Update bcachefs sources to 7bf1ac0d46 bcachefs: Correctly initialize new buckets on device resize

Signed-off-by: Hunter Shaffer <huntershaffer182456@gmail.com>

.bcachefs_revision

@@ -1 +1 @@
-a1b6677dca574a8bf904d9eea2b108474dc378d1
+7bf1ac0d46ebede68561e0476f7af9c07ac21de8

cmd_device.c

@@ -414,7 +414,7 @@ int cmd_device_set_state(int argc, char *argv[])
 		if (ret)
 			die("error opening %s: %s", dev_str, bch2_err_str(ret));
 
-		struct bch_member *m = bch2_sb_get_members(sb.sb)->members + sb.sb->dev_idx;
+		struct bch_member *m = bch2_members_v2_get_mut(sb.sb, sb.sb->dev_idx);
 
 		SET_BCH_MEMBER_STATE(m, new_state);
 
@@ -510,16 +510,11 @@ int cmd_device_resize(int argc, char *argv[])
 		if (idx >= sb->nr_devices)
 			die("error reading superblock: dev idx >= sb->nr_devices");
 
-		struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
+		struct bch_member m = bch2_sb_member_get(sb, idx);
-		if (!mi)
-			die("error reading superblock: no member info");
 
-		/* could also just read this out of sysfs... meh */
+		u64 nbuckets = size / le16_to_cpu(m.bucket_size);
-		struct bch_member *m = mi->members + idx;
 
-		u64 nbuckets = size / le16_to_cpu(m->bucket_size);
+		if (nbuckets < le64_to_cpu(m.nbuckets))
-
-		if (nbuckets < le64_to_cpu(m->nbuckets))
 			die("Shrinking not supported yet");
 
 		printf("resizing %s to %llu buckets\n", dev, nbuckets);
@@ -616,14 +611,9 @@ int cmd_device_resize_journal(int argc, char *argv[])
 		if (idx >= sb->nr_devices)
 			die("error reading superblock: dev idx >= sb->nr_devices");
 
-		struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
+		struct bch_member m = bch2_sb_member_get(sb, idx);
-		if (!mi)
-			die("error reading superblock: no member info");
 
-		/* could also just read this out of sysfs... meh */
+		u64 nbuckets = size / le16_to_cpu(m.bucket_size);
-		struct bch_member *m = mi->members + idx;
-
-		u64 nbuckets = size / le16_to_cpu(m->bucket_size);
 
 		printf("resizing journal on %s to %llu buckets\n", dev, nbuckets);
 		bchu_disk_resize_journal(fs, idx, nbuckets);

cmd_format.c

@@ -245,7 +245,7 @@ int cmd_format(int argc, char *argv[])
 
 		buf.human_readable_units = true;
 
-		bch2_sb_to_text(&buf, sb, false, 1 << BCH_SB_FIELD_members);
+		bch2_sb_to_text(&buf, sb, false, 1 << BCH_SB_FIELD_members_v2);
 		printf("%s", buf.buf);
 
 		printbuf_exit(&buf);
@@ -305,8 +305,9 @@ int cmd_show_super(int argc, char *argv[])
 		{ "help",			0, NULL, 'h' },
 		{ NULL }
 	};
-	unsigned fields = 1 << BCH_SB_FIELD_members;
+	unsigned fields = 0;
 	bool print_layout = false;
+	bool print_default_fields = true;
 	int opt;
 
 	while ((opt = getopt_long(argc, argv, "f:lh", longopts, NULL)) != -1)
@@ -316,6 +317,7 @@ int cmd_show_super(int argc, char *argv[])
 				? ~0
 				: read_flag_list_or_die(optarg,
 					bch2_sb_fields, "superblock field");
+			print_default_fields = false;
 			break;
 		case 'l':
 			print_layout = true;
@@ -342,6 +344,11 @@ int cmd_show_super(int argc, char *argv[])
 	if (ret)
 		die("Error opening %s: %s", dev, bch2_err_str(ret));
 
+	if (print_default_fields)
+		fields = bch2_sb_get_members_v2(sb.sb)
+			? 1 << BCH_SB_FIELD_members_v2
+			: 1 << BCH_SB_FIELD_members_v1;
+
 	struct printbuf buf = PRINTBUF;
 
 	buf.human_readable_units = true;

libbcachefs.c

@@ -150,7 +150,6 @@ struct bch_sb *bch2_format(struct bch_opt_strs	fs_opt_strs,
 {
 	struct bch_sb_handle sb = { NULL };
 	struct dev_opts *i;
-	struct bch_sb_field_members *mi;
 	unsigned max_dev_block_size = 0;
 	unsigned opt_id;
 
@@ -222,12 +221,13 @@ struct bch_sb *bch2_format(struct bch_opt_strs	fs_opt_strs,
 	sb.sb->time_precision	= cpu_to_le32(1);
 
 	/* Member info: */
-	mi = bch2_sb_resize_members(&sb,
+	struct bch_sb_field_members_v2 *mi =
+		bch2_sb_resize_members_v2(&sb,
 			(sizeof(*mi) + sizeof(struct bch_member) *
 			nr_devs) / sizeof(u64));
-
+	mi->member_bytes = cpu_to_le16(sizeof(struct bch_member));
 	for (i = devs; i < devs + nr_devs; i++) {
-		struct bch_member *m = mi->members + (i - devs);
+		struct bch_member *m = bch2_members_v2_get_mut(sb.sb, (i - devs));
 
 		uuid_generate(m->uuid.b);
 		m->nbuckets	= cpu_to_le64(i->nbuckets);
@@ -255,9 +255,7 @@ struct bch_sb *bch2_format(struct bch_opt_strs	fs_opt_strs,
 		 * Recompute mi and m after each sb modification: its location
 		 * in memory may have changed due to reallocation.
 		 */
-		mi = bch2_sb_get_members(sb.sb);
+		m = bch2_members_v2_get_mut(sb.sb, (i - devs));
-		m = mi->members + (i - devs);
-
 		SET_BCH_MEMBER_GROUP(m,	idx + 1);
 	}
 
@@ -279,6 +277,8 @@ struct bch_sb *bch2_format(struct bch_opt_strs	fs_opt_strs,
 		SET_BCH_SB_ENCRYPTION_TYPE(sb.sb, 1);
 	}
 
+	bch_members_cpy_v2_v1(&sb);
+
 	for (i = devs; i < devs + nr_devs; i++) {
 		u64 size_sectors = i->size >> 9;
 

libbcachefs/alloc_background.c

@@ -1831,29 +1831,33 @@ void bch2_do_invalidates(struct bch_fs *c)
 		bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
 }
 
-static int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca,
+int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca,
-				   unsigned long *last_updated)
+			    u64 bucket_start, u64 bucket_end)
 {
 	struct btree_trans *trans = bch2_trans_get(c);
 	struct btree_iter iter;
 	struct bkey_s_c k;
 	struct bkey hole;
-	struct bpos end = POS(ca->dev_idx, ca->mi.nbuckets);
+	struct bpos end = POS(ca->dev_idx, bucket_end);
-	struct bch_member *m;
+	struct bch_member m;
+	unsigned long last_updated = jiffies;
 	int ret;
 
+	BUG_ON(bucket_start > bucket_end);
+	BUG_ON(bucket_end > ca->mi.nbuckets);
+
 	bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
-			     POS(ca->dev_idx, ca->mi.first_bucket),
+		POS(ca->dev_idx, max_t(u64, ca->mi.first_bucket, bucket_start)),
-			     BTREE_ITER_PREFETCH);
+		BTREE_ITER_PREFETCH);
 	/*
 	 * Scan the alloc btree for every bucket on @ca, and add buckets to the
 	 * freespace/need_discard/need_gc_gens btrees as needed:
 	 */
 	while (1) {
-		if (*last_updated + HZ * 10 < jiffies) {
+		if (last_updated + HZ * 10 < jiffies) {
 			bch_info(ca, "%s: currently at %llu/%llu",
 				 __func__, iter.pos.offset, ca->mi.nbuckets);
-			*last_updated = jiffies;
+			last_updated = jiffies;
 		}
 
 		bch2_trans_begin(trans);
@@ -1922,8 +1926,8 @@ bkey_err:
 	}
 
 	mutex_lock(&c->sb_lock);
-	m = bch2_sb_get_members(c->disk_sb.sb)->members + ca->dev_idx;
+	m = bch2_sb_member_get(c->disk_sb.sb, ca->dev_idx);
-	SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, true);
+	SET_BCH_MEMBER_FREESPACE_INITIALIZED(&m, true);
 	mutex_unlock(&c->sb_lock);
 
 	return 0;
@@ -1935,7 +1939,6 @@ int bch2_fs_freespace_init(struct bch_fs *c)
 	unsigned i;
 	int ret = 0;
 	bool doing_init = false;
-	unsigned long last_updated = jiffies;
 
 	/*
 	 * We can crash during the device add path, so we need to check this on
@@ -1951,7 +1954,7 @@ int bch2_fs_freespace_init(struct bch_fs *c)
 			doing_init = true;
 		}
 
-		ret = bch2_dev_freespace_init(c, ca, &last_updated);
+		ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
 		if (ret) {
 			percpu_ref_put(&ca->ref);
 			bch_err_fn(c, ret);

libbcachefs/alloc_background.h

@@ -245,6 +245,7 @@ static inline const struct bch_backpointer *alloc_v4_backpointers_c(const struct
 	return (void *) ((u64 *) &a->v + BCH_ALLOC_V4_BACKPOINTERS_START(a));
 }
 
+int bch2_dev_freespace_init(struct bch_fs *, struct bch_dev *, u64, u64);
 int bch2_fs_freespace_init(struct bch_fs *);
 
 void bch2_recalc_capacity(struct bch_fs *);

libbcachefs/bcachefs.h

@@ -379,7 +379,7 @@ BCH_DEBUG_PARAMS()
 #undef BCH_DEBUG_PARAM
 
 #ifndef CONFIG_BCACHEFS_DEBUG
-#define BCH_DEBUG_PARAM(name, description) static const bool bch2_##name;
+#define BCH_DEBUG_PARAM(name, description) static const __maybe_unused bool bch2_##name;
 BCH_DEBUG_PARAMS_DEBUG()
 #undef BCH_DEBUG_PARAM
 #endif

libbcachefs/bcachefs_format.h

@@ -1222,7 +1222,7 @@ struct bch_sb_field {
 
 #define BCH_SB_FIELDS()				\
 	x(journal,	0)			\
-	x(members,	1)			\
+	x(members_v1,	1)			\
 	x(crypt,	2)			\
 	x(replicas_v0,	3)			\
 	x(quota,	4)			\
@@ -1231,7 +1231,8 @@ struct bch_sb_field {
 	x(replicas,	7)			\
 	x(journal_seq_blacklist, 8)		\
 	x(journal_v2,	9)			\
-	x(counters,	10)
+	x(counters,	10)			\
+	x(members_v2,	11)
 
 enum bch_sb_field_type {
 #define x(f, nr)	BCH_SB_FIELD_##f = nr,
@@ -1264,10 +1265,23 @@ struct bch_sb_field_journal_v2 {
 	}			d[];
 };
 
-/* BCH_SB_FIELD_members: */
+/* BCH_SB_FIELD_members_v1: */
 
 #define BCH_MIN_NR_NBUCKETS	(1 << 6)
 
+#define BCH_IOPS_MEASUREMENTS()			\
+	x(seqread,	0)			\
+	x(seqwrite,	1)			\
+	x(randread,	2)			\
+	x(randwrite,	3)
+
+enum bch_iops_measurement {
+#define x(t, n) BCH_IOPS_##t = n,
+	BCH_IOPS_MEASUREMENTS()
+#undef x
+	BCH_IOPS_NR
+};
+
 struct bch_member {
 	__uuid_t		uuid;
 	__le64			nbuckets;	/* device size */
@@ -1276,17 +1290,20 @@ struct bch_member {
 	__le32			pad;
 	__le64			last_mount;	/* time_t */
 
-	__le64			flags[2];
+	__le64			flags;
+	__le32			iops[4];
 };
 
-LE64_BITMASK(BCH_MEMBER_STATE,		struct bch_member, flags[0],  0,  4)
+#define BCH_MEMBER_V1_BYTES	56
+
+LE64_BITMASK(BCH_MEMBER_STATE,		struct bch_member, flags,  0,  4)
 /* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */
-LE64_BITMASK(BCH_MEMBER_DISCARD,	struct bch_member, flags[0], 14, 15)
+LE64_BITMASK(BCH_MEMBER_DISCARD,	struct bch_member, flags, 14, 15)
-LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED,	struct bch_member, flags[0], 15, 20)
+LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED,	struct bch_member, flags, 15, 20)
-LE64_BITMASK(BCH_MEMBER_GROUP,		struct bch_member, flags[0], 20, 28)
+LE64_BITMASK(BCH_MEMBER_GROUP,		struct bch_member, flags, 20, 28)
-LE64_BITMASK(BCH_MEMBER_DURABILITY,	struct bch_member, flags[0], 28, 30)
+LE64_BITMASK(BCH_MEMBER_DURABILITY,	struct bch_member, flags, 28, 30)
 LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED,
-					struct bch_member, flags[0], 30, 31)
+					struct bch_member, flags, 30, 31)
 
 #if 0
 LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS,	struct bch_member, flags[1], 0,  20);
@@ -1306,9 +1323,16 @@ enum bch_member_state {
 	BCH_MEMBER_STATE_NR
 };
 
-struct bch_sb_field_members {
+struct bch_sb_field_members_v1 {
 	struct bch_sb_field	field;
-	struct bch_member	members[];
+	struct bch_member	_members[]; //Members are now variable size
+};
+
+struct bch_sb_field_members_v2 {
+	struct bch_sb_field	field;
+	__le16			member_bytes; //size of single member entry
+	u8			pad[6];
+	struct bch_member	_members[];
 };
 
 /* BCH_SB_FIELD_crypt: */

libbcachefs/buckets.c

@@ -367,7 +367,6 @@ static inline int update_replicas(struct bch_fs *c, struct bkey_s_c k,
 	struct printbuf buf = PRINTBUF;
 
 	percpu_down_read(&c->mark_lock);
-	buf.atomic++;
 
 	idx = bch2_replicas_entry_idx(c, r);
 	if (idx < 0 &&
@@ -795,7 +794,6 @@ static int mark_stripe_bucket(struct btree_trans *trans,
 	/* * XXX doesn't handle deletion */
 
 	percpu_down_read(&c->mark_lock);
-	buf.atomic++;
 	g = PTR_GC_BUCKET(ca, ptr);
 
 	if (g->dirty_sectors ||

libbcachefs/buckets.h

@@ -70,12 +70,15 @@ union ulong_byte_assert {
 static inline void bucket_unlock(struct bucket *b)
 {
 	BUILD_BUG_ON(!((union ulong_byte_assert) { .ulong = 1UL << BUCKET_LOCK_BITNR }).byte);
-	bit_spin_unlock(BUCKET_LOCK_BITNR, (void *) &b->lock);
+
+	clear_bit_unlock(BUCKET_LOCK_BITNR, (void *) &b->lock);
+	wake_up_bit((void *) &b->lock, BUCKET_LOCK_BITNR);
 }
 
 static inline void bucket_lock(struct bucket *b)
 {
-	bit_spin_lock(BUCKET_LOCK_BITNR, (void *) &b->lock);
+	wait_on_bit_lock((void *) &b->lock, BUCKET_LOCK_BITNR,
+			 TASK_UNINTERRUPTIBLE);
 }
 
 static inline struct bucket_array *gc_bucket_array(struct bch_dev *ca)

libbcachefs/checksum.c

@@ -362,7 +362,7 @@ struct bch_csum bch2_checksum_merge(unsigned type, struct bch_csum a,
 
 	state.type = type;
 	bch2_checksum_init(&state);
-	state.seed = (u64 __force) a.lo;
+	state.seed = le64_to_cpu(a.lo);
 
 	BUG_ON(!bch2_checksum_mergeable(type));
 
@@ -373,7 +373,7 @@ struct bch_csum bch2_checksum_merge(unsigned type, struct bch_csum a,
 				page_address(ZERO_PAGE(0)), page_len);
 		b_len -= page_len;
 	}
-	a.lo = (__le64 __force) bch2_checksum_final(&state);
+	a.lo = cpu_to_le64(bch2_checksum_final(&state));
 	a.lo ^= b.lo;
 	a.hi ^= b.hi;
 	return a;
@@ -534,16 +534,31 @@ static int __bch2_request_key(char *key_description, struct bch_key *key)
 {
 	key_serial_t key_id;
 
+	key_id = request_key("user", key_description, NULL,
+			     KEY_SPEC_SESSION_KEYRING);
+	if (key_id >= 0)
+		goto got_key;
+
 	key_id = request_key("user", key_description, NULL,
 			     KEY_SPEC_USER_KEYRING);
-	if (key_id < 0)
+	if (key_id >= 0)
-		return -errno;
+		goto got_key;
+
+	key_id = request_key("user", key_description, NULL,
+			     KEY_SPEC_USER_SESSION_KEYRING);
+	if (key_id >= 0)
+		goto got_key;
+
+	return -errno;
+got_key:
 
 	if (keyctl_read(key_id, (void *) key, sizeof(*key)) != sizeof(*key))
 		return -1;
 
 	return 0;
 }
+
+#include "../crypto.h"
 #endif
 
 int bch2_request_key(struct bch_sb *sb, struct bch_key *key)
@@ -556,6 +571,20 @@ int bch2_request_key(struct bch_sb *sb, struct bch_key *key)
 
 	ret = __bch2_request_key(key_description.buf, key);
 	printbuf_exit(&key_description);
+
+#ifndef __KERNEL__
+	if (ret) {
+		char *passphrase = read_passphrase("Enter passphrase: ");
+		struct bch_encrypted_key sb_key;
+
+		bch2_passphrase_check(sb, passphrase,
+				      key, &sb_key);
+		ret = 0;
+	}
+#endif
+
+	/* stash with memfd, pass memfd fd to mount */
+
 	return ret;
 }
 

libbcachefs/disk_groups.c

@@ -25,19 +25,18 @@ static int bch2_sb_disk_groups_validate(struct bch_sb *sb,
 	struct bch_sb_field_disk_groups *groups =
 		field_to_type(f, disk_groups);
 	struct bch_disk_group *g, *sorted = NULL;
-	struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 	unsigned nr_groups = disk_groups_nr(groups);
 	unsigned i, len;
 	int ret = 0;
 
 	for (i = 0; i < sb->nr_devices; i++) {
-		struct bch_member *m = mi->members + i;
+		struct bch_member m = bch2_sb_member_get(sb, i);
 		unsigned group_id;
 
-		if (!BCH_MEMBER_GROUP(m))
+		if (!BCH_MEMBER_GROUP(&m))
 			continue;
 
-		group_id = BCH_MEMBER_GROUP(m) - 1;
+		group_id = BCH_MEMBER_GROUP(&m) - 1;
 
 		if (group_id >= nr_groups) {
 			prt_printf(err, "disk %u has invalid label %u (have %u)",
@@ -152,14 +151,12 @@ const struct bch_sb_field_ops bch_sb_field_ops_disk_groups = {
 
 int bch2_sb_disk_groups_to_cpu(struct bch_fs *c)
 {
-	struct bch_sb_field_members *mi;
 	struct bch_sb_field_disk_groups *groups;
 	struct bch_disk_groups_cpu *cpu_g, *old_g;
 	unsigned i, g, nr_groups;
 
 	lockdep_assert_held(&c->sb_lock);
 
-	mi		= bch2_sb_get_members(c->disk_sb.sb);
 	groups		= bch2_sb_get_disk_groups(c->disk_sb.sb);
 	nr_groups	= disk_groups_nr(groups);
 
@@ -182,13 +179,13 @@ int bch2_sb_disk_groups_to_cpu(struct bch_fs *c)
 	}
 
 	for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
-		struct bch_member *m = mi->members + i;
+		struct bch_member m = bch2_sb_member_get(c->disk_sb.sb, i);
 		struct bch_disk_group_cpu *dst;
 
-		if (!bch2_member_exists(m))
+		if (!bch2_member_exists(&m))
 			continue;
 
-		g = BCH_MEMBER_GROUP(m);
+		g = BCH_MEMBER_GROUP(&m);
 		while (g) {
 			dst = &cpu_g->entries[g - 1];
 			__set_bit(i, dst->devs.d);
@@ -443,7 +440,7 @@ int __bch2_dev_group_set(struct bch_fs *c, struct bch_dev *ca, const char *name)
 	if (ret)
 		return ret;
 
-	mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
+	mi = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
 	SET_BCH_MEMBER_GROUP(mi, v + 1);
 	return 0;
 }
@@ -528,12 +525,11 @@ void bch2_opt_target_to_text(struct printbuf *out,
 
 			rcu_read_unlock();
 		} else {
-			struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
+			struct bch_member m = bch2_sb_member_get(sb, t.dev);
-			struct bch_member *m = mi->members + t.dev;
 
-			if (bch2_dev_exists(sb, mi, t.dev)) {
+			if (bch2_dev_exists(sb, t.dev)) {
 				prt_printf(out, "Device ");
-				pr_uuid(out, m->uuid.b);
+				pr_uuid(out, m.uuid.b);
 				prt_printf(out, " (%u)", t.dev);
 			} else {
 				prt_printf(out, "Bad device %u", t.dev);

libbcachefs/errcode.h

@@ -91,6 +91,7 @@
 	x(ENOSPC,			ENOSPC_sb_quota)			\
 	x(ENOSPC,			ENOSPC_sb_replicas)			\
 	x(ENOSPC,			ENOSPC_sb_members)			\
+	x(ENOSPC,			ENOSPC_sb_members_v2)			\
 	x(ENOSPC,			ENOSPC_sb_crypt)			\
 	x(ENOSPC,			ENOSPC_btree_slot)			\
 	x(ENOSPC,			ENOSPC_snapshot_tree)			\

libbcachefs/fs.c

@@ -1595,7 +1595,7 @@ static struct bch_fs *bch2_path_to_fs(const char *path)
 static char **split_devs(const char *_dev_name, unsigned *nr)
 {
 	char *dev_name = NULL, **devs = NULL, *s;
-	size_t i, nr_devs = 0;
+	size_t i = 0, nr_devs = 0;
 
 	dev_name = kstrdup(_dev_name, GFP_KERNEL);
 	if (!dev_name)
@@ -1610,9 +1610,7 @@ static char **split_devs(const char *_dev_name, unsigned *nr)
 		return NULL;
 	}
 
-	for (i = 0, s = dev_name;
+	while ((s = strsep(&dev_name, ":")))
-	     s;
-	     (s = strchr(s, ':')) && (*s++ = '\0'))
 		devs[i++] = s;
 
 	*nr = nr_devs;

libbcachefs/inode.c

@@ -780,6 +780,7 @@ static int bch2_inode_delete_keys(struct btree_trans *trans,
 	struct btree_iter iter;
 	struct bkey_s_c k;
 	struct bkey_i delete;
+	struct bpos end = POS(inum.inum, U64_MAX);
 	u32 snapshot;
 	int ret = 0;
 
@@ -788,7 +789,7 @@ static int bch2_inode_delete_keys(struct btree_trans *trans,
 	 * extent iterator:
 	 */
 	bch2_trans_iter_init(trans, &iter, id, POS(inum.inum, 0),
-			     BTREE_ITER_INTENT|BTREE_ITER_NOT_EXTENTS);
+			     BTREE_ITER_INTENT);
 
 	while (1) {
 		bch2_trans_begin(trans);
@@ -799,7 +800,7 @@ static int bch2_inode_delete_keys(struct btree_trans *trans,
 
 		bch2_btree_iter_set_snapshot(&iter, snapshot);
 
-		k = bch2_btree_iter_peek_upto(&iter, POS(inum.inum, U64_MAX));
+		k = bch2_btree_iter_peek_upto(&iter, end);
 		ret = bkey_err(k);
 		if (ret)
 			goto err;
@@ -810,6 +811,11 @@ static int bch2_inode_delete_keys(struct btree_trans *trans,
 		bkey_init(&delete.k);
 		delete.k.p = iter.pos;
 
+		if (iter.flags & BTREE_ITER_IS_EXTENTS)
+			bch2_key_resize(&delete.k,
+					bpos_min(end, k.k->p).offset -
+					iter.pos.offset);
+
 		ret = bch2_trans_update(trans, &iter, &delete, 0) ?:
 		      bch2_trans_commit(trans, NULL, NULL,
 					BTREE_INSERT_NOFAIL);

libbcachefs/journal_sb.c

@@ -21,7 +21,7 @@ static int bch2_sb_journal_validate(struct bch_sb *sb,
 				    struct printbuf *err)
 {
 	struct bch_sb_field_journal *journal = field_to_type(f, journal);
-	struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
+	struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
 	int ret = -BCH_ERR_invalid_sb_journal;
 	unsigned nr;
 	unsigned i;
@@ -45,15 +45,15 @@ static int bch2_sb_journal_validate(struct bch_sb *sb,
 		goto err;
 	}
 
-	if (b[0] < le16_to_cpu(m->first_bucket)) {
+	if (b[0] < le16_to_cpu(m.first_bucket)) {
 		prt_printf(err, "journal bucket %llu before first bucket %u",
-		       b[0], le16_to_cpu(m->first_bucket));
+		       b[0], le16_to_cpu(m.first_bucket));
 		goto err;
 	}
 
-	if (b[nr - 1] >= le64_to_cpu(m->nbuckets)) {
+	if (b[nr - 1] >= le64_to_cpu(m.nbuckets)) {
 		prt_printf(err, "journal bucket %llu past end of device (nbuckets %llu)",
-		       b[nr - 1], le64_to_cpu(m->nbuckets));
+		       b[nr - 1], le64_to_cpu(m.nbuckets));
 		goto err;
 	}
 
@@ -104,7 +104,7 @@ static int bch2_sb_journal_v2_validate(struct bch_sb *sb,
 				    struct printbuf *err)
 {
 	struct bch_sb_field_journal_v2 *journal = field_to_type(f, journal_v2);
-	struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
+	struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
 	int ret = -BCH_ERR_invalid_sb_journal;
 	unsigned nr;
 	unsigned i;
@@ -130,15 +130,15 @@ static int bch2_sb_journal_v2_validate(struct bch_sb *sb,
 		goto err;
 	}
 
-	if (b[0].start < le16_to_cpu(m->first_bucket)) {
+	if (b[0].start < le16_to_cpu(m.first_bucket)) {
 		prt_printf(err, "journal bucket %llu before first bucket %u",
-		       b[0].start, le16_to_cpu(m->first_bucket));
+		       b[0].start, le16_to_cpu(m.first_bucket));
 		goto err;
 	}
 
-	if (b[nr - 1].end > le64_to_cpu(m->nbuckets)) {
+	if (b[nr - 1].end > le64_to_cpu(m.nbuckets)) {
 		prt_printf(err, "journal bucket %llu past end of device (nbuckets %llu)",
-		       b[nr - 1].end - 1, le64_to_cpu(m->nbuckets));
+		       b[nr - 1].end - 1, le64_to_cpu(m.nbuckets));
 		goto err;
 	}
 

libbcachefs/opts.c

@@ -12,6 +12,11 @@
 
 #define x(t, n, ...) [n] = #t,
 
+const char * const bch2_iops_measurements[] = {
+	BCH_IOPS_MEASUREMENTS()
+	NULL
+};
+
 const char * const bch2_error_actions[] = {
 	BCH_ERROR_ACTIONS()
 	NULL

libbcachefs/opts.h

@@ -10,6 +10,7 @@
 
 struct bch_fs;
 
+extern const char * const bch2_iops_measurements[];
 extern const char * const bch2_error_actions[];
 extern const char * const bch2_fsck_fix_opts[];
 extern const char * const bch2_version_upgrade_opts[];

libbcachefs/recovery.c

@@ -374,13 +374,12 @@ static int read_btree_roots(struct bch_fs *c)
 
 		ret = bch2_btree_root_read(c, i, &r->key, r->level);
 		if (ret) {
-			__fsck_err(c,
+			fsck_err(c,
-				   btree_id_is_alloc(i)
+				 "error reading btree root %s",
-				   ? FSCK_CAN_IGNORE : 0,
+				 bch2_btree_ids[i]);
-				   "error reading btree root %s",
-				   bch2_btree_ids[i]);
 			if (btree_id_is_alloc(i))
 				c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
+			ret = 0;
 		}
 	}
 
@@ -645,7 +644,7 @@ int bch2_fs_recovery(struct bch_fs *c)
 {
 	struct bch_sb_field_clean *clean = NULL;
 	struct jset *last_journal_entry = NULL;
-	u64 last_seq, blacklist_seq, journal_seq;
+	u64 last_seq = 0, blacklist_seq, journal_seq;
 	bool write_sb = false;
 	int ret = 0;
 

libbcachefs/replicas.c

@@ -805,7 +805,6 @@ static int bch2_cpu_replicas_validate(struct bch_replicas_cpu *cpu_r,
 				      struct bch_sb *sb,
 				      struct printbuf *err)
 {
-	struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 	unsigned i, j;
 
 	sort_cmp_size(cpu_r->entries,
@@ -837,7 +836,7 @@ static int bch2_cpu_replicas_validate(struct bch_replicas_cpu *cpu_r,
 		}
 
 		for (j = 0; j < e->nr_devs; j++)
-			if (!bch2_dev_exists(sb, mi, e->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;

libbcachefs/sb-members.c

@@ -2,172 +2,338 @@
 
 #include "bcachefs.h"
 #include "disk_groups.h"
+#include "opts.h"
 #include "replicas.h"
 #include "sb-members.h"
 #include "super-io.h"
 
-/* Code for bch_sb_field_members: */
+/* Code for bch_sb_field_members_v1: */
 
-static int bch2_sb_members_validate(struct bch_sb *sb,
+static struct bch_member *members_v2_get_mut(struct bch_sb_field_members_v2 *mi, int i)
+{
+	return (void *) mi->_members + (i * le16_to_cpu(mi->member_bytes));
+}
+
+struct bch_member *bch2_members_v2_get_mut(struct bch_sb *sb, int i)
+{
+	return members_v2_get_mut(bch2_sb_get_members_v2(sb), i);
+}
+
+static struct bch_member members_v2_get(struct bch_sb_field_members_v2 *mi, int i)
+{
+	struct bch_member ret, *p = members_v2_get_mut(mi, i);
+	memset(&ret, 0, sizeof(ret));
+	memcpy(&ret, p, min_t(size_t, le16_to_cpu(mi->member_bytes), sizeof(ret)));
+	return ret;
+}
+
+static struct bch_member *members_v1_get_mut(struct bch_sb_field_members_v1 *mi, int i)
+{
+	return (void *) mi->_members + (i * BCH_MEMBER_V1_BYTES);
+}
+
+static struct bch_member members_v1_get(struct bch_sb_field_members_v1 *mi, int i)
+{
+	struct bch_member ret, *p = members_v1_get_mut(mi, i);
+	memset(&ret, 0, sizeof(ret));
+	memcpy(&ret, p, min_t(size_t, sizeof(struct bch_member), sizeof(ret))); return ret;
+}
+
+struct bch_member bch2_sb_member_get(struct bch_sb *sb, int i)
+{
+	struct bch_sb_field_members_v2 *mi2 = bch2_sb_get_members_v2(sb);
+	if (mi2)
+		return members_v2_get(mi2, i);
+	struct bch_sb_field_members_v1 *mi1 = bch2_sb_get_members_v1(sb);
+	return members_v1_get(mi1, i);
+}
+
+static int sb_members_v2_resize_entries(struct bch_fs *c)
+{
+	struct bch_sb_field_members_v2 *mi = bch2_sb_get_members_v2(c->disk_sb.sb);
+
+	if (le16_to_cpu(mi->member_bytes) < sizeof(struct bch_member)) {
+		unsigned u64s = DIV_ROUND_UP((sizeof(*mi) + sizeof(mi->_members[0]) *
+					      c->disk_sb.sb->nr_devices), 8);
+
+		mi = bch2_sb_resize_members_v2(&c->disk_sb, u64s);
+		if (!mi)
+			return -BCH_ERR_ENOSPC_sb_members_v2;
+
+		for (int i = c->disk_sb.sb->nr_devices - 1; i >= 0; --i) {
+			void *dst = (void *) mi->_members + (i * sizeof(struct bch_member));
+			memmove(dst, members_v2_get_mut(mi, i), le16_to_cpu(mi->member_bytes));
+			memset(dst + le16_to_cpu(mi->member_bytes),
+			       0, (sizeof(struct bch_member) - le16_to_cpu(mi->member_bytes)));
+		}
+		mi->member_bytes = cpu_to_le16(sizeof(struct bch_member));
+	}
+	return 0;
+}
+
+int bch2_members_v2_init(struct bch_fs *c)
+{
+	struct bch_sb_field_members_v1 *mi1;
+	struct bch_sb_field_members_v2 *mi2;
+
+	if (!bch2_sb_get_members_v2(c->disk_sb.sb)) {
+		mi2 = bch2_sb_resize_members_v2(&c->disk_sb,
+				DIV_ROUND_UP(sizeof(*mi2) +
+					     sizeof(struct bch_member) * c->sb.nr_devices,
+					     sizeof(u64)));
+		mi1 = bch2_sb_get_members_v1(c->disk_sb.sb);
+		memcpy(&mi2->_members[0], &mi1->_members[0],
+		       BCH_MEMBER_V1_BYTES * c->sb.nr_devices);
+		memset(&mi2->pad[0], 0, sizeof(mi2->pad));
+		mi2->member_bytes = cpu_to_le16(BCH_MEMBER_V1_BYTES);
+	}
+
+	return sb_members_v2_resize_entries(c);
+}
+
+int bch_members_cpy_v2_v1(struct bch_sb_handle *disk_sb)
+{
+	struct bch_sb_field_members_v1 *mi1;
+	struct bch_sb_field_members_v2 *mi2;
+
+	mi1 = bch2_sb_resize_members_v1(disk_sb,
+			DIV_ROUND_UP(sizeof(*mi1) + BCH_MEMBER_V1_BYTES *
+				     disk_sb->sb->nr_devices, sizeof(u64)));
+	if (!mi1)
+		return -BCH_ERR_ENOSPC_sb_members;
+
+	mi2 = bch2_sb_get_members_v2(disk_sb->sb);
+
+	for (unsigned i = 0; i < disk_sb->sb->nr_devices; i++)
+		memcpy(members_v1_get_mut(mi1, i), members_v2_get_mut(mi2, i), BCH_MEMBER_V1_BYTES);
+
+	return 0;
+}
+
+static int validate_member(struct printbuf *err,
+			   struct bch_member m,
+			   struct bch_sb *sb,
+			   int i)
+{
+	if (le64_to_cpu(m.nbuckets) > LONG_MAX) {
+		prt_printf(err, "device %u: too many buckets (got %llu, max %lu)",
+			   i, le64_to_cpu(m.nbuckets), LONG_MAX);
+		return -BCH_ERR_invalid_sb_members;
+	}
+
+	if (le64_to_cpu(m.nbuckets) -
+	    le16_to_cpu(m.first_bucket) < BCH_MIN_NR_NBUCKETS) {
+		prt_printf(err, "device %u: not enough buckets (got %llu, max %u)",
+			   i, le64_to_cpu(m.nbuckets), BCH_MIN_NR_NBUCKETS);
+		return -BCH_ERR_invalid_sb_members;
+	}
+
+	if (le16_to_cpu(m.bucket_size) <
+	    le16_to_cpu(sb->block_size)) {
+		prt_printf(err, "device %u: bucket size %u smaller than block size %u",
+			   i, le16_to_cpu(m.bucket_size), le16_to_cpu(sb->block_size));
+		return -BCH_ERR_invalid_sb_members;
+	}
+
+	if (le16_to_cpu(m.bucket_size) <
+	    BCH_SB_BTREE_NODE_SIZE(sb)) {
+		prt_printf(err, "device %u: bucket size %u smaller than btree node size %llu",
+			   i, le16_to_cpu(m.bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
+		return -BCH_ERR_invalid_sb_members;
+	}
+
+	return 0;
+}
+
+static void member_to_text(struct printbuf *out,
+			   struct bch_member m,
+			   struct bch_sb_field_disk_groups *gi,
+			   struct bch_sb *sb,
+			   int i)
+{
+	unsigned data_have = bch2_sb_dev_has_data(sb, i);
+	u64 bucket_size = le16_to_cpu(m.bucket_size);
+	u64 device_size = le64_to_cpu(m.nbuckets) * bucket_size;
+
+
+	prt_printf(out, "Device:");
+	prt_tab(out);
+	prt_printf(out, "%u", i);
+	prt_newline(out);
+
+	printbuf_indent_add(out, 2);
+
+	prt_printf(out, "UUID:");
+	prt_tab(out);
+	pr_uuid(out, m.uuid.b);
+	prt_newline(out);
+
+	prt_printf(out, "Size:");
+	prt_tab(out);
+	prt_units_u64(out, device_size << 9);
+	prt_newline(out);
+
+	for (unsigned i = 0; i < BCH_IOPS_NR; i++) {
+		prt_printf(out, "%s iops:", bch2_iops_measurements[i]);
+		prt_tab(out);
+		prt_printf(out, "%u", le32_to_cpu(m.iops[i]));
+		prt_newline(out);
+	}
+
+	prt_printf(out, "Bucket size:");
+	prt_tab(out);
+	prt_units_u64(out, bucket_size << 9);
+	prt_newline(out);
+
+	prt_printf(out, "First bucket:");
+	prt_tab(out);
+	prt_printf(out, "%u", le16_to_cpu(m.first_bucket));
+	prt_newline(out);
+
+	prt_printf(out, "Buckets:");
+	prt_tab(out);
+	prt_printf(out, "%llu", le64_to_cpu(m.nbuckets));
+	prt_newline(out);
+
+	prt_printf(out, "Last mount:");
+	prt_tab(out);
+	if (m.last_mount)
+		pr_time(out, le64_to_cpu(m.last_mount));
+	else
+		prt_printf(out, "(never)");
+	prt_newline(out);
+
+	prt_printf(out, "State:");
+	prt_tab(out);
+	prt_printf(out, "%s",
+		   BCH_MEMBER_STATE(&m) < BCH_MEMBER_STATE_NR
+		   ? bch2_member_states[BCH_MEMBER_STATE(&m)]
+		   : "unknown");
+	prt_newline(out);
+
+	prt_printf(out, "Label:");
+	prt_tab(out);
+	if (BCH_MEMBER_GROUP(&m)) {
+		unsigned idx = BCH_MEMBER_GROUP(&m) - 1;
+
+		if (idx < disk_groups_nr(gi))
+			prt_printf(out, "%s (%u)",
+				   gi->entries[idx].label, idx);
+		else
+			prt_printf(out, "(bad disk labels section)");
+	} else {
+		prt_printf(out, "(none)");
+	}
+	prt_newline(out);
+
+	prt_printf(out, "Data allowed:");
+	prt_tab(out);
+	if (BCH_MEMBER_DATA_ALLOWED(&m))
+		prt_bitflags(out, bch2_data_types, BCH_MEMBER_DATA_ALLOWED(&m));
+	else
+		prt_printf(out, "(none)");
+	prt_newline(out);
+
+	prt_printf(out, "Has data:");
+	prt_tab(out);
+	if (data_have)
+		prt_bitflags(out, bch2_data_types, data_have);
+	else
+		prt_printf(out, "(none)");
+	prt_newline(out);
+
+	prt_printf(out, "Discard:");
+	prt_tab(out);
+	prt_printf(out, "%llu", BCH_MEMBER_DISCARD(&m));
+	prt_newline(out);
+
+	prt_printf(out, "Freespace initialized:");
+	prt_tab(out);
+	prt_printf(out, "%llu", BCH_MEMBER_FREESPACE_INITIALIZED(&m));
+	prt_newline(out);
+
+	printbuf_indent_sub(out, 2);
+}
+
+static int bch2_sb_members_v1_validate(struct bch_sb *sb,
 				    struct bch_sb_field *f,
 				    struct printbuf *err)
 {
-	struct bch_sb_field_members *mi = field_to_type(f, members);
+	struct bch_sb_field_members_v1 *mi = field_to_type(f, members_v1);
 	unsigned i;
 
-	if ((void *) (mi->members + sb->nr_devices) >
+	if ((void *) members_v1_get_mut(mi, sb->nr_devices)  >
 	    vstruct_end(&mi->field)) {
 		prt_printf(err, "too many devices for section size");
 		return -BCH_ERR_invalid_sb_members;
 	}
 
 	for (i = 0; i < sb->nr_devices; i++) {
-		struct bch_member *m = mi->members + i;
+		struct bch_member m = members_v1_get(mi, i);
 
-		if (!bch2_member_exists(m))
+		int ret = validate_member(err, m, sb, i);
-			continue;
+		if (ret)
-
+			return ret;
-		if (le64_to_cpu(m->nbuckets) > LONG_MAX) {
-			prt_printf(err, "device %u: too many buckets (got %llu, max %lu)",
-			       i, le64_to_cpu(m->nbuckets), LONG_MAX);
-			return -BCH_ERR_invalid_sb_members;
-		}
-
-		if (le64_to_cpu(m->nbuckets) -
-		    le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS) {
-			prt_printf(err, "device %u: not enough buckets (got %llu, max %u)",
-			       i, le64_to_cpu(m->nbuckets), BCH_MIN_NR_NBUCKETS);
-			return -BCH_ERR_invalid_sb_members;
-		}
-
-		if (le16_to_cpu(m->bucket_size) <
-		    le16_to_cpu(sb->block_size)) {
-			prt_printf(err, "device %u: bucket size %u smaller than block size %u",
-			       i, le16_to_cpu(m->bucket_size), le16_to_cpu(sb->block_size));
-			return -BCH_ERR_invalid_sb_members;
-		}
-
-		if (le16_to_cpu(m->bucket_size) <
-		    BCH_SB_BTREE_NODE_SIZE(sb)) {
-			prt_printf(err, "device %u: bucket size %u smaller than btree node size %llu",
-			       i, le16_to_cpu(m->bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
-			return -BCH_ERR_invalid_sb_members;
-		}
 	}
 
 	return 0;
 }
 
-static void bch2_sb_members_to_text(struct printbuf *out, struct bch_sb *sb,
+static void bch2_sb_members_v1_to_text(struct printbuf *out, struct bch_sb *sb,
-				    struct bch_sb_field *f)
+				       struct bch_sb_field *f)
 {
-	struct bch_sb_field_members *mi = field_to_type(f, members);
+	struct bch_sb_field_members_v1 *mi = field_to_type(f, members_v1);
 	struct bch_sb_field_disk_groups *gi = bch2_sb_get_disk_groups(sb);
 	unsigned i;
 
 	for (i = 0; i < sb->nr_devices; i++) {
-		struct bch_member *m = mi->members + i;
+		struct bch_member m = members_v1_get(mi, i);
-		unsigned data_have = bch2_sb_dev_has_data(sb, i);
+		member_to_text(out, m, gi, sb, i);
-		u64 bucket_size = le16_to_cpu(m->bucket_size);
-		u64 device_size = le64_to_cpu(m->nbuckets) * bucket_size;
-
-		if (!bch2_member_exists(m))
-			continue;
-
-		prt_printf(out, "Device:");
-		prt_tab(out);
-		prt_printf(out, "%u", i);
-		prt_newline(out);
-
-		printbuf_indent_add(out, 2);
-
-		prt_printf(out, "UUID:");
-		prt_tab(out);
-		pr_uuid(out, m->uuid.b);
-		prt_newline(out);
-
-		prt_printf(out, "Size:");
-		prt_tab(out);
-		prt_units_u64(out, device_size << 9);
-		prt_newline(out);
-
-		prt_printf(out, "Bucket size:");
-		prt_tab(out);
-		prt_units_u64(out, bucket_size << 9);
-		prt_newline(out);
-
-		prt_printf(out, "First bucket:");
-		prt_tab(out);
-		prt_printf(out, "%u", le16_to_cpu(m->first_bucket));
-		prt_newline(out);
-
-		prt_printf(out, "Buckets:");
-		prt_tab(out);
-		prt_printf(out, "%llu", le64_to_cpu(m->nbuckets));
-		prt_newline(out);
-
-		prt_printf(out, "Last mount:");
-		prt_tab(out);
-		if (m->last_mount)
-			pr_time(out, le64_to_cpu(m->last_mount));
-		else
-			prt_printf(out, "(never)");
-		prt_newline(out);
-
-		prt_printf(out, "State:");
-		prt_tab(out);
-		prt_printf(out, "%s",
-		       BCH_MEMBER_STATE(m) < BCH_MEMBER_STATE_NR
-		       ? bch2_member_states[BCH_MEMBER_STATE(m)]
-		       : "unknown");
-		prt_newline(out);
-
-		prt_printf(out, "Label:");
-		prt_tab(out);
-		if (BCH_MEMBER_GROUP(m)) {
-			unsigned idx = BCH_MEMBER_GROUP(m) - 1;
-
-			if (idx < disk_groups_nr(gi))
-				prt_printf(out, "%s (%u)",
-				       gi->entries[idx].label, idx);
-			else
-				prt_printf(out, "(bad disk labels section)");
-		} else {
-			prt_printf(out, "(none)");
-		}
-		prt_newline(out);
-
-		prt_printf(out, "Data allowed:");
-		prt_tab(out);
-		if (BCH_MEMBER_DATA_ALLOWED(m))
-			prt_bitflags(out, bch2_data_types, BCH_MEMBER_DATA_ALLOWED(m));
-		else
-			prt_printf(out, "(none)");
-		prt_newline(out);
-
-		prt_printf(out, "Has data:");
-		prt_tab(out);
-		if (data_have)
-			prt_bitflags(out, bch2_data_types, data_have);
-		else
-			prt_printf(out, "(none)");
-		prt_newline(out);
-
-		prt_printf(out, "Discard:");
-		prt_tab(out);
-		prt_printf(out, "%llu", BCH_MEMBER_DISCARD(m));
-		prt_newline(out);
-
-		prt_printf(out, "Freespace initialized:");
-		prt_tab(out);
-		prt_printf(out, "%llu", BCH_MEMBER_FREESPACE_INITIALIZED(m));
-		prt_newline(out);
-
-		printbuf_indent_sub(out, 2);
 	}
 }
 
-const struct bch_sb_field_ops bch_sb_field_ops_members = {
+const struct bch_sb_field_ops bch_sb_field_ops_members_v1 = {
-	.validate	= bch2_sb_members_validate,
+	.validate	= bch2_sb_members_v1_validate,
-	.to_text	= bch2_sb_members_to_text,
+	.to_text	= bch2_sb_members_v1_to_text,
+};
+
+static void bch2_sb_members_v2_to_text(struct printbuf *out, struct bch_sb *sb,
+				       struct bch_sb_field *f)
+{
+	struct bch_sb_field_members_v2 *mi = field_to_type(f, members_v2);
+	struct bch_sb_field_disk_groups *gi = bch2_sb_get_disk_groups(sb);
+	unsigned i;
+
+	for (i = 0; i < sb->nr_devices; i++) {
+		struct bch_member m = members_v2_get(mi, i);
+		member_to_text(out, m, gi, sb, i);
+	}
+}
+
+static int bch2_sb_members_v2_validate(struct bch_sb *sb,
+				       struct bch_sb_field *f,
+				       struct printbuf *err)
+{
+	struct bch_sb_field_members_v2 *mi = field_to_type(f, members_v2);
+	size_t mi_bytes = (void *) members_v2_get_mut(mi, sb->nr_devices) -
+		(void *) mi;
+
+	if (mi_bytes > vstruct_bytes(&mi->field)) {
+		prt_printf(err, "section too small (%zu > %zu)",
+			   mi_bytes, vstruct_bytes(&mi->field));
+		return -BCH_ERR_invalid_sb_members;
+	}
+
+	for (unsigned i = 0; i < sb->nr_devices; i++) {
+		int ret = validate_member(err, members_v2_get(mi, i), sb, i);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+const struct bch_sb_field_ops bch_sb_field_ops_members_v2 = {
+	.validate	= bch2_sb_members_v2_validate,
+	.to_text	= bch2_sb_members_v2_to_text,
 };

libbcachefs/sb-members.h

@@ -2,6 +2,11 @@
 #ifndef _BCACHEFS_SB_MEMBERS_H
 #define _BCACHEFS_SB_MEMBERS_H
 
+int bch2_members_v2_init(struct bch_fs *c);
+int bch_members_cpy_v2_v1(struct bch_sb_handle *disk_sb);
+struct bch_member *bch2_members_v2_get_mut(struct bch_sb *sb, int i);
+struct bch_member bch2_sb_member_get(struct bch_sb *sb, int i);
+
 static inline bool bch2_dev_is_online(struct bch_dev *ca)
 {
 	return !percpu_ref_is_zero(&ca->io_ref);
@@ -171,6 +176,7 @@ static inline struct bch_devs_mask bch2_online_devs(struct bch_fs *c)
 	return devs;
 }
 
-extern const struct bch_sb_field_ops bch_sb_field_ops_members;
+extern const struct bch_sb_field_ops bch_sb_field_ops_members_v1;
+extern const struct bch_sb_field_ops bch_sb_field_ops_members_v2;
 
 #endif /* _BCACHEFS_SB_MEMBERS_H */

libbcachefs/super-io.c

@@ -355,7 +355,7 @@ static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out,
 {
 	struct bch_sb *sb = disk_sb->sb;
 	struct bch_sb_field *f;
-	struct bch_sb_field_members *mi;
+	struct bch_sb_field_members_v1 *mi;
 	enum bch_opt_id opt_id;
 	u16 block_size;
 	int ret;
@@ -458,7 +458,7 @@ static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out,
 	}
 
 	/* members must be validated first: */
-	mi = bch2_sb_get_members(sb);
+	mi = bch2_sb_get_members_v1(sb);
 	if (!mi) {
 		prt_printf(out, "Invalid superblock: member info area missing");
 		return -BCH_ERR_invalid_sb_members_missing;
@@ -469,7 +469,7 @@ static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out,
 		return ret;
 
 	vstruct_for_each(sb, f) {
-		if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
+		if (le32_to_cpu(f->type) == BCH_SB_FIELD_members_v1)
 			continue;
 
 		ret = bch2_sb_field_validate(sb, f, out);
@@ -485,7 +485,6 @@ static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out,
 static void bch2_sb_update(struct bch_fs *c)
 {
 	struct bch_sb *src = c->disk_sb.sb;
-	struct bch_sb_field_members *mi = bch2_sb_get_members(src);
 	struct bch_dev *ca;
 	unsigned i;
 
@@ -511,8 +510,10 @@ static void bch2_sb_update(struct bch_fs *c)
 	c->sb.features		= le64_to_cpu(src->features[0]);
 	c->sb.compat		= le64_to_cpu(src->compat[0]);
 
-	for_each_member_device(ca, c, i)
+	for_each_member_device(ca, c, i) {
-		ca->mi = bch2_mi_to_cpu(mi->members + i);
+		struct bch_member m = bch2_sb_member_get(src, i);
+		ca->mi = bch2_mi_to_cpu(&m);
+	}
 }
 
 static int __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
@@ -891,6 +892,7 @@ int bch2_write_super(struct bch_fs *c)
 	SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
 
 	bch2_sb_counters_from_cpu(c);
+	bch_members_cpy_v2_v1(&c->disk_sb);
 
 	for_each_online_member(ca, c, i)
 		bch2_sb_from_fs(c, ca);
@@ -1125,7 +1127,6 @@ void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
 		     bool print_layout, unsigned fields)
 {
-	struct bch_sb_field_members *mi;
 	struct bch_sb_field *f;
 	u64 fields_have = 0;
 	unsigned nr_devices = 0;
@@ -1133,15 +1134,8 @@ void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
 	if (!out->nr_tabstops)
 		printbuf_tabstop_push(out, 44);
 
-	mi = bch2_sb_get_members(sb);
+	for (int i = 0; i < sb->nr_devices; i++)
-	if (mi) {
+		nr_devices += bch2_dev_exists(sb, i);
-		struct bch_member *m;
-
-		for (m = mi->members;
-		     m < mi->members + sb->nr_devices;
-		     m++)
-			nr_devices += bch2_member_exists(m);
-	}
 
 	prt_printf(out, "External UUID:");
 	prt_tab(out);

libbcachefs/super-io.h

@@ -6,6 +6,7 @@
 #include "eytzinger.h"
 #include "super_types.h"
 #include "super.h"
+#include "sb-members.h"
 
 #include <asm/byteorder.h>
 
@@ -89,7 +90,7 @@ static inline void bch2_check_set_feature(struct bch_fs *c, unsigned feat)
 		__bch2_check_set_feature(c, feat);
 }
 
-/* BCH_SB_FIELD_members: */
+/* BCH_SB_FIELD_members_v1: */
 
 static inline bool bch2_member_exists(struct bch_member *m)
 {
@@ -97,11 +98,13 @@ static inline bool bch2_member_exists(struct bch_member *m)
 }
 
 static inline bool bch2_dev_exists(struct bch_sb *sb,
-				   struct bch_sb_field_members *mi,
 				   unsigned dev)
 {
-	return dev < sb->nr_devices &&
+	if (dev < sb->nr_devices) {
-		bch2_member_exists(&mi->members[dev]);
+	struct bch_member m = bch2_sb_member_get(sb, dev);
+		return bch2_member_exists(&m);
+	}
+	return false;
 }
 
 static inline struct bch_member_cpu bch2_mi_to_cpu(struct bch_member *mi)

libbcachefs/super.c

@@ -49,6 +49,7 @@
 #include "recovery.h"
 #include "replicas.h"
 #include "sb-clean.h"
+#include "sb-members.h"
 #include "snapshot.h"
 #include "subvolume.h"
 #include "super.h"
@@ -399,6 +400,10 @@ static int __bch2_fs_read_write(struct bch_fs *c, bool early)
 
 	bch_info(c, "going read-write");
 
+	ret = bch2_members_v2_init(c);
+	if (ret)
+		goto err;
+
 	ret = bch2_fs_mark_dirty(c);
 	if (ret)
 		goto err;
@@ -662,7 +667,6 @@ err:
 
 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 {
-	struct bch_sb_field_members *mi;
 	struct bch_fs *c;
 	struct printbuf name = PRINTBUF;
 	unsigned i, iter_size;
@@ -858,9 +862,8 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 	if (ret)
 		goto err;
 
-	mi = bch2_sb_get_members(c->disk_sb.sb);
 	for (i = 0; i < c->sb.nr_devices; i++)
-		if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
+		if (bch2_dev_exists(c->disk_sb.sb, i) &&
 		    bch2_dev_alloc(c, i)) {
 			ret = -EEXIST;
 			goto err;
@@ -925,7 +928,6 @@ static void print_mount_opts(struct bch_fs *c)
 
 int bch2_fs_start(struct bch_fs *c)
 {
-	struct bch_sb_field_members *mi;
 	struct bch_dev *ca;
 	time64_t now = ktime_get_real_seconds();
 	unsigned i;
@@ -939,12 +941,17 @@ int bch2_fs_start(struct bch_fs *c)
 
 	mutex_lock(&c->sb_lock);
 
+	ret = bch2_members_v2_init(c);
+	if (ret) {
+		mutex_unlock(&c->sb_lock);
+		goto err;
+	}
+
 	for_each_online_member(ca, c, i)
 		bch2_sb_from_fs(c, ca);
 
-	mi = bch2_sb_get_members(c->disk_sb.sb);
 	for_each_online_member(ca, c, i)
-		mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
+		bch2_members_v2_get_mut(c->disk_sb.sb, i)->last_mount = cpu_to_le64(now);
 
 	mutex_unlock(&c->sb_lock);
 
@@ -997,16 +1004,12 @@ err:
 
 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
 {
-	struct bch_sb_field_members *sb_mi;
+	struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
-
-	sb_mi = bch2_sb_get_members(sb);
-	if (!sb_mi)
-		return -BCH_ERR_member_info_missing;
 
 	if (le16_to_cpu(sb->block_size) != block_sectors(c))
 		return -BCH_ERR_mismatched_block_size;
 
-	if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
+	if (le16_to_cpu(m.bucket_size) <
 	    BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
 		return -BCH_ERR_bucket_size_too_small;
 
@@ -1017,12 +1020,11 @@ static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
 {
 	struct bch_sb *newest =
 		le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
-	struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
 
 	if (!uuid_equal(&fs->uuid, &sb->uuid))
 		return -BCH_ERR_device_not_a_member_of_filesystem;
 
-	if (!bch2_dev_exists(newest, mi, sb->dev_idx))
+	if (!bch2_dev_exists(newest, sb->dev_idx))
 		return -BCH_ERR_device_has_been_removed;
 
 	if (fs->block_size != sb->block_size)
@@ -1192,15 +1194,14 @@ static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
 
 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
 {
-	struct bch_member *member =
+	struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
-		bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
 	struct bch_dev *ca = NULL;
 	int ret = 0;
 
 	if (bch2_fs_init_fault("dev_alloc"))
 		goto err;
 
-	ca = __bch2_dev_alloc(c, member);
+	ca = __bch2_dev_alloc(c, &member);
 	if (!ca)
 		goto err;
 
@@ -1335,7 +1336,6 @@ bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
 
 static bool bch2_fs_may_start(struct bch_fs *c)
 {
-	struct bch_sb_field_members *mi;
 	struct bch_dev *ca;
 	unsigned i, flags = 0;
 
@@ -1348,10 +1348,9 @@ static bool bch2_fs_may_start(struct bch_fs *c)
 	if (!c->opts.degraded &&
 	    !c->opts.very_degraded) {
 		mutex_lock(&c->sb_lock);
-		mi = bch2_sb_get_members(c->disk_sb.sb);
 
 		for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
-			if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
+			if (!bch2_dev_exists(c->disk_sb.sb, i))
 				continue;
 
 			ca = bch_dev_locked(c, i);
@@ -1391,7 +1390,7 @@ static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
 			 enum bch_member_state new_state, int flags)
 {
-	struct bch_sb_field_members *mi;
+	struct bch_member *m;
 	int ret = 0;
 
 	if (ca->mi.state == new_state)
@@ -1406,8 +1405,8 @@ int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
 	bch_notice(ca, "%s", bch2_member_states[new_state]);
 
 	mutex_lock(&c->sb_lock);
-	mi = bch2_sb_get_members(c->disk_sb.sb);
+	m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
-	SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
+	SET_BCH_MEMBER_STATE(m, new_state);
 	bch2_write_super(c);
 	mutex_unlock(&c->sb_lock);
 
@@ -1463,7 +1462,7 @@ static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
 
 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
 {
-	struct bch_sb_field_members *mi;
+	struct bch_member *m;
 	unsigned dev_idx = ca->dev_idx, data;
 	int ret;
 
@@ -1551,8 +1550,8 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
 	 * this device must be gone:
 	 */
 	mutex_lock(&c->sb_lock);
-	mi = bch2_sb_get_members(c->disk_sb.sb);
+	m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
-	memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
+	memset(&m->uuid, 0, sizeof(m->uuid));
 
 	bch2_write_super(c);
 
@@ -1575,7 +1574,7 @@ int bch2_dev_add(struct bch_fs *c, const char *path)
 	struct bch_opts opts = bch2_opts_empty();
 	struct bch_sb_handle sb;
 	struct bch_dev *ca = NULL;
-	struct bch_sb_field_members *mi;
+	struct bch_sb_field_members_v2 *mi;
 	struct bch_member dev_mi;
 	unsigned dev_idx, nr_devices, u64s;
 	struct printbuf errbuf = PRINTBUF;
@@ -1588,7 +1587,7 @@ int bch2_dev_add(struct bch_fs *c, const char *path)
 		goto err;
 	}
 
-	dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
+	dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
 
 	if (BCH_MEMBER_GROUP(&dev_mi)) {
 		bch2_disk_path_to_text(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
@@ -1631,9 +1630,9 @@ int bch2_dev_add(struct bch_fs *c, const char *path)
 		goto err_unlock;
 	}
 
-	mi = bch2_sb_get_members(ca->disk_sb.sb);
+	mi = bch2_sb_get_members_v2(ca->disk_sb.sb);
 
-	if (!bch2_sb_resize_members(&ca->disk_sb,
+	if (!bch2_sb_resize_members_v2(&ca->disk_sb,
 				le32_to_cpu(mi->field.u64s) +
 				sizeof(dev_mi) / sizeof(u64))) {
 		ret = -BCH_ERR_ENOSPC_sb_members;
@@ -1644,9 +1643,8 @@ int bch2_dev_add(struct bch_fs *c, const char *path)
 	if (dynamic_fault("bcachefs:add:no_slot"))
 		goto no_slot;
 
-	mi = bch2_sb_get_members(c->disk_sb.sb);
 	for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
-		if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
+		if (!bch2_dev_exists(c->disk_sb.sb, dev_idx))
 			goto have_slot;
 no_slot:
 	ret = -BCH_ERR_ENOSPC_sb_members;
@@ -1655,20 +1653,21 @@ no_slot:
 
 have_slot:
 	nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
-	u64s = (sizeof(struct bch_sb_field_members) +
+	u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
-		sizeof(struct bch_member) * nr_devices) / sizeof(u64);
+			    le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
 
-	mi = bch2_sb_resize_members(&c->disk_sb, u64s);
+	mi = bch2_sb_resize_members_v2(&c->disk_sb, u64s);
 	if (!mi) {
 		ret = -BCH_ERR_ENOSPC_sb_members;
 		bch_err_msg(c, ret, "setting up new superblock");
 		goto err_unlock;
 	}
+	struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
 
 	/* success: */
 
-	mi->members[dev_idx] = dev_mi;
+	*m = dev_mi;
-	mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
+	m->last_mount = cpu_to_le64(ktime_get_real_seconds());
 	c->disk_sb.sb->nr_devices	= nr_devices;
 
 	ca->disk_sb.sb->dev_idx	= dev_idx;
@@ -1728,7 +1727,6 @@ int bch2_dev_online(struct bch_fs *c, const char *path)
 {
 	struct bch_opts opts = bch2_opts_empty();
 	struct bch_sb_handle sb = { NULL };
-	struct bch_sb_field_members *mi;
 	struct bch_dev *ca;
 	unsigned dev_idx;
 	int ret;
@@ -1765,9 +1763,9 @@ int bch2_dev_online(struct bch_fs *c, const char *path)
 		__bch2_dev_read_write(c, ca);
 
 	mutex_lock(&c->sb_lock);
-	mi = bch2_sb_get_members(c->disk_sb.sb);
+	struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
 
-	mi->members[ca->dev_idx].last_mount =
+	m->last_mount =
 		cpu_to_le64(ktime_get_real_seconds());
 
 	bch2_write_super(c);
@@ -1809,10 +1807,12 @@ int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
 
 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
 {
-	struct bch_member *mi;
+	struct bch_member *m;
+	u64 old_nbuckets;
 	int ret = 0;
 
 	down_write(&c->state_lock);
+	old_nbuckets = ca->mi.nbuckets;
 
 	if (nbuckets < ca->mi.nbuckets) {
 		bch_err(ca, "Cannot shrink yet");
@@ -1839,12 +1839,24 @@ int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
 		goto err;
 
 	mutex_lock(&c->sb_lock);
-	mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
+	m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
-	mi->nbuckets = cpu_to_le64(nbuckets);
+	m->nbuckets = cpu_to_le64(nbuckets);
 
 	bch2_write_super(c);
 	mutex_unlock(&c->sb_lock);
 
+	if (ca->mi.freespace_initialized) {
+		ret = bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
+		if (ret)
+			goto err;
+
+		/*
+		 * XXX: this is all wrong transactionally - we'll be able to do
+		 * this correctly after the disk space accounting rewrite
+		 */
+		ca->usage_base->d[BCH_DATA_free].buckets += nbuckets - old_nbuckets;
+	}
+
 	bch2_recalc_capacity(c);
 err:
 	up_write(&c->state_lock);
@@ -1875,7 +1887,6 @@ struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
 {
 	struct bch_sb_handle *sb = NULL;
 	struct bch_fs *c = NULL;
-	struct bch_sb_field_members *mi;
 	unsigned i, best_sb = 0;
 	struct printbuf errbuf = PRINTBUF;
 	int ret = 0;
@@ -1906,12 +1917,10 @@ struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
 		    le64_to_cpu(sb[best_sb].sb->seq))
 			best_sb = i;
 
-	mi = bch2_sb_get_members(sb[best_sb].sb);
-
 	i = 0;
 	while (i < nr_devices) {
 		if (i != best_sb &&
-		    !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
+		    !bch2_dev_exists(sb[best_sb].sb, sb[i].sb->dev_idx)) {
 			pr_info("%pg has been removed, skipping", sb[i].bdev);
 			bch2_free_super(&sb[i]);
 			array_remove_item(sb, nr_devices, i);

libbcachefs/sysfs.c

@@ -962,7 +962,7 @@ STORE(bch2_dev)
 		bool v = strtoul_or_return(buf);
 
 		mutex_lock(&c->sb_lock);
-		mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
+		mi = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
 
 		if (v != BCH_MEMBER_DISCARD(mi)) {
 			SET_BCH_MEMBER_DISCARD(mi, v);
@@ -975,7 +975,7 @@ STORE(bch2_dev)
 		u64 v = strtoul_or_return(buf);
 
 		mutex_lock(&c->sb_lock);
-		mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
+		mi = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
 
 		if (v + 1 != BCH_MEMBER_DURABILITY(mi)) {
 			SET_BCH_MEMBER_DURABILITY(mi, v + 1);