines(), to avoid messages getting truncated
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

This adds a helper for printing a large buffer one line at a time, to
avoid the 1k printk limit.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Most of the node_relock_fail trace events are generated from
bch2_btree_path_verify_level(), when debugcheck_iterators is enabled -
but we're not interested in these trace events, they don't indicate that
we're in a slowpath.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

We're still seeing OOM issues caused by the btree node cache shrinker
not sufficiently freeing memory: thus, this patch changes the shrinker
to not exit if __GFP_FS was not supplied.

Instead, tweak btree node memory allocation so that we never invoke
memory reclaim while holding the btree node cache lock.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

This is a major oopsy - we should always be unlocking before calling
closure_sync(), else we'll cause a deadlock.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

This fixes an obvious deadlock - whoops.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

We were checking for -EAGAIN, but we're not returned that when we didn't
pass a closure to wait with - oops.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

This is just a formatting/readability improvement.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Before we had the deadlock cycle detector, we didn't want to be holding
read locks when taking intent locks, because blocking on an intent lock
while holding a read lock was a lock ordering violation that could
cause a deadlock.

With the cycle detector this is no longer an issue, so this code can be
deleted.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>

In order for bch2_btree_node_lock_write_nofail() to never produce a
deadlock, we must ensure we're never holding read locks when using it.
Fortunately, it's only used from code paths where any read locks may be
safely dropped.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

This deletes our old lock ordering based deadlock avoidance code.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>

In the event that we're not finished debugging the cycle detector, this
adds a new file to debugfs that shows what the cycle detector finds, if
anything. By comparing this with btree_transactions, which shows held
locks for every btree_transaction, we'll be able to determine if it's
the cycle detector that's buggy or something else.
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>