Merge tag 'bcachefs-2025-04-17' of git://evilpiepirate.org/bcachefs

Pull bcachefs fixes from Kent Overstreet:
 "Usual set of small fixes/logging improvements.

  One bigger user reported fix, for inode <-> dirent inconsistencies
  reported in fsck, after moving a subvolume that had been snapshotted"

* tag 'bcachefs-2025-04-17' of git://evilpiepirate.org/bcachefs:
  bcachefs: Fix snapshotting a subvolume, then renaming it
  bcachefs: Add missing READ_ONCE() for metadata replicas
  bcachefs: snapshot_node_missing is now autofix
  bcachefs: Log message when incompat version requested but not enabled
  bcachefs: Print version_incompat_allowed on startup
  bcachefs: Silence extent_poisoned error messages
  bcachefs: btree_root_unreadable_and_scan_found_nothing now AUTOFIX
  bcachefs: fix bch2_dev_usage_full_read_fast()
  bcachefs: Don't print data read retry success on non-errors
  bcachefs: Add missing error handling
  bcachefs: Prevent granting write refs when filesystem is read-only

Author: torvalds

fs/bcachefs/bcachefs.h

@@ -788,6 +788,8 @@ struct bch_fs {
 		unsigned long	errors_silent[BITS_TO_LONGS(BCH_FSCK_ERR_MAX)];
 		u64		btrees_lost_data;
 	}			sb;
+	DARRAY(enum bcachefs_metadata_version)
+				incompat_versions_requested;
 
 #ifdef CONFIG_UNICODE
 	struct unicode_map	*cf_encoding;

fs/bcachefs/btree_update_interior.c

@@ -1221,7 +1221,7 @@ bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
 
 	ret = bch2_disk_reservation_get(c, &as->disk_res,
 			(nr_nodes[0] + nr_nodes[1]) * btree_sectors(c),
-			c->opts.metadata_replicas,
+			READ_ONCE(c->opts.metadata_replicas),
 			disk_res_flags);
 	if (ret)
 		goto err;

fs/bcachefs/buckets.c

@@ -37,7 +37,8 @@ void bch2_dev_usage_read_fast(struct bch_dev *ca, struct bch_dev_usage *usage)
 void bch2_dev_usage_full_read_fast(struct bch_dev *ca, struct bch_dev_usage_full *usage)
 {
 	memset(usage, 0, sizeof(*usage));
-	acc_u64s_percpu((u64 *) usage, (u64 __percpu *) ca->usage, dev_usage_u64s());
+	acc_u64s_percpu((u64 *) usage, (u64 __percpu *) ca->usage,
+			sizeof(struct bch_dev_usage_full) / sizeof(u64));
 }
 
 static u64 reserve_factor(u64 r)

fs/bcachefs/buckets.h

@@ -242,11 +242,6 @@ static inline u64 dev_buckets_available(struct bch_dev *ca,
 
 /* Filesystem usage: */
 
-static inline unsigned dev_usage_u64s(void)
-{
-	return sizeof(struct bch_dev_usage) / sizeof(u64);
-}
-
 struct bch_fs_usage_short
 bch2_fs_usage_read_short(struct bch_fs *);
 

fs/bcachefs/errcode.h

@@ -287,7 +287,7 @@
 	x(EIO,				mark_stripe)				\
 	x(EIO,				stripe_reconstruct)			\
 	x(EIO,				key_type_error)				\
-	x(EIO,				extent_poisened)			\
+	x(EIO,				extent_poisoned)			\
 	x(EIO,				missing_indirect_extent)		\
 	x(EIO,				invalidate_stripe_to_dev)		\
 	x(EIO,				no_encryption_key)			\

fs/bcachefs/extents.c

@@ -139,7 +139,7 @@ int bch2_bkey_pick_read_device(struct bch_fs *c, struct bkey_s_c k,
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 
 	if (bch2_bkey_extent_ptrs_flags(ptrs) & BIT_ULL(BCH_EXTENT_FLAG_poisoned))
-		return -BCH_ERR_extent_poisened;
+		return -BCH_ERR_extent_poisoned;
 
 	rcu_read_lock();
 	const union bch_extent_entry *entry;

fs/bcachefs/fs-ioctl.c

@@ -69,7 +69,7 @@ static int bch2_inode_flags_set(struct btree_trans *trans,
 		if (ret < 0)
 			return ret;
 
-		ret = bch2_request_incompat_feature(c,bcachefs_metadata_version_casefolding);
+		ret = bch2_request_incompat_feature(c, bcachefs_metadata_version_casefolding);
 		if (ret)
 			return ret;
 

fs/bcachefs/fsck.c

@@ -321,6 +321,31 @@ static inline bool inode_should_reattach(struct bch_inode_unpacked *inode)
 	    inode->bi_subvol == BCACHEFS_ROOT_SUBVOL)
 		return false;
 
+	/*
+	 * Subvolume roots are special: older versions of subvolume roots may be
+	 * disconnected, it's only the newest version that matters.
+	 *
+	 * We only keep a single dirent pointing to a subvolume root, i.e.
+	 * older versions of snapshots will not have a different dirent pointing
+	 * to the same subvolume root.
+	 *
+	 * This is because dirents that point to subvolumes are only visible in
+	 * the parent subvolume - versioning is not needed - and keeping them
+	 * around would break fsck, because when we're crossing subvolumes we
+	 * don't have a consistent snapshot ID to do check the inode <-> dirent
+	 * relationships.
+	 *
+	 * Thus, a subvolume root that's been renamed after a snapshot will have
+	 * a disconnected older version - that's expected.
+	 *
+	 * Note that taking a snapshot always updates the root inode (to update
+	 * the dirent backpointer), so a subvolume root inode with
+	 * BCH_INODE_has_child_snapshot is never visible.
+	 */
+	if (inode->bi_subvol &&
+	    (inode->bi_flags & BCH_INODE_has_child_snapshot))
+		return false;
+
 	return !inode->bi_dir && !(inode->bi_flags & BCH_INODE_unlinked);
 }
 
@@ -1007,6 +1032,23 @@ static int check_inode_dirent_inode(struct btree_trans *trans,
 	if (ret && !bch2_err_matches(ret, ENOENT))
 		return ret;
 
+	if ((ret || dirent_points_to_inode_nowarn(d, inode)) &&
+	    inode->bi_subvol &&
+	    (inode->bi_flags & BCH_INODE_has_child_snapshot)) {
+		/* Older version of a renamed subvolume root: we won't have a
+		 * correct dirent for it. That's expected, see
+		 * inode_should_reattach().
+		 *
+		 * We don't clear the backpointer field when doing the rename
+		 * because there might be arbitrarily many versions in older
+		 * snapshots.
+		 */
+		inode->bi_dir = 0;
+		inode->bi_dir_offset = 0;
+		*write_inode = true;
+		goto out;
+	}
+
 	if (fsck_err_on(ret,
 			trans, inode_points_to_missing_dirent,
 			"inode points to missing dirent\n%s",
@@ -1027,7 +1069,7 @@ static int check_inode_dirent_inode(struct btree_trans *trans,
 		inode->bi_dir_offset = 0;
 		*write_inode = true;
 	}
-
+out:
 	ret = 0;
 fsck_err:
 	bch2_trans_iter_exit(trans, &dirent_iter);

fs/bcachefs/io_read.c

@@ -487,6 +487,8 @@ static void bch2_rbio_retry(struct work_struct *work)
 		.inum	= rbio->read_pos.inode,
 	};
 	struct bch_io_failures failed = { .nr = 0 };
+	int orig_error = rbio->ret;
+
 	struct btree_trans *trans = bch2_trans_get(c);
 
 	trace_io_read_retry(&rbio->bio);
@@ -519,7 +521,9 @@ static void bch2_rbio_retry(struct work_struct *work)
 	if (ret) {
 		rbio->ret = ret;
 		rbio->bio.bi_status = BLK_STS_IOERR;
-	} else {
+	} else if (orig_error != -BCH_ERR_data_read_retry_csum_err_maybe_userspace &&
+		   orig_error != -BCH_ERR_data_read_ptr_stale_race &&
+		   !failed.nr) {
 		struct printbuf buf = PRINTBUF;
 
 		lockrestart_do(trans,
@@ -1345,14 +1349,16 @@ int __bch2_read(struct btree_trans *trans, struct bch_read_bio *rbio,
 
 	bch2_trans_iter_exit(trans, &iter);
 
-	if (ret) {
-		struct printbuf buf = PRINTBUF;
-		lockrestart_do(trans,
-			bch2_inum_offset_err_msg_trans(trans, &buf, inum,
-						       bvec_iter.bi_sector << 9));
-		prt_printf(&buf, "read error: %s", bch2_err_str(ret));
-		bch_err_ratelimited(c, "%s", buf.buf);
-		printbuf_exit(&buf);
+	if (unlikely(ret)) {
+		if (ret != -BCH_ERR_extent_poisoned) {
+			struct printbuf buf = PRINTBUF;
+			lockrestart_do(trans,
+				       bch2_inum_offset_err_msg_trans(trans, &buf, inum,
+								      bvec_iter.bi_sector << 9));
+			prt_printf(&buf, "data read error: %s", bch2_err_str(ret));
+			bch_err_ratelimited(c, "%s", buf.buf);
+			printbuf_exit(&buf);
+		}
 
 		rbio->bio.bi_status	= BLK_STS_IOERR;
 		rbio->ret		= ret;

fs/bcachefs/recovery.c

@@ -1125,7 +1125,10 @@ int bch2_fs_initialize(struct bch_fs *c)
 	 * journal_res_get() will crash if called before this has
 	 * set up the journal.pin FIFO and journal.cur pointer:
 	 */
-	bch2_fs_journal_start(&c->journal, 1);
+	ret = bch2_fs_journal_start(&c->journal, 1);
+	if (ret)
+		goto err;
+
 	set_bit(BCH_FS_accounting_replay_done, &c->flags);
 	bch2_journal_set_replay_done(&c->journal);