Kent Overstreet [Sat, 25 Feb 2023 07:22:49 +0000 (02:22 -0500)]
bcachefs: Rework open bucket partial list allocation
Now, any open_bucket can go on the partial list: allocating from the
partial list has been moved to its own dedicated function,
open_bucket_add_bucets() -> bucket_alloc_set_partial().
In particular, this means that erasure coded buckets can safely go on
the partial list; the new location works with the "allocate an ec bucket
first, then the rest" logic.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Brian Foster [Thu, 2 Mar 2023 14:03:37 +0000 (09:03 -0500)]
bcachefs: don't bump key cache journal seq on nojournal commits
fstest generic/388 occasionally reproduces corruptions where an
inode has extents beyond i_size. This is a deliberate crash and
recovery test, and the post crash+recovery characteristics are
usually the same: the inode exists on disk in an early (i.e. just
allocated) state based on the journal sequence number associated
with the inode. Subsequent inode updates exist in the journal at
higher sequence numbers, but the inode hadn't been written back
before the associated crash and the post-crash recovery processes a
set of journal sequence numbers that doesn't include updates to the
inode. In fact, the sequence with the most recent inode key update
always happens to be the sequence just before the front of the
journal processed by recovery.
This last bit is a significant hint that the problem relates to an
on-disk journal update of the front of the journal. The root cause
of this problem is basically that the inode is updated (multiple
times) in-core and in the key cache, each time bumping the key cache
sequence number used to control the cache flush. The cache flush
skips one or more times, bumping the associated key cache journal
pin to the key cache seq value. This has a side effect of holding
the inode in memory a bit longer than normal, which helps exacerbate
this problem, but is also unsafe in certain cases where the key
cache seq may have been updated by a transaction commit that didn't
journal the associated key.
For example, consider an inode that has been allocated, updated
several times in the key cache, journaled, but not yet written back.
At this stage, everything should be consistent if the fs happens to
crash because the latest update has been journal. Now consider a key
update via bch2_extent_update_i_size_sectors() that uses the
BTREE_UPDATE_NOJOURNAL flag. While this update may not change inode
state, it can have the side effect of bumping ck->seq in
bch2_btree_insert_key_cached(). In turn, if a subsequent key cache
flush skips due to seq not matching the former, the ck->journal pin
is updated to ck->seq even though the most recent key update was not
journaled. If this pin happens to reside at the front (tail) of the
journal, this means a subsequent journal write can update last_seq
to a value beyond that which includes the most recent update to the
inode. If this occurs and the fs happens to crash before the inode
happens to flush, recovery will see the latest last_seq, fail to
recover the inode and leave the inode in the inconsistent state
described above.
To avoid this problem, skip the key cache seq update on NOJOURNAL
commits, except on initial pin add. Pass the insert entry directly
to bch2_btree_insert_key_cached() to make the associated flag
available and be consistent with btree_insert_key_leaf().
Signed-off-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Mon, 6 Mar 2023 10:29:12 +0000 (05:29 -0500)]
bcachefs: Journal resize fixes
- Fix a sleeping-in-atomic bug due to calling
bch2_journal_buckets_to_sb() under the journal lock.
- Additionally, now we mark buckets as journal buckets before adding
them to the journal in memory and the superblock. This ensures that
if we crash part way through we'll never be writing to journal
buckets that aren't marked correctly.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Tue, 28 Feb 2023 03:58:01 +0000 (22:58 -0500)]
bcachefs: Improved copygc pipelining
This improves copygc pipelining across multiple buckets: we now track
each in flight bucket we're evacuating, with separate moving_contexts.
This means that whereas previously we had to wait for outstanding moves
to complete to ensure we didn't try to evacuate the same bucket twice,
we can now just check buckets we want to evacuate against the pending
list.
This also mean we can run the verify_bucket_evacuated() check without
killing pipelining - meaning it can now always be enabled, not just on
debug builds.
This is going to be important for the upcoming erasure coding work,
where moving IOs that are being erasure coded will now skip the initial
replication step; instead the IOs will wait on the stripe to complete.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Sun, 5 Mar 2023 07:52:40 +0000 (02:52 -0500)]
bcachefs: Fix stripe reuse path
It's possible that we reuse a stripe that doesn't have quite the same
configuration as the stripe_head we're allocating from. In that case, we
have to make sure that the new stripe uses the settings from the stripe
we resue, not the stripe head, and make sure the buffer is allocated
correctly.
This fixes the ec_mixed_tiers test.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Sun, 5 Mar 2023 04:05:55 +0000 (23:05 -0500)]
bcachefs: Drop some anonymous structs, unions
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>
Kent Overstreet [Sun, 5 Mar 2023 03:36:02 +0000 (22:36 -0500)]
bcachefs: BKEY_PADDED_ONSTACK()
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>
Kent Overstreet [Thu, 2 Mar 2023 06:54:17 +0000 (01:54 -0500)]
bcachefs: RESERVE_stripe
Rework stripe creation path - new algorithm for deciding when to create
new stripes or reuse existing stripes.
We add a new allocation watermark, RESERVE_stripe, above RESERVE_none.
Then we always try to create a new stripe by doing RESERVE_stripe
allocations; if this fails, we reuse an existing stripe and allocate
buckets for it with the reserve watermark for the given write
(RESERVE_none or RESERVE_movinggc).
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Thu, 2 Mar 2023 02:47:07 +0000 (21:47 -0500)]
bcachefs: Mark stripe buckets with correct data type
Currently, we don't use bucket data type for tracking whether buckets
are part of a stripe; parity buckets are BCH_DATA_parity, but data
buckets in a stripe are BCH_DATA_user. There's a separate counter,
buckets_ec, outside the BCH_DATA_TYPES system for tracking number of
buckets on a device that are part of a stripe.
The trouble with this approach is that it's too coarse grained, and we
need better information on fragmentation for debugging copygc.
With this patch, data buckets in a stripe are now tracked as
BCH_DATA_stripe buckets.
This doesn't yet differentiate between erasure coded and non-erasure
coded data in a stripe bucket, nor do we yet track empty data buckets in
stripes.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Thu, 2 Mar 2023 07:12:18 +0000 (02:12 -0500)]
bcachefs: Plumb btree_trans through btree cache code
Soon, __bch2_btree_node_write() is going to require a btree_trans: zoned
device support is going to require a new allocation for every btree node
write. This is a bit of prep work.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Sun, 26 Feb 2023 20:48:39 +0000 (15:48 -0500)]
bcachefs: Fix for shared paths in write buffer flush
It's possible for bch2_write_buffer_flush_one() to end up with a shared
path, if called from a context that already has a btree iterator
pointing to a key being flushed. We have to be careful when that
happens, since we can't clone a path that holds write locks.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Wed, 22 Feb 2023 00:22:44 +0000 (19:22 -0500)]
bcachefs: Cached pointers should not be erasure coded
There's no reason to erasure code cached pointers: we'll always have
another copy, and it'll be cheaper to read the other copy than do a
reconstruct read. And erasure coded cached pointers would add
complications that we'd rather not have to deal with, so let's make sure
to disallow them.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Sun, 19 Feb 2023 05:49:51 +0000 (00:49 -0500)]
bcachefs: bch2_journal_entries_postprocess()
This brings back journal_entries_compact(), but in a more efficient form
- we need to do multiple postprocess steps, so iterate over the
journal entries being written just once to make it more efficient.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Sun, 19 Feb 2023 02:07:25 +0000 (21:07 -0500)]
bcachefs: Erasure coding: Track open stripes
This adds a new hash table for stripes being created or updated, instead
of hackily relying on the stripes heap.
This lets us reserve the slot for the new stripe up front, at the same
time as we would pick an existing stripe - if we were updating an
existing stripe - making the overall code more consistent.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Sat, 18 Feb 2023 01:50:55 +0000 (20:50 -0500)]
bcachefs: Erasure coding now uses bch2_bucket_alloc_trans
This code predates plumbing btree_trans through the bucket allocation
path: switching to it fixes a deadlock due to using multiple btree_trans
at the same time, which we never want to do.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Mon, 5 Dec 2022 15:24:19 +0000 (10:24 -0500)]
bcachefs: Fragmentation LRU
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>
Kent Overstreet [Fri, 17 Feb 2023 05:39:12 +0000 (00:39 -0500)]
bcachefs: bch2_mark_snapshot() now called like other triggers
This fixes a bug where bch2_mark_snapshot() wasn't called for existing
snapshot nodes being updated when child nodes were added.
This led to the data update path thinking the key being updated was for
a snapshot that didn't have children, causing it to fail to insert
whiteouts when splitting existing extents.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Fri, 17 Feb 2023 04:36:41 +0000 (23:36 -0500)]
bcachefs: Snapshot whiteout fix
When fully overwriting an existing extent, we may need to generate a
whiteout - not just if the extent being overwritten was in an older
snapshot, but also if it was overwriting an extent in an older snapshot.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Mon, 13 Feb 2023 00:24:34 +0000 (19:24 -0500)]
bcachefs: Fix failure to read btree roots
If failed to read a btree root - or if we're not using a btree root,
because of the reconstruct_alloc option - make sure we update the
corresponding info for the key/level for the root on disk.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Kent Overstreet [Thu, 11 Nov 2021 20:50:22 +0000 (15:50 -0500)]
bcachefs: Ensure btree node cache is not more than half dirty
Tweak journal reclaim to ensure the btree node cache isn't more
than half dirty so that memory reclaim can always make progress - the
same as we do for the btree key cache.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Daniel Hill [Fri, 6 Jan 2023 08:11:07 +0000 (21:11 +1300)]
bcachefs: don't block reads if we're promoting
The promote path calls data_update_init() and now that we take locks here,
there's potential for promote to block our read path, just error
when we can't take the lock instead of blocking.
Signed-off-by: Daniel Hill <daniel@gluo.nz> Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Make sure to check for lru entries that point to buckets that don't
exist as well as buckets in the wrong state, and improve the error
message we print out.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>