bcachefs: Stripes now properly subject to gc

gc now verifies the contents of the stripes radix tree, important for
persistent alloc info
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/bcachefs.h

@@ -225,6 +225,8 @@
 	printk(KERN_WARNING bch2_fmt(c, fmt), ##__VA_ARGS__)
 #define bch_err(c, fmt, ...) \
 	printk(KERN_ERR bch2_fmt(c, fmt), ##__VA_ARGS__)
+#define bch_err_ratelimited(c, fmt, ...) \
+	printk_ratelimited(KERN_ERR bch2_fmt(c, fmt), ##__VA_ARGS__)
 
 #define bch_verbose(c, fmt, ...)					\
 do {									\
@@ -334,6 +336,7 @@ enum bch_time_stats {
 struct btree;
 
 enum gc_phase {
+	GC_PHASE_NOT_RUNNING,
 	GC_PHASE_START,
 	GC_PHASE_SB,
 
@@ -687,16 +690,17 @@ struct bch_fs {
 	/* REBALANCE */
 	struct bch_fs_rebalance	rebalance;
 
-	/* ERASURE CODING */
-	struct list_head	ec_new_stripe_list;
-	struct mutex		ec_new_stripe_lock;
-
-	GENRADIX(struct ec_stripe) ec_stripes;
-	struct mutex		ec_stripes_lock;
+	/* STRIPES: */
+	GENRADIX(struct stripe) stripes[2];
+	struct mutex		ec_stripe_create_lock;
 
 	ec_stripes_heap		ec_stripes_heap;
 	spinlock_t		ec_stripes_heap_lock;
 
+	/* ERASURE CODING */
+	struct list_head	ec_new_stripe_list;
+	struct mutex		ec_new_stripe_lock;
+
 	struct bio_set		ec_bioset;
 
 	struct work_struct	ec_stripe_delete_work;

fs/bcachefs/btree_gc.c

@@ -332,9 +332,6 @@ static int bch2_gc_btree(struct bch_fs *c, enum btree_id btree_id,
 
 	gc_pos_set(c, gc_pos_btree(btree_id, POS_MIN, 0));
 
-	if (!c->btree_roots[btree_id].b)
-		return 0;
-
 	/*
 	 * if expensive_debug_checks is on, run range_checks on all leaf nodes:
 	 *
@@ -582,6 +579,8 @@ static void bch2_gc_free(struct bch_fs *c)
 	struct bch_dev *ca;
 	unsigned i;
 
+	genradix_free(&c->stripes[1]);
+
 	for_each_member_device(ca, c, i) {
 		kvpfree(rcu_dereference_protected(ca->buckets[1], 1),
 			sizeof(struct bucket_array) +
@@ -602,6 +601,25 @@ static void bch2_gc_done_nocheck(struct bch_fs *c)
 	unsigned i;
 	int cpu;
 
+	{
+		struct genradix_iter dst_iter = genradix_iter_init(&c->stripes[0], 0);
+		struct genradix_iter src_iter = genradix_iter_init(&c->stripes[1], 0);
+		struct stripe *dst, *src;
+
+		c->ec_stripes_heap.used = 0;
+
+		while ((dst = genradix_iter_peek(&dst_iter, &c->stripes[0])) &&
+		       (src = genradix_iter_peek(&src_iter, &c->stripes[1]))) {
+			*dst = *src;
+
+			if (dst->alive)
+				bch2_stripes_heap_insert(c, dst, dst_iter.pos);
+
+			genradix_iter_advance(&dst_iter, &c->stripes[0]);
+			genradix_iter_advance(&src_iter, &c->stripes[1]);
+		}
+	}
+
 	for_each_member_device(ca, c, i) {
 		struct bucket_array *src = __bucket_array(ca, 1);
 
@@ -649,13 +667,21 @@ static void bch2_gc_done(struct bch_fs *c, bool initial)
 
 #define copy_field(_f, _msg, ...)					\
 	if (dst._f != src._f) {						\
-		pr_info(_msg ": got %llu, should be %llu, fixing"	\
+		bch_err(c, _msg ": got %llu, should be %llu, fixing"\
 			, ##__VA_ARGS__, dst._f, src._f);		\
 		dst._f = src._f;					\
 	}
+#define copy_stripe_field(_f, _msg, ...)				\
+	if (dst->_f != src->_f) {					\
+		bch_err_ratelimited(c, "stripe %zu has wrong "_msg	\
+			": got %u, should be %u, fixing",		\
+			dst_iter.pos, ##__VA_ARGS__,			\
+			dst->_f, src->_f);				\
+		dst->_f = src->_f;					\
+	}
 #define copy_bucket_field(_f)						\
 	if (dst->b[b].mark._f != src->b[b].mark._f) {			\
-		pr_info("dev %u bucket %zu has wrong " #_f		\
+		bch_err_ratelimited(c, "dev %u bucket %zu has wrong " #_f\
 			": got %u, should be %u, fixing",		\
 			i, b, dst->b[b].mark._f, src->b[b].mark._f);	\
 		dst->b[b]._mark._f = src->b[b].mark._f;			\
@@ -672,6 +698,36 @@ static void bch2_gc_done(struct bch_fs *c, bool initial)
 		goto out;
 	}
 
+	{
+		struct genradix_iter dst_iter = genradix_iter_init(&c->stripes[0], 0);
+		struct genradix_iter src_iter = genradix_iter_init(&c->stripes[1], 0);
+		struct stripe *dst, *src;
+		unsigned i;
+
+		c->ec_stripes_heap.used = 0;
+
+		while ((dst = genradix_iter_peek(&dst_iter, &c->stripes[0])) &&
+		       (src = genradix_iter_peek(&src_iter, &c->stripes[1]))) {
+			copy_stripe_field(alive,	"alive");
+			copy_stripe_field(sectors,	"sectors");
+			copy_stripe_field(algorithm,	"algorithm");
+			copy_stripe_field(nr_blocks,	"nr_blocks");
+			copy_stripe_field(nr_redundant,	"nr_redundant");
+			copy_stripe_field(blocks_nonempty.counter,
+					  "blocks_nonempty");
+
+			for (i = 0; i < ARRAY_SIZE(dst->block_sectors); i++)
+				copy_stripe_field(block_sectors[i].counter,
+						  "block_sectors[%u]", i);
+
+			if (dst->alive)
+				bch2_stripes_heap_insert(c, dst, dst_iter.pos);
+
+			genradix_iter_advance(&dst_iter, &c->stripes[0]);
+			genradix_iter_advance(&src_iter, &c->stripes[1]);
+		}
+	}
+
 	for_each_member_device(ca, c, i) {
 		struct bucket_array *dst = __bucket_array(ca, 0);
 		struct bucket_array *src = __bucket_array(ca, 1);
@@ -756,10 +812,11 @@ static void bch2_gc_done(struct bch_fs *c, bool initial)
 out:
 	percpu_up_write(&c->usage_lock);
 
-#undef copy_field
 #undef copy_fs_field
 #undef copy_dev_field
 #undef copy_bucket_field
+#undef copy_stripe_field
+#undef copy_field
 }
 
 static int bch2_gc_start(struct bch_fs *c)
@@ -767,6 +824,12 @@ static int bch2_gc_start(struct bch_fs *c)
 	struct bch_dev *ca;
 	unsigned i;
 
+	/*
+	 * indicate to stripe code that we need to allocate for the gc stripes
+	 * radix tree, too
+	 */
+	gc_pos_set(c, gc_phase(GC_PHASE_START));
+
 	BUG_ON(c->usage[1]);
 
 	c->usage[1] = alloc_percpu(struct bch_fs_usage);
@@ -808,7 +871,7 @@ static int bch2_gc_start(struct bch_fs *c)
 
 	percpu_up_write(&c->usage_lock);
 
-	return 0;
+	return bch2_ec_mem_alloc(c, true);
 }
 
 /**
@@ -873,7 +936,7 @@ int bch2_gc(struct bch_fs *c, struct list_head *journal, bool initial)
 		bch2_gc_done(c, initial);
 
 	/* Indicates that gc is no longer in progress: */
-	__gc_pos_set(c, gc_phase(GC_PHASE_START));
+	__gc_pos_set(c, gc_phase(GC_PHASE_NOT_RUNNING));
 
 	bch2_gc_free(c);
 	up_write(&c->gc_lock);

fs/bcachefs/buckets.c

@@ -303,7 +303,7 @@ static inline int is_fragmented_bucket(struct bucket_mark m,
 static inline enum bch_data_type bucket_type(struct bucket_mark m)
 {
 	return m.cached_sectors && !m.dirty_sectors
-		?  BCH_DATA_CACHED
+		? BCH_DATA_CACHED
 		: m.data_type;
 }
 
@@ -375,14 +375,14 @@ static void bch2_dev_usage_update(struct bch_fs *c, struct bch_dev *ca,
 	preempt_disable();
 	dev_usage = this_cpu_ptr(ca->usage[gc]);
 
-	if (bucket_type(old) != bucket_type(new)) {
-		if (bucket_type(old)) {
-			fs_usage->buckets[bucket_type(old)] -= ca->mi.bucket_size;
-			dev_usage->buckets[bucket_type(old)]--;
-		} else {
-			fs_usage->buckets[bucket_type(new)] += ca->mi.bucket_size;
-			dev_usage->buckets[bucket_type(new)]++;
-		}
+	if (bucket_type(old)) {
+		fs_usage->buckets[bucket_type(old)] -= ca->mi.bucket_size;
+		dev_usage->buckets[bucket_type(old)]--;
+	}
+
+	if (bucket_type(new)) {
+		fs_usage->buckets[bucket_type(new)] += ca->mi.bucket_size;
+		dev_usage->buckets[bucket_type(new)]++;
 	}
 
 	dev_usage->buckets_alloc +=
@@ -406,11 +406,11 @@ static void bch2_dev_usage_update(struct bch_fs *c, struct bch_dev *ca,
 	bch2_dev_stats_verify(ca);
 }
 
-#define bucket_data_cmpxchg(c, ca, stats, g, new, expr)		\
+#define bucket_data_cmpxchg(c, ca, fs_usage, g, new, expr)		\
 ({								\
 	struct bucket_mark _old = bucket_cmpxchg(g, new, expr);	\
 								\
-	bch2_dev_usage_update(c, ca, stats, _old, new, gc);	\
+	bch2_dev_usage_update(c, ca, fs_usage, _old, new, gc);	\
 	_old;							\
 })
 
@@ -638,23 +638,25 @@ static void bch2_mark_pointer(struct bch_fs *c,
 	BUG_ON(!gc && bucket_became_unavailable(old, new));
 }
 
-static void bch2_mark_stripe_ptr(struct bch_fs *c,
-				 struct bch_extent_stripe_ptr p,
-				 s64 sectors, unsigned flags,
-				 s64 *adjusted_disk_sectors,
-				 unsigned *redundancy)
+static int bch2_mark_stripe_ptr(struct bch_fs *c,
+				struct bch_extent_stripe_ptr p,
+				s64 sectors, unsigned flags,
+				s64 *adjusted_disk_sectors,
+				unsigned *redundancy,
+				bool gc)
 {
-	struct ec_stripe *m;
+	struct stripe *m;
 	unsigned old, new, nr_data;
 	int blocks_nonempty_delta;
 	s64 parity_sectors;
 
-	m = genradix_ptr(&c->ec_stripes, p.idx);
-	if (WARN_ON(!m))
-		return;
+	m = genradix_ptr(&c->stripes[gc], p.idx);
 
-	if (WARN_ON(!m->alive))
-		return;
+	if (!m || !m->alive) {
+		bch_err_ratelimited(c, "pointer to nonexistent stripe %llu",
+				    (u64) p.idx);
+		return -1;
+	}
 
 	nr_data = m->nr_blocks - m->nr_redundant;
 
@@ -672,20 +674,23 @@ static void bch2_mark_stripe_ptr(struct bch_fs *c,
 
 	blocks_nonempty_delta = (int) !!new - (int) !!old;
 	if (!blocks_nonempty_delta)
-		return;
+		return 0;
 
 	atomic_add(blocks_nonempty_delta, &m->blocks_nonempty);
 
 	BUG_ON(atomic_read(&m->blocks_nonempty) < 0);
 
-	bch2_stripes_heap_update(c, m, p.idx);
+	if (!gc)
+		bch2_stripes_heap_update(c, m, p.idx);
+
+	return 0;
 }
 
-static void bch2_mark_extent(struct bch_fs *c, struct bkey_s_c k,
-			     s64 sectors, enum bch_data_type data_type,
-			     struct bch_fs_usage *stats,
-			     u64 journal_seq, unsigned flags,
-			     bool gc)
+static int bch2_mark_extent(struct bch_fs *c, struct bkey_s_c k,
+			    s64 sectors, enum bch_data_type data_type,
+			    struct bch_fs_usage *stats,
+			    u64 journal_seq, unsigned flags,
+			    bool gc)
 {
 	BUG_ON(!sectors);
 
@@ -701,6 +706,7 @@ static void bch2_mark_extent(struct bch_fs *c, struct bkey_s_c k,
 		unsigned replicas	= 0;
 		unsigned ec_redundancy	= 0;
 		unsigned i;
+		int ret;
 
 		extent_for_each_ptr_decode(e, p, entry) {
 			s64 disk_sectors = ptr_disk_sectors(e, p, sectors);
@@ -710,11 +716,14 @@ static void bch2_mark_extent(struct bch_fs *c, struct bkey_s_c k,
 					  stats, journal_seq, flags, gc);
 
 			if (!p.ptr.cached)
-				for (i = 0; i < p.ec_nr; i++)
-					bch2_mark_stripe_ptr(c, p.ec[i],
+				for (i = 0; i < p.ec_nr; i++) {
+					ret = bch2_mark_stripe_ptr(c, p.ec[i],
 							disk_sectors, flags,
 							&adjusted_disk_sectors,
-							&ec_redundancy);
+							&ec_redundancy, gc);
+					if (ret)
+						return ret;
+				}
 			if (!p.ptr.cached)
 				replicas++;
 
@@ -747,6 +756,8 @@ static void bch2_mark_extent(struct bch_fs *c, struct bkey_s_c k,
 		break;
 	}
 	}
+
+	return 0;
 }
 
 static void bucket_set_stripe(struct bch_fs *c,
@@ -767,7 +778,7 @@ static void bucket_set_stripe(struct bch_fs *c,
 
 		BUG_ON(ptr_stale(ca, ptr));
 
-		old = bucket_cmpxchg(g, new, ({
+		old = bucket_data_cmpxchg(c, ca, fs_usage, g, new, ({
 			new.stripe			= enabled;
 			if (journal_seq) {
 				new.journal_seq_valid	= 1;
@@ -776,26 +787,33 @@ static void bucket_set_stripe(struct bch_fs *c,
 		}));
 
 		BUG_ON(old.stripe == enabled);
-
-		bch2_dev_usage_update(c, ca, fs_usage, old, new, gc);
 	}
 }
 
-static void bch2_mark_stripe(struct bch_fs *c, struct bkey_s_c k,
-			     bool inserting,
-			     struct bch_fs_usage *fs_usage,
-			     u64 journal_seq, unsigned flags,
-			     bool gc)
+static int bch2_mark_stripe(struct bch_fs *c, struct bkey_s_c k,
+			    bool inserting,
+			    struct bch_fs_usage *fs_usage,
+			    u64 journal_seq, unsigned flags,
+			    bool gc)
 {
 	switch (k.k->type) {
 	case BCH_STRIPE: {
 		struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
 		size_t idx = s.k->p.offset;
-		struct ec_stripe *m = genradix_ptr(&c->ec_stripes, idx);
+		struct stripe *m = genradix_ptr(&c->stripes[gc], idx);
 		unsigned i;
 
-		BUG_ON(!m);
-		BUG_ON(m->alive == inserting);
+		if (!m || (!inserting && !m->alive)) {
+			bch_err_ratelimited(c, "error marking nonexistent stripe %zu",
+					    idx);
+			return -1;
+		}
+
+		if (inserting && m->alive) {
+			bch_err_ratelimited(c, "error marking stripe %zu: already exists",
+					    idx);
+			return -1;
+		}
 
 		BUG_ON(atomic_read(&m->blocks_nonempty));
 
@@ -809,70 +827,88 @@ static void bch2_mark_stripe(struct bch_fs *c, struct bkey_s_c k,
 			m->nr_redundant	= s.v->nr_redundant;
 		}
 
-		if (inserting)
-			bch2_stripes_heap_insert(c, m, idx);
-		else
-			bch2_stripes_heap_del(c, m, idx);
+		if (!gc) {
+			if (inserting)
+				bch2_stripes_heap_insert(c, m, idx);
+			else
+				bch2_stripes_heap_del(c, m, idx);
+		} else {
+			m->alive = inserting;
+		}
 
 		bucket_set_stripe(c, s.v, inserting, fs_usage, 0, gc);
 		break;
 	}
 	}
+
+	return 0;
 }
 
-static void __bch2_mark_key(struct bch_fs *c,
-			    enum bkey_type type, struct bkey_s_c k,
-			    bool inserting, s64 sectors,
-			    struct bch_fs_usage *stats,
-			    u64 journal_seq, unsigned flags,
-			    bool gc)
+static int __bch2_mark_key(struct bch_fs *c,
+			   enum bkey_type type, struct bkey_s_c k,
+			   bool inserting, s64 sectors,
+			   struct bch_fs_usage *stats,
+			   u64 journal_seq, unsigned flags,
+			   bool gc)
 {
+	int ret = 0;
+
 	switch (type) {
 	case BKEY_TYPE_BTREE:
-		bch2_mark_extent(c, k, inserting
-				 ?  c->opts.btree_node_size
-				 : -c->opts.btree_node_size,
-				 BCH_DATA_BTREE,
-				 stats, journal_seq, flags, gc);
+		ret = bch2_mark_extent(c, k, inserting
+				       ?  c->opts.btree_node_size
+				       : -c->opts.btree_node_size,
+				       BCH_DATA_BTREE,
+				       stats, journal_seq, flags, gc);
 		break;
 	case BKEY_TYPE_EXTENTS:
-		bch2_mark_extent(c, k, sectors, BCH_DATA_USER,
-				 stats, journal_seq, flags, gc);
+		ret = bch2_mark_extent(c, k, sectors, BCH_DATA_USER,
+				       stats, journal_seq, flags, gc);
 		break;
 	case BKEY_TYPE_EC:
-		bch2_mark_stripe(c, k, inserting,
-				 stats, journal_seq, flags, gc);
+		ret = bch2_mark_stripe(c, k, inserting,
+				       stats, journal_seq, flags, gc);
 		break;
 	default:
 		break;
 	}
+
+	return ret;
 }
 
-void bch2_mark_key(struct bch_fs *c,
-		   enum bkey_type type, struct bkey_s_c k,
-		   bool inserting, s64 sectors,
-		   struct gc_pos pos,
-		   struct bch_fs_usage *stats,
-		   u64 journal_seq, unsigned flags)
+int bch2_mark_key(struct bch_fs *c,
+		  enum bkey_type type, struct bkey_s_c k,
+		  bool inserting, s64 sectors,
+		  struct gc_pos pos,
+		  struct bch_fs_usage *stats,
+		  u64 journal_seq, unsigned flags)
 {
+	int ret = 0;
+
 	percpu_down_read(&c->usage_lock);
 
 	if (!(flags & BCH_BUCKET_MARK_GC)) {
 		if (!stats)
 			stats = this_cpu_ptr(c->usage[0]);
 
-		__bch2_mark_key(c, type, k, inserting, sectors,
-				stats, journal_seq, flags, false);
+		ret = __bch2_mark_key(c, type, k, inserting, sectors,
+				      stats, journal_seq, flags, false);
+		if (ret)
+			goto out;
 	}
 
 	if ((flags & BCH_BUCKET_MARK_GC) ||
 	    gc_visited(c, pos)) {
-		__bch2_mark_key(c, type, k, inserting, sectors,
-				this_cpu_ptr(c->usage[1]),
-				journal_seq, flags, true);
+		ret = __bch2_mark_key(c, type, k, inserting, sectors,
+				      this_cpu_ptr(c->usage[1]),
+				      journal_seq, flags, true);
+		if (ret)
+			goto out;
 	}
-
+out:
 	percpu_up_read(&c->usage_lock);
+
+	return ret;
 }
 
 void bch2_mark_update(struct btree_insert *trans,

fs/bcachefs/buckets.h

@@ -220,9 +220,9 @@ void bch2_mark_metadata_bucket(struct bch_fs *, struct bch_dev *,
 #define BCH_BUCKET_MARK_NOATOMIC		(1 << 0)
 #define BCH_BUCKET_MARK_GC			(1 << 1)
 
-void bch2_mark_key(struct bch_fs *, enum bkey_type, struct bkey_s_c,
-		   bool, s64, struct gc_pos,
-		   struct bch_fs_usage *, u64, unsigned);
+int bch2_mark_key(struct bch_fs *, enum bkey_type, struct bkey_s_c,
+		  bool, s64, struct gc_pos,
+		  struct bch_fs_usage *, u64, unsigned);
 void bch2_mark_update(struct btree_insert *, struct btree_insert_entry *);
 
 void __bch2_disk_reservation_put(struct bch_fs *, struct disk_reservation *);

fs/bcachefs/ec.c

@@ -530,7 +530,7 @@ int bch2_ec_read_extent(struct bch_fs *c, struct bch_read_bio *rbio)
 	return ret;
 }
 
-/* ec_stripe bucket accounting: */
+/* stripe bucket accounting: */
 
 static int __ec_stripe_mem_alloc(struct bch_fs *c, size_t idx, gfp_t gfp)
 {
@@ -551,7 +551,11 @@ static int __ec_stripe_mem_alloc(struct bch_fs *c, size_t idx, gfp_t gfp)
 		free_heap(&n);
 	}
 
-	if (!genradix_ptr_alloc(&c->ec_stripes, idx, gfp))
+	if (!genradix_ptr_alloc(&c->stripes[0], idx, gfp))
+		return -ENOMEM;
+
+	if (c->gc_pos.phase != GC_PHASE_NOT_RUNNING &&
+	    !genradix_ptr_alloc(&c->stripes[1], idx, gfp))
 		return -ENOMEM;
 
 	return 0;
@@ -592,27 +596,26 @@ static inline void ec_stripes_heap_set_backpointer(ec_stripes_heap *h,
 {
 	struct bch_fs *c = container_of(h, struct bch_fs, ec_stripes_heap);
 
-	genradix_ptr(&c->ec_stripes, h->data[i].idx)->heap_idx = i;
+	genradix_ptr(&c->stripes[0], h->data[i].idx)->heap_idx = i;
 }
 
 static void heap_verify_backpointer(struct bch_fs *c, size_t idx)
 {
 	ec_stripes_heap *h = &c->ec_stripes_heap;
-	struct ec_stripe *m = genradix_ptr(&c->ec_stripes, idx);
+	struct stripe *m = genradix_ptr(&c->stripes[0], idx);
 
 	BUG_ON(!m->alive);
 	BUG_ON(m->heap_idx >= h->used);
 	BUG_ON(h->data[m->heap_idx].idx != idx);
 }
 
-static inline unsigned stripe_entry_blocks(struct ec_stripe *m)
+static inline unsigned stripe_entry_blocks(struct stripe *m)
 {
-	return atomic_read(&m->pin)
-		? UINT_MAX : atomic_read(&m->blocks_nonempty);
+	return atomic_read(&m->blocks_nonempty);
 }
 
 void bch2_stripes_heap_update(struct bch_fs *c,
-			      struct ec_stripe *m, size_t idx)
+			      struct stripe *m, size_t idx)
 {
 	ec_stripes_heap *h = &c->ec_stripes_heap;
 	bool queue_delete;
@@ -646,7 +649,7 @@ void bch2_stripes_heap_update(struct bch_fs *c,
 }
 
 void bch2_stripes_heap_del(struct bch_fs *c,
-			   struct ec_stripe *m, size_t idx)
+			   struct stripe *m, size_t idx)
 {
 	spin_lock(&c->ec_stripes_heap_lock);
 	heap_verify_backpointer(c, idx);
@@ -659,7 +662,7 @@ void bch2_stripes_heap_del(struct bch_fs *c,
 }
 
 void bch2_stripes_heap_insert(struct bch_fs *c,
-			      struct ec_stripe *m, size_t idx)
+			      struct stripe *m, size_t idx)
 {
 	spin_lock(&c->ec_stripes_heap_lock);
 
@@ -678,7 +681,9 @@ void bch2_stripes_heap_insert(struct bch_fs *c,
 	spin_unlock(&c->ec_stripes_heap_lock);
 }
 
-static void ec_stripe_delete(struct bch_fs *c, unsigned idx)
+/* stripe deletion */
+
+static void ec_stripe_delete(struct bch_fs *c, size_t idx)
 {
 	struct btree_iter iter;
 	struct bch_stripe *v = NULL;
@@ -717,6 +722,7 @@ static void ec_stripe_delete_work(struct work_struct *work)
 	ssize_t idx;
 
 	down_read(&c->gc_lock);
+	mutex_lock(&c->ec_stripe_create_lock);
 
 	while (1) {
 		spin_lock(&c->ec_stripes_heap_lock);
@@ -729,13 +735,15 @@ static void ec_stripe_delete_work(struct work_struct *work)
 		ec_stripe_delete(c, idx);
 	}
 
+	mutex_unlock(&c->ec_stripe_create_lock);
 	up_read(&c->gc_lock);
 }
 
+/* stripe creation: */
+
 static int ec_stripe_bkey_insert(struct bch_fs *c,
 				 struct bkey_i_stripe *stripe)
 {
-	struct ec_stripe *m;
 	struct btree_iter iter;
 	struct bkey_s_c k;
 	int ret;
@@ -755,18 +763,13 @@ static int ec_stripe_bkey_insert(struct bch_fs *c,
 
 	return bch2_btree_iter_unlock(&iter) ?: -ENOSPC;
 found_slot:
-	mutex_lock(&c->ec_stripes_lock);
 	ret = ec_stripe_mem_alloc(c, &iter);
-	mutex_unlock(&c->ec_stripes_lock);
 
 	if (ret == -EINTR)
 		goto retry;
 	if (ret)
 		return ret;
 
-	m = genradix_ptr(&c->ec_stripes, iter.pos.offset);
-	atomic_inc(&m->pin);
-
 	stripe->k.p = iter.pos;
 
 	ret = bch2_btree_insert_at(c, NULL, NULL,
@@ -775,14 +778,9 @@ static int ec_stripe_bkey_insert(struct bch_fs *c,
 				   BTREE_INSERT_ENTRY(&iter, &stripe->k_i));
 	bch2_btree_iter_unlock(&iter);
 
-	if (ret)
-		atomic_dec(&m->pin);
-
 	return ret;
 }
 
-/* stripe creation: */
-
 static void extent_stripe_ptr_add(struct bkey_s_extent e,
 				  struct ec_stripe_buf *s,
 				  struct bch_extent_ptr *ptr,
@@ -858,7 +856,6 @@ static int ec_stripe_update_ptrs(struct bch_fs *c,
  */
 static void ec_stripe_create(struct ec_stripe_new *s)
 {
-	struct ec_stripe *ec_stripe;
 	struct bch_fs *c = s->c;
 	struct open_bucket *ob;
 	struct bkey_i *k;
@@ -898,10 +895,12 @@ static void ec_stripe_create(struct ec_stripe_new *s)
 			goto err_put_writes;
 		}
 
+	mutex_lock(&c->ec_stripe_create_lock);
+
 	ret = ec_stripe_bkey_insert(c, &s->stripe.key);
 	if (ret) {
 		bch_err(c, "error creating stripe: error creating stripe key");
-		goto err_put_writes;
+		goto err_unlock;
 	}
 
 	for_each_keylist_key(&s->keys, k) {
@@ -910,12 +909,8 @@ static void ec_stripe_create(struct ec_stripe_new *s)
 			break;
 	}
 
-	ec_stripe = genradix_ptr(&c->ec_stripes, s->stripe.key.k.p.offset);
-
-	atomic_dec(&ec_stripe->pin);
-	bch2_stripes_heap_update(c, ec_stripe,
-				 s->stripe.key.k.p.offset);
-
+err_unlock:
+	mutex_unlock(&c->ec_stripe_create_lock);
 err_put_writes:
 	percpu_ref_put(&c->writes);
 err:
@@ -1222,7 +1217,7 @@ void bch2_ec_stop_dev(struct bch_fs *c, struct bch_dev *ca)
 	mutex_unlock(&c->ec_new_stripe_lock);
 }
 
-int bch2_fs_ec_start(struct bch_fs *c)
+int bch2_ec_mem_alloc(struct bch_fs *c, bool gc)
 {
 	struct btree_iter iter;
 	struct bkey_s_c k;
@@ -1238,19 +1233,25 @@ int bch2_fs_ec_start(struct bch_fs *c)
 	if (ret)
 		return ret;
 
-	if (!init_heap(&c->ec_stripes_heap, roundup_pow_of_two(idx),
+	if (!gc &&
+	    !init_heap(&c->ec_stripes_heap, roundup_pow_of_two(idx),
 		       GFP_KERNEL))
 		return -ENOMEM;
 #if 0
-	ret = genradix_prealloc(&c->ec_stripes, idx, GFP_KERNEL);
+	ret = genradix_prealloc(&c->stripes[gc], idx, GFP_KERNEL);
 #else
 	for (i = 0; i < idx; i++)
-		if (!genradix_ptr_alloc(&c->ec_stripes, i, GFP_KERNEL))
+		if (!genradix_ptr_alloc(&c->stripes[gc], i, GFP_KERNEL))
 			return -ENOMEM;
 #endif
 	return 0;
 }
 
+int bch2_fs_ec_start(struct bch_fs *c)
+{
+	return bch2_ec_mem_alloc(c, false);
+}
+
 void bch2_fs_ec_exit(struct bch_fs *c)
 {
 	struct ec_stripe_head *h;
@@ -1271,7 +1272,7 @@ void bch2_fs_ec_exit(struct bch_fs *c)
 	}
 
 	free_heap(&c->ec_stripes_heap);
-	genradix_free(&c->ec_stripes);
+	genradix_free(&c->stripes[0]);
 	bioset_exit(&c->ec_bioset);
 }
 

fs/bcachefs/ec.h

@@ -93,14 +93,16 @@ void bch2_ec_stripe_head_put(struct ec_stripe_head *);
 struct ec_stripe_head *bch2_ec_stripe_head_get(struct bch_fs *, unsigned,
 					       unsigned, unsigned);
 
-void bch2_stripes_heap_update(struct bch_fs *, struct ec_stripe *, size_t);
-void bch2_stripes_heap_del(struct bch_fs *, struct ec_stripe *, size_t);
-void bch2_stripes_heap_insert(struct bch_fs *, struct ec_stripe *, size_t);
+void bch2_stripes_heap_update(struct bch_fs *, struct stripe *, size_t);
+void bch2_stripes_heap_del(struct bch_fs *, struct stripe *, size_t);
+void bch2_stripes_heap_insert(struct bch_fs *, struct stripe *, size_t);
 
 void bch2_ec_stop_dev(struct bch_fs *, struct bch_dev *);
 
 void bch2_ec_flush_new_stripes(struct bch_fs *);
 
+int bch2_ec_mem_alloc(struct bch_fs *, bool);
+
 int bch2_fs_ec_start(struct bch_fs *);
 
 void bch2_fs_ec_exit(struct bch_fs *);

fs/bcachefs/ec_types.h

@@ -6,7 +6,7 @@
 
 #define EC_STRIPE_MAX	16
 
-struct ec_stripe {
+struct stripe {
 	size_t			heap_idx;
 
 	u16			sectors;
@@ -16,7 +16,6 @@ struct ec_stripe {
 	u8			nr_redundant;
 
 	u8			alive;
-	atomic_t		pin;
 	atomic_t		blocks_nonempty;
 	atomic_t		block_sectors[EC_STRIPE_MAX];
 };

fs/bcachefs/extents.c

@@ -208,8 +208,8 @@ static unsigned bch2_extent_ptr_durability(struct bch_fs *c,
 		durability = max_t(unsigned, durability, ca->mi.durability);
 
 	for (i = 0; i < p.ec_nr; i++) {
-		struct ec_stripe *s =
-			genradix_ptr(&c->ec_stripes, p.idx);
+		struct stripe *s =
+			genradix_ptr(&c->stripes[0], p.idx);
 
 		if (WARN_ON(!s))
 			continue;

fs/bcachefs/quota.c

@@ -700,22 +700,19 @@ static int bch2_get_next_quota(struct super_block *sb, struct kqid *kqid,
 	struct bch_fs *c		= sb->s_fs_info;
 	struct bch_memquota_type *q	= &c->quotas[kqid->type];
 	qid_t qid			= from_kqid(&init_user_ns, *kqid);
-	struct genradix_iter iter	= genradix_iter_init(&q->table, qid);
+	struct genradix_iter iter;
 	struct bch_memquota *mq;
 	int ret = 0;
 
 	mutex_lock(&q->lock);
 
-	while ((mq = genradix_iter_peek(&iter, &q->table))) {
+	genradix_for_each_from(&q->table, iter, mq, qid)
 		if (memcmp(mq, page_address(ZERO_PAGE(0)), sizeof(*mq))) {
 			__bch2_quota_get(qdq, mq);
 			*kqid = make_kqid(current_user_ns(), kqid->type, iter.pos);
 			goto found;
 		}
 
-		genradix_iter_advance(&iter, &q->table);
-	}
-
 	ret = -ENOENT;
 found:
 	mutex_unlock(&q->lock);

fs/bcachefs/recovery.c

@@ -305,6 +305,9 @@ int bch2_fs_initialize(struct bch_fs *c)
 
 	set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 
+	for (i = 0; i < BTREE_ID_NR; i++)
+		bch2_btree_root_alloc(c, i);
+
 	ret = bch2_initial_gc(c, &journal);
 	if (ret)
 		goto err;
@@ -316,9 +319,6 @@ int bch2_fs_initialize(struct bch_fs *c)
 			goto err;
 		}
 
-	for (i = 0; i < BTREE_ID_NR; i++)
-		bch2_btree_root_alloc(c, i);
-
 	/*
 	 * journal_res_get() will crash if called before this has
 	 * set up the journal.pin FIFO and journal.cur pointer:

fs/bcachefs/super.c

@@ -548,7 +548,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 
 	INIT_LIST_HEAD(&c->ec_new_stripe_list);
 	mutex_init(&c->ec_new_stripe_lock);
-	mutex_init(&c->ec_stripes_lock);
+	mutex_init(&c->ec_stripe_create_lock);
 	spin_lock_init(&c->ec_stripes_heap_lock);
 
 	seqcount_init(&c->gc_pos_lock);