btree node scan needs to not use the btree node cache: that causes
interference from prior failed reads and parallel workers.
Instead we need to allocate btree nodes that don't live in the btree
cache, so that we can call bch2_btree_node_read_done() directly.
This patch tweaks the low level helpers so they don't touch the btree
cache lists.
Cc: Nikita Ofitserov <himikof@gmail.com>
Reviewed-by: Nikita Ofitserov <himikof@gmail.com>
Reported-and-tested-by: Edoardo Codeglia <bcachefs@404.blue>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
We want recovery.curr_pass to be private to the recovery passes code,
for better showing recovery pass status; also, it may rewind and is
generally not the correct member to use.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
We'd like users to be able to debug without building custom kernels, so
this will help us get rid of CONFIG_BCACHEFS_DEBUG, at least for most
things.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Make the output slightly clearer, and include a counter for "nodes we
couldn't free because we would have gone under our reserve".
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Prevent jobs that do lots of scanning (i.e. evacuatee, scrub) from
causing OOMs.
The shrinker code seems to be having issues when it doesn't do any
freeing because it's just flipping off the acccessed bit - and the
accessed bit shouldn't be set on first use anyways.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Add the new helper printbuf_indent_add_nextline(), and use it in
__bch2_fsck_err() to centralize setting the indentation of multiline
fsck errors.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Factor out a version of bch2_btree_pos_to_text() that doesn't take a
pointer to a in-memory btree node, to be used for btree node scrub.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Instead of walking every extent and every backpointer it points to,
first sum up backpointers in each bucket and check for mismatches, and
only look for missing backpointers if mismatches were detected, and only
check extents in those buckets.
This is a major fsck scalability improvement, since the two backpointers
passes (backpointers -> extents and extents -> backpointers) are the
most expensive fsck passes by far.
Additionally, to speed up the upgrade for backpointer bucket gens, or in
situations when we have to rebuild alloc info, add a special case for
when no backpointers are found in a bucket - don't check each individual
backpointer (in particular, avoiding the write buffer flushes), just
recreate them.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Fix an assertion pop from the recent btree cache freelist fixes.
Fixes: baefd3f849 ("bcachefs: btree_cache.freeable list fixes")
Reported-by: Tyler <th020394@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
btree_root entries for unknown btree IDs are created during recovery,
before reading those btree roots.
But btree_node_scan may find btree nodes with unknown btree IDs when we
haven't seen roots for those btrees.
Reported-by: syzbot+1f202d4da221ec6ebf8e@syzkaller.appspotmail.com
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
When allocating new btree nodes, we were leaving them on the freeable
list - unlocked - allowing them to be reclaimed: ouch.
Additionally, bch2_btree_node_free_never_used() ->
bch2_btree_node_hash_remove was putting it on the freelist, while
bch2_btree_node_free_never_used() was putting it back on the btree
update reserve list - ouch.
Originally, the code was written to always keep btree nodes on a list -
live or freeable - and this worked when new nodes were kept locked.
But now with the cycle detector, we can't keep nodes locked that aren't
tracked by the cycle detector; and this is fine as long as they're not
reachable.
We also have better and more robust leak detection now, with memory
allocation profiling, so the original justification no longer applies.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
In backpointers fsck, we do a seqential scan of one btree, and check
references to another: extents <-> backpointers
Checking references generates random lookups, so we want to pin that
btree in memory (or only a range, if it doesn't fit in ram).
Previously, this was done with a simple check in the shrinker - "if
btree node is in range being pinned, don't free it" - but this generated
OOMs, as our shrinker wasn't well behaved if there was less memory
available than expected.
Instead, we now have two different shrinkers and lru lists; the second
shrinker being for pinned nodes, with seeks set much higher than normal
- so they can still be freed if necessary, but we'll prefer not to.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
32 bits won't overflow any time soon, but size_t is the correct type for
counting objects in memory.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
When freeing in a shrinker callback, we need to notify memory reclaim,
so it knows forward progress has been made.
Normally this is done in e.g. slab code, but we're not freeing through
slab - or rather we are, but these allocations are big, and use the
kmalloc_large() path.
This is really a bug in the slub code, but we're working around it here
for now.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
It's really not needed: the only locks used here are the btree cache
lock, which we drop for GFP_WAIT allocations, and btree node locks - but
we also drop those for GFP_WAIT allocations.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
We weren't always so strict about trans->locked state - but now we are,
and new assertions are shaking some bugs out.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Use try_cmpxchg() family of functions instead of
cmpxchg (*ptr, old, new) == old. x86 CMPXCHG instruction returns
success in ZF flag, so this change saves a compare after cmpxchg
(and related move instruction in front of cmpxchg).
Also, try_cmpxchg() implicitly assigns old *ptr value to "old" when
cmpxchg fails. There is no need to re-read the value in the loop.
No functional change intended.
Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
Cc: Kent Overstreet <kent.overstreet@linux.dev>
Cc: Brian Foster <bfoster@redhat.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Since the key cache shrinker walks the rhashtable, a mostly empty
rhashtable leads to really nasty reclaim performance issues.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This adds distinct counters for every reason the btree node shrinker can
fail to free an object - if our shrinker isn't making progress, this
will tell us why.
Signed-off-by: Daniel Hill <daniel@gluo.nz>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Currently, the reflink_p gc trigger does repair as well - turning a
reflink_p key into an error key if the reflink_v it points to doesn't
exist.
This won't work with online check/repair, because the repair path once
online will be subject to transaction restarts, but BTREE_TRIGGER_gc is
not idempotant - we can't run it multiple times if we get a transaction
restart.
So we need to split these paths; to do so this patch calls
check_fix_ptrs() by a new general path - a new trigger type,
BTREE_TRIGGER_check_repair.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
We shouldn't be doing the unlock/relock dance when we're not using a
path - this fixes an assertion pop when called from btree node scan.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Consolidate bch2_gc_check_topology() and btree_node_interior_verify(),
and replace them with an improved version,
bch2_btree_node_check_topology().
This checks that children of an interior node correctly span the full
range of the parent node with no overlaps.
Also, ensure that topology repairs at runtime are always a fatal error;
in particular, this adds a check in btree_iter_down() - if we don't find
a key while walking down the btree that's indicative of a topology error
and should be flagged as such, not a null ptr deref.
Some checks in btree_update_interior.c remaining BUG_ONS(), because we
already checked the node for topology errors when starting the update,
and the assertions indicate that we _just_ corrupted the btree node -
i.e. the problem can't be that existing on disk corruption, they
indicate an actual algorithmic bug.
In the future, we'll be annotating the fsck errors list with which
recovery pass corrects them; the open coded "run explicit recovery pass
or fatal error" in bch2_btree_node_check_topology() will in the future
be done for every fsck_err() call.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Various phases of fsck involve checking references from one btree to
another: this means doing a sequential scan of one btree, and then
mostly random access into the second.
This is particularly painful for checking extents <-> backpointers; we
can prefetch btree node access on the sequential scan, but not on the
random access portion, and this is particularly painful on spinning
rust, where we'd like to keep the pipeline fairly full of btree node
reads so that the elevator can reduce seeking.
This patch implements prefetching and pinning of the portion of the
btree that we'll be doing random access to. We already calculate how
much of the random access btree will fit in memory so it's a fairly
straightforward change.
This will put more pressure on system memory usage, so we introduce a
new option, fsck_memory_usage_percent, which is the percentage of total
system ram that fsck is allowed to pin.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This converts -EIOs related to btree node errors to private error codes,
which will help with some ongoing debugging by giving us better error
messages.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
btree_and_journal_iter is old code that we want to get rid of, but we're
not ready to yet.
lack of btree node prefetching is, it turns out, a real performance
issue for fsck on spinning rust, so - add it.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
bcachefs btree nodes are big - typically 256k - and btree roots are
pinned in memory. As we're now up to 18 btrees, we now have significant
memory overhead in mostly empty btree roots.
And in the future we're going to start enforcing that certain btree node
boundaries exist, to solve lock contention issues - analagous to XFS's
AGIs.
Thus, we need to start allocating smaller btree node buffers when we
can. This patch changes code that refers to the filesystem constant
c->opts.btree_node_size to refer to the btree node buffer size -
btree_buf_bytes() - where appropriate.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
