Merge 6.9-rc5 into usb-next

We need the usb/thunderbolt fixes in here as well.

Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

160 files changed, 4205 insertions, 1923 deletions

diff --git a/fs/9p/fid.h b/fs/9p/fid.h
index 29281b7c3887..0d6138bee2a3 100644
--- a/fs/9p/fid.h
+++ b/fs/9p/fid.h
@@ -49,9 +49,6 @@ static inline struct p9_fid *v9fs_fid_clone(struct dentry *dentry)
 static inline void v9fs_fid_add_modes(struct p9_fid *fid, unsigned int s_flags,
 	unsigned int s_cache, unsigned int f_flags)
 {
-	if (fid->qid.type != P9_QTFILE)
-		return;
-
 	if ((!s_cache) ||
 	   ((fid->qid.version == 0) && !(s_flags & V9FS_IGNORE_QV)) ||
 	   (s_flags & V9FS_DIRECT_IO) || (f_flags & O_DIRECT)) {
diff --git a/fs/9p/vfs_file.c b/fs/9p/vfs_file.c
index abdbbaee5184..348cc90bf9c5 100644
--- a/fs/9p/vfs_file.c
+++ b/fs/9p/vfs_file.c
@@ -520,6 +520,7 @@ const struct file_operations v9fs_file_operations = {
 	.splice_read = v9fs_file_splice_read,
 	.splice_write = iter_file_splice_write,
 	.fsync = v9fs_file_fsync,
+	.setlease = simple_nosetlease,
 };
 
 const struct file_operations v9fs_file_operations_dotl = {
@@ -534,4 +535,5 @@ const struct file_operations v9fs_file_operations_dotl = {
 	.splice_read = v9fs_file_splice_read,
 	.splice_write = iter_file_splice_write,
 	.fsync = v9fs_file_fsync_dotl,
+	.setlease = simple_nosetlease,
 };
diff --git a/fs/9p/vfs_inode.c b/fs/9p/vfs_inode.c
index b01b1bbf2493..47bd77199e20 100644
--- a/fs/9p/vfs_inode.c
+++ b/fs/9p/vfs_inode.c
@@ -83,7 +83,7 @@ static int p9mode2perm(struct v9fs_session_info *v9ses,
 	int res;
 	int mode = stat->mode;
 
-	res = mode & S_IALLUGO;
+	res = mode & 0777; /* S_IRWXUGO */
 	if (v9fs_proto_dotu(v9ses)) {
 		if ((mode & P9_DMSETUID) == P9_DMSETUID)
 			res |= S_ISUID;
@@ -178,6 +178,9 @@ int v9fs_uflags2omode(int uflags, int extended)
 		break;
 	}
 
+	if (uflags & O_TRUNC)
+		ret |= P9_OTRUNC;
+
 	if (extended) {
 		if (uflags & O_EXCL)
 			ret |= P9_OEXCL;
@@ -1061,8 +1064,6 @@ v9fs_stat2inode(struct p9_wstat *stat, struct inode *inode,
 	struct v9fs_session_info *v9ses = sb->s_fs_info;
 	struct v9fs_inode *v9inode = V9FS_I(inode);
 
-	set_nlink(inode, 1);
-
 	inode_set_atime(inode, stat->atime, 0);
 	inode_set_mtime(inode, stat->mtime, 0);
 	inode_set_ctime(inode, stat->mtime, 0);
diff --git a/fs/9p/vfs_super.c b/fs/9p/vfs_super.c
index 4236058c7bbd..55e67e36ae68 100644
--- a/fs/9p/vfs_super.c
+++ b/fs/9p/vfs_super.c
@@ -244,6 +244,21 @@ done:
 	return res;
 }
 
+static int v9fs_drop_inode(struct inode *inode)
+{
+	struct v9fs_session_info *v9ses;
+
+	v9ses = v9fs_inode2v9ses(inode);
+	if (v9ses->cache & (CACHE_META|CACHE_LOOSE))
+		return generic_drop_inode(inode);
+	/*
+	 * in case of non cached mode always drop the
+	 * inode because we want the inode attribute
+	 * to always match that on the server.
+	 */
+	return 1;
+}
+
 static int v9fs_write_inode(struct inode *inode,
 			    struct writeback_control *wbc)
 {
@@ -268,6 +283,7 @@ static const struct super_operations v9fs_super_ops = {
 	.alloc_inode = v9fs_alloc_inode,
 	.free_inode = v9fs_free_inode,
 	.statfs = simple_statfs,
+	.drop_inode = v9fs_drop_inode,
 	.evict_inode = v9fs_evict_inode,
 	.show_options = v9fs_show_options,
 	.umount_begin = v9fs_umount_begin,
@@ -278,6 +294,7 @@ static const struct super_operations v9fs_super_ops_dotl = {
 	.alloc_inode = v9fs_alloc_inode,
 	.free_inode = v9fs_free_inode,
 	.statfs = v9fs_statfs,
+	.drop_inode = v9fs_drop_inode,
 	.evict_inode = v9fs_evict_inode,
 	.show_options = v9fs_show_options,
 	.umount_begin = v9fs_umount_begin,
diff --git a/fs/aio.c b/fs/aio.c
index 9cdaa2faa536..0f4f531c9780 100644
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -1202,8 +1202,8 @@ static void aio_complete(struct aio_kiocb *iocb)
 		spin_lock_irqsave(&ctx->wait.lock, flags);
 		list_for_each_entry_safe(curr, next, &ctx->wait.head, w.entry)
 			if (avail >= curr->min_nr) {
-				list_del_init_careful(&curr->w.entry);
 				wake_up_process(curr->w.private);
+				list_del_init_careful(&curr->w.entry);
 			}
 		spin_unlock_irqrestore(&ctx->wait.lock, flags);
 	}
diff --git a/fs/bcachefs/Makefile b/fs/bcachefs/Makefile
index b02796c8a595..66ca0bbee639 100644
--- a/fs/bcachefs/Makefile
+++ b/fs/bcachefs/Makefile
@@ -17,6 +17,7 @@ bcachefs-y		:=	\
 	btree_journal_iter.o	\
 	btree_key_cache.o	\
 	btree_locking.o		\
+	btree_node_scan.o	\
 	btree_trans_commit.o	\
 	btree_update.o		\
 	btree_update_interior.o	\
@@ -37,6 +38,7 @@ bcachefs-y		:=	\
 	error.o			\
 	extents.o		\
 	extent_update.o		\
+	eytzinger.o		\
 	fs.o			\
 	fs-common.o		\
 	fs-ioctl.o		\
@@ -67,6 +69,7 @@ bcachefs-y		:=	\
 	quota.o			\
 	rebalance.o		\
 	recovery.o		\
+	recovery_passes.o	\
 	reflink.o		\
 	replicas.o		\
 	sb-clean.o		\
diff --git a/fs/bcachefs/acl.c b/fs/bcachefs/acl.c
index 3640f417cce1..5c180fdc3efb 100644
--- a/fs/bcachefs/acl.c
+++ b/fs/bcachefs/acl.c
@@ -281,7 +281,6 @@ struct posix_acl *bch2_get_acl(struct mnt_idmap *idmap,
 	struct xattr_search_key search = X_SEARCH(acl_to_xattr_type(type), "", 0);
 	struct btree_trans *trans = bch2_trans_get(c);
 	struct btree_iter iter = { NULL };
-	struct bkey_s_c_xattr xattr;
 	struct posix_acl *acl = NULL;
 	struct bkey_s_c k;
 	int ret;
@@ -290,28 +289,27 @@ retry:
 
 	ret = bch2_hash_lookup(trans, &iter, bch2_xattr_hash_desc,
 			&hash, inode_inum(inode), &search, 0);
-	if (ret) {
-		if (!bch2_err_matches(ret, ENOENT))
-			acl = ERR_PTR(ret);
-		goto out;
-	}
+	if (ret)
+		goto err;
 
 	k = bch2_btree_iter_peek_slot(&iter);
 	ret = bkey_err(k);
-	if (ret) {
-		acl = ERR_PTR(ret);
-		goto out;
-	}
+	if (ret)
+		goto err;
 
-	xattr = bkey_s_c_to_xattr(k);
+	struct bkey_s_c_xattr xattr = bkey_s_c_to_xattr(k);
 	acl = bch2_acl_from_disk(trans, xattr_val(xattr.v),
-			le16_to_cpu(xattr.v->x_val_len));
+				 le16_to_cpu(xattr.v->x_val_len));
+	ret = PTR_ERR_OR_ZERO(acl);
+err:
+	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
+		goto retry;
 
-	if (!IS_ERR(acl))
+	if (ret)
+		acl = !bch2_err_matches(ret, ENOENT) ? ERR_PTR(ret) : NULL;
+
+	if (!IS_ERR_OR_NULL(acl))
 		set_cached_acl(&inode->v, type, acl);
-out:
-	if (bch2_err_matches(PTR_ERR_OR_ZERO(acl), BCH_ERR_transaction_restart))
-		goto retry;
 
 	bch2_trans_iter_exit(trans, &iter);
 	bch2_trans_put(trans);
diff --git a/fs/bcachefs/alloc_background.c b/fs/bcachefs/alloc_background.c
index 893e38f9db80..4ff56fa4d539 100644
--- a/fs/bcachefs/alloc_background.c
+++ b/fs/bcachefs/alloc_background.c
@@ -1713,34 +1713,37 @@ static int bch2_discard_one_bucket(struct btree_trans *trans,
 	if (ret)
 		goto out;
 
-	if (BCH_ALLOC_V4_NEED_INC_GEN(&a->v)) {
-		a->v.gen++;
-		SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
-		goto write;
-	}
-
-	if (a->v.journal_seq > c->journal.flushed_seq_ondisk) {
-		if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
-			bch2_trans_inconsistent(trans,
-				"clearing need_discard but journal_seq %llu > flushed_seq %llu\n"
-				"%s",
-				a->v.journal_seq,
-				c->journal.flushed_seq_ondisk,
-				(bch2_bkey_val_to_text(&buf, c, k), buf.buf));
+	if (a->v.dirty_sectors) {
+		if (bch2_trans_inconsistent_on(c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info,
+					       trans, "attempting to discard bucket with dirty data\n%s",
+					       (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
 			ret = -EIO;
-		}
 		goto out;
 	}
 
 	if (a->v.data_type != BCH_DATA_need_discard) {
-		if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
-			bch2_trans_inconsistent(trans,
-				"bucket incorrectly set in need_discard btree\n"
-				"%s",
-				(bch2_bkey_val_to_text(&buf, c, k), buf.buf));
-			ret = -EIO;
+		if (data_type_is_empty(a->v.data_type) &&
+		    BCH_ALLOC_V4_NEED_INC_GEN(&a->v)) {
+			a->v.gen++;
+			SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
+			goto write;
 		}
 
+		if (bch2_trans_inconsistent_on(c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info,
+					       trans, "bucket incorrectly set in need_discard btree\n"
+					       "%s",
+					       (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
+			ret = -EIO;
+		goto out;
+	}
+
+	if (a->v.journal_seq > c->journal.flushed_seq_ondisk) {
+		if (bch2_trans_inconsistent_on(c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info,
+					       trans, "clearing need_discard but journal_seq %llu > flushed_seq %llu\n%s",
+					       a->v.journal_seq,
+					       c->journal.flushed_seq_ondisk,
+					       (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
+			ret = -EIO;
 		goto out;
 	}
 
@@ -1835,6 +1838,7 @@ static int bch2_clear_bucket_needs_discard(struct btree_trans *trans, struct bpo
 	if (ret)
 		goto err;
 
+	BUG_ON(a->v.dirty_sectors);
 	SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false);
 	a->v.data_type = alloc_data_type(a->v, a->v.data_type);
 
@@ -1942,6 +1946,7 @@ static int invalidate_one_bucket(struct btree_trans *trans,
 		goto out;
 
 	BUG_ON(a->v.data_type != BCH_DATA_cached);
+	BUG_ON(a->v.dirty_sectors);
 
 	if (!a->v.cached_sectors)
 		bch_err(c, "invalidating empty bucket, confused");
diff --git a/fs/bcachefs/alloc_foreground.c b/fs/bcachefs/alloc_foreground.c
index 214b15c84d1f..a1fc30adf912 100644
--- a/fs/bcachefs/alloc_foreground.c
+++ b/fs/bcachefs/alloc_foreground.c
@@ -188,8 +188,10 @@ long bch2_bucket_alloc_new_fs(struct bch_dev *ca)
 static inline unsigned open_buckets_reserved(enum bch_watermark watermark)
 {
 	switch (watermark) {
-	case BCH_WATERMARK_reclaim:
+	case BCH_WATERMARK_interior_updates:
 		return 0;
+	case BCH_WATERMARK_reclaim:
+		return OPEN_BUCKETS_COUNT / 6;
 	case BCH_WATERMARK_btree:
 	case BCH_WATERMARK_btree_copygc:
 		return OPEN_BUCKETS_COUNT / 4;
diff --git a/fs/bcachefs/alloc_types.h b/fs/bcachefs/alloc_types.h
index b91b7a461056..c2226e947c41 100644
--- a/fs/bcachefs/alloc_types.h
+++ b/fs/bcachefs/alloc_types.h
@@ -22,7 +22,8 @@ struct bucket_alloc_state {
 	x(copygc)			\
 	x(btree)			\
 	x(btree_copygc)			\
-	x(reclaim)
+	x(reclaim)			\
+	x(interior_updates)
 
 enum bch_watermark {
 #define x(name)	BCH_WATERMARK_##name,
diff --git a/fs/bcachefs/backpointers.c b/fs/bcachefs/backpointers.c
index 8cb35ea572cb..fadb1078903d 100644
--- a/fs/bcachefs/backpointers.c
+++ b/fs/bcachefs/backpointers.c
@@ -8,6 +8,7 @@
 #include "btree_update.h"
 #include "btree_update_interior.h"
 #include "btree_write_buffer.h"
+#include "checksum.h"
 #include "error.h"
 
 #include <linux/mm.h>
@@ -29,8 +30,7 @@ static bool extent_matches_bp(struct bch_fs *c,
 		if (p.ptr.cached)
 			continue;
 
-		bch2_extent_ptr_to_bp(c, btree_id, level, k, p,
-				      &bucket2, &bp2);
+		bch2_extent_ptr_to_bp(c, btree_id, level, k, p, entry, &bucket2, &bp2);
 		if (bpos_eq(bucket, bucket2) &&
 		    !memcmp(&bp, &bp2, sizeof(bp)))
 			return true;
@@ -44,13 +44,20 @@ int bch2_backpointer_invalid(struct bch_fs *c, struct bkey_s_c k,
 			     struct printbuf *err)
 {
 	struct bkey_s_c_backpointer bp = bkey_s_c_to_backpointer(k);
+
+	/* these will be caught by fsck */
+	if (!bch2_dev_exists2(c, bp.k->p.inode))
+		return 0;
+
+	struct bch_dev *ca = bch_dev_bkey_exists(c, bp.k->p.inode);
 	struct bpos bucket = bp_pos_to_bucket(c, bp.k->p);
 	int ret = 0;
 
-	bkey_fsck_err_on(!bpos_eq(bp.k->p, bucket_pos_to_bp(c, bucket, bp.v->bucket_offset)),
+	bkey_fsck_err_on((bp.v->bucket_offset >> MAX_EXTENT_COMPRESS_RATIO_SHIFT) >= ca->mi.bucket_size ||
+			 !bpos_eq(bp.k->p, bucket_pos_to_bp(c, bucket, bp.v->bucket_offset)),
 			 c, err,
-			 backpointer_pos_wrong,
-			 "backpointer at wrong pos");
+			 backpointer_bucket_offset_wrong,
+			 "backpointer bucket_offset wrong");
 fsck_err:
 	return ret;
 }
@@ -378,7 +385,7 @@ static int bch2_check_btree_backpointer(struct btree_trans *trans, struct btree_
 			backpointer_to_missing_alloc,
 			"backpointer for nonexistent alloc key: %llu:%llu:0\n%s",
 			alloc_iter.pos.inode, alloc_iter.pos.offset,
-			(bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
+			(bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
 		ret = bch2_btree_delete_at(trans, bp_iter, 0);
 		goto out;
 	}
@@ -414,6 +421,84 @@ struct extents_to_bp_state {
 	struct bkey_buf last_flushed;
 };
 
+static int drop_dev_and_update(struct btree_trans *trans, enum btree_id btree,
+			       struct bkey_s_c extent, unsigned dev)
+{
+	struct bkey_i *n = bch2_bkey_make_mut_noupdate(trans, extent);
+	int ret = PTR_ERR_OR_ZERO(n);
+	if (ret)
+		return ret;
+
+	bch2_bkey_drop_device(bkey_i_to_s(n), dev);
+	return bch2_btree_insert_trans(trans, btree, n, 0);
+}
+
+static int check_extent_checksum(struct btree_trans *trans,
+				 enum btree_id btree, struct bkey_s_c extent,
+				 enum btree_id o_btree, struct bkey_s_c extent2, unsigned dev)
+{
+	struct bch_fs *c = trans->c;
+	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(extent);
+	const union bch_extent_entry *entry;
+	struct extent_ptr_decoded p;
+	struct printbuf buf = PRINTBUF;
+	void *data_buf = NULL;
+	struct bio *bio = NULL;
+	size_t bytes;
+	int ret = 0;
+
+	if (bkey_is_btree_ptr(extent.k))
+		return false;
+
+	bkey_for_each_ptr_decode(extent.k, ptrs, p, entry)
+		if (p.ptr.dev == dev)
+			goto found;
+	BUG();
+found:
+	if (!p.crc.csum_type)
+		return false;
+
+	bytes = p.crc.compressed_size << 9;
+
+	struct bch_dev *ca = bch_dev_bkey_exists(c, dev);
+	if (!bch2_dev_get_ioref(ca, READ))
+		return false;
+
+	data_buf = kvmalloc(bytes, GFP_KERNEL);
+	if (!data_buf) {
+		ret = -ENOMEM;
+		goto err;
+	}
+
+	bio = bio_alloc(ca->disk_sb.bdev, 1, REQ_OP_READ, GFP_KERNEL);
+	bio->bi_iter.bi_sector = p.ptr.offset;
+	bch2_bio_map(bio, data_buf, bytes);
+	ret = submit_bio_wait(bio);
+	if (ret)
+		goto err;
+
+	prt_str(&buf, "extents pointing to same space, but first extent checksum bad:");
+	prt_printf(&buf, "\n  %s ", bch2_btree_id_str(btree));
+	bch2_bkey_val_to_text(&buf, c, extent);
+	prt_printf(&buf, "\n  %s ", bch2_btree_id_str(o_btree));
+	bch2_bkey_val_to_text(&buf, c, extent2);
+
+	struct nonce nonce = extent_nonce(extent.k->version, p.crc);
+	struct bch_csum csum = bch2_checksum(c, p.crc.csum_type, nonce, data_buf, bytes);
+	if (fsck_err_on(bch2_crc_cmp(csum, p.crc.csum),
+			c, dup_backpointer_to_bad_csum_extent,
+			"%s", buf.buf))
+		ret = drop_dev_and_update(trans, btree, extent, dev) ?: 1;
+fsck_err:
+err:
+	if (bio)
+		bio_put(bio);
+	kvfree(data_buf);
+	percpu_ref_put(&ca->io_ref);
+	printbuf_exit(&buf);
+	return ret;
+}
+
 static int check_bp_exists(struct btree_trans *trans,
 			   struct extents_to_bp_state *s,
 			   struct bpos bucket,
@@ -421,7 +506,8 @@ static int check_bp_exists(struct btree_trans *trans,
 			   struct bkey_s_c orig_k)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter bp_iter = { NULL };
+	struct btree_iter bp_iter = {};
+	struct btree_iter other_extent_iter = {};
 	struct printbuf buf = PRINTBUF;
 	struct bkey_s_c bp_k;
 	struct bkey_buf tmp;
@@ -429,13 +515,19 @@ static int check_bp_exists(struct btree_trans *trans,
 
 	bch2_bkey_buf_init(&tmp);
 
+	if (!bch2_dev_bucket_exists(c, bucket)) {
+		prt_str(&buf, "extent for nonexistent device:bucket ");
+		bch2_bpos_to_text(&buf, bucket);
+		prt_str(&buf, "\n  ");
+		bch2_bkey_val_to_text(&buf, c, orig_k);
+		bch_err(c, "%s", buf.buf);
+		return -BCH_ERR_fsck_repair_unimplemented;
+	}
+
 	if (bpos_lt(bucket, s->bucket_start) ||
 	    bpos_gt(bucket, s->bucket_end))
 		return 0;
 
-	if (!bch2_dev_bucket_exists(c, bucket))
-		goto missing;
-
 	bp_k = bch2_bkey_get_iter(trans, &bp_iter, BTREE_ID_backpointers,
 				  bucket_pos_to_bp(c, bucket, bp.bucket_offset),
 				  0);
@@ -461,21 +553,94 @@ static int check_bp_exists(struct btree_trans *trans,
 			ret = -BCH_ERR_transaction_restart_write_buffer_flush;
 			goto out;
 		}
-		goto missing;
+
+		goto check_existing_bp;
 	}
 out:
 err:
 fsck_err:
+	bch2_trans_iter_exit(trans, &other_extent_iter);
 	bch2_trans_iter_exit(trans, &bp_iter);
 	bch2_bkey_buf_exit(&tmp, c);
 	printbuf_exit(&buf);
 	return ret;
+check_existing_bp:
+	/* Do we have a backpointer for a different extent? */
+	if (bp_k.k->type != KEY_TYPE_backpointer)
+		goto missing;
+
+	struct bch_backpointer other_bp = *bkey_s_c_to_backpointer(bp_k).v;
+
+	struct bkey_s_c other_extent =
+		bch2_backpointer_get_key(trans, &other_extent_iter, bp_k.k->p, other_bp, 0);
+	ret = bkey_err(other_extent);
+	if (ret == -BCH_ERR_backpointer_to_overwritten_btree_node)
+		ret = 0;
+	if (ret)
+		goto err;
+
+	if (!other_extent.k)
+		goto missing;
+
+	if (bch2_extents_match(orig_k, other_extent)) {
+		printbuf_reset(&buf);
+		prt_printf(&buf, "duplicate versions of same extent, deleting smaller\n  ");
+		bch2_bkey_val_to_text(&buf, c, orig_k);
+		prt_str(&buf, "\n  ");
+		bch2_bkey_val_to_text(&buf, c, other_extent);
+		bch_err(c, "%s", buf.buf);
+
+		if (other_extent.k->size <= orig_k.k->size) {
+			ret = drop_dev_and_update(trans, other_bp.btree_id, other_extent, bucket.inode);
+			if (ret)
+				goto err;
+			goto out;
+		} else {
+			ret = drop_dev_and_update(trans, bp.btree_id, orig_k, bucket.inode);
+			if (ret)
+				goto err;
+			goto missing;
+		}
+	}
+
+	ret = check_extent_checksum(trans, other_bp.btree_id, other_extent, bp.btree_id, orig_k, bucket.inode);
+	if (ret < 0)
+		goto err;
+	if (ret) {
+		ret = 0;
+		goto missing;
+	}
+
+	ret = check_extent_checksum(trans, bp.btree_id, orig_k, other_bp.btree_id, other_extent, bucket.inode);
+	if (ret < 0)
+		goto err;
+	if (ret) {
+		ret = 0;
+		goto out;
+	}
+
+	printbuf_reset(&buf);
+	prt_printf(&buf, "duplicate extents pointing to same space on dev %llu\n  ", bucket.inode);
+	bch2_bkey_val_to_text(&buf, c, orig_k);
+	prt_str(&buf, "\n  ");
+	bch2_bkey_val_to_text(&buf, c, other_extent);
+	bch_err(c, "%s", buf.buf);
+	ret = -BCH_ERR_fsck_repair_unimplemented;
+	goto err;
 missing:
+	printbuf_reset(&buf);
 	prt_printf(&buf, "missing backpointer for btree=%s l=%u ",
 	       bch2_btree_id_str(bp.btree_id), bp.level);
 	bch2_bkey_val_to_text(&buf, c, orig_k);
-	prt_printf(&buf, "\nbp pos ");
-	bch2_bpos_to_text(&buf, bp_iter.pos);
+	prt_printf(&buf, "\n  got:   ");
+	bch2_bkey_val_to_text(&buf, c, bp_k);
+
+	struct bkey_i_backpointer n_bp_k;
+	bkey_backpointer_init(&n_bp_k.k_i);
+	n_bp_k.k.p = bucket_pos_to_bp(trans->c, bucket, bp.bucket_offset);
+	n_bp_k.v = bp;
+	prt_printf(&buf, "\n  want:  ");
+	bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&n_bp_k.k_i));
 
 	if (fsck_err(c, ptr_to_missing_backpointer, "%s", buf.buf))
 		ret = bch2_bucket_backpointer_mod(trans, bucket, bp, orig_k, true);
@@ -502,8 +667,7 @@ static int check_extent_to_backpointers(struct btree_trans *trans,
 		if (p.ptr.cached)
 			continue;
 
-		bch2_extent_ptr_to_bp(c, btree, level,
-				      k, p, &bucket_pos, &bp);
+		bch2_extent_ptr_to_bp(c, btree, level, k, p, entry, &bucket_pos, &bp);
 
 		ret = check_bp_exists(trans, s, bucket_pos, bp, k);
 		if (ret)
diff --git a/fs/bcachefs/backpointers.h b/fs/bcachefs/backpointers.h
index 327365a9feac..85949b9fd880 100644
--- a/fs/bcachefs/backpointers.h
+++ b/fs/bcachefs/backpointers.h
@@ -53,14 +53,11 @@ static inline struct bpos bucket_pos_to_bp(const struct bch_fs *c,
 					   u64 bucket_offset)
 {
 	struct bch_dev *ca = bch_dev_bkey_exists(c, bucket.inode);
-	struct bpos ret;
-
-	ret = POS(bucket.inode,
-		  (bucket_to_sector(ca, bucket.offset) <<
-		   MAX_EXTENT_COMPRESS_RATIO_SHIFT) + bucket_offset);
+	struct bpos ret = POS(bucket.inode,
+			      (bucket_to_sector(ca, bucket.offset) <<
+			       MAX_EXTENT_COMPRESS_RATIO_SHIFT) + bucket_offset);
 
 	EBUG_ON(!bkey_eq(bucket, bp_pos_to_bucket(c, ret)));
-
 	return ret;
 }
 
@@ -90,20 +87,40 @@ static inline int bch2_bucket_backpointer_mod(struct btree_trans *trans,
 	return bch2_trans_update_buffered(trans, BTREE_ID_backpointers, &bp_k.k_i);
 }
 
-static inline enum bch_data_type bkey_ptr_data_type(enum btree_id btree_id, unsigned level,
-						    struct bkey_s_c k, struct extent_ptr_decoded p)
+static inline enum bch_data_type bch2_bkey_ptr_data_type(struct bkey_s_c k,
+							 struct extent_ptr_decoded p,
+							 const union bch_extent_entry *entry)
 {
-	return  level		? BCH_DATA_btree :
-		p.has_ec	? BCH_DATA_stripe :
-				  BCH_DATA_user;
+	switch (k.k->type) {
+	case KEY_TYPE_btree_ptr:
+	case KEY_TYPE_btree_ptr_v2:
+		return BCH_DATA_btree;
+	case KEY_TYPE_extent:
+	case KEY_TYPE_reflink_v:
+		return p.has_ec ? BCH_DATA_stripe : BCH_DATA_user;
+	case KEY_TYPE_stripe: {
+		const struct bch_extent_ptr *ptr = &entry->ptr;
+		struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
+
+		BUG_ON(ptr < s.v->ptrs ||
+		       ptr >= s.v->ptrs + s.v->nr_blocks);
+
+		return ptr >= s.v->ptrs + s.v->nr_blocks - s.v->nr_redundant
+			? BCH_DATA_parity
+			: BCH_DATA_user;
+	}
+	default:
+		BUG();
+	}
 }
 
 static inline void bch2_extent_ptr_to_bp(struct bch_fs *c,
 			   enum btree_id btree_id, unsigned level,
 			   struct bkey_s_c k, struct extent_ptr_decoded p,
+			   const union bch_extent_entry *entry,
 			   struct bpos *bucket_pos, struct bch_backpointer *bp)
 {
-	enum bch_data_type data_type = bkey_ptr_data_type(btree_id, level, k, p);
+	enum bch_data_type data_type = bch2_bkey_ptr_data_type(k, p, entry);
 	s64 sectors = level ? btree_sectors(c) : k.k->size;
 	u32 bucket_offset;
 
diff --git a/fs/bcachefs/bcachefs.h b/fs/bcachefs/bcachefs.h
index 799aa32b6b4d..91c3c1fef233 100644
--- a/fs/bcachefs/bcachefs.h
+++ b/fs/bcachefs/bcachefs.h
@@ -209,7 +209,7 @@
 #include "fifo.h"
 #include "nocow_locking_types.h"
 #include "opts.h"
-#include "recovery_types.h"
+#include "recovery_passes_types.h"
 #include "sb-errors_types.h"
 #include "seqmutex.h"
 #include "time_stats.h"
@@ -456,6 +456,7 @@ enum bch_time_stats {
 
 #include "alloc_types.h"
 #include "btree_types.h"
+#include "btree_node_scan_types.h"
 #include "btree_write_buffer_types.h"
 #include "buckets_types.h"
 #include "buckets_waiting_for_journal_types.h"
@@ -614,6 +615,7 @@ struct bch_dev {
  */
 
 #define BCH_FS_FLAGS()			\
+	x(new_fs)			\
 	x(started)			\
 	x(may_go_rw)			\
 	x(rw)				\
@@ -707,6 +709,8 @@ struct btree_trans_buf {
 	x(stripe_delete)						\
 	x(reflink)							\
 	x(fallocate)							\
+	x(fsync)							\
+	x(dio_write)							\
 	x(discard)							\
 	x(discard_fast)							\
 	x(invalidate)							\
@@ -796,6 +800,7 @@ struct bch_fs {
 		u64		features;
 		u64		compat;
 		unsigned long	errors_silent[BITS_TO_LONGS(BCH_SB_ERR_MAX)];
+		u64		btrees_lost_data;
 	}			sb;
 
 
@@ -810,7 +815,6 @@ struct bch_fs {
 
 	/* snapshot.c: */
 	struct snapshot_table __rcu *snapshots;
-	size_t			snapshot_table_size;
 	struct mutex		snapshot_table_lock;
 	struct rw_semaphore	snapshot_create_lock;
 
@@ -1104,6 +1108,8 @@ struct bch_fs {
 	struct journal_keys	journal_keys;
 	struct list_head	journal_iters;
 
+	struct find_btree_nodes	found_btree_nodes;
+
 	u64			last_bucket_seq_cleanup;
 
 	u64			counters_on_mount[BCH_COUNTER_NR];
diff --git a/fs/bcachefs/bcachefs_format.h b/fs/bcachefs/bcachefs_format.h
index bff8750ac0d7..085987435a5e 100644
--- a/fs/bcachefs/bcachefs_format.h
+++ b/fs/bcachefs/bcachefs_format.h
@@ -578,7 +578,8 @@ struct bch_member {
 	__le64			nbuckets;	/* device size */
 	__le16			first_bucket;   /* index of first bucket used */
 	__le16			bucket_size;	/* sectors */
-	__le32			pad;
+	__u8			btree_bitmap_shift;
+	__u8			pad[3];
 	__le64			last_mount;	/* time_t */
 
 	__le64			flags;
@@ -587,6 +588,7 @@ struct bch_member {
 	__le64			errors_at_reset[BCH_MEMBER_ERROR_NR];
 	__le64			errors_reset_time;
 	__le64			seq;
+	__le64			btree_allocated_bitmap;
 };
 
 #define BCH_MEMBER_V1_BYTES	56
@@ -818,6 +820,7 @@ struct bch_sb_field_ext {
 	struct bch_sb_field	field;
 	__le64			recovery_passes_required[2];
 	__le64			errors_silent[8];
+	__le64			btrees_lost_data;
 };
 
 struct bch_sb_field_downgrade_entry {
@@ -875,7 +878,8 @@ struct bch_sb_field_downgrade {
 	x(rebalance_work,		BCH_VERSION(1,  3))		\
 	x(member_seq,			BCH_VERSION(1,  4))		\
 	x(subvolume_fs_parent,		BCH_VERSION(1,  5))		\
-	x(btree_subvolume_children,	BCH_VERSION(1,  6))
+	x(btree_subvolume_children,	BCH_VERSION(1,  6))		\
+	x(mi_btree_bitmap,		BCH_VERSION(1,  7))
 
 enum bcachefs_metadata_version {
 	bcachefs_metadata_version_min = 9,
@@ -1313,7 +1317,7 @@ static inline __u64 __bset_magic(struct bch_sb *sb)
 	x(write_buffer_keys,	11)		\
 	x(datetime,		12)
 
-enum {
+enum bch_jset_entry_type {
 #define x(f, nr)	BCH_JSET_ENTRY_##f	= nr,
 	BCH_JSET_ENTRY_TYPES()
 #undef x
@@ -1359,7 +1363,7 @@ struct jset_entry_blacklist_v2 {
 	x(inodes,		1)		\
 	x(key_version,		2)
 
-enum {
+enum bch_fs_usage_type {
 #define x(f, nr)	BCH_FS_USAGE_##f	= nr,
 	BCH_FS_USAGE_TYPES()
 #undef x
@@ -1534,6 +1538,20 @@ enum btree_id {
 	BTREE_ID_NR
 };
 
+static inline bool btree_id_is_alloc(enum btree_id id)
+{
+	switch (id) {
+	case BTREE_ID_alloc:
+	case BTREE_ID_backpointers:
+	case BTREE_ID_need_discard:
+	case BTREE_ID_freespace:
+	case BTREE_ID_bucket_gens:
+		return true;
+	default:
+		return false;
+	}
+}
+
 #define BTREE_MAX_DEPTH		4U
 
 /* Btree nodes */
diff --git a/fs/bcachefs/bkey.h b/fs/bcachefs/bkey.h
index cf23ff47bed8..3a45d128f608 100644
--- a/fs/bcachefs/bkey.h
+++ b/fs/bcachefs/bkey.h
@@ -314,6 +314,12 @@ static inline unsigned bkeyp_key_u64s(const struct bkey_format *format,
 	return bkey_packed(k) ? format->key_u64s : BKEY_U64s;
 }
 
+static inline bool bkeyp_u64s_valid(const struct bkey_format *f,
+				    const struct bkey_packed *k)
+{
+	return ((unsigned) k->u64s - bkeyp_key_u64s(f, k) <= U8_MAX - BKEY_U64s);
+}
+
 static inline unsigned bkeyp_key_bytes(const struct bkey_format *format,
 				       const struct bkey_packed *k)
 {
diff --git a/fs/bcachefs/bkey_methods.c b/fs/bcachefs/bkey_methods.c
index 5e52684764eb..db336a43fc08 100644
--- a/fs/bcachefs/bkey_methods.c
+++ b/fs/bcachefs/bkey_methods.c
@@ -171,11 +171,15 @@ int __bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
 	if (type >= BKEY_TYPE_NR)
 		return 0;
 
-	bkey_fsck_err_on((flags & BKEY_INVALID_COMMIT) &&
+	bkey_fsck_err_on((type == BKEY_TYPE_btree ||
+			  (flags & BKEY_INVALID_COMMIT)) &&
 			 !(bch2_key_types_allowed[type] & BIT_ULL(k.k->type)), c, err,
 			 bkey_invalid_type_for_btree,
 			 "invalid key type for btree %s (%s)",
-			 bch2_btree_node_type_str(type), bch2_bkey_types[k.k->type]);
+			 bch2_btree_node_type_str(type),
+			 k.k->type < KEY_TYPE_MAX
+			 ? bch2_bkey_types[k.k->type]
+			 : "(unknown)");
 
 	if (btree_node_type_is_extents(type) && !bkey_whiteout(k.k)) {
 		bkey_fsck_err_on(k.k->size == 0, c, err,
diff --git a/fs/bcachefs/bset.c b/fs/bcachefs/bset.c
index 3fd1085b6c61..3bb477840eab 100644
--- a/fs/bcachefs/bset.c
+++ b/fs/bcachefs/bset.c
@@ -134,18 +134,24 @@ void bch2_dump_btree_node_iter(struct btree *b,
 	printbuf_exit(&buf);
 }
 
-#ifdef CONFIG_BCACHEFS_DEBUG
-
-void __bch2_verify_btree_nr_keys(struct btree *b)
+struct btree_nr_keys bch2_btree_node_count_keys(struct btree *b)
 {
 	struct bset_tree *t;
 	struct bkey_packed *k;
-	struct btree_nr_keys nr = { 0 };
+	struct btree_nr_keys nr = {};
 
 	for_each_bset(b, t)
 		bset_tree_for_each_key(b, t, k)
 			if (!bkey_deleted(k))
 				btree_keys_account_key_add(&nr, t - b->set, k);
+	return nr;
+}
+
+#ifdef CONFIG_BCACHEFS_DEBUG
+
+void __bch2_verify_btree_nr_keys(struct btree *b)
+{
+	struct btree_nr_keys nr = bch2_btree_node_count_keys(b);
 
 	BUG_ON(memcmp(&nr, &b->nr, sizeof(nr)));
 }
diff --git a/fs/bcachefs/bset.h b/fs/bcachefs/bset.h
index 79c77baaa383..120a79fd456b 100644
--- a/fs/bcachefs/bset.h
+++ b/fs/bcachefs/bset.h
@@ -458,6 +458,8 @@ struct bkey_s_c bch2_btree_node_iter_peek_unpack(struct btree_node_iter *,
 
 /* Accounting: */
 
+struct btree_nr_keys bch2_btree_node_count_keys(struct btree *);
+
 static inline void btree_keys_account_key(struct btree_nr_keys *n,
 					  unsigned bset,
 					  struct bkey_packed *k,
diff --git a/fs/bcachefs/btree_cache.c b/fs/bcachefs/btree_cache.c
index 562561a9a510..02c70e813fac 100644
--- a/fs/bcachefs/btree_cache.c
+++ b/fs/bcachefs/btree_cache.c
@@ -709,9 +709,31 @@ static noinline struct btree *bch2_btree_node_fill(struct btree_trans *trans,
 	struct bch_fs *c = trans->c;
 	struct btree_cache *bc = &c->btree_cache;
 	struct btree *b;
-	u32 seq;
 
-	BUG_ON(level + 1 >= BTREE_MAX_DEPTH);
+	if (unlikely(level >= BTREE_MAX_DEPTH)) {
+		int ret = bch2_fs_topology_error(c, "attempting to get btree node at level %u, >= max depth %u",
+						 level, BTREE_MAX_DEPTH);
+		return ERR_PTR(ret);
+	}
+
+	if (unlikely(!bkey_is_btree_ptr(&k->k))) {
+		struct printbuf buf = PRINTBUF;
+		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(k));
+
+		int ret = bch2_fs_topology_error(c, "attempting to get btree node with non-btree key %s", buf.buf);
+		printbuf_exit(&buf);
+		return ERR_PTR(ret);
+	}
+
+	if (unlikely(k->k.u64s > BKEY_BTREE_PTR_U64s_MAX)) {
+		struct printbuf buf = PRINTBUF;
+		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(k));
+
+		int ret = bch2_fs_topology_error(c, "attempting to get btree node with too big key %s", buf.buf);
+		printbuf_exit(&buf);
+		return ERR_PTR(ret);
+	}
+
 	/*
 	 * Parent node must be locked, else we could read in a btree node that's
 	 * been freed:
@@ -752,34 +774,26 @@ static noinline struct btree *bch2_btree_node_fill(struct btree_trans *trans,
 	}
 
 	set_btree_node_read_in_flight(b);
-
 	six_unlock_write(&b->c.lock);
-	seq = six_lock_seq(&b->c.lock);
-	six_unlock_intent(&b->c.lock);
 
-	/* Unlock before doing IO: */
-	if (path && sync)
-		bch2_trans_unlock_noassert(trans);
-
-	bch2_btree_node_read(trans, b, sync);
+	if (path) {
+		u32 seq = six_lock_seq(&b->c.lock);
 
-	if (!sync)
-		return NULL;
+		/* Unlock before doing IO: */
+		six_unlock_intent(&b->c.lock);
+		bch2_trans_unlock_noassert(trans);
 
-	if (path) {
-		int ret = bch2_trans_relock(trans) ?:
-			bch2_btree_path_relock_intent(trans, path);
-		if (ret) {
-			BUG_ON(!trans->restarted);
-			return ERR_PTR(ret);
-		}
-	}
+		bch2_btree_node_read(trans, b, sync);
 
-	if (!six_relock_type(&b->c.lock, lock_type, seq)) {
-		BUG_ON(!path);
+		if (!sync)
+			return NULL;
 
-		trace_and_count(c, trans_restart_relock_after_fill, trans, _THIS_IP_, path);
-		return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_relock_after_fill));
+		if (!six_relock_type(&b->c.lock, lock_type, seq))
+			b = NULL;
+	} else {
+		bch2_btree_node_read(trans, b, sync);
+		if (lock_type == SIX_LOCK_read)
+			six_lock_downgrade(&b->c.lock);
 	}
 
 	return b;
@@ -808,7 +822,8 @@ static noinline void btree_bad_header(struct bch_fs *c, struct btree *b)
 	prt_printf(&buf, "\nmax ");
 	bch2_bpos_to_text(&buf, b->data->max_key);
 
-	bch2_fs_inconsistent(c, "%s", buf.buf);
+	bch2_fs_topology_error(c, "%s", buf.buf);
+
 	printbuf_exit(&buf);
 }
 
@@ -1111,18 +1126,19 @@ int bch2_btree_node_prefetch(struct btree_trans *trans,
 {
 	struct bch_fs *c = trans->c;
 	struct btree_cache *bc = &c->btree_cache;
-	struct btree *b;
 
 	BUG_ON(path && !btree_node_locked(path, level + 1));
 	BUG_ON(level >= BTREE_MAX_DEPTH);
 
-	b = btree_cache_find(bc, k);
+	struct btree *b = btree_cache_find(bc, k);
 	if (b)
 		return 0;
 
 	b = bch2_btree_node_fill(trans, path, k, btree_id,
 				 level, SIX_LOCK_read, false);
-	return PTR_ERR_OR_ZERO(b);
+	if (!IS_ERR_OR_NULL(b))
+		six_unlock_read(&b->c.lock);
+	return bch2_trans_relock(trans) ?: PTR_ERR_OR_ZERO(b);
 }
 
 void bch2_btree_node_evict(struct btree_trans *trans, const struct bkey_i *k)
@@ -1134,6 +1150,8 @@ void bch2_btree_node_evict(struct btree_trans *trans, const struct bkey_i *k)
 	b = btree_cache_find(bc, k);
 	if (!b)
 		return;
+
+	BUG_ON(b == btree_node_root(trans->c, b));
 wait_on_io:
 	/* not allowed to wait on io with btree locks held: */
 
@@ -1145,6 +1163,8 @@ wait_on_io:
 
 	btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_intent);
 	btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_write);
+	if (unlikely(b->hash_val != btree_ptr_hash_val(k)))
+		goto out;
 
 	if (btree_node_dirty(b)) {
 		__bch2_btree_node_write(c, b, BTREE_WRITE_cache_reclaim);
@@ -1159,7 +1179,7 @@ wait_on_io:
 	btree_node_data_free(c, b);
 	bch2_btree_node_hash_remove(bc, b);
 	mutex_unlock(&bc->lock);
-
+out:
 	six_unlock_write(&b->c.lock);
 	six_unlock_intent(&b->c.lock);
 }
diff --git a/fs/bcachefs/btree_gc.c b/fs/bcachefs/btree_gc.c
index bdaed29f084a..ecbd9598f69f 100644
--- a/fs/bcachefs/btree_gc.c
+++ b/fs/bcachefs/btree_gc.c
@@ -7,11 +7,13 @@
 #include "bcachefs.h"
 #include "alloc_background.h"
 #include "alloc_foreground.h"
+#include "backpointers.h"
 #include "bkey_methods.h"
 #include "bkey_buf.h"
 #include "btree_journal_iter.h"
 #include "btree_key_cache.h"
 #include "btree_locking.h"
+#include "btree_node_scan.h"
 #include "btree_update_interior.h"
 #include "btree_io.h"
 #include "btree_gc.h"
@@ -24,7 +26,7 @@
 #include "journal.h"
 #include "keylist.h"
 #include "move.h"
-#include "recovery.h"
+#include "recovery_passes.h"
 #include "reflink.h"
 #include "replicas.h"
 #include "super-io.h"
@@ -40,6 +42,7 @@
 
 #define DROP_THIS_NODE		10
 #define DROP_PREV_NODE		11
+#define DID_FILL_FROM_SCAN	12
 
 static struct bkey_s unsafe_bkey_s_c_to_s(struct bkey_s_c k)
 {
@@ -70,90 +73,6 @@ static inline void gc_pos_set(struct bch_fs *c, struct gc_pos new_pos)
 	__gc_pos_set(c, new_pos);
 }
 
-/*
- * Missing: if an interior btree node is empty, we need to do something -
- * perhaps just kill it
- */
-static int bch2_gc_check_topology(struct bch_fs *c,
-				  struct btree *b,
-				  struct bkey_buf *prev,
-				  struct bkey_buf cur,
-				  bool is_last)
-{
-	struct bpos node_start	= b->data->min_key;
-	struct bpos node_end	= b->data->max_key;
-	struct bpos expected_start = bkey_deleted(&prev->k->k)
-		? node_start
-		: bpos_successor(prev->k->k.p);
-	struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
-	int ret = 0;
-
-	if (cur.k->k.type == KEY_TYPE_btree_ptr_v2) {
-		struct bkey_i_btree_ptr_v2 *bp = bkey_i_to_btree_ptr_v2(cur.k);
-
-		if (!bpos_eq(expected_start, bp->v.min_key)) {
-			bch2_topology_error(c);
-
-			if (bkey_deleted(&prev->k->k)) {
-				prt_printf(&buf1, "start of node: ");
-				bch2_bpos_to_text(&buf1, node_start);
-			} else {
-				bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(prev->k));
-			}
-			bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(cur.k));
-
-			if (__fsck_err(c,
-				       FSCK_CAN_FIX|
-				       FSCK_CAN_IGNORE|
-				       FSCK_NO_RATELIMIT,
-				       btree_node_topology_bad_min_key,
-				       "btree node with incorrect min_key at btree %s level %u:\n"
-				       "  prev %s\n"
-				       "  cur %s",
-				       bch2_btree_id_str(b->c.btree_id), b->c.level,
-				       buf1.buf, buf2.buf) && should_restart_for_topology_repair(c)) {
-				bch_info(c, "Halting mark and sweep to start topology repair pass");
-				ret = bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_topology);
-				goto err;
-			} else {
-				set_bit(BCH_FS_initial_gc_unfixed, &c->flags);
-			}
-		}
-	}
-
-	if (is_last && !bpos_eq(cur.k->k.p, node_end)) {
-		bch2_topology_error(c);
-
-		printbuf_reset(&buf1);
-		printbuf_reset(&buf2);
-
-		bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(cur.k));
-		bch2_bpos_to_text(&buf2, node_end);
-
-		if (__fsck_err(c, FSCK_CAN_FIX|FSCK_CAN_IGNORE|FSCK_NO_RATELIMIT,
-			  btree_node_topology_bad_max_key,
-			  "btree node with incorrect max_key at btree %s level %u:\n"
-			  "  %s\n"
-			  "  expected %s",
-			  bch2_btree_id_str(b->c.btree_id), b->c.level,
-			  buf1.buf, buf2.buf) &&
-		    should_restart_for_topology_repair(c)) {
-			bch_info(c, "Halting mark and sweep to start topology repair pass");
-			ret = bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_topology);
-			goto err;
-		} else {
-			set_bit(BCH_FS_initial_gc_unfixed, &c->flags);
-		}
-	}
-
-	bch2_bkey_buf_copy(prev, c, cur.k);
-err:
-fsck_err:
-	printbuf_exit(&buf2);
-	printbuf_exit(&buf1);
-	return ret;
-}
-
 static void btree_ptr_to_v2(struct btree *b, struct bkey_i_btree_ptr_v2 *dst)
 {
 	switch (b->key.k.type) {
@@ -212,6 +131,17 @@ static int set_node_min(struct bch_fs *c, struct btree *b, struct bpos new_min)
 	struct bkey_i_btree_ptr_v2 *new;
 	int ret;
 
+	if (c->opts.verbose) {
+		struct printbuf buf = PRINTBUF;
+
+		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
+		prt_str(&buf, " -> ");
+		bch2_bpos_to_text(&buf, new_min);
+
+		bch_info(c, "%s(): %s", __func__, buf.buf);
+		printbuf_exit(&buf);
+	}
+
 	new = kmalloc_array(BKEY_BTREE_PTR_U64s_MAX, sizeof(u64), GFP_KERNEL);
 	if (!new)
 		return -BCH_ERR_ENOMEM_gc_repair_key;
@@ -237,6 +167,17 @@ static int set_node_max(struct bch_fs *c, struct btree *b, struct bpos new_max)
 	struct bkey_i_btree_ptr_v2 *new;
 	int ret;
 
+	if (c->opts.verbose) {
+		struct printbuf buf = PRINTBUF;
+
+		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
+		prt_str(&buf, " -> ");
+		bch2_bpos_to_text(&buf, new_max);
+
+		bch_info(c, "%s(): %s", __func__, buf.buf);
+		printbuf_exit(&buf);
+	}
+
 	ret = bch2_journal_key_delete(c, b->c.btree_id, b->c.level + 1, b->key.k.p);
 	if (ret)
 		return ret;
@@ -268,127 +209,138 @@ static int set_node_max(struct bch_fs *c, struct btree *b, struct bpos new_max)
 	return 0;
 }
 
-static int btree_repair_node_boundaries(struct bch_fs *c, struct btree *b,
-					struct btree *prev, struct btree *cur)
+static int btree_check_node_boundaries(struct bch_fs *c, struct btree *b,
+				       struct btree *prev, struct btree *cur,
+				       struct bpos *pulled_from_scan)
 {
 	struct bpos expected_start = !prev
 		? b->data->min_key
 		: bpos_successor(prev->key.k.p);
-	struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
+	struct printbuf buf = PRINTBUF;
 	int ret = 0;
 
-	if (!prev) {
-		prt_printf(&buf1, "start of node: ");
-		bch2_bpos_to_text(&buf1, b->data->min_key);
-	} else {
-		bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(&prev->key));
+	BUG_ON(b->key.k.type == KEY_TYPE_btree_ptr_v2 &&
+	       !bpos_eq(bkey_i_to_btree_ptr_v2(&b->key)->v.min_key,
+			b->data->min_key));
+
+	if (bpos_eq(expected_start, cur->data->min_key))
+		return 0;
+
+	prt_printf(&buf, "  at btree %s level %u:\n  parent: ",
+		   bch2_btree_id_str(b->c.btree_id), b->c.level);
+	bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
+
+	if (prev) {
+		prt_printf(&buf, "\n  prev: ");
+		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&prev->key));
 	}
 
-	bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(&cur->key));
-
-	if (prev &&
-	    bpos_gt(expected_start, cur->data->min_key) &&
-	    BTREE_NODE_SEQ(cur->data) > BTREE_NODE_SEQ(prev->data)) {
-		/* cur overwrites prev: */
-
-		if (mustfix_fsck_err_on(bpos_ge(prev->data->min_key,
-						cur->data->min_key), c,
-				btree_node_topology_overwritten_by_next_node,
-				"btree node overwritten by next node at btree %s level %u:\n"
-				"  node %s\n"
-				"  next %s",
-				bch2_btree_id_str(b->c.btree_id), b->c.level,
-				buf1.buf, buf2.buf)) {
-			ret = DROP_PREV_NODE;
-			goto out;
-		}
+	prt_str(&buf, "\n  next: ");
+	bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&cur->key));
 
-		if (mustfix_fsck_err_on(!bpos_eq(prev->key.k.p,
-						 bpos_predecessor(cur->data->min_key)), c,
-				btree_node_topology_bad_max_key,
-				"btree node with incorrect max_key at btree %s level %u:\n"
-				"  node %s\n"
-				"  next %s",
-				bch2_btree_id_str(b->c.btree_id), b->c.level,
-				buf1.buf, buf2.buf))
-			ret = set_node_max(c, prev,
-					   bpos_predecessor(cur->data->min_key));
-	} else {
-		/* prev overwrites cur: */
-
-		if (mustfix_fsck_err_on(bpos_ge(expected_start,
-						cur->data->max_key), c,
-				btree_node_topology_overwritten_by_prev_node,
-				"btree node overwritten by prev node at btree %s level %u:\n"
-				"  prev %s\n"
-				"  node %s",
-				bch2_btree_id_str(b->c.btree_id), b->c.level,
-				buf1.buf, buf2.buf)) {
-			ret = DROP_THIS_NODE;
-			goto out;
-		}
+	if (bpos_lt(expected_start, cur->data->min_key)) {				/* gap */
+		if (b->c.level == 1 &&
+		    bpos_lt(*pulled_from_scan, cur->data->min_key)) {
+			ret = bch2_get_scanned_nodes(c, b->c.btree_id, 0,
+						     expected_start,
+						     bpos_predecessor(cur->data->min_key));
+			if (ret)
+				goto err;
 
-		if (mustfix_fsck_err_on(!bpos_eq(expected_start, cur->data->min_key), c,
-				btree_node_topology_bad_min_key,
-				"btree node with incorrect min_key at btree %s level %u:\n"
-				"  prev %s\n"
-				"  node %s",
-				bch2_btree_id_str(b->c.btree_id), b->c.level,
-				buf1.buf, buf2.buf))
-			ret = set_node_min(c, cur, expected_start);
+			*pulled_from_scan = cur->data->min_key;
+			ret = DID_FILL_FROM_SCAN;
+		} else {
+			if (mustfix_fsck_err(c, btree_node_topology_bad_min_key,
+					     "btree node with incorrect min_key%s", buf.buf))
+				ret = set_node_min(c, cur, expected_start);
+		}
+	} else {									/* overlap */
+		if (prev && BTREE_NODE_SEQ(cur->data) > BTREE_NODE_SEQ(prev->data)) {	/* cur overwrites prev */
+			if (bpos_ge(prev->data->min_key, cur->data->min_key)) {		/* fully? */
+				if (mustfix_fsck_err(c, btree_node_topology_overwritten_by_next_node,
+						     "btree node overwritten by next node%s", buf.buf))
+					ret = DROP_PREV_NODE;
+			} else {
+				if (mustfix_fsck_err(c, btree_node_topology_bad_max_key,
+						     "btree node with incorrect max_key%s", buf.buf))
+					ret = set_node_max(c, prev,
+							   bpos_predecessor(cur->data->min_key));
+			}
+		} else {
+			if (bpos_ge(expected_start, cur->data->max_key)) {		/* fully? */
+				if (mustfix_fsck_err(c, btree_node_topology_overwritten_by_prev_node,
+						     "btree node overwritten by prev node%s", buf.buf))
+					ret = DROP_THIS_NODE;
+			} else {
+				if (mustfix_fsck_err(c, btree_node_topology_bad_min_key,
+						     "btree node with incorrect min_key%s", buf.buf))
+					ret = set_node_min(c, cur, expected_start);
+			}
+		}
 	}
-out:
+err:
 fsck_err:
-	printbuf_exit(&buf2);
-	printbuf_exit(&buf1);
+	printbuf_exit(&buf);
 	return ret;
 }
 
 static int btree_repair_node_end(struct bch_fs *c, struct btree *b,
-				 struct btree *child)
+				 struct btree *child, struct bpos *pulled_from_scan)
 {
-	struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
+	struct printbuf buf = PRINTBUF;
 	int ret = 0;
 
-	bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(&child->key));
-	bch2_bpos_to_text(&buf2, b->key.k.p);
+	if (bpos_eq(child->key.k.p, b->key.k.p))
+		return 0;
 
-	if (mustfix_fsck_err_on(!bpos_eq(child->key.k.p, b->key.k.p), c,
-				btree_node_topology_bad_max_key,
-			"btree node with incorrect max_key at btree %s level %u:\n"
-			"  %s\n"
-			"  expected %s",
-			bch2_btree_id_str(b->c.btree_id), b->c.level,
-			buf1.buf, buf2.buf)) {
-		ret = set_node_max(c, child, b->key.k.p);
-		if (ret)
-			goto err;
+	prt_printf(&buf, "at btree %s level %u:\n  parent: ",
+		   bch2_btree_id_str(b->c.btree_id), b->c.level);
+	bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
+
+	prt_str(&buf, "\n  child: ");
+	bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&child->key));
+
+	if (mustfix_fsck_err(c, btree_node_topology_bad_max_key,
+			     "btree node with incorrect max_key%s", buf.buf)) {
+		if (b->c.level == 1 &&
+		    bpos_lt(*pulled_from_scan, b->key.k.p)) {
+			ret = bch2_get_scanned_nodes(c, b->c.btree_id, 0,
+						bpos_successor(child->key.k.p), b->key.k.p);
+			if (ret)
+				goto err;
+
+			*pulled_from_scan = b->key.k.p;
+			ret = DID_FILL_FROM_SCAN;
+		} else {
+			ret = set_node_max(c, child, b->key.k.p);
+		}
 	}
 err:
 fsck_err:
-	printbuf_exit(&buf2);
-	printbuf_exit(&buf1);
+	printbuf_exit(&buf);
 	return ret;
 }
 
-static int bch2_btree_repair_topology_recurse(struct btree_trans *trans, struct btree *b)
+static int bch2_btree_repair_topology_recurse(struct btree_trans *trans, struct btree *b,
+					      struct bpos *pulled_from_scan)
 {
 	struct bch_fs *c = trans->c;
 	struct btree_and_journal_iter iter;
 	struct bkey_s_c k;
 	struct bkey_buf prev_k, cur_k;
 	struct btree *prev = NULL, *cur = NULL;
-	bool have_child, dropped_children = false;
+	bool have_child, new_pass = false;
 	struct printbuf buf = PRINTBUF;
 	int ret = 0;
 
 	if (!b->c.level)
 		return 0;
-again:
-	prev = NULL;
-	have_child = dropped_children = false;
+
 	bch2_bkey_buf_init(&prev_k);
 	bch2_bkey_buf_init(&cur_k);
+again:
+	cur = prev = NULL;
+	have_child = new_pass = false;
 	bch2_btree_and_journal_iter_init_node_iter(trans, &iter, b);
 	iter.prefetch = true;
 
@@ -415,11 +367,17 @@ again:
 				b->c.level - 1,
 				buf.buf)) {
 			bch2_btree_node_evict(trans, cur_k.k);
+			cur = NULL;
 			ret = bch2_journal_key_delete(c, b->c.btree_id,
 						      b->c.level, cur_k.k->k.p);
-			cur = NULL;
 			if (ret)
 				break;
+
+			if (!btree_id_is_alloc(b->c.btree_id)) {
+				ret = bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_scan_for_btree_nodes);
+				if (ret)
+					break;
+			}
 			continue;
 		}
 
@@ -427,7 +385,23 @@ again:
 		if (ret)
 			break;
 
-		ret = btree_repair_node_boundaries(c, b, prev, cur);
+		if (bch2_btree_node_is_stale(c, cur)) {
+			bch_info(c, "btree node %s older than nodes found by scanning", buf.buf);
+			six_unlock_read(&cur->c.lock);
+			bch2_btree_node_evict(trans, cur_k.k);
+			ret = bch2_journal_key_delete(c, b->c.btree_id,
+						      b->c.level, cur_k.k->k.p);
+			cur = NULL;
+			if (ret)
+				break;
+			continue;
+		}
+
+		ret = btree_check_node_boundaries(c, b, prev, cur, pulled_from_scan);
+		if (ret == DID_FILL_FROM_SCAN) {
+			new_pass = true;
+			ret = 0;
+		}
 
 		if (ret == DROP_THIS_NODE) {
 			six_unlock_read(&cur->c.lock);
@@ -445,6 +419,7 @@ again:
 		prev = NULL;
 
 		if (ret == DROP_PREV_NODE) {
+			bch_info(c, "dropped prev node");
 			bch2_btree_node_evict(trans, prev_k.k);
 			ret = bch2_journal_key_delete(c, b->c.btree_id,
 						      b->c.level, prev_k.k->k.p);
@@ -452,8 +427,6 @@ again:
 				break;
 
 			bch2_btree_and_journal_iter_exit(&iter);
-			bch2_bkey_buf_exit(&prev_k, c);
-			bch2_bkey_buf_exit(&cur_k, c);
 			goto again;
 		} else if (ret)
 			break;
@@ -465,7 +438,11 @@ again:
 
 	if (!ret && !IS_ERR_OR_NULL(prev)) {
 		BUG_ON(cur);
-		ret = btree_repair_node_end(c, b, prev);
+		ret = btree_repair_node_end(c, b, prev, pulled_from_scan);
+		if (ret == DID_FILL_FROM_SCAN) {
+			new_pass = true;
+			ret = 0;
+		}
 	}
 
 	if (!IS_ERR_OR_NULL(prev))
@@ -479,6 +456,10 @@ again:
 		goto err;
 
 	bch2_btree_and_journal_iter_exit(&iter);
+
+	if (new_pass)
+		goto again;
+
 	bch2_btree_and_journal_iter_init_node_iter(trans, &iter, b);
 	iter.prefetch = true;
 
@@ -495,7 +476,7 @@ again:
 		if (ret)
 			goto err;
 
-		ret = bch2_btree_repair_topology_recurse(trans, cur);
+		ret = bch2_btree_repair_topology_recurse(trans, cur, pulled_from_scan);
 		six_unlock_read(&cur->c.lock);
 		cur = NULL;
 
@@ -503,7 +484,7 @@ again:
 			bch2_btree_node_evict(trans, cur_k.k);
 			ret = bch2_journal_key_delete(c, b->c.btree_id,
 						      b->c.level, cur_k.k->k.p);
-			dropped_children = true;
+			new_pass = true;
 		}
 
 		if (ret)
@@ -530,12 +511,14 @@ fsck_err:
 		six_unlock_read(&cur->c.lock);
 
 	bch2_btree_and_journal_iter_exit(&iter);
-	bch2_bkey_buf_exit(&prev_k, c);
-	bch2_bkey_buf_exit(&cur_k, c);
 
-	if (!ret && dropped_children)
+	if (!ret && new_pass)
 		goto again;
 
+	BUG_ON(!ret && bch2_btree_node_check_topology(trans, b));
+
+	bch2_bkey_buf_exit(&prev_k, c);
+	bch2_bkey_buf_exit(&cur_k, c);
 	printbuf_exit(&buf);
 	return ret;
 }
@@ -543,32 +526,63 @@ fsck_err:
 int bch2_check_topology(struct bch_fs *c)
 {
 	struct btree_trans *trans = bch2_trans_get(c);
-	struct btree *b;
-	unsigned i;
+	struct bpos pulled_from_scan = POS_MIN;
 	int ret = 0;
 
-	for (i = 0; i < btree_id_nr_alive(c) && !ret; i++) {
+	for (unsigned i = 0; i < btree_id_nr_alive(c) && !ret; i++) {
 		struct btree_root *r = bch2_btree_id_root(c, i);
+		bool reconstructed_root = false;
 
-		if (!r->alive)
-			continue;
+		if (r->error) {
+			ret = bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_scan_for_btree_nodes);
+			if (ret)
+				break;
+reconstruct_root:
+			bch_info(c, "btree root %s unreadable, must recover from scan", bch2_btree_id_str(i));
 
-		b = r->b;
-		if (btree_node_fake(b))
-			continue;
+			r->alive = false;
+			r->error = 0;
+
+			if (!bch2_btree_has_scanned_nodes(c, i)) {
+				mustfix_fsck_err(c, btree_root_unreadable_and_scan_found_nothing,
+						 "no nodes found for btree %s, continue?", bch2_btree_id_str(i));
+				bch2_btree_root_alloc_fake(c, i, 0);
+			} else {
+				bch2_btree_root_alloc_fake(c, i, 1);
+				bch2_shoot_down_journal_keys(c, i, 1, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
+				ret = bch2_get_scanned_nodes(c, i, 0, POS_MIN, SPOS_MAX);
+				if (ret)
+					break;
+			}
+
+			reconstructed_root = true;
+		}
+
+		struct btree *b = r->b;
 
 		btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_read);
-		ret = bch2_btree_repair_topology_recurse(trans, b);
+		ret = bch2_btree_repair_topology_recurse(trans, b, &pulled_from_scan);
 		six_unlock_read(&b->c.lock);
 
 		if (ret == DROP_THIS_NODE) {
-			bch_err(c, "empty btree root - repair unimplemented");
-			ret = -BCH_ERR_fsck_repair_unimplemented;
+			bch2_btree_node_hash_remove(&c->btree_cache, b);
+			mutex_lock(&c->btree_cache.lock);
+			list_move(&b->list, &c->btree_cache.freeable);
+			mutex_unlock(&c->btree_cache.lock);
+
+			r->b = NULL;
+
+			if (!reconstructed_root)
+				goto reconstruct_root;
+
+			bch_err(c, "empty btree root %s", bch2_btree_id_str(i));
+			bch2_btree_root_alloc_fake(c, i, 0);
+			r->alive = false;
+			ret = 0;
 		}
 	}
-
+fsck_err:
 	bch2_trans_put(trans);
-
 	return ret;
 }
 
@@ -591,7 +605,7 @@ static int bch2_check_fix_ptrs(struct btree_trans *trans, enum btree_id btree_id
 	bkey_for_each_ptr_decode(k->k, ptrs_c, p, entry_c) {
 		struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
 		struct bucket *g = PTR_GC_BUCKET(ca, &p.ptr);
-		enum bch_data_type data_type = bch2_bkey_ptr_data_type(*k, &entry_c->ptr);
+		enum bch_data_type data_type = bch2_bkey_ptr_data_type(*k, p, entry_c);
 
 		if (fsck_err_on(!g->gen_valid,
 				c, ptr_to_missing_alloc_key,
@@ -657,7 +671,8 @@ static int bch2_check_fix_ptrs(struct btree_trans *trans, enum btree_id btree_id
 			continue;
 
 		if (fsck_err_on(bucket_data_type(g->data_type) &&
-				bucket_data_type(g->data_type) != data_type, c,
+				bucket_data_type(g->data_type) !=
+				bucket_data_type(data_type), c,
 				ptr_bucket_data_type_mismatch,
 				"bucket %u:%zu different types of data in same bucket: %s, %s\n"
 				"while marking %s",
@@ -698,18 +713,13 @@ static int bch2_check_fix_ptrs(struct btree_trans *trans, enum btree_id btree_id
 	}
 
 	if (do_update) {
-		struct bkey_ptrs ptrs;
-		union bch_extent_entry *entry;
-		struct bch_extent_ptr *ptr;
-		struct bkey_i *new;
-
 		if (is_root) {
 			bch_err(c, "cannot update btree roots yet");
 			ret = -EINVAL;
 			goto err;
 		}
 
-		new = kmalloc(bkey_bytes(k->k), GFP_KERNEL);
+		struct bkey_i *new = kmalloc(bkey_bytes(k->k), GFP_KERNEL);
 		if (!new) {
 			ret = -BCH_ERR_ENOMEM_gc_repair_key;
 			bch_err_msg(c, ret, "allocating new key");
@@ -724,7 +734,7 @@ static int bch2_check_fix_ptrs(struct btree_trans *trans, enum btree_id btree_id
 			 * btree node isn't there anymore, the read path will
 			 * sort it out:
 			 */
-			ptrs = bch2_bkey_ptrs(bkey_i_to_s(new));
+			struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(new));
 			bkey_for_each_ptr(ptrs, ptr) {
 				struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
 				struct bucket *g = PTR_GC_BUCKET(ca, ptr);
@@ -732,19 +742,26 @@ static int bch2_check_fix_ptrs(struct btree_trans *trans, enum btree_id btree_id
 				ptr->gen = g->gen;
 			}
 		} else {
-			bch2_bkey_drop_ptrs(bkey_i_to_s(new), ptr, ({
-				struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
-				struct bucket *g = PTR_GC_BUCKET(ca, ptr);
-				enum bch_data_type data_type = bch2_bkey_ptr_data_type(*k, ptr);
-
-				(ptr->cached &&
-				 (!g->gen_valid || gen_cmp(ptr->gen, g->gen) > 0)) ||
-				(!ptr->cached &&
-				 gen_cmp(ptr->gen, g->gen) < 0) ||
-				gen_cmp(g->gen, ptr->gen) > BUCKET_GC_GEN_MAX ||
-				(g->data_type &&
-				 g->data_type != data_type);
-			}));
+			struct bkey_ptrs ptrs;
+			union bch_extent_entry *entry;
+restart_drop_ptrs:
+			ptrs = bch2_bkey_ptrs(bkey_i_to_s(new));
+			bkey_for_each_ptr_decode(bkey_i_to_s(new).k, ptrs, p, entry) {
+				struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
+				struct bucket *g = PTR_GC_BUCKET(ca, &p.ptr);
+				enum bch_data_type data_type = bch2_bkey_ptr_data_type(bkey_i_to_s_c(new), p, entry);
+
+				if ((p.ptr.cached &&
+				     (!g->gen_valid || gen_cmp(p.ptr.gen, g->gen) > 0)) ||
+				    (!p.ptr.cached &&
+				     gen_cmp(p.ptr.gen, g->gen) < 0) ||
+				    gen_cmp(g->gen, p.ptr.gen) > BUCKET_GC_GEN_MAX ||
+				    (g->data_type &&
+				     g->data_type != data_type)) {
+					bch2_bkey_drop_ptr(bkey_i_to_s(new), &entry->ptr);
+					goto restart_drop_ptrs;
+				}
+			}
 again:
 			ptrs = bch2_bkey_ptrs(bkey_i_to_s(new));
 			bkey_extent_entry_for_each(ptrs, entry) {
@@ -774,12 +791,6 @@ found:
 			}
 		}
 
-		ret = bch2_journal_key_insert_take(c, btree_id, level, new);
-		if (ret) {
-			kfree(new);
-			goto err;
-		}
-
 		if (level)
 			bch2_btree_node_update_key_early(trans, btree_id, level - 1, *k, new);
 
@@ -793,6 +804,12 @@ found:
 			bch_info(c, "new key %s", buf.buf);
 		}
 
+		ret = bch2_journal_key_insert_take(c, btree_id, level, new);
+		if (ret) {
+			kfree(new);
+			goto err;
+		}
+
 		*k = bkey_i_to_s_c(new);
 	}
 err:
@@ -811,6 +828,7 @@ static int bch2_gc_mark_key(struct btree_trans *trans, enum btree_id btree_id,
 	struct bch_fs *c = trans->c;
 	struct bkey deleted = KEY(0, 0, 0);
 	struct bkey_s_c old = (struct bkey_s_c) { &deleted, NULL };
+	struct printbuf buf = PRINTBUF;
 	int ret = 0;
 
 	deleted.p = k->k->p;
@@ -819,10 +837,6 @@ static int bch2_gc_mark_key(struct btree_trans *trans, enum btree_id btree_id,
 		BUG_ON(bch2_journal_seq_verify &&
 		       k->k->version.lo > atomic64_read(&c->journal.seq));
 
-		ret = bch2_check_fix_ptrs(trans, btree_id, level, is_root, k);
-		if (ret)
-			goto err;
-
 		if (fsck_err_on(k->k->version.lo > atomic64_read(&c->key_version), c,
 				bkey_version_in_future,
 				"key version number higher than recorded: %llu > %llu",
@@ -831,52 +845,57 @@ static int bch2_gc_mark_key(struct btree_trans *trans, enum btree_id btree_id,
 			atomic64_set(&c->key_version, k->k->version.lo);
 	}
 
+	ret = bch2_check_fix_ptrs(trans, btree_id, level, is_root, k);
+	if (ret)
+		goto err;
+
+	if (mustfix_fsck_err_on(level && !bch2_dev_btree_bitmap_marked(c, *k),
+				c, btree_bitmap_not_marked,
+				"btree ptr not marked in member info btree allocated bitmap\n  %s",
+				(bch2_bkey_val_to_text(&buf, c, *k),
+				 buf.buf))) {
+		mutex_lock(&c->sb_lock);
+		bch2_dev_btree_bitmap_mark(c, *k);
+		bch2_write_super(c);
+		mutex_unlock(&c->sb_lock);
+	}
+
 	ret = commit_do(trans, NULL, NULL, 0,
-			bch2_key_trigger(trans, btree_id, level, old, unsafe_bkey_s_c_to_s(*k), BTREE_TRIGGER_GC));
+			bch2_key_trigger(trans, btree_id, level, old,
+					 unsafe_bkey_s_c_to_s(*k), BTREE_TRIGGER_GC));
 fsck_err:
 err:
+	printbuf_exit(&buf);
 	bch_err_fn(c, ret);
 	return ret;
 }
 
 static int btree_gc_mark_node(struct btree_trans *trans, struct btree *b, bool initial)
 {
-	struct bch_fs *c = trans->c;
 	struct btree_node_iter iter;
 	struct bkey unpacked;
 	struct bkey_s_c k;
-	struct bkey_buf prev, cur;
 	int ret = 0;
 
+	ret = bch2_btree_node_check_topology(trans, b);
+	if (ret)
+		return ret;
+
 	if (!btree_node_type_needs_gc(btree_node_type(b)))
 		return 0;
 
 	bch2_btree_node_iter_init_from_start(&iter, b);
-	bch2_bkey_buf_init(&prev);
-	bch2_bkey_buf_init(&cur);
-	bkey_init(&prev.k->k);
 
 	while ((k = bch2_btree_node_iter_peek_unpack(&iter, b, &unpacked)).k) {
 		ret = bch2_gc_mark_key(trans, b->c.btree_id, b->c.level, false,
 				       &k, initial);
 		if (ret)
-			break;
+			return ret;
 
 		bch2_btree_node_iter_advance(&iter, b);
-
-		if (b->c.level) {
-			bch2_bkey_buf_reassemble(&cur, c, k);
-
-			ret = bch2_gc_check_topology(c, b, &prev, cur,
-					bch2_btree_node_iter_end(&iter));
-			if (ret)
-				break;
-		}
 	}
 
-	bch2_bkey_buf_exit(&cur, c);
-	bch2_bkey_buf_exit(&prev, c);
-	return ret;
+	return 0;
 }
 
 static int bch2_gc_btree(struct btree_trans *trans, enum btree_id btree_id,
@@ -925,14 +944,16 @@ static int bch2_gc_btree_init_recurse(struct btree_trans *trans, struct btree *b
 	struct bch_fs *c = trans->c;
 	struct btree_and_journal_iter iter;
 	struct bkey_s_c k;
-	struct bkey_buf cur, prev;
+	struct bkey_buf cur;
 	struct printbuf buf = PRINTBUF;
 	int ret = 0;
 
+	ret = bch2_btree_node_check_topology(trans, b);
+	if (ret)
+		return ret;
+
 	bch2_btree_and_journal_iter_init_node_iter(trans, &iter, b);
-	bch2_bkey_buf_init(&prev);
 	bch2_bkey_buf_init(&cur);
-	bkey_init(&prev.k->k);
 
 	while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 		BUG_ON(bpos_lt(k.k->p, b->data->min_key));
@@ -943,20 +964,7 @@ static int bch2_gc_btree_init_recurse(struct btree_trans *trans, struct btree *b
 		if (ret)
 			goto fsck_err;
 
-		if (b->c.level) {
-			bch2_bkey_buf_reassemble(&cur, c, k);
-			k = bkey_i_to_s_c(cur.k);
-
-			bch2_btree_and_journal_iter_advance(&iter);
-
-			ret = bch2_gc_check_topology(c, b,
-					&prev, cur,
-					!bch2_btree_and_journal_iter_peek(&iter).k);
-			if (ret)
-				goto fsck_err;
-		} else {
-			bch2_btree_and_journal_iter_advance(&iter);
-		}
+		bch2_btree_and_journal_iter_advance(&iter);
 	}
 
 	if (b->c.level > target_depth) {
@@ -1015,7 +1023,6 @@ static int bch2_gc_btree_init_recurse(struct btree_trans *trans, struct btree *b
 	}
 fsck_err:
 	bch2_bkey_buf_exit(&cur, c);
-	bch2_bkey_buf_exit(&prev, c);
 	bch2_btree_and_journal_iter_exit(&iter);
 	printbuf_exit(&buf);
 	return ret;
@@ -1033,9 +1040,6 @@ static int bch2_gc_btree_init(struct btree_trans *trans,
 
 	b = bch2_btree_id_root(c, btree_id)->b;
 
-	if (btree_node_fake(b))
-		return 0;
-
 	six_lock_read(&b->c.lock, NULL, NULL);
 	printbuf_reset(&buf);
 	bch2_bpos_to_text(&buf, b->data->min_key);
diff --git a/fs/bcachefs/btree_io.c b/fs/bcachefs/btree_io.c
index 34df8ccc5fec..9678b2375bed 100644
--- a/fs/bcachefs/btree_io.c
+++ b/fs/bcachefs/btree_io.c
@@ -654,6 +654,7 @@ void bch2_btree_node_drop_keys_outside_node(struct btree *b)
 	 */
 	bch2_bset_set_no_aux_tree(b, b->set);
 	bch2_btree_build_aux_trees(b);
+	b->nr = bch2_btree_node_count_keys(b);
 
 	struct bkey_s_c k;
 	struct bkey unpacked;
@@ -830,7 +831,7 @@ static int bset_key_invalid(struct bch_fs *c, struct btree *b,
 		(rw == WRITE ? bch2_bkey_val_invalid(c, k, READ, err) : 0);
 }
 
-static bool __bkey_valid(struct bch_fs *c, struct btree *b,
+static bool bkey_packed_valid(struct bch_fs *c, struct btree *b,
 			 struct bset *i, struct bkey_packed *k)
 {
 	if (bkey_p_next(k) > vstruct_last(i))
@@ -839,7 +840,7 @@ static bool __bkey_valid(struct bch_fs *c, struct btree *b,
 	if (k->format > KEY_FORMAT_CURRENT)
 		return false;
 
-	if (k->u64s < bkeyp_key_u64s(&b->format, k))
+	if (!bkeyp_u64s_valid(&b->format, k))
 		return false;
 
 	struct printbuf buf = PRINTBUF;
@@ -883,11 +884,13 @@ static int validate_bset_keys(struct bch_fs *c, struct btree *b,
 				 "invalid bkey format %u", k->format))
 			goto drop_this_key;
 
-		if (btree_err_on(k->u64s < bkeyp_key_u64s(&b->format, k),
+		if (btree_err_on(!bkeyp_u64s_valid(&b->format, k),
 				 -BCH_ERR_btree_node_read_err_fixable,
 				 c, NULL, b, i,
 				 btree_node_bkey_bad_u64s,
-				 "k->u64s too small (%u < %u)", k->u64s, bkeyp_key_u64s(&b->format, k)))
+				 "bad k->u64s %u (min %u max %lu)", k->u64s,
+				 bkeyp_key_u64s(&b->format, k),
+				 U8_MAX - BKEY_U64s + bkeyp_key_u64s(&b->format, k)))
 			goto drop_this_key;
 
 		if (!write)
@@ -946,13 +949,12 @@ drop_this_key:
 			 * do
 			 */
 
-			if (!__bkey_valid(c, b, i, (void *) ((u64 *) k + next_good_key))) {
+			if (!bkey_packed_valid(c, b, i, (void *) ((u64 *) k + next_good_key))) {
 				for (next_good_key = 1;
 				     next_good_key < (u64 *) vstruct_last(i) - (u64 *) k;
 				     next_good_key++)
-					if (__bkey_valid(c, b, i, (void *) ((u64 *) k + next_good_key)))
+					if (bkey_packed_valid(c, b, i, (void *) ((u64 *) k + next_good_key)))
 						goto got_good_key;
-
 			}
 
 			/*
@@ -1263,10 +1265,12 @@ out:
 	return retry_read;
 fsck_err:
 	if (ret == -BCH_ERR_btree_node_read_err_want_retry ||
-	    ret == -BCH_ERR_btree_node_read_err_must_retry)
+	    ret == -BCH_ERR_btree_node_read_err_must_retry) {
 		retry_read = 1;
-	else
+	} else {
 		set_btree_node_read_error(b);
+		bch2_btree_lost_data(c, b->c.btree_id);
+	}
 	goto out;
 }
 
@@ -1327,6 +1331,7 @@ start:
 
 		if (!can_retry) {
 			set_btree_node_read_error(b);
+			bch2_btree_lost_data(c, b->c.btree_id);
 			break;
 		}
 	}
@@ -1335,7 +1340,9 @@ start:
 			       rb->start_time);
 	bio_put(&rb->bio);
 
-	if (saw_error && !btree_node_read_error(b)) {
+	if (saw_error &&
+	    !btree_node_read_error(b) &&
+	    c->curr_recovery_pass != BCH_RECOVERY_PASS_scan_for_btree_nodes) {
 		printbuf_reset(&buf);
 		bch2_bpos_to_text(&buf, b->key.k.p);
 		bch_err_ratelimited(c, "%s: rewriting btree node at btree=%s level=%u %s due to error",
@@ -1526,9 +1533,10 @@ fsck_err:
 		ret = -1;
 	}
 
-	if (ret)
+	if (ret) {
 		set_btree_node_read_error(b);
-	else if (*saw_error)
+		bch2_btree_lost_data(c, b->c.btree_id);
+	} else if (*saw_error)
 		bch2_btree_node_rewrite_async(c, b);
 
 	for (i = 0; i < ra->nr; i++) {
@@ -1657,13 +1665,14 @@ void bch2_btree_node_read(struct btree_trans *trans, struct btree *b,
 
 		prt_str(&buf, "btree node read error: no device to read from\n at ");
 		bch2_btree_pos_to_text(&buf, c, b);
-		bch_err(c, "%s", buf.buf);
+		bch_err_ratelimited(c, "%s", buf.buf);
 
 		if (c->recovery_passes_explicit & BIT_ULL(BCH_RECOVERY_PASS_check_topology) &&
 		    c->curr_recovery_pass > BCH_RECOVERY_PASS_check_topology)
 			bch2_fatal_error(c);
 
 		set_btree_node_read_error(b);
+		bch2_btree_lost_data(c, b->c.btree_id);
 		clear_btree_node_read_in_flight(b);
 		wake_up_bit(&b->flags, BTREE_NODE_read_in_flight);
 		printbuf_exit(&buf);
@@ -1860,7 +1869,7 @@ static void btree_node_write_work(struct work_struct *work)
 	} else {
 		ret = bch2_trans_do(c, NULL, NULL, 0,
 			bch2_btree_node_update_key_get_iter(trans, b, &wbio->key,
-					BCH_WATERMARK_reclaim|
+					BCH_WATERMARK_interior_updates|
 					BCH_TRANS_COMMIT_journal_reclaim|
 					BCH_TRANS_COMMIT_no_enospc|
 					BCH_TRANS_COMMIT_no_check_rw,
diff --git a/fs/bcachefs/btree_iter.c b/fs/bcachefs/btree_iter.c
index 51bcdc6c6d1c..2a211a4bebd1 100644
--- a/fs/bcachefs/btree_iter.c
+++ b/fs/bcachefs/btree_iter.c
@@ -927,8 +927,22 @@ static __always_inline int btree_path_down(struct btree_trans *trans,
 		if (ret)
 			goto err;
 	} else {
-		bch2_bkey_buf_unpack(&tmp, c, l->b,
-				 bch2_btree_node_iter_peek(&l->iter, l->b));
+		struct bkey_packed *k = bch2_btree_node_iter_peek(&l->iter, l->b);
+		if (!k) {
+			struct printbuf buf = PRINTBUF;
+
+			prt_str(&buf, "node not found at pos ");
+			bch2_bpos_to_text(&buf, path->pos);
+			prt_str(&buf, " within parent node ");
+			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&l->b->key));
+
+			bch2_fs_fatal_error(c, "%s", buf.buf);
+			printbuf_exit(&buf);
+			ret = -BCH_ERR_btree_need_topology_repair;
+			goto err;
+		}
+
+		bch2_bkey_buf_unpack(&tmp, c, l->b, k);
 
 		if ((flags & BTREE_ITER_PREFETCH) &&
 		    c->opts.btree_node_prefetch) {
@@ -962,7 +976,6 @@ err:
 	return ret;
 }
 
-
 static int bch2_btree_path_traverse_all(struct btree_trans *trans)
 {
 	struct bch_fs *c = trans->c;
@@ -2790,6 +2803,31 @@ void *__bch2_trans_kmalloc(struct btree_trans *trans, size_t size)
 	struct btree_transaction_stats *s = btree_trans_stats(trans);
 	s->max_mem = max(s->max_mem, new_bytes);
 
+	if (trans->used_mempool) {
+		if (trans->mem_bytes >= new_bytes)
+			goto out_change_top;
+
+		/* No more space from mempool item, need malloc new one */
+		new_mem = kmalloc(new_bytes, GFP_NOWAIT|__GFP_NOWARN);
+		if (unlikely(!new_mem)) {
+			bch2_trans_unlock(trans);
+
+			new_mem = kmalloc(new_bytes, GFP_KERNEL);
+			if (!new_mem)
+				return ERR_PTR(-BCH_ERR_ENOMEM_trans_kmalloc);
+
+			ret = bch2_trans_relock(trans);
+			if (ret) {
+				kfree(new_mem);
+				return ERR_PTR(ret);
+			}
+		}
+		memcpy(new_mem, trans->mem, trans->mem_top);
+		trans->used_mempool = false;
+		mempool_free(trans->mem, &c->btree_trans_mem_pool);
+		goto out_new_mem;
+	}
+
 	new_mem = krealloc(trans->mem, new_bytes, GFP_NOWAIT|__GFP_NOWARN);
 	if (unlikely(!new_mem)) {
 		bch2_trans_unlock(trans);
@@ -2798,6 +2836,8 @@ void *__bch2_trans_kmalloc(struct btree_trans *trans, size_t size)
 		if (!new_mem && new_bytes <= BTREE_TRANS_MEM_MAX) {
 			new_mem = mempool_alloc(&c->btree_trans_mem_pool, GFP_KERNEL);
 			new_bytes = BTREE_TRANS_MEM_MAX;
+			memcpy(new_mem, trans->mem, trans->mem_top);
+			trans->used_mempool = true;
 			kfree(trans->mem);
 		}
 
@@ -2811,7 +2851,7 @@ void *__bch2_trans_kmalloc(struct btree_trans *trans, size_t size)
 		if (ret)
 			return ERR_PTR(ret);
 	}
-
+out_new_mem:
 	trans->mem = new_mem;
 	trans->mem_bytes = new_bytes;
 
@@ -2819,7 +2859,7 @@ void *__bch2_trans_kmalloc(struct btree_trans *trans, size_t size)
 		trace_and_count(c, trans_restart_mem_realloced, trans, _RET_IP_, new_bytes);
 		return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_mem_realloced));
 	}
-
+out_change_top:
 	p = trans->mem + trans->mem_top;
 	trans->mem_top += size;
 	memset(p, 0, size);
@@ -3093,7 +3133,7 @@ void bch2_trans_put(struct btree_trans *trans)
 	if (paths_allocated != trans->_paths_allocated)
 		kvfree_rcu_mightsleep(paths_allocated);
 
-	if (trans->mem_bytes == BTREE_TRANS_MEM_MAX)
+	if (trans->used_mempool)
 		mempool_free(trans->mem, &c->btree_trans_mem_pool);
 	else
 		kfree(trans->mem);
diff --git a/fs/bcachefs/btree_iter.h b/fs/bcachefs/btree_iter.h
index 24772538e4cc..1c70836dd7cc 100644
--- a/fs/bcachefs/btree_iter.h
+++ b/fs/bcachefs/btree_iter.h
@@ -498,8 +498,13 @@ static inline void set_btree_iter_dontneed(struct btree_iter *iter)
 {
 	struct btree_trans *trans = iter->trans;
 
-	if (!trans->restarted)
-		btree_iter_path(trans, iter)->preserve = false;
+	if (!iter->path || trans->restarted)
+		return;
+
+	struct btree_path *path = btree_iter_path(trans, iter);
+	path->preserve		= false;
+	if (path->ref == 1)
+		path->should_be_locked	= false;
 }
 
 void *__bch2_trans_kmalloc(struct btree_trans *, size_t);
@@ -642,7 +647,7 @@ int __bch2_btree_trans_too_many_iters(struct btree_trans *);
 
 static inline int btree_trans_too_many_iters(struct btree_trans *trans)
 {
-	if (bitmap_weight(trans->paths_allocated, trans->nr_paths) > BTREE_ITER_INITIAL - 8)
+	if (bitmap_weight(trans->paths_allocated, trans->nr_paths) > BTREE_ITER_NORMAL_LIMIT - 8)
 		return __bch2_btree_trans_too_many_iters(trans);
 
 	return 0;
diff --git a/fs/bcachefs/btree_journal_iter.c b/fs/bcachefs/btree_journal_iter.c
index 50e04356d72c..1e8cf49a6935 100644
--- a/fs/bcachefs/btree_journal_iter.c
+++ b/fs/bcachefs/btree_journal_iter.c
@@ -130,12 +130,30 @@ struct bkey_i *bch2_journal_keys_peek_slot(struct bch_fs *c, enum btree_id btree
 	return bch2_journal_keys_peek_upto(c, btree_id, level, pos, pos, &idx);
 }
 
+static void journal_iter_verify(struct journal_iter *iter)
+{
+	struct journal_keys *keys = iter->keys;
+	size_t gap_size = keys->size - keys->nr;
+
+	BUG_ON(iter->idx >= keys->gap &&
+	       iter->idx <  keys->gap + gap_size);
+
+	if (iter->idx < keys->size) {
+		struct journal_key *k = keys->data + iter->idx;
+
+		int cmp = cmp_int(k->btree_id,	iter->btree_id) ?:
+			  cmp_int(k->level,	iter->level);
+		BUG_ON(cmp < 0);
+	}
+}
+
 static void journal_iters_fix(struct bch_fs *c)
 {
 	struct journal_keys *keys = &c->journal_keys;
 	/* The key we just inserted is immediately before the gap: */
 	size_t gap_end = keys->gap + (keys->size - keys->nr);
-	struct btree_and_journal_iter *iter;
+	struct journal_key *new_key = &keys->data[keys->gap - 1];
+	struct journal_iter *iter;
 
 	/*
 	 * If an iterator points one after the key we just inserted, decrement
@@ -143,9 +161,14 @@ static void journal_iters_fix(struct bch_fs *c)
 	 * decrement was unnecessary, bch2_btree_and_journal_iter_peek() will
 	 * handle that:
 	 */
-	list_for_each_entry(iter, &c->journal_iters, journal.list)
-		if (iter->journal.idx == gap_end)
-			iter->journal.idx = keys->gap - 1;
+	list_for_each_entry(iter, &c->journal_iters, list) {
+		journal_iter_verify(iter);
+		if (iter->idx		== gap_end &&
+		    new_key->btree_id	== iter->btree_id &&
+		    new_key->level	== iter->level)
+			iter->idx = keys->gap - 1;
+		journal_iter_verify(iter);
+	}
 }
 
 static void journal_iters_move_gap(struct bch_fs *c, size_t old_gap, size_t new_gap)
@@ -192,7 +215,12 @@ int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
 	if (idx > keys->gap)
 		idx -= keys->size - keys->nr;
 
+	size_t old_gap = keys->gap;
+
 	if (keys->nr == keys->size) {
+		journal_iters_move_gap(c, old_gap, keys->size);
+		old_gap = keys->size;
+
 		struct journal_keys new_keys = {
 			.nr			= keys->nr,
 			.size			= max_t(size_t, keys->size, 8) * 2,
@@ -216,7 +244,7 @@ int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
 		keys->gap	= keys->nr;
 	}
 
-	journal_iters_move_gap(c, keys->gap, idx);
+	journal_iters_move_gap(c, old_gap, idx);
 
 	move_gap(keys, idx);
 
@@ -261,6 +289,22 @@ int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
 	return bch2_journal_key_insert(c, id, level, &whiteout);
 }
 
+bool bch2_key_deleted_in_journal(struct btree_trans *trans, enum btree_id btree,
+				 unsigned level, struct bpos pos)
+{
+	struct journal_keys *keys = &trans->c->journal_keys;
+	size_t idx = bch2_journal_key_search(keys, btree, level, pos);
+
+	if (!trans->journal_replay_not_finished)
+		return false;
+
+	return (idx < keys->size &&
+		keys->data[idx].btree_id	== btree &&
+		keys->data[idx].level		== level &&
+		bpos_eq(keys->data[idx].k->k.p, pos) &&
+		bkey_deleted(&keys->data[idx].k->k));
+}
+
 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
 				  unsigned level, struct bpos pos)
 {
@@ -285,16 +329,21 @@ static void bch2_journal_iter_advance(struct journal_iter *iter)
 
 static struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
 {
-	struct journal_key *k = iter->keys->data + iter->idx;
+	journal_iter_verify(iter);
+
+	while (iter->idx < iter->keys->size) {
+		struct journal_key *k = iter->keys->data + iter->idx;
+
+		int cmp = cmp_int(k->btree_id,	iter->btree_id) ?:
+			  cmp_int(k->level,	iter->level);
+		if (cmp > 0)
+			break;
+		BUG_ON(cmp);
 
-	while (k < iter->keys->data + iter->keys->size &&
-	       k->btree_id	== iter->btree_id &&
-	       k->level		== iter->level) {
 		if (!k->overwritten)
 			return bkey_i_to_s_c(k->k);
 
 		bch2_journal_iter_advance(iter);
-		k = iter->keys->data + iter->idx;
 	}
 
 	return bkey_s_c_null;
@@ -314,6 +363,8 @@ static void bch2_journal_iter_init(struct bch_fs *c,
 	iter->level	= level;
 	iter->keys	= &c->journal_keys;
 	iter->idx	= bch2_journal_key_search(&c->journal_keys, id, level, pos);
+
+	journal_iter_verify(iter);
 }
 
 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
@@ -363,7 +414,7 @@ static void btree_and_journal_iter_prefetch(struct btree_and_journal_iter *_iter
 
 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 {
-	struct bkey_s_c btree_k, journal_k, ret;
+	struct bkey_s_c btree_k, journal_k = bkey_s_c_null, ret;
 
 	if (iter->prefetch && iter->journal.level)
 		btree_and_journal_iter_prefetch(iter);
@@ -375,9 +426,10 @@ again:
 	       bpos_lt(btree_k.k->p, iter->pos))
 		bch2_journal_iter_advance_btree(iter);
 
-	while ((journal_k = bch2_journal_iter_peek(&iter->journal)).k &&
-	       bpos_lt(journal_k.k->p, iter->pos))
-		bch2_journal_iter_advance(&iter->journal);
+	if (iter->trans->journal_replay_not_finished)
+		while ((journal_k = bch2_journal_iter_peek(&iter->journal)).k &&
+		       bpos_lt(journal_k.k->p, iter->pos))
+			bch2_journal_iter_advance(&iter->journal);
 
 	ret = journal_k.k &&
 		(!btree_k.k || bpos_le(journal_k.k->p, btree_k.k->p))
@@ -417,10 +469,15 @@ void __bch2_btree_and_journal_iter_init_node_iter(struct btree_trans *trans,
 	iter->trans = trans;
 	iter->b = b;
 	iter->node_iter = node_iter;
-	bch2_journal_iter_init(trans->c, &iter->journal, b->c.btree_id, b->c.level, pos);
-	INIT_LIST_HEAD(&iter->journal.list);
 	iter->pos = b->data->min_key;
 	iter->at_end = false;
+	INIT_LIST_HEAD(&iter->journal.list);
+
+	if (trans->journal_replay_not_finished) {
+		bch2_journal_iter_init(trans->c, &iter->journal, b->c.btree_id, b->c.level, pos);
+		if (!test_bit(BCH_FS_may_go_rw, &trans->c->flags))
+			list_add(&iter->journal.list, &trans->c->journal_iters);
+	}
 }
 
 /*
@@ -435,7 +492,6 @@ void bch2_btree_and_journal_iter_init_node_iter(struct btree_trans *trans,
 
 	bch2_btree_node_iter_init_from_start(&node_iter, b);
 	__bch2_btree_and_journal_iter_init_node_iter(trans, iter, b, node_iter, b->data->min_key);
-	list_add(&iter->journal.list, &trans->c->journal_iters);
 }
 
 /* sort and dedup all keys in the journal: */
@@ -548,3 +604,22 @@ int bch2_journal_keys_sort(struct bch_fs *c)
 	bch_verbose(c, "Journal keys: %zu read, %zu after sorting and compacting", nr_read, keys->nr);
 	return 0;
 }
+
+void bch2_shoot_down_journal_keys(struct bch_fs *c, enum btree_id btree,
+				  unsigned level_min, unsigned level_max,
+				  struct bpos start, struct bpos end)
+{
+	struct journal_keys *keys = &c->journal_keys;
+	size_t dst = 0;
+
+	move_gap(keys, keys->nr);
+
+	darray_for_each(*keys, i)
+		if (!(i->btree_id == btree &&
+		      i->level >= level_min &&
+		      i->level <= level_max &&
+		      bpos_ge(i->k->k.p, start) &&
+		      bpos_le(i->k->k.p, end)))
+			keys->data[dst++] = *i;
+	keys->nr = keys->gap = dst;
+}
diff --git a/fs/bcachefs/btree_journal_iter.h b/fs/bcachefs/btree_journal_iter.h
index c9d19da3ea04..af25046ebcaa 100644
--- a/fs/bcachefs/btree_journal_iter.h
+++ b/fs/bcachefs/btree_journal_iter.h
@@ -40,8 +40,8 @@ int bch2_journal_key_insert(struct bch_fs *, enum btree_id,
 			    unsigned, struct bkey_i *);
 int bch2_journal_key_delete(struct bch_fs *, enum btree_id,
 			    unsigned, struct bpos);
-void bch2_journal_key_overwritten(struct bch_fs *, enum btree_id,
-				  unsigned, struct bpos);
+bool bch2_key_deleted_in_journal(struct btree_trans *, enum btree_id, unsigned, struct bpos);
+void bch2_journal_key_overwritten(struct bch_fs *, enum btree_id, unsigned, struct bpos);
 
 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *);
 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *);
@@ -66,4 +66,8 @@ void bch2_journal_entries_free(struct bch_fs *);
 
 int bch2_journal_keys_sort(struct bch_fs *);
 
+void bch2_shoot_down_journal_keys(struct bch_fs *, enum btree_id,
+				  unsigned, unsigned,
+				  struct bpos, struct bpos);
+
 #endif /* _BCACHEFS_BTREE_JOURNAL_ITER_H */
diff --git a/fs/bcachefs/btree_key_cache.c b/fs/bcachefs/btree_key_cache.c
index 581edcb0911b..88a3582a3275 100644
--- a/fs/bcachefs/btree_key_cache.c
+++ b/fs/bcachefs/btree_key_cache.c
@@ -169,6 +169,7 @@ static void bkey_cached_move_to_freelist(struct btree_key_cache *bc,
 	} else {
 		mutex_lock(&bc->lock);
 		list_move_tail(&ck->list, &bc->freed_pcpu);
+		bc->nr_freed_pcpu++;
 		mutex_unlock(&bc->lock);
 	}
 }
@@ -245,6 +246,7 @@ bkey_cached_alloc(struct btree_trans *trans, struct btree_path *path,
 		if (!list_empty(&bc->freed_pcpu)) {
 			ck = list_last_entry(&bc->freed_pcpu, struct bkey_cached, list);
 			list_del_init(&ck->list);
+			bc->nr_freed_pcpu--;
 		}
 		mutex_unlock(&bc->lock);
 	}
@@ -659,7 +661,7 @@ static int btree_key_cache_flush_pos(struct btree_trans *trans,
 		commit_flags |= BCH_WATERMARK_reclaim;
 
 	if (ck->journal.seq != journal_last_seq(j) ||
-	    j->watermark == BCH_WATERMARK_stripe)
+	    !test_bit(JOURNAL_SPACE_LOW, &c->journal.flags))
 		commit_flags |= BCH_TRANS_COMMIT_no_journal_res;
 
 	ret   = bch2_btree_iter_traverse(&b_iter) ?:
diff --git a/fs/bcachefs/btree_locking.c b/fs/bcachefs/btree_locking.c
index b9b151e693ed..f2caf491957e 100644
--- a/fs/bcachefs/btree_locking.c
+++ b/fs/bcachefs/btree_locking.c
@@ -440,33 +440,7 @@ void bch2_btree_node_lock_write_nofail(struct btree_trans *trans,
 				       struct btree_path *path,
 				       struct btree_bkey_cached_common *b)
 {
-	struct btree_path *linked;
-	unsigned i, iter;
-	int ret;
-
-	/*
-	 * XXX BIG FAT NOTICE
-	 *
-	 * Drop all read locks before taking a write lock:
-	 *
-	 * This is a hack, because bch2_btree_node_lock_write_nofail() is a
-	 * hack - but by dropping read locks first, this should never fail, and
-	 * we only use this in code paths where whatever read locks we've
-	 * already taken are no longer needed:
-	 */
-
-	trans_for_each_path(trans, linked, iter) {
-		if (!linked->nodes_locked)
-			continue;
-
-		for (i = 0; i < BTREE_MAX_DEPTH; i++)
-			if (btree_node_read_locked(linked, i)) {
-				btree_node_unlock(trans, linked, i);
-				btree_path_set_dirty(linked, BTREE_ITER_NEED_RELOCK);
-			}
-	}
-
-	ret = __btree_node_lock_write(trans, path, b, true);
+	int ret = __btree_node_lock_write(trans, path, b, true);
 	BUG_ON(ret);
 }
 
diff --git a/fs/bcachefs/btree_node_scan.c b/fs/bcachefs/btree_node_scan.c
new file mode 100644
index 000000000000..866bd278439f
--- /dev/null
+++ b/fs/bcachefs/btree_node_scan.c
@@ -0,0 +1,519 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include "bcachefs.h"
+#include "btree_cache.h"
+#include "btree_io.h"
+#include "btree_journal_iter.h"
+#include "btree_node_scan.h"
+#include "btree_update_interior.h"
+#include "buckets.h"
+#include "error.h"
+#include "journal_io.h"
+#include "recovery_passes.h"
+
+#include <linux/kthread.h>
+#include <linux/sort.h>
+
+struct find_btree_nodes_worker {
+	struct closure		*cl;
+	struct find_btree_nodes	*f;
+	struct bch_dev		*ca;
+};
+
+static void found_btree_node_to_text(struct printbuf *out, struct bch_fs *c, const struct found_btree_node *n)
+{
+	prt_printf(out, "%s l=%u seq=%u cookie=%llx ", bch2_btree_id_str(n->btree_id), n->level, n->seq, n->cookie);
+	bch2_bpos_to_text(out, n->min_key);
+	prt_str(out, "-");
+	bch2_bpos_to_text(out, n->max_key);
+
+	if (n->range_updated)
+		prt_str(out, " range updated");
+	if (n->overwritten)
+		prt_str(out, " overwritten");
+
+	for (unsigned i = 0; i < n->nr_ptrs; i++) {
+		prt_char(out, ' ');
+		bch2_extent_ptr_to_text(out, c, n->ptrs + i);
+	}
+}
+
+static void found_btree_nodes_to_text(struct printbuf *out, struct bch_fs *c, found_btree_nodes nodes)
+{
+	printbuf_indent_add(out, 2);
+	darray_for_each(nodes, i) {
+		found_btree_node_to_text(out, c, i);
+		prt_newline(out);
+	}
+	printbuf_indent_sub(out, 2);
+}
+
+static void found_btree_node_to_key(struct bkey_i *k, const struct found_btree_node *f)
+{
+	struct bkey_i_btree_ptr_v2 *bp = bkey_btree_ptr_v2_init(k);
+
+	set_bkey_val_u64s(&bp->k, sizeof(struct bch_btree_ptr_v2) / sizeof(u64) + f->nr_ptrs);
+	bp->k.p			= f->max_key;
+	bp->v.seq		= cpu_to_le64(f->cookie);
+	bp->v.sectors_written	= 0;
+	bp->v.flags		= 0;
+	bp->v.min_key		= f->min_key;
+	SET_BTREE_PTR_RANGE_UPDATED(&bp->v, f->range_updated);
+	memcpy(bp->v.start, f->ptrs, sizeof(struct bch_extent_ptr) * f->nr_ptrs);
+}
+
+static bool found_btree_node_is_readable(struct btree_trans *trans,
+					 const struct found_btree_node *f)
+{
+	struct { __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX); } k;
+
+	found_btree_node_to_key(&k.k, f);
+
+	struct btree *b = bch2_btree_node_get_noiter(trans, &k.k, f->btree_id, f->level, false);
+	bool ret = !IS_ERR_OR_NULL(b);
+	if (ret)
+		six_unlock_read(&b->c.lock);
+
+	/*
+	 * We might update this node's range; if that happens, we need the node
+	 * to be re-read so the read path can trim keys that are no longer in
+	 * this node
+	 */
+	if (b != btree_node_root(trans->c, b))
+		bch2_btree_node_evict(trans, &k.k);
+	return ret;
+}
+
+static int found_btree_node_cmp_cookie(const void *_l, const void *_r)
+{
+	const struct found_btree_node *l = _l;
+	const struct found_btree_node *r = _r;
+
+	return  cmp_int(l->btree_id,	r->btree_id) ?:
+		cmp_int(l->level,	r->level) ?:
+		cmp_int(l->cookie,	r->cookie);
+}
+
+/*
+ * Given two found btree nodes, if their sequence numbers are equal, take the
+ * one that's readable:
+ */
+static int found_btree_node_cmp_time(const struct found_btree_node *l,
+				     const struct found_btree_node *r)
+{
+	return cmp_int(l->seq, r->seq);
+}
+
+static int found_btree_node_cmp_pos(const void *_l, const void *_r)
+{
+	const struct found_btree_node *l = _l;
+	const struct found_btree_node *r = _r;
+
+	return  cmp_int(l->btree_id,	r->btree_id) ?:
+	       -cmp_int(l->level,	r->level) ?:
+		bpos_cmp(l->min_key,	r->min_key) ?:
+	       -found_btree_node_cmp_time(l, r);
+}
+
+static void try_read_btree_node(struct find_btree_nodes *f, struct bch_dev *ca,
+				struct bio *bio, struct btree_node *bn, u64 offset)
+{
+	struct bch_fs *c = container_of(f, struct bch_fs, found_btree_nodes);
+
+	bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ);
+	bio->bi_iter.bi_sector	= offset;
+	bch2_bio_map(bio, bn, PAGE_SIZE);
+
+	submit_bio_wait(bio);
+	if (bch2_dev_io_err_on(bio->bi_status, ca, BCH_MEMBER_ERROR_read,
+			       "IO error in try_read_btree_node() at %llu: %s",
+			       offset, bch2_blk_status_to_str(bio->bi_status)))
+		return;
+
+	if (le64_to_cpu(bn->magic) != bset_magic(c))
+		return;
+
+	if (bch2_csum_type_is_encryption(BSET_CSUM_TYPE(&bn->keys))) {
+		struct nonce nonce = btree_nonce(&bn->keys, 0);
+		unsigned bytes = (void *) &bn->keys - (void *) &bn->flags;
+
+		bch2_encrypt(c, BSET_CSUM_TYPE(&bn->keys), nonce, &bn->flags, bytes);
+	}
+
+	if (btree_id_is_alloc(BTREE_NODE_ID(bn)))
+		return;
+
+	if (BTREE_NODE_LEVEL(bn) >= BTREE_MAX_DEPTH)
+		return;
+
+	rcu_read_lock();
+	struct found_btree_node n = {
+		.btree_id	= BTREE_NODE_ID(bn),
+		.level		= BTREE_NODE_LEVEL(bn),
+		.seq		= BTREE_NODE_SEQ(bn),
+		.cookie		= le64_to_cpu(bn->keys.seq),
+		.min_key	= bn->min_key,
+		.max_key	= bn->max_key,
+		.nr_ptrs	= 1,
+		.ptrs[0].type	= 1 << BCH_EXTENT_ENTRY_ptr,
+		.ptrs[0].offset	= offset,
+		.ptrs[0].dev	= ca->dev_idx,
+		.ptrs[0].gen	= *bucket_gen(ca, sector_to_bucket(ca, offset)),
+	};
+	rcu_read_unlock();
+
+	if (bch2_trans_run(c, found_btree_node_is_readable(trans, &n))) {
+		mutex_lock(&f->lock);
+		if (BSET_BIG_ENDIAN(&bn->keys) != CPU_BIG_ENDIAN) {
+			bch_err(c, "try_read_btree_node() can't handle endian conversion");
+			f->ret = -EINVAL;
+			goto unlock;
+		}
+
+		if (darray_push(&f->nodes, n))
+			f->ret = -ENOMEM;
+unlock:
+		mutex_unlock(&f->lock);
+	}
+}
+
+static int read_btree_nodes_worker(void *p)
+{
+	struct find_btree_nodes_worker *w = p;
+	struct bch_fs *c = container_of(w->f, struct bch_fs, found_btree_nodes);
+	struct bch_dev *ca = w->ca;
+	void *buf = (void *) __get_free_page(GFP_KERNEL);
+	struct bio *bio = bio_alloc(NULL, 1, 0, GFP_KERNEL);
+	unsigned long last_print = jiffies;
+
+	if (!buf || !bio) {
+		bch_err(c, "read_btree_nodes_worker: error allocating bio/buf");
+		w->f->ret = -ENOMEM;
+		goto err;
+	}
+
+	for (u64 bucket = ca->mi.first_bucket; bucket < ca->mi.nbuckets; bucket++)
+		for (unsigned bucket_offset = 0;
+		     bucket_offset + btree_sectors(c) <= ca->mi.bucket_size;
+		     bucket_offset += btree_sectors(c)) {
+			if (time_after(jiffies, last_print + HZ * 30)) {
+				u64 cur_sector = bucket * ca->mi.bucket_size + bucket_offset;
+				u64 end_sector = ca->mi.nbuckets * ca->mi.bucket_size;
+
+				bch_info(ca, "%s: %2u%% done", __func__,
+					 (unsigned) div64_u64(cur_sector * 100, end_sector));
+				last_print = jiffies;
+			}
+
+			u64 sector = bucket * ca->mi.bucket_size + bucket_offset;
+
+			if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_mi_btree_bitmap &&
+			    !bch2_dev_btree_bitmap_marked_sectors(ca, sector, btree_sectors(c)))
+				continue;
+
+			try_read_btree_node(w->f, ca, bio, buf, sector);
+		}
+err:
+	bio_put(bio);
+	free_page((unsigned long) buf);
+	percpu_ref_get(&ca->io_ref);
+	closure_put(w->cl);
+	kfree(w);
+	return 0;
+}
+
+static int read_btree_nodes(struct find_btree_nodes *f)
+{
+	struct bch_fs *c = container_of(f, struct bch_fs, found_btree_nodes);
+	struct closure cl;
+	int ret = 0;
+
+	closure_init_stack(&cl);
+
+	for_each_online_member(c, ca) {
+		if (!(ca->mi.data_allowed & BIT(BCH_DATA_btree)))
+			continue;
+
+		struct find_btree_nodes_worker *w = kmalloc(sizeof(*w), GFP_KERNEL);
+		struct task_struct *t;
+
+		if (!w) {
+			percpu_ref_put(&ca->io_ref);
+			ret = -ENOMEM;
+			goto err;
+		}
+
+		percpu_ref_get(&ca->io_ref);
+		closure_get(&cl);
+		w->cl		= &cl;
+		w->f		= f;
+		w->ca		= ca;
+
+		t = kthread_run(read_btree_nodes_worker, w, "read_btree_nodes/%s", ca->name);
+		ret = IS_ERR_OR_NULL(t);
+		if (ret) {
+			percpu_ref_put(&ca->io_ref);
+			closure_put(&cl);
+			f->ret = ret;
+			bch_err(c, "error starting kthread: %i", ret);
+			break;
+		}
+	}
+err:
+	closure_sync(&cl);
+	return f->ret ?: ret;
+}
+
+static void bubble_up(struct found_btree_node *n, struct found_btree_node *end)
+{
+	while (n + 1 < end &&
+	       found_btree_node_cmp_pos(n, n + 1) > 0) {
+		swap(n[0], n[1]);
+		n++;
+	}
+}
+
+static int handle_overwrites(struct bch_fs *c,
+			     struct found_btree_node *start,
+			     struct found_btree_node *end)
+{
+	struct found_btree_node *n;
+again:
+	for (n = start + 1;
+	     n < end &&
+	     n->btree_id	== start->btree_id &&
+	     n->level		== start->level &&
+	     bpos_lt(n->min_key, start->max_key);
+	     n++)  {
+		int cmp = found_btree_node_cmp_time(start, n);
+
+		if (cmp > 0) {
+			if (bpos_cmp(start->max_key, n->max_key) >= 0)
+				n->overwritten = true;
+			else {
+				n->range_updated = true;
+				n->min_key = bpos_successor(start->max_key);
+				n->range_updated = true;
+				bubble_up(n, end);
+				goto again;
+			}
+		} else if (cmp < 0) {
+			BUG_ON(bpos_cmp(n->min_key, start->min_key) <= 0);
+
+			start->max_key = bpos_predecessor(n->min_key);
+			start->range_updated = true;
+		} else {
+			struct printbuf buf = PRINTBUF;
+
+			prt_str(&buf, "overlapping btree nodes with same seq! halting\n  ");
+			found_btree_node_to_text(&buf, c, start);
+			prt_str(&buf, "\n  ");
+			found_btree_node_to_text(&buf, c, n);
+			bch_err(c, "%s", buf.buf);
+			printbuf_exit(&buf);
+			return -BCH_ERR_fsck_repair_unimplemented;
+		}
+	}
+
+	return 0;
+}
+
+int bch2_scan_for_btree_nodes(struct bch_fs *c)
+{
+	struct find_btree_nodes *f = &c->found_btree_nodes;
+	struct printbuf buf = PRINTBUF;
+	size_t dst;
+	int ret = 0;
+
+	if (f->nodes.nr)
+		return 0;
+
+	mutex_init(&f->lock);
+
+	ret = read_btree_nodes(f);
+	if (ret)
+		return ret;
+
+	if (!f->nodes.nr) {
+		bch_err(c, "%s: no btree nodes found", __func__);
+		ret = -EINVAL;
+		goto err;
+	}
+
+	if (0 && c->opts.verbose) {
+		printbuf_reset(&buf);
+		prt_printf(&buf, "%s: nodes found:\n", __func__);
+		found_btree_nodes_to_text(&buf, c, f->nodes);
+		bch2_print_string_as_lines(KERN_INFO, buf.buf);
+	}
+
+	sort(f->nodes.data, f->nodes.nr, sizeof(f->nodes.data[0]), found_btree_node_cmp_cookie, NULL);
+
+	dst = 0;
+	darray_for_each(f->nodes, i) {
+		struct found_btree_node *prev = dst ? f->nodes.data + dst - 1 : NULL;
+
+		if (prev &&
+		    prev->cookie == i->cookie) {
+			if (prev->nr_ptrs == ARRAY_SIZE(prev->ptrs)) {
+				bch_err(c, "%s: found too many replicas for btree node", __func__);
+				ret = -EINVAL;
+				goto err;
+			}
+			prev->ptrs[prev->nr_ptrs++] = i->ptrs[0];
+		} else {
+			f->nodes.data[dst++] = *i;
+		}
+	}
+	f->nodes.nr = dst;
+
+	sort(f->nodes.data, f->nodes.nr, sizeof(f->nodes.data[0]), found_btree_node_cmp_pos, NULL);
+
+	if (0 && c->opts.verbose) {
+		printbuf_reset(&buf);
+		prt_printf(&buf, "%s: nodes after merging replicas:\n", __func__);
+		found_btree_nodes_to_text(&buf, c, f->nodes);
+		bch2_print_string_as_lines(KERN_INFO, buf.buf);
+	}
+
+	dst = 0;
+	darray_for_each(f->nodes, i) {
+		if (i->overwritten)
+			continue;
+
+		ret = handle_overwrites(c, i, &darray_top(f->nodes));
+		if (ret)
+			goto err;
+
+		BUG_ON(i->overwritten);
+		f->nodes.data[dst++] = *i;
+	}
+	f->nodes.nr = dst;
+
+	if (c->opts.verbose) {
+		printbuf_reset(&buf);
+		prt_printf(&buf, "%s: nodes found after overwrites:\n", __func__);
+		found_btree_nodes_to_text(&buf, c, f->nodes);
+		bch2_print_string_as_lines(KERN_INFO, buf.buf);
+	}
+
+	eytzinger0_sort(f->nodes.data, f->nodes.nr, sizeof(f->nodes.data[0]), found_btree_node_cmp_pos, NULL);
+err:
+	printbuf_exit(&buf);
+	return ret;
+}
+
+static int found_btree_node_range_start_cmp(const void *_l, const void *_r)
+{
+	const struct found_btree_node *l = _l;
+	const struct found_btree_node *r = _r;
+
+	return  cmp_int(l->btree_id,	r->btree_id) ?:
+	       -cmp_int(l->level,	r->level) ?:
+		bpos_cmp(l->max_key,	r->min_key);
+}
+
+#define for_each_found_btree_node_in_range(_f, _search, _idx)				\
+	for (size_t _idx = eytzinger0_find_gt((_f)->nodes.data, (_f)->nodes.nr,		\
+					sizeof((_f)->nodes.data[0]),			\
+					found_btree_node_range_start_cmp, &search);	\
+	     _idx < (_f)->nodes.nr &&							\
+	     (_f)->nodes.data[_idx].btree_id == _search.btree_id &&			\
+	     (_f)->nodes.data[_idx].level == _search.level &&				\
+	     bpos_lt((_f)->nodes.data[_idx].min_key, _search.max_key);			\
+	     _idx = eytzinger0_next(_idx, (_f)->nodes.nr))
+
+bool bch2_btree_node_is_stale(struct bch_fs *c, struct btree *b)
+{
+	struct find_btree_nodes *f = &c->found_btree_nodes;
+
+	struct found_btree_node search = {
+		.btree_id	= b->c.btree_id,
+		.level		= b->c.level,
+		.min_key	= b->data->min_key,
+		.max_key	= b->key.k.p,
+	};
+
+	for_each_found_btree_node_in_range(f, search, idx)
+		if (f->nodes.data[idx].seq > BTREE_NODE_SEQ(b->data))
+			return true;
+	return false;
+}
+
+bool bch2_btree_has_scanned_nodes(struct bch_fs *c, enum btree_id btree)
+{
+	struct found_btree_node search = {
+		.btree_id	= btree,
+		.level		= 0,
+		.min_key	= POS_MIN,
+		.max_key	= SPOS_MAX,
+	};
+
+	for_each_found_btree_node_in_range(&c->found_btree_nodes, search, idx)
+		return true;
+	return false;
+}
+
+int bch2_get_scanned_nodes(struct bch_fs *c, enum btree_id btree,
+			   unsigned level, struct bpos node_min, struct bpos node_max)
+{
+	if (btree_id_is_alloc(btree))
+		return 0;
+
+	struct find_btree_nodes *f = &c->found_btree_nodes;
+
+	int ret = bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_scan_for_btree_nodes);
+	if (ret)
+		return ret;
+
+	if (c->opts.verbose) {
+		struct printbuf buf = PRINTBUF;
+
+		prt_printf(&buf, "recovering %s l=%u ", bch2_btree_id_str(btree), level);
+		bch2_bpos_to_text(&buf, node_min);
+		prt_str(&buf, " - ");
+		bch2_bpos_to_text(&buf, node_max);
+
+		bch_info(c, "%s(): %s", __func__, buf.buf);
+		printbuf_exit(&buf);
+	}
+
+	struct found_btree_node search = {
+		.btree_id	= btree,
+		.level		= level,
+		.min_key	= node_min,
+		.max_key	= node_max,
+	};
+
+	for_each_found_btree_node_in_range(f, search, idx) {
+		struct found_btree_node n = f->nodes.data[idx];
+
+		n.range_updated |= bpos_lt(n.min_key, node_min);
+		n.min_key = bpos_max(n.min_key, node_min);
+
+		n.range_updated |= bpos_gt(n.max_key, node_max);
+		n.max_key = bpos_min(n.max_key, node_max);
+
+		struct { __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX); } tmp;
+
+		found_btree_node_to_key(&tmp.k, &n);
+
+		struct printbuf buf = PRINTBUF;
+		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&tmp.k));
+		bch_verbose(c, "%s(): recovering %s", __func__, buf.buf);
+		printbuf_exit(&buf);
+
+		BUG_ON(bch2_bkey_invalid(c, bkey_i_to_s_c(&tmp.k), BKEY_TYPE_btree, 0, NULL));
+
+		ret = bch2_journal_key_insert(c, btree, level + 1, &tmp.k);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+void bch2_find_btree_nodes_exit(struct find_btree_nodes *f)
+{
+	darray_exit(&f->nodes);
+}
diff --git a/fs/bcachefs/btree_node_scan.h b/fs/bcachefs/btree_node_scan.h
new file mode 100644
index 000000000000..08687b209787
--- /dev/null
+++ b/fs/bcachefs/btree_node_scan.h
@@ -0,0 +1,11 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BCACHEFS_BTREE_NODE_SCAN_H
+#define _BCACHEFS_BTREE_NODE_SCAN_H
+
+int bch2_scan_for_btree_nodes(struct bch_fs *);
+bool bch2_btree_node_is_stale(struct bch_fs *, struct btree *);
+bool bch2_btree_has_scanned_nodes(struct bch_fs *, enum btree_id);
+int bch2_get_scanned_nodes(struct bch_fs *, enum btree_id, unsigned, struct bpos, struct bpos);
+void bch2_find_btree_nodes_exit(struct find_btree_nodes *);
+
+#endif /* _BCACHEFS_BTREE_NODE_SCAN_H */
diff --git a/fs/bcachefs/btree_node_scan_types.h b/fs/bcachefs/btree_node_scan_types.h
new file mode 100644
index 000000000000..abb7b27d556a
--- /dev/null
+++ b/fs/bcachefs/btree_node_scan_types.h
@@ -0,0 +1,30 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BCACHEFS_BTREE_NODE_SCAN_TYPES_H
+#define _BCACHEFS_BTREE_NODE_SCAN_TYPES_H
+
+#include "darray.h"
+
+struct found_btree_node {
+	bool			range_updated:1;
+	bool			overwritten:1;
+	u8			btree_id;
+	u8			level;
+	u32			seq;
+	u64			cookie;
+
+	struct bpos		min_key;
+	struct bpos		max_key;
+
+	unsigned		nr_ptrs;
+	struct bch_extent_ptr	ptrs[BCH_REPLICAS_MAX];
+};
+
+typedef DARRAY(struct found_btree_node)	found_btree_nodes;
+
+struct find_btree_nodes {
+	int			ret;
+	struct mutex		lock;
+	found_btree_nodes	nodes;
+};
+
+#endif /* _BCACHEFS_BTREE_NODE_SCAN_TYPES_H */
diff --git a/fs/bcachefs/btree_trans_commit.c b/fs/bcachefs/btree_trans_commit.c
index 30d69a6d133e..bbec91e8e650 100644
--- a/fs/bcachefs/btree_trans_commit.c
+++ b/fs/bcachefs/btree_trans_commit.c
@@ -318,7 +318,7 @@ static inline void btree_insert_entry_checks(struct btree_trans *trans,
 		!(i->flags & BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) &&
 		test_bit(JOURNAL_REPLAY_DONE, &trans->c->journal.flags) &&
 		i->k->k.p.snapshot &&
-		bch2_snapshot_is_internal_node(trans->c, i->k->k.p.snapshot));
+		bch2_snapshot_is_internal_node(trans->c, i->k->k.p.snapshot) > 0);
 }
 
 static __always_inline int bch2_trans_journal_res_get(struct btree_trans *trans,
@@ -397,12 +397,13 @@ static int btree_key_can_insert_cached(struct btree_trans *trans, unsigned flags
 	struct bkey_cached *ck = (void *) path->l[0].b;
 	unsigned new_u64s;
 	struct bkey_i *new_k;
+	unsigned watermark = flags & BCH_WATERMARK_MASK;
 
 	EBUG_ON(path->level);
 
-	if (!test_bit(BKEY_CACHED_DIRTY, &ck->flags) &&
-	    bch2_btree_key_cache_must_wait(c) &&
-	    !(flags & BCH_TRANS_COMMIT_journal_reclaim))
+	if (watermark < BCH_WATERMARK_reclaim &&
+	    !test_bit(BKEY_CACHED_DIRTY, &ck->flags) &&
+	    bch2_btree_key_cache_must_wait(c))
 		return -BCH_ERR_btree_insert_need_journal_reclaim;
 
 	/*
@@ -499,9 +500,8 @@ static int run_one_trans_trigger(struct btree_trans *trans, struct btree_insert_
 }
 
 static int run_btree_triggers(struct btree_trans *trans, enum btree_id btree_id,
-			      struct btree_insert_entry *btree_id_start)
+			      unsigned btree_id_start)
 {
-	struct btree_insert_entry *i;
 	bool trans_trigger_run;
 	int ret, overwrite;
 
@@ -514,13 +514,13 @@ static int run_btree_triggers(struct btree_trans *trans, enum btree_id btree_id,
 		do {
 			trans_trigger_run = false;
 
-			for (i = btree_id_start;
-			     i < trans->updates + trans->nr_updates && i->btree_id <= btree_id;
+			for (unsigned i = btree_id_start;
+			     i < trans->nr_updates && trans->updates[i].btree_id <= btree_id;
 			     i++) {
-				if (i->btree_id != btree_id)
+				if (trans->updates[i].btree_id != btree_id)
 					continue;
 
-				ret = run_one_trans_trigger(trans, i, overwrite);
+				ret = run_one_trans_trigger(trans, trans->updates + i, overwrite);
 				if (ret < 0)
 					return ret;
 				if (ret)
@@ -534,8 +534,7 @@ static int run_btree_triggers(struct btree_trans *trans, enum btree_id btree_id,
 
 static int bch2_trans_commit_run_triggers(struct btree_trans *trans)
 {
-	struct btree_insert_entry *btree_id_start = trans->updates;
-	unsigned btree_id = 0;
+	unsigned btree_id = 0, btree_id_start = 0;
 	int ret = 0;
 
 	/*
@@ -549,8 +548,8 @@ static int bch2_trans_commit_run_triggers(struct btree_trans *trans)
 		if (btree_id == BTREE_ID_alloc)
 			continue;
 
-		while (btree_id_start < trans->updates + trans->nr_updates &&
-		       btree_id_start->btree_id < btree_id)
+		while (btree_id_start < trans->nr_updates &&
+		       trans->updates[btree_id_start].btree_id < btree_id)
 			btree_id_start++;
 
 		ret = run_btree_triggers(trans, btree_id, btree_id_start);
@@ -558,11 +557,13 @@ static int bch2_trans_commit_run_triggers(struct btree_trans *trans)
 			return ret;
 	}
 
-	trans_for_each_update(trans, i) {
+	for (unsigned idx = 0; idx < trans->nr_updates; idx++) {
+		struct btree_insert_entry *i = trans->updates + idx;
+
 		if (i->btree_id > BTREE_ID_alloc)
 			break;
 		if (i->btree_id == BTREE_ID_alloc) {
-			ret = run_btree_triggers(trans, BTREE_ID_alloc, i);
+			ret = run_btree_triggers(trans, BTREE_ID_alloc, idx);
 			if (ret)
 				return ret;
 			break;
@@ -826,7 +827,8 @@ static inline int do_bch2_trans_commit(struct btree_trans *trans, unsigned flags
 	struct bch_fs *c = trans->c;
 	int ret = 0, u64s_delta = 0;
 
-	trans_for_each_update(trans, i) {
+	for (unsigned idx = 0; idx < trans->nr_updates; idx++) {
+		struct btree_insert_entry *i = trans->updates + idx;
 		if (i->cached)
 			continue;
 
@@ -887,6 +889,7 @@ int bch2_trans_commit_error(struct btree_trans *trans, unsigned flags,
 			    int ret, unsigned long trace_ip)
 {
 	struct bch_fs *c = trans->c;
+	enum bch_watermark watermark = flags & BCH_WATERMARK_MASK;
 
 	switch (ret) {
 	case -BCH_ERR_btree_insert_btree_node_full:
@@ -905,7 +908,7 @@ int bch2_trans_commit_error(struct btree_trans *trans, unsigned flags,
 		 * flag
 		 */
 		if ((flags & BCH_TRANS_COMMIT_journal_reclaim) &&
-		    (flags & BCH_WATERMARK_MASK) != BCH_WATERMARK_reclaim) {
+		    watermark < BCH_WATERMARK_reclaim) {
 			ret = -BCH_ERR_journal_reclaim_would_deadlock;
 			break;
 		}
diff --git a/fs/bcachefs/btree_types.h b/fs/bcachefs/btree_types.h
index 9404d96c38f3..e0c982a4195c 100644
--- a/fs/bcachefs/btree_types.h
+++ b/fs/bcachefs/btree_types.h
@@ -364,7 +364,21 @@ struct btree_insert_entry {
 	unsigned long		ip_allocated;
 };
 
+/* Number of btree paths we preallocate, usually enough */
 #define BTREE_ITER_INITIAL		64
+/*
+ * Lmiit for btree_trans_too_many_iters(); this is enough that almost all code
+ * paths should run inside this limit, and if they don't it usually indicates a
+ * bug (leaking/duplicated btree paths).
+ *
+ * exception: some fsck paths
+ *
+ * bugs with excessive path usage seem to have possibly been eliminated now, so
+ * we might consider eliminating this (and btree_trans_too_many_iter()) at some
+ * point.
+ */
+#define BTREE_ITER_NORMAL_LIMIT		256
+/* never exceed limit */
 #define BTREE_ITER_MAX			(1U << 10)
 
 struct btree_trans_commit_hook;
diff --git a/fs/bcachefs/btree_update.c b/fs/bcachefs/btree_update.c
index a4b40c1656a5..8e47e260eba5 100644
--- a/fs/bcachefs/btree_update.c
+++ b/fs/bcachefs/btree_update.c
@@ -38,6 +38,9 @@ static noinline int extent_front_merge(struct btree_trans *trans,
 	struct bkey_i *update;
 	int ret;
 
+	if (unlikely(trans->journal_replay_not_finished))
+		return 0;
+
 	update = bch2_bkey_make_mut_noupdate(trans, k);
 	ret = PTR_ERR_OR_ZERO(update);
 	if (ret)
@@ -69,6 +72,9 @@ static noinline int extent_back_merge(struct btree_trans *trans,
 	struct bch_fs *c = trans->c;
 	int ret;
 
+	if (unlikely(trans->journal_replay_not_finished))
+		return 0;
+
 	ret =   bch2_key_has_snapshot_overwrites(trans, iter->btree_id, insert->k.p) ?:
 		bch2_key_has_snapshot_overwrites(trans, iter->btree_id, k.k->p);
 	if (ret < 0)
diff --git a/fs/bcachefs/btree_update_interior.c b/fs/bcachefs/btree_update_interior.c
index b2f5f2e50f7e..6030c396754f 100644
--- a/fs/bcachefs/btree_update_interior.c
+++ b/fs/bcachefs/btree_update_interior.c
@@ -2,6 +2,7 @@
 
 #include "bcachefs.h"
 #include "alloc_foreground.h"
+#include "bkey_buf.h"
 #include "bkey_methods.h"
 #include "btree_cache.h"
 #include "btree_gc.h"
@@ -18,12 +19,21 @@
 #include "journal.h"
 #include "journal_reclaim.h"
 #include "keylist.h"
+#include "recovery_passes.h"
 #include "replicas.h"
+#include "sb-members.h"
 #include "super-io.h"
 #include "trace.h"
 
 #include <linux/random.h>
 
+static const char * const bch2_btree_update_modes[] = {
+#define x(t) #t,
+	BTREE_UPDATE_MODES()
+#undef x
+	NULL
+};
+
 static int bch2_btree_insert_node(struct btree_update *, struct btree_trans *,
 				  btree_path_idx_t, struct btree *, struct keylist *);
 static void bch2_btree_update_add_new_node(struct btree_update *, struct btree *);
@@ -44,56 +54,103 @@ static btree_path_idx_t get_unlocked_mut_path(struct btree_trans *trans,
 	return path_idx;
 }
 
-/* Debug code: */
-
 /*
  * Verify that child nodes correctly span parent node's range:
  */
-static void btree_node_interior_verify(struct bch_fs *c, struct btree *b)
+int bch2_btree_node_check_topology(struct btree_trans *trans, struct btree *b)
 {
-#ifdef CONFIG_BCACHEFS_DEBUG
-	struct bpos next_node = b->data->min_key;
-	struct btree_node_iter iter;
+	struct bch_fs *c = trans->c;
+	struct bpos node_min = b->key.k.type == KEY_TYPE_btree_ptr_v2
+		? bkey_i_to_btree_ptr_v2(&b->key)->v.min_key
+		: b->data->min_key;
+	struct btree_and_journal_iter iter;
 	struct bkey_s_c k;
-	struct bkey_s_c_btree_ptr_v2 bp;
-	struct bkey unpacked;
-	struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
+	struct printbuf buf = PRINTBUF;
+	struct bkey_buf prev;
+	int ret = 0;
 
-	BUG_ON(!b->c.level);
+	BUG_ON(b->key.k.type == KEY_TYPE_btree_ptr_v2 &&
+	       !bpos_eq(bkey_i_to_btree_ptr_v2(&b->key)->v.min_key,
+			b->data->min_key));
 
-	if (!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags))
-		return;
+	if (!b->c.level)
+		return 0;
 
-	bch2_btree_node_iter_init_from_start(&iter, b);
+	bch2_bkey_buf_init(&prev);
+	bkey_init(&prev.k->k);
+	bch2_btree_and_journal_iter_init_node_iter(trans, &iter, b);
 
-	while (1) {
-		k = bch2_btree_node_iter_peek_unpack(&iter, b, &unpacked);
+	while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 		if (k.k->type != KEY_TYPE_btree_ptr_v2)
-			break;
-		bp = bkey_s_c_to_btree_ptr_v2(k);
+			goto out;
 
-		if (!bpos_eq(next_node, bp.v->min_key)) {
-			bch2_dump_btree_node(c, b);
-			bch2_bpos_to_text(&buf1, next_node);
-			bch2_bpos_to_text(&buf2, bp.v->min_key);
-			panic("expected next min_key %s got %s\n", buf1.buf, buf2.buf);
-		}
+		struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
 
-		bch2_btree_node_iter_advance(&iter, b);
+		struct bpos expected_min = bkey_deleted(&prev.k->k)
+			? node_min
+			: bpos_successor(prev.k->k.p);
 
-		if (bch2_btree_node_iter_end(&iter)) {
-			if (!bpos_eq(k.k->p, b->key.k.p)) {
-				bch2_dump_btree_node(c, b);
-				bch2_bpos_to_text(&buf1, b->key.k.p);
-				bch2_bpos_to_text(&buf2, k.k->p);
-				panic("expected end %s got %s\n", buf1.buf, buf2.buf);
-			}
-			break;
+		if (!bpos_eq(expected_min, bp.v->min_key)) {
+			bch2_topology_error(c);
+
+			printbuf_reset(&buf);
+			prt_str(&buf, "end of prev node doesn't match start of next node\n"),
+			prt_printf(&buf, "  in btree %s level %u node ",
+				   bch2_btree_id_str(b->c.btree_id), b->c.level);
+			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
+			prt_str(&buf, "\n  prev ");
+			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(prev.k));
+			prt_str(&buf, "\n  next ");
+			bch2_bkey_val_to_text(&buf, c, k);
+
+			need_fsck_err(c, btree_node_topology_bad_min_key, "%s", buf.buf);
+			goto topology_repair;
 		}
 
-		next_node = bpos_successor(k.k->p);
+		bch2_bkey_buf_reassemble(&prev, c, k);
+		bch2_btree_and_journal_iter_advance(&iter);
+	}
+
+	if (bkey_deleted(&prev.k->k)) {
+		bch2_topology_error(c);
+
+		printbuf_reset(&buf);
+		prt_str(&buf, "empty interior node\n");
+		prt_printf(&buf, "  in btree %s level %u node ",
+			   bch2_btree_id_str(b->c.btree_id), b->c.level);
+		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
+
+		need_fsck_err(c, btree_node_topology_empty_interior_node, "%s", buf.buf);
+		goto topology_repair;
+	} else if (!bpos_eq(prev.k->k.p, b->key.k.p)) {
+		bch2_topology_error(c);
+
+		printbuf_reset(&buf);
+		prt_str(&buf, "last child node doesn't end at end of parent node\n");
+		prt_printf(&buf, "  in btree %s level %u node ",
+			   bch2_btree_id_str(b->c.btree_id), b->c.level);
+		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
+		prt_str(&buf, "\n  last key ");
+		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(prev.k));
+
+		need_fsck_err(c, btree_node_topology_bad_max_key, "%s", buf.buf);
+		goto topology_repair;
 	}
-#endif
+out:
+fsck_err:
+	bch2_btree_and_journal_iter_exit(&iter);
+	bch2_bkey_buf_exit(&prev, c);
+	printbuf_exit(&buf);
+	return ret;
+topology_repair:
+	if ((c->recovery_passes_explicit & BIT_ULL(BCH_RECOVERY_PASS_check_topology)) &&
+	    c->curr_recovery_pass > BCH_RECOVERY_PASS_check_topology) {
+		bch2_inconsistent_error(c);
+		ret = -BCH_ERR_btree_need_topology_repair;
+	} else {
+		ret = bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_topology);
+	}
+	goto out;
 }
 
 /* Calculate ideal packed bkey format for new btree nodes: */
@@ -254,7 +311,7 @@ static struct btree *__bch2_btree_node_alloc(struct btree_trans *trans,
 	struct open_buckets obs = { .nr = 0 };
 	struct bch_devs_list devs_have = (struct bch_devs_list) { 0 };
 	enum bch_watermark watermark = flags & BCH_WATERMARK_MASK;
-	unsigned nr_reserve = watermark > BCH_WATERMARK_reclaim
+	unsigned nr_reserve = watermark < BCH_WATERMARK_reclaim
 		? BTREE_NODE_RESERVE
 		: 0;
 	int ret;
@@ -549,6 +606,26 @@ static void btree_update_add_key(struct btree_update *as,
 	bch2_keylist_push(keys);
 }
 
+static bool btree_update_new_nodes_marked_sb(struct btree_update *as)
+{
+	for_each_keylist_key(&as->new_keys, k)
+		if (!bch2_dev_btree_bitmap_marked(as->c, bkey_i_to_s_c(k)))
+			return false;
+	return true;
+}
+
+static void btree_update_new_nodes_mark_sb(struct btree_update *as)
+{
+	struct bch_fs *c = as->c;
+
+	mutex_lock(&c->sb_lock);
+	for_each_keylist_key(&as->new_keys, k)
+		bch2_dev_btree_bitmap_mark(c, bkey_i_to_s_c(k));
+
+	bch2_write_super(c);
+	mutex_unlock(&c->sb_lock);
+}
+
 /*
  * The transactional part of an interior btree node update, where we journal the
  * update we did to the interior node and update alloc info:
@@ -606,6 +683,9 @@ static void btree_update_nodes_written(struct btree_update *as)
 	if (ret)
 		goto err;
 
+	if (!btree_update_new_nodes_marked_sb(as))
+		btree_update_new_nodes_mark_sb(as);
+
 	/*
 	 * Wait for any in flight writes to finish before we free the old nodes
 	 * on disk:
@@ -638,7 +718,7 @@ static void btree_update_nodes_written(struct btree_update *as)
 	 * which may require allocations as well.
 	 */
 	ret = commit_do(trans, &as->disk_res, &journal_seq,
-			BCH_WATERMARK_reclaim|
+			BCH_WATERMARK_interior_updates|
 			BCH_TRANS_COMMIT_no_enospc|
 			BCH_TRANS_COMMIT_no_check_rw|
 			BCH_TRANS_COMMIT_journal_reclaim,
@@ -648,9 +728,13 @@ static void btree_update_nodes_written(struct btree_update *as)
 	bch2_fs_fatal_err_on(ret && !bch2_journal_error(&c->journal), c,
 			     "%s", bch2_err_str(ret));
 err:
-	if (as->b) {
-
-		b = as->b;
+	/*
+	 * We have to be careful because another thread might be getting ready
+	 * to free as->b and calling btree_update_reparent() on us - we'll
+	 * recheck under btree_update_lock below:
+	 */
+	b = READ_ONCE(as->b);
+	if (b) {
 		btree_path_idx_t path_idx = get_unlocked_mut_path(trans,
 						as->btree_id, b->c.level, b->key.k.p);
 		struct btree_path *path = trans->paths + path_idx;
@@ -794,15 +878,17 @@ static void btree_update_updated_node(struct btree_update *as, struct btree *b)
 {
 	struct bch_fs *c = as->c;
 
-	mutex_lock(&c->btree_interior_update_lock);
-	list_add_tail(&as->unwritten_list, &c->btree_interior_updates_unwritten);
-
-	BUG_ON(as->mode != BTREE_INTERIOR_NO_UPDATE);
+	BUG_ON(as->mode != BTREE_UPDATE_none);
+	BUG_ON(as->update_level_end < b->c.level);
 	BUG_ON(!btree_node_dirty(b));
 	BUG_ON(!b->c.level);
 
-	as->mode	= BTREE_INTERIOR_UPDATING_NODE;
+	mutex_lock(&c->btree_interior_update_lock);
+	list_add_tail(&as->unwritten_list, &c->btree_interior_updates_unwritten);
+
+	as->mode	= BTREE_UPDATE_node;
 	as->b		= b;
+	as->update_level_end = b->c.level;
 
 	set_btree_node_write_blocked(b);
 	list_add(&as->write_blocked_list, &b->write_blocked);
@@ -824,7 +910,7 @@ static void btree_update_reparent(struct btree_update *as,
 	lockdep_assert_held(&c->btree_interior_update_lock);
 
 	child->b = NULL;
-	child->mode = BTREE_INTERIOR_UPDATING_AS;
+	child->mode = BTREE_UPDATE_update;
 
 	bch2_journal_pin_copy(&c->journal, &as->journal, &child->journal,
 			      bch2_update_reparent_journal_pin_flush);
@@ -835,7 +921,7 @@ static void btree_update_updated_root(struct btree_update *as, struct btree *b)
 	struct bkey_i *insert = &b->key;
 	struct bch_fs *c = as->c;
 
-	BUG_ON(as->mode != BTREE_INTERIOR_NO_UPDATE);
+	BUG_ON(as->mode != BTREE_UPDATE_none);
 
 	BUG_ON(as->journal_u64s + jset_u64s(insert->k.u64s) >
 	       ARRAY_SIZE(as->journal_entries));
@@ -849,7 +935,7 @@ static void btree_update_updated_root(struct btree_update *as, struct btree *b)
 	mutex_lock(&c->btree_interior_update_lock);
 	list_add_tail(&as->unwritten_list, &c->btree_interior_updates_unwritten);
 
-	as->mode	= BTREE_INTERIOR_UPDATING_ROOT;
+	as->mode	= BTREE_UPDATE_root;
 	mutex_unlock(&c->btree_interior_update_lock);
 }
 
@@ -1027,7 +1113,7 @@ static void bch2_btree_update_done(struct btree_update *as, struct btree_trans *
 	struct bch_fs *c = as->c;
 	u64 start_time = as->start_time;
 
-	BUG_ON(as->mode == BTREE_INTERIOR_NO_UPDATE);
+	BUG_ON(as->mode == BTREE_UPDATE_none);
 
 	if (as->took_gc_lock)
 		up_read(&as->c->gc_lock);
@@ -1044,7 +1130,7 @@ static void bch2_btree_update_done(struct btree_update *as, struct btree_trans *
 
 static struct btree_update *
 bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
-			unsigned level, bool split, unsigned flags)
+			unsigned level_start, bool split, unsigned flags)
 {
 	struct bch_fs *c = trans->c;
 	struct btree_update *as;
@@ -1052,7 +1138,7 @@ bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
 	int disk_res_flags = (flags & BCH_TRANS_COMMIT_no_enospc)
 		? BCH_DISK_RESERVATION_NOFAIL : 0;
 	unsigned nr_nodes[2] = { 0, 0 };
-	unsigned update_level = level;
+	unsigned level_end = level_start;
 	enum bch_watermark watermark = flags & BCH_WATERMARK_MASK;
 	int ret = 0;
 	u32 restart_count = trans->restart_count;
@@ -1067,34 +1153,30 @@ bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
 	flags &= ~BCH_WATERMARK_MASK;
 	flags |= watermark;
 
-	if (watermark < c->journal.watermark) {
-		struct journal_res res = { 0 };
-		unsigned journal_flags = watermark|JOURNAL_RES_GET_CHECK;
-
-		if ((flags & BCH_TRANS_COMMIT_journal_reclaim) &&
-		    watermark != BCH_WATERMARK_reclaim)
-			journal_flags |= JOURNAL_RES_GET_NONBLOCK;
+	if (watermark < BCH_WATERMARK_reclaim &&
+	    test_bit(JOURNAL_SPACE_LOW, &c->journal.flags)) {
+		if (flags & BCH_TRANS_COMMIT_journal_reclaim)
+			return ERR_PTR(-BCH_ERR_journal_reclaim_would_deadlock);
 
-		ret = drop_locks_do(trans,
-			bch2_journal_res_get(&c->journal, &res, 1, journal_flags));
-		if (bch2_err_matches(ret, BCH_ERR_operation_blocked))
-			ret = -BCH_ERR_journal_reclaim_would_deadlock;
+		bch2_trans_unlock(trans);
+		wait_event(c->journal.wait, !test_bit(JOURNAL_SPACE_LOW, &c->journal.flags));
+		ret = bch2_trans_relock(trans);
 		if (ret)
 			return ERR_PTR(ret);
 	}
 
 	while (1) {
-		nr_nodes[!!update_level] += 1 + split;
-		update_level++;
+		nr_nodes[!!level_end] += 1 + split;
+		level_end++;
 
-		ret = bch2_btree_path_upgrade(trans, path, update_level + 1);
+		ret = bch2_btree_path_upgrade(trans, path, level_end + 1);
 		if (ret)
 			return ERR_PTR(ret);
 
-		if (!btree_path_node(path, update_level)) {
+		if (!btree_path_node(path, level_end)) {
 			/* Allocating new root? */
 			nr_nodes[1] += split;
-			update_level = BTREE_MAX_DEPTH;
+			level_end = BTREE_MAX_DEPTH;
 			break;
 		}
 
@@ -1102,11 +1184,11 @@ bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
 		 * Always check for space for two keys, even if we won't have to
 		 * split at prior level - it might have been a merge instead:
 		 */
-		if (bch2_btree_node_insert_fits(path->l[update_level].b,
+		if (bch2_btree_node_insert_fits(path->l[level_end].b,
 						BKEY_BTREE_PTR_U64s_MAX * 2))
 			break;
 
-		split = path->l[update_level].b->nr.live_u64s > BTREE_SPLIT_THRESHOLD(c);
+		split = path->l[level_end].b->nr.live_u64s > BTREE_SPLIT_THRESHOLD(c);
 	}
 
 	if (!down_read_trylock(&c->gc_lock)) {
@@ -1120,13 +1202,15 @@ bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
 	as = mempool_alloc(&c->btree_interior_update_pool, GFP_NOFS);
 	memset(as, 0, sizeof(*as));
 	closure_init(&as->cl, NULL);
-	as->c		= c;
-	as->start_time	= start_time;
-	as->ip_started	= _RET_IP_;
-	as->mode	= BTREE_INTERIOR_NO_UPDATE;
-	as->took_gc_lock = true;
-	as->btree_id	= path->btree_id;
-	as->update_level = update_level;
+	as->c			= c;
+	as->start_time		= start_time;
+	as->ip_started		= _RET_IP_;
+	as->mode		= BTREE_UPDATE_none;
+	as->watermark		= watermark;
+	as->took_gc_lock	= true;
+	as->btree_id		= path->btree_id;
+	as->update_level_start	= level_start;
+	as->update_level_end	= level_end;
 	INIT_LIST_HEAD(&as->list);
 	INIT_LIST_HEAD(&as->unwritten_list);
 	INIT_LIST_HEAD(&as->write_blocked_list);
@@ -1168,7 +1252,7 @@ bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
 		 */
 		if (bch2_err_matches(ret, ENOSPC) &&
 		    (flags & BCH_TRANS_COMMIT_journal_reclaim) &&
-		    watermark != BCH_WATERMARK_reclaim) {
+		    watermark < BCH_WATERMARK_reclaim) {
 			ret = -BCH_ERR_journal_reclaim_would_deadlock;
 			goto err;
 		}
@@ -1220,23 +1304,29 @@ static void bch2_btree_set_root_inmem(struct bch_fs *c, struct btree *b)
 	bch2_recalc_btree_reserve(c);
 }
 
-static void bch2_btree_set_root(struct btree_update *as,
-				struct btree_trans *trans,
-				struct btree_path *path,
-				struct btree *b)
+static int bch2_btree_set_root(struct btree_update *as,
+			       struct btree_trans *trans,
+			       struct btree_path *path,
+			       struct btree *b,
+			       bool nofail)
 {
 	struct bch_fs *c = as->c;
-	struct btree *old;
 
 	trace_and_count(c, btree_node_set_root, trans, b);
 
-	old = btree_node_root(c, b);
+	struct btree *old = btree_node_root(c, b);
 
 	/*
 	 * Ensure no one is using the old root while we switch to the
 	 * new root:
 	 */
-	bch2_btree_node_lock_write_nofail(trans, path, &old->c);
+	if (nofail) {
+		bch2_btree_node_lock_write_nofail(trans, path, &old->c);
+	} else {
+		int ret = bch2_btree_node_lock_write(trans, path, &old->c);
+		if (ret)
+			return ret;
+	}
 
 	bch2_btree_set_root_inmem(c, b);
 
@@ -1250,6 +1340,7 @@ static void bch2_btree_set_root(struct btree_update *as,
 	 * depend on the new root would have to update the new root.
 	 */
 	bch2_btree_node_unlock_write(trans, path, old);
+	return 0;
 }
 
 /* Interior node updates: */
@@ -1316,12 +1407,12 @@ static void bch2_insert_fixup_btree_ptr(struct btree_update *as,
 }
 
 static void
-__bch2_btree_insert_keys_interior(struct btree_update *as,
-				  struct btree_trans *trans,
-				  struct btree_path *path,
-				  struct btree *b,
-				  struct btree_node_iter node_iter,
-				  struct keylist *keys)
+bch2_btree_insert_keys_interior(struct btree_update *as,
+				struct btree_trans *trans,
+				struct btree_path *path,
+				struct btree *b,
+				struct btree_node_iter node_iter,
+				struct keylist *keys)
 {
 	struct bkey_i *insert = bch2_keylist_front(keys);
 	struct bkey_packed *k;
@@ -1380,9 +1471,16 @@ static void __btree_split_node(struct btree_update *as,
 		if (bkey_deleted(k))
 			continue;
 
+		uk = bkey_unpack_key(b, k);
+
+		if (b->c.level &&
+		    u64s < n1_u64s &&
+		    u64s + k->u64s >= n1_u64s &&
+		    bch2_key_deleted_in_journal(trans, b->c.btree_id, b->c.level, uk.p))
+			n1_u64s += k->u64s;
+
 		i = u64s >= n1_u64s;
 		u64s += k->u64s;
-		uk = bkey_unpack_key(b, k);
 		if (!i)
 			n1_pos = uk.p;
 		bch2_bkey_format_add_key(&format[i], &uk);
@@ -1441,8 +1539,7 @@ static void __btree_split_node(struct btree_update *as,
 
 		bch2_verify_btree_nr_keys(n[i]);
 
-		if (b->c.level)
-			btree_node_interior_verify(as->c, n[i]);
+		BUG_ON(bch2_btree_node_check_topology(trans, n[i]));
 	}
 }
 
@@ -1471,9 +1568,9 @@ static void btree_split_insert_keys(struct btree_update *as,
 
 		bch2_btree_node_iter_init(&node_iter, b, &bch2_keylist_front(keys)->k.p);
 
-		__bch2_btree_insert_keys_interior(as, trans, path, b, node_iter, keys);
+		bch2_btree_insert_keys_interior(as, trans, path, b, node_iter, keys);
 
-		btree_node_interior_verify(as->c, b);
+		BUG_ON(bch2_btree_node_check_topology(trans, b));
 	}
 }
 
@@ -1488,9 +1585,14 @@ static int btree_split(struct btree_update *as, struct btree_trans *trans,
 	u64 start_time = local_clock();
 	int ret = 0;
 
+	bch2_verify_btree_nr_keys(b);
 	BUG_ON(!parent && (b != btree_node_root(c, b)));
 	BUG_ON(parent && !btree_node_intent_locked(trans->paths + path, b->c.level + 1));
 
+	ret = bch2_btree_node_check_topology(trans, b);
+	if (ret)
+		return ret;
+
 	bch2_btree_interior_update_will_free_node(as, b);
 
 	if (b->nr.live_u64s > BTREE_SPLIT_THRESHOLD(c)) {
@@ -1581,15 +1683,16 @@ static int btree_split(struct btree_update *as, struct btree_trans *trans,
 	if (parent) {
 		/* Split a non root node */
 		ret = bch2_btree_insert_node(as, trans, path, parent, &as->parent_keys);
-		if (ret)
-			goto err;
 	} else if (n3) {
-		bch2_btree_set_root(as, trans, trans->paths + path, n3);
+		ret = bch2_btree_set_root(as, trans, trans->paths + path, n3, false);
 	} else {
 		/* Root filled up but didn't need to be split */
-		bch2_btree_set_root(as, trans, trans->paths + path, n1);
+		ret = bch2_btree_set_root(as, trans, trans->paths + path, n1, false);
 	}
 
+	if (ret)
+		goto err;
+
 	if (n3) {
 		bch2_btree_update_get_open_buckets(as, n3);
 		bch2_btree_node_write(c, n3, SIX_LOCK_intent, 0);
@@ -1646,27 +1749,6 @@ err:
 	goto out;
 }
 
-static void
-bch2_btree_insert_keys_interior(struct btree_update *as,
-				struct btree_trans *trans,
-				struct btree_path *path,
-				struct btree *b,
-				struct keylist *keys)
-{
-	struct btree_path *linked;
-	unsigned i;
-
-	__bch2_btree_insert_keys_interior(as, trans, path, b,
-					  path->l[b->c.level].iter, keys);
-
-	btree_update_updated_node(as, b);
-
-	trans_for_each_path_with_node(trans, b, linked, i)
-		bch2_btree_node_iter_peek(&linked->l[b->c.level].iter, b);
-
-	bch2_trans_verify_paths(trans);
-}
-
 /**
  * bch2_btree_insert_node - insert bkeys into a given btree node
  *
@@ -1687,7 +1769,8 @@ static int bch2_btree_insert_node(struct btree_update *as, struct btree_trans *t
 				  struct keylist *keys)
 {
 	struct bch_fs *c = as->c;
-	struct btree_path *path = trans->paths + path_idx;
+	struct btree_path *path = trans->paths + path_idx, *linked;
+	unsigned i;
 	int old_u64s = le16_to_cpu(btree_bset_last(b)->u64s);
 	int old_live_u64s = b->nr.live_u64s;
 	int live_u64s_added, u64s_added;
@@ -1710,9 +1793,19 @@ static int bch2_btree_insert_node(struct btree_update *as, struct btree_trans *t
 		goto split;
 	}
 
-	btree_node_interior_verify(c, b);
+	ret = bch2_btree_node_check_topology(trans, b);
+	if (ret) {
+		bch2_btree_node_unlock_write(trans, path, b);
+		return ret;
+	}
+
+	bch2_btree_insert_keys_interior(as, trans, path, b,
+					path->l[b->c.level].iter, keys);
+
+	trans_for_each_path_with_node(trans, b, linked, i)
+		bch2_btree_node_iter_peek(&linked->l[b->c.level].iter, b);
 
-	bch2_btree_insert_keys_interior(as, trans, path, b, keys);
+	bch2_trans_verify_paths(trans);
 
 	live_u64s_added = (int) b->nr.live_u64s - old_live_u64s;
 	u64s_added = (int) le16_to_cpu(btree_bset_last(b)->u64s) - old_u64s;
@@ -1726,16 +1819,17 @@ static int bch2_btree_insert_node(struct btree_update *as, struct btree_trans *t
 	    bch2_maybe_compact_whiteouts(c, b))
 		bch2_trans_node_reinit_iter(trans, b);
 
+	btree_update_updated_node(as, b);
 	bch2_btree_node_unlock_write(trans, path, b);
 
-	btree_node_interior_verify(c, b);
+	BUG_ON(bch2_btree_node_check_topology(trans, b));
 	return 0;
 split:
 	/*
 	 * We could attempt to avoid the transaction restart, by calling
 	 * bch2_btree_path_upgrade() and allocating more nodes:
 	 */
-	if (b->c.level >= as->update_level) {
+	if (b->c.level >= as->update_level_end) {
 		trace_and_count(c, trans_restart_split_race, trans, _THIS_IP_, b);
 		return btree_trans_restart(trans, BCH_ERR_transaction_restart_split_race);
 	}
@@ -1801,7 +1895,9 @@ static void __btree_increase_depth(struct btree_update *as, struct btree_trans *
 	bch2_keylist_add(&as->parent_keys, &b->key);
 	btree_split_insert_keys(as, trans, path_idx, n, &as->parent_keys);
 
-	bch2_btree_set_root(as, trans, path, n);
+	int ret = bch2_btree_set_root(as, trans, path, n, true);
+	BUG_ON(ret);
+
 	bch2_btree_update_get_open_buckets(as, n);
 	bch2_btree_node_write(c, n, SIX_LOCK_intent, 0);
 	bch2_trans_node_add(trans, path, n);
@@ -1818,9 +1914,12 @@ int bch2_btree_increase_depth(struct btree_trans *trans, btree_path_idx_t path,
 {
 	struct bch_fs *c = trans->c;
 	struct btree *b = bch2_btree_id_root(c, trans->paths[path].btree_id)->b;
+
+	if (btree_node_fake(b))
+		return bch2_btree_split_leaf(trans, path, flags);
+
 	struct btree_update *as =
-		bch2_btree_update_start(trans, trans->paths + path,
-					b->c.level, true, flags);
+		bch2_btree_update_start(trans, trans->paths + path, b->c.level, true, flags);
 	if (IS_ERR(as))
 		return PTR_ERR(as);
 
@@ -1851,6 +1950,18 @@ int __bch2_foreground_maybe_merge(struct btree_trans *trans,
 	BUG_ON(!trans->paths[path].should_be_locked);
 	BUG_ON(!btree_node_locked(&trans->paths[path], level));
 
+	/*
+	 * Work around a deadlock caused by the btree write buffer not doing
+	 * merges and leaving tons of merges for us to do - we really don't need
+	 * to be doing merges at all from the interior update path, and if the
+	 * interior update path is generating too many new interior updates we
+	 * deadlock:
+	 */
+	if ((flags & BCH_WATERMARK_MASK) == BCH_WATERMARK_interior_updates)
+		return 0;
+
+	flags &= ~BCH_WATERMARK_MASK;
+
 	b = trans->paths[path].l[level].b;
 
 	if ((sib == btree_prev_sib && bpos_eq(b->data->min_key, POS_MIN)) ||
@@ -1996,6 +2107,10 @@ err:
 		bch2_path_put(trans, new_path, true);
 	bch2_path_put(trans, sib_path, true);
 	bch2_trans_verify_locks(trans);
+	if (ret == -BCH_ERR_journal_reclaim_would_deadlock)
+		ret = 0;
+	if (!ret)
+		ret = bch2_trans_relock(trans);
 	return ret;
 err_free_update:
 	bch2_btree_node_free_never_used(as, trans, n);
@@ -2041,12 +2156,13 @@ int bch2_btree_node_rewrite(struct btree_trans *trans,
 	if (parent) {
 		bch2_keylist_add(&as->parent_keys, &n->key);
 		ret = bch2_btree_insert_node(as, trans, iter->path, parent, &as->parent_keys);
-		if (ret)
-			goto err;
 	} else {
-		bch2_btree_set_root(as, trans, btree_iter_path(trans, iter), n);
+		ret = bch2_btree_set_root(as, trans, btree_iter_path(trans, iter), n, false);
 	}
 
+	if (ret)
+		goto err;
+
 	bch2_btree_update_get_open_buckets(as, n);
 	bch2_btree_node_write(c, n, SIX_LOCK_intent, 0);
 
@@ -2391,7 +2507,7 @@ void bch2_btree_set_root_for_read(struct bch_fs *c, struct btree *b)
 	bch2_btree_set_root_inmem(c, b);
 }
 
-static int __bch2_btree_root_alloc(struct btree_trans *trans, enum btree_id id)
+static int __bch2_btree_root_alloc_fake(struct btree_trans *trans, enum btree_id id, unsigned level)
 {
 	struct bch_fs *c = trans->c;
 	struct closure cl;
@@ -2410,7 +2526,7 @@ static int __bch2_btree_root_alloc(struct btree_trans *trans, enum btree_id id)
 
 	set_btree_node_fake(b);
 	set_btree_node_need_rewrite(b);
-	b->c.level	= 0;
+	b->c.level	= level;
 	b->c.btree_id	= id;
 
 	bkey_btree_ptr_init(&b->key);
@@ -2437,9 +2553,23 @@ static int __bch2_btree_root_alloc(struct btree_trans *trans, enum btree_id id)
 	return 0;
 }
 
-void bch2_btree_root_alloc(struct bch_fs *c, enum btree_id id)
+void bch2_btree_root_alloc_fake(struct bch_fs *c, enum btree_id id, unsigned level)
+{
+	bch2_trans_run(c, __bch2_btree_root_alloc_fake(trans, id, level));
+}
+
+static void bch2_btree_update_to_text(struct printbuf *out, struct btree_update *as)
 {
-	bch2_trans_run(c, __bch2_btree_root_alloc(trans, id));
+	prt_printf(out, "%ps: btree=%s l=%u-%u watermark=%s mode=%s nodes_written=%u cl.remaining=%u journal_seq=%llu\n",
+		   (void *) as->ip_started,
+		   bch2_btree_id_str(as->btree_id),
+		   as->update_level_start,
+		   as->update_level_end,
+		   bch2_watermarks[as->watermark],
+		   bch2_btree_update_modes[as->mode],
+		   as->nodes_written,
+		   closure_nr_remaining(&as->cl),
+		   as->journal.seq);
 }
 
 void bch2_btree_updates_to_text(struct printbuf *out, struct bch_fs *c)
@@ -2448,12 +2578,7 @@ void bch2_btree_updates_to_text(struct printbuf *out, struct bch_fs *c)
 
 	mutex_lock(&c->btree_interior_update_lock);
 	list_for_each_entry(as, &c->btree_interior_update_list, list)
-		prt_printf(out, "%ps: mode=%u nodes_written=%u cl.remaining=%u journal_seq=%llu\n",
-			   (void *) as->ip_started,
-			   as->mode,
-			   as->nodes_written,
-			   closure_nr_remaining(&as->cl),
-			   as->journal.seq);
+		bch2_btree_update_to_text(out, as);
 	mutex_unlock(&c->btree_interior_update_lock);
 }
 
diff --git a/fs/bcachefs/btree_update_interior.h b/fs/bcachefs/btree_update_interior.h
index f651dd48aaa0..c1a479ebaad1 100644
--- a/fs/bcachefs/btree_update_interior.h
+++ b/fs/bcachefs/btree_update_interior.h
@@ -10,6 +10,20 @@
 
 #define BTREE_UPDATE_JOURNAL_RES	(BTREE_UPDATE_NODES_MAX * (BKEY_BTREE_PTR_U64s_MAX + 1))
 
+int bch2_btree_node_check_topology(struct btree_trans *, struct btree *);
+
+#define BTREE_UPDATE_MODES()	\
+	x(none)			\
+	x(node)			\
+	x(root)			\
+	x(update)
+
+enum btree_update_mode {
+#define x(n)	BTREE_UPDATE_##n,
+	BTREE_UPDATE_MODES()
+#undef x
+};
+
 /*
  * Tracks an in progress split/rewrite of a btree node and the update to the
  * parent node:
@@ -37,24 +51,19 @@ struct btree_update {
 	struct list_head		list;
 	struct list_head		unwritten_list;
 
-	/* What kind of update are we doing? */
-	enum {
-		BTREE_INTERIOR_NO_UPDATE,
-		BTREE_INTERIOR_UPDATING_NODE,
-		BTREE_INTERIOR_UPDATING_ROOT,
-		BTREE_INTERIOR_UPDATING_AS,
-	} mode;
-
+	enum btree_update_mode		mode;
+	enum bch_watermark		watermark;
 	unsigned			nodes_written:1;
 	unsigned			took_gc_lock:1;
 
 	enum btree_id			btree_id;
-	unsigned			update_level;
+	unsigned			update_level_start;
+	unsigned			update_level_end;
 
 	struct disk_reservation		disk_res;
 
 	/*
-	 * BTREE_INTERIOR_UPDATING_NODE:
+	 * BTREE_UPDATE_node:
 	 * The update that made the new nodes visible was a regular update to an
 	 * existing interior node - @b. We can't write out the update to @b
 	 * until the new nodes we created are finished writing, so we block @b
@@ -163,7 +172,7 @@ int bch2_btree_node_update_key_get_iter(struct btree_trans *, struct btree *,
 					struct bkey_i *, unsigned, bool);
 
 void bch2_btree_set_root_for_read(struct bch_fs *, struct btree *);
-void bch2_btree_root_alloc(struct bch_fs *, enum btree_id);
+void bch2_btree_root_alloc_fake(struct bch_fs *, enum btree_id, unsigned);
 
 static inline unsigned btree_update_reserve_required(struct bch_fs *c,
 						     struct btree *b)
diff --git a/fs/bcachefs/btree_write_buffer.c b/fs/bcachefs/btree_write_buffer.c
index 5cbad8445782..36a6f42aba5e 100644
--- a/fs/bcachefs/btree_write_buffer.c
+++ b/fs/bcachefs/btree_write_buffer.c
@@ -11,6 +11,7 @@
 #include "journal_reclaim.h"
 
 #include <linux/prefetch.h>
+#include <linux/sort.h>
 
 static int bch2_btree_write_buffer_journal_flush(struct journal *,
 				struct journal_entry_pin *, u64);
@@ -46,6 +47,14 @@ static inline bool wb_key_ref_cmp(const struct wb_key_ref *l, const struct wb_ke
 #endif
 }
 
+static int wb_key_seq_cmp(const void *_l, const void *_r)
+{
+	const struct btree_write_buffered_key *l = _l;
+	const struct btree_write_buffered_key *r = _r;
+
+	return cmp_int(l->journal_seq, r->journal_seq);
+}
+
 /* Compare excluding idx, the low 24 bits: */
 static inline bool wb_key_eq(const void *_l, const void *_r)
 {
@@ -307,6 +316,16 @@ static int bch2_btree_write_buffer_flush_locked(struct btree_trans *trans)
 			    bpos_gt(k->k.k.p, path->l[0].b->key.k.p)) {
 				bch2_btree_node_unlock_write(trans, path, path->l[0].b);
 				write_locked = false;
+
+				ret = lockrestart_do(trans,
+					bch2_btree_iter_traverse(&iter) ?:
+					bch2_foreground_maybe_merge(trans, iter.path, 0,
+							BCH_WATERMARK_reclaim|
+							BCH_TRANS_COMMIT_journal_reclaim|
+							BCH_TRANS_COMMIT_no_check_rw|
+							BCH_TRANS_COMMIT_no_enospc));
+				if (ret)
+					goto err;
 			}
 		}
 
@@ -357,6 +376,11 @@ static int bch2_btree_write_buffer_flush_locked(struct btree_trans *trans)
 		 */
 		trace_and_count(c, write_buffer_flush_slowpath, trans, slowpath, wb->flushing.keys.nr);
 
+		sort(wb->flushing.keys.data,
+		     wb->flushing.keys.nr,
+		     sizeof(wb->flushing.keys.data[0]),
+		     wb_key_seq_cmp, NULL);
+
 		darray_for_each(wb->flushing.keys, i) {
 			if (!i->journal_seq)
 				continue;
@@ -368,10 +392,10 @@ static int bch2_btree_write_buffer_flush_locked(struct btree_trans *trans)
 
 			ret = commit_do(trans, NULL, NULL,
 					BCH_WATERMARK_reclaim|
+					BCH_TRANS_COMMIT_journal_reclaim|
 					BCH_TRANS_COMMIT_no_check_rw|
 					BCH_TRANS_COMMIT_no_enospc|
-					BCH_TRANS_COMMIT_no_journal_res|
-					BCH_TRANS_COMMIT_journal_reclaim,
+					BCH_TRANS_COMMIT_no_journal_res ,
 					btree_write_buffered_insert(trans, i));
 			if (ret)
 				goto err;
diff --git a/fs/bcachefs/buckets.c b/fs/bcachefs/buckets.c
index 96edf2c34d43..941401a210f5 100644
--- a/fs/bcachefs/buckets.c
+++ b/fs/bcachefs/buckets.c
@@ -525,6 +525,7 @@ int bch2_mark_metadata_bucket(struct bch_fs *c, struct bch_dev *ca,
 			"different types of data in same bucket: %s, %s",
 			bch2_data_type_str(g->data_type),
 			bch2_data_type_str(data_type))) {
+		BUG();
 		ret = -EIO;
 		goto err;
 	}
@@ -628,6 +629,7 @@ int bch2_check_bucket_ref(struct btree_trans *trans,
 			bch2_data_type_str(ptr_data_type),
 			(printbuf_reset(&buf),
 			 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
+		BUG();
 		ret = -EIO;
 		goto err;
 	}
@@ -815,14 +817,14 @@ static int __mark_pointer(struct btree_trans *trans,
 static int bch2_trigger_pointer(struct btree_trans *trans,
 			enum btree_id btree_id, unsigned level,
 			struct bkey_s_c k, struct extent_ptr_decoded p,
-			s64 *sectors,
-			unsigned flags)
+			const union bch_extent_entry *entry,
+			s64 *sectors, unsigned flags)
 {
 	bool insert = !(flags & BTREE_TRIGGER_OVERWRITE);
 	struct bpos bucket;
 	struct bch_backpointer bp;
 
-	bch2_extent_ptr_to_bp(trans->c, btree_id, level, k, p, &bucket, &bp);
+	bch2_extent_ptr_to_bp(trans->c, btree_id, level, k, p, entry, &bucket, &bp);
 	*sectors = insert ? bp.bucket_len : -((s64) bp.bucket_len);
 
 	if (flags & BTREE_TRIGGER_TRANSACTIONAL) {
@@ -851,7 +853,7 @@ static int bch2_trigger_pointer(struct btree_trans *trans,
 	if (flags & BTREE_TRIGGER_GC) {
 		struct bch_fs *c = trans->c;
 		struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
-		enum bch_data_type data_type = bkey_ptr_data_type(btree_id, level, k, p);
+		enum bch_data_type data_type = bch2_bkey_ptr_data_type(k, p, entry);
 
 		percpu_down_read(&c->mark_lock);
 		struct bucket *g = PTR_GC_BUCKET(ca, &p.ptr);
@@ -979,7 +981,7 @@ static int __trigger_extent(struct btree_trans *trans,
 
 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 		s64 disk_sectors;
-		ret = bch2_trigger_pointer(trans, btree_id, level, k, p, &disk_sectors, flags);
+		ret = bch2_trigger_pointer(trans, btree_id, level, k, p, entry, &disk_sectors, flags);
 		if (ret < 0)
 			return ret;
 
diff --git a/fs/bcachefs/buckets.h b/fs/bcachefs/buckets.h
index 6387e039f789..f9af5adabe83 100644
--- a/fs/bcachefs/buckets.h
+++ b/fs/bcachefs/buckets.h
@@ -226,6 +226,7 @@ static inline u64 bch2_dev_buckets_reserved(struct bch_dev *ca, enum bch_waterma
 		fallthrough;
 	case BCH_WATERMARK_btree_copygc:
 	case BCH_WATERMARK_reclaim:
+	case BCH_WATERMARK_interior_updates:
 		break;
 	}
 
@@ -394,14 +395,6 @@ static inline const char *bch2_data_type_str(enum bch_data_type type)
 		: "(invalid data type)";
 }
 
-static inline void bch2_prt_data_type(struct printbuf *out, enum bch_data_type type)
-{
-	if (type < BCH_DATA_NR)
-		prt_str(out, __bch2_data_types[type]);
-	else
-		prt_printf(out, "(invalid data type %u)", type);
-}
-
 /* disk reservations: */
 
 static inline void bch2_disk_reservation_put(struct bch_fs *c,
diff --git a/fs/bcachefs/chardev.c b/fs/bcachefs/chardev.c
index 38defa19d52d..72781aad6ba7 100644
--- a/fs/bcachefs/chardev.c
+++ b/fs/bcachefs/chardev.c
@@ -7,7 +7,7 @@
 #include "chardev.h"
 #include "journal.h"
 #include "move.h"
-#include "recovery.h"
+#include "recovery_passes.h"
 #include "replicas.h"
 #include "super.h"
 #include "super-io.h"
@@ -134,42 +134,38 @@ static long bch2_ioctl_incremental(struct bch_ioctl_incremental __user *user_arg
 struct fsck_thread {
 	struct thread_with_stdio thr;
 	struct bch_fs		*c;
-	char			**devs;
-	size_t			nr_devs;
 	struct bch_opts		opts;
 };
 
 static void bch2_fsck_thread_exit(struct thread_with_stdio *_thr)
 {
 	struct fsck_thread *thr = container_of(_thr, struct fsck_thread, thr);
-	if (thr->devs)
-		for (size_t i = 0; i < thr->nr_devs; i++)
-			kfree(thr->devs[i]);
-	kfree(thr->devs);
 	kfree(thr);
 }
 
 static int bch2_fsck_offline_thread_fn(struct thread_with_stdio *stdio)
 {
 	struct fsck_thread *thr = container_of(stdio, struct fsck_thread, thr);
-	struct bch_fs *c = bch2_fs_open(thr->devs, thr->nr_devs, thr->opts);
-
-	if (IS_ERR(c))
-		return PTR_ERR(c);
+	struct bch_fs *c = thr->c;
 
-	int ret = 0;
-	if (test_bit(BCH_FS_errors_fixed, &c->flags))
-		ret |= 1;
-	if (test_bit(BCH_FS_error, &c->flags))
-		ret |= 4;
+	int ret = PTR_ERR_OR_ZERO(c);
+	if (ret)
+		return ret;
 
-	bch2_fs_stop(c);
+	ret = bch2_fs_start(thr->c);
+	if (ret)
+		goto err;
 
-	if (ret & 1)
+	if (test_bit(BCH_FS_errors_fixed, &c->flags)) {
 		bch2_stdio_redirect_printf(&stdio->stdio, false, "%s: errors fixed\n", c->name);
-	if (ret & 4)
+		ret |= 1;
+	}
+	if (test_bit(BCH_FS_error, &c->flags)) {
 		bch2_stdio_redirect_printf(&stdio->stdio, false, "%s: still has errors\n", c->name);
-
+		ret |= 4;
+	}
+err:
+	bch2_fs_stop(c);
 	return ret;
 }
 
@@ -182,7 +178,7 @@ static long bch2_ioctl_fsck_offline(struct bch_ioctl_fsck_offline __user *user_a
 {
 	struct bch_ioctl_fsck_offline arg;
 	struct fsck_thread *thr = NULL;
-	u64 *devs = NULL;
+	darray_str(devs) = {};
 	long ret = 0;
 
 	if (copy_from_user(&arg, user_arg, sizeof(arg)))
@@ -194,29 +190,32 @@ static long bch2_ioctl_fsck_offline(struct bch_ioctl_fsck_offline __user *user_a
 	if (!capable(CAP_SYS_ADMIN))
 		return -EPERM;
 
-	if (!(devs = kcalloc(arg.nr_devs, sizeof(*devs), GFP_KERNEL)) ||
-	    !(thr = kzalloc(sizeof(*thr), GFP_KERNEL)) ||
-	    !(thr->devs = kcalloc(arg.nr_devs, sizeof(*thr->devs), GFP_KERNEL))) {
-		ret = -ENOMEM;
-		goto err;
-	}
+	for (size_t i = 0; i < arg.nr_devs; i++) {
+		u64 dev_u64;
+		ret = copy_from_user_errcode(&dev_u64, &user_arg->devs[i], sizeof(u64));
+		if (ret)
+			goto err;
 
-	thr->opts = bch2_opts_empty();
-	thr->nr_devs = arg.nr_devs;
+		char *dev_str = strndup_user((char __user *)(unsigned long) dev_u64, PATH_MAX);
+		ret = PTR_ERR_OR_ZERO(dev_str);
+		if (ret)
+			goto err;
 
-	if (copy_from_user(devs, &user_arg->devs[0],
-			   array_size(sizeof(user_arg->devs[0]), arg.nr_devs))) {
-		ret = -EINVAL;
-		goto err;
+		ret = darray_push(&devs, dev_str);
+		if (ret) {
+			kfree(dev_str);
+			goto err;
+		}
 	}
 
-	for (size_t i = 0; i < arg.nr_devs; i++) {
-		thr->devs[i] = strndup_user((char __user *)(unsigned long) devs[i], PATH_MAX);
-		ret = PTR_ERR_OR_ZERO(thr->devs[i]);
-		if (ret)
-			goto err;
+	thr = kzalloc(sizeof(*thr), GFP_KERNEL);
+	if (!thr) {
+		ret = -ENOMEM;
+		goto err;
 	}
 
+	thr->opts = bch2_opts_empty();
+
 	if (arg.opts) {
 		char *optstr = strndup_user((char __user *)(unsigned long) arg.opts, 1 << 16);
 
@@ -230,15 +229,26 @@ static long bch2_ioctl_fsck_offline(struct bch_ioctl_fsck_offline __user *user_a
 
 	opt_set(thr->opts, stdio, (u64)(unsigned long)&thr->thr.stdio);
 
+	/* We need request_key() to be called before we punt to kthread: */
+	opt_set(thr->opts, nostart, true);
+
+	thr->c = bch2_fs_open(devs.data, arg.nr_devs, thr->opts);
+
+	if (!IS_ERR(thr->c) &&
+	    thr->c->opts.errors == BCH_ON_ERROR_panic)
+		thr->c->opts.errors = BCH_ON_ERROR_ro;
+
 	ret = bch2_run_thread_with_stdio(&thr->thr, &bch2_offline_fsck_ops);
-err:
-	if (ret < 0) {
-		if (thr)
-			bch2_fsck_thread_exit(&thr->thr);
-		pr_err("ret %s", bch2_err_str(ret));
-	}
-	kfree(devs);
+out:
+	darray_for_each(devs, i)
+		kfree(*i);
+	darray_exit(&devs);
 	return ret;
+err:
+	if (thr)
+		bch2_fsck_thread_exit(&thr->thr);
+	pr_err("ret %s", bch2_err_str(ret));
+	goto out;
 }
 
 static long bch2_global_ioctl(unsigned cmd, void __user *arg)
diff --git a/fs/bcachefs/checksum.c b/fs/bcachefs/checksum.c
index 4701457f6381..7ed779b411f6 100644
--- a/fs/bcachefs/checksum.c
+++ b/fs/bcachefs/checksum.c
@@ -429,15 +429,20 @@ int bch2_rechecksum_bio(struct bch_fs *c, struct bio *bio,
 				extent_nonce(version, crc_old), bio);
 
 	if (bch2_crc_cmp(merged, crc_old.csum) && !c->opts.no_data_io) {
-		bch_err(c, "checksum error in %s() (memory corruption or bug?)\n"
-			"expected %0llx:%0llx got %0llx:%0llx (old type %s new type %s)",
-			__func__,
-			crc_old.csum.hi,
-			crc_old.csum.lo,
-			merged.hi,
-			merged.lo,
-			bch2_csum_types[crc_old.csum_type],
-			bch2_csum_types[new_csum_type]);
+		struct printbuf buf = PRINTBUF;
+		prt_printf(&buf, "checksum error in %s() (memory corruption or bug?)\n"
+			   "expected %0llx:%0llx got %0llx:%0llx (old type ",
+			   __func__,
+			   crc_old.csum.hi,
+			   crc_old.csum.lo,
+			   merged.hi,
+			   merged.lo);
+		bch2_prt_csum_type(&buf, crc_old.csum_type);
+		prt_str(&buf, " new type ");
+		bch2_prt_csum_type(&buf, new_csum_type);
+		prt_str(&buf, ")");
+		bch_err(c, "%s", buf.buf);
+		printbuf_exit(&buf);
 		return -EIO;
 	}
 
diff --git a/fs/bcachefs/checksum.h b/fs/bcachefs/checksum.h
index 1b8c2c1016dc..e40499fde9a4 100644
--- a/fs/bcachefs/checksum.h
+++ b/fs/bcachefs/checksum.h
@@ -61,11 +61,12 @@ static inline void bch2_csum_err_msg(struct printbuf *out,
 				     struct bch_csum expected,
 				     struct bch_csum got)
 {
-	prt_printf(out, "checksum error: got ");
+	prt_str(out, "checksum error, type ");
+	bch2_prt_csum_type(out, type);
+	prt_str(out, ": got ");
 	bch2_csum_to_text(out, type, got);
 	prt_str(out, " should be ");
 	bch2_csum_to_text(out, type, expected);
-	prt_printf(out, " type %s", bch2_csum_types[type]);
 }
 
 int bch2_chacha_encrypt_key(struct bch_key *, struct nonce, void *, size_t);
diff --git a/fs/bcachefs/compress.h b/fs/bcachefs/compress.h
index 58c2eb45570f..607fd5e232c9 100644
--- a/fs/bcachefs/compress.h
+++ b/fs/bcachefs/compress.h
@@ -47,14 +47,6 @@ static inline enum bch_compression_type bch2_compression_opt_to_type(unsigned v)
 	return __bch2_compression_opt_to_type[bch2_compression_decode(v).type];
 }
 
-static inline void bch2_prt_compression_type(struct printbuf *out, enum bch_compression_type type)
-{
-	if (type < BCH_COMPRESSION_TYPE_NR)
-		prt_str(out, __bch2_compression_types[type]);
-	else
-		prt_printf(out, "(invalid compression type %u)", type);
-}
-
 int bch2_bio_uncompress_inplace(struct bch_fs *, struct bio *,
 				struct bch_extent_crc_unpacked *);
 int bch2_bio_uncompress(struct bch_fs *, struct bio *, struct bio *,
diff --git a/fs/bcachefs/data_update.c b/fs/bcachefs/data_update.c
index 4150feca42a2..0022b51ce3c0 100644
--- a/fs/bcachefs/data_update.c
+++ b/fs/bcachefs/data_update.c
@@ -14,6 +14,7 @@
 #include "move.h"
 #include "nocow_locking.h"
 #include "rebalance.h"
+#include "snapshot.h"
 #include "subvolume.h"
 #include "trace.h"
 
@@ -509,6 +510,14 @@ int bch2_data_update_init(struct btree_trans *trans,
 	unsigned ptrs_locked = 0;
 	int ret = 0;
 
+	/*
+	 * fs is corrupt  we have a key for a snapshot node that doesn't exist,
+	 * and we have to check for this because we go rw before repairing the
+	 * snapshots table - just skip it, we can move it later.
+	 */
+	if (unlikely(k.k->p.snapshot && !bch2_snapshot_equiv(c, k.k->p.snapshot)))
+		return -BCH_ERR_data_update_done;
+
 	bch2_bkey_buf_init(&m->k);
 	bch2_bkey_buf_reassemble(&m->k, c, k);
 	m->btree_id	= btree_id;
@@ -571,8 +580,7 @@ int bch2_data_update_init(struct btree_trans *trans,
 				move_ctxt_wait_event(ctxt,
 						(locked = bch2_bucket_nocow_trylock(&c->nocow_locks,
 									  PTR_BUCKET_POS(c, &p.ptr), 0)) ||
-						(!atomic_read(&ctxt->read_sectors) &&
-						 !atomic_read(&ctxt->write_sectors)));
+						list_empty(&ctxt->ios));
 
 				if (!locked)
 					bch2_bucket_nocow_lock(&c->nocow_locks,
@@ -590,6 +598,8 @@ int bch2_data_update_init(struct btree_trans *trans,
 		i++;
 	}
 
+	unsigned durability_required = max(0, (int) (io_opts.data_replicas - durability_have));
+
 	/*
 	 * If current extent durability is less than io_opts.data_replicas,
 	 * we're not trying to rereplicate the extent up to data_replicas here -
@@ -599,7 +609,7 @@ int bch2_data_update_init(struct btree_trans *trans,
 	 * rereplicate, currently, so that users don't get an unexpected -ENOSPC
 	 */
 	if (!(m->data_opts.write_flags & BCH_WRITE_CACHED) &&
-	    durability_have >= io_opts.data_replicas) {
+	    !durability_required) {
 		m->data_opts.kill_ptrs |= m->data_opts.rewrite_ptrs;
 		m->data_opts.rewrite_ptrs = 0;
 		/* if iter == NULL, it's just a promote */
@@ -608,11 +618,18 @@ int bch2_data_update_init(struct btree_trans *trans,
 		goto done;
 	}
 
-	m->op.nr_replicas = min(durability_removing, io_opts.data_replicas - durability_have) +
+	m->op.nr_replicas = min(durability_removing, durability_required) +
 		m->data_opts.extra_replicas;
-	m->op.nr_replicas_required = m->op.nr_replicas;
 
-	BUG_ON(!m->op.nr_replicas);
+	/*
+	 * If device(s) were set to durability=0 after data was written to them
+	 * we can end up with a duribilty=0 extent, and the normal algorithm
+	 * that tries not to increase durability doesn't work:
+	 */
+	if (!(durability_have + durability_removing))
+		m->op.nr_replicas = max((unsigned) m->op.nr_replicas, 1);
+
+	m->op.nr_replicas_required = m->op.nr_replicas;
 
 	if (reserve_sectors) {
 		ret = bch2_disk_reservation_add(c, &m->op.res, reserve_sectors,
diff --git a/fs/bcachefs/debug.c b/fs/bcachefs/debug.c
index 208ce6f0fc43..cd99b7399414 100644
--- a/fs/bcachefs/debug.c
+++ b/fs/bcachefs/debug.c
@@ -13,6 +13,7 @@
 #include "btree_iter.h"
 #include "btree_locking.h"
 #include "btree_update.h"
+#include "btree_update_interior.h"
 #include "buckets.h"
 #include "debug.h"
 #include "error.h"
@@ -668,7 +669,7 @@ static ssize_t bch2_journal_pins_read(struct file *file, char __user *buf,
 	i->size	= size;
 	i->ret	= 0;
 
-	do {
+	while (1) {
 		err = flush_buf(i);
 		if (err)
 			return err;
@@ -676,9 +677,12 @@ static ssize_t bch2_journal_pins_read(struct file *file, char __user *buf,
 		if (!i->size)
 			break;
 
+		if (done)
+			break;
+
 		done = bch2_journal_seq_pins_to_text(&i->buf, &c->journal, &i->iter);
 		i->iter++;
-	} while (!done);
+	}
 
 	if (i->buf.allocation_failure)
 		return -ENOMEM;
@@ -693,13 +697,45 @@ static const struct file_operations journal_pins_ops = {
 	.read		= bch2_journal_pins_read,
 };
 
+static ssize_t bch2_btree_updates_read(struct file *file, char __user *buf,
+				       size_t size, loff_t *ppos)
+{
+	struct dump_iter *i = file->private_data;
+	struct bch_fs *c = i->c;
+	int err;
+
+	i->ubuf = buf;
+	i->size	= size;
+	i->ret	= 0;
+
+	if (!i->iter) {
+		bch2_btree_updates_to_text(&i->buf, c);
+		i->iter++;
+	}
+
+	err = flush_buf(i);
+	if (err)
+		return err;
+
+	if (i->buf.allocation_failure)
+		return -ENOMEM;
+
+	return i->ret;
+}
+
+static const struct file_operations btree_updates_ops = {
+	.owner		= THIS_MODULE,
+	.open		= bch2_dump_open,
+	.release	= bch2_dump_release,
+	.read		= bch2_btree_updates_read,
+};
+
 static int btree_transaction_stats_open(struct inode *inode, struct file *file)
 {
 	struct bch_fs *c = inode->i_private;
 	struct dump_iter *i;
 
 	i = kzalloc(sizeof(struct dump_iter), GFP_KERNEL);
-
 	if (!i)
 		return -ENOMEM;
 
@@ -866,6 +902,20 @@ void bch2_fs_debug_exit(struct bch_fs *c)
 		debugfs_remove_recursive(c->fs_debug_dir);
 }
 
+static void bch2_fs_debug_btree_init(struct bch_fs *c, struct btree_debug *bd)
+{
+	struct dentry *d;
+
+	d = debugfs_create_dir(bch2_btree_id_str(bd->id), c->btree_debug_dir);
+
+	debugfs_create_file("keys", 0400, d, bd, &btree_debug_ops);
+
+	debugfs_create_file("formats", 0400, d, bd, &btree_format_debug_ops);
+
+	debugfs_create_file("bfloat-failed", 0400, d, bd,
+			    &bfloat_failed_debug_ops);
+}
+
 void bch2_fs_debug_init(struct bch_fs *c)
 {
 	struct btree_debug *bd;
@@ -888,6 +938,9 @@ void bch2_fs_debug_init(struct bch_fs *c)
 	debugfs_create_file("journal_pins", 0400, c->fs_debug_dir,
 			    c->btree_debug, &journal_pins_ops);
 
+	debugfs_create_file("btree_updates", 0400, c->fs_debug_dir,
+			    c->btree_debug, &btree_updates_ops);
+
 	debugfs_create_file("btree_transaction_stats", 0400, c->fs_debug_dir,
 			    c, &btree_transaction_stats_op);
 
@@ -902,21 +955,7 @@ void bch2_fs_debug_init(struct bch_fs *c)
 	     bd < c->btree_debug + ARRAY_SIZE(c->btree_debug);
 	     bd++) {
 		bd->id = bd - c->btree_debug;
-		debugfs_create_file(bch2_btree_id_str(bd->id),
-				    0400, c->btree_debug_dir, bd,
-				    &btree_debug_ops);
-
-		snprintf(name, sizeof(name), "%s-formats",
-			 bch2_btree_id_str(bd->id));
-
-		debugfs_create_file(name, 0400, c->btree_debug_dir, bd,
-				    &btree_format_debug_ops);
-
-		snprintf(name, sizeof(name), "%s-bfloat-failed",
-			 bch2_btree_id_str(bd->id));
-
-		debugfs_create_file(name, 0400, c->btree_debug_dir, bd,
-				    &bfloat_failed_debug_ops);
+		bch2_fs_debug_btree_init(c, bd);
 	}
 }
 
diff --git a/fs/bcachefs/ec.c b/fs/bcachefs/ec.c
index 082075244e16..556a217108d3 100644
--- a/fs/bcachefs/ec.c
+++ b/fs/bcachefs/ec.c
@@ -131,29 +131,33 @@ fsck_err:
 void bch2_stripe_to_text(struct printbuf *out, struct bch_fs *c,
 			 struct bkey_s_c k)
 {
-	const struct bch_stripe *s = bkey_s_c_to_stripe(k).v;
-	unsigned i, nr_data = s->nr_blocks - s->nr_redundant;
+	const struct bch_stripe *sp = bkey_s_c_to_stripe(k).v;
+	struct bch_stripe s = {};
+
+	memcpy(&s, sp, min(sizeof(s), bkey_val_bytes(k.k)));
+
+	unsigned nr_data = s.nr_blocks - s.nr_redundant;
+
+	prt_printf(out, "algo %u sectors %u blocks %u:%u csum ",
+		   s.algorithm,
+		   le16_to_cpu(s.sectors),
+		   nr_data,
+		   s.nr_redundant);
+	bch2_prt_csum_type(out, s.csum_type);
+	prt_printf(out, " gran %u", 1U << s.csum_granularity_bits);
+
+	for (unsigned i = 0; i < s.nr_blocks; i++) {
+		const struct bch_extent_ptr *ptr = sp->ptrs + i;
+
+		if ((void *) ptr >= bkey_val_end(k))
+			break;
+
+		bch2_extent_ptr_to_text(out, c, ptr);
 
-	prt_printf(out, "algo %u sectors %u blocks %u:%u csum %u gran %u",
-	       s->algorithm,
-	       le16_to_cpu(s->sectors),
-	       nr_data,
-	       s->nr_redundant,
-	       s->csum_type,
-	       1U << s->csum_granularity_bits);
-
-	for (i = 0; i < s->nr_blocks; i++) {
-		const struct bch_extent_ptr *ptr = s->ptrs + i;
-		struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
-		u32 offset;
-		u64 b = sector_to_bucket_and_offset(ca, ptr->offset, &offset);
-
-		prt_printf(out, " %u:%llu:%u", ptr->dev, b, offset);
-		if (i < nr_data)
-			prt_printf(out, "#%u", stripe_blockcount_get(s, i));
-		prt_printf(out, " gen %u", ptr->gen);
-		if (ptr_stale(ca, ptr))
-			prt_printf(out, " stale");
+		if (s.csum_type < BCH_CSUM_NR &&
+		    i < nr_data &&
+		    stripe_blockcount_offset(&s, i) < bkey_val_bytes(k.k))
+			prt_printf(out,  "#%u", stripe_blockcount_get(sp, i));
 	}
 }
 
@@ -607,10 +611,8 @@ static void ec_validate_checksums(struct bch_fs *c, struct ec_stripe_buf *buf)
 				struct printbuf err = PRINTBUF;
 				struct bch_dev *ca = bch_dev_bkey_exists(c, v->ptrs[i].dev);
 
-				prt_printf(&err, "stripe checksum error: expected %0llx:%0llx got %0llx:%0llx (type %s)\n",
-					   want.hi, want.lo,
-					   got.hi, got.lo,
-					   bch2_csum_types[v->csum_type]);
+				prt_str(&err, "stripe ");
+				bch2_csum_err_msg(&err, v->csum_type, want, got);
 				prt_printf(&err, "  for %ps at %u of\n  ", (void *) _RET_IP_, i);
 				bch2_bkey_val_to_text(&err, c, bkey_i_to_s_c(&buf->key));
 				bch_err_ratelimited(ca, "%s", err.buf);
diff --git a/fs/bcachefs/ec.h b/fs/bcachefs/ec.h
index f4369b02e805..f042616888b0 100644
--- a/fs/bcachefs/ec.h
+++ b/fs/bcachefs/ec.h
@@ -32,6 +32,8 @@ static inline unsigned stripe_csums_per_device(const struct bch_stripe *s)
 static inline unsigned stripe_csum_offset(const struct bch_stripe *s,
 					  unsigned dev, unsigned csum_idx)
 {
+	EBUG_ON(s->csum_type >= BCH_CSUM_NR);
+
 	unsigned csum_bytes = bch_crc_bytes[s->csum_type];
 
 	return sizeof(struct bch_stripe) +
diff --git a/fs/bcachefs/errcode.h b/fs/bcachefs/errcode.h
index af25d8ec60f2..01a79fa3eacb 100644
--- a/fs/bcachefs/errcode.h
+++ b/fs/bcachefs/errcode.h
@@ -252,7 +252,8 @@
 	x(BCH_ERR_nopromote,		nopromote_in_flight)			\
 	x(BCH_ERR_nopromote,		nopromote_no_writes)			\
 	x(BCH_ERR_nopromote,		nopromote_enomem)			\
-	x(0,				need_inode_lock)
+	x(0,				need_inode_lock)			\
+	x(0,				invalid_snapshot_node)
 
 enum bch_errcode {
 	BCH_ERR_START		= 2048,
diff --git a/fs/bcachefs/error.c b/fs/bcachefs/error.c
index 043431206799..82a6656c941c 100644
--- a/fs/bcachefs/error.c
+++ b/fs/bcachefs/error.c
@@ -1,7 +1,8 @@
 // SPDX-License-Identifier: GPL-2.0
 #include "bcachefs.h"
 #include "error.h"
-#include "recovery.h"
+#include "journal.h"
+#include "recovery_passes.h"
 #include "super.h"
 #include "thread_with_file.h"
 
@@ -16,7 +17,8 @@ bool bch2_inconsistent_error(struct bch_fs *c)
 		return false;
 	case BCH_ON_ERROR_ro:
 		if (bch2_fs_emergency_read_only(c))
-			bch_err(c, "inconsistency detected - emergency read only");
+			bch_err(c, "inconsistency detected - emergency read only at journal seq %llu",
+				journal_cur_seq(&c->journal));
 		return true;
 	case BCH_ON_ERROR_panic:
 		panic(bch2_fmt(c, "panic after error"));
diff --git a/fs/bcachefs/error.h b/fs/bcachefs/error.h
index ae1d6674c512..36caedf72d89 100644
--- a/fs/bcachefs/error.h
+++ b/fs/bcachefs/error.h
@@ -32,6 +32,12 @@ bool bch2_inconsistent_error(struct bch_fs *);
 
 int bch2_topology_error(struct bch_fs *);
 
+#define bch2_fs_topology_error(c, ...)					\
+({									\
+	bch_err(c, "btree topology error: " __VA_ARGS__);		\
+	bch2_topology_error(c);						\
+})
+
 #define bch2_fs_inconsistent(c, ...)					\
 ({									\
 	bch_err(c, __VA_ARGS__);					\
diff --git a/fs/bcachefs/extents.c b/fs/bcachefs/extents.c
index 61395b113df9..1a331e539204 100644
--- a/fs/bcachefs/extents.c
+++ b/fs/bcachefs/extents.c
@@ -189,13 +189,18 @@ int bch2_btree_ptr_v2_invalid(struct bch_fs *c, struct bkey_s_c k,
 			      enum bkey_invalid_flags flags,
 			      struct printbuf *err)
 {
+	struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
 	int ret = 0;
 
-	bkey_fsck_err_on(bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX, c, err,
-			 btree_ptr_v2_val_too_big,
+	bkey_fsck_err_on(bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX,
+			 c, err, btree_ptr_v2_val_too_big,
 			 "value too big (%zu > %zu)",
 			 bkey_val_u64s(k.k), BKEY_BTREE_PTR_VAL_U64s_MAX);
 
+	bkey_fsck_err_on(bpos_ge(bp.v->min_key, bp.k->p),
+			 c, err, btree_ptr_v2_min_key_bad,
+			 "min_key > key");
+
 	ret = bch2_bkey_ptrs_invalid(c, k, flags, err);
 fsck_err:
 	return ret;
@@ -973,6 +978,33 @@ bool bch2_extent_normalize(struct bch_fs *c, struct bkey_s k)
 	return bkey_deleted(k.k);
 }
 
+void bch2_extent_ptr_to_text(struct printbuf *out, struct bch_fs *c, const struct bch_extent_ptr *ptr)
+{
+	struct bch_dev *ca = c && ptr->dev < c->sb.nr_devices && c->devs[ptr->dev]
+		? bch_dev_bkey_exists(c, ptr->dev)
+		: NULL;
+
+	if (!ca) {
+		prt_printf(out, "ptr: %u:%llu gen %u%s", ptr->dev,
+			   (u64) ptr->offset, ptr->gen,
+			   ptr->cached ? " cached" : "");
+	} else {
+		u32 offset;
+		u64 b = sector_to_bucket_and_offset(ca, ptr->offset, &offset);
+
+		prt_printf(out, "ptr: %u:%llu:%u gen %u",
+			   ptr->dev, b, offset, ptr->gen);
+		if (ptr->cached)
+			prt_str(out, " cached");
+		if (ptr->unwritten)
+			prt_str(out, " unwritten");
+		if (b >= ca->mi.first_bucket &&
+		    b <  ca->mi.nbuckets &&
+		    ptr_stale(ca, ptr))
+			prt_printf(out, " stale");
+	}
+}
+
 void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 			    struct bkey_s_c k)
 {
@@ -988,42 +1020,22 @@ void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 			prt_printf(out, " ");
 
 		switch (__extent_entry_type(entry)) {
-		case BCH_EXTENT_ENTRY_ptr: {
-			const struct bch_extent_ptr *ptr = entry_to_ptr(entry);
-			struct bch_dev *ca = c && ptr->dev < c->sb.nr_devices && c->devs[ptr->dev]
-				? bch_dev_bkey_exists(c, ptr->dev)
-				: NULL;
-
-			if (!ca) {
-				prt_printf(out, "ptr: %u:%llu gen %u%s", ptr->dev,
-				       (u64) ptr->offset, ptr->gen,
-				       ptr->cached ? " cached" : "");
-			} else {
-				u32 offset;
-				u64 b = sector_to_bucket_and_offset(ca, ptr->offset, &offset);
-
-				prt_printf(out, "ptr: %u:%llu:%u gen %u",
-					   ptr->dev, b, offset, ptr->gen);
-				if (ptr->cached)
-					prt_str(out, " cached");
-				if (ptr->unwritten)
-					prt_str(out, " unwritten");
-				if (ca && ptr_stale(ca, ptr))
-					prt_printf(out, " stale");
-			}
+		case BCH_EXTENT_ENTRY_ptr:
+			bch2_extent_ptr_to_text(out, c, entry_to_ptr(entry));
 			break;
-		}
+
 		case BCH_EXTENT_ENTRY_crc32:
 		case BCH_EXTENT_ENTRY_crc64:
 		case BCH_EXTENT_ENTRY_crc128: {
 			struct bch_extent_crc_unpacked crc =
 				bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
 
-			prt_printf(out, "crc: c_size %u size %u offset %u nonce %u csum %s compress ",
+			prt_printf(out, "crc: c_size %u size %u offset %u nonce %u csum ",
 			       crc.compressed_size,
 			       crc.uncompressed_size,
-			       crc.offset, crc.nonce,
-			       bch2_csum_types[crc.csum_type]);
+			       crc.offset, crc.nonce);
+			bch2_prt_csum_type(out, crc.csum_type);
+			prt_str(out, " compress ");
 			bch2_prt_compression_type(out, crc.compression_type);
 			break;
 		}
diff --git a/fs/bcachefs/extents.h b/fs/bcachefs/extents.h
index fd2669cdd76f..528e817eacbd 100644
--- a/fs/bcachefs/extents.h
+++ b/fs/bcachefs/extents.h
@@ -596,30 +596,6 @@ static inline struct bch_devs_list bch2_bkey_cached_devs(struct bkey_s_c k)
 	return ret;
 }
 
-static inline unsigned bch2_bkey_ptr_data_type(struct bkey_s_c k, const struct bch_extent_ptr *ptr)
-{
-	switch (k.k->type) {
-	case KEY_TYPE_btree_ptr:
-	case KEY_TYPE_btree_ptr_v2:
-		return BCH_DATA_btree;
-	case KEY_TYPE_extent:
-	case KEY_TYPE_reflink_v:
-		return BCH_DATA_user;
-	case KEY_TYPE_stripe: {
-		struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
-
-		BUG_ON(ptr < s.v->ptrs ||
-		       ptr >= s.v->ptrs + s.v->nr_blocks);
-
-		return ptr >= s.v->ptrs + s.v->nr_blocks - s.v->nr_redundant
-			? BCH_DATA_parity
-			: BCH_DATA_user;
-	}
-	default:
-		BUG();
-	}
-}
-
 unsigned bch2_bkey_nr_ptrs(struct bkey_s_c);
 unsigned bch2_bkey_nr_ptrs_allocated(struct bkey_s_c);
 unsigned bch2_bkey_nr_ptrs_fully_allocated(struct bkey_s_c);
@@ -700,6 +676,7 @@ bch2_extent_has_ptr(struct bkey_s_c, struct extent_ptr_decoded, struct bkey_s);
 void bch2_extent_ptr_set_cached(struct bkey_s, struct bch_extent_ptr *);
 
 bool bch2_extent_normalize(struct bch_fs *, struct bkey_s);
+void bch2_extent_ptr_to_text(struct printbuf *out, struct bch_fs *, const struct bch_extent_ptr *);
 void bch2_bkey_ptrs_to_text(struct printbuf *, struct bch_fs *,
 			    struct bkey_s_c);
 int bch2_bkey_ptrs_invalid(struct bch_fs *, struct bkey_s_c,
diff --git a/fs/bcachefs/eytzinger.c b/fs/bcachefs/eytzinger.c
new file mode 100644
index 000000000000..0f955c3c76a7
--- /dev/null
+++ b/fs/bcachefs/eytzinger.c
@@ -0,0 +1,234 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include "eytzinger.h"
+
+/**
+ * is_aligned - is this pointer & size okay for word-wide copying?
+ * @base: pointer to data
+ * @size: size of each element
+ * @align: required alignment (typically 4 or 8)
+ *
+ * Returns true if elements can be copied using word loads and stores.
+ * The size must be a multiple of the alignment, and the base address must
+ * be if we do not have CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS.
+ *
+ * For some reason, gcc doesn't know to optimize "if (a & mask || b & mask)"
+ * to "if ((a | b) & mask)", so we do that by hand.
+ */
+__attribute_const__ __always_inline
+static bool is_aligned(const void *base, size_t size, unsigned char align)
+{
+	unsigned char lsbits = (unsigned char)size;
+
+	(void)base;
+#ifndef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
+	lsbits |= (unsigned char)(uintptr_t)base;
+#endif
+	return (lsbits & (align - 1)) == 0;
+}
+
+/**
+ * swap_words_32 - swap two elements in 32-bit chunks
+ * @a: pointer to the first element to swap
+ * @b: pointer to the second element to swap
+ * @n: element size (must be a multiple of 4)
+ *
+ * Exchange the two objects in memory.  This exploits base+index addressing,
+ * which basically all CPUs have, to minimize loop overhead computations.
+ *
+ * For some reason, on x86 gcc 7.3.0 adds a redundant test of n at the
+ * bottom of the loop, even though the zero flag is still valid from the
+ * subtract (since the intervening mov instructions don't alter the flags).
+ * Gcc 8.1.0 doesn't have that problem.
+ */
+static void swap_words_32(void *a, void *b, size_t n)
+{
+	do {
+		u32 t = *(u32 *)(a + (n -= 4));
+		*(u32 *)(a + n) = *(u32 *)(b + n);
+		*(u32 *)(b + n) = t;
+	} while (n);
+}
+
+/**
+ * swap_words_64 - swap two elements in 64-bit chunks
+ * @a: pointer to the first element to swap
+ * @b: pointer to the second element to swap
+ * @n: element size (must be a multiple of 8)
+ *
+ * Exchange the two objects in memory.  This exploits base+index
+ * addressing, which basically all CPUs have, to minimize loop overhead
+ * computations.
+ *
+ * We'd like to use 64-bit loads if possible.  If they're not, emulating
+ * one requires base+index+4 addressing which x86 has but most other
+ * processors do not.  If CONFIG_64BIT, we definitely have 64-bit loads,
+ * but it's possible to have 64-bit loads without 64-bit pointers (e.g.
+ * x32 ABI).  Are there any cases the kernel needs to worry about?
+ */
+static void swap_words_64(void *a, void *b, size_t n)
+{
+	do {
+#ifdef CONFIG_64BIT
+		u64 t = *(u64 *)(a + (n -= 8));
+		*(u64 *)(a + n) = *(u64 *)(b + n);
+		*(u64 *)(b + n) = t;
+#else
+		/* Use two 32-bit transfers to avoid base+index+4 addressing */
+		u32 t = *(u32 *)(a + (n -= 4));
+		*(u32 *)(a + n) = *(u32 *)(b + n);
+		*(u32 *)(b + n) = t;
+
+		t = *(u32 *)(a + (n -= 4));
+		*(u32 *)(a + n) = *(u32 *)(b + n);
+		*(u32 *)(b + n) = t;
+#endif
+	} while (n);
+}
+
+/**
+ * swap_bytes - swap two elements a byte at a time
+ * @a: pointer to the first element to swap
+ * @b: pointer to the second element to swap
+ * @n: element size
+ *
+ * This is the fallback if alignment doesn't allow using larger chunks.
+ */
+static void swap_bytes(void *a, void *b, size_t n)
+{
+	do {
+		char t = ((char *)a)[--n];
+		((char *)a)[n] = ((char *)b)[n];
+		((char *)b)[n] = t;
+	} while (n);
+}
+
+/*
+ * The values are arbitrary as long as they can't be confused with
+ * a pointer, but small integers make for the smallest compare
+ * instructions.
+ */
+#define SWAP_WORDS_64 (swap_r_func_t)0
+#define SWAP_WORDS_32 (swap_r_func_t)1
+#define SWAP_BYTES    (swap_r_func_t)2
+#define SWAP_WRAPPER  (swap_r_func_t)3
+
+struct wrapper {
+	cmp_func_t cmp;
+	swap_func_t swap_func;
+};
+
+/*
+ * The function pointer is last to make tail calls most efficient if the
+ * compiler decides not to inline this function.
+ */
+static void do_swap(void *a, void *b, size_t size, swap_r_func_t swap_func, const void *priv)
+{
+	if (swap_func == SWAP_WRAPPER) {
+		((const struct wrapper *)priv)->swap_func(a, b, (int)size);
+		return;
+	}
+
+	if (swap_func == SWAP_WORDS_64)
+		swap_words_64(a, b, size);
+	else if (swap_func == SWAP_WORDS_32)
+		swap_words_32(a, b, size);
+	else if (swap_func == SWAP_BYTES)
+		swap_bytes(a, b, size);
+	else
+		swap_func(a, b, (int)size, priv);
+}
+
+#define _CMP_WRAPPER ((cmp_r_func_t)0L)
+
+static int do_cmp(const void *a, const void *b, cmp_r_func_t cmp, const void *priv)
+{
+	if (cmp == _CMP_WRAPPER)
+		return ((const struct wrapper *)priv)->cmp(a, b);
+	return cmp(a, b, priv);
+}
+
+static inline int eytzinger0_do_cmp(void *base, size_t n, size_t size,
+			 cmp_r_func_t cmp_func, const void *priv,
+			 size_t l, size_t r)
+{
+	return do_cmp(base + inorder_to_eytzinger0(l, n) * size,
+		      base + inorder_to_eytzinger0(r, n) * size,
+		      cmp_func, priv);
+}
+
+static inline void eytzinger0_do_swap(void *base, size_t n, size_t size,
+			   swap_r_func_t swap_func, const void *priv,
+			   size_t l, size_t r)
+{
+	do_swap(base + inorder_to_eytzinger0(l, n) * size,
+		base + inorder_to_eytzinger0(r, n) * size,
+		size, swap_func, priv);
+}
+
+void eytzinger0_sort_r(void *base, size_t n, size_t size,
+		       cmp_r_func_t cmp_func,
+		       swap_r_func_t swap_func,
+		       const void *priv)
+{
+	int i, c, r;
+
+	/* called from 'sort' without swap function, let's pick the default */
+	if (swap_func == SWAP_WRAPPER && !((struct wrapper *)priv)->swap_func)
+		swap_func = NULL;
+
+	if (!swap_func) {
+		if (is_aligned(base, size, 8))
+			swap_func = SWAP_WORDS_64;
+		else if (is_aligned(base, size, 4))
+			swap_func = SWAP_WORDS_32;
+		else
+			swap_func = SWAP_BYTES;
+	}
+
+	/* heapify */
+	for (i = n / 2 - 1; i >= 0; --i) {
+		for (r = i; r * 2 + 1 < n; r = c) {
+			c = r * 2 + 1;
+
+			if (c + 1 < n &&
+			    eytzinger0_do_cmp(base, n, size, cmp_func, priv, c, c + 1) < 0)
+				c++;
+
+			if (eytzinger0_do_cmp(base, n, size, cmp_func, priv, r, c) >= 0)
+				break;
+
+			eytzinger0_do_swap(base, n, size, swap_func, priv, r, c);
+		}
+	}
+
+	/* sort */
+	for (i = n - 1; i > 0; --i) {
+		eytzinger0_do_swap(base, n, size, swap_func, priv, 0, i);
+
+		for (r = 0; r * 2 + 1 < i; r = c) {
+			c = r * 2 + 1;
+
+			if (c + 1 < i &&
+			    eytzinger0_do_cmp(base, n, size, cmp_func, priv, c, c + 1) < 0)
+				c++;
+
+			if (eytzinger0_do_cmp(base, n, size, cmp_func, priv, r, c) >= 0)
+				break;
+
+			eytzinger0_do_swap(base, n, size, swap_func, priv, r, c);
+		}
+	}
+}
+
+void eytzinger0_sort(void *base, size_t n, size_t size,
+		     cmp_func_t cmp_func,
+		     swap_func_t swap_func)
+{
+	struct wrapper w = {
+		.cmp  = cmp_func,
+		.swap_func = swap_func,
+	};
+
+	return eytzinger0_sort_r(base, n, size, _CMP_WRAPPER, SWAP_WRAPPER, &w);
+}
diff --git a/fs/bcachefs/eytzinger.h b/fs/bcachefs/eytzinger.h
index b04750dbf870..24840aee335c 100644
--- a/fs/bcachefs/eytzinger.h
+++ b/fs/bcachefs/eytzinger.h
@@ -5,23 +5,33 @@
 #include <linux/bitops.h>
 #include <linux/log2.h>
 
-#include "util.h"
+#ifdef EYTZINGER_DEBUG
+#define EYTZINGER_BUG_ON(cond)		BUG_ON(cond)
+#else
+#define EYTZINGER_BUG_ON(cond)
+#endif
 
 /*
  * Traversal for trees in eytzinger layout - a full binary tree layed out in an
- * array
- */
-
-/*
- * One based indexing version:
+ * array.
+ *
+ * Consider using an eytzinger tree any time you would otherwise be doing binary
+ * search over an array. Binary search is a worst case scenario for branch
+ * prediction and prefetching, but in an eytzinger tree every node's children
+ * are adjacent in memory, thus we can prefetch children before knowing the
+ * result of the comparison, assuming multiple nodes fit on a cacheline.
  *
- * With one based indexing each level of the tree starts at a power of two -
- * good for cacheline alignment:
+ * Two variants are provided, for one based indexing and zero based indexing.
+ *
+ * Zero based indexing is more convenient, but one based indexing has better
+ * alignment and thus better performance because each new level of the tree
+ * starts at a power of two, and thus if element 0 was cacheline aligned, each
+ * new level will be as well.
  */
 
 static inline unsigned eytzinger1_child(unsigned i, unsigned child)
 {
-	EBUG_ON(child > 1);
+	EYTZINGER_BUG_ON(child > 1);
 
 	return (i << 1) + child;
 }
@@ -58,7 +68,7 @@ static inline unsigned eytzinger1_last(unsigned size)
 
 static inline unsigned eytzinger1_next(unsigned i, unsigned size)
 {
-	EBUG_ON(i > size);
+	EYTZINGER_BUG_ON(i > size);
 
 	if (eytzinger1_right_child(i) <= size) {
 		i = eytzinger1_right_child(i);
@@ -74,7 +84,7 @@ static inline unsigned eytzinger1_next(unsigned i, unsigned size)
 
 static inline unsigned eytzinger1_prev(unsigned i, unsigned size)
 {
-	EBUG_ON(i > size);
+	EYTZINGER_BUG_ON(i > size);
 
 	if (eytzinger1_left_child(i) <= size) {
 		i = eytzinger1_left_child(i) + 1;
@@ -101,7 +111,7 @@ static inline unsigned __eytzinger1_to_inorder(unsigned i, unsigned size,
 	unsigned shift = __fls(size) - b;
 	int s;
 
-	EBUG_ON(!i || i > size);
+	EYTZINGER_BUG_ON(!i || i > size);
 
 	i  ^= 1U << b;
 	i <<= 1;
@@ -126,7 +136,7 @@ static inline unsigned __inorder_to_eytzinger1(unsigned i, unsigned size,
 	unsigned shift;
 	int s;
 
-	EBUG_ON(!i || i > size);
+	EYTZINGER_BUG_ON(!i || i > size);
 
 	/*
 	 * sign bit trick:
@@ -164,7 +174,7 @@ static inline unsigned inorder_to_eytzinger1(unsigned i, unsigned size)
 
 static inline unsigned eytzinger0_child(unsigned i, unsigned child)
 {
-	EBUG_ON(child > 1);
+	EYTZINGER_BUG_ON(child > 1);
 
 	return (i << 1) + 1 + child;
 }
@@ -231,11 +241,9 @@ static inline unsigned inorder_to_eytzinger0(unsigned i, unsigned size)
 	     (_i) != -1;				\
 	     (_i) = eytzinger0_next((_i), (_size)))
 
-typedef int (*eytzinger_cmp_fn)(const void *l, const void *r, size_t size);
-
 /* return greatest node <= @search, or -1 if not found */
-static inline ssize_t eytzinger0_find_le(void *base, size_t nr, size_t size,
-					 eytzinger_cmp_fn cmp, const void *search)
+static inline int eytzinger0_find_le(void *base, size_t nr, size_t size,
+				     cmp_func_t cmp, const void *search)
 {
 	unsigned i, n = 0;
 
@@ -244,21 +252,38 @@ static inline ssize_t eytzinger0_find_le(void *base, size_t nr, size_t size,
 
 	do {
 		i = n;
-		n = eytzinger0_child(i, cmp(search, base + i * size, size) >= 0);
+		n = eytzinger0_child(i, cmp(base + i * size, search) <= 0);
 	} while (n < nr);
 
 	if (n & 1) {
-		/* @i was greater than @search, return previous node: */
-
-		if (i == eytzinger0_first(nr))
-			return -1;
-
+		/*
+		 * @i was greater than @search, return previous node:
+		 *
+		 * if @i was leftmost/smallest element,
+		 * eytzinger0_prev(eytzinger0_first())) returns -1, as expected
+		 */
 		return eytzinger0_prev(i, nr);
 	} else {
 		return i;
 	}
 }
 
+static inline int eytzinger0_find_gt(void *base, size_t nr, size_t size,
+				     cmp_func_t cmp, const void *search)
+{
+	ssize_t idx = eytzinger0_find_le(base, nr, size, cmp, search);
+
+	/*
+	 * if eytitzinger0_find_le() returned -1 - no element was <= search - we
+	 * want to return the first element; next/prev identities mean this work
+	 * as expected
+	 *
+	 * similarly if find_le() returns last element, we should return -1;
+	 * identities mean this all works out:
+	 */
+	return eytzinger0_next(idx, nr);
+}
+
 #define eytzinger0_find(base, nr, size, _cmp, search)			\
 ({									\
 	void *_base		= (base);				\
@@ -269,13 +294,13 @@ static inline ssize_t eytzinger0_find_le(void *base, size_t nr, size_t size,
 	int _res;							\
 									\
 	while (_i < _nr &&						\
-	       (_res = _cmp(_search, _base + _i * _size, _size)))	\
+	       (_res = _cmp(_search, _base + _i * _size)))		\
 		_i = eytzinger0_child(_i, _res > 0);			\
 	_i;								\
 })
 
-void eytzinger0_sort(void *, size_t, size_t,
-		    int (*cmp_func)(const void *, const void *, size_t),
-		    void (*swap_func)(void *, void *, size_t));
+void eytzinger0_sort_r(void *, size_t, size_t,
+		       cmp_r_func_t, swap_r_func_t, const void *);
+void eytzinger0_sort(void *, size_t, size_t, cmp_func_t, swap_func_t);
 
 #endif /* _EYTZINGER_H */
diff --git a/fs/bcachefs/fs-io-direct.c b/fs/bcachefs/fs-io-direct.c
index 33cb6da3a5ad..b889370a5088 100644
--- a/fs/bcachefs/fs-io-direct.c
+++ b/fs/bcachefs/fs-io-direct.c
@@ -387,6 +387,8 @@ static __always_inline long bch2_dio_write_done(struct dio_write *dio)
 	ret = dio->op.error ?: ((long) dio->written << 9);
 	bio_put(&dio->op.wbio.bio);
 
+	bch2_write_ref_put(dio->op.c, BCH_WRITE_REF_dio_write);
+
 	/* inode->i_dio_count is our ref on inode and thus bch_fs */
 	inode_dio_end(&inode->v);
 
@@ -536,7 +538,7 @@ static __always_inline long bch2_dio_write_loop(struct dio_write *dio)
 		if (likely(!dio->iter.count) || dio->op.error)
 			break;
 
-		bio_reset(bio, NULL, REQ_OP_WRITE);
+		bio_reset(bio, NULL, REQ_OP_WRITE | REQ_SYNC | REQ_IDLE);
 	}
 out:
 	return bch2_dio_write_done(dio);
@@ -590,22 +592,25 @@ ssize_t bch2_direct_write(struct kiocb *req, struct iov_iter *iter)
 	prefetch(&inode->ei_inode);
 	prefetch((void *) &inode->ei_inode + 64);
 
+	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_dio_write))
+		return -EROFS;
+
 	inode_lock(&inode->v);
 
 	ret = generic_write_checks(req, iter);
 	if (unlikely(ret <= 0))
-		goto err;
+		goto err_put_write_ref;
 
 	ret = file_remove_privs(file);
 	if (unlikely(ret))
-		goto err;
+		goto err_put_write_ref;
 
 	ret = file_update_time(file);
 	if (unlikely(ret))
-		goto err;
+		goto err_put_write_ref;
 
 	if (unlikely((req->ki_pos|iter->count) & (block_bytes(c) - 1)))
-		goto err;
+		goto err_put_write_ref;
 
 	inode_dio_begin(&inode->v);
 	bch2_pagecache_block_get(inode);
@@ -618,7 +623,7 @@ ssize_t bch2_direct_write(struct kiocb *req, struct iov_iter *iter)
 
 	bio = bio_alloc_bioset(NULL,
 			       bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS),
-			       REQ_OP_WRITE,
+			       REQ_OP_WRITE | REQ_SYNC | REQ_IDLE,
 			       GFP_KERNEL,
 			       &c->dio_write_bioset);
 	dio = container_of(bio, struct dio_write, op.wbio.bio);
@@ -645,7 +650,7 @@ ssize_t bch2_direct_write(struct kiocb *req, struct iov_iter *iter)
 	}
 
 	ret = bch2_dio_write_loop(dio);
-err:
+out:
 	if (locked)
 		inode_unlock(&inode->v);
 	return ret;
@@ -653,7 +658,9 @@ err_put_bio:
 	bch2_pagecache_block_put(inode);
 	bio_put(bio);
 	inode_dio_end(&inode->v);
-	goto err;
+err_put_write_ref:
+	bch2_write_ref_put(c, BCH_WRITE_REF_dio_write);
+	goto out;
 }
 
 void bch2_fs_fs_io_direct_exit(struct bch_fs *c)
diff --git a/fs/bcachefs/fs-io.c b/fs/bcachefs/fs-io.c
index 8c70123b6a0c..20b40477425f 100644
--- a/fs/bcachefs/fs-io.c
+++ b/fs/bcachefs/fs-io.c
@@ -174,18 +174,18 @@ void __bch2_i_sectors_acct(struct bch_fs *c, struct bch_inode_info *inode,
 static int bch2_flush_inode(struct bch_fs *c,
 			    struct bch_inode_info *inode)
 {
-	struct bch_inode_unpacked u;
-	int ret;
-
 	if (c->opts.journal_flush_disabled)
 		return 0;
 
-	ret = bch2_inode_find_by_inum(c, inode_inum(inode), &u);
-	if (ret)
-		return ret;
+	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_fsync))
+		return -EROFS;
 
-	return bch2_journal_flush_seq(&c->journal, u.bi_journal_seq) ?:
-		bch2_inode_flush_nocow_writes(c, inode);
+	struct bch_inode_unpacked u;
+	int ret = bch2_inode_find_by_inum(c, inode_inum(inode), &u) ?:
+		  bch2_journal_flush_seq(&c->journal, u.bi_journal_seq) ?:
+		  bch2_inode_flush_nocow_writes(c, inode);
+	bch2_write_ref_put(c, BCH_WRITE_REF_fsync);
+	return ret;
 }
 
 int bch2_fsync(struct file *file, loff_t start, loff_t end, int datasync)
diff --git a/fs/bcachefs/fs.c b/fs/bcachefs/fs.c
index 0ccee05f6887..b5ea9fa1259d 100644
--- a/fs/bcachefs/fs.c
+++ b/fs/bcachefs/fs.c
@@ -1997,6 +1997,7 @@ out:
 	return dget(sb->s_root);
 
 err_put_super:
+	__bch2_fs_stop(c);
 	deactivate_locked_super(sb);
 	return ERR_PTR(bch2_err_class(ret));
 }
diff --git a/fs/bcachefs/fsck.c b/fs/bcachefs/fsck.c
index 47d4eefaba7b..8e2010212cc3 100644
--- a/fs/bcachefs/fsck.c
+++ b/fs/bcachefs/fsck.c
@@ -12,7 +12,7 @@
 #include "fsck.h"
 #include "inode.h"
 #include "keylist.h"
-#include "recovery.h"
+#include "recovery_passes.h"
 #include "snapshot.h"
 #include "super.h"
 #include "xattr.h"
@@ -63,9 +63,7 @@ static int subvol_lookup(struct btree_trans *trans, u32 subvol,
 			 u32 *snapshot, u64 *inum)
 {
 	struct bch_subvolume s;
-	int ret;
-
-	ret = bch2_subvolume_get(trans, subvol, false, 0, &s);
+	int ret = bch2_subvolume_get(trans, subvol, false, 0, &s);
 
 	*snapshot = le32_to_cpu(s.snapshot);
 	*inum = le64_to_cpu(s.inode);
@@ -158,9 +156,10 @@ static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
 
 	bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_INTENT);
 
-	ret = bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
-				  &dir_hash_info, &iter,
-				  BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
+	ret =   bch2_btree_iter_traverse(&iter) ?:
+		bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
+				    &dir_hash_info, &iter,
+				    BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 	bch2_trans_iter_exit(trans, &iter);
 err:
 	bch_err_fn(c, ret);
@@ -169,7 +168,8 @@ err:
 
 /* Get lost+found, create if it doesn't exist: */
 static int lookup_lostfound(struct btree_trans *trans, u32 snapshot,
-			    struct bch_inode_unpacked *lostfound)
+			    struct bch_inode_unpacked *lostfound,
+			    u64 reattaching_inum)
 {
 	struct bch_fs *c = trans->c;
 	struct qstr lostfound_str = QSTR("lost+found");
@@ -184,19 +184,36 @@ static int lookup_lostfound(struct btree_trans *trans, u32 snapshot,
 		return ret;
 
 	subvol_inum root_inum = { .subvol = le32_to_cpu(st.master_subvol) };
-	u32 subvol_snapshot;
 
-	ret = subvol_lookup(trans, le32_to_cpu(st.master_subvol),
-			    &subvol_snapshot, &root_inum.inum);
-	bch_err_msg(c, ret, "looking up root subvol");
+	struct bch_subvolume subvol;
+	ret = bch2_subvolume_get(trans, le32_to_cpu(st.master_subvol),
+				 false, 0, &subvol);
+	bch_err_msg(c, ret, "looking up root subvol %u for snapshot %u",
+		    le32_to_cpu(st.master_subvol), snapshot);
 	if (ret)
 		return ret;
 
+	if (!subvol.inode) {
+		struct btree_iter iter;
+		struct bkey_i_subvolume *subvol = bch2_bkey_get_mut_typed(trans, &iter,
+				BTREE_ID_subvolumes, POS(0, le32_to_cpu(st.master_subvol)),
+				0, subvolume);
+		ret = PTR_ERR_OR_ZERO(subvol);
+		if (ret)
+			return ret;
+
+		subvol->v.inode = cpu_to_le64(reattaching_inum);
+		bch2_trans_iter_exit(trans, &iter);
+	}
+
+	root_inum.inum = le64_to_cpu(subvol.inode);
+
 	struct bch_inode_unpacked root_inode;
 	struct bch_hash_info root_hash_info;
 	u32 root_inode_snapshot = snapshot;
 	ret = lookup_inode(trans, root_inum.inum, &root_inode, &root_inode_snapshot);
-	bch_err_msg(c, ret, "looking up root inode");
+	bch_err_msg(c, ret, "looking up root inode %llu for subvol %u",
+		    root_inum.inum, le32_to_cpu(st.master_subvol));
 	if (ret)
 		return ret;
 
@@ -292,7 +309,7 @@ static int reattach_inode(struct btree_trans *trans,
 		snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
 	}
 
-	ret = lookup_lostfound(trans, dirent_snapshot, &lostfound);
+	ret = lookup_lostfound(trans, dirent_snapshot, &lostfound, inode->bi_inum);
 	if (ret)
 		return ret;
 
@@ -363,6 +380,112 @@ static int reattach_subvol(struct btree_trans *trans, struct bkey_s_c_subvolume
 	return ret;
 }
 
+static int reconstruct_subvol(struct btree_trans *trans, u32 snapshotid, u32 subvolid, u64 inum)
+{
+	struct bch_fs *c = trans->c;
+
+	if (!bch2_snapshot_is_leaf(c, snapshotid)) {
+		bch_err(c, "need to reconstruct subvol, but have interior node snapshot");
+		return -BCH_ERR_fsck_repair_unimplemented;
+	}
+
+	/*
+	 * If inum isn't set, that means we're being called from check_dirents,
+	 * not check_inodes - the root of this subvolume doesn't exist or we
+	 * would have found it there:
+	 */
+	if (!inum) {
+		struct btree_iter inode_iter = {};
+		struct bch_inode_unpacked new_inode;
+		u64 cpu = raw_smp_processor_id();
+
+		bch2_inode_init_early(c, &new_inode);
+		bch2_inode_init_late(&new_inode, bch2_current_time(c), 0, 0, S_IFDIR|0755, 0, NULL);
+
+		new_inode.bi_subvol = subvolid;
+
+		int ret = bch2_inode_create(trans, &inode_iter, &new_inode, snapshotid, cpu) ?:
+			  bch2_btree_iter_traverse(&inode_iter) ?:
+			  bch2_inode_write(trans, &inode_iter, &new_inode);
+		bch2_trans_iter_exit(trans, &inode_iter);
+		if (ret)
+			return ret;
+
+		inum = new_inode.bi_inum;
+	}
+
+	bch_info(c, "reconstructing subvol %u with root inode %llu", subvolid, inum);
+
+	struct bkey_i_subvolume *new_subvol = bch2_trans_kmalloc(trans, sizeof(*new_subvol));
+	int ret = PTR_ERR_OR_ZERO(new_subvol);
+	if (ret)
+		return ret;
+
+	bkey_subvolume_init(&new_subvol->k_i);
+	new_subvol->k.p.offset	= subvolid;
+	new_subvol->v.snapshot	= cpu_to_le32(snapshotid);
+	new_subvol->v.inode	= cpu_to_le64(inum);
+	ret = bch2_btree_insert_trans(trans, BTREE_ID_subvolumes, &new_subvol->k_i, 0);
+	if (ret)
+		return ret;
+
+	struct btree_iter iter;
+	struct bkey_i_snapshot *s = bch2_bkey_get_mut_typed(trans, &iter,
+			BTREE_ID_snapshots, POS(0, snapshotid),
+			0, snapshot);
+	ret = PTR_ERR_OR_ZERO(s);
+	bch_err_msg(c, ret, "getting snapshot %u", snapshotid);
+	if (ret)
+		return ret;
+
+	u32 snapshot_tree = le32_to_cpu(s->v.tree);
+
+	s->v.subvol = cpu_to_le32(subvolid);
+	SET_BCH_SNAPSHOT_SUBVOL(&s->v, true);
+	bch2_trans_iter_exit(trans, &iter);
+
+	struct bkey_i_snapshot_tree *st = bch2_bkey_get_mut_typed(trans, &iter,
+			BTREE_ID_snapshot_trees, POS(0, snapshot_tree),
+			0, snapshot_tree);
+	ret = PTR_ERR_OR_ZERO(st);
+	bch_err_msg(c, ret, "getting snapshot tree %u", snapshot_tree);
+	if (ret)
+		return ret;
+
+	if (!st->v.master_subvol)
+		st->v.master_subvol = cpu_to_le32(subvolid);
+
+	bch2_trans_iter_exit(trans, &iter);
+	return 0;
+}
+
+static int reconstruct_inode(struct btree_trans *trans, u32 snapshot, u64 inum, u64 size, unsigned mode)
+{
+	struct bch_fs *c = trans->c;
+	struct bch_inode_unpacked new_inode;
+
+	bch2_inode_init_early(c, &new_inode);
+	bch2_inode_init_late(&new_inode, bch2_current_time(c), 0, 0, mode|0755, 0, NULL);
+	new_inode.bi_size = size;
+	new_inode.bi_inum = inum;
+
+	return __bch2_fsck_write_inode(trans, &new_inode, snapshot);
+}
+
+static int reconstruct_reg_inode(struct btree_trans *trans, u32 snapshot, u64 inum)
+{
+	struct btree_iter iter = {};
+
+	bch2_trans_iter_init(trans, &iter, BTREE_ID_extents, SPOS(inum, U64_MAX, snapshot), 0);
+	struct bkey_s_c k = bch2_btree_iter_peek_prev(&iter);
+	bch2_trans_iter_exit(trans, &iter);
+	int ret = bkey_err(k);
+	if (ret)
+		return ret;
+
+	return reconstruct_inode(trans, snapshot, inum, k.k->p.offset << 9, S_IFREG);
+}
+
 struct snapshots_seen_entry {
 	u32				id;
 	u32				equiv;
@@ -1064,6 +1187,11 @@ static int check_inode(struct btree_trans *trans,
 		if (ret && !bch2_err_matches(ret, ENOENT))
 			goto err;
 
+		if (ret && (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_subvolumes))) {
+			ret = reconstruct_subvol(trans, k.k->p.snapshot, u.bi_subvol, u.bi_inum);
+			goto do_update;
+		}
+
 		if (fsck_err_on(ret,
 				c, inode_bi_subvol_missing,
 				"inode %llu:%u bi_subvol points to missing subvolume %u",
@@ -1081,7 +1209,7 @@ static int check_inode(struct btree_trans *trans,
 			do_update = true;
 		}
 	}
-
+do_update:
 	if (do_update) {
 		ret = __bch2_fsck_write_inode(trans, &u, iter->pos.snapshot);
 		bch_err_msg(c, ret, "in fsck updating inode");
@@ -1130,8 +1258,8 @@ static int check_i_sectors_notnested(struct btree_trans *trans, struct inode_wal
 			i->count = count2;
 
 		if (i->count != count2) {
-			bch_err(c, "fsck counted i_sectors wrong for inode %llu:%u: got %llu should be %llu",
-				w->last_pos.inode, i->snapshot, i->count, count2);
+			bch_err_ratelimited(c, "fsck counted i_sectors wrong for inode %llu:%u: got %llu should be %llu",
+					    w->last_pos.inode, i->snapshot, i->count, count2);
 			return -BCH_ERR_internal_fsck_err;
 		}
 
@@ -1371,10 +1499,6 @@ static int check_overlapping_extents(struct btree_trans *trans,
 			goto err;
 	}
 
-	ret = extent_ends_at(c, extent_ends, seen, k);
-	if (ret)
-		goto err;
-
 	extent_ends->last_pos = k.k->p;
 err:
 	return ret;
@@ -1438,6 +1562,17 @@ static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
 		goto err;
 
 	if (k.k->type != KEY_TYPE_whiteout) {
+		if (!i && (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_inodes))) {
+			ret =   reconstruct_reg_inode(trans, k.k->p.snapshot, k.k->p.inode) ?:
+				bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
+			if (ret)
+				goto err;
+
+			inode->last_pos.inode--;
+			ret = -BCH_ERR_transaction_restart_nested;
+			goto err;
+		}
+
 		if (fsck_err_on(!i, c, extent_in_missing_inode,
 				"extent in missing inode:\n  %s",
 				(printbuf_reset(&buf),
@@ -1504,6 +1639,12 @@ static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
 
 		i->seen_this_pos = true;
 	}
+
+	if (k.k->type != KEY_TYPE_whiteout) {
+		ret = extent_ends_at(c, extent_ends, s, k);
+		if (ret)
+			goto err;
+	}
 out:
 err:
 fsck_err:
@@ -1584,8 +1725,8 @@ static int check_subdir_count_notnested(struct btree_trans *trans, struct inode_
 			return count2;
 
 		if (i->count != count2) {
-			bch_err(c, "fsck counted subdirectories wrong: got %llu should be %llu",
-				i->count, count2);
+			bch_err_ratelimited(c, "fsck counted subdirectories wrong for inum %llu:%u: got %llu should be %llu",
+					    w->last_pos.inode, i->snapshot, i->count, count2);
 			i->count = count2;
 			if (i->inode.bi_nlink == i->count)
 				continue;
@@ -1782,6 +1923,7 @@ static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter *
 	u32 parent_subvol = le32_to_cpu(d.v->d_parent_subvol);
 	u32 target_subvol = le32_to_cpu(d.v->d_child_subvol);
 	u32 parent_snapshot;
+	u32 new_parent_subvol = 0;
 	u64 parent_inum;
 	struct printbuf buf = PRINTBUF;
 	int ret = 0;
@@ -1790,6 +1932,27 @@ static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter *
 	if (ret && !bch2_err_matches(ret, ENOENT))
 		return ret;
 
+	if (ret ||
+	    (!ret && !bch2_snapshot_is_ancestor(c, parent_snapshot, d.k->p.snapshot))) {
+		int ret2 = find_snapshot_subvol(trans, d.k->p.snapshot, &new_parent_subvol);
+		if (ret2 && !bch2_err_matches(ret, ENOENT))
+			return ret2;
+	}
+
+	if (ret &&
+	    !new_parent_subvol &&
+	    (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_subvolumes))) {
+		/*
+		 * Couldn't find a subvol for dirent's snapshot - but we lost
+		 * subvols, so we need to reconstruct:
+		 */
+		ret = reconstruct_subvol(trans, d.k->p.snapshot, parent_subvol, 0);
+		if (ret)
+			return ret;
+
+		parent_snapshot = d.k->p.snapshot;
+	}
+
 	if (fsck_err_on(ret, c, dirent_to_missing_parent_subvol,
 			"dirent parent_subvol points to missing subvolume\n%s",
 			(bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf)) ||
@@ -1798,10 +1961,10 @@ static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter *
 			"dirent not visible in parent_subvol (not an ancestor of subvol snap %u)\n%s",
 			parent_snapshot,
 			(bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
-		u32 new_parent_subvol;
-		ret = find_snapshot_subvol(trans, d.k->p.snapshot, &new_parent_subvol);
-		if (ret)
-			goto err;
+		if (!new_parent_subvol) {
+			bch_err(c, "could not find a subvol for snapshot %u", d.k->p.snapshot);
+			return -BCH_ERR_fsck_repair_unimplemented;
+		}
 
 		struct bkey_i_dirent *new_dirent = bch2_bkey_make_mut_typed(trans, iter, &d.s_c, 0, dirent);
 		ret = PTR_ERR_OR_ZERO(new_dirent);
@@ -1847,9 +2010,16 @@ static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter *
 
 	ret = lookup_inode(trans, target_inum, &subvol_root, &target_snapshot);
 	if (ret && !bch2_err_matches(ret, ENOENT))
-		return ret;
+		goto err;
+
+	if (ret) {
+		bch_err(c, "subvol %u points to missing inode root %llu", target_subvol, target_inum);
+		ret = -BCH_ERR_fsck_repair_unimplemented;
+		ret = 0;
+		goto err;
+	}
 
-	if (fsck_err_on(parent_subvol != subvol_root.bi_parent_subvol,
+	if (fsck_err_on(!ret && parent_subvol != subvol_root.bi_parent_subvol,
 			c, inode_bi_parent_wrong,
 			"subvol root %llu has wrong bi_parent_subvol: got %u, should be %u",
 			target_inum,
@@ -1857,13 +2027,13 @@ static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter *
 		subvol_root.bi_parent_subvol = parent_subvol;
 		ret = __bch2_fsck_write_inode(trans, &subvol_root, target_snapshot);
 		if (ret)
-			return ret;
+			goto err;
 	}
 
 	ret = check_dirent_target(trans, iter, d, &subvol_root,
 				  target_snapshot);
 	if (ret)
-		return ret;
+		goto err;
 out:
 err:
 fsck_err:
@@ -1880,7 +2050,6 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
 			struct snapshots_seen *s)
 {
 	struct bch_fs *c = trans->c;
-	struct bkey_s_c_dirent d;
 	struct inode_walker_entry *i;
 	struct printbuf buf = PRINTBUF;
 	struct bpos equiv;
@@ -1919,6 +2088,17 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
 		*hash_info = bch2_hash_info_init(c, &dir->inodes.data[0].inode);
 	dir->first_this_inode = false;
 
+	if (!i && (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_inodes))) {
+		ret =   reconstruct_inode(trans, k.k->p.snapshot, k.k->p.inode, 0, S_IFDIR) ?:
+			bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
+		if (ret)
+			goto err;
+
+		dir->last_pos.inode--;
+		ret = -BCH_ERR_transaction_restart_nested;
+		goto err;
+	}
+
 	if (fsck_err_on(!i, c, dirent_in_missing_dir_inode,
 			"dirent in nonexisting directory:\n%s",
 			(printbuf_reset(&buf),
@@ -1953,7 +2133,7 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
 	if (k.k->type != KEY_TYPE_dirent)
 		goto out;
 
-	d = bkey_s_c_to_dirent(k);
+	struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
 
 	if (d.v->d_type == DT_SUBVOL) {
 		ret = check_dirent_to_subvol(trans, iter, d);
@@ -2098,17 +2278,21 @@ static int check_root_trans(struct btree_trans *trans)
 
 	if (mustfix_fsck_err_on(ret, c, root_subvol_missing,
 				"root subvol missing")) {
-		struct bkey_i_subvolume root_subvol;
+		struct bkey_i_subvolume *root_subvol =
+			bch2_trans_kmalloc(trans, sizeof(*root_subvol));
+		ret = PTR_ERR_OR_ZERO(root_subvol);
+		if (ret)
+			goto err;
 
 		snapshot	= U32_MAX;
 		inum		= BCACHEFS_ROOT_INO;
 
-		bkey_subvolume_init(&root_subvol.k_i);
-		root_subvol.k.p.offset = BCACHEFS_ROOT_SUBVOL;
-		root_subvol.v.flags	= 0;
-		root_subvol.v.snapshot	= cpu_to_le32(snapshot);
-		root_subvol.v.inode	= cpu_to_le64(inum);
-		ret = bch2_btree_insert_trans(trans, BTREE_ID_subvolumes, &root_subvol.k_i, 0);
+		bkey_subvolume_init(&root_subvol->k_i);
+		root_subvol->k.p.offset = BCACHEFS_ROOT_SUBVOL;
+		root_subvol->v.flags	= 0;
+		root_subvol->v.snapshot	= cpu_to_le32(snapshot);
+		root_subvol->v.inode	= cpu_to_le64(inum);
+		ret = bch2_btree_insert_trans(trans, BTREE_ID_subvolumes, &root_subvol->k_i, 0);
 		bch_err_msg(c, ret, "writing root subvol");
 		if (ret)
 			goto err;
diff --git a/fs/bcachefs/inode.c b/fs/bcachefs/inode.c
index 2b5e06770ab3..ca4a066e9a54 100644
--- a/fs/bcachefs/inode.c
+++ b/fs/bcachefs/inode.c
@@ -552,8 +552,8 @@ static void __bch2_inode_unpacked_to_text(struct printbuf *out,
 	prt_printf(out, "bi_sectors=%llu", inode->bi_sectors);
 	prt_newline(out);
 
-	prt_newline(out);
 	prt_printf(out, "bi_version=%llu", inode->bi_version);
+	prt_newline(out);
 
 #define x(_name, _bits)						\
 	prt_printf(out, #_name "=%llu", (u64) inode->_name);	\
diff --git a/fs/bcachefs/io_misc.c b/fs/bcachefs/io_misc.c
index 1baf78594cca..82f9170dab3f 100644
--- a/fs/bcachefs/io_misc.c
+++ b/fs/bcachefs/io_misc.c
@@ -264,6 +264,7 @@ static int __bch2_resume_logged_op_truncate(struct btree_trans *trans,
 		ret = 0;
 err:
 	bch2_logged_op_finish(trans, op_k);
+	bch_err_fn(c, ret);
 	return ret;
 }
 
@@ -476,6 +477,7 @@ case LOGGED_OP_FINSERT_finish:
 	break;
 	}
 err:
+	bch_err_fn(c, ret);
 	bch2_logged_op_finish(trans, op_k);
 	bch2_trans_iter_exit(trans, &iter);
 	return ret;
diff --git a/fs/bcachefs/journal_io.c b/fs/bcachefs/journal_io.c
index 725fcf46f631..9aa28b52ab92 100644
--- a/fs/bcachefs/journal_io.c
+++ b/fs/bcachefs/journal_io.c
@@ -247,7 +247,7 @@ static void journal_entry_err_msg(struct printbuf *out,
 
 	if (entry) {
 		prt_str(out, " type=");
-		prt_str(out, bch2_jset_entry_types[entry->type]);
+		bch2_prt_jset_entry_type(out, entry->type);
 	}
 
 	if (!jset) {
@@ -403,7 +403,8 @@ static void journal_entry_btree_keys_to_text(struct printbuf *out, struct bch_fs
 	jset_entry_for_each_key(entry, k) {
 		if (!first) {
 			prt_newline(out);
-			prt_printf(out, "%s: ", bch2_jset_entry_types[entry->type]);
+			bch2_prt_jset_entry_type(out, entry->type);
+			prt_str(out, ": ");
 		}
 		prt_printf(out, "btree=%s l=%u ", bch2_btree_id_str(entry->btree_id), entry->level);
 		bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(k));
@@ -563,9 +564,9 @@ static void journal_entry_usage_to_text(struct printbuf *out, struct bch_fs *c,
 	struct jset_entry_usage *u =
 		container_of(entry, struct jset_entry_usage, entry);
 
-	prt_printf(out, "type=%s v=%llu",
-	       bch2_fs_usage_types[u->entry.btree_id],
-	       le64_to_cpu(u->v));
+	prt_str(out, "type=");
+	bch2_prt_fs_usage_type(out, u->entry.btree_id);
+	prt_printf(out, " v=%llu", le64_to_cpu(u->v));
 }
 
 static int journal_entry_data_usage_validate(struct bch_fs *c,
@@ -827,11 +828,11 @@ int bch2_journal_entry_validate(struct bch_fs *c,
 void bch2_journal_entry_to_text(struct printbuf *out, struct bch_fs *c,
 				struct jset_entry *entry)
 {
+	bch2_prt_jset_entry_type(out, entry->type);
+
 	if (entry->type < BCH_JSET_ENTRY_NR) {
-		prt_printf(out, "%s: ", bch2_jset_entry_types[entry->type]);
+		prt_str(out, ": ");
 		bch2_jset_entry_ops[entry->type].to_text(out, c, entry);
-	} else {
-		prt_printf(out, "(unknown type %u)", entry->type);
 	}
 }
 
diff --git a/fs/bcachefs/journal_reclaim.c b/fs/bcachefs/journal_reclaim.c
index ab811c0dad26..04a577848b01 100644
--- a/fs/bcachefs/journal_reclaim.c
+++ b/fs/bcachefs/journal_reclaim.c
@@ -67,6 +67,8 @@ void bch2_journal_set_watermark(struct journal *j)
 	    track_event_change(&c->times[BCH_TIME_blocked_write_buffer_full], low_on_wb))
 		trace_and_count(c, journal_full, c);
 
+	mod_bit(JOURNAL_SPACE_LOW, &j->flags, low_on_space || low_on_pin);
+
 	swap(watermark, j->watermark);
 	if (watermark > j->watermark)
 		journal_wake(j);
diff --git a/fs/bcachefs/journal_seq_blacklist.c b/fs/bcachefs/journal_seq_blacklist.c
index b5303874fc35..37a024e034d4 100644
--- a/fs/bcachefs/journal_seq_blacklist.c
+++ b/fs/bcachefs/journal_seq_blacklist.c
@@ -95,8 +95,7 @@ out:
 	return ret ?: bch2_blacklist_table_initialize(c);
 }
 
-static int journal_seq_blacklist_table_cmp(const void *_l,
-					   const void *_r, size_t size)
+static int journal_seq_blacklist_table_cmp(const void *_l, const void *_r)
 {
 	const struct journal_seq_blacklist_table_entry *l = _l;
 	const struct journal_seq_blacklist_table_entry *r = _r;
diff --git a/fs/bcachefs/journal_types.h b/fs/bcachefs/journal_types.h
index 8c053cb64ca5..b5161b5d76a0 100644
--- a/fs/bcachefs/journal_types.h
+++ b/fs/bcachefs/journal_types.h
@@ -134,6 +134,7 @@ enum journal_flags {
 	JOURNAL_STARTED,
 	JOURNAL_MAY_SKIP_FLUSH,
 	JOURNAL_NEED_FLUSH_WRITE,
+	JOURNAL_SPACE_LOW,
 };
 
 /* Reasons we may fail to get a journal reservation: */
diff --git a/fs/bcachefs/logged_ops.c b/fs/bcachefs/logged_ops.c
index 9fac838d123e..b82f8209041f 100644
--- a/fs/bcachefs/logged_ops.c
+++ b/fs/bcachefs/logged_ops.c
@@ -37,7 +37,6 @@ static int resume_logged_op(struct btree_trans *trans, struct btree_iter *iter,
 	const struct bch_logged_op_fn *fn = logged_op_fn(k.k->type);
 	struct bkey_buf sk;
 	u32 restart_count = trans->restart_count;
-	int ret;
 
 	if (!fn)
 		return 0;
@@ -45,11 +44,11 @@ static int resume_logged_op(struct btree_trans *trans, struct btree_iter *iter,
 	bch2_bkey_buf_init(&sk);
 	bch2_bkey_buf_reassemble(&sk, c, k);
 
-	ret =   drop_locks_do(trans, (bch2_fs_lazy_rw(c), 0)) ?:
-		fn->resume(trans, sk.k) ?: trans_was_restarted(trans, restart_count);
+	fn->resume(trans, sk.k);
 
 	bch2_bkey_buf_exit(&sk, c);
-	return ret;
+
+	return trans_was_restarted(trans, restart_count);
 }
 
 int bch2_resume_logged_ops(struct bch_fs *c)
diff --git a/fs/bcachefs/mean_and_variance_test.c b/fs/bcachefs/mean_and_variance_test.c
index db63b3f3b338..4c298e74723d 100644
--- a/fs/bcachefs/mean_and_variance_test.c
+++ b/fs/bcachefs/mean_and_variance_test.c
@@ -136,20 +136,8 @@ static void mean_and_variance_test_1(struct kunit *test)
 			d, mean, stddev, weighted_mean, weighted_stddev);
 }
 
-static void mean_and_variance_test_2(struct kunit *test)
-{
-	s64 d[]			= { 100, 10, 10, 10, 10, 10, 10 };
-	s64 mean[]		= {  10, 10, 10, 10, 10, 10, 10 };
-	s64 stddev[]		= {   9,  9,  9,  9,  9,  9,  9 };
-	s64 weighted_mean[]	= {  32, 27, 22, 19, 17, 15, 14 };
-	s64 weighted_stddev[]	= {  38, 35, 31, 27, 24, 21, 18 };
-
-	do_mean_and_variance_test(test, 10, 6, ARRAY_SIZE(d), 2,
-			d, mean, stddev, weighted_mean, weighted_stddev);
-}
-
 /* Test behaviour where we switch from one steady state to another: */
-static void mean_and_variance_test_3(struct kunit *test)
+static void mean_and_variance_test_2(struct kunit *test)
 {
 	s64 d[]			= { 100, 100, 100, 100, 100 };
 	s64 mean[]		= {  22,  32,  40,  46,  50 };
@@ -161,18 +149,6 @@ static void mean_and_variance_test_3(struct kunit *test)
 			d, mean, stddev, weighted_mean, weighted_stddev);
 }
 
-static void mean_and_variance_test_4(struct kunit *test)
-{
-	s64 d[]			= { 100, 100, 100, 100, 100 };
-	s64 mean[]		= {  10,  11,  12,  13,  14 };
-	s64 stddev[]		= {   9,  13,  15,  17,  19 };
-	s64 weighted_mean[]	= {  32,  49,  61,  71,  78 };
-	s64 weighted_stddev[]	= {  38,  44,  44,  41,  38 };
-
-	do_mean_and_variance_test(test, 10, 6, ARRAY_SIZE(d), 2,
-			d, mean, stddev, weighted_mean, weighted_stddev);
-}
-
 static void mean_and_variance_fast_divpow2(struct kunit *test)
 {
 	s64 i;
@@ -230,8 +206,6 @@ static struct kunit_case mean_and_variance_test_cases[] = {
 	KUNIT_CASE(mean_and_variance_weighted_advanced_test),
 	KUNIT_CASE(mean_and_variance_test_1),
 	KUNIT_CASE(mean_and_variance_test_2),
-	KUNIT_CASE(mean_and_variance_test_3),
-	KUNIT_CASE(mean_and_variance_test_4),
 	{}
 };
 
diff --git a/fs/bcachefs/opts.c b/fs/bcachefs/opts.c
index 08ea0cfc4aef..bb068fd72465 100644
--- a/fs/bcachefs/opts.c
+++ b/fs/bcachefs/opts.c
@@ -7,6 +7,7 @@
 #include "disk_groups.h"
 #include "error.h"
 #include "opts.h"
+#include "recovery_passes.h"
 #include "super-io.h"
 #include "util.h"
 
@@ -42,7 +43,7 @@ const char * const __bch2_btree_ids[] = {
 	NULL
 };
 
-const char * const bch2_csum_types[] = {
+static const char * const __bch2_csum_types[] = {
 	BCH_CSUM_TYPES()
 	NULL
 };
@@ -52,7 +53,7 @@ const char * const bch2_csum_opts[] = {
 	NULL
 };
 
-const char * const __bch2_compression_types[] = {
+static const char * const __bch2_compression_types[] = {
 	BCH_COMPRESSION_TYPES()
 	NULL
 };
@@ -82,18 +83,39 @@ const char * const bch2_member_states[] = {
 	NULL
 };
 
-const char * const bch2_jset_entry_types[] = {
+static const char * const __bch2_jset_entry_types[] = {
 	BCH_JSET_ENTRY_TYPES()
 	NULL
 };
 
-const char * const bch2_fs_usage_types[] = {
+static const char * const __bch2_fs_usage_types[] = {
 	BCH_FS_USAGE_TYPES()
 	NULL
 };
 
 #undef x
 
+static void prt_str_opt_boundscheck(struct printbuf *out, const char * const opts[],
+				    unsigned nr, const char *type, unsigned idx)
+{
+	if (idx < nr)
+		prt_str(out, opts[idx]);
+	else
+		prt_printf(out, "(unknown %s %u)", type, idx);
+}
+
+#define PRT_STR_OPT_BOUNDSCHECKED(name, type)					\
+void bch2_prt_##name(struct printbuf *out, type t)				\
+{										\
+	prt_str_opt_boundscheck(out, __bch2_##name##s, ARRAY_SIZE(__bch2_##name##s) - 1, #name, t);\
+}
+
+PRT_STR_OPT_BOUNDSCHECKED(jset_entry_type,	enum bch_jset_entry_type);
+PRT_STR_OPT_BOUNDSCHECKED(fs_usage_type,	enum bch_fs_usage_type);
+PRT_STR_OPT_BOUNDSCHECKED(data_type,		enum bch_data_type);
+PRT_STR_OPT_BOUNDSCHECKED(csum_type,		enum bch_csum_type);
+PRT_STR_OPT_BOUNDSCHECKED(compression_type,	enum bch_compression_type);
+
 static int bch2_opt_fix_errors_parse(struct bch_fs *c, const char *val, u64 *res,
 				     struct printbuf *err)
 {
@@ -205,6 +227,9 @@ const struct bch_option bch2_opt_table[] = {
 #define OPT_STR(_choices)	.type = BCH_OPT_STR,			\
 				.min = 0, .max = ARRAY_SIZE(_choices),	\
 				.choices = _choices
+#define OPT_STR_NOLIMIT(_choices)	.type = BCH_OPT_STR,		\
+				.min = 0, .max = U64_MAX,		\
+				.choices = _choices
 #define OPT_FN(_fn)		.type = BCH_OPT_FN, .fn	= _fn
 
 #define x(_name, _bits, _flags, _type, _sb_opt, _default, _hint, _help)	\
diff --git a/fs/bcachefs/opts.h b/fs/bcachefs/opts.h
index 136083c11f3a..84e452835a17 100644
--- a/fs/bcachefs/opts.h
+++ b/fs/bcachefs/opts.h
@@ -16,18 +16,20 @@ extern const char * const bch2_version_upgrade_opts[];
 extern const char * const bch2_sb_features[];
 extern const char * const bch2_sb_compat[];
 extern const char * const __bch2_btree_ids[];
-extern const char * const bch2_csum_types[];
 extern const char * const bch2_csum_opts[];
-extern const char * const __bch2_compression_types[];
 extern const char * const bch2_compression_opts[];
 extern const char * const bch2_str_hash_types[];
 extern const char * const bch2_str_hash_opts[];
 extern const char * const __bch2_data_types[];
 extern const char * const bch2_member_states[];
-extern const char * const bch2_jset_entry_types[];
-extern const char * const bch2_fs_usage_types[];
 extern const char * const bch2_d_types[];
 
+void bch2_prt_jset_entry_type(struct printbuf *,	enum bch_jset_entry_type);
+void bch2_prt_fs_usage_type(struct printbuf *,		enum bch_fs_usage_type);
+void bch2_prt_data_type(struct printbuf *,		enum bch_data_type);
+void bch2_prt_csum_type(struct printbuf *,		enum bch_csum_type);
+void bch2_prt_compression_type(struct printbuf *,	enum bch_compression_type);
+
 static inline const char *bch2_d_type_str(unsigned d_type)
 {
 	return (d_type < BCH_DT_MAX ? bch2_d_types[d_type] : NULL) ?: "(bad d_type)";
@@ -362,12 +364,17 @@ enum fsck_err_opts {
 	  OPT_FS|OPT_MOUNT,						\
 	  OPT_BOOL(),							\
 	  BCH2_NO_SB_OPT,		false,				\
-	  NULL,		"Don't replay the journal")			\
-	x(keep_journal,			u8,				\
+	  NULL,		"Exit recovery immediately prior to journal replay")\
+	x(recovery_pass_last,		u8,				\
+	  OPT_FS|OPT_MOUNT,						\
+	  OPT_STR_NOLIMIT(bch2_recovery_passes),			\
+	  BCH2_NO_SB_OPT,		0,				\
+	  NULL,		"Exit recovery after specified pass")		\
+	x(retain_recovery_info,		u8,				\
 	  0,								\
 	  OPT_BOOL(),							\
 	  BCH2_NO_SB_OPT,		false,				\
-	  NULL,		"Don't free journal entries/keys after startup")\
+	  NULL,		"Don't free journal entries/keys, scanned btree nodes after startup")\
 	x(read_entire_journal,		u8,				\
 	  0,								\
 	  OPT_BOOL(),							\
diff --git a/fs/bcachefs/recovery.c b/fs/bcachefs/recovery.c
index 03f9d6afe467..0f328aba9760 100644
--- a/fs/bcachefs/recovery.c
+++ b/fs/bcachefs/recovery.c
@@ -1,35 +1,31 @@
 // SPDX-License-Identifier: GPL-2.0
 
 #include "bcachefs.h"
-#include "backpointers.h"
-#include "bkey_buf.h"
 #include "alloc_background.h"
-#include "btree_gc.h"
+#include "bkey_buf.h"
 #include "btree_journal_iter.h"
+#include "btree_node_scan.h"
 #include "btree_update.h"
 #include "btree_update_interior.h"
 #include "btree_io.h"
 #include "buckets.h"
 #include "dirent.h"
-#include "ec.h"
 #include "errcode.h"
 #include "error.h"
 #include "fs-common.h"
-#include "fsck.h"
 #include "journal_io.h"
 #include "journal_reclaim.h"
 #include "journal_seq_blacklist.h"
-#include "lru.h"
 #include "logged_ops.h"
 #include "move.h"
 #include "quota.h"
 #include "rebalance.h"
 #include "recovery.h"
+#include "recovery_passes.h"
 #include "replicas.h"
 #include "sb-clean.h"
 #include "sb-downgrade.h"
 #include "snapshot.h"
-#include "subvolume.h"
 #include "super-io.h"
 
 #include <linux/sort.h>
@@ -37,22 +33,22 @@
 
 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
 
-static bool btree_id_is_alloc(enum btree_id id)
+void bch2_btree_lost_data(struct bch_fs *c, enum btree_id btree)
 {
-	switch (id) {
-	case BTREE_ID_alloc:
-	case BTREE_ID_backpointers:
-	case BTREE_ID_need_discard:
-	case BTREE_ID_freespace:
-	case BTREE_ID_bucket_gens:
-		return true;
-	default:
-		return false;
+	u64 b = BIT_ULL(btree);
+
+	if (!(c->sb.btrees_lost_data & b)) {
+		bch_err(c, "flagging btree %s lost data", bch2_btree_id_str(btree));
+
+		mutex_lock(&c->sb_lock);
+		bch2_sb_field_get(c->disk_sb.sb, ext)->btrees_lost_data |= cpu_to_le64(b);
+		bch2_write_super(c);
+		mutex_unlock(&c->sb_lock);
 	}
 }
 
 /* for -o reconstruct_alloc: */
-static void do_reconstruct_alloc(struct bch_fs *c)
+static void bch2_reconstruct_alloc(struct bch_fs *c)
 {
 	bch2_journal_log_msg(c, "dropping alloc info");
 	bch_info(c, "dropping and reconstructing all alloc info");
@@ -87,15 +83,17 @@ static void do_reconstruct_alloc(struct bch_fs *c)
 
 	c->recovery_passes_explicit |= bch2_recovery_passes_from_stable(le64_to_cpu(ext->recovery_passes_required[0]));
 
-	struct journal_keys *keys = &c->journal_keys;
-	size_t src, dst;
 
-	move_gap(keys, keys->nr);
-
-	for (src = 0, dst = 0; src < keys->nr; src++)
-		if (!btree_id_is_alloc(keys->data[src].btree_id))
-			keys->data[dst++] = keys->data[src];
-	keys->nr = keys->gap = dst;
+	bch2_shoot_down_journal_keys(c, BTREE_ID_alloc,
+				     0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
+	bch2_shoot_down_journal_keys(c, BTREE_ID_backpointers,
+				     0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
+	bch2_shoot_down_journal_keys(c, BTREE_ID_need_discard,
+				     0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
+	bch2_shoot_down_journal_keys(c, BTREE_ID_freespace,
+				     0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
+	bch2_shoot_down_journal_keys(c, BTREE_ID_bucket_gens,
+				     0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
 }
 
 /*
@@ -186,7 +184,7 @@ static int journal_sort_seq_cmp(const void *_l, const void *_r)
 	return cmp_int(l->journal_seq, r->journal_seq);
 }
 
-static int bch2_journal_replay(struct bch_fs *c)
+int bch2_journal_replay(struct bch_fs *c)
 {
 	struct journal_keys *keys = &c->journal_keys;
 	DARRAY(struct journal_key *) keys_sorted = { 0 };
@@ -194,6 +192,7 @@ static int bch2_journal_replay(struct bch_fs *c)
 	u64 start_seq	= c->journal_replay_seq_start;
 	u64 end_seq	= c->journal_replay_seq_start;
 	struct btree_trans *trans = bch2_trans_get(c);
+	bool immediate_flush = false;
 	int ret = 0;
 
 	if (keys->nr) {
@@ -215,6 +214,13 @@ static int bch2_journal_replay(struct bch_fs *c)
 	darray_for_each(*keys, k) {
 		cond_resched();
 
+		/*
+		 * k->allocated means the key wasn't read in from the journal,
+		 * rather it was from early repair code
+		 */
+		if (k->allocated)
+			immediate_flush = true;
+
 		/* Skip fastpath if we're low on space in the journal */
 		ret = c->journal.watermark ? -1 :
 			commit_do(trans, NULL, NULL,
@@ -266,7 +272,8 @@ static int bch2_journal_replay(struct bch_fs *c)
 	bch2_trans_put(trans);
 	trans = NULL;
 
-	if (!c->opts.keep_journal)
+	if (!c->opts.retain_recovery_info &&
+	    c->recovery_pass_done >= BCH_RECOVERY_PASS_journal_replay)
 		bch2_journal_keys_put_initial(c);
 
 	replay_now_at(j, j->replay_journal_seq_end);
@@ -274,6 +281,12 @@ static int bch2_journal_replay(struct bch_fs *c)
 
 	bch2_journal_set_replay_done(j);
 
+	/* if we did any repair, flush it immediately */
+	if (immediate_flush) {
+		bch2_journal_flush_all_pins(&c->journal);
+		ret = bch2_journal_meta(&c->journal);
+	}
+
 	if (keys->nr)
 		bch2_journal_log_msg(c, "journal replay finished");
 err:
@@ -423,10 +436,9 @@ static int journal_replay_early(struct bch_fs *c,
 
 static int read_btree_roots(struct bch_fs *c)
 {
-	unsigned i;
 	int ret = 0;
 
-	for (i = 0; i < btree_id_nr_alive(c); i++) {
+	for (unsigned i = 0; i < btree_id_nr_alive(c); i++) {
 		struct btree_root *r = bch2_btree_id_root(c, i);
 
 		if (!r->alive)
@@ -435,186 +447,46 @@ static int read_btree_roots(struct bch_fs *c)
 		if (btree_id_is_alloc(i) && c->opts.reconstruct_alloc)
 			continue;
 
-		if (r->error) {
-			__fsck_err(c,
-				   btree_id_is_alloc(i)
-				   ? FSCK_CAN_IGNORE : 0,
-				   btree_root_bkey_invalid,
-				   "invalid btree root %s",
-				   bch2_btree_id_str(i));
-			if (i == BTREE_ID_alloc)
+		if (mustfix_fsck_err_on((ret = r->error),
+					c, btree_root_bkey_invalid,
+					"invalid btree root %s",
+					bch2_btree_id_str(i)) ||
+		    mustfix_fsck_err_on((ret = r->error = bch2_btree_root_read(c, i, &r->key, r->level)),
+					c, btree_root_read_error,
+					"error reading btree root %s l=%u: %s",
+					bch2_btree_id_str(i), r->level, bch2_err_str(ret))) {
+			if (btree_id_is_alloc(i)) {
+				c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_allocations);
+				c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_alloc_info);
+				c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_lrus);
+				c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_extents_to_backpointers);
+				c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_alloc_to_lru_refs);
 				c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
-		}
+				r->error = 0;
+			} else if (!(c->recovery_passes_explicit & BIT_ULL(BCH_RECOVERY_PASS_scan_for_btree_nodes))) {
+				bch_info(c, "will run btree node scan");
+				c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_scan_for_btree_nodes);
+				c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_topology);
+			}
 
-		ret = bch2_btree_root_read(c, i, &r->key, r->level);
-		if (ret) {
-			fsck_err(c,
-				 btree_root_read_error,
-				 "error reading btree root %s",
-				 bch2_btree_id_str(i));
-			if (btree_id_is_alloc(i))
-				c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 			ret = 0;
+			bch2_btree_lost_data(c, i);
 		}
 	}
 
-	for (i = 0; i < BTREE_ID_NR; i++) {
+	for (unsigned i = 0; i < BTREE_ID_NR; i++) {
 		struct btree_root *r = bch2_btree_id_root(c, i);
 
-		if (!r->b) {
+		if (!r->b && !r->error) {
 			r->alive = false;
 			r->level = 0;
-			bch2_btree_root_alloc(c, i);
+			bch2_btree_root_alloc_fake(c, i, 0);
 		}
 	}
 fsck_err:
 	return ret;
 }
 
-static int bch2_initialize_subvolumes(struct bch_fs *c)
-{
-	struct bkey_i_snapshot_tree	root_tree;
-	struct bkey_i_snapshot		root_snapshot;
-	struct bkey_i_subvolume		root_volume;
-	int ret;
-
-	bkey_snapshot_tree_init(&root_tree.k_i);
-	root_tree.k.p.offset		= 1;
-	root_tree.v.master_subvol	= cpu_to_le32(1);
-	root_tree.v.root_snapshot	= cpu_to_le32(U32_MAX);
-
-	bkey_snapshot_init(&root_snapshot.k_i);
-	root_snapshot.k.p.offset = U32_MAX;
-	root_snapshot.v.flags	= 0;
-	root_snapshot.v.parent	= 0;
-	root_snapshot.v.subvol	= cpu_to_le32(BCACHEFS_ROOT_SUBVOL);
-	root_snapshot.v.tree	= cpu_to_le32(1);
-	SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
-
-	bkey_subvolume_init(&root_volume.k_i);
-	root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
-	root_volume.v.flags	= 0;
-	root_volume.v.snapshot	= cpu_to_le32(U32_MAX);
-	root_volume.v.inode	= cpu_to_le64(BCACHEFS_ROOT_INO);
-
-	ret =   bch2_btree_insert(c, BTREE_ID_snapshot_trees,	&root_tree.k_i, NULL, 0) ?:
-		bch2_btree_insert(c, BTREE_ID_snapshots,	&root_snapshot.k_i, NULL, 0) ?:
-		bch2_btree_insert(c, BTREE_ID_subvolumes,	&root_volume.k_i, NULL, 0);
-	bch_err_fn(c, ret);
-	return ret;
-}
-
-static int __bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
-{
-	struct btree_iter iter;
-	struct bkey_s_c k;
-	struct bch_inode_unpacked inode;
-	int ret;
-
-	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
-			       SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
-	ret = bkey_err(k);
-	if (ret)
-		return ret;
-
-	if (!bkey_is_inode(k.k)) {
-		bch_err(trans->c, "root inode not found");
-		ret = -BCH_ERR_ENOENT_inode;
-		goto err;
-	}
-
-	ret = bch2_inode_unpack(k, &inode);
-	BUG_ON(ret);
-
-	inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
-
-	ret = bch2_inode_write(trans, &iter, &inode);
-err:
-	bch2_trans_iter_exit(trans, &iter);
-	return ret;
-}
-
-/* set bi_subvol on root inode */
-noinline_for_stack
-static int bch2_fs_upgrade_for_subvolumes(struct bch_fs *c)
-{
-	int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_lazy_rw,
-				__bch2_fs_upgrade_for_subvolumes(trans));
-	bch_err_fn(c, ret);
-	return ret;
-}
-
-const char * const bch2_recovery_passes[] = {
-#define x(_fn, ...)	#_fn,
-	BCH_RECOVERY_PASSES()
-#undef x
-	NULL
-};
-
-static int bch2_check_allocations(struct bch_fs *c)
-{
-	return bch2_gc(c, true, c->opts.norecovery);
-}
-
-static int bch2_set_may_go_rw(struct bch_fs *c)
-{
-	struct journal_keys *keys = &c->journal_keys;
-
-	/*
-	 * After we go RW, the journal keys buffer can't be modified (except for
-	 * setting journal_key->overwritten: it will be accessed by multiple
-	 * threads
-	 */
-	move_gap(keys, keys->nr);
-
-	set_bit(BCH_FS_may_go_rw, &c->flags);
-
-	if (keys->nr || c->opts.fsck || !c->sb.clean)
-		return bch2_fs_read_write_early(c);
-	return 0;
-}
-
-struct recovery_pass_fn {
-	int		(*fn)(struct bch_fs *);
-	unsigned	when;
-};
-
-static struct recovery_pass_fn recovery_pass_fns[] = {
-#define x(_fn, _id, _when)	{ .fn = bch2_##_fn, .when = _when },
-	BCH_RECOVERY_PASSES()
-#undef x
-};
-
-u64 bch2_recovery_passes_to_stable(u64 v)
-{
-	static const u8 map[] = {
-#define x(n, id, ...)	[BCH_RECOVERY_PASS_##n] = BCH_RECOVERY_PASS_STABLE_##n,
-	BCH_RECOVERY_PASSES()
-#undef x
-	};
-
-	u64 ret = 0;
-	for (unsigned i = 0; i < ARRAY_SIZE(map); i++)
-		if (v & BIT_ULL(i))
-			ret |= BIT_ULL(map[i]);
-	return ret;
-}
-
-u64 bch2_recovery_passes_from_stable(u64 v)
-{
-	static const u8 map[] = {
-#define x(n, id, ...)	[BCH_RECOVERY_PASS_STABLE_##n] = BCH_RECOVERY_PASS_##n,
-	BCH_RECOVERY_PASSES()
-#undef x
-	};
-
-	u64 ret = 0;
-	for (unsigned i = 0; i < ARRAY_SIZE(map); i++)
-		if (v & BIT_ULL(i))
-			ret |= BIT_ULL(map[i]);
-	return ret;
-}
-
 static bool check_version_upgrade(struct bch_fs *c)
 {
 	unsigned latest_version	= bcachefs_metadata_version_current;
@@ -687,96 +559,6 @@ static bool check_version_upgrade(struct bch_fs *c)
 	return false;
 }
 
-u64 bch2_fsck_recovery_passes(void)
-{
-	u64 ret = 0;
-
-	for (unsigned i = 0; i < ARRAY_SIZE(recovery_pass_fns); i++)
-		if (recovery_pass_fns[i].when & PASS_FSCK)
-			ret |= BIT_ULL(i);
-	return ret;
-}
-
-static bool should_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
-{
-	struct recovery_pass_fn *p = recovery_pass_fns + pass;
-
-	if (c->opts.norecovery && pass > BCH_RECOVERY_PASS_snapshots_read)
-		return false;
-	if (c->recovery_passes_explicit & BIT_ULL(pass))
-		return true;
-	if ((p->when & PASS_FSCK) && c->opts.fsck)
-		return true;
-	if ((p->when & PASS_UNCLEAN) && !c->sb.clean)
-		return true;
-	if (p->when & PASS_ALWAYS)
-		return true;
-	return false;
-}
-
-static int bch2_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
-{
-	struct recovery_pass_fn *p = recovery_pass_fns + pass;
-	int ret;
-
-	if (!(p->when & PASS_SILENT))
-		bch2_print(c, KERN_INFO bch2_log_msg(c, "%s..."),
-			   bch2_recovery_passes[pass]);
-	ret = p->fn(c);
-	if (ret)
-		return ret;
-	if (!(p->when & PASS_SILENT))
-		bch2_print(c, KERN_CONT " done\n");
-
-	return 0;
-}
-
-static int bch2_run_recovery_passes(struct bch_fs *c)
-{
-	int ret = 0;
-
-	while (c->curr_recovery_pass < ARRAY_SIZE(recovery_pass_fns)) {
-		if (should_run_recovery_pass(c, c->curr_recovery_pass)) {
-			unsigned pass = c->curr_recovery_pass;
-
-			ret = bch2_run_recovery_pass(c, c->curr_recovery_pass);
-			if (bch2_err_matches(ret, BCH_ERR_restart_recovery) ||
-			    (ret && c->curr_recovery_pass < pass))
-				continue;
-			if (ret)
-				break;
-
-			c->recovery_passes_complete |= BIT_ULL(c->curr_recovery_pass);
-		}
-		c->curr_recovery_pass++;
-		c->recovery_pass_done = max(c->recovery_pass_done, c->curr_recovery_pass);
-	}
-
-	return ret;
-}
-
-int bch2_run_online_recovery_passes(struct bch_fs *c)
-{
-	int ret = 0;
-
-	for (unsigned i = 0; i < ARRAY_SIZE(recovery_pass_fns); i++) {
-		struct recovery_pass_fn *p = recovery_pass_fns + i;
-
-		if (!(p->when & PASS_ONLINE))
-			continue;
-
-		ret = bch2_run_recovery_pass(c, i);
-		if (bch2_err_matches(ret, BCH_ERR_restart_recovery)) {
-			i = c->curr_recovery_pass;
-			continue;
-		}
-		if (ret)
-			break;
-	}
-
-	return ret;
-}
-
 int bch2_fs_recovery(struct bch_fs *c)
 {
 	struct bch_sb_field_clean *clean = NULL;
@@ -809,24 +591,14 @@ int bch2_fs_recovery(struct bch_fs *c)
 		goto err;
 	}
 
-	if (c->opts.fsck && c->opts.norecovery) {
-		bch_err(c, "cannot select both norecovery and fsck");
-		ret = -EINVAL;
-		goto err;
-	}
+	if (c->opts.norecovery)
+		c->opts.recovery_pass_last = BCH_RECOVERY_PASS_journal_replay - 1;
 
 	if (!c->opts.nochanges) {
 		mutex_lock(&c->sb_lock);
+		struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
 		bool write_sb = false;
 
-		struct bch_sb_field_ext *ext =
-			bch2_sb_field_get_minsize(&c->disk_sb, ext, sizeof(*ext) / sizeof(u64));
-		if (!ext) {
-			ret = -BCH_ERR_ENOSPC_sb;
-			mutex_unlock(&c->sb_lock);
-			goto err;
-		}
-
 		if (BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb)) {
 			ext->recovery_passes_required[0] |=
 				cpu_to_le64(bch2_recovery_passes_to_stable(BIT_ULL(BCH_RECOVERY_PASS_check_topology)));
@@ -885,7 +657,7 @@ int bch2_fs_recovery(struct bch_fs *c)
 		goto err;
 	}
 
-	if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
+	if (!c->sb.clean || c->opts.fsck || c->opts.retain_recovery_info) {
 		struct genradix_iter iter;
 		struct journal_replay **i;
 
@@ -965,7 +737,7 @@ use_clean:
 	c->journal_replay_seq_end	= blacklist_seq - 1;
 
 	if (c->opts.reconstruct_alloc)
-		do_reconstruct_alloc(c);
+		bch2_reconstruct_alloc(c);
 
 	zero_out_btree_mem_ptr(&c->journal_keys);
 
@@ -1017,6 +789,12 @@ use_clean:
 
 	clear_bit(BCH_FS_fsck_running, &c->flags);
 
+	/* fsync if we fixed errors */
+	if (test_bit(BCH_FS_errors_fixed, &c->flags)) {
+		bch2_journal_flush_all_pins(&c->journal);
+		bch2_journal_meta(&c->journal);
+	}
+
 	/* If we fixed errors, verify that fs is actually clean now: */
 	if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) &&
 	    test_bit(BCH_FS_errors_fixed, &c->flags) &&
@@ -1051,6 +829,7 @@ use_clean:
 	}
 
 	mutex_lock(&c->sb_lock);
+	struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
 	bool write_sb = false;
 
 	if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) != le16_to_cpu(c->disk_sb.sb->version)) {
@@ -1064,15 +843,18 @@ use_clean:
 		write_sb = true;
 	}
 
-	if (!test_bit(BCH_FS_error, &c->flags)) {
-		struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
-		if (ext &&
-		    (!bch2_is_zero(ext->recovery_passes_required, sizeof(ext->recovery_passes_required)) ||
-		     !bch2_is_zero(ext->errors_silent, sizeof(ext->errors_silent)))) {
-			memset(ext->recovery_passes_required, 0, sizeof(ext->recovery_passes_required));
-			memset(ext->errors_silent, 0, sizeof(ext->errors_silent));
-			write_sb = true;
-		}
+	if (!test_bit(BCH_FS_error, &c->flags) &&
+	    !bch2_is_zero(ext->errors_silent, sizeof(ext->errors_silent))) {
+		memset(ext->errors_silent, 0, sizeof(ext->errors_silent));
+		write_sb = true;
+	}
+
+	if (c->opts.fsck &&
+	    !test_bit(BCH_FS_error, &c->flags) &&
+	    c->recovery_pass_done == BCH_RECOVERY_PASS_NR - 1 &&
+	    ext->btrees_lost_data) {
+		ext->btrees_lost_data = 0;
+		write_sb = true;
 	}
 
 	if (c->opts.fsck &&
@@ -1113,9 +895,10 @@ use_clean:
 out:
 	bch2_flush_fsck_errs(c);
 
-	if (!c->opts.keep_journal &&
-	    test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags))
+	if (!c->opts.retain_recovery_info) {
 		bch2_journal_keys_put_initial(c);
+		bch2_find_btree_nodes_exit(&c->found_btree_nodes);
+	}
 	kfree(clean);
 
 	if (!ret &&
@@ -1141,6 +924,7 @@ int bch2_fs_initialize(struct bch_fs *c)
 	int ret;
 
 	bch_notice(c, "initializing new filesystem");
+	set_bit(BCH_FS_new_fs, &c->flags);
 
 	mutex_lock(&c->sb_lock);
 	c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
@@ -1155,11 +939,11 @@ int bch2_fs_initialize(struct bch_fs *c)
 	}
 	mutex_unlock(&c->sb_lock);
 
-	c->curr_recovery_pass = ARRAY_SIZE(recovery_pass_fns);
+	c->curr_recovery_pass = BCH_RECOVERY_PASS_NR;
 	set_bit(BCH_FS_may_go_rw, &c->flags);
 
 	for (unsigned i = 0; i < BTREE_ID_NR; i++)
-		bch2_btree_root_alloc(c, i);
+		bch2_btree_root_alloc_fake(c, i, 0);
 
 	for_each_member_device(c, ca)
 		bch2_dev_usage_init(ca);
@@ -1230,7 +1014,7 @@ int bch2_fs_initialize(struct bch_fs *c)
 	if (ret)
 		goto err;
 
-	c->recovery_pass_done = ARRAY_SIZE(recovery_pass_fns) - 1;
+	c->recovery_pass_done = BCH_RECOVERY_PASS_NR - 1;
 
 	if (enabled_qtypes(c)) {
 		ret = bch2_fs_quota_read(c);
diff --git a/fs/bcachefs/recovery.h b/fs/bcachefs/recovery.h
index 4e9d24719b2e..4bf818de1f2f 100644
--- a/fs/bcachefs/recovery.h
+++ b/fs/bcachefs/recovery.h
@@ -2,37 +2,9 @@
 #ifndef _BCACHEFS_RECOVERY_H
 #define _BCACHEFS_RECOVERY_H
 
-extern const char * const bch2_recovery_passes[];
+void bch2_btree_lost_data(struct bch_fs *, enum btree_id);
 
-u64 bch2_recovery_passes_to_stable(u64 v);
-u64 bch2_recovery_passes_from_stable(u64 v);
-
-/*
- * For when we need to rewind recovery passes and run a pass we skipped:
- */
-static inline int bch2_run_explicit_recovery_pass(struct bch_fs *c,
-						  enum bch_recovery_pass pass)
-{
-	if (c->recovery_passes_explicit & BIT_ULL(pass))
-		return 0;
-
-	bch_info(c, "running explicit recovery pass %s (%u), currently at %s (%u)",
-		 bch2_recovery_passes[pass], pass,
-		 bch2_recovery_passes[c->curr_recovery_pass], c->curr_recovery_pass);
-
-	c->recovery_passes_explicit |= BIT_ULL(pass);
-
-	if (c->curr_recovery_pass >= pass) {
-		c->curr_recovery_pass = pass;
-		c->recovery_passes_complete &= (1ULL << pass) >> 1;
-		return -BCH_ERR_restart_recovery;
-	} else {
-		return 0;
-	}
-}
-
-int bch2_run_online_recovery_passes(struct bch_fs *);
-u64 bch2_fsck_recovery_passes(void);
+int bch2_journal_replay(struct bch_fs *);
 
 int bch2_fs_recovery(struct bch_fs *);
 int bch2_fs_initialize(struct bch_fs *);
diff --git a/fs/bcachefs/recovery_passes.c b/fs/bcachefs/recovery_passes.c
new file mode 100644
index 000000000000..0cec0f7d9703
--- /dev/null
+++ b/fs/bcachefs/recovery_passes.c
@@ -0,0 +1,249 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include "bcachefs.h"
+#include "alloc_background.h"
+#include "backpointers.h"
+#include "btree_gc.h"
+#include "btree_node_scan.h"
+#include "ec.h"
+#include "fsck.h"
+#include "inode.h"
+#include "journal.h"
+#include "lru.h"
+#include "logged_ops.h"
+#include "rebalance.h"
+#include "recovery.h"
+#include "recovery_passes.h"
+#include "snapshot.h"
+#include "subvolume.h"
+#include "super.h"
+#include "super-io.h"
+
+const char * const bch2_recovery_passes[] = {
+#define x(_fn, ...)	#_fn,
+	BCH_RECOVERY_PASSES()
+#undef x
+	NULL
+};
+
+static int bch2_check_allocations(struct bch_fs *c)
+{
+	return bch2_gc(c, true, false);
+}
+
+static int bch2_set_may_go_rw(struct bch_fs *c)
+{
+	struct journal_keys *keys = &c->journal_keys;
+
+	/*
+	 * After we go RW, the journal keys buffer can't be modified (except for
+	 * setting journal_key->overwritten: it will be accessed by multiple
+	 * threads
+	 */
+	move_gap(keys, keys->nr);
+
+	set_bit(BCH_FS_may_go_rw, &c->flags);
+
+	if (keys->nr || c->opts.fsck || !c->sb.clean || c->recovery_passes_explicit)
+		return bch2_fs_read_write_early(c);
+	return 0;
+}
+
+struct recovery_pass_fn {
+	int		(*fn)(struct bch_fs *);
+	unsigned	when;
+};
+
+static struct recovery_pass_fn recovery_pass_fns[] = {
+#define x(_fn, _id, _when)	{ .fn = bch2_##_fn, .when = _when },
+	BCH_RECOVERY_PASSES()
+#undef x
+};
+
+static const u8 passes_to_stable_map[] = {
+#define x(n, id, ...)	[BCH_RECOVERY_PASS_##n] = BCH_RECOVERY_PASS_STABLE_##n,
+	BCH_RECOVERY_PASSES()
+#undef x
+};
+
+static enum bch_recovery_pass_stable bch2_recovery_pass_to_stable(enum bch_recovery_pass pass)
+{
+	return passes_to_stable_map[pass];
+}
+
+u64 bch2_recovery_passes_to_stable(u64 v)
+{
+	u64 ret = 0;
+	for (unsigned i = 0; i < ARRAY_SIZE(passes_to_stable_map); i++)
+		if (v & BIT_ULL(i))
+			ret |= BIT_ULL(passes_to_stable_map[i]);
+	return ret;
+}
+
+u64 bch2_recovery_passes_from_stable(u64 v)
+{
+	static const u8 map[] = {
+#define x(n, id, ...)	[BCH_RECOVERY_PASS_STABLE_##n] = BCH_RECOVERY_PASS_##n,
+	BCH_RECOVERY_PASSES()
+#undef x
+	};
+
+	u64 ret = 0;
+	for (unsigned i = 0; i < ARRAY_SIZE(map); i++)
+		if (v & BIT_ULL(i))
+			ret |= BIT_ULL(map[i]);
+	return ret;
+}
+
+/*
+ * For when we need to rewind recovery passes and run a pass we skipped:
+ */
+int bch2_run_explicit_recovery_pass(struct bch_fs *c,
+				    enum bch_recovery_pass pass)
+{
+	if (c->recovery_passes_explicit & BIT_ULL(pass))
+		return 0;
+
+	bch_info(c, "running explicit recovery pass %s (%u), currently at %s (%u)",
+		 bch2_recovery_passes[pass], pass,
+		 bch2_recovery_passes[c->curr_recovery_pass], c->curr_recovery_pass);
+
+	c->recovery_passes_explicit |= BIT_ULL(pass);
+
+	if (c->curr_recovery_pass >= pass) {
+		c->curr_recovery_pass = pass;
+		c->recovery_passes_complete &= (1ULL << pass) >> 1;
+		return -BCH_ERR_restart_recovery;
+	} else {
+		return 0;
+	}
+}
+
+int bch2_run_explicit_recovery_pass_persistent(struct bch_fs *c,
+					       enum bch_recovery_pass pass)
+{
+	enum bch_recovery_pass_stable s = bch2_recovery_pass_to_stable(pass);
+
+	mutex_lock(&c->sb_lock);
+	struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
+
+	if (!test_bit_le64(s, ext->recovery_passes_required)) {
+		__set_bit_le64(s, ext->recovery_passes_required);
+		bch2_write_super(c);
+	}
+	mutex_unlock(&c->sb_lock);
+
+	return bch2_run_explicit_recovery_pass(c, pass);
+}
+
+static void bch2_clear_recovery_pass_required(struct bch_fs *c,
+					      enum bch_recovery_pass pass)
+{
+	enum bch_recovery_pass_stable s = bch2_recovery_pass_to_stable(pass);
+
+	mutex_lock(&c->sb_lock);
+	struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
+
+	if (test_bit_le64(s, ext->recovery_passes_required)) {
+		__clear_bit_le64(s, ext->recovery_passes_required);
+		bch2_write_super(c);
+	}
+	mutex_unlock(&c->sb_lock);
+}
+
+u64 bch2_fsck_recovery_passes(void)
+{
+	u64 ret = 0;
+
+	for (unsigned i = 0; i < ARRAY_SIZE(recovery_pass_fns); i++)
+		if (recovery_pass_fns[i].when & PASS_FSCK)
+			ret |= BIT_ULL(i);
+	return ret;
+}
+
+static bool should_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
+{
+	struct recovery_pass_fn *p = recovery_pass_fns + pass;
+
+	if (c->recovery_passes_explicit & BIT_ULL(pass))
+		return true;
+	if ((p->when & PASS_FSCK) && c->opts.fsck)
+		return true;
+	if ((p->when & PASS_UNCLEAN) && !c->sb.clean)
+		return true;
+	if (p->when & PASS_ALWAYS)
+		return true;
+	return false;
+}
+
+static int bch2_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
+{
+	struct recovery_pass_fn *p = recovery_pass_fns + pass;
+	int ret;
+
+	if (!(p->when & PASS_SILENT))
+		bch2_print(c, KERN_INFO bch2_log_msg(c, "%s..."),
+			   bch2_recovery_passes[pass]);
+	ret = p->fn(c);
+	if (ret)
+		return ret;
+	if (!(p->when & PASS_SILENT))
+		bch2_print(c, KERN_CONT " done\n");
+
+	return 0;
+}
+
+int bch2_run_online_recovery_passes(struct bch_fs *c)
+{
+	int ret = 0;
+
+	for (unsigned i = 0; i < ARRAY_SIZE(recovery_pass_fns); i++) {
+		struct recovery_pass_fn *p = recovery_pass_fns + i;
+
+		if (!(p->when & PASS_ONLINE))
+			continue;
+
+		ret = bch2_run_recovery_pass(c, i);
+		if (bch2_err_matches(ret, BCH_ERR_restart_recovery)) {
+			i = c->curr_recovery_pass;
+			continue;
+		}
+		if (ret)
+			break;
+	}
+
+	return ret;
+}
+
+int bch2_run_recovery_passes(struct bch_fs *c)
+{
+	int ret = 0;
+
+	while (c->curr_recovery_pass < ARRAY_SIZE(recovery_pass_fns)) {
+		if (c->opts.recovery_pass_last &&
+		    c->curr_recovery_pass > c->opts.recovery_pass_last)
+			break;
+
+		if (should_run_recovery_pass(c, c->curr_recovery_pass)) {
+			unsigned pass = c->curr_recovery_pass;
+
+			ret = bch2_run_recovery_pass(c, c->curr_recovery_pass);
+			if (bch2_err_matches(ret, BCH_ERR_restart_recovery) ||
+			    (ret && c->curr_recovery_pass < pass))
+				continue;
+			if (ret)
+				break;
+
+			c->recovery_passes_complete |= BIT_ULL(c->curr_recovery_pass);
+		}
+
+		c->recovery_pass_done = max(c->recovery_pass_done, c->curr_recovery_pass);
+
+		if (!test_bit(BCH_FS_error, &c->flags))
+			bch2_clear_recovery_pass_required(c, c->curr_recovery_pass);
+
+		c->curr_recovery_pass++;
+	}
+
+	return ret;
+}
diff --git a/fs/bcachefs/recovery_passes.h b/fs/bcachefs/recovery_passes.h
new file mode 100644
index 000000000000..99b464e127b8
--- /dev/null
+++ b/fs/bcachefs/recovery_passes.h
@@ -0,0 +1,17 @@
+#ifndef _BCACHEFS_RECOVERY_PASSES_H
+#define _BCACHEFS_RECOVERY_PASSES_H
+
+extern const char * const bch2_recovery_passes[];
+
+u64 bch2_recovery_passes_to_stable(u64 v);
+u64 bch2_recovery_passes_from_stable(u64 v);
+
+u64 bch2_fsck_recovery_passes(void);
+
+int bch2_run_explicit_recovery_pass(struct bch_fs *, enum bch_recovery_pass);
+int bch2_run_explicit_recovery_pass_persistent(struct bch_fs *, enum bch_recovery_pass);
+
+int bch2_run_online_recovery_passes(struct bch_fs *);
+int bch2_run_recovery_passes(struct bch_fs *);
+
+#endif /* _BCACHEFS_RECOVERY_PASSES_H */
diff --git a/fs/bcachefs/recovery_types.h b/fs/bcachefs/recovery_passes_types.h
index 4959e95e7c74..773aea9a0080 100644
--- a/fs/bcachefs/recovery_types.h
+++ b/fs/bcachefs/recovery_passes_types.h
@@ -1,6 +1,6 @@
 /* SPDX-License-Identifier: GPL-2.0 */
-#ifndef _BCACHEFS_RECOVERY_TYPES_H
-#define _BCACHEFS_RECOVERY_TYPES_H
+#ifndef _BCACHEFS_RECOVERY_PASSES_TYPES_H
+#define _BCACHEFS_RECOVERY_PASSES_TYPES_H
 
 #define PASS_SILENT		BIT(0)
 #define PASS_FSCK		BIT(1)
@@ -13,6 +13,7 @@
  * must never change:
  */
 #define BCH_RECOVERY_PASSES()							\
+	x(scan_for_btree_nodes,			37, 0)				\
 	x(check_topology,			 4, 0)				\
 	x(alloc_read,				 0, PASS_ALWAYS)		\
 	x(stripes_read,				 1, PASS_ALWAYS)		\
@@ -31,13 +32,13 @@
 	x(check_alloc_to_lru_refs,		15, PASS_ONLINE|PASS_FSCK)	\
 	x(fs_freespace_init,			16, PASS_ALWAYS|PASS_SILENT)	\
 	x(bucket_gens_init,			17, 0)				\
+	x(reconstruct_snapshots,		38, 0)				\
 	x(check_snapshot_trees,			18, PASS_ONLINE|PASS_FSCK)	\
 	x(check_snapshots,			19, PASS_ONLINE|PASS_FSCK)	\
 	x(check_subvols,			20, PASS_ONLINE|PASS_FSCK)	\
 	x(check_subvol_children,		35, PASS_ONLINE|PASS_FSCK)	\
 	x(delete_dead_snapshots,		21, PASS_ONLINE|PASS_FSCK)	\
 	x(fs_upgrade_for_subvolumes,		22, 0)				\
-	x(resume_logged_ops,			23, PASS_ALWAYS)		\
 	x(check_inodes,				24, PASS_FSCK)			\
 	x(check_extents,			25, PASS_FSCK)			\
 	x(check_indirect_extents,		26, PASS_FSCK)			\
@@ -47,6 +48,7 @@
 	x(check_subvolume_structure,		36, PASS_ONLINE|PASS_FSCK)	\
 	x(check_directory_structure,		30, PASS_ONLINE|PASS_FSCK)	\
 	x(check_nlinks,				31, PASS_FSCK)			\
+	x(resume_logged_ops,			23, PASS_ALWAYS)		\
 	x(delete_dead_inodes,			32, PASS_FSCK|PASS_UNCLEAN)	\
 	x(fix_reflink_p,			33, 0)				\
 	x(set_fs_needs_rebalance,		34, 0)				\
@@ -56,6 +58,7 @@ enum bch_recovery_pass {
 #define x(n, id, when)	BCH_RECOVERY_PASS_##n,
 	BCH_RECOVERY_PASSES()
 #undef x
+	BCH_RECOVERY_PASS_NR
 };
 
 /* But we also need stable identifiers that can be used in the superblock */
@@ -65,4 +68,4 @@ enum bch_recovery_pass_stable {
 #undef x
 };
 
-#endif /* _BCACHEFS_RECOVERY_TYPES_H */
+#endif /* _BCACHEFS_RECOVERY_PASSES_TYPES_H */
diff --git a/fs/bcachefs/reflink.c b/fs/bcachefs/reflink.c
index c47c66c2b394..ff7864731a07 100644
--- a/fs/bcachefs/reflink.c
+++ b/fs/bcachefs/reflink.c
@@ -185,8 +185,7 @@ not_found:
 		} else {
 			bkey_error_init(update);
 			update->k.p		= p.k->p;
-			update->k.p.offset	= next_idx;
-			update->k.size		= next_idx - *idx;
+			update->k.size		= p.k->size;
 			set_bkey_val_u64s(&update->k, 0);
 		}
 
diff --git a/fs/bcachefs/replicas.c b/fs/bcachefs/replicas.c
index cc2672c12031..678b9c20e251 100644
--- a/fs/bcachefs/replicas.c
+++ b/fs/bcachefs/replicas.c
@@ -6,12 +6,15 @@
 #include "replicas.h"
 #include "super-io.h"
 
+#include <linux/sort.h>
+
 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *,
 					    struct bch_replicas_cpu *);
 
 /* Some (buggy!) compilers don't allow memcmp to be passed as a pointer */
-static int bch2_memcmp(const void *l, const void *r, size_t size)
+static int bch2_memcmp(const void *l, const void *r,  const void *priv)
 {
+	size_t size = (size_t) priv;
 	return memcmp(l, r, size);
 }
 
@@ -39,7 +42,8 @@ void bch2_replicas_entry_sort(struct bch_replicas_entry_v1 *e)
 
 static void bch2_cpu_replicas_sort(struct bch_replicas_cpu *r)
 {
-	eytzinger0_sort(r->entries, r->nr, r->entry_size, bch2_memcmp, NULL);
+	eytzinger0_sort_r(r->entries, r->nr, r->entry_size,
+			  bch2_memcmp, NULL, (void *)(size_t)r->entry_size);
 }
 
 static void bch2_replicas_entry_v0_to_text(struct printbuf *out,
@@ -228,7 +232,7 @@ static inline int __replicas_entry_idx(struct bch_replicas_cpu *r,
 
 	verify_replicas_entry(search);
 
-#define entry_cmp(_l, _r, size)	memcmp(_l, _r, entry_size)
+#define entry_cmp(_l, _r)	memcmp(_l, _r, entry_size)
 	idx = eytzinger0_find(r->entries, r->nr, r->entry_size,
 			      entry_cmp, search);
 #undef entry_cmp
@@ -824,10 +828,11 @@ static int bch2_cpu_replicas_validate(struct bch_replicas_cpu *cpu_r,
 {
 	unsigned i;
 
-	sort_cmp_size(cpu_r->entries,
-		      cpu_r->nr,
-		      cpu_r->entry_size,
-		      bch2_memcmp, NULL);
+	sort_r(cpu_r->entries,
+	       cpu_r->nr,
+	       cpu_r->entry_size,
+	       bch2_memcmp, NULL,
+	       (void *)(size_t)cpu_r->entry_size);
 
 	for (i = 0; i < cpu_r->nr; i++) {
 		struct bch_replicas_entry_v1 *e =
diff --git a/fs/bcachefs/sb-downgrade.c b/fs/bcachefs/sb-downgrade.c
index e4396cb0bacb..a98ef940b7a3 100644
--- a/fs/bcachefs/sb-downgrade.c
+++ b/fs/bcachefs/sb-downgrade.c
@@ -7,7 +7,7 @@
 
 #include "bcachefs.h"
 #include "darray.h"
-#include "recovery.h"
+#include "recovery_passes.h"
 #include "sb-downgrade.h"
 #include "sb-errors.h"
 #include "super-io.h"
@@ -51,7 +51,10 @@
 	  BCH_FSCK_ERR_subvol_fs_path_parent_wrong)		\
 	x(btree_subvolume_children,				\
 	  BIT_ULL(BCH_RECOVERY_PASS_check_subvols),		\
-	  BCH_FSCK_ERR_subvol_children_not_set)
+	  BCH_FSCK_ERR_subvol_children_not_set)			\
+	x(mi_btree_bitmap,					\
+	  BIT_ULL(BCH_RECOVERY_PASS_check_allocations),		\
+	  BCH_FSCK_ERR_btree_bitmap_not_marked)
 
 #define DOWNGRADE_TABLE()
 
diff --git a/fs/bcachefs/sb-errors_types.h b/fs/bcachefs/sb-errors_types.h
index 5178bf579f7c..4ca6e7b0d8aa 100644
--- a/fs/bcachefs/sb-errors_types.h
+++ b/fs/bcachefs/sb-errors_types.h
@@ -130,7 +130,7 @@
 	x(bucket_gens_nonzero_for_invalid_buckets,		122)	\
 	x(need_discard_freespace_key_to_invalid_dev_bucket,	123)	\
 	x(need_discard_freespace_key_bad,			124)	\
-	x(backpointer_pos_wrong,				125)	\
+	x(backpointer_bucket_offset_wrong,			125)	\
 	x(backpointer_to_missing_device,			126)	\
 	x(backpointer_to_missing_alloc,				127)	\
 	x(backpointer_to_missing_ptr,				128)	\
@@ -265,7 +265,13 @@
 	x(subvol_children_bad,					257)	\
 	x(subvol_loop,						258)	\
 	x(subvol_unreachable,					259)	\
-	x(btree_node_bkey_bad_u64s,				260)
+	x(btree_node_bkey_bad_u64s,				260)	\
+	x(btree_node_topology_empty_interior_node,		261)	\
+	x(btree_ptr_v2_min_key_bad,				262)	\
+	x(btree_root_unreadable_and_scan_found_nothing,		263)	\
+	x(snapshot_node_missing,				264)	\
+	x(dup_backpointer_to_bad_csum_extent,			265)	\
+	x(btree_bitmap_not_marked,				266)
 
 enum bch_sb_error_id {
 #define x(t, n) BCH_FSCK_ERR_##t = n,
diff --git a/fs/bcachefs/sb-members.c b/fs/bcachefs/sb-members.c
index eff5ce18c69c..522a969345e5 100644
--- a/fs/bcachefs/sb-members.c
+++ b/fs/bcachefs/sb-members.c
@@ -1,6 +1,7 @@
 // SPDX-License-Identifier: GPL-2.0
 
 #include "bcachefs.h"
+#include "btree_cache.h"
 #include "disk_groups.h"
 #include "opts.h"
 #include "replicas.h"
@@ -426,3 +427,55 @@ void bch2_dev_errors_reset(struct bch_dev *ca)
 	bch2_write_super(c);
 	mutex_unlock(&c->sb_lock);
 }
+
+/*
+ * Per member "range has btree nodes" bitmap:
+ *
+ * This is so that if we ever have to run the btree node scan to repair we don't
+ * have to scan full devices:
+ */
+
+bool bch2_dev_btree_bitmap_marked(struct bch_fs *c, struct bkey_s_c k)
+{
+	bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr)
+		if (!bch2_dev_btree_bitmap_marked_sectors(bch_dev_bkey_exists(c, ptr->dev),
+							  ptr->offset, btree_sectors(c)))
+			return false;
+	return true;
+}
+
+static void __bch2_dev_btree_bitmap_mark(struct bch_sb_field_members_v2 *mi, unsigned dev,
+				u64 start, unsigned sectors)
+{
+	struct bch_member *m = __bch2_members_v2_get_mut(mi, dev);
+	u64 bitmap = le64_to_cpu(m->btree_allocated_bitmap);
+
+	u64 end = start + sectors;
+
+	int resize = ilog2(roundup_pow_of_two(end)) - (m->btree_bitmap_shift + 6);
+	if (resize > 0) {
+		u64 new_bitmap = 0;
+
+		for (unsigned i = 0; i < 64; i++)
+			if (bitmap & BIT_ULL(i))
+				new_bitmap |= BIT_ULL(i >> resize);
+		bitmap = new_bitmap;
+		m->btree_bitmap_shift += resize;
+	}
+
+	for (unsigned bit = sectors >> m->btree_bitmap_shift;
+	     bit << m->btree_bitmap_shift < end;
+	     bit++)
+		bitmap |= BIT_ULL(bit);
+
+	m->btree_allocated_bitmap = cpu_to_le64(bitmap);
+}
+
+void bch2_dev_btree_bitmap_mark(struct bch_fs *c, struct bkey_s_c k)
+{
+	lockdep_assert_held(&c->sb_lock);
+
+	struct bch_sb_field_members_v2 *mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
+	bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr)
+		__bch2_dev_btree_bitmap_mark(mi, ptr->dev, ptr->offset, btree_sectors(c));
+}
diff --git a/fs/bcachefs/sb-members.h b/fs/bcachefs/sb-members.h
index be0a94183271..b27c3e4467cf 100644
--- a/fs/bcachefs/sb-members.h
+++ b/fs/bcachefs/sb-members.h
@@ -3,6 +3,7 @@
 #define _BCACHEFS_SB_MEMBERS_H
 
 #include "darray.h"
+#include "bkey_types.h"
 
 extern char * const bch2_member_error_strs[];
 
@@ -220,6 +221,8 @@ static inline struct bch_member_cpu bch2_mi_to_cpu(struct bch_member *mi)
 			: 1,
 		.freespace_initialized = BCH_MEMBER_FREESPACE_INITIALIZED(mi),
 		.valid		= bch2_member_exists(mi),
+		.btree_bitmap_shift	= mi->btree_bitmap_shift,
+		.btree_allocated_bitmap = le64_to_cpu(mi->btree_allocated_bitmap),
 	};
 }
 
@@ -228,4 +231,22 @@ void bch2_sb_members_from_cpu(struct bch_fs *);
 void bch2_dev_io_errors_to_text(struct printbuf *, struct bch_dev *);
 void bch2_dev_errors_reset(struct bch_dev *);
 
+static inline bool bch2_dev_btree_bitmap_marked_sectors(struct bch_dev *ca, u64 start, unsigned sectors)
+{
+	u64 end = start + sectors;
+
+	if (end > 64 << ca->mi.btree_bitmap_shift)
+		return false;
+
+	for (unsigned bit = sectors >> ca->mi.btree_bitmap_shift;
+	     bit << ca->mi.btree_bitmap_shift < end;
+	     bit++)
+		if (!(ca->mi.btree_allocated_bitmap & BIT_ULL(bit)))
+			return false;
+	return true;
+}
+
+bool bch2_dev_btree_bitmap_marked(struct bch_fs *, struct bkey_s_c);
+void bch2_dev_btree_bitmap_mark(struct bch_fs *, struct bkey_s_c);
+
 #endif /* _BCACHEFS_SB_MEMBERS_H */
diff --git a/fs/bcachefs/snapshot.c b/fs/bcachefs/snapshot.c
index 39debe814bf3..544322d5c251 100644
--- a/fs/bcachefs/snapshot.c
+++ b/fs/bcachefs/snapshot.c
@@ -8,6 +8,7 @@
 #include "errcode.h"
 #include "error.h"
 #include "fs.h"
+#include "recovery_passes.h"
 #include "snapshot.h"
 
 #include <linux/random.h>
@@ -93,8 +94,10 @@ static int bch2_snapshot_tree_create(struct btree_trans *trans,
 
 static bool __bch2_snapshot_is_ancestor_early(struct snapshot_table *t, u32 id, u32 ancestor)
 {
-	while (id && id < ancestor)
-		id = __snapshot_t(t, id)->parent;
+	while (id && id < ancestor) {
+		const struct snapshot_t *s = __snapshot_t(t, id);
+		id = s ? s->parent : 0;
+	}
 	return id == ancestor;
 }
 
@@ -110,6 +113,8 @@ static bool bch2_snapshot_is_ancestor_early(struct bch_fs *c, u32 id, u32 ancest
 static inline u32 get_ancestor_below(struct snapshot_table *t, u32 id, u32 ancestor)
 {
 	const struct snapshot_t *s = __snapshot_t(t, id);
+	if (!s)
+		return 0;
 
 	if (s->skip[2] <= ancestor)
 		return s->skip[2];
@@ -120,6 +125,15 @@ static inline u32 get_ancestor_below(struct snapshot_table *t, u32 id, u32 ances
 	return s->parent;
 }
 
+static bool test_ancestor_bitmap(struct snapshot_table *t, u32 id, u32 ancestor)
+{
+	const struct snapshot_t *s = __snapshot_t(t, id);
+	if (!s)
+		return false;
+
+	return test_bit(ancestor - id - 1, s->is_ancestor);
+}
+
 bool __bch2_snapshot_is_ancestor(struct bch_fs *c, u32 id, u32 ancestor)
 {
 	bool ret;
@@ -127,7 +141,7 @@ bool __bch2_snapshot_is_ancestor(struct bch_fs *c, u32 id, u32 ancestor)
 	rcu_read_lock();
 	struct snapshot_table *t = rcu_dereference(c->snapshots);
 
-	if (unlikely(c->recovery_pass_done <= BCH_RECOVERY_PASS_check_snapshots)) {
+	if (unlikely(c->recovery_pass_done < BCH_RECOVERY_PASS_check_snapshots)) {
 		ret = __bch2_snapshot_is_ancestor_early(t, id, ancestor);
 		goto out;
 	}
@@ -135,13 +149,11 @@ bool __bch2_snapshot_is_ancestor(struct bch_fs *c, u32 id, u32 ancestor)
 	while (id && id < ancestor - IS_ANCESTOR_BITMAP)
 		id = get_ancestor_below(t, id, ancestor);
 
-	if (id && id < ancestor) {
-		ret = test_bit(ancestor - id - 1, __snapshot_t(t, id)->is_ancestor);
+	ret = id && id < ancestor
+		? test_ancestor_bitmap(t, id, ancestor)
+		: id == ancestor;
 
-		EBUG_ON(ret != __bch2_snapshot_is_ancestor_early(t, id, ancestor));
-	} else {
-		ret = id == ancestor;
-	}
+	EBUG_ON(ret != __bch2_snapshot_is_ancestor_early(t, id, ancestor));
 out:
 	rcu_read_unlock();
 
@@ -151,36 +163,39 @@ out:
 static noinline struct snapshot_t *__snapshot_t_mut(struct bch_fs *c, u32 id)
 {
 	size_t idx = U32_MAX - id;
-	size_t new_size;
 	struct snapshot_table *new, *old;
 
-	new_size = max(16UL, roundup_pow_of_two(idx + 1));
+	size_t new_bytes = kmalloc_size_roundup(struct_size(new, s, idx + 1));
+	size_t new_size = (new_bytes - sizeof(*new)) / sizeof(new->s[0]);
 
-	new = kvzalloc(struct_size(new, s, new_size), GFP_KERNEL);
+	new = kvzalloc(new_bytes, GFP_KERNEL);
 	if (!new)
 		return NULL;
 
+	new->nr = new_size;
+
 	old = rcu_dereference_protected(c->snapshots, true);
 	if (old)
-		memcpy(new->s,
-		       rcu_dereference_protected(c->snapshots, true)->s,
-		       sizeof(new->s[0]) * c->snapshot_table_size);
+		memcpy(new->s, old->s, sizeof(old->s[0]) * old->nr);
 
 	rcu_assign_pointer(c->snapshots, new);
-	c->snapshot_table_size = new_size;
-	kvfree_rcu_mightsleep(old);
+	kvfree_rcu(old, rcu);
 
-	return &rcu_dereference_protected(c->snapshots, true)->s[idx];
+	return &rcu_dereference_protected(c->snapshots,
+				lockdep_is_held(&c->snapshot_table_lock))->s[idx];
 }
 
 static inline struct snapshot_t *snapshot_t_mut(struct bch_fs *c, u32 id)
 {
 	size_t idx = U32_MAX - id;
+	struct snapshot_table *table =
+		rcu_dereference_protected(c->snapshots,
+				lockdep_is_held(&c->snapshot_table_lock));
 
 	lockdep_assert_held(&c->snapshot_table_lock);
 
-	if (likely(idx < c->snapshot_table_size))
-		return &rcu_dereference_protected(c->snapshots, true)->s[idx];
+	if (likely(table && idx < table->nr))
+		return &table->s[idx];
 
 	return __snapshot_t_mut(c, id);
 }
@@ -567,6 +582,13 @@ static int check_snapshot_tree(struct btree_trans *trans,
 		u32 subvol_id;
 
 		ret = bch2_snapshot_tree_master_subvol(trans, root_id, &subvol_id);
+		bch_err_fn(c, ret);
+
+		if (bch2_err_matches(ret, ENOENT)) { /* nothing to be done here */
+			ret = 0;
+			goto err;
+		}
+
 		if (ret)
 			goto err;
 
@@ -724,7 +746,6 @@ static int check_snapshot(struct btree_trans *trans,
 	u32 parent_id = bch2_snapshot_parent_early(c, k.k->p.offset);
 	u32 real_depth;
 	struct printbuf buf = PRINTBUF;
-	bool should_have_subvol;
 	u32 i, id;
 	int ret = 0;
 
@@ -770,7 +791,7 @@ static int check_snapshot(struct btree_trans *trans,
 		}
 	}
 
-	should_have_subvol = BCH_SNAPSHOT_SUBVOL(&s) &&
+	bool should_have_subvol = BCH_SNAPSHOT_SUBVOL(&s) &&
 		!BCH_SNAPSHOT_DELETED(&s);
 
 	if (should_have_subvol) {
@@ -872,6 +893,154 @@ int bch2_check_snapshots(struct bch_fs *c)
 	return ret;
 }
 
+static int check_snapshot_exists(struct btree_trans *trans, u32 id)
+{
+	struct bch_fs *c = trans->c;
+
+	if (bch2_snapshot_equiv(c, id))
+		return 0;
+
+	u32 tree_id;
+	int ret = bch2_snapshot_tree_create(trans, id, 0, &tree_id);
+	if (ret)
+		return ret;
+
+	struct bkey_i_snapshot *snapshot = bch2_trans_kmalloc(trans, sizeof(*snapshot));
+	ret = PTR_ERR_OR_ZERO(snapshot);
+	if (ret)
+		return ret;
+
+	bkey_snapshot_init(&snapshot->k_i);
+	snapshot->k.p		= POS(0, id);
+	snapshot->v.tree	= cpu_to_le32(tree_id);
+	snapshot->v.btime.lo	= cpu_to_le64(bch2_current_time(c));
+
+	return  bch2_btree_insert_trans(trans, BTREE_ID_snapshots, &snapshot->k_i, 0) ?:
+		bch2_mark_snapshot(trans, BTREE_ID_snapshots, 0,
+				   bkey_s_c_null, bkey_i_to_s(&snapshot->k_i), 0) ?:
+		bch2_snapshot_set_equiv(trans, bkey_i_to_s_c(&snapshot->k_i));
+}
+
+/* Figure out which snapshot nodes belong in the same tree: */
+struct snapshot_tree_reconstruct {
+	enum btree_id			btree;
+	struct bpos			cur_pos;
+	snapshot_id_list		cur_ids;
+	DARRAY(snapshot_id_list)	trees;
+};
+
+static void snapshot_tree_reconstruct_exit(struct snapshot_tree_reconstruct *r)
+{
+	darray_for_each(r->trees, i)
+		darray_exit(i);
+	darray_exit(&r->trees);
+	darray_exit(&r->cur_ids);
+}
+
+static inline bool same_snapshot(struct snapshot_tree_reconstruct *r, struct bpos pos)
+{
+	return r->btree == BTREE_ID_inodes
+		? r->cur_pos.offset == pos.offset
+		: r->cur_pos.inode == pos.inode;
+}
+
+static inline bool snapshot_id_lists_have_common(snapshot_id_list *l, snapshot_id_list *r)
+{
+	darray_for_each(*l, i)
+		if (snapshot_list_has_id(r, *i))
+			return true;
+	return false;
+}
+
+static void snapshot_id_list_to_text(struct printbuf *out, snapshot_id_list *s)
+{
+	bool first = true;
+	darray_for_each(*s, i) {
+		if (!first)
+			prt_char(out, ' ');
+		first = false;
+		prt_printf(out, "%u", *i);
+	}
+}
+
+static int snapshot_tree_reconstruct_next(struct bch_fs *c, struct snapshot_tree_reconstruct *r)
+{
+	if (r->cur_ids.nr) {
+		darray_for_each(r->trees, i)
+			if (snapshot_id_lists_have_common(i, &r->cur_ids)) {
+				int ret = snapshot_list_merge(c, i, &r->cur_ids);
+				if (ret)
+					return ret;
+				goto out;
+			}
+		darray_push(&r->trees, r->cur_ids);
+		darray_init(&r->cur_ids);
+	}
+out:
+	r->cur_ids.nr = 0;
+	return 0;
+}
+
+static int get_snapshot_trees(struct bch_fs *c, struct snapshot_tree_reconstruct *r, struct bpos pos)
+{
+	if (!same_snapshot(r, pos))
+		snapshot_tree_reconstruct_next(c, r);
+	r->cur_pos = pos;
+	return snapshot_list_add_nodup(c, &r->cur_ids, pos.snapshot);
+}
+
+int bch2_reconstruct_snapshots(struct bch_fs *c)
+{
+	struct btree_trans *trans = bch2_trans_get(c);
+	struct printbuf buf = PRINTBUF;
+	struct snapshot_tree_reconstruct r = {};
+	int ret = 0;
+
+	for (unsigned btree = 0; btree < BTREE_ID_NR; btree++) {
+		if (btree_type_has_snapshots(btree)) {
+			r.btree = btree;
+
+			ret = for_each_btree_key(trans, iter, btree, POS_MIN,
+					BTREE_ITER_ALL_SNAPSHOTS|BTREE_ITER_PREFETCH, k, ({
+				get_snapshot_trees(c, &r, k.k->p);
+			}));
+			if (ret)
+				goto err;
+
+			snapshot_tree_reconstruct_next(c, &r);
+		}
+	}
+
+	darray_for_each(r.trees, t) {
+		printbuf_reset(&buf);
+		snapshot_id_list_to_text(&buf, t);
+
+		darray_for_each(*t, id) {
+			if (fsck_err_on(!bch2_snapshot_equiv(c, *id),
+					c, snapshot_node_missing,
+					"snapshot node %u from tree %s missing", *id, buf.buf)) {
+				if (t->nr > 1) {
+					bch_err(c, "cannot reconstruct snapshot trees with multiple nodes");
+					ret = -BCH_ERR_fsck_repair_unimplemented;
+					goto err;
+				}
+
+				ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+						check_snapshot_exists(trans, *id));
+				if (ret)
+					goto err;
+			}
+		}
+	}
+fsck_err:
+err:
+	bch2_trans_put(trans);
+	snapshot_tree_reconstruct_exit(&r);
+	printbuf_exit(&buf);
+	bch_err_fn(c, ret);
+	return ret;
+}
+
 /*
  * Mark a snapshot as deleted, for future cleanup:
  */
@@ -1682,6 +1851,20 @@ int bch2_snapshots_read(struct bch_fs *c)
 				   POS_MIN, 0, k,
 			   (set_is_ancestor_bitmap(c, k.k->p.offset), 0)));
 	bch_err_fn(c, ret);
+
+	/*
+	 * It's important that we check if we need to reconstruct snapshots
+	 * before going RW, so we mark that pass as required in the superblock -
+	 * otherwise, we could end up deleting keys with missing snapshot nodes
+	 * instead
+	 */
+	BUG_ON(!test_bit(BCH_FS_new_fs, &c->flags) &&
+	       test_bit(BCH_FS_may_go_rw, &c->flags));
+
+	if (bch2_err_matches(ret, EIO) ||
+	    (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_snapshots)))
+		ret = bch2_run_explicit_recovery_pass_persistent(c, BCH_RECOVERY_PASS_reconstruct_snapshots);
+
 	return ret;
 }
 
diff --git a/fs/bcachefs/snapshot.h b/fs/bcachefs/snapshot.h
index 7c66ffc06385..b7d2fed37c4f 100644
--- a/fs/bcachefs/snapshot.h
+++ b/fs/bcachefs/snapshot.h
@@ -33,7 +33,11 @@ int bch2_mark_snapshot(struct btree_trans *, enum btree_id, unsigned,
 
 static inline struct snapshot_t *__snapshot_t(struct snapshot_table *t, u32 id)
 {
-	return &t->s[U32_MAX - id];
+	u32 idx = U32_MAX - id;
+
+	return likely(t && idx < t->nr)
+		? &t->s[idx]
+		: NULL;
 }
 
 static inline const struct snapshot_t *snapshot_t(struct bch_fs *c, u32 id)
@@ -44,7 +48,8 @@ static inline const struct snapshot_t *snapshot_t(struct bch_fs *c, u32 id)
 static inline u32 bch2_snapshot_tree(struct bch_fs *c, u32 id)
 {
 	rcu_read_lock();
-	id = snapshot_t(c, id)->tree;
+	const struct snapshot_t *s = snapshot_t(c, id);
+	id = s ? s->tree : 0;
 	rcu_read_unlock();
 
 	return id;
@@ -52,7 +57,8 @@ static inline u32 bch2_snapshot_tree(struct bch_fs *c, u32 id)
 
 static inline u32 __bch2_snapshot_parent_early(struct bch_fs *c, u32 id)
 {
-	return snapshot_t(c, id)->parent;
+	const struct snapshot_t *s = snapshot_t(c, id);
+	return s ? s->parent : 0;
 }
 
 static inline u32 bch2_snapshot_parent_early(struct bch_fs *c, u32 id)
@@ -66,19 +72,19 @@ static inline u32 bch2_snapshot_parent_early(struct bch_fs *c, u32 id)
 
 static inline u32 __bch2_snapshot_parent(struct bch_fs *c, u32 id)
 {
-#ifdef CONFIG_BCACHEFS_DEBUG
-	u32 parent = snapshot_t(c, id)->parent;
+	const struct snapshot_t *s = snapshot_t(c, id);
+	if (!s)
+		return 0;
 
-	if (parent &&
-	    snapshot_t(c, id)->depth != snapshot_t(c, parent)->depth + 1)
+	u32 parent = s->parent;
+	if (IS_ENABLED(CONFIG_BCACHEFS_DEBU) &&
+	    parent &&
+	    s->depth != snapshot_t(c, parent)->depth + 1)
 		panic("id %u depth=%u parent %u depth=%u\n",
 		      id, snapshot_t(c, id)->depth,
 		      parent, snapshot_t(c, parent)->depth);
 
 	return parent;
-#else
-	return snapshot_t(c, id)->parent;
-#endif
 }
 
 static inline u32 bch2_snapshot_parent(struct bch_fs *c, u32 id)
@@ -116,7 +122,8 @@ static inline u32 bch2_snapshot_root(struct bch_fs *c, u32 id)
 
 static inline u32 __bch2_snapshot_equiv(struct bch_fs *c, u32 id)
 {
-	return snapshot_t(c, id)->equiv;
+	const struct snapshot_t *s = snapshot_t(c, id);
+	return s ? s->equiv : 0;
 }
 
 static inline u32 bch2_snapshot_equiv(struct bch_fs *c, u32 id)
@@ -133,38 +140,22 @@ static inline bool bch2_snapshot_is_equiv(struct bch_fs *c, u32 id)
 	return id == bch2_snapshot_equiv(c, id);
 }
 
-static inline bool bch2_snapshot_is_internal_node(struct bch_fs *c, u32 id)
+static inline int bch2_snapshot_is_internal_node(struct bch_fs *c, u32 id)
 {
-	const struct snapshot_t *s;
-	bool ret;
-
 	rcu_read_lock();
-	s = snapshot_t(c, id);
-	ret = s->children[0];
+	const struct snapshot_t *s = snapshot_t(c, id);
+	int ret = s ? s->children[0] : -BCH_ERR_invalid_snapshot_node;
 	rcu_read_unlock();
 
 	return ret;
 }
 
-static inline u32 bch2_snapshot_is_leaf(struct bch_fs *c, u32 id)
-{
-	return !bch2_snapshot_is_internal_node(c, id);
-}
-
-static inline u32 bch2_snapshot_sibling(struct bch_fs *c, u32 id)
+static inline int bch2_snapshot_is_leaf(struct bch_fs *c, u32 id)
 {
-	const struct snapshot_t *s;
-	u32 parent = __bch2_snapshot_parent(c, id);
-
-	if (!parent)
-		return 0;
-
-	s = snapshot_t(c, __bch2_snapshot_parent(c, id));
-	if (id == s->children[0])
-		return s->children[1];
-	if (id == s->children[1])
-		return s->children[0];
-	return 0;
+	int ret = bch2_snapshot_is_internal_node(c, id);
+	if (ret < 0)
+		return ret;
+	return !ret;
 }
 
 static inline u32 bch2_snapshot_depth(struct bch_fs *c, u32 parent)
@@ -218,15 +209,34 @@ static inline bool snapshot_list_has_ancestor(struct bch_fs *c, snapshot_id_list
 
 static inline int snapshot_list_add(struct bch_fs *c, snapshot_id_list *s, u32 id)
 {
-	int ret;
-
 	BUG_ON(snapshot_list_has_id(s, id));
-	ret = darray_push(s, id);
+	int ret = darray_push(s, id);
 	if (ret)
 		bch_err(c, "error reallocating snapshot_id_list (size %zu)", s->size);
 	return ret;
 }
 
+static inline int snapshot_list_add_nodup(struct bch_fs *c, snapshot_id_list *s, u32 id)
+{
+	int ret = snapshot_list_has_id(s, id)
+		? 0
+		: darray_push(s, id);
+	if (ret)
+		bch_err(c, "error reallocating snapshot_id_list (size %zu)", s->size);
+	return ret;
+}
+
+static inline int snapshot_list_merge(struct bch_fs *c, snapshot_id_list *dst, snapshot_id_list *src)
+{
+	darray_for_each(*src, i) {
+		int ret = snapshot_list_add_nodup(c, dst, *i);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
 int bch2_snapshot_lookup(struct btree_trans *trans, u32 id,
 			 struct bch_snapshot *s);
 int bch2_snapshot_get_subvol(struct btree_trans *, u32,
@@ -238,6 +248,7 @@ int bch2_snapshot_node_create(struct btree_trans *, u32,
 
 int bch2_check_snapshot_trees(struct bch_fs *);
 int bch2_check_snapshots(struct bch_fs *);
+int bch2_reconstruct_snapshots(struct bch_fs *);
 
 int bch2_snapshot_node_set_deleted(struct btree_trans *, u32);
 void bch2_delete_dead_snapshots_work(struct work_struct *);
@@ -249,7 +260,7 @@ static inline int bch2_key_has_snapshot_overwrites(struct btree_trans *trans,
 					  struct bpos pos)
 {
 	if (!btree_type_has_snapshots(id) ||
-	    bch2_snapshot_is_leaf(trans->c, pos.snapshot))
+	    bch2_snapshot_is_leaf(trans->c, pos.snapshot) > 0)
 		return 0;
 
 	return __bch2_key_has_snapshot_overwrites(trans, id, pos);
diff --git a/fs/bcachefs/subvolume.c b/fs/bcachefs/subvolume.c
index ce7aed121942..88a79c823276 100644
--- a/fs/bcachefs/subvolume.c
+++ b/fs/bcachefs/subvolume.c
@@ -595,6 +595,78 @@ err:
 	return ret;
 }
 
+int bch2_initialize_subvolumes(struct bch_fs *c)
+{
+	struct bkey_i_snapshot_tree	root_tree;
+	struct bkey_i_snapshot		root_snapshot;
+	struct bkey_i_subvolume		root_volume;
+	int ret;
+
+	bkey_snapshot_tree_init(&root_tree.k_i);
+	root_tree.k.p.offset		= 1;
+	root_tree.v.master_subvol	= cpu_to_le32(1);
+	root_tree.v.root_snapshot	= cpu_to_le32(U32_MAX);
+
+	bkey_snapshot_init(&root_snapshot.k_i);
+	root_snapshot.k.p.offset = U32_MAX;
+	root_snapshot.v.flags	= 0;
+	root_snapshot.v.parent	= 0;
+	root_snapshot.v.subvol	= cpu_to_le32(BCACHEFS_ROOT_SUBVOL);
+	root_snapshot.v.tree	= cpu_to_le32(1);
+	SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
+
+	bkey_subvolume_init(&root_volume.k_i);
+	root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
+	root_volume.v.flags	= 0;
+	root_volume.v.snapshot	= cpu_to_le32(U32_MAX);
+	root_volume.v.inode	= cpu_to_le64(BCACHEFS_ROOT_INO);
+
+	ret =   bch2_btree_insert(c, BTREE_ID_snapshot_trees,	&root_tree.k_i, NULL, 0) ?:
+		bch2_btree_insert(c, BTREE_ID_snapshots,	&root_snapshot.k_i, NULL, 0) ?:
+		bch2_btree_insert(c, BTREE_ID_subvolumes,	&root_volume.k_i, NULL, 0);
+	bch_err_fn(c, ret);
+	return ret;
+}
+
+static int __bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
+{
+	struct btree_iter iter;
+	struct bkey_s_c k;
+	struct bch_inode_unpacked inode;
+	int ret;
+
+	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
+			       SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
+	ret = bkey_err(k);
+	if (ret)
+		return ret;
+
+	if (!bkey_is_inode(k.k)) {
+		bch_err(trans->c, "root inode not found");
+		ret = -BCH_ERR_ENOENT_inode;
+		goto err;
+	}
+
+	ret = bch2_inode_unpack(k, &inode);
+	BUG_ON(ret);
+
+	inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
+
+	ret = bch2_inode_write(trans, &iter, &inode);
+err:
+	bch2_trans_iter_exit(trans, &iter);
+	return ret;
+}
+
+/* set bi_subvol on root inode */
+int bch2_fs_upgrade_for_subvolumes(struct bch_fs *c)
+{
+	int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_lazy_rw,
+				__bch2_fs_upgrade_for_subvolumes(trans));
+	bch_err_fn(c, ret);
+	return ret;
+}
+
 int bch2_fs_subvolumes_init(struct bch_fs *c)
 {
 	INIT_WORK(&c->snapshot_delete_work, bch2_delete_dead_snapshots_work);
diff --git a/fs/bcachefs/subvolume.h b/fs/bcachefs/subvolume.h
index 903c05162c06..d2015d549bd2 100644
--- a/fs/bcachefs/subvolume.h
+++ b/fs/bcachefs/subvolume.h
@@ -37,6 +37,9 @@ void bch2_delete_dead_snapshots_async(struct bch_fs *);
 int bch2_subvolume_unlink(struct btree_trans *, u32);
 int bch2_subvolume_create(struct btree_trans *, u64, u32, u32, u32 *, u32 *, bool);
 
+int bch2_initialize_subvolumes(struct bch_fs *);
+int bch2_fs_upgrade_for_subvolumes(struct bch_fs *);
+
 int bch2_fs_subvolumes_init(struct bch_fs *);
 
 #endif /* _BCACHEFS_SUBVOLUME_H */
diff --git a/fs/bcachefs/subvolume_types.h b/fs/bcachefs/subvolume_types.h
index ae644adfc391..9b10c8947828 100644
--- a/fs/bcachefs/subvolume_types.h
+++ b/fs/bcachefs/subvolume_types.h
@@ -20,6 +20,8 @@ struct snapshot_t {
 };
 
 struct snapshot_table {
+	struct rcu_head		rcu;
+	size_t			nr;
 #ifndef RUST_BINDGEN
 	DECLARE_FLEX_ARRAY(struct snapshot_t, s);
 #else
diff --git a/fs/bcachefs/super-io.c b/fs/bcachefs/super-io.c
index ad28e370b640..08ea3dbbbe97 100644
--- a/fs/bcachefs/super-io.c
+++ b/fs/bcachefs/super-io.c
@@ -8,7 +8,7 @@
 #include "journal.h"
 #include "journal_sb.h"
 #include "journal_seq_blacklist.h"
-#include "recovery.h"
+#include "recovery_passes.h"
 #include "replicas.h"
 #include "quota.h"
 #include "sb-clean.h"
@@ -143,7 +143,7 @@ void bch2_free_super(struct bch_sb_handle *sb)
 {
 	kfree(sb->bio);
 	if (!IS_ERR_OR_NULL(sb->s_bdev_file))
-		fput(sb->s_bdev_file);
+		bdev_fput(sb->s_bdev_file);
 	kfree(sb->holder);
 	kfree(sb->sb_name);
 
@@ -527,9 +527,11 @@ static void bch2_sb_update(struct bch_fs *c)
 	memset(c->sb.errors_silent, 0, sizeof(c->sb.errors_silent));
 
 	struct bch_sb_field_ext *ext = bch2_sb_field_get(src, ext);
-	if (ext)
+	if (ext) {
 		le_bitvector_to_cpu(c->sb.errors_silent, (void *) ext->errors_silent,
 				    sizeof(c->sb.errors_silent) * 8);
+		c->sb.btrees_lost_data = le64_to_cpu(ext->btrees_lost_data);
+	}
 
 	for_each_member_device(c, ca) {
 		struct bch_member m = bch2_sb_member_get(src, ca->dev_idx);
@@ -698,8 +700,11 @@ retry:
 		return -ENOMEM;
 
 	sb->sb_name = kstrdup(path, GFP_KERNEL);
-	if (!sb->sb_name)
-		return -ENOMEM;
+	if (!sb->sb_name) {
+		ret = -ENOMEM;
+		prt_printf(&err, "error allocating memory for sb_name");
+		goto err;
+	}
 
 #ifndef __KERNEL__
 	if (opt_get(*opts, direct_io) == false)
@@ -1162,6 +1167,11 @@ static void bch2_sb_ext_to_text(struct printbuf *out, struct bch_sb *sb,
 
 		kfree(errors_silent);
 	}
+
+	prt_printf(out, "Btrees with missing data:");
+	prt_tab(out);
+	prt_bitflags(out, __bch2_btree_ids, le64_to_cpu(e->btrees_lost_data));
+	prt_newline(out);
 }
 
 static const struct bch_sb_field_ops bch_sb_field_ops_ext = {
diff --git a/fs/bcachefs/super.c b/fs/bcachefs/super.c
index 1ad6e5cd9476..8daf80a38d60 100644
--- a/fs/bcachefs/super.c
+++ b/fs/bcachefs/super.c
@@ -15,6 +15,7 @@
 #include "btree_gc.h"
 #include "btree_journal_iter.h"
 #include "btree_key_cache.h"
+#include "btree_node_scan.h"
 #include "btree_update_interior.h"
 #include "btree_io.h"
 #include "btree_write_buffer.h"
@@ -287,8 +288,13 @@ static void __bch2_fs_read_only(struct bch_fs *c)
 	if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
 	    !test_bit(BCH_FS_emergency_ro, &c->flags))
 		set_bit(BCH_FS_clean_shutdown, &c->flags);
+
 	bch2_fs_journal_stop(&c->journal);
 
+	bch_info(c, "%sshutdown complete, journal seq %llu",
+		 test_bit(BCH_FS_clean_shutdown, &c->flags) ? "" : "un",
+		 c->journal.seq_ondisk);
+
 	/*
 	 * After stopping journal:
 	 */
@@ -365,7 +371,7 @@ void bch2_fs_read_only(struct bch_fs *c)
 	    !test_bit(BCH_FS_emergency_ro, &c->flags) &&
 	    test_bit(BCH_FS_started, &c->flags) &&
 	    test_bit(BCH_FS_clean_shutdown, &c->flags) &&
-	    !c->opts.norecovery) {
+	    c->recovery_pass_done >= BCH_RECOVERY_PASS_journal_replay) {
 		BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
 		BUG_ON(atomic_read(&c->btree_cache.dirty));
 		BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
@@ -510,7 +516,8 @@ err:
 
 int bch2_fs_read_write(struct bch_fs *c)
 {
-	if (c->opts.norecovery)
+	if (c->opts.recovery_pass_last &&
+	    c->opts.recovery_pass_last < BCH_RECOVERY_PASS_journal_replay)
 		return -BCH_ERR_erofs_norecovery;
 
 	if (c->opts.nochanges)
@@ -535,6 +542,7 @@ static void __bch2_fs_free(struct bch_fs *c)
 	for (i = 0; i < BCH_TIME_STAT_NR; i++)
 		bch2_time_stats_exit(&c->times[i]);
 
+	bch2_find_btree_nodes_exit(&c->found_btree_nodes);
 	bch2_free_pending_node_rewrites(c);
 	bch2_fs_sb_errors_exit(c);
 	bch2_fs_counters_exit(c);
@@ -559,6 +567,7 @@ static void __bch2_fs_free(struct bch_fs *c)
 	bch2_io_clock_exit(&c->io_clock[READ]);
 	bch2_fs_compress_exit(c);
 	bch2_journal_keys_put_initial(c);
+	bch2_find_btree_nodes_exit(&c->found_btree_nodes);
 	BUG_ON(atomic_read(&c->journal_keys.ref));
 	bch2_fs_btree_write_buffer_exit(c);
 	percpu_free_rwsem(&c->mark_lock);
@@ -1015,8 +1024,16 @@ int bch2_fs_start(struct bch_fs *c)
 	for_each_online_member(c, ca)
 		bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now);
 
+	struct bch_sb_field_ext *ext =
+		bch2_sb_field_get_minsize(&c->disk_sb, ext, sizeof(*ext) / sizeof(u64));
 	mutex_unlock(&c->sb_lock);
 
+	if (!ext) {
+		bch_err(c, "insufficient space in superblock for sb_field_ext");
+		ret = -BCH_ERR_ENOSPC_sb;
+		goto err;
+	}
+
 	for_each_rw_member(c, ca)
 		bch2_dev_allocator_add(c, ca);
 	bch2_recalc_capacity(c);
diff --git a/fs/bcachefs/super_types.h b/fs/bcachefs/super_types.h
index ec784d975f66..11bcef170c2c 100644
--- a/fs/bcachefs/super_types.h
+++ b/fs/bcachefs/super_types.h
@@ -37,6 +37,8 @@ struct bch_member_cpu {
 	u8			durability;
 	u8			freespace_initialized;
 	u8			valid;
+	u8			btree_bitmap_shift;
+	u64			btree_allocated_bitmap;
 };
 
 #endif /* _BCACHEFS_SUPER_TYPES_H */
diff --git a/fs/bcachefs/sysfs.c b/fs/bcachefs/sysfs.c
index c86a93a8d8fc..5be92fe3f4ea 100644
--- a/fs/bcachefs/sysfs.c
+++ b/fs/bcachefs/sysfs.c
@@ -17,7 +17,6 @@
 #include "btree_iter.h"
 #include "btree_key_cache.h"
 #include "btree_update.h"
-#include "btree_update_interior.h"
 #include "btree_gc.h"
 #include "buckets.h"
 #include "clock.h"
@@ -26,6 +25,7 @@
 #include "ec.h"
 #include "inode.h"
 #include "journal.h"
+#include "journal_reclaim.h"
 #include "keylist.h"
 #include "move.h"
 #include "movinggc.h"
@@ -139,6 +139,7 @@ do {									\
 write_attribute(trigger_gc);
 write_attribute(trigger_discards);
 write_attribute(trigger_invalidates);
+write_attribute(trigger_journal_flush);
 write_attribute(prune_cache);
 write_attribute(btree_wakeup);
 rw_attribute(btree_gc_periodic);
@@ -166,7 +167,6 @@ read_attribute(btree_write_stats);
 read_attribute(btree_cache_size);
 read_attribute(compression_stats);
 read_attribute(journal_debug);
-read_attribute(btree_updates);
 read_attribute(btree_cache);
 read_attribute(btree_key_cache);
 read_attribute(stripes_heap);
@@ -415,9 +415,6 @@ SHOW(bch2_fs)
 	if (attr == &sysfs_journal_debug)
 		bch2_journal_debug_to_text(out, &c->journal);
 
-	if (attr == &sysfs_btree_updates)
-		bch2_btree_updates_to_text(out, c);
-
 	if (attr == &sysfs_btree_cache)
 		bch2_btree_cache_to_text(out, c);
 
@@ -505,7 +502,7 @@ STORE(bch2_fs)
 
 	/* Debugging: */
 
-	if (!test_bit(BCH_FS_rw, &c->flags))
+	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_sysfs))
 		return -EROFS;
 
 	if (attr == &sysfs_prune_cache) {
@@ -538,6 +535,11 @@ STORE(bch2_fs)
 	if (attr == &sysfs_trigger_invalidates)
 		bch2_do_invalidates(c);
 
+	if (attr == &sysfs_trigger_journal_flush) {
+		bch2_journal_flush_all_pins(&c->journal);
+		bch2_journal_meta(&c->journal);
+	}
+
 #ifdef CONFIG_BCACHEFS_TESTS
 	if (attr == &sysfs_perf_test) {
 		char *tmp = kstrdup(buf, GFP_KERNEL), *p = tmp;
@@ -558,6 +560,7 @@ STORE(bch2_fs)
 			size = ret;
 	}
 #endif
+	bch2_write_ref_put(c, BCH_WRITE_REF_sysfs);
 	return size;
 }
 SYSFS_OPS(bch2_fs);
@@ -639,7 +642,6 @@ SYSFS_OPS(bch2_fs_internal);
 struct attribute *bch2_fs_internal_files[] = {
 	&sysfs_flags,
 	&sysfs_journal_debug,
-	&sysfs_btree_updates,
 	&sysfs_btree_cache,
 	&sysfs_btree_key_cache,
 	&sysfs_new_stripes,
@@ -657,6 +659,7 @@ struct attribute *bch2_fs_internal_files[] = {
 	&sysfs_trigger_gc,
 	&sysfs_trigger_discards,
 	&sysfs_trigger_invalidates,
+	&sysfs_trigger_journal_flush,
 	&sysfs_prune_cache,
 	&sysfs_btree_wakeup,
 
diff --git a/fs/bcachefs/tests.c b/fs/bcachefs/tests.c
index b3fe9fc57747..bfec656f94c0 100644
--- a/fs/bcachefs/tests.c
+++ b/fs/bcachefs/tests.c
@@ -672,7 +672,7 @@ static int __do_delete(struct btree_trans *trans, struct bpos pos)
 
 	bch2_trans_iter_init(trans, &iter, BTREE_ID_xattrs, pos,
 			     BTREE_ITER_INTENT);
-	k = bch2_btree_iter_peek(&iter);
+	k = bch2_btree_iter_peek_upto(&iter, POS(0, U64_MAX));
 	ret = bkey_err(k);
 	if (ret)
 		goto err;
diff --git a/fs/bcachefs/util.c b/fs/bcachefs/util.c
index 216fadf16928..92c6ad75e702 100644
--- a/fs/bcachefs/util.c
+++ b/fs/bcachefs/util.c
@@ -707,149 +707,6 @@ void memcpy_from_bio(void *dst, struct bio *src, struct bvec_iter src_iter)
 	}
 }
 
-static int alignment_ok(const void *base, size_t align)
-{
-	return IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) ||
-		((unsigned long)base & (align - 1)) == 0;
-}
-
-static void u32_swap(void *a, void *b, size_t size)
-{
-	u32 t = *(u32 *)a;
-	*(u32 *)a = *(u32 *)b;
-	*(u32 *)b = t;
-}
-
-static void u64_swap(void *a, void *b, size_t size)
-{
-	u64 t = *(u64 *)a;
-	*(u64 *)a = *(u64 *)b;
-	*(u64 *)b = t;
-}
-
-static void generic_swap(void *a, void *b, size_t size)
-{
-	char t;
-
-	do {
-		t = *(char *)a;
-		*(char *)a++ = *(char *)b;
-		*(char *)b++ = t;
-	} while (--size > 0);
-}
-
-static inline int do_cmp(void *base, size_t n, size_t size,
-			 int (*cmp_func)(const void *, const void *, size_t),
-			 size_t l, size_t r)
-{
-	return cmp_func(base + inorder_to_eytzinger0(l, n) * size,
-			base + inorder_to_eytzinger0(r, n) * size,
-			size);
-}
-
-static inline void do_swap(void *base, size_t n, size_t size,
-			   void (*swap_func)(void *, void *, size_t),
-			   size_t l, size_t r)
-{
-	swap_func(base + inorder_to_eytzinger0(l, n) * size,
-		  base + inorder_to_eytzinger0(r, n) * size,
-		  size);
-}
-
-void eytzinger0_sort(void *base, size_t n, size_t size,
-		     int (*cmp_func)(const void *, const void *, size_t),
-		     void (*swap_func)(void *, void *, size_t))
-{
-	int i, c, r;
-
-	if (!swap_func) {
-		if (size == 4 && alignment_ok(base, 4))
-			swap_func = u32_swap;
-		else if (size == 8 && alignment_ok(base, 8))
-			swap_func = u64_swap;
-		else
-			swap_func = generic_swap;
-	}
-
-	/* heapify */
-	for (i = n / 2 - 1; i >= 0; --i) {
-		for (r = i; r * 2 + 1 < n; r = c) {
-			c = r * 2 + 1;
-
-			if (c + 1 < n &&
-			    do_cmp(base, n, size, cmp_func, c, c + 1) < 0)
-				c++;
-
-			if (do_cmp(base, n, size, cmp_func, r, c) >= 0)
-				break;
-
-			do_swap(base, n, size, swap_func, r, c);
-		}
-	}
-
-	/* sort */
-	for (i = n - 1; i > 0; --i) {
-		do_swap(base, n, size, swap_func, 0, i);
-
-		for (r = 0; r * 2 + 1 < i; r = c) {
-			c = r * 2 + 1;
-
-			if (c + 1 < i &&
-			    do_cmp(base, n, size, cmp_func, c, c + 1) < 0)
-				c++;
-
-			if (do_cmp(base, n, size, cmp_func, r, c) >= 0)
-				break;
-
-			do_swap(base, n, size, swap_func, r, c);
-		}
-	}
-}
-
-void sort_cmp_size(void *base, size_t num, size_t size,
-	  int (*cmp_func)(const void *, const void *, size_t),
-	  void (*swap_func)(void *, void *, size_t size))
-{
-	/* pre-scale counters for performance */
-	int i = (num/2 - 1) * size, n = num * size, c, r;
-
-	if (!swap_func) {
-		if (size == 4 && alignment_ok(base, 4))
-			swap_func = u32_swap;
-		else if (size == 8 && alignment_ok(base, 8))
-			swap_func = u64_swap;
-		else
-			swap_func = generic_swap;
-	}
-
-	/* heapify */
-	for ( ; i >= 0; i -= size) {
-		for (r = i; r * 2 + size < n; r  = c) {
-			c = r * 2 + size;
-			if (c < n - size &&
-			    cmp_func(base + c, base + c + size, size) < 0)
-				c += size;
-			if (cmp_func(base + r, base + c, size) >= 0)
-				break;
-			swap_func(base + r, base + c, size);
-		}
-	}
-
-	/* sort */
-	for (i = n - size; i > 0; i -= size) {
-		swap_func(base, base + i, size);
-		for (r = 0; r * 2 + size < i; r = c) {
-			c = r * 2 + size;
-			if (c < i - size &&
-			    cmp_func(base + c, base + c + size, size) < 0)
-				c += size;
-			if (cmp_func(base + r, base + c, size) >= 0)
-				break;
-			swap_func(base + r, base + c, size);
-		}
-	}
-}
-
 #if 0
 void eytzinger1_test(void)
 {
diff --git a/fs/bcachefs/util.h b/fs/bcachefs/util.h
index 175aee3074c7..5cf885b09986 100644
--- a/fs/bcachefs/util.h
+++ b/fs/bcachefs/util.h
@@ -631,10 +631,6 @@ static inline void memset_u64s_tail(void *s, int c, unsigned bytes)
 	memset(s + bytes, c, rem);
 }
 
-void sort_cmp_size(void *base, size_t num, size_t size,
-	  int (*cmp_func)(const void *, const void *, size_t),
-	  void (*swap_func)(void *, void *, size_t));
-
 /* just the memmove, doesn't update @_nr */
 #define __array_insert_item(_array, _nr, _pos)				\
 	memmove(&(_array)[(_pos) + 1],					\
@@ -792,9 +788,27 @@ static inline int copy_from_user_errcode(void *to, const void __user *from, unsi
 
 #endif
 
+static inline void mod_bit(long nr, volatile unsigned long *addr, bool v)
+{
+	if (v)
+		set_bit(nr, addr);
+	else
+		clear_bit(nr, addr);
+}
+
 static inline void __set_bit_le64(size_t bit, __le64 *addr)
 {
 	addr[bit / 64] |= cpu_to_le64(BIT_ULL(bit % 64));
 }
 
+static inline void __clear_bit_le64(size_t bit, __le64 *addr)
+{
+	addr[bit / 64] &= ~cpu_to_le64(BIT_ULL(bit % 64));
+}
+
+static inline bool test_bit_le64(size_t bit, __le64 *addr)
+{
+	return (addr[bit / 64] & cpu_to_le64(BIT_ULL(bit % 64))) != 0;
+}
+
 #endif /* _BCACHEFS_UTIL_H */
diff --git a/fs/btrfs/delayed-inode.c b/fs/btrfs/delayed-inode.c
index dd6f566a383f..121ab890bd05 100644
--- a/fs/btrfs/delayed-inode.c
+++ b/fs/btrfs/delayed-inode.c
@@ -1133,6 +1133,9 @@ __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
 	if (ret)
 		return ret;
 
+	ret = btrfs_record_root_in_trans(trans, node->root);
+	if (ret)
+		return ret;
 	ret = btrfs_update_delayed_inode(trans, node->root, path, node);
 	return ret;
 }
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index beedd6ed64d3..257d044bca91 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -3464,6 +3464,14 @@ void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
 	if (root_id != BTRFS_TREE_LOG_OBJECTID) {
 		struct btrfs_ref generic_ref = { 0 };
 
+		/*
+		 * Assert that the extent buffer is not cleared due to
+		 * EXTENT_BUFFER_ZONED_ZEROOUT. Please refer
+		 * btrfs_clear_buffer_dirty() and btree_csum_one_bio() for
+		 * detail.
+		 */
+		ASSERT(btrfs_header_bytenr(buf) != 0);
+
 		btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF,
 				       buf->start, buf->len, parent,
 				       btrfs_header_owner(buf));
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 61594eaf1f89..2776112dbdf8 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -681,31 +681,21 @@ static void end_bbio_data_read(struct btrfs_bio *bbio)
 int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array,
 			   gfp_t extra_gfp)
 {
+	const gfp_t gfp = GFP_NOFS | extra_gfp;
 	unsigned int allocated;
 
 	for (allocated = 0; allocated < nr_pages;) {
 		unsigned int last = allocated;
 
-		allocated = alloc_pages_bulk_array(GFP_NOFS | extra_gfp,
-						   nr_pages, page_array);
-
-		if (allocated == nr_pages)
-			return 0;
-
-		/*
-		 * During this iteration, no page could be allocated, even
-		 * though alloc_pages_bulk_array() falls back to alloc_page()
-		 * if  it could not bulk-allocate. So we must be out of memory.
-		 */
-		if (allocated == last) {
+		allocated = alloc_pages_bulk_array(gfp, nr_pages, page_array);
+		if (unlikely(allocated == last)) {
+			/* No progress, fail and do cleanup. */
 			for (int i = 0; i < allocated; i++) {
 				__free_page(page_array[i]);
 				page_array[i] = NULL;
 			}
 			return -ENOMEM;
 		}
-
-		memalloc_retry_wait(GFP_NOFS);
 	}
 	return 0;
 }
@@ -4154,7 +4144,7 @@ void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans,
 	 * The actual zeroout of the buffer will happen later in
 	 * btree_csum_one_bio.
 	 */
-	if (btrfs_is_zoned(fs_info)) {
+	if (btrfs_is_zoned(fs_info) && test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
 		set_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags);
 		return;
 	}
@@ -4193,6 +4183,7 @@ void set_extent_buffer_dirty(struct extent_buffer *eb)
 	num_folios = num_extent_folios(eb);
 	WARN_ON(atomic_read(&eb->refs) == 0);
 	WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
+	WARN_ON(test_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags));
 
 	if (!was_dirty) {
 		bool subpage = eb->fs_info->nodesize < PAGE_SIZE;
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 37701531eeb1..c65fe5de4022 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -2533,7 +2533,7 @@ void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
 		 */
 		if (bits & EXTENT_CLEAR_META_RESV &&
 		    root != fs_info->tree_root)
-			btrfs_delalloc_release_metadata(inode, len, false);
+			btrfs_delalloc_release_metadata(inode, len, true);
 
 		/* For sanity tests. */
 		if (btrfs_is_testing(fs_info))
@@ -4503,6 +4503,7 @@ int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
 	struct btrfs_trans_handle *trans;
 	struct btrfs_block_rsv block_rsv;
 	u64 root_flags;
+	u64 qgroup_reserved = 0;
 	int ret;
 
 	down_write(&fs_info->subvol_sem);
@@ -4547,12 +4548,20 @@ int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
 	ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
 	if (ret)
 		goto out_undead;
+	qgroup_reserved = block_rsv.qgroup_rsv_reserved;
 
 	trans = btrfs_start_transaction(root, 0);
 	if (IS_ERR(trans)) {
 		ret = PTR_ERR(trans);
 		goto out_release;
 	}
+	ret = btrfs_record_root_in_trans(trans, root);
+	if (ret) {
+		btrfs_abort_transaction(trans, ret);
+		goto out_end_trans;
+	}
+	btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved);
+	qgroup_reserved = 0;
 	trans->block_rsv = &block_rsv;
 	trans->bytes_reserved = block_rsv.size;
 
@@ -4611,7 +4620,9 @@ out_end_trans:
 	ret = btrfs_end_transaction(trans);
 	inode->i_flags |= S_DEAD;
 out_release:
-	btrfs_subvolume_release_metadata(root, &block_rsv);
+	btrfs_block_rsv_release(fs_info, &block_rsv, (u64)-1, NULL);
+	if (qgroup_reserved)
+		btrfs_qgroup_free_meta_prealloc(root, qgroup_reserved);
 out_undead:
 	if (ret) {
 		spin_lock(&dest->root_item_lock);
diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 294e31edec9d..55f3ba6a831c 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -613,6 +613,7 @@ static noinline int create_subvol(struct mnt_idmap *idmap,
 	int ret;
 	dev_t anon_dev;
 	u64 objectid;
+	u64 qgroup_reserved = 0;
 
 	root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 	if (!root_item)
@@ -650,13 +651,18 @@ static noinline int create_subvol(struct mnt_idmap *idmap,
 					       trans_num_items, false);
 	if (ret)
 		goto out_new_inode_args;
+	qgroup_reserved = block_rsv.qgroup_rsv_reserved;
 
 	trans = btrfs_start_transaction(root, 0);
 	if (IS_ERR(trans)) {
 		ret = PTR_ERR(trans);
-		btrfs_subvolume_release_metadata(root, &block_rsv);
-		goto out_new_inode_args;
+		goto out_release_rsv;
 	}
+	ret = btrfs_record_root_in_trans(trans, BTRFS_I(dir)->root);
+	if (ret)
+		goto out;
+	btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved);
+	qgroup_reserved = 0;
 	trans->block_rsv = &block_rsv;
 	trans->bytes_reserved = block_rsv.size;
 	/* Tree log can't currently deal with an inode which is a new root. */
@@ -767,9 +773,11 @@ static noinline int create_subvol(struct mnt_idmap *idmap,
 out:
 	trans->block_rsv = NULL;
 	trans->bytes_reserved = 0;
-	btrfs_subvolume_release_metadata(root, &block_rsv);
-
 	btrfs_end_transaction(trans);
+out_release_rsv:
+	btrfs_block_rsv_release(fs_info, &block_rsv, (u64)-1, NULL);
+	if (qgroup_reserved)
+		btrfs_qgroup_free_meta_prealloc(root, qgroup_reserved);
 out_new_inode_args:
 	btrfs_new_inode_args_destroy(&new_inode_args);
 out_inode:
@@ -791,6 +799,8 @@ static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 	struct btrfs_pending_snapshot *pending_snapshot;
 	unsigned int trans_num_items;
 	struct btrfs_trans_handle *trans;
+	struct btrfs_block_rsv *block_rsv;
+	u64 qgroup_reserved = 0;
 	int ret;
 
 	/* We do not support snapshotting right now. */
@@ -827,19 +837,19 @@ static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 		goto free_pending;
 	}
 
-	btrfs_init_block_rsv(&pending_snapshot->block_rsv,
-			     BTRFS_BLOCK_RSV_TEMP);
+	block_rsv = &pending_snapshot->block_rsv;
+	btrfs_init_block_rsv(block_rsv, BTRFS_BLOCK_RSV_TEMP);
 	/*
 	 * 1 to add dir item
 	 * 1 to add dir index
 	 * 1 to update parent inode item
 	 */
 	trans_num_items = create_subvol_num_items(inherit) + 3;
-	ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
-					       &pending_snapshot->block_rsv,
+	ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root, block_rsv,
 					       trans_num_items, false);
 	if (ret)
 		goto free_pending;
+	qgroup_reserved = block_rsv->qgroup_rsv_reserved;
 
 	pending_snapshot->dentry = dentry;
 	pending_snapshot->root = root;
@@ -852,6 +862,13 @@ static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 		ret = PTR_ERR(trans);
 		goto fail;
 	}
+	ret = btrfs_record_root_in_trans(trans, BTRFS_I(dir)->root);
+	if (ret) {
+		btrfs_end_transaction(trans);
+		goto fail;
+	}
+	btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved);
+	qgroup_reserved = 0;
 
 	trans->pending_snapshot = pending_snapshot;
 
@@ -881,7 +898,9 @@ fail:
 	if (ret && pending_snapshot->snap)
 		pending_snapshot->snap->anon_dev = 0;
 	btrfs_put_root(pending_snapshot->snap);
-	btrfs_subvolume_release_metadata(root, &pending_snapshot->block_rsv);
+	btrfs_block_rsv_release(fs_info, block_rsv, (u64)-1, NULL);
+	if (qgroup_reserved)
+		btrfs_qgroup_free_meta_prealloc(root, qgroup_reserved);
 free_pending:
 	if (pending_snapshot->anon_dev)
 		free_anon_bdev(pending_snapshot->anon_dev);
diff --git a/fs/btrfs/qgroup.c b/fs/btrfs/qgroup.c
index 5f90f0605b12..cf8820ce7aa2 100644
--- a/fs/btrfs/qgroup.c
+++ b/fs/btrfs/qgroup.c
@@ -4495,6 +4495,8 @@ void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
 				      BTRFS_QGROUP_RSV_META_PREALLOC);
 	trace_qgroup_meta_convert(root, num_bytes);
 	qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
+	if (!sb_rdonly(fs_info->sb))
+		add_root_meta_rsv(root, num_bytes, BTRFS_QGROUP_RSV_META_PERTRANS);
 }
 
 /*
diff --git a/fs/btrfs/root-tree.c b/fs/btrfs/root-tree.c
index 4bb538a372ce..7007f9e0c972 100644
--- a/fs/btrfs/root-tree.c
+++ b/fs/btrfs/root-tree.c
@@ -548,13 +548,3 @@ int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
 	}
 	return ret;
 }
-
-void btrfs_subvolume_release_metadata(struct btrfs_root *root,
-				      struct btrfs_block_rsv *rsv)
-{
-	struct btrfs_fs_info *fs_info = root->fs_info;
-	u64 qgroup_to_release;
-
-	btrfs_block_rsv_release(fs_info, rsv, (u64)-1, &qgroup_to_release);
-	btrfs_qgroup_convert_reserved_meta(root, qgroup_to_release);
-}
diff --git a/fs/btrfs/root-tree.h b/fs/btrfs/root-tree.h
index 6f929cf3bd49..8f5739e732b9 100644
--- a/fs/btrfs/root-tree.h
+++ b/fs/btrfs/root-tree.h
@@ -18,8 +18,6 @@ struct btrfs_trans_handle;
 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
 				     struct btrfs_block_rsv *rsv,
 				     int nitems, bool use_global_rsv);
-void btrfs_subvolume_release_metadata(struct btrfs_root *root,
-				      struct btrfs_block_rsv *rsv);
 int btrfs_add_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
 		       u64 ref_id, u64 dirid, u64 sequence,
 		       const struct fscrypt_str *name);
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 46e8426adf4f..85f359e0e0a7 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -745,14 +745,6 @@ again:
 		h->reloc_reserved = reloc_reserved;
 	}
 
-	/*
-	 * Now that we have found a transaction to be a part of, convert the
-	 * qgroup reservation from prealloc to pertrans. A different transaction
-	 * can't race in and free our pertrans out from under us.
-	 */
-	if (qgroup_reserved)
-		btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved);
-
 got_it:
 	if (!current->journal_info)
 		current->journal_info = h;
@@ -786,8 +778,15 @@ got_it:
 		 * not just freed.
 		 */
 		btrfs_end_transaction(h);
-		return ERR_PTR(ret);
+		goto reserve_fail;
 	}
+	/*
+	 * Now that we have found a transaction to be a part of, convert the
+	 * qgroup reservation from prealloc to pertrans. A different transaction
+	 * can't race in and free our pertrans out from under us.
+	 */
+	if (qgroup_reserved)
+		btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved);
 
 	return h;
 
@@ -1495,6 +1494,7 @@ static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
 			radix_tree_tag_clear(&fs_info->fs_roots_radix,
 					(unsigned long)root->root_key.objectid,
 					BTRFS_ROOT_TRANS_TAG);
+			btrfs_qgroup_free_meta_all_pertrans(root);
 			spin_unlock(&fs_info->fs_roots_radix_lock);
 
 			btrfs_free_log(trans, root);
@@ -1519,7 +1519,6 @@ static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
 			if (ret2)
 				return ret2;
 			spin_lock(&fs_info->fs_roots_radix_lock);
-			btrfs_qgroup_free_meta_all_pertrans(root);
 		}
 	}
 	spin_unlock(&fs_info->fs_roots_radix_lock);
diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
index 1340d77124ae..ee9caf7916fb 100644
--- a/fs/ceph/addr.c
+++ b/fs/ceph/addr.c
@@ -795,8 +795,10 @@ static int ceph_writepage(struct page *page, struct writeback_control *wbc)
 	ihold(inode);
 
 	if (wbc->sync_mode == WB_SYNC_NONE &&
-	    ceph_inode_to_fs_client(inode)->write_congested)
+	    ceph_inode_to_fs_client(inode)->write_congested) {
+		redirty_page_for_writepage(wbc, page);
 		return AOP_WRITEPAGE_ACTIVATE;
+	}
 
 	wait_on_page_fscache(page);
 
diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c
index 55051ad09c19..c4941ba245ac 100644
--- a/fs/ceph/caps.c
+++ b/fs/ceph/caps.c
@@ -4783,13 +4783,13 @@ int ceph_drop_caps_for_unlink(struct inode *inode)
 
 			doutc(mdsc->fsc->client, "%p %llx.%llx\n", inode,
 			      ceph_vinop(inode));
-			spin_lock(&mdsc->cap_unlink_delay_lock);
+			spin_lock(&mdsc->cap_delay_lock);
 			ci->i_ceph_flags |= CEPH_I_FLUSH;
 			if (!list_empty(&ci->i_cap_delay_list))
 				list_del_init(&ci->i_cap_delay_list);
 			list_add_tail(&ci->i_cap_delay_list,
 				      &mdsc->cap_unlink_delay_list);
-			spin_unlock(&mdsc->cap_unlink_delay_lock);
+			spin_unlock(&mdsc->cap_delay_lock);
 
 			/*
 			 * Fire the work immediately, because the MDS maybe
diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
index 3ab9c268a8bb..360b686c3c67 100644
--- a/fs/ceph/mds_client.c
+++ b/fs/ceph/mds_client.c
@@ -2504,7 +2504,7 @@ static void ceph_cap_unlink_work(struct work_struct *work)
 	struct ceph_client *cl = mdsc->fsc->client;
 
 	doutc(cl, "begin\n");
-	spin_lock(&mdsc->cap_unlink_delay_lock);
+	spin_lock(&mdsc->cap_delay_lock);
 	while (!list_empty(&mdsc->cap_unlink_delay_list)) {
 		struct ceph_inode_info *ci;
 		struct inode *inode;
@@ -2516,15 +2516,15 @@ static void ceph_cap_unlink_work(struct work_struct *work)
 
 		inode = igrab(&ci->netfs.inode);
 		if (inode) {
-			spin_unlock(&mdsc->cap_unlink_delay_lock);
+			spin_unlock(&mdsc->cap_delay_lock);
 			doutc(cl, "on %p %llx.%llx\n", inode,
 			      ceph_vinop(inode));
 			ceph_check_caps(ci, CHECK_CAPS_FLUSH);
 			iput(inode);
-			spin_lock(&mdsc->cap_unlink_delay_lock);
+			spin_lock(&mdsc->cap_delay_lock);
 		}
 	}
-	spin_unlock(&mdsc->cap_unlink_delay_lock);
+	spin_unlock(&mdsc->cap_delay_lock);
 	doutc(cl, "done\n");
 }
 
@@ -5404,7 +5404,6 @@ int ceph_mdsc_init(struct ceph_fs_client *fsc)
 	INIT_LIST_HEAD(&mdsc->cap_wait_list);
 	spin_lock_init(&mdsc->cap_delay_lock);
 	INIT_LIST_HEAD(&mdsc->cap_unlink_delay_list);
-	spin_lock_init(&mdsc->cap_unlink_delay_lock);
 	INIT_LIST_HEAD(&mdsc->snap_flush_list);
 	spin_lock_init(&mdsc->snap_flush_lock);
 	mdsc->last_cap_flush_tid = 1;
diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h
index 03f8ff00874f..b88e80415224 100644
--- a/fs/ceph/mds_client.h
+++ b/fs/ceph/mds_client.h
@@ -461,9 +461,8 @@ struct ceph_mds_client {
 	struct delayed_work    delayed_work;  /* delayed work */
 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
 	struct list_head cap_delay_list;   /* caps with delayed release */
-	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
 	struct list_head cap_unlink_delay_list;  /* caps with delayed release for unlink */
-	spinlock_t       cap_unlink_delay_lock;  /* protects cap_unlink_delay_list */
+	spinlock_t       cap_delay_lock;   /* protects cap_delay_list and cap_unlink_delay_list */
 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
 	spinlock_t       snap_flush_lock;
 
diff --git a/fs/cramfs/inode.c b/fs/cramfs/inode.c
index 39e75131fd5a..9901057a15ba 100644
--- a/fs/cramfs/inode.c
+++ b/fs/cramfs/inode.c
@@ -495,7 +495,7 @@ static void cramfs_kill_sb(struct super_block *sb)
 		sb->s_mtd = NULL;
 	} else if (IS_ENABLED(CONFIG_CRAMFS_BLOCKDEV) && sb->s_bdev) {
 		sync_blockdev(sb->s_bdev);
-		fput(sb->s_bdev_file);
+		bdev_fput(sb->s_bdev_file);
 	}
 	kfree(sbi);
 }
diff --git a/fs/ext4/super.c b/fs/ext4/super.c
index cfb8449c731f..044135796f2b 100644
--- a/fs/ext4/super.c
+++ b/fs/ext4/super.c
@@ -5668,7 +5668,7 @@ failed_mount:
 	brelse(sbi->s_sbh);
 	if (sbi->s_journal_bdev_file) {
 		invalidate_bdev(file_bdev(sbi->s_journal_bdev_file));
-		fput(sbi->s_journal_bdev_file);
+		bdev_fput(sbi->s_journal_bdev_file);
 	}
 out_fail:
 	invalidate_bdev(sb->s_bdev);
@@ -5913,7 +5913,7 @@ static struct file *ext4_get_journal_blkdev(struct super_block *sb,
 out_bh:
 	brelse(bh);
 out_bdev:
-	fput(bdev_file);
+	bdev_fput(bdev_file);
 	return ERR_PTR(errno);
 }
 
@@ -5952,7 +5952,7 @@ static journal_t *ext4_open_dev_journal(struct super_block *sb,
 out_journal:
 	jbd2_journal_destroy(journal);
 out_bdev:
-	fput(bdev_file);
+	bdev_fput(bdev_file);
 	return ERR_PTR(errno);
 }
 
@@ -7327,7 +7327,7 @@ static void ext4_kill_sb(struct super_block *sb)
 	kill_block_super(sb);
 
 	if (bdev_file)
-		fput(bdev_file);
+		bdev_fput(bdev_file);
 }
 
 static struct file_system_type ext4_fs_type = {
diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
index a6867f26f141..a4bc26dfdb1a 100644
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -1558,7 +1558,7 @@ static void destroy_device_list(struct f2fs_sb_info *sbi)
 
 	for (i = 0; i < sbi->s_ndevs; i++) {
 		if (i > 0)
-			fput(FDEV(i).bdev_file);
+			bdev_fput(FDEV(i).bdev_file);
 #ifdef CONFIG_BLK_DEV_ZONED
 		kvfree(FDEV(i).blkz_seq);
 #endif
diff --git a/fs/fuse/cuse.c b/fs/fuse/cuse.c
index b6cad106c37e..0b2da7b7e2ad 100644
--- a/fs/fuse/cuse.c
+++ b/fs/fuse/cuse.c
@@ -310,6 +310,10 @@ struct cuse_init_args {
 /**
  * cuse_process_init_reply - finish initializing CUSE channel
  *
+ * @fm: The fuse mount information containing the CUSE connection.
+ * @args: The arguments passed to the init reply.
+ * @error: The error code signifying if any error occurred during the process.
+ *
  * This function creates the character device and sets up all the
  * required data structures for it.  Please read the comment at the
  * top of this file for high level overview.
diff --git a/fs/fuse/dir.c b/fs/fuse/dir.c
index 4a6df591add6..2b0d4781f394 100644
--- a/fs/fuse/dir.c
+++ b/fs/fuse/dir.c
@@ -1321,6 +1321,7 @@ retry:
 			err = fuse_do_statx(inode, file, stat);
 			if (err == -ENOSYS) {
 				fc->no_statx = 1;
+				err = 0;
 				goto retry;
 			}
 		} else {
diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index a56e7bffd000..b57ce4157640 100644
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -1362,7 +1362,7 @@ static void fuse_dio_lock(struct kiocb *iocb, struct iov_iter *from,
 			  bool *exclusive)
 {
 	struct inode *inode = file_inode(iocb->ki_filp);
-	struct fuse_file *ff = iocb->ki_filp->private_data;
+	struct fuse_inode *fi = get_fuse_inode(inode);
 
 	*exclusive = fuse_dio_wr_exclusive_lock(iocb, from);
 	if (*exclusive) {
@@ -1377,7 +1377,7 @@ static void fuse_dio_lock(struct kiocb *iocb, struct iov_iter *from,
 		 * have raced, so check it again.
 		 */
 		if (fuse_io_past_eof(iocb, from) ||
-		    fuse_file_uncached_io_start(inode, ff, NULL) != 0) {
+		    fuse_inode_uncached_io_start(fi, NULL) != 0) {
 			inode_unlock_shared(inode);
 			inode_lock(inode);
 			*exclusive = true;
@@ -1388,13 +1388,13 @@ static void fuse_dio_lock(struct kiocb *iocb, struct iov_iter *from,
 static void fuse_dio_unlock(struct kiocb *iocb, bool exclusive)
 {
 	struct inode *inode = file_inode(iocb->ki_filp);
-	struct fuse_file *ff = iocb->ki_filp->private_data;
+	struct fuse_inode *fi = get_fuse_inode(inode);
 
 	if (exclusive) {
 		inode_unlock(inode);
 	} else {
 		/* Allow opens in caching mode after last parallel dio end */
-		fuse_file_uncached_io_end(inode, ff);
+		fuse_inode_uncached_io_end(fi);
 		inode_unlock_shared(inode);
 	}
 }
@@ -2574,8 +2574,10 @@ static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
 		 * First mmap of direct_io file enters caching inode io mode.
 		 * Also waits for parallel dio writers to go into serial mode
 		 * (exclusive instead of shared lock).
+		 * After first mmap, the inode stays in caching io mode until
+		 * the direct_io file release.
 		 */
-		rc = fuse_file_cached_io_start(inode, ff);
+		rc = fuse_file_cached_io_open(inode, ff);
 		if (rc)
 			return rc;
 	}
diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
index b24084b60864..f23919610313 100644
--- a/fs/fuse/fuse_i.h
+++ b/fs/fuse/fuse_i.h
@@ -1394,9 +1394,10 @@ int fuse_fileattr_set(struct mnt_idmap *idmap,
 		      struct dentry *dentry, struct fileattr *fa);
 
 /* iomode.c */
-int fuse_file_cached_io_start(struct inode *inode, struct fuse_file *ff);
-int fuse_file_uncached_io_start(struct inode *inode, struct fuse_file *ff, struct fuse_backing *fb);
-void fuse_file_uncached_io_end(struct inode *inode, struct fuse_file *ff);
+int fuse_file_cached_io_open(struct inode *inode, struct fuse_file *ff);
+int fuse_inode_uncached_io_start(struct fuse_inode *fi,
+				 struct fuse_backing *fb);
+void fuse_inode_uncached_io_end(struct fuse_inode *fi);
 
 int fuse_file_io_open(struct file *file, struct inode *inode);
 void fuse_file_io_release(struct fuse_file *ff, struct inode *inode);
diff --git a/fs/fuse/inode.c b/fs/fuse/inode.c
index 3a5d88878335..99e44ea7d875 100644
--- a/fs/fuse/inode.c
+++ b/fs/fuse/inode.c
@@ -175,6 +175,7 @@ static void fuse_evict_inode(struct inode *inode)
 		}
 	}
 	if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
+		WARN_ON(fi->iocachectr != 0);
 		WARN_ON(!list_empty(&fi->write_files));
 		WARN_ON(!list_empty(&fi->queued_writes));
 	}
diff --git a/fs/fuse/iomode.c b/fs/fuse/iomode.c
index c653ddcf0578..c99e285f3183 100644
--- a/fs/fuse/iomode.c
+++ b/fs/fuse/iomode.c
@@ -21,12 +21,13 @@ static inline bool fuse_is_io_cache_wait(struct fuse_inode *fi)
 }
 
 /*
- * Start cached io mode.
+ * Called on cached file open() and on first mmap() of direct_io file.
+ * Takes cached_io inode mode reference to be dropped on file release.
  *
  * Blocks new parallel dio writes and waits for the in-progress parallel dio
  * writes to complete.
  */
-int fuse_file_cached_io_start(struct inode *inode, struct fuse_file *ff)
+int fuse_file_cached_io_open(struct inode *inode, struct fuse_file *ff)
 {
 	struct fuse_inode *fi = get_fuse_inode(inode);
 
@@ -67,10 +68,9 @@ int fuse_file_cached_io_start(struct inode *inode, struct fuse_file *ff)
 	return 0;
 }
 
-static void fuse_file_cached_io_end(struct inode *inode, struct fuse_file *ff)
+static void fuse_file_cached_io_release(struct fuse_file *ff,
+					struct fuse_inode *fi)
 {
-	struct fuse_inode *fi = get_fuse_inode(inode);
-
 	spin_lock(&fi->lock);
 	WARN_ON(fi->iocachectr <= 0);
 	WARN_ON(ff->iomode != IOM_CACHED);
@@ -82,16 +82,15 @@ static void fuse_file_cached_io_end(struct inode *inode, struct fuse_file *ff)
 }
 
 /* Start strictly uncached io mode where cache access is not allowed */
-int fuse_file_uncached_io_start(struct inode *inode, struct fuse_file *ff, struct fuse_backing *fb)
+int fuse_inode_uncached_io_start(struct fuse_inode *fi, struct fuse_backing *fb)
 {
-	struct fuse_inode *fi = get_fuse_inode(inode);
 	struct fuse_backing *oldfb;
 	int err = 0;
 
 	spin_lock(&fi->lock);
 	/* deny conflicting backing files on same fuse inode */
 	oldfb = fuse_inode_backing(fi);
-	if (oldfb && oldfb != fb) {
+	if (fb && oldfb && oldfb != fb) {
 		err = -EBUSY;
 		goto unlock;
 	}
@@ -99,12 +98,10 @@ int fuse_file_uncached_io_start(struct inode *inode, struct fuse_file *ff, struc
 		err = -ETXTBSY;
 		goto unlock;
 	}
-	WARN_ON(ff->iomode != IOM_NONE);
 	fi->iocachectr--;
-	ff->iomode = IOM_UNCACHED;
 
 	/* fuse inode holds a single refcount of backing file */
-	if (!oldfb) {
+	if (fb && !oldfb) {
 		oldfb = fuse_inode_backing_set(fi, fb);
 		WARN_ON_ONCE(oldfb != NULL);
 	} else {
@@ -115,15 +112,29 @@ unlock:
 	return err;
 }
 
-void fuse_file_uncached_io_end(struct inode *inode, struct fuse_file *ff)
+/* Takes uncached_io inode mode reference to be dropped on file release */
+static int fuse_file_uncached_io_open(struct inode *inode,
+				      struct fuse_file *ff,
+				      struct fuse_backing *fb)
 {
 	struct fuse_inode *fi = get_fuse_inode(inode);
+	int err;
+
+	err = fuse_inode_uncached_io_start(fi, fb);
+	if (err)
+		return err;
+
+	WARN_ON(ff->iomode != IOM_NONE);
+	ff->iomode = IOM_UNCACHED;
+	return 0;
+}
+
+void fuse_inode_uncached_io_end(struct fuse_inode *fi)
+{
 	struct fuse_backing *oldfb = NULL;
 
 	spin_lock(&fi->lock);
 	WARN_ON(fi->iocachectr >= 0);
-	WARN_ON(ff->iomode != IOM_UNCACHED);
-	ff->iomode = IOM_NONE;
 	fi->iocachectr++;
 	if (!fi->iocachectr) {
 		wake_up(&fi->direct_io_waitq);
@@ -134,6 +145,15 @@ void fuse_file_uncached_io_end(struct inode *inode, struct fuse_file *ff)
 		fuse_backing_put(oldfb);
 }
 
+/* Drop uncached_io reference from passthrough open */
+static void fuse_file_uncached_io_release(struct fuse_file *ff,
+					  struct fuse_inode *fi)
+{
+	WARN_ON(ff->iomode != IOM_UNCACHED);
+	ff->iomode = IOM_NONE;
+	fuse_inode_uncached_io_end(fi);
+}
+
 /*
  * Open flags that are allowed in combination with FOPEN_PASSTHROUGH.
  * A combination of FOPEN_PASSTHROUGH and FOPEN_DIRECT_IO means that read/write
@@ -163,7 +183,7 @@ static int fuse_file_passthrough_open(struct inode *inode, struct file *file)
 		return PTR_ERR(fb);
 
 	/* First passthrough file open denies caching inode io mode */
-	err = fuse_file_uncached_io_start(inode, ff, fb);
+	err = fuse_file_uncached_io_open(inode, ff, fb);
 	if (!err)
 		return 0;
 
@@ -216,7 +236,7 @@ int fuse_file_io_open(struct file *file, struct inode *inode)
 	if (ff->open_flags & FOPEN_PASSTHROUGH)
 		err = fuse_file_passthrough_open(inode, file);
 	else
-		err = fuse_file_cached_io_start(inode, ff);
+		err = fuse_file_cached_io_open(inode, ff);
 	if (err)
 		goto fail;
 
@@ -236,8 +256,10 @@ fail:
 /* No more pending io and no new io possible to inode via open/mmapped file */
 void fuse_file_io_release(struct fuse_file *ff, struct inode *inode)
 {
+	struct fuse_inode *fi = get_fuse_inode(inode);
+
 	/*
-	 * Last parallel dio close allows caching inode io mode.
+	 * Last passthrough file close allows caching inode io mode.
 	 * Last caching file close exits caching inode io mode.
 	 */
 	switch (ff->iomode) {
@@ -245,10 +267,10 @@ void fuse_file_io_release(struct fuse_file *ff, struct inode *inode)
 		/* Nothing to do */
 		break;
 	case IOM_UNCACHED:
-		fuse_file_uncached_io_end(inode, ff);
+		fuse_file_uncached_io_release(ff, fi);
 		break;
 	case IOM_CACHED:
-		fuse_file_cached_io_end(inode, ff);
+		fuse_file_cached_io_release(ff, fi);
 		break;
 	}
 }
diff --git a/fs/jfs/jfs_logmgr.c b/fs/jfs/jfs_logmgr.c
index 73389c68e251..9609349e92e5 100644
--- a/fs/jfs/jfs_logmgr.c
+++ b/fs/jfs/jfs_logmgr.c
@@ -1141,7 +1141,7 @@ journal_found:
 	lbmLogShutdown(log);
 
       close:		/* close external log device */
-	fput(bdev_file);
+	bdev_fput(bdev_file);
 
       free:		/* free log descriptor */
 	mutex_unlock(&jfs_log_mutex);
@@ -1485,7 +1485,7 @@ int lmLogClose(struct super_block *sb)
 	bdev_file = log->bdev_file;
 	rc = lmLogShutdown(log);
 
-	fput(bdev_file);
+	bdev_fput(bdev_file);
 
 	kfree(log);
 
diff --git a/fs/kernfs/file.c b/fs/kernfs/file.c
index e9df2f87072c..8502ef68459b 100644
--- a/fs/kernfs/file.c
+++ b/fs/kernfs/file.c
@@ -636,11 +636,18 @@ static int kernfs_fop_open(struct inode *inode, struct file *file)
 	 * each file a separate locking class.  Let's differentiate on
 	 * whether the file has mmap or not for now.
 	 *
-	 * Both paths of the branch look the same.  They're supposed to
+	 * For similar reasons, writable and readonly files are given different
+	 * lockdep key, because the writable file /sys/power/resume may call vfs
+	 * lookup helpers for arbitrary paths and readonly files can be read by
+	 * overlayfs from vfs helpers when sysfs is a lower layer of overalyfs.
+	 *
+	 * All three cases look the same.  They're supposed to
 	 * look that way and give @of->mutex different static lockdep keys.
 	 */
 	if (has_mmap)
 		mutex_init(&of->mutex);
+	else if (file->f_mode & FMODE_WRITE)
+		mutex_init(&of->mutex);
 	else
 		mutex_init(&of->mutex);
 
diff --git a/fs/namei.c b/fs/namei.c
index ceb9ddf8dfdd..c5b2a25be7d0 100644
--- a/fs/namei.c
+++ b/fs/namei.c
@@ -4050,6 +4050,8 @@ retry:
 		case 0: case S_IFREG:
 			error = vfs_create(idmap, path.dentry->d_inode,
 					   dentry, mode, true);
+			if (!error)
+				security_path_post_mknod(idmap, dentry);
 			break;
 		case S_IFCHR: case S_IFBLK:
 			error = vfs_mknod(idmap, path.dentry->d_inode,
@@ -4060,11 +4062,6 @@ retry:
 					  dentry, mode, 0);
 			break;
 	}
-
-	if (error)
-		goto out2;
-
-	security_path_post_mknod(idmap, dentry);
 out2:
 	done_path_create(&path, dentry);
 	if (retry_estale(error, lookup_flags)) {
diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
index 2391ab3c3231..84d4093ca713 100644
--- a/fs/nfsd/nfs4state.c
+++ b/fs/nfsd/nfs4state.c
@@ -3042,12 +3042,9 @@ static void
 nfsd4_cb_recall_any_release(struct nfsd4_callback *cb)
 {
 	struct nfs4_client *clp = cb->cb_clp;
-	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 
-	spin_lock(&nn->client_lock);
 	clear_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
-	put_client_renew_locked(clp);
-	spin_unlock(&nn->client_lock);
+	drop_client(clp);
 }
 
 static int
@@ -6616,7 +6613,7 @@ deleg_reaper(struct nfsd_net *nn)
 		list_add(&clp->cl_ra_cblist, &cblist);
 
 		/* release in nfsd4_cb_recall_any_release */
-		atomic_inc(&clp->cl_rpc_users);
+		kref_get(&clp->cl_nfsdfs.cl_ref);
 		set_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
 		clp->cl_ra_time = ktime_get_boottime_seconds();
 	}
diff --git a/fs/nfsd/nfs4xdr.c b/fs/nfsd/nfs4xdr.c
index fac938f563ad..1955481832e0 100644
--- a/fs/nfsd/nfs4xdr.c
+++ b/fs/nfsd/nfs4xdr.c
@@ -3490,11 +3490,13 @@ nfsd4_encode_fattr4(struct svc_rqst *rqstp, struct xdr_stream *xdr,
 		    struct dentry *dentry, const u32 *bmval,
 		    int ignore_crossmnt)
 {
+	DECLARE_BITMAP(attr_bitmap, ARRAY_SIZE(nfsd4_enc_fattr4_encode_ops));
 	struct nfsd4_fattr_args args;
 	struct svc_fh *tempfh = NULL;
 	int starting_len = xdr->buf->len;
 	__be32 *attrlen_p, status;
 	int attrlen_offset;
+	u32 attrmask[3];
 	int err;
 	struct nfsd4_compoundres *resp = rqstp->rq_resp;
 	u32 minorversion = resp->cstate.minorversion;
@@ -3502,10 +3504,6 @@ nfsd4_encode_fattr4(struct svc_rqst *rqstp, struct xdr_stream *xdr,
 		.mnt	= exp->ex_path.mnt,
 		.dentry	= dentry,
 	};
-	union {
-		u32		attrmask[3];
-		unsigned long	mask[2];
-	} u;
 	unsigned long bit;
 	bool file_modified = false;
 	u64 size = 0;
@@ -3521,20 +3519,19 @@ nfsd4_encode_fattr4(struct svc_rqst *rqstp, struct xdr_stream *xdr,
 	/*
 	 * Make a local copy of the attribute bitmap that can be modified.
 	 */
-	memset(&u, 0, sizeof(u));
-	u.attrmask[0] = bmval[0];
-	u.attrmask[1] = bmval[1];
-	u.attrmask[2] = bmval[2];
+	attrmask[0] = bmval[0];
+	attrmask[1] = bmval[1];
+	attrmask[2] = bmval[2];
 
 	args.rdattr_err = 0;
 	if (exp->ex_fslocs.migrated) {
-		status = fattr_handle_absent_fs(&u.attrmask[0], &u.attrmask[1],
-						&u.attrmask[2], &args.rdattr_err);
+		status = fattr_handle_absent_fs(&attrmask[0], &attrmask[1],
+						&attrmask[2], &args.rdattr_err);
 		if (status)
 			goto out;
 	}
 	args.size = 0;
-	if (u.attrmask[0] & (FATTR4_WORD0_CHANGE | FATTR4_WORD0_SIZE)) {
+	if (attrmask[0] & (FATTR4_WORD0_CHANGE | FATTR4_WORD0_SIZE)) {
 		status = nfsd4_deleg_getattr_conflict(rqstp, d_inode(dentry),
 					&file_modified, &size);
 		if (status)
@@ -3553,16 +3550,16 @@ nfsd4_encode_fattr4(struct svc_rqst *rqstp, struct xdr_stream *xdr,
 
 	if (!(args.stat.result_mask & STATX_BTIME))
 		/* underlying FS does not offer btime so we can't share it */
-		u.attrmask[1] &= ~FATTR4_WORD1_TIME_CREATE;
-	if ((u.attrmask[0] & (FATTR4_WORD0_FILES_AVAIL | FATTR4_WORD0_FILES_FREE |
+		attrmask[1] &= ~FATTR4_WORD1_TIME_CREATE;
+	if ((attrmask[0] & (FATTR4_WORD0_FILES_AVAIL | FATTR4_WORD0_FILES_FREE |
 			FATTR4_WORD0_FILES_TOTAL | FATTR4_WORD0_MAXNAME)) ||
-	    (u.attrmask[1] & (FATTR4_WORD1_SPACE_AVAIL | FATTR4_WORD1_SPACE_FREE |
+	    (attrmask[1] & (FATTR4_WORD1_SPACE_AVAIL | FATTR4_WORD1_SPACE_FREE |
 		       FATTR4_WORD1_SPACE_TOTAL))) {
 		err = vfs_statfs(&path, &args.statfs);
 		if (err)
 			goto out_nfserr;
 	}
-	if ((u.attrmask[0] & (FATTR4_WORD0_FILEHANDLE | FATTR4_WORD0_FSID)) &&
+	if ((attrmask[0] & (FATTR4_WORD0_FILEHANDLE | FATTR4_WORD0_FSID)) &&
 	    !fhp) {
 		tempfh = kmalloc(sizeof(struct svc_fh), GFP_KERNEL);
 		status = nfserr_jukebox;
@@ -3577,10 +3574,10 @@ nfsd4_encode_fattr4(struct svc_rqst *rqstp, struct xdr_stream *xdr,
 		args.fhp = fhp;
 
 	args.acl = NULL;
-	if (u.attrmask[0] & FATTR4_WORD0_ACL) {
+	if (attrmask[0] & FATTR4_WORD0_ACL) {
 		err = nfsd4_get_nfs4_acl(rqstp, dentry, &args.acl);
 		if (err == -EOPNOTSUPP)
-			u.attrmask[0] &= ~FATTR4_WORD0_ACL;
+			attrmask[0] &= ~FATTR4_WORD0_ACL;
 		else if (err == -EINVAL) {
 			status = nfserr_attrnotsupp;
 			goto out;
@@ -3592,17 +3589,17 @@ nfsd4_encode_fattr4(struct svc_rqst *rqstp, struct xdr_stream *xdr,
 
 #ifdef CONFIG_NFSD_V4_SECURITY_LABEL
 	args.context = NULL;
-	if ((u.attrmask[2] & FATTR4_WORD2_SECURITY_LABEL) ||
-	     u.attrmask[0] & FATTR4_WORD0_SUPPORTED_ATTRS) {
+	if ((attrmask[2] & FATTR4_WORD2_SECURITY_LABEL) ||
+	     attrmask[0] & FATTR4_WORD0_SUPPORTED_ATTRS) {
 		if (exp->ex_flags & NFSEXP_SECURITY_LABEL)
 			err = security_inode_getsecctx(d_inode(dentry),
 						&args.context, &args.contextlen);
 		else
 			err = -EOPNOTSUPP;
 		args.contextsupport = (err == 0);
-		if (u.attrmask[2] & FATTR4_WORD2_SECURITY_LABEL) {
+		if (attrmask[2] & FATTR4_WORD2_SECURITY_LABEL) {
 			if (err == -EOPNOTSUPP)
-				u.attrmask[2] &= ~FATTR4_WORD2_SECURITY_LABEL;
+				attrmask[2] &= ~FATTR4_WORD2_SECURITY_LABEL;
 			else if (err)
 				goto out_nfserr;
 		}
@@ -3610,8 +3607,8 @@ nfsd4_encode_fattr4(struct svc_rqst *rqstp, struct xdr_stream *xdr,
 #endif /* CONFIG_NFSD_V4_SECURITY_LABEL */
 
 	/* attrmask */
-	status = nfsd4_encode_bitmap4(xdr, u.attrmask[0],
-				      u.attrmask[1], u.attrmask[2]);
+	status = nfsd4_encode_bitmap4(xdr, attrmask[0], attrmask[1],
+				      attrmask[2]);
 	if (status)
 		goto out;
 
@@ -3620,7 +3617,9 @@ nfsd4_encode_fattr4(struct svc_rqst *rqstp, struct xdr_stream *xdr,
 	attrlen_p = xdr_reserve_space(xdr, XDR_UNIT);
 	if (!attrlen_p)
 		goto out_resource;
-	for_each_set_bit(bit, (const unsigned long *)&u.mask,
+	bitmap_from_arr32(attr_bitmap, attrmask,
+			  ARRAY_SIZE(nfsd4_enc_fattr4_encode_ops));
+	for_each_set_bit(bit, attr_bitmap,
 			 ARRAY_SIZE(nfsd4_enc_fattr4_encode_ops)) {
 		status = nfsd4_enc_fattr4_encode_ops[bit](xdr, &args);
 		if (status != nfs_ok)
diff --git a/fs/nilfs2/dir.c b/fs/nilfs2/dir.c
index bc846b904b68..aee40db7a036 100644
--- a/fs/nilfs2/dir.c
+++ b/fs/nilfs2/dir.c
@@ -240,7 +240,7 @@ nilfs_filetype_table[NILFS_FT_MAX] = {
 
 #define S_SHIFT 12
 static unsigned char
-nilfs_type_by_mode[S_IFMT >> S_SHIFT] = {
+nilfs_type_by_mode[(S_IFMT >> S_SHIFT) + 1] = {
 	[S_IFREG >> S_SHIFT]	= NILFS_FT_REG_FILE,
 	[S_IFDIR >> S_SHIFT]	= NILFS_FT_DIR,
 	[S_IFCHR >> S_SHIFT]	= NILFS_FT_CHRDEV,
diff --git a/fs/proc/bootconfig.c b/fs/proc/bootconfig.c
index 902b326e1e56..87dcaae32ff8 100644
--- a/fs/proc/bootconfig.c
+++ b/fs/proc/bootconfig.c
@@ -62,12 +62,12 @@ static int __init copy_xbc_key_value_list(char *dst, size_t size)
 				break;
 			dst += ret;
 		}
-		if (ret >= 0 && boot_command_line[0]) {
-			ret = snprintf(dst, rest(dst, end), "# Parameters from bootloader:\n# %s\n",
-				       boot_command_line);
-			if (ret > 0)
-				dst += ret;
-		}
+	}
+	if (cmdline_has_extra_options() && ret >= 0 && boot_command_line[0]) {
+		ret = snprintf(dst, rest(dst, end), "# Parameters from bootloader:\n# %s\n",
+			       boot_command_line);
+		if (ret > 0)
+			dst += ret;
 	}
 out:
 	kfree(key);
diff --git a/fs/reiserfs/journal.c b/fs/reiserfs/journal.c
index 6474529c4253..e539ccd39e1e 100644
--- a/fs/reiserfs/journal.c
+++ b/fs/reiserfs/journal.c
@@ -2589,7 +2589,7 @@ static void journal_list_init(struct super_block *sb)
 static void release_journal_dev(struct reiserfs_journal *journal)
 {
 	if (journal->j_bdev_file) {
-		fput(journal->j_bdev_file);
+		bdev_fput(journal->j_bdev_file);
 		journal->j_bdev_file = NULL;
 	}
 }
diff --git a/fs/romfs/super.c b/fs/romfs/super.c
index 2be227532f39..2cbb92462074 100644
--- a/fs/romfs/super.c
+++ b/fs/romfs/super.c
@@ -594,7 +594,7 @@ static void romfs_kill_sb(struct super_block *sb)
 #ifdef CONFIG_ROMFS_ON_BLOCK
 	if (sb->s_bdev) {
 		sync_blockdev(sb->s_bdev);
-		fput(sb->s_bdev_file);
+		bdev_fput(sb->s_bdev_file);
 	}
 #endif
 }
diff --git a/fs/smb/client/cached_dir.c b/fs/smb/client/cached_dir.c
index a0017724d523..0ff2491c311d 100644
--- a/fs/smb/client/cached_dir.c
+++ b/fs/smb/client/cached_dir.c
@@ -417,6 +417,7 @@ smb2_close_cached_fid(struct kref *ref)
 {
 	struct cached_fid *cfid = container_of(ref, struct cached_fid,
 					       refcount);
+	int rc;
 
 	spin_lock(&cfid->cfids->cfid_list_lock);
 	if (cfid->on_list) {
@@ -430,9 +431,10 @@ smb2_close_cached_fid(struct kref *ref)
 	cfid->dentry = NULL;
 
 	if (cfid->is_open) {
-		SMB2_close(0, cfid->tcon, cfid->fid.persistent_fid,
+		rc = SMB2_close(0, cfid->tcon, cfid->fid.persistent_fid,
 			   cfid->fid.volatile_fid);
-		atomic_dec(&cfid->tcon->num_remote_opens);
+		if (rc) /* should we retry on -EBUSY or -EAGAIN? */
+			cifs_dbg(VFS, "close cached dir rc %d\n", rc);
 	}
 
 	free_cached_dir(cfid);
diff --git a/fs/smb/client/cifs_debug.c b/fs/smb/client/cifs_debug.c
index 226d4835c92d..c71ae5c04306 100644
--- a/fs/smb/client/cifs_debug.c
+++ b/fs/smb/client/cifs_debug.c
@@ -250,6 +250,8 @@ static int cifs_debug_files_proc_show(struct seq_file *m, void *v)
 	spin_lock(&cifs_tcp_ses_lock);
 	list_for_each_entry(server, &cifs_tcp_ses_list, tcp_ses_list) {
 		list_for_each_entry(ses, &server->smb_ses_list, smb_ses_list) {
+			if (cifs_ses_exiting(ses))
+				continue;
 			list_for_each_entry(tcon, &ses->tcon_list, tcon_list) {
 				spin_lock(&tcon->open_file_lock);
 				list_for_each_entry(cfile, &tcon->openFileList, tlist) {
@@ -676,6 +678,8 @@ static ssize_t cifs_stats_proc_write(struct file *file,
 			}
 #endif /* CONFIG_CIFS_STATS2 */
 			list_for_each_entry(ses, &server->smb_ses_list, smb_ses_list) {
+				if (cifs_ses_exiting(ses))
+					continue;
 				list_for_each_entry(tcon, &ses->tcon_list, tcon_list) {
 					atomic_set(&tcon->num_smbs_sent, 0);
 					spin_lock(&tcon->stat_lock);
@@ -755,6 +759,8 @@ static int cifs_stats_proc_show(struct seq_file *m, void *v)
 			}
 #endif /* STATS2 */
 		list_for_each_entry(ses, &server->smb_ses_list, smb_ses_list) {
+			if (cifs_ses_exiting(ses))
+				continue;
 			list_for_each_entry(tcon, &ses->tcon_list, tcon_list) {
 				i++;
 				seq_printf(m, "\n%d) %s", i, tcon->tree_name);
diff --git a/fs/smb/client/cifsfs.c b/fs/smb/client/cifsfs.c
index aa6f1ecb7c0e..d41eedbff674 100644
--- a/fs/smb/client/cifsfs.c
+++ b/fs/smb/client/cifsfs.c
@@ -156,6 +156,7 @@ struct workqueue_struct	*decrypt_wq;
 struct workqueue_struct	*fileinfo_put_wq;
 struct workqueue_struct	*cifsoplockd_wq;
 struct workqueue_struct	*deferredclose_wq;
+struct workqueue_struct	*serverclose_wq;
 __u32 cifs_lock_secret;
 
 /*
@@ -1888,6 +1889,13 @@ init_cifs(void)
 		goto out_destroy_cifsoplockd_wq;
 	}
 
+	serverclose_wq = alloc_workqueue("serverclose",
+					   WQ_FREEZABLE|WQ_MEM_RECLAIM, 0);
+	if (!serverclose_wq) {
+		rc = -ENOMEM;
+		goto out_destroy_serverclose_wq;
+	}
+
 	rc = cifs_init_inodecache();
 	if (rc)
 		goto out_destroy_deferredclose_wq;
@@ -1962,6 +1970,8 @@ out_destroy_decrypt_wq:
 	destroy_workqueue(decrypt_wq);
 out_destroy_cifsiod_wq:
 	destroy_workqueue(cifsiod_wq);
+out_destroy_serverclose_wq:
+	destroy_workqueue(serverclose_wq);
 out_clean_proc:
 	cifs_proc_clean();
 	return rc;
@@ -1991,6 +2001,7 @@ exit_cifs(void)
 	destroy_workqueue(cifsoplockd_wq);
 	destroy_workqueue(decrypt_wq);
 	destroy_workqueue(fileinfo_put_wq);
+	destroy_workqueue(serverclose_wq);
 	destroy_workqueue(cifsiod_wq);
 	cifs_proc_clean();
 }
diff --git a/fs/smb/client/cifsglob.h b/fs/smb/client/cifsglob.h
index 7ed9d05f6890..d6669ce4ae87 100644
--- a/fs/smb/client/cifsglob.h
+++ b/fs/smb/client/cifsglob.h
@@ -442,10 +442,10 @@ struct smb_version_operations {
 	/* set fid protocol-specific info */
 	void (*set_fid)(struct cifsFileInfo *, struct cifs_fid *, __u32);
 	/* close a file */
-	void (*close)(const unsigned int, struct cifs_tcon *,
+	int (*close)(const unsigned int, struct cifs_tcon *,
 		      struct cifs_fid *);
 	/* close a file, returning file attributes and timestamps */
-	void (*close_getattr)(const unsigned int xid, struct cifs_tcon *tcon,
+	int (*close_getattr)(const unsigned int xid, struct cifs_tcon *tcon,
 		      struct cifsFileInfo *pfile_info);
 	/* send a flush request to the server */
 	int (*flush)(const unsigned int, struct cifs_tcon *, struct cifs_fid *);
@@ -1077,6 +1077,7 @@ struct cifs_ses {
 				   and after mount option parsing we fill it */
 	char *domainName;
 	char *password;
+	char *password2; /* When key rotation used, new password may be set before it expires */
 	char workstation_name[CIFS_MAX_WORKSTATION_LEN];
 	struct session_key auth_key;
 	struct ntlmssp_auth *ntlmssp; /* ciphertext, flags, server challenge */
@@ -1281,7 +1282,6 @@ struct cifs_tcon {
 	struct cached_fids *cfids;
 	/* BB add field for back pointer to sb struct(s)? */
 #ifdef CONFIG_CIFS_DFS_UPCALL
-	struct list_head dfs_ses_list;
 	struct delayed_work dfs_cache_work;
 #endif
 	struct delayed_work	query_interfaces; /* query interfaces workqueue job */
@@ -1440,6 +1440,7 @@ struct cifsFileInfo {
 	bool swapfile:1;
 	bool oplock_break_cancelled:1;
 	bool status_file_deleted:1; /* file has been deleted */
+	bool offload:1; /* offload final part of _put to a wq */
 	unsigned int oplock_epoch; /* epoch from the lease break */
 	__u32 oplock_level; /* oplock/lease level from the lease break */
 	int count;
@@ -1448,6 +1449,7 @@ struct cifsFileInfo {
 	struct cifs_search_info srch_inf;
 	struct work_struct oplock_break; /* work for oplock breaks */
 	struct work_struct put; /* work for the final part of _put */
+	struct work_struct serverclose; /* work for serverclose */
 	struct delayed_work deferred;
 	bool deferred_close_scheduled; /* Flag to indicate close is scheduled */
 	char *symlink_target;
@@ -1804,7 +1806,6 @@ struct cifs_mount_ctx {
 	struct TCP_Server_Info *server;
 	struct cifs_ses *ses;
 	struct cifs_tcon *tcon;
-	struct list_head dfs_ses_list;
 };
 
 static inline void __free_dfs_info_param(struct dfs_info3_param *param)
@@ -2105,6 +2106,7 @@ extern struct workqueue_struct *decrypt_wq;
 extern struct workqueue_struct *fileinfo_put_wq;
 extern struct workqueue_struct *cifsoplockd_wq;
 extern struct workqueue_struct *deferredclose_wq;
+extern struct workqueue_struct *serverclose_wq;
 extern __u32 cifs_lock_secret;
 
 extern mempool_t *cifs_sm_req_poolp;
@@ -2324,4 +2326,14 @@ struct smb2_compound_vars {
 	struct kvec ea_iov;
 };
 
+static inline bool cifs_ses_exiting(struct cifs_ses *ses)
+{
+	bool ret;
+
+	spin_lock(&ses->ses_lock);
+	ret = ses->ses_status == SES_EXITING;
+	spin_unlock(&ses->ses_lock);
+	return ret;
+}
+
 #endif	/* _CIFS_GLOB_H */
diff --git a/fs/smb/client/cifsproto.h b/fs/smb/client/cifsproto.h
index 0723e1b57256..8e0a348f1f66 100644
--- a/fs/smb/client/cifsproto.h
+++ b/fs/smb/client/cifsproto.h
@@ -725,31 +725,31 @@ struct super_block *cifs_get_tcon_super(struct cifs_tcon *tcon);
 void cifs_put_tcon_super(struct super_block *sb);
 int cifs_wait_for_server_reconnect(struct TCP_Server_Info *server, bool retry);
 
-/* Put references of @ses and @ses->dfs_root_ses */
+/* Put references of @ses and its children */
 static inline void cifs_put_smb_ses(struct cifs_ses *ses)
 {
-	struct cifs_ses *rses = ses->dfs_root_ses;
+	struct cifs_ses *next;
 
-	__cifs_put_smb_ses(ses);
-	if (rses)
-		__cifs_put_smb_ses(rses);
+	do {
+		next = ses->dfs_root_ses;
+		__cifs_put_smb_ses(ses);
+	} while ((ses = next));
 }
 
-/* Get an active reference of @ses and @ses->dfs_root_ses.
+/* Get an active reference of @ses and its children.
  *
  * NOTE: make sure to call this function when incrementing reference count of
  * @ses to ensure that any DFS root session attached to it (@ses->dfs_root_ses)
  * will also get its reference count incremented.
  *
- * cifs_put_smb_ses() will put both references, so call it when you're done.
+ * cifs_put_smb_ses() will put all references, so call it when you're done.
  */
 static inline void cifs_smb_ses_inc_refcount(struct cifs_ses *ses)
 {
 	lockdep_assert_held(&cifs_tcp_ses_lock);
 
-	ses->ses_count++;
-	if (ses->dfs_root_ses)
-		ses->dfs_root_ses->ses_count++;
+	for (; ses; ses = ses->dfs_root_ses)
+		ses->ses_count++;
 }
 
 static inline bool dfs_src_pathname_equal(const char *s1, const char *s2)
diff --git a/fs/smb/client/cifssmb.c b/fs/smb/client/cifssmb.c
index 5aee55551573..23b5709ddc31 100644
--- a/fs/smb/client/cifssmb.c
+++ b/fs/smb/client/cifssmb.c
@@ -5854,10 +5854,8 @@ SetEARetry:
 	parm_data->list.EA_flags = 0;
 	/* we checked above that name len is less than 255 */
 	parm_data->list.name_len = (__u8)name_len;
-	/* EA names are always ASCII */
-	if (ea_name)
-		strncpy(parm_data->list.name, ea_name, name_len);
-	parm_data->list.name[name_len] = '\0';
+	/* EA names are always ASCII and NUL-terminated */
+	strscpy(parm_data->list.name, ea_name ?: "", name_len + 1);
 	parm_data->list.value_len = cpu_to_le16(ea_value_len);
 	/* caller ensures that ea_value_len is less than 64K but
 	we need to ensure that it fits within the smb */
diff --git a/fs/smb/client/connect.c b/fs/smb/client/connect.c
index 9b85b5341822..4e35970681bf 100644
--- a/fs/smb/client/connect.c
+++ b/fs/smb/client/connect.c
@@ -175,6 +175,8 @@ cifs_signal_cifsd_for_reconnect(struct TCP_Server_Info *server,
 
 	spin_lock(&cifs_tcp_ses_lock);
 	list_for_each_entry(ses, &pserver->smb_ses_list, smb_ses_list) {
+		if (cifs_ses_exiting(ses))
+			continue;
 		spin_lock(&ses->chan_lock);
 		for (i = 0; i < ses->chan_count; i++) {
 			if (!ses->chans[i].server)
@@ -232,7 +234,13 @@ cifs_mark_tcp_ses_conns_for_reconnect(struct TCP_Server_Info *server,
 
 	spin_lock(&cifs_tcp_ses_lock);
 	list_for_each_entry_safe(ses, nses, &pserver->smb_ses_list, smb_ses_list) {
-		/* check if iface is still active */
+		spin_lock(&ses->ses_lock);
+		if (ses->ses_status == SES_EXITING) {
+			spin_unlock(&ses->ses_lock);
+			continue;
+		}
+		spin_unlock(&ses->ses_lock);
+
 		spin_lock(&ses->chan_lock);
 		if (cifs_ses_get_chan_index(ses, server) ==
 		    CIFS_INVAL_CHAN_INDEX) {
@@ -1860,6 +1868,9 @@ static int match_session(struct cifs_ses *ses, struct smb3_fs_context *ctx)
 	    ctx->sectype != ses->sectype)
 		return 0;
 
+	if (ctx->dfs_root_ses != ses->dfs_root_ses)
+		return 0;
+
 	/*
 	 * If an existing session is limited to less channels than
 	 * requested, it should not be reused
@@ -1963,31 +1974,6 @@ out:
 	return rc;
 }
 
-/**
- * cifs_free_ipc - helper to release the session IPC tcon
- * @ses: smb session to unmount the IPC from
- *
- * Needs to be called everytime a session is destroyed.
- *
- * On session close, the IPC is closed and the server must release all tcons of the session.
- * No need to send a tree disconnect here.
- *
- * Besides, it will make the server to not close durable and resilient files on session close, as
- * specified in MS-SMB2 3.3.5.6 Receiving an SMB2 LOGOFF Request.
- */
-static int
-cifs_free_ipc(struct cifs_ses *ses)
-{
-	struct cifs_tcon *tcon = ses->tcon_ipc;
-
-	if (tcon == NULL)
-		return 0;
-
-	tconInfoFree(tcon);
-	ses->tcon_ipc = NULL;
-	return 0;
-}
-
 static struct cifs_ses *
 cifs_find_smb_ses(struct TCP_Server_Info *server, struct smb3_fs_context *ctx)
 {
@@ -2019,48 +2005,52 @@ cifs_find_smb_ses(struct TCP_Server_Info *server, struct smb3_fs_context *ctx)
 void __cifs_put_smb_ses(struct cifs_ses *ses)
 {
 	struct TCP_Server_Info *server = ses->server;
+	struct cifs_tcon *tcon;
 	unsigned int xid;
 	size_t i;
+	bool do_logoff;
 	int rc;
 
+	spin_lock(&cifs_tcp_ses_lock);
 	spin_lock(&ses->ses_lock);
-	if (ses->ses_status == SES_EXITING) {
+	cifs_dbg(FYI, "%s: id=0x%llx ses_count=%d ses_status=%u ipc=%s\n",
+		 __func__, ses->Suid, ses->ses_count, ses->ses_status,
+		 ses->tcon_ipc ? ses->tcon_ipc->tree_name : "none");
+	if (ses->ses_status == SES_EXITING || --ses->ses_count > 0) {
 		spin_unlock(&ses->ses_lock);
+		spin_unlock(&cifs_tcp_ses_lock);
 		return;
 	}
-	spin_unlock(&ses->ses_lock);
+	/* ses_count can never go negative */
+	WARN_ON(ses->ses_count < 0);
 
-	cifs_dbg(FYI, "%s: ses_count=%d\n", __func__, ses->ses_count);
-	cifs_dbg(FYI,
-		 "%s: ses ipc: %s\n", __func__, ses->tcon_ipc ? ses->tcon_ipc->tree_name : "NONE");
+	spin_lock(&ses->chan_lock);
+	cifs_chan_clear_need_reconnect(ses, server);
+	spin_unlock(&ses->chan_lock);
 
-	spin_lock(&cifs_tcp_ses_lock);
-	if (--ses->ses_count > 0) {
-		spin_unlock(&cifs_tcp_ses_lock);
-		return;
-	}
-	spin_lock(&ses->ses_lock);
-	if (ses->ses_status == SES_GOOD)
-		ses->ses_status = SES_EXITING;
+	do_logoff = ses->ses_status == SES_GOOD && server->ops->logoff;
+	ses->ses_status = SES_EXITING;
+	tcon = ses->tcon_ipc;
+	ses->tcon_ipc = NULL;
 	spin_unlock(&ses->ses_lock);
 	spin_unlock(&cifs_tcp_ses_lock);
 
-	/* ses_count can never go negative */
-	WARN_ON(ses->ses_count < 0);
-
-	spin_lock(&ses->ses_lock);
-	if (ses->ses_status == SES_EXITING && server->ops->logoff) {
-		spin_unlock(&ses->ses_lock);
-		cifs_free_ipc(ses);
+	/*
+	 * On session close, the IPC is closed and the server must release all
+	 * tcons of the session.  No need to send a tree disconnect here.
+	 *
+	 * Besides, it will make the server to not close durable and resilient
+	 * files on session close, as specified in MS-SMB2 3.3.5.6 Receiving an
+	 * SMB2 LOGOFF Request.
+	 */
+	tconInfoFree(tcon);
+	if (do_logoff) {
 		xid = get_xid();
 		rc = server->ops->logoff(xid, ses);
 		if (rc)
 			cifs_server_dbg(VFS, "%s: Session Logoff failure rc=%d\n",
 				__func__, rc);
 		_free_xid(xid);
-	} else {
-		spin_unlock(&ses->ses_lock);
-		cifs_free_ipc(ses);
 	}
 
 	spin_lock(&cifs_tcp_ses_lock);
@@ -2193,6 +2183,7 @@ cifs_set_cifscreds(struct smb3_fs_context *ctx, struct cifs_ses *ses)
 	}
 
 	++delim;
+	/* BB consider adding support for password2 (Key Rotation) for multiuser in future */
 	ctx->password = kstrndup(delim, len, GFP_KERNEL);
 	if (!ctx->password) {
 		cifs_dbg(FYI, "Unable to allocate %zd bytes for password\n",
@@ -2216,6 +2207,7 @@ cifs_set_cifscreds(struct smb3_fs_context *ctx, struct cifs_ses *ses)
 			kfree(ctx->username);
 			ctx->username = NULL;
 			kfree_sensitive(ctx->password);
+			/* no need to free ctx->password2 since not allocated in this path */
 			ctx->password = NULL;
 			goto out_key_put;
 		}
@@ -2327,6 +2319,12 @@ cifs_get_smb_ses(struct TCP_Server_Info *server, struct smb3_fs_context *ctx)
 		if (!ses->password)
 			goto get_ses_fail;
 	}
+	/* ctx->password freed at unmount */
+	if (ctx->password2) {
+		ses->password2 = kstrdup(ctx->password2, GFP_KERNEL);
+		if (!ses->password2)
+			goto get_ses_fail;
+	}
 	if (ctx->domainname) {
 		ses->domainName = kstrdup(ctx->domainname, GFP_KERNEL);
 		if (!ses->domainName)
@@ -2373,9 +2371,9 @@ cifs_get_smb_ses(struct TCP_Server_Info *server, struct smb3_fs_context *ctx)
 	 * need to lock before changing something in the session.
 	 */
 	spin_lock(&cifs_tcp_ses_lock);
+	if (ctx->dfs_root_ses)
+		cifs_smb_ses_inc_refcount(ctx->dfs_root_ses);
 	ses->dfs_root_ses = ctx->dfs_root_ses;
-	if (ses->dfs_root_ses)
-		ses->dfs_root_ses->ses_count++;
 	list_add(&ses->smb_ses_list, &server->smb_ses_list);
 	spin_unlock(&cifs_tcp_ses_lock);
 
@@ -3326,6 +3324,9 @@ void cifs_mount_put_conns(struct cifs_mount_ctx *mnt_ctx)
 		cifs_put_smb_ses(mnt_ctx->ses);
 	else if (mnt_ctx->server)
 		cifs_put_tcp_session(mnt_ctx->server, 0);
+	mnt_ctx->ses = NULL;
+	mnt_ctx->tcon = NULL;
+	mnt_ctx->server = NULL;
 	mnt_ctx->cifs_sb->mnt_cifs_flags &= ~CIFS_MOUNT_POSIX_PATHS;
 	free_xid(mnt_ctx->xid);
 }
@@ -3604,8 +3605,6 @@ int cifs_mount(struct cifs_sb_info *cifs_sb, struct smb3_fs_context *ctx)
 	bool isdfs;
 	int rc;
 
-	INIT_LIST_HEAD(&mnt_ctx.dfs_ses_list);
-
 	rc = dfs_mount_share(&mnt_ctx, &isdfs);
 	if (rc)
 		goto error;
@@ -3636,7 +3635,6 @@ out:
 	return rc;
 
 error:
-	dfs_put_root_smb_sessions(&mnt_ctx.dfs_ses_list);
 	cifs_mount_put_conns(&mnt_ctx);
 	return rc;
 }
@@ -3651,6 +3649,18 @@ int cifs_mount(struct cifs_sb_info *cifs_sb, struct smb3_fs_context *ctx)
 		goto error;
 
 	rc = cifs_mount_get_tcon(&mnt_ctx);
+	if (!rc) {
+		/*
+		 * Prevent superblock from being created with any missing
+		 * connections.
+		 */
+		if (WARN_ON(!mnt_ctx.server))
+			rc = -EHOSTDOWN;
+		else if (WARN_ON(!mnt_ctx.ses))
+			rc = -EACCES;
+		else if (WARN_ON(!mnt_ctx.tcon))
+			rc = -ENOENT;
+	}
 	if (rc)
 		goto error;
 
@@ -3988,13 +3998,14 @@ cifs_set_vol_auth(struct smb3_fs_context *ctx, struct cifs_ses *ses)
 }
 
 static struct cifs_tcon *
-cifs_construct_tcon(struct cifs_sb_info *cifs_sb, kuid_t fsuid)
+__cifs_construct_tcon(struct cifs_sb_info *cifs_sb, kuid_t fsuid)
 {
 	int rc;
 	struct cifs_tcon *master_tcon = cifs_sb_master_tcon(cifs_sb);
 	struct cifs_ses *ses;
 	struct cifs_tcon *tcon = NULL;
 	struct smb3_fs_context *ctx;
+	char *origin_fullpath = NULL;
 
 	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
 	if (ctx == NULL)
@@ -4018,6 +4029,7 @@ cifs_construct_tcon(struct cifs_sb_info *cifs_sb, kuid_t fsuid)
 	ctx->sign = master_tcon->ses->sign;
 	ctx->seal = master_tcon->seal;
 	ctx->witness = master_tcon->use_witness;
+	ctx->dfs_root_ses = master_tcon->ses->dfs_root_ses;
 
 	rc = cifs_set_vol_auth(ctx, master_tcon->ses);
 	if (rc) {
@@ -4037,12 +4049,39 @@ cifs_construct_tcon(struct cifs_sb_info *cifs_sb, kuid_t fsuid)
 		goto out;
 	}
 
+#ifdef CONFIG_CIFS_DFS_UPCALL
+	spin_lock(&master_tcon->tc_lock);
+	if (master_tcon->origin_fullpath) {
+		spin_unlock(&master_tcon->tc_lock);
+		origin_fullpath = dfs_get_path(cifs_sb, cifs_sb->ctx->source);
+		if (IS_ERR(origin_fullpath)) {
+			tcon = ERR_CAST(origin_fullpath);
+			origin_fullpath = NULL;
+			cifs_put_smb_ses(ses);
+			goto out;
+		}
+	} else {
+		spin_unlock(&master_tcon->tc_lock);
+	}
+#endif
+
 	tcon = cifs_get_tcon(ses, ctx);
 	if (IS_ERR(tcon)) {
 		cifs_put_smb_ses(ses);
 		goto out;
 	}
 
+#ifdef CONFIG_CIFS_DFS_UPCALL
+	if (origin_fullpath) {
+		spin_lock(&tcon->tc_lock);
+		tcon->origin_fullpath = origin_fullpath;
+		spin_unlock(&tcon->tc_lock);
+		origin_fullpath = NULL;
+		queue_delayed_work(dfscache_wq, &tcon->dfs_cache_work,
+				   dfs_cache_get_ttl() * HZ);
+	}
+#endif
+
 #ifdef CONFIG_CIFS_ALLOW_INSECURE_LEGACY
 	if (cap_unix(ses))
 		reset_cifs_unix_caps(0, tcon, NULL, ctx);
@@ -4051,11 +4090,23 @@ cifs_construct_tcon(struct cifs_sb_info *cifs_sb, kuid_t fsuid)
 out:
 	kfree(ctx->username);
 	kfree_sensitive(ctx->password);
+	kfree(origin_fullpath);
 	kfree(ctx);
 
 	return tcon;
 }
 
+static struct cifs_tcon *
+cifs_construct_tcon(struct cifs_sb_info *cifs_sb, kuid_t fsuid)
+{
+	struct cifs_tcon *ret;
+
+	cifs_mount_lock();
+	ret = __cifs_construct_tcon(cifs_sb, fsuid);
+	cifs_mount_unlock();
+	return ret;
+}
+
 struct cifs_tcon *
 cifs_sb_master_tcon(struct cifs_sb_info *cifs_sb)
 {
diff --git a/fs/smb/client/dfs.c b/fs/smb/client/dfs.c
index 449c59830039..3ec965547e3d 100644
--- a/fs/smb/client/dfs.c
+++ b/fs/smb/client/dfs.c
@@ -66,33 +66,20 @@ static int get_session(struct cifs_mount_ctx *mnt_ctx, const char *full_path)
 }
 
 /*
- * Track individual DFS referral servers used by new DFS mount.
- *
- * On success, their lifetime will be shared by final tcon (dfs_ses_list).
- * Otherwise, they will be put by dfs_put_root_smb_sessions() in cifs_mount().
+ * Get an active reference of @ses so that next call to cifs_put_tcon() won't
+ * release it as any new DFS referrals must go through its IPC tcon.
  */
-static int add_root_smb_session(struct cifs_mount_ctx *mnt_ctx)
+static void add_root_smb_session(struct cifs_mount_ctx *mnt_ctx)
 {
 	struct smb3_fs_context *ctx = mnt_ctx->fs_ctx;
-	struct dfs_root_ses *root_ses;
 	struct cifs_ses *ses = mnt_ctx->ses;
 
 	if (ses) {
-		root_ses = kmalloc(sizeof(*root_ses), GFP_KERNEL);
-		if (!root_ses)
-			return -ENOMEM;
-
-		INIT_LIST_HEAD(&root_ses->list);
-
 		spin_lock(&cifs_tcp_ses_lock);
 		cifs_smb_ses_inc_refcount(ses);
 		spin_unlock(&cifs_tcp_ses_lock);
-		root_ses->ses = ses;
-		list_add_tail(&root_ses->list, &mnt_ctx->dfs_ses_list);
 	}
-	/* Select new DFS referral server so that new referrals go through it */
 	ctx->dfs_root_ses = ses;
-	return 0;
 }
 
 static inline int parse_dfs_target(struct smb3_fs_context *ctx,
@@ -185,11 +172,8 @@ again:
 					continue;
 			}
 
-			if (is_refsrv) {
-				rc = add_root_smb_session(mnt_ctx);
-				if (rc)
-					goto out;
-			}
+			if (is_refsrv)
+				add_root_smb_session(mnt_ctx);
 
 			rc = ref_walk_advance(rw);
 			if (!rc) {
@@ -232,6 +216,7 @@ static int __dfs_mount_share(struct cifs_mount_ctx *mnt_ctx)
 	struct smb3_fs_context *ctx = mnt_ctx->fs_ctx;
 	struct cifs_tcon *tcon;
 	char *origin_fullpath;
+	bool new_tcon = true;
 	int rc;
 
 	origin_fullpath = dfs_get_path(cifs_sb, ctx->source);
@@ -239,6 +224,18 @@ static int __dfs_mount_share(struct cifs_mount_ctx *mnt_ctx)
 		return PTR_ERR(origin_fullpath);
 
 	rc = dfs_referral_walk(mnt_ctx);
+	if (!rc) {
+		/*
+		 * Prevent superblock from being created with any missing
+		 * connections.
+		 */
+		if (WARN_ON(!mnt_ctx->server))
+			rc = -EHOSTDOWN;
+		else if (WARN_ON(!mnt_ctx->ses))
+			rc = -EACCES;
+		else if (WARN_ON(!mnt_ctx->tcon))
+			rc = -ENOENT;
+	}
 	if (rc)
 		goto out;
 
@@ -247,15 +244,14 @@ static int __dfs_mount_share(struct cifs_mount_ctx *mnt_ctx)
 	if (!tcon->origin_fullpath) {
 		tcon->origin_fullpath = origin_fullpath;
 		origin_fullpath = NULL;
+	} else {
+		new_tcon = false;
 	}
 	spin_unlock(&tcon->tc_lock);
 
-	if (list_empty(&tcon->dfs_ses_list)) {
-		list_replace_init(&mnt_ctx->dfs_ses_list, &tcon->dfs_ses_list);
+	if (new_tcon) {
 		queue_delayed_work(dfscache_wq, &tcon->dfs_cache_work,
 				   dfs_cache_get_ttl() * HZ);
-	} else {
-		dfs_put_root_smb_sessions(&mnt_ctx->dfs_ses_list);
 	}
 
 out:
@@ -298,7 +294,6 @@ int dfs_mount_share(struct cifs_mount_ctx *mnt_ctx, bool *isdfs)
 	if (rc)
 		return rc;
 
-	ctx->dfs_root_ses = mnt_ctx->ses;
 	/*
 	 * If called with 'nodfs' mount option, then skip DFS resolving.  Otherwise unconditionally
 	 * try to get an DFS referral (even cached) to determine whether it is an DFS mount.
@@ -324,7 +319,9 @@ int dfs_mount_share(struct cifs_mount_ctx *mnt_ctx, bool *isdfs)
 
 	*isdfs = true;
 	add_root_smb_session(mnt_ctx);
-	return __dfs_mount_share(mnt_ctx);
+	rc = __dfs_mount_share(mnt_ctx);
+	dfs_put_root_smb_sessions(mnt_ctx);
+	return rc;
 }
 
 /* Update dfs referral path of superblock */
diff --git a/fs/smb/client/dfs.h b/fs/smb/client/dfs.h
index 875ab7ae57fc..e5c4dcf83750 100644
--- a/fs/smb/client/dfs.h
+++ b/fs/smb/client/dfs.h
@@ -7,7 +7,9 @@
 #define _CIFS_DFS_H
 
 #include "cifsglob.h"
+#include "cifsproto.h"
 #include "fs_context.h"
+#include "dfs_cache.h"
 #include "cifs_unicode.h"
 #include <linux/namei.h>
 
@@ -114,11 +116,6 @@ static inline void ref_walk_set_tgt_hint(struct dfs_ref_walk *rw)
 				       ref_walk_tit(rw));
 }
 
-struct dfs_root_ses {
-	struct list_head list;
-	struct cifs_ses *ses;
-};
-
 int dfs_parse_target_referral(const char *full_path, const struct dfs_info3_param *ref,
 			      struct smb3_fs_context *ctx);
 int dfs_mount_share(struct cifs_mount_ctx *mnt_ctx, bool *isdfs);
@@ -133,20 +130,32 @@ static inline int dfs_get_referral(struct cifs_mount_ctx *mnt_ctx, const char *p
 {
 	struct smb3_fs_context *ctx = mnt_ctx->fs_ctx;
 	struct cifs_sb_info *cifs_sb = mnt_ctx->cifs_sb;
+	struct cifs_ses *rses = ctx->dfs_root_ses ?: mnt_ctx->ses;
 
-	return dfs_cache_find(mnt_ctx->xid, ctx->dfs_root_ses, cifs_sb->local_nls,
+	return dfs_cache_find(mnt_ctx->xid, rses, cifs_sb->local_nls,
 			      cifs_remap(cifs_sb), path, ref, tl);
 }
 
-static inline void dfs_put_root_smb_sessions(struct list_head *head)
+/*
+ * cifs_get_smb_ses() already guarantees an active reference of
+ * @ses->dfs_root_ses when a new session is created, so we need to put extra
+ * references of all DFS root sessions that were used across the mount process
+ * in dfs_mount_share().
+ */
+static inline void dfs_put_root_smb_sessions(struct cifs_mount_ctx *mnt_ctx)
 {
-	struct dfs_root_ses *root, *tmp;
+	const struct smb3_fs_context *ctx = mnt_ctx->fs_ctx;
+	struct cifs_ses *ses = ctx->dfs_root_ses;
+	struct cifs_ses *cur;
+
+	if (!ses)
+		return;
 
-	list_for_each_entry_safe(root, tmp, head, list) {
-		list_del_init(&root->list);
-		cifs_put_smb_ses(root->ses);
-		kfree(root);
+	for (cur = ses; cur; cur = cur->dfs_root_ses) {
+		if (cur->dfs_root_ses)
+			cifs_put_smb_ses(cur->dfs_root_ses);
 	}
+	cifs_put_smb_ses(ses);
 }
 
 #endif /* _CIFS_DFS_H */
diff --git a/fs/smb/client/dfs_cache.c b/fs/smb/client/dfs_cache.c
index 508d831fabe3..11c8efecf7aa 100644
--- a/fs/smb/client/dfs_cache.c
+++ b/fs/smb/client/dfs_cache.c
@@ -1172,8 +1172,8 @@ static bool is_ses_good(struct cifs_ses *ses)
 	return ret;
 }
 
-/* Refresh dfs referral of tcon and mark it for reconnect if needed */
-static int __refresh_tcon(const char *path, struct cifs_ses *ses, bool force_refresh)
+/* Refresh dfs referral of @ses and mark it for reconnect if needed */
+static void __refresh_ses_referral(struct cifs_ses *ses, bool force_refresh)
 {
 	struct TCP_Server_Info *server = ses->server;
 	DFS_CACHE_TGT_LIST(old_tl);
@@ -1181,10 +1181,21 @@ static int __refresh_tcon(const char *path, struct cifs_ses *ses, bool force_ref
 	bool needs_refresh = false;
 	struct cache_entry *ce;
 	unsigned int xid;
+	char *path = NULL;
 	int rc = 0;
 
 	xid = get_xid();
 
+	mutex_lock(&server->refpath_lock);
+	if (server->leaf_fullpath) {
+		path = kstrdup(server->leaf_fullpath + 1, GFP_ATOMIC);
+		if (!path)
+			rc = -ENOMEM;
+	}
+	mutex_unlock(&server->refpath_lock);
+	if (!path)
+		goto out;
+
 	down_read(&htable_rw_lock);
 	ce = lookup_cache_entry(path);
 	needs_refresh = force_refresh || IS_ERR(ce) || cache_entry_expired(ce);
@@ -1218,19 +1229,17 @@ out:
 	free_xid(xid);
 	dfs_cache_free_tgts(&old_tl);
 	dfs_cache_free_tgts(&new_tl);
-	return rc;
+	kfree(path);
 }
 
-static int refresh_tcon(struct cifs_tcon *tcon, bool force_refresh)
+static inline void refresh_ses_referral(struct cifs_ses *ses)
 {
-	struct TCP_Server_Info *server = tcon->ses->server;
-	struct cifs_ses *ses = tcon->ses;
+	__refresh_ses_referral(ses, false);
+}
 
-	mutex_lock(&server->refpath_lock);
-	if (server->leaf_fullpath)
-		__refresh_tcon(server->leaf_fullpath + 1, ses, force_refresh);
-	mutex_unlock(&server->refpath_lock);
-	return 0;
+static inline void force_refresh_ses_referral(struct cifs_ses *ses)
+{
+	__refresh_ses_referral(ses, true);
 }
 
 /**
@@ -1271,34 +1280,20 @@ int dfs_cache_remount_fs(struct cifs_sb_info *cifs_sb)
 	 */
 	cifs_sb->mnt_cifs_flags |= CIFS_MOUNT_USE_PREFIX_PATH;
 
-	return refresh_tcon(tcon, true);
+	force_refresh_ses_referral(tcon->ses);
+	return 0;
 }
 
 /* Refresh all DFS referrals related to DFS tcon */
 void dfs_cache_refresh(struct work_struct *work)
 {
-	struct TCP_Server_Info *server;
-	struct dfs_root_ses *rses;
 	struct cifs_tcon *tcon;
 	struct cifs_ses *ses;
 
 	tcon = container_of(work, struct cifs_tcon, dfs_cache_work.work);
-	ses = tcon->ses;
-	server = ses->server;
 
-	mutex_lock(&server->refpath_lock);
-	if (server->leaf_fullpath)
-		__refresh_tcon(server->leaf_fullpath + 1, ses, false);
-	mutex_unlock(&server->refpath_lock);
-
-	list_for_each_entry(rses, &tcon->dfs_ses_list, list) {
-		ses = rses->ses;
-		server = ses->server;
-		mutex_lock(&server->refpath_lock);
-		if (server->leaf_fullpath)
-			__refresh_tcon(server->leaf_fullpath + 1, ses, false);
-		mutex_unlock(&server->refpath_lock);
-	}
+	for (ses = tcon->ses; ses; ses = ses->dfs_root_ses)
+		refresh_ses_referral(ses);
 
 	queue_delayed_work(dfscache_wq, &tcon->dfs_cache_work,
 			   atomic_read(&dfs_cache_ttl) * HZ);
diff --git a/fs/smb/client/dir.c b/fs/smb/client/dir.c
index d11dc3aa458b..864b194dbaa0 100644
--- a/fs/smb/client/dir.c
+++ b/fs/smb/client/dir.c
@@ -189,6 +189,7 @@ static int cifs_do_create(struct inode *inode, struct dentry *direntry, unsigned
 	int disposition;
 	struct TCP_Server_Info *server = tcon->ses->server;
 	struct cifs_open_parms oparms;
+	int rdwr_for_fscache = 0;
 
 	*oplock = 0;
 	if (tcon->ses->server->oplocks)
@@ -200,6 +201,10 @@ static int cifs_do_create(struct inode *inode, struct dentry *direntry, unsigned
 		return PTR_ERR(full_path);
 	}
 
+	/* If we're caching, we need to be able to fill in around partial writes. */
+	if (cifs_fscache_enabled(inode) && (oflags & O_ACCMODE) == O_WRONLY)
+		rdwr_for_fscache = 1;
+
 #ifdef CONFIG_CIFS_ALLOW_INSECURE_LEGACY
 	if (tcon->unix_ext && cap_unix(tcon->ses) && !tcon->broken_posix_open &&
 	    (CIFS_UNIX_POSIX_PATH_OPS_CAP &
@@ -276,6 +281,8 @@ static int cifs_do_create(struct inode *inode, struct dentry *direntry, unsigned
 		desired_access |= GENERIC_READ; /* is this too little? */
 	if (OPEN_FMODE(oflags) & FMODE_WRITE)
 		desired_access |= GENERIC_WRITE;
+	if (rdwr_for_fscache == 1)
+		desired_access |= GENERIC_READ;
 
 	disposition = FILE_OVERWRITE_IF;
 	if ((oflags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
@@ -304,6 +311,7 @@ static int cifs_do_create(struct inode *inode, struct dentry *direntry, unsigned
 	if (!tcon->unix_ext && (mode & S_IWUGO) == 0)
 		create_options |= CREATE_OPTION_READONLY;
 
+retry_open:
 	oparms = (struct cifs_open_parms) {
 		.tcon = tcon,
 		.cifs_sb = cifs_sb,
@@ -317,8 +325,15 @@ static int cifs_do_create(struct inode *inode, struct dentry *direntry, unsigned
 	rc = server->ops->open(xid, &oparms, oplock, buf);
 	if (rc) {
 		cifs_dbg(FYI, "cifs_create returned 0x%x\n", rc);
+		if (rc == -EACCES && rdwr_for_fscache == 1) {
+			desired_access &= ~GENERIC_READ;
+			rdwr_for_fscache = 2;
+			goto retry_open;
+		}
 		goto out;
 	}
+	if (rdwr_for_fscache == 2)
+		cifs_invalidate_cache(inode, FSCACHE_INVAL_DIO_WRITE);
 
 #ifdef CONFIG_CIFS_ALLOW_INSECURE_LEGACY
 	/*
diff --git a/fs/smb/client/file.c b/fs/smb/client/file.c
index 16aadce492b2..9be37d0fe724 100644
--- a/fs/smb/client/file.c
+++ b/fs/smb/client/file.c
@@ -206,12 +206,12 @@ cifs_mark_open_files_invalid(struct cifs_tcon *tcon)
 	 */
 }
 
-static inline int cifs_convert_flags(unsigned int flags)
+static inline int cifs_convert_flags(unsigned int flags, int rdwr_for_fscache)
 {
 	if ((flags & O_ACCMODE) == O_RDONLY)
 		return GENERIC_READ;
 	else if ((flags & O_ACCMODE) == O_WRONLY)
-		return GENERIC_WRITE;
+		return rdwr_for_fscache == 1 ? (GENERIC_READ | GENERIC_WRITE) : GENERIC_WRITE;
 	else if ((flags & O_ACCMODE) == O_RDWR) {
 		/* GENERIC_ALL is too much permission to request
 		   can cause unnecessary access denied on create */
@@ -348,11 +348,16 @@ static int cifs_nt_open(const char *full_path, struct inode *inode, struct cifs_
 	int create_options = CREATE_NOT_DIR;
 	struct TCP_Server_Info *server = tcon->ses->server;
 	struct cifs_open_parms oparms;
+	int rdwr_for_fscache = 0;
 
 	if (!server->ops->open)
 		return -ENOSYS;
 
-	desired_access = cifs_convert_flags(f_flags);
+	/* If we're caching, we need to be able to fill in around partial writes. */
+	if (cifs_fscache_enabled(inode) && (f_flags & O_ACCMODE) == O_WRONLY)
+		rdwr_for_fscache = 1;
+
+	desired_access = cifs_convert_flags(f_flags, rdwr_for_fscache);
 
 /*********************************************************************
  *  open flag mapping table:
@@ -389,6 +394,7 @@ static int cifs_nt_open(const char *full_path, struct inode *inode, struct cifs_
 	if (f_flags & O_DIRECT)
 		create_options |= CREATE_NO_BUFFER;
 
+retry_open:
 	oparms = (struct cifs_open_parms) {
 		.tcon = tcon,
 		.cifs_sb = cifs_sb,
@@ -400,8 +406,16 @@ static int cifs_nt_open(const char *full_path, struct inode *inode, struct cifs_
 	};
 
 	rc = server->ops->open(xid, &oparms, oplock, buf);
-	if (rc)
+	if (rc) {
+		if (rc == -EACCES && rdwr_for_fscache == 1) {
+			desired_access = cifs_convert_flags(f_flags, 0);
+			rdwr_for_fscache = 2;
+			goto retry_open;
+		}
 		return rc;
+	}
+	if (rdwr_for_fscache == 2)
+		cifs_invalidate_cache(inode, FSCACHE_INVAL_DIO_WRITE);
 
 	/* TODO: Add support for calling posix query info but with passing in fid */
 	if (tcon->unix_ext)
@@ -445,6 +459,7 @@ cifs_down_write(struct rw_semaphore *sem)
 }
 
 static void cifsFileInfo_put_work(struct work_struct *work);
+void serverclose_work(struct work_struct *work);
 
 struct cifsFileInfo *cifs_new_fileinfo(struct cifs_fid *fid, struct file *file,
 				       struct tcon_link *tlink, __u32 oplock,
@@ -491,6 +506,7 @@ struct cifsFileInfo *cifs_new_fileinfo(struct cifs_fid *fid, struct file *file,
 	cfile->tlink = cifs_get_tlink(tlink);
 	INIT_WORK(&cfile->oplock_break, cifs_oplock_break);
 	INIT_WORK(&cfile->put, cifsFileInfo_put_work);
+	INIT_WORK(&cfile->serverclose, serverclose_work);
 	INIT_DELAYED_WORK(&cfile->deferred, smb2_deferred_work_close);
 	mutex_init(&cfile->fh_mutex);
 	spin_lock_init(&cfile->file_info_lock);
@@ -582,6 +598,40 @@ static void cifsFileInfo_put_work(struct work_struct *work)
 	cifsFileInfo_put_final(cifs_file);
 }
 
+void serverclose_work(struct work_struct *work)
+{
+	struct cifsFileInfo *cifs_file = container_of(work,
+			struct cifsFileInfo, serverclose);
+
+	struct cifs_tcon *tcon = tlink_tcon(cifs_file->tlink);
+
+	struct TCP_Server_Info *server = tcon->ses->server;
+	int rc = 0;
+	int retries = 0;
+	int MAX_RETRIES = 4;
+
+	do {
+		if (server->ops->close_getattr)
+			rc = server->ops->close_getattr(0, tcon, cifs_file);
+		else if (server->ops->close)
+			rc = server->ops->close(0, tcon, &cifs_file->fid);
+
+		if (rc == -EBUSY || rc == -EAGAIN) {
+			retries++;
+			msleep(250);
+		}
+	} while ((rc == -EBUSY || rc == -EAGAIN) && (retries < MAX_RETRIES)
+	);
+
+	if (retries == MAX_RETRIES)
+		pr_warn("Serverclose failed %d times, giving up\n", MAX_RETRIES);
+
+	if (cifs_file->offload)
+		queue_work(fileinfo_put_wq, &cifs_file->put);
+	else
+		cifsFileInfo_put_final(cifs_file);
+}
+
 /**
  * cifsFileInfo_put - release a reference of file priv data
  *
@@ -622,10 +672,13 @@ void _cifsFileInfo_put(struct cifsFileInfo *cifs_file,
 	struct cifs_fid fid = {};
 	struct cifs_pending_open open;
 	bool oplock_break_cancelled;
+	bool serverclose_offloaded = false;
 
 	spin_lock(&tcon->open_file_lock);
 	spin_lock(&cifsi->open_file_lock);
 	spin_lock(&cifs_file->file_info_lock);
+
+	cifs_file->offload = offload;
 	if (--cifs_file->count > 0) {
 		spin_unlock(&cifs_file->file_info_lock);
 		spin_unlock(&cifsi->open_file_lock);
@@ -667,13 +720,20 @@ void _cifsFileInfo_put(struct cifsFileInfo *cifs_file,
 	if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
 		struct TCP_Server_Info *server = tcon->ses->server;
 		unsigned int xid;
+		int rc = 0;
 
 		xid = get_xid();
 		if (server->ops->close_getattr)
-			server->ops->close_getattr(xid, tcon, cifs_file);
+			rc = server->ops->close_getattr(xid, tcon, cifs_file);
 		else if (server->ops->close)
-			server->ops->close(xid, tcon, &cifs_file->fid);
+			rc = server->ops->close(xid, tcon, &cifs_file->fid);
 		_free_xid(xid);
+
+		if (rc == -EBUSY || rc == -EAGAIN) {
+			// Server close failed, hence offloading it as an async op
+			queue_work(serverclose_wq, &cifs_file->serverclose);
+			serverclose_offloaded = true;
+		}
 	}
 
 	if (oplock_break_cancelled)
@@ -681,10 +741,15 @@ void _cifsFileInfo_put(struct cifsFileInfo *cifs_file,
 
 	cifs_del_pending_open(&open);
 
-	if (offload)
-		queue_work(fileinfo_put_wq, &cifs_file->put);
-	else
-		cifsFileInfo_put_final(cifs_file);
+	// if serverclose has been offloaded to wq (on failure), it will
+	// handle offloading put as well. If serverclose not offloaded,
+	// we need to handle offloading put here.
+	if (!serverclose_offloaded) {
+		if (offload)
+			queue_work(fileinfo_put_wq, &cifs_file->put);
+		else
+			cifsFileInfo_put_final(cifs_file);
+	}
 }
 
 int cifs_open(struct inode *inode, struct file *file)
@@ -834,11 +899,11 @@ int cifs_open(struct inode *inode, struct file *file)
 use_cache:
 	fscache_use_cookie(cifs_inode_cookie(file_inode(file)),
 			   file->f_mode & FMODE_WRITE);
-	if (file->f_flags & O_DIRECT &&
-	    (!((file->f_flags & O_ACCMODE) != O_RDONLY) ||
-	     file->f_flags & O_APPEND))
-		cifs_invalidate_cache(file_inode(file),
-				      FSCACHE_INVAL_DIO_WRITE);
+	if (!(file->f_flags & O_DIRECT))
+		goto out;
+	if ((file->f_flags & (O_ACCMODE | O_APPEND)) == O_RDONLY)
+		goto out;
+	cifs_invalidate_cache(file_inode(file), FSCACHE_INVAL_DIO_WRITE);
 
 out:
 	free_dentry_path(page);
@@ -903,6 +968,7 @@ cifs_reopen_file(struct cifsFileInfo *cfile, bool can_flush)
 	int disposition = FILE_OPEN;
 	int create_options = CREATE_NOT_DIR;
 	struct cifs_open_parms oparms;
+	int rdwr_for_fscache = 0;
 
 	xid = get_xid();
 	mutex_lock(&cfile->fh_mutex);
@@ -966,7 +1032,11 @@ cifs_reopen_file(struct cifsFileInfo *cfile, bool can_flush)
 	}
 #endif /* CONFIG_CIFS_ALLOW_INSECURE_LEGACY */
 
-	desired_access = cifs_convert_flags(cfile->f_flags);
+	/* If we're caching, we need to be able to fill in around partial writes. */
+	if (cifs_fscache_enabled(inode) && (cfile->f_flags & O_ACCMODE) == O_WRONLY)
+		rdwr_for_fscache = 1;
+
+	desired_access = cifs_convert_flags(cfile->f_flags, rdwr_for_fscache);
 
 	/* O_SYNC also has bit for O_DSYNC so following check picks up either */
 	if (cfile->f_flags & O_SYNC)
@@ -978,6 +1048,7 @@ cifs_reopen_file(struct cifsFileInfo *cfile, bool can_flush)
 	if (server->ops->get_lease_key)
 		server->ops->get_lease_key(inode, &cfile->fid);
 
+retry_open:
 	oparms = (struct cifs_open_parms) {
 		.tcon = tcon,
 		.cifs_sb = cifs_sb,
@@ -1003,6 +1074,11 @@ cifs_reopen_file(struct cifsFileInfo *cfile, bool can_flush)
 		/* indicate that we need to relock the file */
 		oparms.reconnect = true;
 	}
+	if (rc == -EACCES && rdwr_for_fscache == 1) {
+		desired_access = cifs_convert_flags(cfile->f_flags, 0);
+		rdwr_for_fscache = 2;
+		goto retry_open;
+	}
 
 	if (rc) {
 		mutex_unlock(&cfile->fh_mutex);
@@ -1011,6 +1087,9 @@ cifs_reopen_file(struct cifsFileInfo *cfile, bool can_flush)
 		goto reopen_error_exit;
 	}
 
+	if (rdwr_for_fscache == 2)
+		cifs_invalidate_cache(inode, FSCACHE_INVAL_DIO_WRITE);
+
 #ifdef CONFIG_CIFS_ALLOW_INSECURE_LEGACY
 reopen_success:
 #endif /* CONFIG_CIFS_ALLOW_INSECURE_LEGACY */
diff --git a/fs/smb/client/fs_context.c b/fs/smb/client/fs_context.c
index bdcbe6ff2739..6c727d8c31e8 100644
--- a/fs/smb/client/fs_context.c
+++ b/fs/smb/client/fs_context.c
@@ -37,7 +37,7 @@
 #include "rfc1002pdu.h"
 #include "fs_context.h"
 
-static DEFINE_MUTEX(cifs_mount_mutex);
+DEFINE_MUTEX(cifs_mount_mutex);
 
 static const match_table_t cifs_smb_version_tokens = {
 	{ Smb_1, SMB1_VERSION_STRING },
@@ -162,6 +162,7 @@ const struct fs_parameter_spec smb3_fs_parameters[] = {
 	fsparam_string("username", Opt_user),
 	fsparam_string("pass", Opt_pass),
 	fsparam_string("password", Opt_pass),
+	fsparam_string("password2", Opt_pass2),
 	fsparam_string("ip", Opt_ip),
 	fsparam_string("addr", Opt_ip),
 	fsparam_string("domain", Opt_domain),
@@ -345,6 +346,7 @@ smb3_fs_context_dup(struct smb3_fs_context *new_ctx, struct smb3_fs_context *ctx
 	new_ctx->nodename = NULL;
 	new_ctx->username = NULL;
 	new_ctx->password = NULL;
+	new_ctx->password2 = NULL;
 	new_ctx->server_hostname = NULL;
 	new_ctx->domainname = NULL;
 	new_ctx->UNC = NULL;
@@ -357,6 +359,7 @@ smb3_fs_context_dup(struct smb3_fs_context *new_ctx, struct smb3_fs_context *ctx
 	DUP_CTX_STR(prepath);
 	DUP_CTX_STR(username);
 	DUP_CTX_STR(password);
+	DUP_CTX_STR(password2);
 	DUP_CTX_STR(server_hostname);
 	DUP_CTX_STR(UNC);
 	DUP_CTX_STR(source);
@@ -783,9 +786,9 @@ static int smb3_get_tree(struct fs_context *fc)
 
 	if (err)
 		return err;
-	mutex_lock(&cifs_mount_mutex);
+	cifs_mount_lock();
 	ret = smb3_get_tree_common(fc);
-	mutex_unlock(&cifs_mount_mutex);
+	cifs_mount_unlock();
 	return ret;
 }
 
@@ -905,6 +908,8 @@ static int smb3_reconfigure(struct fs_context *fc)
 	else  {
 		kfree_sensitive(ses->password);
 		ses->password = kstrdup(ctx->password, GFP_KERNEL);
+		kfree_sensitive(ses->password2);
+		ses->password2 = kstrdup(ctx->password2, GFP_KERNEL);
 	}
 	STEAL_STRING(cifs_sb, ctx, domainname);
 	STEAL_STRING(cifs_sb, ctx, nodename);
@@ -1305,6 +1310,18 @@ static int smb3_fs_context_parse_param(struct fs_context *fc,
 			goto cifs_parse_mount_err;
 		}
 		break;
+	case Opt_pass2:
+		kfree_sensitive(ctx->password2);
+		ctx->password2 = NULL;
+		if (strlen(param->string) == 0)
+			break;
+
+		ctx->password2 = kstrdup(param->string, GFP_KERNEL);
+		if (ctx->password2 == NULL) {
+			cifs_errorf(fc, "OOM when copying password2 string\n");
+			goto cifs_parse_mount_err;
+		}
+		break;
 	case Opt_ip:
 		if (strlen(param->string) == 0) {
 			ctx->got_ip = false;
@@ -1608,6 +1625,8 @@ static int smb3_fs_context_parse_param(struct fs_context *fc,
  cifs_parse_mount_err:
 	kfree_sensitive(ctx->password);
 	ctx->password = NULL;
+	kfree_sensitive(ctx->password2);
+	ctx->password2 = NULL;
 	return -EINVAL;
 }
 
@@ -1713,6 +1732,8 @@ smb3_cleanup_fs_context_contents(struct smb3_fs_context *ctx)
 	ctx->username = NULL;
 	kfree_sensitive(ctx->password);
 	ctx->password = NULL;
+	kfree_sensitive(ctx->password2);
+	ctx->password2 = NULL;
 	kfree(ctx->server_hostname);
 	ctx->server_hostname = NULL;
 	kfree(ctx->UNC);
diff --git a/fs/smb/client/fs_context.h b/fs/smb/client/fs_context.h
index 7863f2248c4d..a947bddeba27 100644
--- a/fs/smb/client/fs_context.h
+++ b/fs/smb/client/fs_context.h
@@ -145,6 +145,7 @@ enum cifs_param {
 	Opt_source,
 	Opt_user,
 	Opt_pass,
+	Opt_pass2,
 	Opt_ip,
 	Opt_domain,
 	Opt_srcaddr,
@@ -177,6 +178,7 @@ struct smb3_fs_context {
 
 	char *username;
 	char *password;
+	char *password2;
 	char *domainname;
 	char *source;
 	char *server_hostname;
@@ -304,4 +306,16 @@ extern void smb3_update_mnt_flags(struct cifs_sb_info *cifs_sb);
 #define MAX_CACHED_FIDS 16
 extern char *cifs_sanitize_prepath(char *prepath, gfp_t gfp);
 
+extern struct mutex cifs_mount_mutex;
+
+static inline void cifs_mount_lock(void)
+{
+	mutex_lock(&cifs_mount_mutex);
+}
+
+static inline void cifs_mount_unlock(void)
+{
+	mutex_unlock(&cifs_mount_mutex);
+}
+
 #endif
diff --git a/fs/smb/client/fscache.h b/fs/smb/client/fscache.h
index a3d73720914f..1f2ea9f5cc9a 100644
--- a/fs/smb/client/fscache.h
+++ b/fs/smb/client/fscache.h
@@ -109,6 +109,11 @@ static inline void cifs_readahead_to_fscache(struct inode *inode,
 		__cifs_readahead_to_fscache(inode, pos, len);
 }
 
+static inline bool cifs_fscache_enabled(struct inode *inode)
+{
+	return fscache_cookie_enabled(cifs_inode_cookie(inode));
+}
+
 #else /* CONFIG_CIFS_FSCACHE */
 static inline
 void cifs_fscache_fill_coherency(struct inode *inode,
@@ -124,6 +129,7 @@ static inline void cifs_fscache_release_inode_cookie(struct inode *inode) {}
 static inline void cifs_fscache_unuse_inode_cookie(struct inode *inode, bool update) {}
 static inline struct fscache_cookie *cifs_inode_cookie(struct inode *inode) { return NULL; }
 static inline void cifs_invalidate_cache(struct inode *inode, unsigned int flags) {}
+static inline bool cifs_fscache_enabled(struct inode *inode) { return false; }
 
 static inline int cifs_fscache_query_occupancy(struct inode *inode,
 					       pgoff_t first, unsigned int nr_pages,
diff --git a/fs/smb/client/inode.c b/fs/smb/client/inode.c
index 91b07ef9e25c..60afab5c83d4 100644
--- a/fs/smb/client/inode.c
+++ b/fs/smb/client/inode.c
@@ -1105,7 +1105,8 @@ static int cifs_get_fattr(struct cifs_open_info_data *data,
 		} else {
 			cifs_open_info_to_fattr(fattr, data, sb);
 		}
-		if (!rc && fattr->cf_flags & CIFS_FATTR_DELETE_PENDING)
+		if (!rc && *inode &&
+		    (fattr->cf_flags & CIFS_FATTR_DELETE_PENDING))
 			cifs_mark_open_handles_for_deleted_file(*inode, full_path);
 		break;
 	case -EREMOTE:
diff --git a/fs/smb/client/ioctl.c b/fs/smb/client/ioctl.c
index c012dfdba80d..855ac5a62edf 100644
--- a/fs/smb/client/ioctl.c
+++ b/fs/smb/client/ioctl.c
@@ -247,7 +247,9 @@ static int cifs_dump_full_key(struct cifs_tcon *tcon, struct smb3_full_key_debug
 		spin_lock(&cifs_tcp_ses_lock);
 		list_for_each_entry(server_it, &cifs_tcp_ses_list, tcp_ses_list) {
 			list_for_each_entry(ses_it, &server_it->smb_ses_list, smb_ses_list) {
-				if (ses_it->Suid == out.session_id) {
+				spin_lock(&ses_it->ses_lock);
+				if (ses_it->ses_status != SES_EXITING &&
+				    ses_it->Suid == out.session_id) {
 					ses = ses_it;
 					/*
 					 * since we are using the session outside the crit
@@ -255,9 +257,11 @@ static int cifs_dump_full_key(struct cifs_tcon *tcon, struct smb3_full_key_debug
 					 * so increment its refcount
 					 */
 					cifs_smb_ses_inc_refcount(ses);
+					spin_unlock(&ses_it->ses_lock);
 					found = true;
 					goto search_end;
 				}
+				spin_unlock(&ses_it->ses_lock);
 			}
 		}
 search_end:
diff --git a/fs/smb/client/misc.c b/fs/smb/client/misc.c
index c3771fc81328..7d15a1969b81 100644
--- a/fs/smb/client/misc.c
+++ b/fs/smb/client/misc.c
@@ -98,6 +98,7 @@ sesInfoFree(struct cifs_ses *buf_to_free)
 	kfree(buf_to_free->serverDomain);
 	kfree(buf_to_free->serverNOS);
 	kfree_sensitive(buf_to_free->password);
+	kfree_sensitive(buf_to_free->password2);
 	kfree(buf_to_free->user_name);
 	kfree(buf_to_free->domainName);
 	kfree_sensitive(buf_to_free->auth_key.response);
@@ -138,9 +139,6 @@ tcon_info_alloc(bool dir_leases_enabled)
 	atomic_set(&ret_buf->num_local_opens, 0);
 	atomic_set(&ret_buf->num_remote_opens, 0);
 	ret_buf->stats_from_time = ktime_get_real_seconds();
-#ifdef CONFIG_CIFS_DFS_UPCALL
-	INIT_LIST_HEAD(&ret_buf->dfs_ses_list);
-#endif
 
 	return ret_buf;
 }
@@ -156,9 +154,6 @@ tconInfoFree(struct cifs_tcon *tcon)
 	atomic_dec(&tconInfoAllocCount);
 	kfree(tcon->nativeFileSystem);
 	kfree_sensitive(tcon->password);
-#ifdef CONFIG_CIFS_DFS_UPCALL
-	dfs_put_root_smb_sessions(&tcon->dfs_ses_list);
-#endif
 	kfree(tcon->origin_fullpath);
 	kfree(tcon);
 }
@@ -487,6 +482,8 @@ is_valid_oplock_break(char *buffer, struct TCP_Server_Info *srv)
 	/* look up tcon based on tid & uid */
 	spin_lock(&cifs_tcp_ses_lock);
 	list_for_each_entry(ses, &pserver->smb_ses_list, smb_ses_list) {
+		if (cifs_ses_exiting(ses))
+			continue;
 		list_for_each_entry(tcon, &ses->tcon_list, tcon_list) {
 			if (tcon->tid != buf->Tid)
 				continue;
diff --git a/fs/smb/client/smb1ops.c b/fs/smb/client/smb1ops.c
index a9eaba8083b0..212ec6f66ec6 100644
--- a/fs/smb/client/smb1ops.c
+++ b/fs/smb/client/smb1ops.c
@@ -753,11 +753,11 @@ cifs_set_fid(struct cifsFileInfo *cfile, struct cifs_fid *fid, __u32 oplock)
 	cinode->can_cache_brlcks = CIFS_CACHE_WRITE(cinode);
 }
 
-static void
+static int
 cifs_close_file(const unsigned int xid, struct cifs_tcon *tcon,
 		struct cifs_fid *fid)
 {
-	CIFSSMBClose(xid, tcon, fid->netfid);
+	return CIFSSMBClose(xid, tcon, fid->netfid);
 }
 
 static int
diff --git a/fs/smb/client/smb2misc.c b/fs/smb/client/smb2misc.c
index 82b84a4941dd..cc72be5a93a9 100644
--- a/fs/smb/client/smb2misc.c
+++ b/fs/smb/client/smb2misc.c
@@ -622,6 +622,8 @@ smb2_is_valid_lease_break(char *buffer, struct TCP_Server_Info *server)
 	/* look up tcon based on tid & uid */
 	spin_lock(&cifs_tcp_ses_lock);
 	list_for_each_entry(ses, &pserver->smb_ses_list, smb_ses_list) {
+		if (cifs_ses_exiting(ses))
+			continue;
 		list_for_each_entry(tcon, &ses->tcon_list, tcon_list) {
 			spin_lock(&tcon->open_file_lock);
 			cifs_stats_inc(
@@ -697,6 +699,8 @@ smb2_is_valid_oplock_break(char *buffer, struct TCP_Server_Info *server)
 	/* look up tcon based on tid & uid */
 	spin_lock(&cifs_tcp_ses_lock);
 	list_for_each_entry(ses, &pserver->smb_ses_list, smb_ses_list) {
+		if (cifs_ses_exiting(ses))
+			continue;
 		list_for_each_entry(tcon, &ses->tcon_list, tcon_list) {
 
 			spin_lock(&tcon->open_file_lock);
diff --git a/fs/smb/client/smb2ops.c b/fs/smb/client/smb2ops.c
index 2ed456948f34..78c94d0350fe 100644
--- a/fs/smb/client/smb2ops.c
+++ b/fs/smb/client/smb2ops.c
@@ -1412,14 +1412,14 @@ smb2_set_fid(struct cifsFileInfo *cfile, struct cifs_fid *fid, __u32 oplock)
 	memcpy(cfile->fid.create_guid, fid->create_guid, 16);
 }
 
-static void
+static int
 smb2_close_file(const unsigned int xid, struct cifs_tcon *tcon,
 		struct cifs_fid *fid)
 {
-	SMB2_close(xid, tcon, fid->persistent_fid, fid->volatile_fid);
+	return SMB2_close(xid, tcon, fid->persistent_fid, fid->volatile_fid);
 }
 
-static void
+static int
 smb2_close_getattr(const unsigned int xid, struct cifs_tcon *tcon,
 		   struct cifsFileInfo *cfile)
 {
@@ -1430,7 +1430,7 @@ smb2_close_getattr(const unsigned int xid, struct cifs_tcon *tcon,
 	rc = __SMB2_close(xid, tcon, cfile->fid.persistent_fid,
 		   cfile->fid.volatile_fid, &file_inf);
 	if (rc)
-		return;
+		return rc;
 
 	inode = d_inode(cfile->dentry);
 
@@ -1459,6 +1459,7 @@ smb2_close_getattr(const unsigned int xid, struct cifs_tcon *tcon,
 
 	/* End of file and Attributes should not have to be updated on close */
 	spin_unlock(&inode->i_lock);
+	return rc;
 }
 
 static int
@@ -2480,6 +2481,8 @@ smb2_is_network_name_deleted(char *buf, struct TCP_Server_Info *server)
 
 	spin_lock(&cifs_tcp_ses_lock);
 	list_for_each_entry(ses, &pserver->smb_ses_list, smb_ses_list) {
+		if (cifs_ses_exiting(ses))
+			continue;
 		list_for_each_entry(tcon, &ses->tcon_list, tcon_list) {
 			if (tcon->tid == le32_to_cpu(shdr->Id.SyncId.TreeId)) {
 				spin_lock(&tcon->tc_lock);
@@ -3913,7 +3916,7 @@ smb21_set_oplock_level(struct cifsInodeInfo *cinode, __u32 oplock,
 		strcat(message, "W");
 	}
 	if (!new_oplock)
-		strncpy(message, "None", sizeof(message));
+		strscpy(message, "None");
 
 	cinode->oplock = new_oplock;
 	cifs_dbg(FYI, "%s Lease granted on inode %p\n", message,
@@ -4961,68 +4964,84 @@ static int smb2_next_header(struct TCP_Server_Info *server, char *buf,
 	return 0;
 }
 
-int cifs_sfu_make_node(unsigned int xid, struct inode *inode,
-		       struct dentry *dentry, struct cifs_tcon *tcon,
-		       const char *full_path, umode_t mode, dev_t dev)
+static int __cifs_sfu_make_node(unsigned int xid, struct inode *inode,
+				struct dentry *dentry, struct cifs_tcon *tcon,
+				const char *full_path, umode_t mode, dev_t dev)
 {
-	struct cifs_open_info_data buf = {};
 	struct TCP_Server_Info *server = tcon->ses->server;
 	struct cifs_open_parms oparms;
 	struct cifs_io_parms io_parms = {};
 	struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
 	struct cifs_fid fid;
 	unsigned int bytes_written;
-	struct win_dev *pdev;
+	struct win_dev pdev = {};
 	struct kvec iov[2];
 	__u32 oplock = server->oplocks ? REQ_OPLOCK : 0;
 	int rc;
 
-	if (!S_ISCHR(mode) && !S_ISBLK(mode) && !S_ISFIFO(mode))
+	switch (mode & S_IFMT) {
+	case S_IFCHR:
+		strscpy(pdev.type, "IntxCHR");
+		pdev.major = cpu_to_le64(MAJOR(dev));
+		pdev.minor = cpu_to_le64(MINOR(dev));
+		break;
+	case S_IFBLK:
+		strscpy(pdev.type, "IntxBLK");
+		pdev.major = cpu_to_le64(MAJOR(dev));
+		pdev.minor = cpu_to_le64(MINOR(dev));
+		break;
+	case S_IFIFO:
+		strscpy(pdev.type, "LnxFIFO");
+		break;
+	default:
 		return -EPERM;
+	}
 
-	oparms = (struct cifs_open_parms) {
-		.tcon = tcon,
-		.cifs_sb = cifs_sb,
-		.desired_access = GENERIC_WRITE,
-		.create_options = cifs_create_options(cifs_sb, CREATE_NOT_DIR |
-						      CREATE_OPTION_SPECIAL),
-		.disposition = FILE_CREATE,
-		.path = full_path,
-		.fid = &fid,
-	};
+	oparms = CIFS_OPARMS(cifs_sb, tcon, full_path, GENERIC_WRITE,
+			     FILE_CREATE, CREATE_NOT_DIR |
+			     CREATE_OPTION_SPECIAL, ACL_NO_MODE);
+	oparms.fid = &fid;
 
-	rc = server->ops->open(xid, &oparms, &oplock, &buf);
+	rc = server->ops->open(xid, &oparms, &oplock, NULL);
 	if (rc)
 		return rc;
 
-	/*
-	 * BB Do not bother to decode buf since no local inode yet to put
-	 * timestamps in, but we can reuse it safely.
-	 */
-	pdev = (struct win_dev *)&buf.fi;
 	io_parms.pid = current->tgid;
 	io_parms.tcon = tcon;
-	io_parms.length = sizeof(*pdev);
-	iov[1].iov_base = pdev;
-	iov[1].iov_len = sizeof(*pdev);
-	if (S_ISCHR(mode)) {
-		memcpy(pdev->type, "IntxCHR", 8);
-		pdev->major = cpu_to_le64(MAJOR(dev));
-		pdev->minor = cpu_to_le64(MINOR(dev));
-	} else if (S_ISBLK(mode)) {
-		memcpy(pdev->type, "IntxBLK", 8);
-		pdev->major = cpu_to_le64(MAJOR(dev));
-		pdev->minor = cpu_to_le64(MINOR(dev));
-	} else if (S_ISFIFO(mode)) {
-		memcpy(pdev->type, "LnxFIFO", 8);
-	}
+	io_parms.length = sizeof(pdev);
+	iov[1].iov_base = &pdev;
+	iov[1].iov_len = sizeof(pdev);
 
 	rc = server->ops->sync_write(xid, &fid, &io_parms,
 				     &bytes_written, iov, 1);
 	server->ops->close(xid, tcon, &fid);
-	d_drop(dentry);
-	/* FIXME: add code here to set EAs */
-	cifs_free_open_info(&buf);
+	return rc;
+}
+
+int cifs_sfu_make_node(unsigned int xid, struct inode *inode,
+		       struct dentry *dentry, struct cifs_tcon *tcon,
+		       const char *full_path, umode_t mode, dev_t dev)
+{
+	struct inode *new = NULL;
+	int rc;
+
+	rc = __cifs_sfu_make_node(xid, inode, dentry, tcon,
+				  full_path, mode, dev);
+	if (rc)
+		return rc;
+
+	if (tcon->posix_extensions) {
+		rc = smb311_posix_get_inode_info(&new, full_path, NULL,
+						 inode->i_sb, xid);
+	} else if (tcon->unix_ext) {
+		rc = cifs_get_inode_info_unix(&new, full_path,
+					      inode->i_sb, xid);
+	} else {
+		rc = cifs_get_inode_info(&new, full_path, NULL,
+					 inode->i_sb, xid, NULL);
+	}
+	if (!rc)
+		d_instantiate(dentry, new);
 	return rc;
 }
 
diff --git a/fs/smb/client/smb2pdu.c b/fs/smb/client/smb2pdu.c
index 3ea688558e6c..86c647a947cc 100644
--- a/fs/smb/client/smb2pdu.c
+++ b/fs/smb/client/smb2pdu.c
@@ -367,6 +367,17 @@ again:
 		}
 
 		rc = cifs_setup_session(0, ses, server, nls_codepage);
+		if ((rc == -EACCES) || (rc == -EKEYEXPIRED) || (rc == -EKEYREVOKED)) {
+			/*
+			 * Try alternate password for next reconnect (key rotation
+			 * could be enabled on the server e.g.) if an alternate
+			 * password is available and the current password is expired,
+			 * but do not swap on non pwd related errors like host down
+			 */
+			if (ses->password2)
+				swap(ses->password2, ses->password);
+		}
+
 		if ((rc == -EACCES) && !tcon->retry) {
 			mutex_unlock(&ses->session_mutex);
 			rc = -EHOSTDOWN;
@@ -3628,9 +3639,9 @@ replay_again:
 			memcpy(&pbuf->network_open_info,
 			       &rsp->network_open_info,
 			       sizeof(pbuf->network_open_info));
+		atomic_dec(&tcon->num_remote_opens);
 	}
 
-	atomic_dec(&tcon->num_remote_opens);
 close_exit:
 	SMB2_close_free(&rqst);
 	free_rsp_buf(resp_buftype, rsp);
diff --git a/fs/smb/client/smb2transport.c b/fs/smb/client/smb2transport.c
index 5a3ca62d2f07..1d6e54f7879e 100644
--- a/fs/smb/client/smb2transport.c
+++ b/fs/smb/client/smb2transport.c
@@ -659,7 +659,7 @@ smb2_sign_rqst(struct smb_rqst *rqst, struct TCP_Server_Info *server)
 	}
 	spin_unlock(&server->srv_lock);
 	if (!is_binding && !server->session_estab) {
-		strncpy(shdr->Signature, "BSRSPYL", 8);
+		strscpy(shdr->Signature, "BSRSPYL");
 		return 0;
 	}
 
diff --git a/fs/smb/server/ksmbd_netlink.h b/fs/smb/server/ksmbd_netlink.h
index 8ca8a45c4c62..686b321c5a8b 100644
--- a/fs/smb/server/ksmbd_netlink.h
+++ b/fs/smb/server/ksmbd_netlink.h
@@ -167,7 +167,8 @@ struct ksmbd_share_config_response {
 	__u16	force_uid;
 	__u16	force_gid;
 	__s8	share_name[KSMBD_REQ_MAX_SHARE_NAME];
-	__u32	reserved[112];		/* Reserved room */
+	__u32	reserved[111];		/* Reserved room */
+	__u32	payload_sz;
 	__u32	veto_list_sz;
 	__s8	____payload[];
 };
diff --git a/fs/smb/server/mgmt/share_config.c b/fs/smb/server/mgmt/share_config.c
index 328a412259dc..a2f0a2edceb8 100644
--- a/fs/smb/server/mgmt/share_config.c
+++ b/fs/smb/server/mgmt/share_config.c
@@ -158,7 +158,12 @@ static struct ksmbd_share_config *share_config_request(struct unicode_map *um,
 	share->name = kstrdup(name, GFP_KERNEL);
 
 	if (!test_share_config_flag(share, KSMBD_SHARE_FLAG_PIPE)) {
-		share->path = kstrdup(ksmbd_share_config_path(resp),
+		int path_len = PATH_MAX;
+
+		if (resp->payload_sz)
+			path_len = resp->payload_sz - resp->veto_list_sz;
+
+		share->path = kstrndup(ksmbd_share_config_path(resp), path_len,
 				      GFP_KERNEL);
 		if (share->path)
 			share->path_sz = strlen(share->path);
diff --git a/fs/smb/server/smb2ops.c b/fs/smb/server/smb2ops.c
index a45f7dca482e..606aa3c5189a 100644
--- a/fs/smb/server/smb2ops.c
+++ b/fs/smb/server/smb2ops.c
@@ -228,6 +228,11 @@ void init_smb3_0_server(struct ksmbd_conn *conn)
 	    conn->cli_cap & SMB2_GLOBAL_CAP_ENCRYPTION)
 		conn->vals->capabilities |= SMB2_GLOBAL_CAP_ENCRYPTION;
 
+	if (server_conf.flags & KSMBD_GLOBAL_FLAG_SMB2_ENCRYPTION ||
+	    (!(server_conf.flags & KSMBD_GLOBAL_FLAG_SMB2_ENCRYPTION_OFF) &&
+	     conn->cli_cap & SMB2_GLOBAL_CAP_ENCRYPTION))
+		conn->vals->capabilities |= SMB2_GLOBAL_CAP_ENCRYPTION;
+
 	if (server_conf.flags & KSMBD_GLOBAL_FLAG_SMB3_MULTICHANNEL)
 		conn->vals->capabilities |= SMB2_GLOBAL_CAP_MULTI_CHANNEL;
 }
@@ -278,11 +283,6 @@ int init_smb3_11_server(struct ksmbd_conn *conn)
 		conn->vals->capabilities |= SMB2_GLOBAL_CAP_LEASING |
 			SMB2_GLOBAL_CAP_DIRECTORY_LEASING;
 
-	if (server_conf.flags & KSMBD_GLOBAL_FLAG_SMB2_ENCRYPTION ||
-	    (!(server_conf.flags & KSMBD_GLOBAL_FLAG_SMB2_ENCRYPTION_OFF) &&
-	     conn->cli_cap & SMB2_GLOBAL_CAP_ENCRYPTION))
-		conn->vals->capabilities |= SMB2_GLOBAL_CAP_ENCRYPTION;
-
 	if (server_conf.flags & KSMBD_GLOBAL_FLAG_SMB3_MULTICHANNEL)
 		conn->vals->capabilities |= SMB2_GLOBAL_CAP_MULTI_CHANNEL;
 
diff --git a/fs/smb/server/smb2pdu.c b/fs/smb/server/smb2pdu.c
index d478fa0c57ab..5723bbf372d7 100644
--- a/fs/smb/server/smb2pdu.c
+++ b/fs/smb/server/smb2pdu.c
@@ -5857,8 +5857,9 @@ static int smb2_rename(struct ksmbd_work *work,
 	if (!file_info->ReplaceIfExists)
 		flags = RENAME_NOREPLACE;
 
-	smb_break_all_levII_oplock(work, fp, 0);
 	rc = ksmbd_vfs_rename(work, &fp->filp->f_path, new_name, flags);
+	if (!rc)
+		smb_break_all_levII_oplock(work, fp, 0);
 out:
 	kfree(new_name);
 	return rc;
diff --git a/fs/smb/server/transport_ipc.c b/fs/smb/server/transport_ipc.c
index f29bb03f0dc4..8752ac82c557 100644
--- a/fs/smb/server/transport_ipc.c
+++ b/fs/smb/server/transport_ipc.c
@@ -65,6 +65,7 @@ struct ipc_msg_table_entry {
 	struct hlist_node	ipc_table_hlist;
 
 	void			*response;
+	unsigned int		msg_sz;
 };
 
 static struct delayed_work ipc_timer_work;
@@ -275,6 +276,7 @@ static int handle_response(int type, void *payload, size_t sz)
 		}
 
 		memcpy(entry->response, payload, sz);
+		entry->msg_sz = sz;
 		wake_up_interruptible(&entry->wait);
 		ret = 0;
 		break;
@@ -453,6 +455,34 @@ out:
 	return ret;
 }
 
+static int ipc_validate_msg(struct ipc_msg_table_entry *entry)
+{
+	unsigned int msg_sz = entry->msg_sz;
+
+	if (entry->type == KSMBD_EVENT_RPC_REQUEST) {
+		struct ksmbd_rpc_command *resp = entry->response;
+
+		msg_sz = sizeof(struct ksmbd_rpc_command) + resp->payload_sz;
+	} else if (entry->type == KSMBD_EVENT_SPNEGO_AUTHEN_REQUEST) {
+		struct ksmbd_spnego_authen_response *resp = entry->response;
+
+		msg_sz = sizeof(struct ksmbd_spnego_authen_response) +
+				resp->session_key_len + resp->spnego_blob_len;
+	} else if (entry->type == KSMBD_EVENT_SHARE_CONFIG_REQUEST) {
+		struct ksmbd_share_config_response *resp = entry->response;
+
+		if (resp->payload_sz) {
+			if (resp->payload_sz < resp->veto_list_sz)
+				return -EINVAL;
+
+			msg_sz = sizeof(struct ksmbd_share_config_response) +
+					resp->payload_sz;
+		}
+	}
+
+	return entry->msg_sz != msg_sz ? -EINVAL : 0;
+}
+
 static void *ipc_msg_send_request(struct ksmbd_ipc_msg *msg, unsigned int handle)
 {
 	struct ipc_msg_table_entry entry;
@@ -477,6 +507,13 @@ static void *ipc_msg_send_request(struct ksmbd_ipc_msg *msg, unsigned int handle
 	ret = wait_event_interruptible_timeout(entry.wait,
 					       entry.response != NULL,
 					       IPC_WAIT_TIMEOUT);
+	if (entry.response) {
+		ret = ipc_validate_msg(&entry);
+		if (ret) {
+			kvfree(entry.response);
+			entry.response = NULL;
+		}
+	}
 out:
 	down_write(&ipc_msg_table_lock);
 	hash_del(&entry.ipc_table_hlist);
diff --git a/fs/squashfs/inode.c b/fs/squashfs/inode.c
index aa3411354e66..16bd693d0b3a 100644
--- a/fs/squashfs/inode.c
+++ b/fs/squashfs/inode.c
@@ -48,6 +48,10 @@ static int squashfs_new_inode(struct super_block *sb, struct inode *inode,
 	gid_t i_gid;
 	int err;
 
+	inode->i_ino = le32_to_cpu(sqsh_ino->inode_number);
+	if (inode->i_ino == 0)
+		return -EINVAL;
+
 	err = squashfs_get_id(sb, le16_to_cpu(sqsh_ino->uid), &i_uid);
 	if (err)
 		return err;
@@ -58,7 +62,6 @@ static int squashfs_new_inode(struct super_block *sb, struct inode *inode,
 
 	i_uid_write(inode, i_uid);
 	i_gid_write(inode, i_gid);
-	inode->i_ino = le32_to_cpu(sqsh_ino->inode_number);
 	inode_set_mtime(inode, le32_to_cpu(sqsh_ino->mtime), 0);
 	inode_set_atime(inode, inode_get_mtime_sec(inode), 0);
 	inode_set_ctime(inode, inode_get_mtime_sec(inode), 0);
diff --git a/fs/super.c b/fs/super.c
index 71d9779c42b1..69ce6c600968 100644
--- a/fs/super.c
+++ b/fs/super.c
@@ -1515,29 +1515,11 @@ static int fs_bdev_thaw(struct block_device *bdev)
 	return error;
 }
 
-static void fs_bdev_super_get(void *data)
-{
-	struct super_block *sb = data;
-
-	spin_lock(&sb_lock);
-	sb->s_count++;
-	spin_unlock(&sb_lock);
-}
-
-static void fs_bdev_super_put(void *data)
-{
-	struct super_block *sb = data;
-
-	put_super(sb);
-}
-
 const struct blk_holder_ops fs_holder_ops = {
 	.mark_dead		= fs_bdev_mark_dead,
 	.sync			= fs_bdev_sync,
 	.freeze			= fs_bdev_freeze,
 	.thaw			= fs_bdev_thaw,
-	.get_holder		= fs_bdev_super_get,
-	.put_holder		= fs_bdev_super_put,
 };
 EXPORT_SYMBOL_GPL(fs_holder_ops);
 
@@ -1562,7 +1544,7 @@ int setup_bdev_super(struct super_block *sb, int sb_flags,
 	 * writable from userspace even for a read-only block device.
 	 */
 	if ((mode & BLK_OPEN_WRITE) && bdev_read_only(bdev)) {
-		fput(bdev_file);
+		bdev_fput(bdev_file);
 		return -EACCES;
 	}
 
@@ -1573,7 +1555,7 @@ int setup_bdev_super(struct super_block *sb, int sb_flags,
 	if (atomic_read(&bdev->bd_fsfreeze_count) > 0) {
 		if (fc)
 			warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev);
-		fput(bdev_file);
+		bdev_fput(bdev_file);
 		return -EBUSY;
 	}
 	spin_lock(&sb_lock);
@@ -1693,7 +1675,7 @@ void kill_block_super(struct super_block *sb)
 	generic_shutdown_super(sb);
 	if (bdev) {
 		sync_blockdev(bdev);
-		fput(sb->s_bdev_file);
+		bdev_fput(sb->s_bdev_file);
 	}
 }
 
diff --git a/fs/sysfs/file.c b/fs/sysfs/file.c
index 6b7652fb8050..7cd64021d453 100644
--- a/fs/sysfs/file.c
+++ b/fs/sysfs/file.c
@@ -463,6 +463,8 @@ struct kernfs_node *sysfs_break_active_protection(struct kobject *kobj,
 	kn = kernfs_find_and_get(kobj->sd, attr->name);
 	if (kn)
 		kernfs_break_active_protection(kn);
+	else
+		kobject_put(kobj);
 	return kn;
 }
 EXPORT_SYMBOL_GPL(sysfs_break_active_protection);
diff --git a/fs/tracefs/event_inode.c b/fs/tracefs/event_inode.c
index dc067eeb6387..894c6ca1e500 100644
--- a/fs/tracefs/event_inode.c
+++ b/fs/tracefs/event_inode.c
@@ -336,6 +336,7 @@ static void update_inode_attr(struct dentry *dentry, struct inode *inode,
 
 /**
  * lookup_file - look up a file in the tracefs filesystem
+ * @parent_ei: Pointer to the eventfs_inode that represents parent of the file
  * @dentry: the dentry to look up
  * @mode: the permission that the file should have.
  * @attr: saved attributes changed by user
@@ -389,6 +390,7 @@ static struct dentry *lookup_file(struct eventfs_inode *parent_ei,
 /**
  * lookup_dir_entry - look up a dir in the tracefs filesystem
  * @dentry: the directory to look up
+ * @pei: Pointer to the parent eventfs_inode if available
  * @ei: the eventfs_inode that represents the directory to create
  *
  * This function will look up a dentry for a directory represented by
@@ -478,16 +480,20 @@ void eventfs_d_release(struct dentry *dentry)
 
 /**
  * lookup_file_dentry - create a dentry for a file of an eventfs_inode
+ * @dentry: The parent dentry under which the new file's dentry will be created
  * @ei: the eventfs_inode that the file will be created under
  * @idx: the index into the entry_attrs[] of the @ei
- * @parent: The parent dentry of the created file.
- * @name: The name of the file to create
  * @mode: The mode of the file.
  * @data: The data to use to set the inode of the file with on open()
  * @fops: The fops of the file to be created.
  *
- * Create a dentry for a file of an eventfs_inode @ei and place it into the
- * address located at @e_dentry.
+ * This function creates a dentry for a file associated with an
+ * eventfs_inode @ei. It uses the entry attributes specified by @idx,
+ * if available. The file will have the specified @mode and its inode will be
+ * set up with @data upon open. The file operations will be set to @fops.
+ *
+ * Return: Returns a pointer to the newly created file's dentry or an error
+ * pointer.
  */
 static struct dentry *
 lookup_file_dentry(struct dentry *dentry,
diff --git a/fs/vboxsf/file.c b/fs/vboxsf/file.c
index 2307f8037efc..118dedef8ebe 100644
--- a/fs/vboxsf/file.c
+++ b/fs/vboxsf/file.c
@@ -218,6 +218,7 @@ const struct file_operations vboxsf_reg_fops = {
 	.release = vboxsf_file_release,
 	.fsync = noop_fsync,
 	.splice_read = filemap_splice_read,
+	.setlease = simple_nosetlease,
 };
 
 const struct inode_operations vboxsf_reg_iops = {
diff --git a/fs/vboxsf/super.c b/fs/vboxsf/super.c
index cabe8ac4fefc..ffb1d565da39 100644
--- a/fs/vboxsf/super.c
+++ b/fs/vboxsf/super.c
@@ -151,11 +151,11 @@ static int vboxsf_fill_super(struct super_block *sb, struct fs_context *fc)
 		if (!sbi->nls) {
 			vbg_err("vboxsf: Count not load '%s' nls\n", nls_name);
 			err = -EINVAL;
-			goto fail_free;
+			goto fail_destroy_idr;
 		}
 	}
 
-	sbi->bdi_id = ida_simple_get(&vboxsf_bdi_ida, 0, 0, GFP_KERNEL);
+	sbi->bdi_id = ida_alloc(&vboxsf_bdi_ida, GFP_KERNEL);
 	if (sbi->bdi_id < 0) {
 		err = sbi->bdi_id;
 		goto fail_free;
@@ -221,9 +221,10 @@ fail_unmap:
 	vboxsf_unmap_folder(sbi->root);
 fail_free:
 	if (sbi->bdi_id >= 0)
-		ida_simple_remove(&vboxsf_bdi_ida, sbi->bdi_id);
+		ida_free(&vboxsf_bdi_ida, sbi->bdi_id);
 	if (sbi->nls)
 		unload_nls(sbi->nls);
+fail_destroy_idr:
 	idr_destroy(&sbi->ino_idr);
 	kfree(sbi);
 	return err;
@@ -268,7 +269,7 @@ static void vboxsf_put_super(struct super_block *sb)
 
 	vboxsf_unmap_folder(sbi->root);
 	if (sbi->bdi_id >= 0)
-		ida_simple_remove(&vboxsf_bdi_ida, sbi->bdi_id);
+		ida_free(&vboxsf_bdi_ida, sbi->bdi_id);
 	if (sbi->nls)
 		unload_nls(sbi->nls);
 
diff --git a/fs/vboxsf/utils.c b/fs/vboxsf/utils.c
index 72ac9320e6a3..9515bbf0b54c 100644
--- a/fs/vboxsf/utils.c
+++ b/fs/vboxsf/utils.c
@@ -440,7 +440,6 @@ int vboxsf_nlscpy(struct vboxsf_sbi *sbi, char *name, size_t name_bound_len,
 {
 	const char *in;
 	char *out;
-	size_t out_len;
 	size_t out_bound_len;
 	size_t in_bound_len;
 
@@ -448,7 +447,6 @@ int vboxsf_nlscpy(struct vboxsf_sbi *sbi, char *name, size_t name_bound_len,
 	in_bound_len = utf8_len;
 
 	out = name;
-	out_len = 0;
 	/* Reserve space for terminating 0 */
 	out_bound_len = name_bound_len - 1;
 
@@ -469,7 +467,6 @@ int vboxsf_nlscpy(struct vboxsf_sbi *sbi, char *name, size_t name_bound_len,
 
 		out += nb;
 		out_bound_len -= nb;
-		out_len += nb;
 	}
 
 	*out = 0;
diff --git a/fs/xfs/xfs_buf.c b/fs/xfs/xfs_buf.c
index 1a18c381127e..f0fa02264eda 100644
--- a/fs/xfs/xfs_buf.c
+++ b/fs/xfs/xfs_buf.c
@@ -2030,7 +2030,7 @@ xfs_free_buftarg(
 	fs_put_dax(btp->bt_daxdev, btp->bt_mount);
 	/* the main block device is closed by kill_block_super */
 	if (btp->bt_bdev != btp->bt_mount->m_super->s_bdev)
-		fput(btp->bt_bdev_file);
+		bdev_fput(btp->bt_bdev_file);
 	kfree(btp);
 }
 
diff --git a/fs/xfs/xfs_inode.c b/fs/xfs/xfs_inode.c
index ea48774f6b76..d55b42b2480d 100644
--- a/fs/xfs/xfs_inode.c
+++ b/fs/xfs/xfs_inode.c
@@ -1301,8 +1301,19 @@ xfs_link(
 	 */
 	if (unlikely((tdp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
 		     tdp->i_projid != sip->i_projid)) {
-		error = -EXDEV;
-		goto error_return;
+		/*
+		 * Project quota setup skips special files which can
+		 * leave inodes in a PROJINHERIT directory without a
+		 * project ID set. We need to allow links to be made
+		 * to these "project-less" inodes because userspace
+		 * expects them to succeed after project ID setup,
+		 * but everything else should be rejected.
+		 */
+		if (!special_file(VFS_I(sip)->i_mode) ||
+		    sip->i_projid != 0) {
+			error = -EXDEV;
+			goto error_return;
+		}
 	}
 
 	if (!resblks) {
diff --git a/fs/xfs/xfs_super.c b/fs/xfs/xfs_super.c
index c21f10ab0f5d..bce020374c5e 100644
--- a/fs/xfs/xfs_super.c
+++ b/fs/xfs/xfs_super.c
@@ -485,7 +485,7 @@ xfs_open_devices(
 		mp->m_logdev_targp = mp->m_ddev_targp;
 		/* Handle won't be used, drop it */
 		if (logdev_file)
-			fput(logdev_file);
+			bdev_fput(logdev_file);
 	}
 
 	return 0;
@@ -497,10 +497,10 @@ xfs_open_devices(
 	xfs_free_buftarg(mp->m_ddev_targp);
  out_close_rtdev:
 	 if (rtdev_file)
-		fput(rtdev_file);
+		bdev_fput(rtdev_file);
  out_close_logdev:
 	if (logdev_file)
-		fput(logdev_file);
+		bdev_fput(logdev_file);
 	return error;
 }
 
diff --git a/fs/zonefs/super.c b/fs/zonefs/super.c
index c6a124e8d565..964fa7f24003 100644
--- a/fs/zonefs/super.c
+++ b/fs/zonefs/super.c
@@ -1048,7 +1048,7 @@ static int zonefs_init_zgroup(struct super_block *sb,
 	zonefs_info(sb, "Zone group \"%s\" has %u file%s\n",
 		    zonefs_zgroup_name(ztype),
 		    zgroup->g_nr_zones,
-		    zgroup->g_nr_zones > 1 ? "s" : "");
+		    str_plural(zgroup->g_nr_zones));
 
 	return 0;
 }