X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Freplicas.c;h=cef2a0447b8601a191ac39852ae4a7e0d9fc317d;hb=9799b119c34d7be1ee96d143209cfe5fc543d92a;hp=d0076bd4fc4ea3ad8508a6193c19c07f462017a1;hpb=ddb58076ef4fe4572ab8537785fc67052f47bf5b;p=bcachefs-tools-debian

diff --git a/libbcachefs/replicas.c b/libbcachefs/replicas.c
index d0076bd..cef2a04 100644
--- a/libbcachefs/replicas.c
+++ b/libbcachefs/replicas.c
@@ -1,5 +1,7 @@
+// SPDX-License-Identifier: GPL-2.0
 
 #include "bcachefs.h"
+#include "buckets.h"
 #include "journal.h"
 #include "replicas.h"
 #include "super-io.h"
@@ -9,22 +11,22 @@ static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *,
 
 /* Replicas tracking - in memory: */
 
-static inline int u8_cmp(u8 l, u8 r)
+static void verify_replicas_entry(struct bch_replicas_entry *e)
 {
-	return (l > r) - (l < r);
-}
-
-static void verify_replicas_entry_sorted(struct bch_replicas_entry *e)
-{
-#ifdef CONFIG_BCACHES_DEBUG
+#ifdef CONFIG_BCACHEFS_DEBUG
 	unsigned i;
 
+	BUG_ON(e->data_type >= BCH_DATA_NR);
+	BUG_ON(!e->nr_devs);
+	BUG_ON(e->nr_required > 1 &&
+	       e->nr_required >= e->nr_devs);
+
 	for (i = 0; i + 1 < e->nr_devs; i++)
 		BUG_ON(e->devs[i] >= e->devs[i + 1]);
 #endif
 }
 
-static void replicas_entry_sort(struct bch_replicas_entry *e)
+void bch2_replicas_entry_sort(struct bch_replicas_entry *e)
 {
 	bubble_sort(e->devs, e->nr_devs, u8_cmp);
 }
@@ -34,30 +36,47 @@ static void bch2_cpu_replicas_sort(struct bch_replicas_cpu *r)
 	eytzinger0_sort(r->entries, r->nr, r->entry_size, memcmp, NULL);
 }
 
+static void bch2_replicas_entry_v0_to_text(struct printbuf *out,
+					   struct bch_replicas_entry_v0 *e)
+{
+	unsigned i;
+
+	if (e->data_type < BCH_DATA_NR)
+		prt_printf(out, "%s", bch2_data_types[e->data_type]);
+	else
+		prt_printf(out, "(invalid data type %u)", e->data_type);
+
+	prt_printf(out, ": %u [", e->nr_devs);
+	for (i = 0; i < e->nr_devs; i++)
+		prt_printf(out, i ? " %u" : "%u", e->devs[i]);
+	prt_printf(out, "]");
+}
+
 void bch2_replicas_entry_to_text(struct printbuf *out,
 				 struct bch_replicas_entry *e)
 {
 	unsigned i;
 
-	pr_buf(out, "%s: %u/%u [",
-	       bch2_data_types[e->data_type],
-	       e->nr_required,
-	       e->nr_devs);
+	if (e->data_type < BCH_DATA_NR)
+		prt_printf(out, "%s", bch2_data_types[e->data_type]);
+	else
+		prt_printf(out, "(invalid data type %u)", e->data_type);
 
+	prt_printf(out, ": %u/%u [", e->nr_required, e->nr_devs);
 	for (i = 0; i < e->nr_devs; i++)
-		pr_buf(out, i ? " %u" : "%u", e->devs[i]);
-	pr_buf(out, "]");
+		prt_printf(out, i ? " %u" : "%u", e->devs[i]);
+	prt_printf(out, "]");
 }
 
 void bch2_cpu_replicas_to_text(struct printbuf *out,
-			      struct bch_replicas_cpu *r)
+			       struct bch_replicas_cpu *r)
 {
 	struct bch_replicas_entry *e;
 	bool first = true;
 
 	for_each_cpu_replicas_entry(r, e) {
 		if (!first)
-			pr_buf(out, " ");
+			prt_printf(out, " ");
 		first = false;
 
 		bch2_replicas_entry_to_text(out, e);
@@ -77,12 +96,10 @@ static void extent_to_replicas(struct bkey_s_c k,
 		if (p.ptr.cached)
 			continue;
 
-		if (p.ec_nr) {
-			r->nr_devs = 0;
-			break;
-		}
-
-		r->devs[r->nr_devs++] = p.ptr.dev;
+		if (!p.has_ec)
+			r->devs[r->nr_devs++] = p.ptr.dev;
+		else
+			r->nr_required = 0;
 	}
 }
 
@@ -100,27 +117,29 @@ static void stripe_to_replicas(struct bkey_s_c k,
 		r->devs[r->nr_devs++] = ptr->dev;
 }
 
-static void bkey_to_replicas(struct bch_replicas_entry *e,
-			     struct bkey_s_c k)
+void bch2_bkey_to_replicas(struct bch_replicas_entry *e,
+			   struct bkey_s_c k)
 {
 	e->nr_devs = 0;
 
 	switch (k.k->type) {
 	case KEY_TYPE_btree_ptr:
-		e->data_type = BCH_DATA_BTREE;
+	case KEY_TYPE_btree_ptr_v2:
+		e->data_type = BCH_DATA_btree;
 		extent_to_replicas(k, e);
 		break;
 	case KEY_TYPE_extent:
-		e->data_type = BCH_DATA_USER;
+	case KEY_TYPE_reflink_v:
+		e->data_type = BCH_DATA_user;
 		extent_to_replicas(k, e);
 		break;
 	case KEY_TYPE_stripe:
-		e->data_type = BCH_DATA_USER;
+		e->data_type = BCH_DATA_parity;
 		stripe_to_replicas(k, e);
 		break;
 	}
 
-	replicas_entry_sort(e);
+	bch2_replicas_entry_sort(e);
 }
 
 void bch2_devlist_to_replicas(struct bch_replicas_entry *e,
@@ -130,7 +149,7 @@ void bch2_devlist_to_replicas(struct bch_replicas_entry *e,
 	unsigned i;
 
 	BUG_ON(!data_type ||
-	       data_type == BCH_DATA_SB ||
+	       data_type == BCH_DATA_sb ||
 	       data_type >= BCH_DATA_NR);
 
 	e->data_type	= data_type;
@@ -140,7 +159,7 @@ void bch2_devlist_to_replicas(struct bch_replicas_entry *e,
 	for (i = 0; i < devs.nr; i++)
 		e->devs[e->nr_devs++] = devs.devs[i];
 
-	replicas_entry_sort(e);
+	bch2_replicas_entry_sort(e);
 }
 
 static struct bch_replicas_cpu
@@ -155,9 +174,9 @@ cpu_replicas_add_entry(struct bch_replicas_cpu *old,
 	};
 
 	BUG_ON(!new_entry->data_type);
-	verify_replicas_entry_sorted(new_entry);
+	verify_replicas_entry(new_entry);
 
-	new.entries = kcalloc(new.nr, new.entry_size, GFP_NOIO);
+	new.entries = kcalloc(new.nr, new.entry_size, GFP_KERNEL);
 	if (!new.entries)
 		return new;
 
@@ -182,7 +201,7 @@ static inline int __replicas_entry_idx(struct bch_replicas_cpu *r,
 	if (unlikely(entry_size > r->entry_size))
 		return -1;
 
-	verify_replicas_entry_sorted(search);
+	verify_replicas_entry(search);
 
 #define entry_cmp(_l, _r, size)	memcmp(_l, _r, entry_size)
 	idx = eytzinger0_find(r->entries, r->nr, r->entry_size,
@@ -195,7 +214,7 @@ static inline int __replicas_entry_idx(struct bch_replicas_cpu *r,
 int bch2_replicas_entry_idx(struct bch_fs *c,
 			    struct bch_replicas_entry *search)
 {
-	replicas_entry_sort(search);
+	bch2_replicas_entry_sort(search);
 
 	return __replicas_entry_idx(&c->replicas, search);
 }
@@ -206,48 +225,32 @@ static bool __replicas_has_entry(struct bch_replicas_cpu *r,
 	return __replicas_entry_idx(r, search) >= 0;
 }
 
-static bool bch2_replicas_marked_locked(struct bch_fs *c,
-			  struct bch_replicas_entry *search,
-			  bool check_gc_replicas)
+bool bch2_replicas_marked(struct bch_fs *c,
+			  struct bch_replicas_entry *search)
 {
+	bool marked;
+
 	if (!search->nr_devs)
 		return true;
 
-	verify_replicas_entry_sorted(search);
+	verify_replicas_entry(search);
 
-	return __replicas_has_entry(&c->replicas, search) &&
-		(!check_gc_replicas ||
-		 likely((!c->replicas_gc.entries)) ||
+	percpu_down_read(&c->mark_lock);
+	marked = __replicas_has_entry(&c->replicas, search) &&
+		(likely((!c->replicas_gc.entries)) ||
 		 __replicas_has_entry(&c->replicas_gc, search));
-}
-
-bool bch2_replicas_marked(struct bch_fs *c,
-			  struct bch_replicas_entry *search,
-			  bool check_gc_replicas)
-{
-	bool marked;
-
-	percpu_down_read_preempt_disable(&c->mark_lock);
-	marked = bch2_replicas_marked_locked(c, search, check_gc_replicas);
-	percpu_up_read_preempt_enable(&c->mark_lock);
+	percpu_up_read(&c->mark_lock);
 
 	return marked;
 }
 
-static void __replicas_table_update(struct bch_fs_usage __percpu *dst_p,
+static void __replicas_table_update(struct bch_fs_usage *dst,
 				    struct bch_replicas_cpu *dst_r,
-				    struct bch_fs_usage __percpu *src_p,
+				    struct bch_fs_usage *src,
 				    struct bch_replicas_cpu *src_r)
 {
-	unsigned src_nr = sizeof(struct bch_fs_usage) / sizeof(u64) + src_r->nr;
-	struct bch_fs_usage *dst, *src = (void *)
-		bch2_acc_percpu_u64s((void *) src_p, src_nr);
 	int src_idx, dst_idx;
 
-	preempt_disable();
-	dst = this_cpu_ptr(dst_p);
-	preempt_enable();
-
 	*dst = *src;
 
 	for (src_idx = 0; src_idx < src_r->nr; src_idx++) {
@@ -262,43 +265,79 @@ static void __replicas_table_update(struct bch_fs_usage __percpu *dst_p,
 	}
 }
 
+static void __replicas_table_update_pcpu(struct bch_fs_usage __percpu *dst_p,
+				    struct bch_replicas_cpu *dst_r,
+				    struct bch_fs_usage __percpu *src_p,
+				    struct bch_replicas_cpu *src_r)
+{
+	unsigned src_nr = sizeof(struct bch_fs_usage) / sizeof(u64) + src_r->nr;
+	struct bch_fs_usage *dst, *src = (void *)
+		bch2_acc_percpu_u64s((u64 __percpu *) src_p, src_nr);
+
+	preempt_disable();
+	dst = this_cpu_ptr(dst_p);
+	preempt_enable();
+
+	__replicas_table_update(dst, dst_r, src, src_r);
+}
+
 /*
  * Resize filesystem accounting:
  */
 static int replicas_table_update(struct bch_fs *c,
 				 struct bch_replicas_cpu *new_r)
 {
-	struct bch_fs_usage __percpu *new_usage[2] = { NULL, NULL };
-	struct bch_fs_usage *new_scratch = NULL;
-	unsigned bytes = sizeof(struct bch_fs_usage) +
+	struct bch_fs_usage __percpu *new_usage[JOURNAL_BUF_NR];
+	struct bch_fs_usage_online *new_scratch = NULL;
+	struct bch_fs_usage __percpu *new_gc = NULL;
+	struct bch_fs_usage *new_base = NULL;
+	unsigned i, bytes = sizeof(struct bch_fs_usage) +
 		sizeof(u64) * new_r->nr;
-	int ret = -ENOMEM;
-
-	if (!(new_usage[0] = __alloc_percpu_gfp(bytes, sizeof(u64),
-						GFP_NOIO)) ||
-	    (c->usage[1] &&
-	     !(new_usage[1] = __alloc_percpu_gfp(bytes, sizeof(u64),
-						 GFP_NOIO))) ||
-	    !(new_scratch  = kmalloc(bytes, GFP_NOIO)))
-		goto err;
+	unsigned scratch_bytes = sizeof(struct bch_fs_usage_online) +
+		sizeof(u64) * new_r->nr;
+	int ret = 0;
+
+	memset(new_usage, 0, sizeof(new_usage));
 
-	if (c->usage[0])
-		__replicas_table_update(new_usage[0],	new_r,
-					c->usage[0],	&c->replicas);
-	if (c->usage[1])
-		__replicas_table_update(new_usage[1],	new_r,
-					c->usage[1],	&c->replicas);
+	for (i = 0; i < ARRAY_SIZE(new_usage); i++)
+		if (!(new_usage[i] = __alloc_percpu_gfp(bytes,
+					sizeof(u64), GFP_KERNEL)))
+			goto err;
 
-	swap(c->usage[0],	new_usage[0]);
-	swap(c->usage[1],	new_usage[1]);
+	if (!(new_base = kzalloc(bytes, GFP_KERNEL)) ||
+	    !(new_scratch  = kmalloc(scratch_bytes, GFP_KERNEL)) ||
+	    (c->usage_gc &&
+	     !(new_gc = __alloc_percpu_gfp(bytes, sizeof(u64), GFP_KERNEL))))
+		goto err;
+
+	for (i = 0; i < ARRAY_SIZE(new_usage); i++)
+		if (c->usage[i])
+			__replicas_table_update_pcpu(new_usage[i], new_r,
+						     c->usage[i], &c->replicas);
+	if (c->usage_base)
+		__replicas_table_update(new_base,		new_r,
+					c->usage_base,		&c->replicas);
+	if (c->usage_gc)
+		__replicas_table_update_pcpu(new_gc,		new_r,
+					     c->usage_gc,	&c->replicas);
+
+	for (i = 0; i < ARRAY_SIZE(new_usage); i++)
+		swap(c->usage[i],	new_usage[i]);
+	swap(c->usage_base,	new_base);
 	swap(c->usage_scratch,	new_scratch);
+	swap(c->usage_gc,	new_gc);
 	swap(c->replicas,	*new_r);
-	ret = 0;
-err:
+out:
+	free_percpu(new_gc);
 	kfree(new_scratch);
-	free_percpu(new_usage[1]);
-	free_percpu(new_usage[0]);
+	for (i = 0; i < ARRAY_SIZE(new_usage); i++)
+		free_percpu(new_usage[i]);
+	kfree(new_base);
 	return ret;
+err:
+	bch_err(c, "error updating replicas table: memory allocation failure");
+	ret = -BCH_ERR_ENOMEM_replicas_table;
+	goto out;
 }
 
 static unsigned reserve_journal_replicas(struct bch_fs *c,
@@ -332,7 +371,9 @@ static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 				struct bch_replicas_entry *new_entry)
 {
 	struct bch_replicas_cpu new_r, new_gc;
-	int ret = -ENOMEM;
+	int ret = 0;
+
+	verify_replicas_entry(new_entry);
 
 	memset(&new_r, 0, sizeof(new_r));
 	memset(&new_gc, 0, sizeof(new_gc));
@@ -342,14 +383,18 @@ static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 	if (c->replicas_gc.entries &&
 	    !__replicas_has_entry(&c->replicas_gc, new_entry)) {
 		new_gc = cpu_replicas_add_entry(&c->replicas_gc, new_entry);
-		if (!new_gc.entries)
+		if (!new_gc.entries) {
+			ret = -BCH_ERR_ENOMEM_cpu_replicas;
 			goto err;
+		}
 	}
 
 	if (!__replicas_has_entry(&c->replicas, new_entry)) {
 		new_r = cpu_replicas_add_entry(&c->replicas, new_entry);
-		if (!new_r.entries)
+		if (!new_r.entries) {
+			ret = -BCH_ERR_ENOMEM_cpu_replicas;
 			goto err;
+		}
 
 		ret = bch2_cpu_replicas_to_sb_replicas(c, &new_r);
 		if (ret)
@@ -377,138 +422,68 @@ static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 		swap(new_gc, c->replicas_gc);
 	percpu_up_write(&c->mark_lock);
 out:
-	ret = 0;
-err:
 	mutex_unlock(&c->sb_lock);
 
 	kfree(new_r.entries);
 	kfree(new_gc.entries);
 
 	return ret;
+err:
+	bch_err_msg(c, ret, "adding replicas entry");
+	goto out;
 }
 
-int bch2_mark_replicas(struct bch_fs *c,
-		       struct bch_replicas_entry *r)
+int bch2_mark_replicas(struct bch_fs *c, struct bch_replicas_entry *r)
 {
-	return likely(bch2_replicas_marked(c, r, true))
-		? 0
-		: bch2_mark_replicas_slowpath(c, r);
+	return likely(bch2_replicas_marked(c, r))
+		? 0 : bch2_mark_replicas_slowpath(c, r);
 }
 
-bool bch2_bkey_replicas_marked_locked(struct bch_fs *c,
-				      struct bkey_s_c k,
-				      bool check_gc_replicas)
-{
-	struct bch_replicas_padded search;
-	struct bch_devs_list cached = bch2_bkey_cached_devs(k);
-	unsigned i;
-
-	for (i = 0; i < cached.nr; i++) {
-		bch2_replicas_entry_cached(&search.e, cached.devs[i]);
-
-		if (!bch2_replicas_marked_locked(c, &search.e,
-						 check_gc_replicas))
-			return false;
-	}
-
-	bkey_to_replicas(&search.e, k);
-
-	return bch2_replicas_marked_locked(c, &search.e, check_gc_replicas);
-}
+/* replicas delta list: */
 
-bool bch2_bkey_replicas_marked(struct bch_fs *c,
-			       struct bkey_s_c k,
-			       bool check_gc_replicas)
+int bch2_replicas_delta_list_mark(struct bch_fs *c,
+				  struct replicas_delta_list *r)
 {
-	bool marked;
-
-	percpu_down_read_preempt_disable(&c->mark_lock);
-	marked = bch2_bkey_replicas_marked_locked(c, k, check_gc_replicas);
-	percpu_up_read_preempt_enable(&c->mark_lock);
+	struct replicas_delta *d = r->d;
+	struct replicas_delta *top = (void *) r->d + r->used;
+	int ret = 0;
 
-	return marked;
+	for (d = r->d; !ret && d != top; d = replicas_delta_next(d))
+		ret = bch2_mark_replicas(c, &d->r);
+	return ret;
 }
 
-int bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
-{
-	struct bch_replicas_padded search;
-	struct bch_devs_list cached = bch2_bkey_cached_devs(k);
-	unsigned i;
-	int ret;
-
-	for (i = 0; i < cached.nr; i++) {
-		bch2_replicas_entry_cached(&search.e, cached.devs[i]);
-
-		ret = bch2_mark_replicas(c, &search.e);
-		if (ret)
-			return ret;
-	}
-
-	bkey_to_replicas(&search.e, k);
-
-	return bch2_mark_replicas(c, &search.e);
-}
+/*
+ * Old replicas_gc mechanism: only used for journal replicas entries now, should
+ * die at some point:
+ */
 
 int bch2_replicas_gc_end(struct bch_fs *c, int ret)
 {
-	unsigned i;
-
 	lockdep_assert_held(&c->replicas_gc_lock);
 
-	mutex_lock(&c->sb_lock);
-
 	if (ret)
 		goto err;
 
-	/*
-	 * this is kind of crappy; the replicas gc mechanism needs to be ripped
-	 * out
-	 */
-
-	for (i = 0; i < c->replicas.nr; i++) {
-		struct bch_replicas_entry *e =
-			cpu_replicas_entry(&c->replicas, i);
-		struct bch_replicas_cpu n;
-		u64 v;
-
-		if (__replicas_has_entry(&c->replicas_gc, e))
-			continue;
-
-		v = percpu_u64_get(&c->usage[0]->replicas[i]);
-		if (!v)
-			continue;
-
-		n = cpu_replicas_add_entry(&c->replicas_gc, e);
-		if (!n.entries) {
-			ret = -ENOSPC;
-			goto err;
-		}
-
-		percpu_down_write(&c->mark_lock);
-		swap(n, c->replicas_gc);
-		percpu_up_write(&c->mark_lock);
-
-		kfree(n.entries);
-	}
+	mutex_lock(&c->sb_lock);
+	percpu_down_write(&c->mark_lock);
 
-	if (bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc)) {
-		ret = -ENOSPC;
+	ret = bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc);
+	if (ret)
 		goto err;
-	}
-
-	bch2_write_super(c);
 
-	/* don't update in memory replicas until changes are persistent */
+	ret = replicas_table_update(c, &c->replicas_gc);
 err:
-	percpu_down_write(&c->mark_lock);
-	if (!ret)
-		ret = replicas_table_update(c, &c->replicas_gc);
-
 	kfree(c->replicas_gc.entries);
 	c->replicas_gc.entries = NULL;
+
 	percpu_up_write(&c->mark_lock);
 
+	if (!ret)
+		bch2_write_super(c);
+
 	mutex_unlock(&c->sb_lock);
+
 	return ret;
 }
 
@@ -535,10 +510,11 @@ int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
 
 	c->replicas_gc.entries = kcalloc(c->replicas_gc.nr,
 					 c->replicas_gc.entry_size,
-					 GFP_NOIO);
+					 GFP_KERNEL);
 	if (!c->replicas_gc.entries) {
 		mutex_unlock(&c->sb_lock);
-		return -ENOMEM;
+		bch_err(c, "error allocating c->replicas_gc");
+		return -BCH_ERR_ENOMEM_replicas_gc;
 	}
 
 	for_each_cpu_replicas_entry(&c->replicas, e)
@@ -552,6 +528,75 @@ int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
 	return 0;
 }
 
+/*
+ * New much simpler mechanism for clearing out unneeded replicas entries - drop
+ * replicas entries that have 0 sectors used.
+ *
+ * However, we don't track sector counts for journal usage, so this doesn't drop
+ * any BCH_DATA_journal entries; the old bch2_replicas_gc_(start|end) mechanism
+ * is retained for that.
+ */
+int bch2_replicas_gc2(struct bch_fs *c)
+{
+	struct bch_replicas_cpu new = { 0 };
+	unsigned i, nr;
+	int ret = 0;
+
+	bch2_journal_meta(&c->journal);
+retry:
+	nr		= READ_ONCE(c->replicas.nr);
+	new.entry_size	= READ_ONCE(c->replicas.entry_size);
+	new.entries	= kcalloc(nr, new.entry_size, GFP_KERNEL);
+	if (!new.entries) {
+		bch_err(c, "error allocating c->replicas_gc");
+		return -BCH_ERR_ENOMEM_replicas_gc;
+	}
+
+	mutex_lock(&c->sb_lock);
+	percpu_down_write(&c->mark_lock);
+
+	if (nr			!= c->replicas.nr ||
+	    new.entry_size	!= c->replicas.entry_size) {
+		percpu_up_write(&c->mark_lock);
+		mutex_unlock(&c->sb_lock);
+		kfree(new.entries);
+		goto retry;
+	}
+
+	for (i = 0; i < c->replicas.nr; i++) {
+		struct bch_replicas_entry *e =
+			cpu_replicas_entry(&c->replicas, i);
+
+		if (e->data_type == BCH_DATA_journal ||
+		    c->usage_base->replicas[i] ||
+		    percpu_u64_get(&c->usage[0]->replicas[i]) ||
+		    percpu_u64_get(&c->usage[1]->replicas[i]) ||
+		    percpu_u64_get(&c->usage[2]->replicas[i]) ||
+		    percpu_u64_get(&c->usage[3]->replicas[i]))
+			memcpy(cpu_replicas_entry(&new, new.nr++),
+			       e, new.entry_size);
+	}
+
+	bch2_cpu_replicas_sort(&new);
+
+	ret = bch2_cpu_replicas_to_sb_replicas(c, &new);
+	if (ret)
+		goto err;
+
+	ret = replicas_table_update(c, &new);
+err:
+	kfree(new.entries);
+
+	percpu_up_write(&c->mark_lock);
+
+	if (!ret)
+		bch2_write_super(c);
+
+	mutex_unlock(&c->sb_lock);
+
+	return ret;
+}
+
 int bch2_replicas_set_usage(struct bch_fs *c,
 			    struct bch_replicas_entry *r,
 			    u64 sectors)
@@ -563,7 +608,7 @@ int bch2_replicas_set_usage(struct bch_fs *c,
 
 		n = cpu_replicas_add_entry(&c->replicas, r);
 		if (!n.entries)
-			return -ENOMEM;
+			return -BCH_ERR_ENOMEM_cpu_replicas;
 
 		ret = replicas_table_update(c, &n);
 		if (ret)
@@ -575,7 +620,7 @@ int bch2_replicas_set_usage(struct bch_fs *c,
 		BUG_ON(ret < 0);
 	}
 
-	percpu_u64_set(&c->usage[0]->replicas[idx], sectors);
+	c->usage_base->replicas[idx] = sectors;
 
 	return 0;
 }
@@ -595,9 +640,9 @@ __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r,
 		nr++;
 	}
 
-	cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
+	cpu_r->entries = kcalloc(nr, entry_size, GFP_KERNEL);
 	if (!cpu_r->entries)
-		return -ENOMEM;
+		return -BCH_ERR_ENOMEM_cpu_replicas;
 
 	cpu_r->nr		= nr;
 	cpu_r->entry_size	= entry_size;
@@ -605,7 +650,7 @@ __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r,
 	for_each_replicas_entry(sb_r, e) {
 		dst = cpu_replicas_entry(cpu_r, idx++);
 		memcpy(dst, e, replicas_entry_bytes(e));
-		replicas_entry_sort(dst);
+		bch2_replicas_entry_sort(dst);
 	}
 
 	return 0;
@@ -627,9 +672,9 @@ __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r,
 	entry_size += sizeof(struct bch_replicas_entry) -
 		sizeof(struct bch_replicas_entry_v0);
 
-	cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
+	cpu_r->entries = kcalloc(nr, entry_size, GFP_KERNEL);
 	if (!cpu_r->entries)
-		return -ENOMEM;
+		return -BCH_ERR_ENOMEM_cpu_replicas;
 
 	cpu_r->nr		= nr;
 	cpu_r->entry_size	= entry_size;
@@ -642,7 +687,7 @@ __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r,
 		dst->nr_devs	= e->nr_devs;
 		dst->nr_required = 1;
 		memcpy(dst->devs, e->devs, e->nr_devs);
-		replicas_entry_sort(dst);
+		bch2_replicas_entry_sort(dst);
 	}
 
 	return 0;
@@ -655,13 +700,12 @@ int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
 	struct bch_replicas_cpu new_r = { 0, 0, NULL };
 	int ret = 0;
 
-	if ((sb_v1 = bch2_sb_get_replicas(c->disk_sb.sb)))
+	if ((sb_v1 = bch2_sb_field_get(c->disk_sb.sb, replicas)))
 		ret = __bch2_sb_replicas_to_cpu_replicas(sb_v1, &new_r);
-	else if ((sb_v0 = bch2_sb_get_replicas_v0(c->disk_sb.sb)))
+	else if ((sb_v0 = bch2_sb_field_get(c->disk_sb.sb, replicas_v0)))
 		ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0, &new_r);
-
 	if (ret)
-		return -ENOMEM;
+		return ret;
 
 	bch2_cpu_replicas_sort(&new_r);
 
@@ -688,13 +732,13 @@ static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c,
 	for_each_cpu_replicas_entry(r, src)
 		bytes += replicas_entry_bytes(src) - 1;
 
-	sb_r = bch2_sb_resize_replicas_v0(&c->disk_sb,
+	sb_r = bch2_sb_field_resize(&c->disk_sb, replicas_v0,
 			DIV_ROUND_UP(bytes, sizeof(u64)));
 	if (!sb_r)
-		return -ENOSPC;
+		return -BCH_ERR_ENOSPC_sb_replicas;
 
 	bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
-	sb_r = bch2_sb_get_replicas_v0(c->disk_sb.sb);
+	sb_r = bch2_sb_field_get(c->disk_sb.sb, replicas_v0);
 
 	memset(&sb_r->entries, 0,
 	       vstruct_end(&sb_r->field) -
@@ -733,13 +777,13 @@ static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
 	if (!need_v1)
 		return bch2_cpu_replicas_to_sb_replicas_v0(c, r);
 
-	sb_r = bch2_sb_resize_replicas(&c->disk_sb,
+	sb_r = bch2_sb_field_resize(&c->disk_sb, replicas,
 			DIV_ROUND_UP(bytes, sizeof(u64)));
 	if (!sb_r)
-		return -ENOSPC;
+		return -BCH_ERR_ENOSPC_sb_replicas;
 
 	bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
-	sb_r = bch2_sb_get_replicas(c->disk_sb.sb);
+	sb_r = bch2_sb_field_get(c->disk_sb.sb, replicas);
 
 	memset(&sb_r->entries, 0,
 	       vstruct_end(&sb_r->field) -
@@ -757,68 +801,78 @@ static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
 	return 0;
 }
 
-static const char *check_dup_replicas_entries(struct bch_replicas_cpu *cpu_r)
+static int bch2_cpu_replicas_validate(struct bch_replicas_cpu *cpu_r,
+				      struct bch_sb *sb,
+				      struct printbuf *err)
 {
-	unsigned i;
+	unsigned i, j;
 
 	sort_cmp_size(cpu_r->entries,
 		      cpu_r->nr,
 		      cpu_r->entry_size,
 		      memcmp, NULL);
 
-	for (i = 0; i + 1 < cpu_r->nr; i++) {
-		struct bch_replicas_entry *l =
+	for (i = 0; i < cpu_r->nr; i++) {
+		struct bch_replicas_entry *e =
 			cpu_replicas_entry(cpu_r, i);
-		struct bch_replicas_entry *r =
-			cpu_replicas_entry(cpu_r, i + 1);
 
-		BUG_ON(memcmp(l, r, cpu_r->entry_size) > 0);
-
-		if (!memcmp(l, r, cpu_r->entry_size))
-			return "duplicate replicas entry";
-	}
+		if (e->data_type >= BCH_DATA_NR) {
+			prt_printf(err, "invalid data type in entry ");
+			bch2_replicas_entry_to_text(err, e);
+			return -BCH_ERR_invalid_sb_replicas;
+		}
 
-	return NULL;
-}
+		if (!e->nr_devs) {
+			prt_printf(err, "no devices in entry ");
+			bch2_replicas_entry_to_text(err, e);
+			return -BCH_ERR_invalid_sb_replicas;
+		}
 
-static const char *bch2_sb_validate_replicas(struct bch_sb *sb, struct bch_sb_field *f)
-{
-	struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
-	struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
-	struct bch_replicas_cpu cpu_r = { .entries = NULL };
-	struct bch_replicas_entry *e;
-	const char *err;
-	unsigned i;
+		if (e->nr_required > 1 &&
+		    e->nr_required >= e->nr_devs) {
+			prt_printf(err, "bad nr_required in entry ");
+			bch2_replicas_entry_to_text(err, e);
+			return -BCH_ERR_invalid_sb_replicas;
+		}
 
-	for_each_replicas_entry(sb_r, e) {
-		err = "invalid replicas entry: invalid data type";
-		if (e->data_type >= BCH_DATA_NR)
-			goto err;
+		for (j = 0; j < e->nr_devs; j++)
+			if (!bch2_dev_exists(sb, e->devs[j])) {
+				prt_printf(err, "invalid device %u in entry ", e->devs[j]);
+				bch2_replicas_entry_to_text(err, e);
+				return -BCH_ERR_invalid_sb_replicas;
+			}
 
-		err = "invalid replicas entry: no devices";
-		if (!e->nr_devs)
-			goto err;
+		if (i + 1 < cpu_r->nr) {
+			struct bch_replicas_entry *n =
+				cpu_replicas_entry(cpu_r, i + 1);
 
-		err = "invalid replicas entry: bad nr_required";
-		if (!e->nr_required ||
-		    (e->nr_required > 1 &&
-		     e->nr_required >= e->nr_devs))
-			goto err;
+			BUG_ON(memcmp(e, n, cpu_r->entry_size) > 0);
 
-		err = "invalid replicas entry: invalid device";
-		for (i = 0; i < e->nr_devs; i++)
-			if (!bch2_dev_exists(sb, mi, e->devs[i]))
-				goto err;
+			if (!memcmp(e, n, cpu_r->entry_size)) {
+				prt_printf(err, "duplicate replicas entry ");
+				bch2_replicas_entry_to_text(err, e);
+				return -BCH_ERR_invalid_sb_replicas;
+			}
+		}
 	}
 
-	err = "cannot allocate memory";
-	if (__bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r))
-		goto err;
+	return 0;
+}
 
-	err = check_dup_replicas_entries(&cpu_r);
-err:
+static int bch2_sb_replicas_validate(struct bch_sb *sb, struct bch_sb_field *f,
+				     struct printbuf *err)
+{
+	struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
+	struct bch_replicas_cpu cpu_r;
+	int ret;
+
+	ret = __bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r);
+	if (ret)
+		return ret;
+
+	ret = bch2_cpu_replicas_validate(&cpu_r, sb, err);
 	kfree(cpu_r.entries);
-	return err;
+	return ret;
 }
 
 static void bch2_sb_replicas_to_text(struct printbuf *out,
@@ -831,167 +885,174 @@ static void bch2_sb_replicas_to_text(struct printbuf *out,
 
 	for_each_replicas_entry(r, e) {
 		if (!first)
-			pr_buf(out, " ");
+			prt_printf(out, " ");
 		first = false;
 
 		bch2_replicas_entry_to_text(out, e);
 	}
+	prt_newline(out);
 }
 
 const struct bch_sb_field_ops bch_sb_field_ops_replicas = {
-	.validate	= bch2_sb_validate_replicas,
+	.validate	= bch2_sb_replicas_validate,
 	.to_text	= bch2_sb_replicas_to_text,
 };
 
-static const char *bch2_sb_validate_replicas_v0(struct bch_sb *sb, struct bch_sb_field *f)
+static int bch2_sb_replicas_v0_validate(struct bch_sb *sb, struct bch_sb_field *f,
+					struct printbuf *err)
 {
 	struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
-	struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
-	struct bch_replicas_cpu cpu_r = { .entries = NULL };
-	struct bch_replicas_entry_v0 *e;
-	const char *err;
-	unsigned i;
+	struct bch_replicas_cpu cpu_r;
+	int ret;
 
-	for_each_replicas_entry_v0(sb_r, e) {
-		err = "invalid replicas entry: invalid data type";
-		if (e->data_type >= BCH_DATA_NR)
-			goto err;
+	ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r);
+	if (ret)
+		return ret;
 
-		err = "invalid replicas entry: no devices";
-		if (!e->nr_devs)
-			goto err;
+	ret = bch2_cpu_replicas_validate(&cpu_r, sb, err);
+	kfree(cpu_r.entries);
+	return ret;
+}
 
-		err = "invalid replicas entry: invalid device";
-		for (i = 0; i < e->nr_devs; i++)
-			if (!bch2_dev_exists(sb, mi, e->devs[i]))
-				goto err;
-	}
+static void bch2_sb_replicas_v0_to_text(struct printbuf *out,
+					struct bch_sb *sb,
+					struct bch_sb_field *f)
+{
+	struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
+	struct bch_replicas_entry_v0 *e;
+	bool first = true;
 
-	err = "cannot allocate memory";
-	if (__bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r))
-		goto err;
+	for_each_replicas_entry(sb_r, e) {
+		if (!first)
+			prt_printf(out, " ");
+		first = false;
 
-	err = check_dup_replicas_entries(&cpu_r);
-err:
-	kfree(cpu_r.entries);
-	return err;
+		bch2_replicas_entry_v0_to_text(out, e);
+	}
+	prt_newline(out);
 }
 
 const struct bch_sb_field_ops bch_sb_field_ops_replicas_v0 = {
-	.validate	= bch2_sb_validate_replicas_v0,
+	.validate	= bch2_sb_replicas_v0_validate,
+	.to_text	= bch2_sb_replicas_v0_to_text,
 };
 
 /* Query replicas: */
 
-struct replicas_status __bch2_replicas_status(struct bch_fs *c,
-					      struct bch_devs_mask online_devs)
+bool bch2_have_enough_devs(struct bch_fs *c, struct bch_devs_mask devs,
+			   unsigned flags, bool print)
 {
-	struct bch_sb_field_members *mi;
 	struct bch_replicas_entry *e;
-	unsigned i, nr_online, nr_offline;
-	struct replicas_status ret;
-
-	memset(&ret, 0, sizeof(ret));
-
-	for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
-		ret.replicas[i].redundancy = INT_MAX;
-
-	mi = bch2_sb_get_members(c->disk_sb.sb);
-
-	percpu_down_read_preempt_disable(&c->mark_lock);
+	bool ret = true;
 
+	percpu_down_read(&c->mark_lock);
 	for_each_cpu_replicas_entry(&c->replicas, e) {
-		if (e->data_type >= ARRAY_SIZE(ret.replicas))
-			panic("e %p data_type %u\n", e, e->data_type);
+		unsigned i, nr_online = 0, nr_failed = 0, dflags = 0;
+		bool metadata = e->data_type < BCH_DATA_user;
 
-		nr_online = nr_offline = 0;
+		if (e->data_type == BCH_DATA_cached)
+			continue;
 
 		for (i = 0; i < e->nr_devs; i++) {
-			BUG_ON(!bch2_dev_exists(c->disk_sb.sb, mi,
-						e->devs[i]));
+			struct bch_dev *ca = bch_dev_bkey_exists(c, e->devs[i]);
 
-			if (test_bit(e->devs[i], online_devs.d))
-				nr_online++;
-			else
-				nr_offline++;
+			nr_online += test_bit(e->devs[i], devs.d);
+			nr_failed += ca->mi.state == BCH_MEMBER_STATE_failed;
 		}
 
-		ret.replicas[e->data_type].redundancy =
-			min(ret.replicas[e->data_type].redundancy,
-			    (int) nr_online - (int) e->nr_required);
-
-		ret.replicas[e->data_type].nr_offline =
-			max(ret.replicas[e->data_type].nr_offline,
-			    nr_offline);
-	}
+		if (nr_failed == e->nr_devs)
+			continue;
 
-	percpu_up_read_preempt_enable(&c->mark_lock);
+		if (nr_online < e->nr_required)
+			dflags |= metadata
+				? BCH_FORCE_IF_METADATA_LOST
+				: BCH_FORCE_IF_DATA_LOST;
+
+		if (nr_online < e->nr_devs)
+			dflags |= metadata
+				? BCH_FORCE_IF_METADATA_DEGRADED
+				: BCH_FORCE_IF_DATA_DEGRADED;
+
+		if (dflags & ~flags) {
+			if (print) {
+				struct printbuf buf = PRINTBUF;
+
+				bch2_replicas_entry_to_text(&buf, e);
+				bch_err(c, "insufficient devices online (%u) for replicas entry %s",
+					nr_online, buf.buf);
+				printbuf_exit(&buf);
+			}
+			ret = false;
+			break;
+		}
 
-	for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
-		if (ret.replicas[i].redundancy == INT_MAX)
-			ret.replicas[i].redundancy = 0;
+	}
+	percpu_up_read(&c->mark_lock);
 
 	return ret;
 }
 
-struct replicas_status bch2_replicas_status(struct bch_fs *c)
+unsigned bch2_sb_dev_has_data(struct bch_sb *sb, unsigned dev)
 {
-	return __bch2_replicas_status(c, bch2_online_devs(c));
-}
+	struct bch_sb_field_replicas *replicas;
+	struct bch_sb_field_replicas_v0 *replicas_v0;
+	unsigned i, data_has = 0;
+
+	replicas = bch2_sb_field_get(sb, replicas);
+	replicas_v0 = bch2_sb_field_get(sb, replicas_v0);
+
+	if (replicas) {
+		struct bch_replicas_entry *r;
+
+		for_each_replicas_entry(replicas, r)
+			for (i = 0; i < r->nr_devs; i++)
+				if (r->devs[i] == dev)
+					data_has |= 1 << r->data_type;
+	} else if (replicas_v0) {
+		struct bch_replicas_entry_v0 *r;
+
+		for_each_replicas_entry_v0(replicas_v0, r)
+			for (i = 0; i < r->nr_devs; i++)
+				if (r->devs[i] == dev)
+					data_has |= 1 << r->data_type;
+	}
 
-static bool have_enough_devs(struct replicas_status s,
-			     enum bch_data_type type,
-			     bool force_if_degraded,
-			     bool force_if_lost)
-{
-	return (!s.replicas[type].nr_offline || force_if_degraded) &&
-		(s.replicas[type].redundancy >= 0 || force_if_lost);
-}
 
-bool bch2_have_enough_devs(struct replicas_status s, unsigned flags)
-{
-	return (have_enough_devs(s, BCH_DATA_JOURNAL,
-				 flags & BCH_FORCE_IF_METADATA_DEGRADED,
-				 flags & BCH_FORCE_IF_METADATA_LOST) &&
-		have_enough_devs(s, BCH_DATA_BTREE,
-				 flags & BCH_FORCE_IF_METADATA_DEGRADED,
-				 flags & BCH_FORCE_IF_METADATA_LOST) &&
-		have_enough_devs(s, BCH_DATA_USER,
-				 flags & BCH_FORCE_IF_DATA_DEGRADED,
-				 flags & BCH_FORCE_IF_DATA_LOST));
+	return data_has;
 }
 
-int bch2_replicas_online(struct bch_fs *c, bool meta)
+unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
 {
-	struct replicas_status s = bch2_replicas_status(c);
+	unsigned ret;
 
-	return (meta
-		? min(s.replicas[BCH_DATA_JOURNAL].redundancy,
-		      s.replicas[BCH_DATA_BTREE].redundancy)
-		: s.replicas[BCH_DATA_USER].redundancy) + 1;
+	mutex_lock(&c->sb_lock);
+	ret = bch2_sb_dev_has_data(c->disk_sb.sb, ca->dev_idx);
+	mutex_unlock(&c->sb_lock);
+
+	return ret;
 }
 
-unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
+void bch2_fs_replicas_exit(struct bch_fs *c)
 {
-	struct bch_replicas_entry *e;
-	unsigned i, ret = 0;
-
-	percpu_down_read_preempt_disable(&c->mark_lock);
-
-	for_each_cpu_replicas_entry(&c->replicas, e)
-		for (i = 0; i < e->nr_devs; i++)
-			if (e->devs[i] == ca->dev_idx)
-				ret |= 1 << e->data_type;
+	unsigned i;
 
-	percpu_up_read_preempt_enable(&c->mark_lock);
+	kfree(c->usage_scratch);
+	for (i = 0; i < ARRAY_SIZE(c->usage); i++)
+		free_percpu(c->usage[i]);
+	kfree(c->usage_base);
+	kfree(c->replicas.entries);
+	kfree(c->replicas_gc.entries);
 
-	return ret;
+	mempool_exit(&c->replicas_delta_pool);
 }
 
 int bch2_fs_replicas_init(struct bch_fs *c)
 {
-	c->journal.entry_u64s_reserved +=
-		reserve_journal_replicas(c, &c->replicas);
+	bch2_journal_entry_res_resize(&c->journal,
+			&c->replicas_journal_res,
+			reserve_journal_replicas(c, &c->replicas));
 
-	return replicas_table_update(c, &c->replicas);
+	return mempool_init_kmalloc_pool(&c->replicas_delta_pool, 1,
+					 REPLICAS_DELTA_LIST_MAX) ?:
+		replicas_table_update(c, &c->replicas);
 }