The allocator thread invalidates buckets (increments their generation
number) prior to discarding them and putting them on freelists. We've
had a short circuit path for some time to only update the in-memory
bucket mark when doing the invalidate if we're not invalidating cached
data, but that short-circuit path hasn't really been needed for quite
some time (likely since the btree key cache code was added).
We're deleting it now as part of deleting/converting code that uses the
in memory bucket array.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Implement a hash table, using cuckoo hashing, for empty buckets that are
waiting on a journal commit before they can be reused.
This replaces the journal_seq field of bucket_mark, and is part of
eventually getting rid of the in memory bucket array.
We may need to make bch2_bucket_needs_journal_commit() lockless, pending
profiling and testing.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
With BTREE_ITER_WITH_JOURNAL, there's no longer any restrictions on the
order we have to replay keys from the journal in, and we can also start
up journal reclaim right away - and delete a bunch of code.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
This adds a new btree iterator flag, BTREE_ITER_WITH_JOURNAL, that is
automatically enabled when initializing a btree iterator before journal
replay has completed - it overlays the contents of the journal with the
btree.
This lets us delete bch2_btree_and_journal_walk() and just use the
normal btree iterator interface instead - which also lets us delete a
significant amount of duplicated code.
Note that BTREE_ITER_WITH_JOURNAL is still unoptimized in this patch -
we're redoing the binary search over keys in the journal every time we
call bch2_btree_iter_peek().
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
The allocator needs to wait until the last update touching a bucket has
been commited before writing to it again. However, the code was checking
against the last dirty journal sequence number, not the last flushed
journal sequence number.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
The main in-memory bucket array is going away, but we'll still need to
keep bucket generations in memory, at least for now - ptr_stale() needs
to be an efficient operation.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Prep work for adding a hash table of open buckets - instead of embedding
a bch_extent_ptr, we need to refer to the bucket directly so that we're
not calling sector_to_bucket() in the hash table lookup code, which has
an expensive divide.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
If the allocator threads start before journal replay has finished
replaying alloc keys, journal replay might overwrite the allocator's
btree updates.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
This changes bch2_bucket_alloc_new_fs() to a simple bump allocator that
doesn't need to use the in memory bucket array, part of a larger patch
series to entirely get rid of the in memory bucket array, except for
gc/fsck.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
This moves some common code into alloc_mem_to_key(), which translates
from the in-memory format for a bucket to the btree key format.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
This adds a new helper that much like the one we have for inode updates,
that allocates the packed alloc key, packs it and calls
bch2_trans_update.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
When we added the stripe and stripe_redundancy fields to alloc keys, we
neglected to add them to the functions that convert back and forth with
the in-memory types.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Add fields to inode & alloc keys that record the journal sequence number
when they were most recently modified.
For alloc keys, this is needed to know what journal sequence number we
have to flush before the bucket can be reused. Currently this is tracked
in memory, but we'll be getting rid of the in memory bucket array.
For inodes, this is needed for fsync when the inode has been evicted
from the vfs cache. Currently we use a bloom filter per outstanding
journal buf - but that mechanism has been broken since we added the
ability to not issue a flush/fua for every journal write.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
This helps to unify the interface between bch2_mark_key() and
bch2_trans_mark_key() - and it also gives access to the journal
reservation and journal seq in the mark_key path.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
But we don't need to call it from outside the btree iterator code
anymore, since it's called by bch2_trans_begin() and
bch2_btree_path_traverse().
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
We return 1 to indicate kthread_should_stop() returned true - we
shouldn't be printing an error.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
This splits btree_iter into two components: btree_iter is now the
externally visible componont, and it points to a btree_path which is now
reference counted.
This means we no longer have to clone iterators up front if they might
be mutated - btree_path can be shared by multiple iterators, and cloned
if an iterator would mutate a shared btree_path. This will help us use
iterators more efficiently, as well as slimming down the main long lived
state in btree_trans, and significantly cleans up the logic for iterator
lifetimes.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
On transaction restart iterators won't be locked anymore - make sure
we're always checking for errors.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
This adds safe versions of bch2_varint_(encode|decode) that don't read
or write past the end of the buffer, or varint being encoded.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
This is to help debug a rare shutdown deadlock in the allocator code -
the btree code is leaking open_buckets.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Upcoming refactoring is going to change bch2_trans_update() to start
returning transaction restarts.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
We were missing a kthread_should_stop() check in the loop in
bch2_invalidate_buckets(), very occasionally leading to us getting stuck
while shutting down.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
We really need debug mode assertions that ca->ref and ca->io_ref are
used correctly.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This uses the kthread_wait_freezable() macro to simplify a lot of the
allocator thread code, along with cleaning up bch2_invalidate_bucket2().
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
It's not actually the case that iterators are always checked here -
__bch2_trans_commit() checks for that after running triggers.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Even with runtime gc (which currently isn't supported), runtime gc no
longer clears/recalculates the main set of bucket marks - it allocates
and calculates another set, updating the primary at the end.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
The owned_by_allocator field is a purely in memory thing, even if/when
we bring back GC at runtime there's no need for it to be recalculating
this field. This is prep work for pulling it out of struct bucket, and
eventually getting rid of the bucket array.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
It used to be necessary for the allocator thread to batch up
invalidating buckets when possible - but since we added the btree key
cache that hasn't been a concern, and now it's causing the allocator
thread to livelock when the filesystem is nearly full.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This is mkfs's job. Also, clean up the handling of feature bits some.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
We keep running into occasional bugs with btree transaction iterators
overflowing - this will make those bugs more visible.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This patch standardizes all the enums that have associated string tables
(probably more enums should have string tables).
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
We have a separate mechanism for ratelimiting copygc now - the pd
controller has only been causing problems.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Awhile back the meaning of is_available_bucket() and thus also
bch_dev_usage->buckets_unavailable changed to include buckets that are
owned by the allocator - this was so that the stat could be persisted
like other allocation information, and wouldn't have to be regenerated
by walking each bucket at mount time.
This broke copygc, which needs to consider buckets that are reclaimable
and haven't yet been grabbed by the allocator thread and moved onta
freelist. This patch fixes that by adding dev_buckets_reclaimable() for
copygc and the allocator thread, and cleans up some of the callers a bit.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This is just a band-aid fix for now.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
We're seeing failures to mount because of a failure to start the
allocator threads, which currently happens fairly late in the mount
process, after walking all metadata, and kthread_create() fails if
something has tried to kill the mount process, which is probably not
what we want.
This patch avoids this issue by creating, but not starting, the
allocator threads when we preallocate all of our other in memory data
structures.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This eliminates the need to scan every bucket to regenerate dev_usage at
mount time.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Originally, bcachefs - going back to bcache - stored, for each bucket, a
16 bit counter corresponding to how long it had been since the bucket
was read from. But, this required periodically rescaling counters on
every bucket to avoid wraparound. That wasn't an issue in bcache, where
we'd perodically rewrite the per bucket metadata all at once, but in
bcachefs we're trying to avoid having to walk every single bucket.
This patch switches to persisting 64 bit io clocks, corresponding to the
64 bit bucket timestaps introduced in the previous patch with
KEY_TYPE_alloc_v2.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This introduces a new version of KEY_TYPE_alloc, which uses the new
varint encoding introduced for inodes. This means we'll eventually be
able to support much larger bucket sizes (for SMR devices), and the
read/write time fields are expanded to 64 bits - which will be used in
the next patch to get rid of the periodic rescaling of those fields.
Also, for buckets that are members of erasure coded stripes, this adds
persistent fields for the index of the stripe they're members of and the
stripe redundancy. This is part of work to get rid of having to scan and
read into memory the alloc and stripes btrees at mount time.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
If we're invalidating a bucket that has cached data in it, data_type
won't be 0 - oops.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
More work towards getting rid of the in memory struct bucket: this path
adds code for marking superblock and journal buckets via the btree, and
uses it in the device add and journal resize paths.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This patch is working towards eventually getting rid of the in memory
struct bucket, and relying only on the btree representation.
Since bch2_invalidate_bucket() was only used for incrementing gens, not
invalidating cached data, no other counters were being changed as a side
effect - meaning it's safe for the allocator code to increment the
bucket gen directly.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This is to make it more amenable for serialization.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Alloc info isn't stored on a particular device, it makes no sense to
only be writing it out for rw members - this was causing fsck to not fix
alloc info errors, oops.
Also, make sure we write out alloc info in other repair paths.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
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>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
With the btree key cache code, we don't need to update the alloc btree
lazily - and this will mean we can remove the bch2_alloc_write() call in
the shutdown path.
Future work: we really need to expend the bucket IO clocks from 16 to 64
bits, so that we don't have to rescale them.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
The allocator usually doesn't increment bucket gens right away on
buckets that it's about to hand out (for reasons that need to be
documented), instead deferring that to whatever extent update first
references that bucket.
But stripe pointers reference buckets without changing bucket sector
counts, meaning we could end up with a pointer in a stripe with a gen
newer than the bucket it points to.
Fix this by adding a transactional trigger for KEY_TYPE_stripe that just
writes out the keys in the alloc btree for the buckets it points to.
Also - consolidate the code that checks pointer validity.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This fixes a bug where the clock hands in the journal and superblock
didn't match, because we were still incrementing the read clock hand
while read-only.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Now that we've got transactional alloc info updates (and have for
awhile), we don't need to write it out on shutdown, and we don't need to
write it out on startup except when GC found errors - this is a big
improvement to mount/unmount performance.
This patch also fixes a few bugs where we weren't writing out alloc
info (on new filesystems, and new devices) and should have been.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
On large filesystems reading in the alloc info takes a significant
amount of time. But we don't need to be calling into the fully general
bch2_mark_key() path, just open code what we need in
bch2_alloc_read_fn().
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Per device copygc threads don't move data to different devices and they
make fragmentation works - they don't make much sense anymore.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Turns out it's possible during shutdown for the allocator to get stuck
spinning on bch2_invalidate_buckets() without hitting any of the other
checks.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Also tweak the allocator to be more aggressive about keeping it full.
The recent changes to make updates to interior nodes transactional (and
thus generate updates to the alloc btree) all put more stress on the
btree node reserves.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
It's not needed anymore since we can now write to buckets before
updating the alloc btree.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Now that interior btree updates are fully transactional, we don't need
to write out alloc info in a loop. However, interior btree updates do
put more things in the journal, so we still need a loop in the RO
sequence.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This is to break a circular dependency in the shutdown path.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
We now update the alloc info (bucket sector counts) atomically with
journalling the update to the interior btree nodes, and we also set new
btree roots atomically with the journalled part of the btree update.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
When updates to interior nodes started being journalled, that meant that
after an unclean shutdown, until journal replay is done we can't walk
the btree without overlaying the updates from the journal.
The initial btree gc was changed to walk the btree overlaying keys from
the journal - but bch2_alloc_read() and bch2_stripes_read() were missed.
Major whoops...
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
- Always pass BTREE_INSERT_USE_RESERVE when writing alloc btree keys
- Don't strand buckest on the copygc freelist until after recovery is
done and we're starting copygc.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Introduce a new iterator that iterates over keys in the btree with keys
from the journal overlaid on top. This factors out what the erasure
coding init code was doing manually.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
The trigger flags really belong with individual btree_insert_entries,
not the transaction commit flags - this splits out those flags and
unifies them with the BCH_BUCKET_MARK flags. Todo - split out
btree_trigger.c from buckets.c
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Packed bkeys are padded up to 64 bit alignment, but the alloc bkey type
was not clearing the pad bytes after the last data byte. This left the
key possibly containing some random garbage at the end.
This problem was found using valgrind.
This patch also changes a path with the inode bkey to clear in the same
way.
Signed-off-by: Justin Husted <sigstop@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
The allocator needs to make sure there's buckets available on the
RESERVE_NONE freelist if at all possible - otherwise foreground IO will
get stuck.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Major simplification - gets rid of the need for marking buckets as
dirty, instead we write buckets if the in memory mark is different from
what's in the btree.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This fixes a bug in the journal replay -> extent_replay_key ->
split_compressed path, when we do an update that changes alloc info but
the alloc info in the btree isn't up to date yet.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This is prep work for the btree key cache: btree iterators will point to
either struct btree, or a new struct bkey_cached.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
If the allocator thread exited before bch2_dev_allocator_stop() was
called (because of an error), bch2_dev_allocator_quiesce() could hang.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
- Does not persist alloc info for stripes yet
- Also does not yet include filesystem block/sector counts yet, from
struct fs_usage
- Not made use of just yet
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
this lets us get rid of a lot of extra switch statements - in a lot of
places we dispatch on the btree node type, and then the key type, so
this is a nice cleanup across a lot of code.
Also improve the on disk format versioning stuff.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This means we can now use gc to verify the allocation information -
important for testing persistant alloc info
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
It's now possible to create and use a filesystem on a 512k device with
4k buckets (though at that size we still waste almost half to internal
reserves)
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
bch2_alloc_sectors_start() was a nightmare to work with - it's got some
tricky stuff to do, since it wants to use the buckets the writepoint
already has, unless they're not in the target it wants to write to,
unless it can't allocate from any other devices in which case it will
use those buckets if it has to - et cetera.
This restructures the code to start with a new empty list of open
buckets we're going to use for the new allocation, pulling buckets from
the write point's list as we decide that we really are going to use
them - making the code somewhat more functional and drastically easier
to understand.
Also fixes a bug where we could end up waiting on c->freelist_wait
(because allocating from one device failed) but return success from
bch2_bucket_alloc(), because allocating from a different device
succeeded.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
