Update bcachefs sources to b6d5426551 bcachefs: Fix bch2_btree_iter_peek_with_updates()

.bcachefs_revision

@@ -1 +1 @@
-a5c0e1bb306e79b40b2432a22f164697c8b22110
+b6d54265513ceb7532e3983163338c1ad28a0284

libbcachefs/alloc_background.c

@@ -371,7 +371,7 @@ int bch2_alloc_write(struct bch_fs *c, unsigned flags)
 
 			ret = bch2_alloc_write_key(&trans, iter, flags);
 			if (ret) {
-				percpu_ref_put(&ca->io_ref);
+				percpu_ref_put(&ca->ref);
 				goto err;
 			}
 			bch2_btree_iter_next_slot(iter);

libbcachefs/btree_gc.c

@@ -1020,7 +1020,7 @@ static void bch2_gc_free(struct bch_fs *c)
 static int bch2_gc_done(struct bch_fs *c,
 			bool initial, bool metadata_only)
 {
-	struct bch_dev *ca;
+	struct bch_dev *ca = NULL;
 	bool verify = !metadata_only && (!initial ||
 		       (c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)));
 	unsigned i, dev;
@@ -1166,6 +1166,8 @@ static int bch2_gc_done(struct bch_fs *c,
 #undef copy_stripe_field
 #undef copy_field
 fsck_err:
+	if (ca)
+		percpu_ref_put(&ca->ref);
 	if (ret)
 		bch_err(c, "%s: ret %i", __func__, ret);
 	return ret;
@@ -1174,7 +1176,7 @@ fsck_err:
 static int bch2_gc_start(struct bch_fs *c,
 			 bool metadata_only)
 {
-	struct bch_dev *ca;
+	struct bch_dev *ca = NULL;
 	unsigned i;
 	int ret;
 

libbcachefs/btree_io.c

@@ -1491,6 +1491,9 @@ void __bch2_btree_node_write(struct bch_fs *c, struct btree *b)
 	/* bch2_varint_decode may read up to 7 bytes past the end of the buffer: */
 	bytes += 8;
 
+	/* buffer must be a multiple of the block size */
+	bytes = round_up(bytes, block_bytes(c));
+
 	data = btree_bounce_alloc(c, bytes, &used_mempool);
 
 	if (!b->written) {

libbcachefs/btree_iter.c

@@ -1611,16 +1611,17 @@ static struct bkey_i *btree_trans_peek_updates(struct btree_trans *trans,
 static inline struct bkey_s_c __btree_iter_peek(struct btree_iter *iter, bool with_updates)
 {
 	struct bpos search_key = btree_iter_search_key(iter);
-	struct bkey_i *next_update = with_updates
+	struct bkey_i *next_update;
-		? btree_trans_peek_updates(iter->trans, iter->btree_id, search_key)
-		: NULL;
 	struct bkey_s_c k;
 	int ret;
 
 	EBUG_ON(btree_iter_type(iter) != BTREE_ITER_KEYS);
 	bch2_btree_iter_verify(iter);
 	bch2_btree_iter_verify_entry_exit(iter);
-
+start:
+	next_update = with_updates
+		? btree_trans_peek_updates(iter->trans, iter->btree_id, search_key)
+		: NULL;
 	btree_iter_set_search_pos(iter, search_key);
 
 	while (1) {
@@ -1636,9 +1637,8 @@ static inline struct bkey_s_c __btree_iter_peek(struct btree_iter *iter, bool wi
 
 		if (likely(k.k)) {
 			if (bkey_deleted(k.k)) {
-				btree_iter_set_search_pos(iter,
+				search_key = bkey_successor(iter, k.k->p);
-						bkey_successor(iter, k.k->p));
+				goto start;
-				continue;
 			}
 
 			break;

libbcachefs/buckets.c

@@ -1706,9 +1706,28 @@ static __le64 *bkey_refcount(struct bkey_i *k)
 	}
 }
 
+static bool reflink_p_frag_references(struct bkey_s_c_reflink_p p,
+				      u64 start, u64 end,
+				      struct bkey_s_c k)
+{
+	if (start == end)
+		return false;
+
+	start	+= le64_to_cpu(p.v->idx);
+	end	+= le64_to_cpu(p.v->idx);
+
+	if (end <= bkey_start_offset(k.k))
+		return false;
+	if (start >= k.k->p.offset)
+		return false;
+	return true;
+}
+
 static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
 			struct bkey_s_c_reflink_p p,
 			u64 idx, unsigned sectors,
+			unsigned front_frag,
+			unsigned back_frag,
 			unsigned flags)
 {
 	struct bch_fs *c = trans->c;
@@ -1716,6 +1735,7 @@ static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
 	struct bkey_s_c k;
 	struct bkey_i *n;
 	__le64 *refcount;
+	int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
 	s64 ret;
 
 	ret = trans_get_key(trans, BTREE_ID_reflink,
@@ -1723,12 +1743,17 @@ static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
 	if (ret < 0)
 		return ret;
 
-	if ((flags & BTREE_TRIGGER_OVERWRITE) &&
+	if (reflink_p_frag_references(p, 0, front_frag, k) &&
-	    (bkey_start_offset(k.k) < idx ||
+	    reflink_p_frag_references(p, back_frag, p.k->size, k)) {
-	     k.k->p.offset > idx + sectors))
+		BUG_ON(!(flags & BTREE_TRIGGER_OVERWRITE_SPLIT));
+		add = -add;
+	} else if (reflink_p_frag_references(p, 0, front_frag, k) ||
+		   reflink_p_frag_references(p, back_frag, p.k->size, k)) {
+		BUG_ON(!(flags & BTREE_TRIGGER_OVERWRITE));
 		goto out;
+	}
 
-	sectors = k.k->p.offset - idx;
+	sectors = min_t(u64, sectors, k.k->p.offset - idx);
 
 	n = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
 	ret = PTR_ERR_OR_ZERO(n);
@@ -1747,7 +1772,8 @@ static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
 		goto err;
 	}
 
-	le64_add_cpu(refcount, !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1);
+	BUG_ON(!*refcount && (flags & BTREE_TRIGGER_OVERWRITE));
+	le64_add_cpu(refcount, add);
 
 	if (!*refcount) {
 		n->k.type = KEY_TYPE_deleted;
@@ -1768,13 +1794,18 @@ static int bch2_trans_mark_reflink_p(struct btree_trans *trans,
 			s64 sectors, unsigned flags)
 {
 	u64 idx = le64_to_cpu(p.v->idx) + offset;
+	unsigned front_frag, back_frag;
 	s64 ret = 0;
 
 	sectors = abs(sectors);
 	BUG_ON(offset + sectors > p.k->size);
 
+	front_frag = offset;
+	back_frag = offset + sectors;
+
 	while (sectors) {
-		ret = __bch2_trans_mark_reflink_p(trans, p, idx, sectors, flags);
+		ret = __bch2_trans_mark_reflink_p(trans, p, idx, sectors,
+					front_frag, back_frag, flags);
 		if (ret < 0)
 			break;
 
@@ -2067,7 +2098,7 @@ static int __bch2_trans_mark_dev_sb(struct btree_trans *trans,
 
 int bch2_trans_mark_dev_sb(struct bch_fs *c, struct bch_dev *ca)
 {
-	return bch2_trans_do(c, NULL, NULL, 0,
+	return bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
 			__bch2_trans_mark_dev_sb(&trans, ca));
 }
 

libbcachefs/io.c

@@ -1967,7 +1967,10 @@ int __bch2_read_indirect_extent(struct btree_trans *trans,
 	if (k.k->type != KEY_TYPE_reflink_v &&
 	    k.k->type != KEY_TYPE_indirect_inline_data) {
 		bch_err_inum_ratelimited(trans->c, orig_k->k->k.p.inode,
-				"pointer to nonexistent indirect extent");
+			"%llu len %u points to nonexistent indirect extent %llu",
+			orig_k->k->k.p.offset,
+			orig_k->k->k.size,
+			reflink_offset);
 		bch2_inconsistent_error(trans->c);
 		ret = -EIO;
 		goto err;

libbcachefs/journal.c

@@ -190,7 +190,8 @@ static bool __journal_entry_close(struct journal *j)
 	 * Hence, we want update/set last_seq on the current journal entry right
 	 * before we open a new one:
 	 */
-	buf->data->last_seq	= cpu_to_le64(journal_last_seq(j));
+	buf->last_seq		= journal_last_seq(j);
+	buf->data->last_seq	= cpu_to_le64(buf->last_seq);
 
 	__bch2_journal_pin_put(j, le64_to_cpu(buf->data->seq));
 

libbcachefs/journal_io.c

@@ -1237,7 +1237,7 @@ static void journal_write_done(struct closure *cl)
 		bch2_bkey_devs(bkey_i_to_s_c(&w->key));
 	struct bch_replicas_padded replicas;
 	union journal_res_state old, new;
-	u64 v, seq, last_seq;
+	u64 v, seq;
 	int err = 0;
 
 	bch2_time_stats_update(j->write_time, j->write_start_time);
@@ -1256,7 +1256,6 @@ static void journal_write_done(struct closure *cl)
 
 	spin_lock(&j->lock);
 	seq = le64_to_cpu(w->data->seq);
-	last_seq = le64_to_cpu(w->data->last_seq);
 
 	if (seq >= j->pin.front)
 		journal_seq_pin(j, seq)->devs = devs;
@@ -1267,7 +1266,7 @@ static void journal_write_done(struct closure *cl)
 
 	if (!JSET_NO_FLUSH(w->data)) {
 		j->flushed_seq_ondisk = seq;
-		j->last_seq_ondisk = last_seq;
+		j->last_seq_ondisk = w->last_seq;
 	}
 
 	/*
@@ -1403,7 +1402,7 @@ void bch2_journal_write(struct closure *cl)
 	    test_bit(JOURNAL_MAY_SKIP_FLUSH, &j->flags)) {
 		w->noflush = true;
 		SET_JSET_NO_FLUSH(jset, true);
-		jset->last_seq = 0;
+		jset->last_seq = w->last_seq = 0;
 
 		j->nr_noflush_writes++;
 	} else {

libbcachefs/journal_types.h

@@ -23,6 +23,7 @@ struct journal_buf {
 	__BKEY_PADDED(key, BCH_REPLICAS_MAX);
 
 	struct closure_waitlist	wait;
+	u64			last_seq;	/* copy of data->last_seq */
 
 	unsigned		buf_size;	/* size in bytes of @data */
 	unsigned		sectors;	/* maximum size for current entry */

libbcachefs/recovery.c

@@ -1328,9 +1328,11 @@ int bch2_fs_initialize(struct bch_fs *c)
 	err = "error marking superblock and journal";
 	for_each_member_device(ca, c, i) {
 		ret = bch2_trans_mark_dev_sb(c, ca);
-		if (ret)
+		if (ret) {
+			percpu_ref_put(&ca->ref);
 			goto err;
 		}
+	}
 
 	bch2_inode_init(c, &root_inode, 0, 0,
 			S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);

libbcachefs/s128.h

@@ -1,29 +0,0 @@
-/* SPDX-License-Identifier: GPL-2.0 */
-#ifndef _BCACHEFS_S128_H
-#define _BCACHEFS_S128_H
-
-#include <linux/math64.h>
-
-typedef struct {
-	s64	lo;
-	s64	hi;
-} s128;
-
-typedef struct {
-	s64	lo;
-	s32	hi;
-} s96;
-
-static inline s128 s128_mul(s128 a, s128 b)
-{
-	return a.lo
-
-}
-
-static inline s96 s96_mul(s96 a, s96 b)
-{
-	return a.lo
-
-}
-
-#endif /* _BCACHEFS_S128_H */

libbcachefs/super.c

@@ -627,9 +627,11 @@ static const char *bch2_fs_online(struct bch_fs *c)
 	down_write(&c->state_lock);
 
 	err = "error creating sysfs objects";
-	__for_each_member_device(ca, c, i, NULL)
+	for_each_member_device(ca, c, i)
-		if (bch2_dev_sysfs_online(c, ca))
+		if (bch2_dev_sysfs_online(c, ca)) {
+			percpu_ref_put(&ca->ref);
 			goto err;
+		}
 
 	list_add(&c->list, &bch_fs_list);
 	err = NULL;
@@ -1841,12 +1843,14 @@ struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
 	if (ret)
 		return ERR_PTR(ret);
 
-	for_each_member_device(ca, c, i)
+	rcu_read_lock();
+	for_each_member_device_rcu(ca, c, i, NULL)
 		if (ca->disk_sb.bdev->bd_dev == dev)
 			goto found;
-
 	ca = ERR_PTR(-ENOENT);
 found:
+	rcu_read_unlock();
+
 	return ca;
 }
 

libbcachefs/super.h

@@ -107,11 +107,8 @@ static inline struct bch_dev *__bch2_next_dev(struct bch_fs *c, unsigned *iter,
 	return ca;
 }
 
-#define __for_each_member_device(ca, c, iter, mask)			\
-	for ((iter) = 0; ((ca) = __bch2_next_dev((c), &(iter), mask)); (iter)++)
-
 #define for_each_member_device_rcu(ca, c, iter, mask)			\
-	__for_each_member_device(ca, c, iter, mask)
+	for ((iter) = 0; ((ca) = __bch2_next_dev((c), &(iter), mask)); (iter)++)
 
 static inline struct bch_dev *bch2_get_next_dev(struct bch_fs *c, unsigned *iter)
 {