bcachefs: Improved nocow locking

This improves the nocow lock table so that hash table entries have
multiple locks, and locks specify which bucket they're for - i.e. we can
now resolve hash collisions.

This is important because the allocator has to skip buckets that are
locked in the nocow lock table, and previously hash collisions would
cause it to spuriously skip unlocked buckets.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

10 files changed, 173 insertions, 65 deletions

diff --git a/fs/bcachefs/alloc_background.h b/fs/bcachefs/alloc_background.h
index d67fab0f5dc1..47336e9e1018 100644
--- a/fs/bcachefs/alloc_background.h
+++ b/fs/bcachefs/alloc_background.h
@@ -23,6 +23,16 @@ static inline bool bch2_dev_bucket_exists(struct bch_fs *c, struct bpos pos)
 		pos.offset < ca->mi.nbuckets;
 }
 
+static inline u64 bucket_to_u64(struct bpos bucket)
+{
+	return (bucket.inode << 48) | bucket.offset;
+}
+
+static inline struct bpos u64_to_bucket(u64 bucket)
+{
+	return POS(bucket >> 48, bucket & ~(~0ULL << 48));
+}
+
 static inline u8 alloc_gc_gen(struct bch_alloc_v4 a)
 {
 	return a.gen - a.oldest_gen;
diff --git a/fs/bcachefs/alloc_foreground.c b/fs/bcachefs/alloc_foreground.c
index 2010a9af0eb2..f1cfb90b6d54 100644
--- a/fs/bcachefs/alloc_foreground.c
+++ b/fs/bcachefs/alloc_foreground.c
@@ -28,6 +28,7 @@
 #include "io.h"
 #include "journal.h"
 #include "movinggc.h"
+#include "nocow_locking.h"
 
 #include <linux/math64.h>
 #include <linux/rculist.h>
diff --git a/fs/bcachefs/bcachefs.h b/fs/bcachefs/bcachefs.h
index 71f51459872d..febef9ac254b 100644
--- a/fs/bcachefs/bcachefs.h
+++ b/fs/bcachefs/bcachefs.h
@@ -206,7 +206,7 @@
 #include "bcachefs_format.h"
 #include "errcode.h"
 #include "fifo.h"
-#include "nocow_locking.h"
+#include "nocow_locking_types.h"
 #include "opts.h"
 #include "util.h"
 
diff --git a/fs/bcachefs/data_update.c b/fs/bcachefs/data_update.c
index d4cdfb48ab8f..7ef7bb613347 100644
--- a/fs/bcachefs/data_update.c
+++ b/fs/bcachefs/data_update.c
@@ -11,6 +11,7 @@
 #include "io.h"
 #include "keylist.h"
 #include "move.h"
+#include "nocow_locking.h"
 #include "subvolume.h"
 
 #include <trace/events/bcachefs.h>
diff --git a/fs/bcachefs/io.c b/fs/bcachefs/io.c
index 2bff7a24a830..d215973ae73b 100644
--- a/fs/bcachefs/io.c
+++ b/fs/bcachefs/io.c
@@ -27,6 +27,7 @@
 #include "journal.h"
 #include "keylist.h"
 #include "move.h"
+#include "nocow_locking.h"
 #include "rebalance.h"
 #include "subvolume.h"
 #include "super.h"
@@ -1411,7 +1412,7 @@ static void bch2_nocow_write(struct bch_write_op *op)
 	struct {
 		struct bpos	b;
 		unsigned	gen;
-		two_state_lock_t *l;
+		struct nocow_lock_bucket *l;
 	} buckets[BCH_REPLICAS_MAX];
 	unsigned nr_buckets = 0;
 	u32 snapshot;
@@ -1458,7 +1459,8 @@ retry:
 			buckets[nr_buckets].b = PTR_BUCKET_POS(c, ptr);
 			buckets[nr_buckets].gen = ptr->gen;
 			buckets[nr_buckets].l =
-				bucket_nocow_lock(&c->nocow_locks, buckets[nr_buckets].b);
+				bucket_nocow_lock(&c->nocow_locks,
+						  bucket_to_u64(buckets[nr_buckets].b));
 
 			prefetch(buckets[nr_buckets].l);
 			nr_buckets++;
@@ -1480,11 +1482,12 @@ retry:
 
 		for (i = 0; i < nr_buckets; i++) {
 			struct bch_dev *ca = bch_dev_bkey_exists(c, buckets[i].b.inode);
-			two_state_lock_t *l = buckets[i].l;
+			struct nocow_lock_bucket *l = buckets[i].l;
 			bool stale;
 
-			if (!bch2_two_state_trylock(l, BUCKET_NOCOW_LOCK_UPDATE))
-				__bch2_bucket_nocow_lock(&c->nocow_locks, l, BUCKET_NOCOW_LOCK_UPDATE);
+			__bch2_bucket_nocow_lock(&c->nocow_locks, l,
+						 bucket_to_u64(buckets[i].b),
+						 BUCKET_NOCOW_LOCK_UPDATE);
 
 			rcu_read_lock();
 			stale = gen_after(*bucket_gen(ca, buckets[i].b.offset), buckets[i].gen);
@@ -2910,11 +2913,6 @@ void bch2_fs_io_exit(struct bch_fs *c)
 
 int bch2_fs_io_init(struct bch_fs *c)
 {
-	unsigned i;
-
-	for (i = 0; i < ARRAY_SIZE(c->nocow_locks.l); i++)
-		two_state_lock_init(&c->nocow_locks.l[i]);
-
 	if (bioset_init(&c->bio_read, 1, offsetof(struct bch_read_bio, bio),
 			BIOSET_NEED_BVECS) ||
 	    bioset_init(&c->bio_read_split, 1, offsetof(struct bch_read_bio, bio),
diff --git a/fs/bcachefs/nocow_locking.c b/fs/bcachefs/nocow_locking.c
index b325fb105322..bff6267158cc 100644
--- a/fs/bcachefs/nocow_locking.c
+++ b/fs/bcachefs/nocow_locking.c
@@ -4,12 +4,116 @@
 #include "nocow_locking.h"
 #include "util.h"
 
+#include <linux/closure.h>
+
+bool bch2_bucket_nocow_is_locked(struct bucket_nocow_lock_table *t, struct bpos bucket)
+{
+	u64 dev_bucket = bucket_to_u64(bucket);
+	struct nocow_lock_bucket *l = bucket_nocow_lock(t, dev_bucket);
+	unsigned i;
+
+	for (i = 0; i < ARRAY_SIZE(l->b); i++)
+		if (l->b[i] == dev_bucket && atomic_read(&l->l[i]))
+			return true;
+	return false;
+}
+
+void bch2_bucket_nocow_unlock(struct bucket_nocow_lock_table *t, struct bpos bucket, int flags)
+{
+	u64 dev_bucket = bucket_to_u64(bucket);
+	struct nocow_lock_bucket *l = bucket_nocow_lock(t, dev_bucket);
+	int lock_val = flags ? 1 : -1;
+	unsigned i;
+
+	for (i = 0; i < ARRAY_SIZE(l->b); i++)
+		if (l->b[i] == dev_bucket) {
+			if (!atomic_sub_return(lock_val, &l->l[i]))
+				closure_wake_up(&l->wait);
+			return;
+		}
+
+	BUG();
+}
+
+static bool bch2_bucket_nocow_trylock(struct nocow_lock_bucket *l,
+				      u64 dev_bucket, int flags)
+{
+	int v, lock_val = flags ? 1 : -1;
+	unsigned i;
+
+	spin_lock(&l->lock);
+
+	for (i = 0; i < ARRAY_SIZE(l->b); i++)
+		if (l->b[i] == dev_bucket)
+			goto got_entry;
+
+	for (i = 0; i < ARRAY_SIZE(l->b); i++)
+		if (!atomic_read(&l->l[i])) {
+			l->b[i] = dev_bucket;
+			goto take_lock;
+		}
+fail:
+	spin_unlock(&l->lock);
+	return false;
+got_entry:
+	v = atomic_read(&l->l[i]);
+	if (lock_val > 0 ? v < 0 : v > 0)
+		goto fail;
+take_lock:
+	atomic_add(lock_val, &l->l[i]);
+	spin_unlock(&l->lock);
+	return true;
+}
+
 void __bch2_bucket_nocow_lock(struct bucket_nocow_lock_table *t,
-			      two_state_lock_t *l, int flags)
+			      struct nocow_lock_bucket *l,
+			      u64 dev_bucket, int flags)
+{
+	if (!bch2_bucket_nocow_trylock(l, dev_bucket, flags)) {
+		struct bch_fs *c = container_of(t, struct bch_fs, nocow_locks);
+		u64 start_time = local_clock();
+
+		__closure_wait_event(&l->wait, bch2_bucket_nocow_trylock(l, dev_bucket, flags));
+		bch2_time_stats_update(&c->times[BCH_TIME_nocow_lock_contended], start_time);
+	}
+}
+
+void bch2_nocow_locks_to_text(struct printbuf *out, struct bucket_nocow_lock_table *t)
 {
-	struct bch_fs *c = container_of(t, struct bch_fs, nocow_locks);
-	u64 start_time = local_clock();
+	unsigned i, nr_zero = 0;
+	struct nocow_lock_bucket *l;
+
+	for (l = t->l; l < t->l + ARRAY_SIZE(t->l); l++) {
+		unsigned v = 0;
+
+		for (i = 0; i < ARRAY_SIZE(l->l); i++)
+			v |= atomic_read(&l->l[i]);
+
+		if (!v) {
+			nr_zero++;
+			continue;
+		}
+
+		if (nr_zero)
+			prt_printf(out, "(%u empty entries)\n", nr_zero);
+		nr_zero = 0;
+
+		for (i = 0; i < ARRAY_SIZE(l->l); i++)
+			if (atomic_read(&l->l[i]))
+				prt_printf(out, "%llu: %i ", l->b[i], atomic_read(&l->l[i]));
+		prt_newline(out);
+	}
+
+	if (nr_zero)
+		prt_printf(out, "(%u empty entries)\n", nr_zero);
+}
+
+int bch2_fs_nocow_locking_init(struct bch_fs *c)
+{
+	unsigned i;
+
+	for (i = 0; i < ARRAY_SIZE(c->nocow_locks.l); i++)
+		spin_lock_init(&c->nocow_locks.l[i].lock);
 
-	__bch2_two_state_lock(l, flags & BUCKET_NOCOW_LOCK_UPDATE);
-	bch2_time_stats_update(&c->times[BCH_TIME_nocow_lock_contended], start_time);
+	return 0;
 }
diff --git a/fs/bcachefs/nocow_locking.h b/fs/bcachefs/nocow_locking.h
index 2a7a9f44e88e..45258cc34614 100644
--- a/fs/bcachefs/nocow_locking.h
+++ b/fs/bcachefs/nocow_locking.h
@@ -2,54 +2,38 @@
 #ifndef _BCACHEFS_NOCOW_LOCKING_H
 #define _BCACHEFS_NOCOW_LOCKING_H
 
-#include "bcachefs_format.h"
-#include "two_state_shared_lock.h"
+#include "bcachefs.h"
+#include "alloc_background.h"
+#include "nocow_locking_types.h"
 
 #include <linux/hash.h>
 
-#define BUCKET_NOCOW_LOCKS_BITS		10
-#define BUCKET_NOCOW_LOCKS		(1U << BUCKET_NOCOW_LOCKS_BITS)
-
-struct bucket_nocow_lock_table {
-	two_state_lock_t		l[BUCKET_NOCOW_LOCKS];
-};
-
-#define BUCKET_NOCOW_LOCK_UPDATE	(1 << 0)
-
-static inline two_state_lock_t *bucket_nocow_lock(struct bucket_nocow_lock_table *t,
-						  struct bpos bucket)
+static inline struct nocow_lock_bucket *bucket_nocow_lock(struct bucket_nocow_lock_table *t,
+							  u64 dev_bucket)
 {
-	u64 dev_bucket = bucket.inode << 56 | bucket.offset;
 	unsigned h = hash_64(dev_bucket, BUCKET_NOCOW_LOCKS_BITS);
 
 	return t->l + (h & (BUCKET_NOCOW_LOCKS - 1));
 }
 
-static inline bool bch2_bucket_nocow_is_locked(struct bucket_nocow_lock_table *t,
-					       struct bpos bucket)
-{
-	two_state_lock_t *l = bucket_nocow_lock(t, bucket);
-
-	return atomic_long_read(&l->v) != 0;
-}
-
-static inline void bch2_bucket_nocow_unlock(struct bucket_nocow_lock_table *t,
-					    struct bpos bucket, int flags)
-{
-	two_state_lock_t *l = bucket_nocow_lock(t, bucket);
-
-	bch2_two_state_unlock(l, flags & BUCKET_NOCOW_LOCK_UPDATE);
-}
+#define BUCKET_NOCOW_LOCK_UPDATE	(1 << 0)
 
-void __bch2_bucket_nocow_lock(struct bucket_nocow_lock_table *, two_state_lock_t *, int);
+bool bch2_bucket_nocow_is_locked(struct bucket_nocow_lock_table *, struct bpos);
+void bch2_bucket_nocow_unlock(struct bucket_nocow_lock_table *, struct bpos, int);
+void __bch2_bucket_nocow_lock(struct bucket_nocow_lock_table *,
+			      struct nocow_lock_bucket *, u64, int);
 
 static inline void bch2_bucket_nocow_lock(struct bucket_nocow_lock_table *t,
 					  struct bpos bucket, int flags)
 {
-	two_state_lock_t *l = bucket_nocow_lock(t, bucket);
+	u64 dev_bucket = bucket_to_u64(bucket);
+	struct nocow_lock_bucket *l = bucket_nocow_lock(t, dev_bucket);
 
-	if (!bch2_two_state_trylock(l, flags & BUCKET_NOCOW_LOCK_UPDATE))
-		__bch2_bucket_nocow_lock(t, l, flags);
+	__bch2_bucket_nocow_lock(t, l, dev_bucket, flags);
 }
 
+void bch2_nocow_locks_to_text(struct printbuf *, struct bucket_nocow_lock_table *);
+
+int bch2_fs_nocow_locking_init(struct bch_fs *);
+
 #endif /* _BCACHEFS_NOCOW_LOCKING_H */
diff --git a/fs/bcachefs/nocow_locking_types.h b/fs/bcachefs/nocow_locking_types.h
new file mode 100644
index 000000000000..bd12bf677924
--- /dev/null
+++ b/fs/bcachefs/nocow_locking_types.h
@@ -0,0 +1,20 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BCACHEFS_NOCOW_LOCKING_TYPES_H
+#define _BCACHEFS_NOCOW_LOCKING_TYPES_H
+
+#define BUCKET_NOCOW_LOCKS_BITS		10
+#define BUCKET_NOCOW_LOCKS		(1U << BUCKET_NOCOW_LOCKS_BITS)
+
+struct nocow_lock_bucket {
+	struct closure_waitlist		wait;
+	spinlock_t			lock;
+	u64				b[4];
+	atomic_t			l[4];
+} __aligned(SMP_CACHE_BYTES);
+
+struct bucket_nocow_lock_table {
+	struct nocow_lock_bucket	l[BUCKET_NOCOW_LOCKS];
+};
+
+#endif /* _BCACHEFS_NOCOW_LOCKING_TYPES_H */
+
diff --git a/fs/bcachefs/super.c b/fs/bcachefs/super.c
index 7cac0567037d..95c16f70512f 100644
--- a/fs/bcachefs/super.c
+++ b/fs/bcachefs/super.c
@@ -37,6 +37,7 @@
 #include "move.h"
 #include "migrate.h"
 #include "movinggc.h"
+#include "nocow_locking.h"
 #include "quota.h"
 #include "rebalance.h"
 #include "recovery.h"
@@ -803,6 +804,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 	    bch2_fs_buckets_waiting_for_journal_init(c) ?:
 	    bch2_fs_subvolumes_init(c) ?:
 	    bch2_fs_io_init(c) ?:
+	    bch2_fs_nocow_locking_init(c) ?:
 	    bch2_fs_encryption_init(c) ?:
 	    bch2_fs_compress_init(c) ?:
 	    bch2_fs_ec_init(c) ?:
diff --git a/fs/bcachefs/sysfs.c b/fs/bcachefs/sysfs.c
index bad3eafd32d2..6e49cf98f60d 100644
--- a/fs/bcachefs/sysfs.c
+++ b/fs/bcachefs/sysfs.c
@@ -27,6 +27,7 @@
 #include "journal.h"
 #include "keylist.h"
 #include "move.h"
+#include "nocow_locking.h"
 #include "opts.h"
 #include "rebalance.h"
 #include "replicas.h"
@@ -446,22 +447,9 @@ SHOW(bch2_fs)
 	if (attr == &sysfs_data_jobs)
 		data_progress_to_text(out, c);
 
-	if (attr == &sysfs_nocow_lock_table) {
-		int i, count = 1;
-		long last, curr = 0;
-
-		last = atomic_long_read(&c->nocow_locks.l[0].v);
-		for (i = 1; i < BUCKET_NOCOW_LOCKS; i++) {
-			curr = atomic_long_read(&c->nocow_locks.l[i].v);
-			if (last != curr) {
-				prt_printf(out, "%li: %d\n", last, count);
-				count = 1;
-				last = curr;
-			} else
-				count++;
-		}
-		prt_printf(out, "%li: %d\n", last, count);
-}
+	if (attr == &sysfs_nocow_lock_table)
+		bch2_nocow_locks_to_text(out, &c->nocow_locks);
+
 	return 0;
 }