Merge tag 'bcachefs-2025-07-03' of git://evilpiepirate.org/bcachefs

Pull bcachefs fixes from Kent Overstreet:
 "The 'opts.casefold_disabled' patch is non critical, but would be a
  6.15 backport; it's to address the casefolding + overlayfs
  incompatibility that was discovvered late.

  It's late because I was hoping that this would be addressed on the
  overlayfs side (and will be in 6.17), but user reports keep coming in
  on this one (lots of people are using docker these days)"

* tag 'bcachefs-2025-07-03' of git://evilpiepirate.org/bcachefs:
  bcachefs: opts.casefold_disabled
  bcachefs: Work around deadlock to btree node rewrites in journal replay
  bcachefs: Fix incorrect transaction restart handling
  bcachefs: fix btree_trans_peek_prev_journal()
  bcachefs: mark invalid_btree_id autofix

Author: torvalds

fs/bcachefs/bcachefs.h

@@ -863,9 +863,7 @@ struct bch_fs {
 	DARRAY(enum bcachefs_metadata_version)
 				incompat_versions_requested;
 
-#ifdef CONFIG_UNICODE
 	struct unicode_map	*cf_encoding;
-#endif
 
 	struct bch_sb_handle	disk_sb;
 
@@ -1285,4 +1283,13 @@ static inline bool bch2_discard_opt_enabled(struct bch_fs *c, struct bch_dev *ca
 		: ca->mi.discard;
 }
 
+static inline bool bch2_fs_casefold_enabled(struct bch_fs *c)
+{
+#ifdef CONFIG_UNICODE
+	return !c->opts.casefold_disabled;
+#else
+	return false;
+#endif
+}
+
 #endif /* _BCACHEFS_H */

fs/bcachefs/btree_io.c

@@ -1337,15 +1337,42 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca,
 
 	btree_node_reset_sib_u64s(b);
 
-	scoped_guard(rcu)
-		bkey_for_each_ptr(bch2_bkey_ptrs(bkey_i_to_s(&b->key)), ptr) {
-			struct bch_dev *ca2 = bch2_dev_rcu(c, ptr->dev);
-
-			if (!ca2 || ca2->mi.state != BCH_MEMBER_STATE_rw) {
-				set_btree_node_need_rewrite(b);
-				set_btree_node_need_rewrite_degraded(b);
+	/*
+	 * XXX:
+	 *
+	 * We deadlock if too many btree updates require node rewrites while
+	 * we're still in journal replay.
+	 *
+	 * This is because btree node rewrites generate more updates for the
+	 * interior updates (alloc, backpointers), and if those updates touch
+	 * new nodes and generate more rewrites - well, you see the problem.
+	 *
+	 * The biggest cause is that we don't use the btree write buffer (for
+	 * the backpointer updates - this needs some real thought on locking in
+	 * order to fix.
+	 *
+	 * The problem with this workaround (not doing the rewrite for degraded
+	 * nodes in journal replay) is that those degraded nodes persist, and we
+	 * don't want that (this is a real bug when a btree node write completes
+	 * with fewer replicas than we wanted and leaves a degraded node due to
+	 * device _removal_, i.e. the device went away mid write).
+	 *
+	 * It's less of a bug here, but still a problem because we don't yet
+	 * have a way of tracking degraded data - we another index (all
+	 * extents/btree nodes, by replicas entry) in order to fix properly
+	 * (re-replicate degraded data at the earliest possible time).
+	 */
+	if (c->recovery.passes_complete & BIT_ULL(BCH_RECOVERY_PASS_journal_replay)) {
+		scoped_guard(rcu)
+			bkey_for_each_ptr(bch2_bkey_ptrs(bkey_i_to_s(&b->key)), ptr) {
+				struct bch_dev *ca2 = bch2_dev_rcu(c, ptr->dev);
+
+				if (!ca2 || ca2->mi.state != BCH_MEMBER_STATE_rw) {
+					set_btree_node_need_rewrite(b);
+					set_btree_node_need_rewrite_degraded(b);
+				}
 			}
-		}
+	}
 
 	if (!ptr_written) {
 		set_btree_node_need_rewrite(b);

fs/bcachefs/btree_iter.c

@@ -2189,7 +2189,7 @@ void btree_trans_peek_prev_journal(struct btree_trans *trans,
 	struct btree_path *path = btree_iter_path(trans, iter);
 	struct bkey_i *next_journal =
 		bch2_btree_journal_peek_prev(trans, iter, search_key,
-				k->k ? k->k->p : path_l(path)->b->key.k.p);
+				k->k ? k->k->p : path_l(path)->b->data->min_key);
 
 	if (next_journal) {
 		iter->k = next_journal->k;

fs/bcachefs/dirent.c

@@ -18,7 +18,9 @@ int bch2_casefold(struct btree_trans *trans, const struct bch_hash_info *info,
 {
 	*out_cf = (struct qstr) QSTR_INIT(NULL, 0);
 
-#ifdef CONFIG_UNICODE
+	if (!bch2_fs_casefold_enabled(trans->c))
+		return -EOPNOTSUPP;
+
 	unsigned char *buf = bch2_trans_kmalloc(trans, BCH_NAME_MAX + 1);
 	int ret = PTR_ERR_OR_ZERO(buf);
 	if (ret)
@@ -30,9 +32,6 @@ int bch2_casefold(struct btree_trans *trans, const struct bch_hash_info *info,
 
 	*out_cf = (struct qstr) QSTR_INIT(buf, ret);
 	return 0;
-#else
-	return -EOPNOTSUPP;
-#endif
 }
 
 static unsigned bch2_dirent_name_bytes(struct bkey_s_c_dirent d)
@@ -231,7 +230,8 @@ void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c
 	prt_printf(out, " type %s", bch2_d_type_str(d.v->d_type));
 }
 
-int bch2_dirent_init_name(struct bkey_i_dirent *dirent,
+int bch2_dirent_init_name(struct bch_fs *c,
+			  struct bkey_i_dirent *dirent,
 			  const struct bch_hash_info *hash_info,
 			  const struct qstr *name,
 			  const struct qstr *cf_name)
@@ -251,7 +251,9 @@ int bch2_dirent_init_name(struct bkey_i_dirent *dirent,
 		       offsetof(struct bch_dirent, d_name) -
 		       name->len);
 	} else {
-#ifdef CONFIG_UNICODE
+		if (!bch2_fs_casefold_enabled(c))
+			return -EOPNOTSUPP;
+
 		memcpy(&dirent->v.d_cf_name_block.d_names[0], name->name, name->len);
 
 		char *cf_out = &dirent->v.d_cf_name_block.d_names[name->len];
@@ -277,9 +279,6 @@ int bch2_dirent_init_name(struct bkey_i_dirent *dirent,
 		dirent->v.d_cf_name_block.d_cf_name_len = cpu_to_le16(cf_len);
 
 		EBUG_ON(bch2_dirent_get_casefold_name(dirent_i_to_s_c(dirent)).len != cf_len);
-#else
-	return -EOPNOTSUPP;
-#endif
 	}
 
 	unsigned u64s = dirent_val_u64s(name->len, cf_len);
@@ -313,7 +312,7 @@ struct bkey_i_dirent *bch2_dirent_create_key(struct btree_trans *trans,
 	dirent->v.d_type = type;
 	dirent->v.d_unused = 0;
 
-	int ret = bch2_dirent_init_name(dirent, hash_info, name, cf_name);
+	int ret = bch2_dirent_init_name(trans->c, dirent, hash_info, name, cf_name);
 	if (ret)
 		return ERR_PTR(ret);
 

fs/bcachefs/dirent.h

@@ -59,7 +59,8 @@ static inline void dirent_copy_target(struct bkey_i_dirent *dst,
 	dst->v.d_type = src.v->d_type;
 }
 
-int bch2_dirent_init_name(struct bkey_i_dirent *,
+int bch2_dirent_init_name(struct bch_fs *,
+			  struct bkey_i_dirent *,
 			  const struct bch_hash_info *,
 			  const struct qstr *,
 			  const struct qstr *);

fs/bcachefs/fs.c

@@ -722,7 +722,6 @@ static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
 	if (IS_ERR(inode))
 		inode = NULL;
 
-#ifdef CONFIG_UNICODE
 	if (!inode && IS_CASEFOLDED(vdir)) {
 		/*
 		 * Do not cache a negative dentry in casefolded directories
@@ -737,7 +736,6 @@ static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
 		 */
 		return NULL;
 	}
-#endif
 
 	return d_splice_alias(&inode->v, dentry);
 }
@@ -2566,9 +2564,10 @@ static int bch2_fs_get_tree(struct fs_context *fc)
 	sb->s_shrink->seeks = 0;
 
 #ifdef CONFIG_UNICODE
-	sb->s_encoding = c->cf_encoding;
-#endif
+	if (bch2_fs_casefold_enabled(c))
+		sb->s_encoding = c->cf_encoding;
 	generic_set_sb_d_ops(sb);
+#endif
 
 	vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_SUBVOL_INUM);
 	ret = PTR_ERR_OR_ZERO(vinode);

fs/bcachefs/fsck.c

@@ -2302,9 +2302,7 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
 		*hash_info = bch2_hash_info_init(c, &i->inode);
 	dir->first_this_inode = false;
 
-#ifdef CONFIG_UNICODE
 	hash_info->cf_encoding = bch2_inode_casefold(c, &i->inode) ? c->cf_encoding : NULL;
-#endif
 
 	ret = bch2_str_hash_check_key(trans, s, &bch2_dirent_hash_desc, hash_info,
 				      iter, k, need_second_pass);
@@ -2819,7 +2817,7 @@ static int check_path_loop(struct btree_trans *trans, struct bkey_s_c inode_k)
 				ret = remove_backpointer(trans, &inode);
 				bch_err_msg(c, ret, "removing dirent");
 				if (ret)
-					break;
+					goto out;
 
 				ret = reattach_inode(trans, &inode);
 				bch_err_msg(c, ret, "reattaching inode %llu", inode.bi_inum);

fs/bcachefs/inode.c

@@ -1265,7 +1265,14 @@ int bch2_inode_set_casefold(struct btree_trans *trans, subvol_inum inum,
 {
 	struct bch_fs *c = trans->c;
 
-#ifdef CONFIG_UNICODE
+#ifndef CONFIG_UNICODE
+	bch_err(c, "Cannot use casefolding on a kernel without CONFIG_UNICODE");
+	return -EOPNOTSUPP;
+#endif
+
+	if (c->opts.casefold_disabled)
+		return -EOPNOTSUPP;
+
 	int ret = 0;
 	/* Not supported on individual files. */
 	if (!S_ISDIR(bi->bi_mode))
@@ -1289,10 +1296,6 @@ int bch2_inode_set_casefold(struct btree_trans *trans, subvol_inum inum,
 	bi->bi_fields_set |= BIT(Inode_opt_casefold);
 
 	return bch2_maybe_propagate_has_case_insensitive(trans, inum, bi);
-#else
-	bch_err(c, "Cannot use casefolding on a kernel without CONFIG_UNICODE");
-	return -EOPNOTSUPP;
-#endif
 }
 
 static noinline int __bch2_inode_rm_snapshot(struct btree_trans *trans, u64 inum, u32 snapshot)

fs/bcachefs/opts.h

@@ -234,6 +234,11 @@ enum fsck_err_opts {
 	  OPT_BOOL(),							\
 	  BCH_SB_CASEFOLD,		false,				\
 	  NULL,		"Dirent lookups are casefolded")		\
+	x(casefold_disabled,			u8,			\
+	  OPT_FS|OPT_MOUNT,						\
+	  OPT_BOOL(),							\
+	  BCH2_NO_SB_OPT,		false,				\
+	  NULL,		"Disable casefolding filesystem wide")		\
 	x(inodes_32bit,			u8,				\
 	  OPT_FS|OPT_INODE|OPT_FORMAT|OPT_MOUNT|OPT_RUNTIME,		\
 	  OPT_BOOL(),							\

fs/bcachefs/sb-errors_format.h

@@ -314,7 +314,7 @@ enum bch_fsck_flags {
 	x(accounting_mismatch,					272,	FSCK_AUTOFIX)	\
 	x(accounting_replicas_not_marked,			273,	0)		\
 	x(accounting_to_invalid_device,				289,	0)		\
-	x(invalid_btree_id,					274,	0)		\
+	x(invalid_btree_id,					274,	FSCK_AUTOFIX)		\
 	x(alloc_key_io_time_bad,				275,	0)		\
 	x(alloc_key_fragmentation_lru_wrong,			276,	FSCK_AUTOFIX)	\
 	x(accounting_key_junk_at_end,				277,	FSCK_AUTOFIX)	\