| Age | Commit message (Collapse) | Author |
|---|
|
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This is the last piece for btree key cache coherency:
We already have:
- btree iterator code checks the key cache when iterating over a cached
btree
- update path ensures that updates go to the key cache when updating a
cached btree
But for iterating over a cached btree to work, we need to ensure that if
a key exists in the key cache, it also exists in the btree - otherwise
the iterator code will skip past it and not check the key cache.
This patch implements that last piece: on a key cache update, if
creating a new key, we now also update the underlying btree.
This fixes a device removal bug, where deleting alloc info wasn't
correctly deleting all keys associated with a given device. It also
means we should be able to re-enable the key cache for inodes.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
If we allocate a buffer that's a bit bigger than necessary the
transaction commit path will be much less likely to have to reallocate -
which requires a transaction restart.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
These showed up when building for mips.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Since journal reclaim -> btree key cache flushing may require the
allocation of new btree nodes, it has an implicit dependency on copygc
in order to make forward progress - so we should avoid blocking copygc
unless the journal is really close to full.
This introduces watermarks to replace our single MAY_GET_UNRESERVED bit
in the journal, and adds a watermark for copygc and plumbs it through.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This adds some new assertions that we always take locks in the correct
order while running under traverse_all() - we've been seeing some
livelocks and a bit of strange behaviour, this helps ensure that
everything is working the way we expect.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Six locks have a percpu mode, which we use for interior btree nodes, as
well as btree key cache keys for the subvolumes btree. We've been
switching locks back and forth between percpu and non percpu mode as
needed, but it turns out this is racy - when we're reusing an existing
node, other threads could be attempting to lock it while we're switching
it between modes.
This patch fixes this by never switching 'struct btree' between the two
modes, and instead segragating them between two different freed lists.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
We were emitting two trace events on transaction restart in this code
path - delete the redundant one.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Updates to non key cache iterators will now be transparently redirected
to the key cache for cached btrees.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This is the start of cache coherency with the btree key cache - this
adds a btree iterator flag that causes lookups to also check the key
cache when we're iterating over the btree (not iterating over the key
cache).
Note that we could still race with another thread creating at item in
the key cache and updating it, since we aren't holding the key cache
locked if it wasn't found. The next patch for the update path will
address this by causing the transaction to restart if the key cache is
found to be dirty.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
btree_key_cache_flush_pos() uses BTREE_ITER_CACHED_NOFILL - but it
wasn't checking for !ck->valid. It does check for the entry being dirty,
so it shouldn't matter, but this refactor it a bit and adds and
assertion.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This improves the transaction restart tracepoints - adding distinct
tracepoints for all the locations and reasons a transaction might have
been restarted, and ensures that there's a tracepoint for every
transaction restart.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
- Add a shim uuid_unparse_lower() in the kernel, since %pU doesn't work
in userspace
- We don't need to print the bcachefs: or the filesystem name prefix in
userspace
- Improve a few error messages
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Symbol decoding, via %ps, isn't supported in userspace - this will also
be faster when we're using trans->fn in the fast path, as with the new
BCH_JSET_ENTRY_log journal messages.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This adds some missing diagnostics from rare but annoying to debug
runtime allocation failure paths.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This fixes some bugs when we hit an error very early in the filesystem
startup path, before most things have been initialized.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
For snapshots, we need to implement btree lookups that return the first
key that's an ancestor of the snapshot ID the lookup is being done in -
and filter out keys in unrelated snapshots. This patch adds the btree
iterator flag BTREE_ITER_FILTER_SNAPSHOTS which does that filtering.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This patch adds subvolume.c - support for the subvolumes and snapshots
btrees and related data types and on disk data structures. The next
patches will start hooking up this new code to existing code.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Figured out the bug we were chasing, and it had nothing to do with
locking btree iterators/paths out of order.
This reverts commit ff08733dd298c969aec7c7828095458f73fd5374.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
btree_path_traverse_all() traverses btree iterators in sorted order, and
thus shouldn't see transaction restarts due to potential deadlocks - but
sometimes we do. This patch adds some more assertions and tracks some
more state to help track this down.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
If an extent ends up with a replica that is encrypted an a replica that
isn't encrypted (due the user changing options), and then
copygc/rebalance moves one of the replicas by reading from the
unencrypted replica, we had a bug where we wouldn't correctly initialize
op->nonce - for each crc field in an extent, crc.offset + crc.nonce must
be equal.
This patch fixes that by moving op.nonce initialization to
bch2_migrate_write_init.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This was used for an optimization that hasn't existing in quite awhile
- iter->uptodate will probably be going away as well.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This is prep work for splitting btree_path out from btree_iter -
btree_path will not have a pointer to btree_trans.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Disfavoured, and should go away.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
With the recent transaction restart changes, it's no longer needed - all
transaction commits have BTREE_INSERT_NOUNLOCK semantics.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Start tracking when btree transactions have been restarted - and assert
that we're always calling bch2_trans_begin() immediately after
transaction restart.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
We're working to standardize handling of transaction restarts.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Downgrading of btree iterators is something that should only happen
explicitly.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
We weren't correctly verifying that we had interior node intent locks -
this patch also fixes bugs uncovered by the new assertions.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Add a new flag to control assertions about updating to internal snapshot
nodes, that normally should not be written to - to be used in an
upcoming patch.
Also do some renaming - trigger_flags is now update_flags.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
We're seeing livelocks that appear to be due to
bch2_btree_key_cache_scan repeatedly scanning and blocking other tasks
from using the key cache lock - we probably shouldn't be reporting
objects that can't actually be freed yet.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
bch2_varint_decode() can read up to 7 bytes past the end of the buffer,
which means we need to allocate slightly larger key cache buffers.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Flushing the btree key cache needs to use allocation reserves - journal
reclaim depends on flushing the btree key cache for making forward
progress, and the allocator and copygc depend on journal reclaim making
forward progress.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
We need to flush the btree key cache when it's too dirty, because
otherwise the shrinker won't be able to reclaim memory - this is done by
journal reclaim. But journal reclaim also kicks btree node writes: this
meant that btree node writes were getting kicked much too often just
because we needed to flush btree key cache keys.
This patch splits journal pins into two different lists, and teaches
journal reclaim to not flush btree node writes when it only needs to
flush key cache keys.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
JOURNAL_RES_GET_RESERVED should only be used for updatse that need to be
done to free up space in the journal. In particular, when we're flushing
keys from the key cache, if we're flushing them out of order we
shouldn't be using it, since we're using up our remaining space in the
journal without dropping a pin that will let us make forward progress.
With this patch, BTREE_INSERT_JOURNAL_RECLAIM without
BTREE_INSERT_JOURNAL_RESERVED may return -EAGAIN - we can't wait on
journal reclaim if we're already in journal reclaim.
This means we need to propagate these errors up to journal reclaim,
indicating that flushing a journal pin should be retried in the future.
This is prep work for a patch to change the way journal reclaim works,
to split out flushing key cache keys because the btree key cache is too
dirty from journal reclaim because we need space in the journal.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
With snapshots, we're going to need to differentiate between comparisons
that should and shouldn't include the snapshot field. bpos_cmp is now
the comparison function that does include the snapshot field, used by
core btree code.
Upper level filesystem code generally does _not_ want to compare against
the snapshot field - that code wants keys to compare as equal even when
one of them is in an ancestor snapshot.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
The btree key cache mutex was becoming a significant bottleneck - it was
mainly used to protect the lists of dirty, clean and freed cached keys.
This patch eliminates the dirty and clean lists - instead, when we need
to scan for keys to drop from the cache we iterate over the rhashtable,
and thus we're able to remove most uses of that lock.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This is a bit clearer than using bch2_btree_iter_free().
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
bch2_btree_iter_traverse() is supposed to ensure we have the correct
type of lock - it was downgrading if necessary, but if we entered with a
read lock it wasn't upgrading to an intent lock, oops.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Previously, we were using BTREE_INSERT_RESERVE in a lot of places where
it no longer makes sense.
- we now have more open_buckets than we used to, and the reserves work
better, so we shouldn't need to use BTREE_INSERT_RESERVE just because
we're holding open_buckets pinned anymore.
- We have the btree key cache for updates to the alloc btree, meaning
we no longer need the btree reserve to ensure the allocator can make
forward progress.
This means that we should only need a reserve for btree updates to
ensure that copygc can make forward progress.
Since it's now just for copygc, we can also fold RESERVE_BTREE into
RESERVE_MOVINGGC (the allocator's freelist reserve).
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Particularly on emergency shutdown we can end up having to clean up a
lot of dirty cached btree keys here.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
These only come up when building in userspace, for some reason.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Since we now always preallocate the maximum number of iterators when we
initialize a btree transaction, getting an iterator never fails - we can
delete a fair amount of error path code.
This patch also simplifies the iterator allocation code a bit.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
In the btree key cache code, failing to flush a dirty key is a serious
error, but it doesn't need to be a BUG_ON(), we can stop the filesystem
instead.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
The rhashtable code doesn't like when we destroy an rhashtable that was
never initialized
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This is to make tracing easier.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Ensuring the key cache isn't too dirty is critical for ensuring that the
shrinker can reclaim memory.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
The shrinker should start scanning for entries that can be freed oldest
to newest - this way, we can avoid scanning a lot of entries that are
too new to be freed.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
