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>
While a btree transaction is running, we hold a SRCU read lock on the
btree key cache that prevents btree key cache keys from being freed -
this is so that relock() operations won't access freed memory.
The downside of this is that long running btree transactions prevent
memory from being freed from the key cache. This adds a check in
bch2_trans_begin() - if the transaction has been running longer than 1
second, drop and retake the SRCU read lock and zero out pointers to
unlock key cache paths.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
We shouldn't be overloading standard error codes now that we have
provisions for bcachefs-specific errorcodes: this patch converts super.c
and super-io.c to per error site errcodes, with a bit of cleanup.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
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>
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>
Continuing the saga of introducing private dedicated error codes for
each error path, this patch converts ENOSPC to error codes that are
subtypes of ENOSPC. We've recently had a test failure where we got
-ENOSPC where we shouldn't have, and didn't have enough information to
tell where it came from, so this patch will solve that problem.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
The next patch is going to be adding private error codes for all the
places we return -ENOSPC.
Additionally, this patch updates return paths at all module boundaries
to call bch2_err_class(), to return the standard error code.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Now that we have error codes, with subtypes, we can switch to our own
error code for transaction restarts - and even better, a distinct error
code for each transaction restart reason: clearer code and better
debugging.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Instead of overloading standard error codes (EINTR/EAGAIN), and defining
short lists of error codes in multiple places that potentially end up
overlapping & conflicting, we're now going to have one master list of
error codes.
Error codes are defined with an x-macro: thus we also have
bch2_err_str() now.
Also, error codes have a class field. Now, instead of checking for
errors with ==, code should use bch2_err_matches(), which returns true
if the error is equal to or a sub-error of the error class.
This means we can define unique errors for every source location where
an error is generated, which will help improve our error messages.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Fsck now checks for keys in different snapshot IDs that are now
redundant due to other snapshots being deleted - it needs to for its own
algorithms to not get confused.
When it detects this it should re-run the post snapshot deletion cleanup
- this patch does that.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Start a new header, errcode.h, for bcachefs-private error codes - more
error codes will be converted later.
This patch just converts bucket_alloc_ret so that they can be mixed with
standard error codes and passed as ERR_PTR errors - the ec.c code was
doing this already, but incorrectly.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
