Merge tag 'bcachefs-2025-06-12' of git://evilpiepirate.org/bcachefs

Pull bcachefs fixes from Kent Overstreet:
 "As usual, highlighting the ones users have been noticing:

   - Fix a small issue with has_case_insensitive not being propagated on
     snapshot creation; this led to fsck errors, which we're harmless
     because we're not using this flag yet (it's for overlayfs +
     casefolding).

   - Log the error being corrected in the journal when we're doing fsck
     repair: this was one of the "lessons learned" from the i_nlink 0 ->
     subvolume deletion bug, where reconstructing what had happened by
     analyzing the journal was a bit more difficult than it needed to
     be.

   - Don't schedule btree node scan to run in the superblock: this fixes
     a regression from the 6.16 recovery passes rework, and let to it
     running unnecessarily.

     The real issue here is that we don't have online, "self healing"
     style topology repair yet: topology repair currently has to run
     before we go RW, which means that we may schedule it unnecessarily
     after a transient error. This will be fixed in the future.

   - We now track, in btree node flags, the reason it was scheduled to
     be rewritten. We discovered a deadlock in recovery when many btree
     nodes need to be rewritten because they're degraded: fully fixing
     this will take some work but it's now easier to see what's going
     on.

     For the bug report where this came up, a device had been kicked RO
     due to transient errors: manually setting it back to RW was
     sufficient to allow recovery to succeed.

   - Mark a few more fsck errors as autofix: as a reminder to users,
     please do keep reporting cases where something needs to be repaired
     and is not repaired automatically (i.e. cases where -o fix_errors
     or fsck -y is required).

   - rcu_pending.c now works with PREEMPT_RT

   - 'bcachefs device add', then umount, then remount wasn't working -
     we now emit a uevent so that the new device's new superblock is
     correctly picked up

   - Assorted repair fixes: btree node scan will no longer incorrectly
     update sb->version_min,

   - Assorted syzbot fixes"

* tag 'bcachefs-2025-06-12' of git://evilpiepirate.org/bcachefs: (23 commits)
  bcachefs: Don't trace should_be_locked unless changing
  bcachefs: Ensure that snapshot creation propagates has_case_insensitive
  bcachefs: Print devices we're mounting on multi device filesystems
  bcachefs: Don't trust sb->nr_devices in members_to_text()
  bcachefs: Fix version checks in validate_bset()
  bcachefs: ioctl: avoid stack overflow warning
  bcachefs: Don't pass trans to fsck_err() in gc_accounting_done
  bcachefs: Fix leak in bch2_fs_recovery() error path
  bcachefs: Fix rcu_pending for PREEMPT_RT
  bcachefs: Fix downgrade_table_extra()
  bcachefs: Don't put rhashtable on stack
  bcachefs: Make sure opts.read_only gets propagated back to VFS
  bcachefs: Fix possible console lock involved deadlock
  bcachefs: mark more errors autofix
  bcachefs: Don't persistently run scan_for_btree_nodes
  bcachefs: Read error message now prints if self healing
  bcachefs: Only run 'increase_depth' for keys from btree node csan
  bcachefs: Mark need_discard_freespace_key_bad autofix
  bcachefs: Update /dev/disk/by-uuid on device add
  bcachefs: Add more flags to btree nodes for rewrite reason
  ...

Author: torvalds

fs/bcachefs/bcachefs.h

@@ -296,7 +296,6 @@ do {									\
 #define bch2_fmt(_c, fmt)		bch2_log_msg(_c, fmt "\n")
 
 void bch2_print_str(struct bch_fs *, const char *, const char *);
-void bch2_print_str_nonblocking(struct bch_fs *, const char *, const char *);
 
 __printf(2, 3)
 void bch2_print_opts(struct bch_opts *, const char *, ...);

fs/bcachefs/btree_gc.c

@@ -397,7 +397,11 @@ static int bch2_btree_repair_topology_recurse(struct btree_trans *trans, struct
 			continue;
 		}
 
-		ret = btree_check_node_boundaries(trans, b, prev, cur, pulled_from_scan);
+		ret = lockrestart_do(trans,
+			btree_check_node_boundaries(trans, b, prev, cur, pulled_from_scan));
+		if (ret < 0)
+			goto err;
+
 		if (ret == DID_FILL_FROM_SCAN) {
 			new_pass = true;
 			ret = 0;
@@ -438,7 +442,8 @@ static int bch2_btree_repair_topology_recurse(struct btree_trans *trans, struct
 
 	if (!ret && !IS_ERR_OR_NULL(prev)) {
 		BUG_ON(cur);
-		ret = btree_repair_node_end(trans, b, prev, pulled_from_scan);
+		ret = lockrestart_do(trans,
+			btree_repair_node_end(trans, b, prev, pulled_from_scan));
 		if (ret == DID_FILL_FROM_SCAN) {
 			new_pass = true;
 			ret = 0;
@@ -519,49 +524,65 @@ static int bch2_btree_repair_topology_recurse(struct btree_trans *trans, struct
 	bch2_bkey_buf_exit(&prev_k, c);
 	bch2_bkey_buf_exit(&cur_k, c);
 	printbuf_exit(&buf);
+	bch_err_fn(c, ret);
+	return ret;
+}
+
+static int bch2_check_root(struct btree_trans *trans, enum btree_id i,
+			   bool *reconstructed_root)
+{
+	struct bch_fs *c = trans->c;
+	struct btree_root *r = bch2_btree_id_root(c, i);
+	struct printbuf buf = PRINTBUF;
+	int ret = 0;
+
+	bch2_btree_id_to_text(&buf, i);
+
+	if (r->error) {
+		bch_info(c, "btree root %s unreadable, must recover from scan", buf.buf);
+
+		r->alive = false;
+		r->error = 0;
+
+		if (!bch2_btree_has_scanned_nodes(c, i)) {
+			__fsck_err(trans,
+				   FSCK_CAN_FIX|(!btree_id_important(i) ? FSCK_AUTOFIX : 0),
+				   btree_root_unreadable_and_scan_found_nothing,
+				   "no nodes found for btree %s, continue?", buf.buf);
+			bch2_btree_root_alloc_fake_trans(trans, i, 0);
+		} else {
+			bch2_btree_root_alloc_fake_trans(trans, 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)
+				goto err;
+		}
+
+		*reconstructed_root = true;
+	}
+err:
+fsck_err:
+	printbuf_exit(&buf);
+	bch_err_fn(c, ret);
 	return ret;
 }
 
 int bch2_check_topology(struct bch_fs *c)
 {
 	struct btree_trans *trans = bch2_trans_get(c);
 	struct bpos pulled_from_scan = POS_MIN;
-	struct printbuf buf = PRINTBUF;
 	int ret = 0;
 
 	bch2_trans_srcu_unlock(trans);
 
 	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;
+recover:
+		ret = lockrestart_do(trans, bch2_check_root(trans, i, &reconstructed_root));
+		if (ret)
+			break;
 
-		printbuf_reset(&buf);
-		bch2_btree_id_to_text(&buf, i);
-
-		if (r->error) {
-reconstruct_root:
-			bch_info(c, "btree root %s unreadable, must recover from scan", buf.buf);
-
-			r->alive = false;
-			r->error = 0;
-
-			if (!bch2_btree_has_scanned_nodes(c, i)) {
-				__fsck_err(trans,
-					   FSCK_CAN_FIX|(!btree_id_important(i) ? FSCK_AUTOFIX : 0),
-					   btree_root_unreadable_and_scan_found_nothing,
-					   "no nodes found for btree %s, continue?", buf.buf);
-				bch2_btree_root_alloc_fake_trans(trans, i, 0);
-			} else {
-				bch2_btree_root_alloc_fake_trans(trans, 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_root *r = bch2_btree_id_root(c, i);
 		struct btree *b = r->b;
 
 		btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_read);
@@ -575,17 +596,21 @@ int bch2_check_topology(struct bch_fs *c)
 
 			r->b = NULL;
 
-			if (!reconstructed_root)
-				goto reconstruct_root;
+			if (!reconstructed_root) {
+				r->error = -EIO;
+				goto recover;
+			}
 
+			struct printbuf buf = PRINTBUF;
+			bch2_btree_id_to_text(&buf, i);
 			bch_err(c, "empty btree root %s", buf.buf);
+			printbuf_exit(&buf);
 			bch2_btree_root_alloc_fake_trans(trans, i, 0);
 			r->alive = false;
 			ret = 0;
 		}
 	}
-fsck_err:
-	printbuf_exit(&buf);
+
 	bch2_trans_put(trans);
 	return ret;
 }

fs/bcachefs/btree_io.c

@@ -741,16 +741,22 @@ static int validate_bset(struct bch_fs *c, struct bch_dev *ca,
 		     BCH_VERSION_MAJOR(version),
 		     BCH_VERSION_MINOR(version));
 
-	if (btree_err_on(version < c->sb.version_min,
+	if (c->recovery.curr_pass != BCH_RECOVERY_PASS_scan_for_btree_nodes &&
+	    btree_err_on(version < c->sb.version_min,
 			 -BCH_ERR_btree_node_read_err_fixable,
 			 c, NULL, b, i, NULL,
 			 btree_node_bset_older_than_sb_min,
 			 "bset version %u older than superblock version_min %u",
 			 version, c->sb.version_min)) {
-		mutex_lock(&c->sb_lock);
-		c->disk_sb.sb->version_min = cpu_to_le16(version);
-		bch2_write_super(c);
-		mutex_unlock(&c->sb_lock);
+		if (bch2_version_compatible(version)) {
+			mutex_lock(&c->sb_lock);
+			c->disk_sb.sb->version_min = cpu_to_le16(version);
+			bch2_write_super(c);
+			mutex_unlock(&c->sb_lock);
+		} else {
+			/* We have no idea what's going on: */
+			i->version = cpu_to_le16(c->sb.version);
+		}
 	}
 
 	if (btree_err_on(BCH_VERSION_MAJOR(version) >
@@ -1045,6 +1051,7 @@ static int validate_bset_keys(struct bch_fs *c, struct btree *b,
 		le16_add_cpu(&i->u64s, -next_good_key);
 		memmove_u64s_down(k, (u64 *) k + next_good_key, (u64 *) vstruct_end(i) - (u64 *) k);
 		set_btree_node_need_rewrite(b);
+		set_btree_node_need_rewrite_error(b);
 	}
 fsck_err:
 	printbuf_exit(&buf);
@@ -1305,6 +1312,7 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca,
 					  (u64 *) vstruct_end(i) - (u64 *) k);
 			set_btree_bset_end(b, b->set);
 			set_btree_node_need_rewrite(b);
+			set_btree_node_need_rewrite_error(b);
 			continue;
 		}
 		if (ret)
@@ -1329,12 +1337,16 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca,
 		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)
+			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)
+	if (!ptr_written) {
 		set_btree_node_need_rewrite(b);
+		set_btree_node_need_rewrite_ptr_written_zero(b);
+	}
 fsck_err:
 	mempool_free(iter, &c->fill_iter);
 	printbuf_exit(&buf);

fs/bcachefs/btree_locking.c

@@ -213,7 +213,7 @@ static noinline __noreturn void break_cycle_fail(struct lock_graph *g)
 		prt_newline(&buf);
 	}
 
-	bch2_print_str_nonblocking(g->g->trans->c, KERN_ERR, buf.buf);
+	bch2_print_str(g->g->trans->c, KERN_ERR, buf.buf);
 	printbuf_exit(&buf);
 	BUG();
 }

fs/bcachefs/btree_locking.h

@@ -417,8 +417,10 @@ static inline void btree_path_set_should_be_locked(struct btree_trans *trans, st
 	EBUG_ON(!btree_node_locked(path, path->level));
 	EBUG_ON(path->uptodate);
 
-	path->should_be_locked = true;
-	trace_btree_path_should_be_locked(trans, path);
+	if (!path->should_be_locked) {
+		path->should_be_locked = true;
+		trace_btree_path_should_be_locked(trans, path);
+	}
 }
 
 static inline void __btree_path_set_level_up(struct btree_trans *trans,

fs/bcachefs/btree_types.h

@@ -617,6 +617,9 @@ enum btree_write_type {
 	x(dying)							\
 	x(fake)								\
 	x(need_rewrite)							\
+	x(need_rewrite_error)						\
+	x(need_rewrite_degraded)					\
+	x(need_rewrite_ptr_written_zero)				\
 	x(never_write)							\
 	x(pinned)
 
@@ -641,6 +644,32 @@ static inline void clear_btree_node_ ## flag(struct btree *b)		\
 BTREE_FLAGS()
 #undef x
 
+#define BTREE_NODE_REWRITE_REASON()					\
+	x(none)								\
+	x(unknown)							\
+	x(error)							\
+	x(degraded)							\
+	x(ptr_written_zero)
+
+enum btree_node_rewrite_reason {
+#define x(n)	BTREE_NODE_REWRITE_##n,
+	BTREE_NODE_REWRITE_REASON()
+#undef x
+};
+
+static inline enum btree_node_rewrite_reason btree_node_rewrite_reason(struct btree *b)
+{
+	if (btree_node_need_rewrite_ptr_written_zero(b))
+		return BTREE_NODE_REWRITE_ptr_written_zero;
+	if (btree_node_need_rewrite_degraded(b))
+		return BTREE_NODE_REWRITE_degraded;
+	if (btree_node_need_rewrite_error(b))
+		return BTREE_NODE_REWRITE_error;
+	if (btree_node_need_rewrite(b))
+		return BTREE_NODE_REWRITE_unknown;
+	return BTREE_NODE_REWRITE_none;
+}
+
 static inline struct btree_write *btree_current_write(struct btree *b)
 {
 	return b->writes + btree_node_write_idx(b);

fs/bcachefs/btree_update_interior.c

@@ -1138,6 +1138,13 @@ static void bch2_btree_update_done(struct btree_update *as, struct btree_trans *
 			       start_time);
 }
 
+static const char * const btree_node_reawrite_reason_strs[] = {
+#define x(n)	#n,
+	BTREE_NODE_REWRITE_REASON()
+#undef x
+	NULL,
+};
+
 static struct btree_update *
 bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
 			unsigned level_start, bool split,
@@ -1232,6 +1239,15 @@ bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
 	list_add_tail(&as->list, &c->btree_interior_update_list);
 	mutex_unlock(&c->btree_interior_update_lock);
 
+	struct btree *b = btree_path_node(path, path->level);
+	as->node_start	= b->data->min_key;
+	as->node_end	= b->data->max_key;
+	as->node_needed_rewrite = btree_node_rewrite_reason(b);
+	as->node_written = b->written;
+	as->node_sectors = btree_buf_bytes(b) >> 9;
+	as->node_remaining = __bch2_btree_u64s_remaining(b,
+				btree_bkey_last(b, bset_tree_last(b)));
+
 	/*
 	 * We don't want to allocate if we're in an error state, that can cause
 	 * deadlock on emergency shutdown due to open buckets getting stuck in
@@ -2108,6 +2124,9 @@ int __bch2_foreground_maybe_merge(struct btree_trans *trans,
 	if (ret)
 		goto err;
 
+	as->node_start	= prev->data->min_key;
+	as->node_end	= next->data->max_key;
+
 	trace_and_count(c, btree_node_merge, trans, b);
 
 	n = bch2_btree_node_alloc(as, trans, b->c.level);
@@ -2681,9 +2700,19 @@ static void bch2_btree_update_to_text(struct printbuf *out, struct btree_update
 
 	prt_str(out, " ");
 	bch2_btree_id_to_text(out, as->btree_id);
-	prt_printf(out, " l=%u-%u mode=%s nodes_written=%u cl.remaining=%u journal_seq=%llu\n",
+	prt_printf(out, " l=%u-%u ",
 		   as->update_level_start,
-		   as->update_level_end,
+		   as->update_level_end);
+	bch2_bpos_to_text(out, as->node_start);
+	prt_char(out, ' ');
+	bch2_bpos_to_text(out, as->node_end);
+	prt_printf(out, "\nwritten %u/%u u64s_remaining %u need_rewrite %s",
+		   as->node_written,
+		   as->node_sectors,
+		   as->node_remaining,
+		   btree_node_reawrite_reason_strs[as->node_needed_rewrite]);
+
+	prt_printf(out, "\nmode=%s nodes_written=%u cl.remaining=%u journal_seq=%llu\n",
 		   bch2_btree_update_modes[as->mode],
 		   as->nodes_written,
 		   closure_nr_remaining(&as->cl),

fs/bcachefs/btree_update_interior.h

@@ -57,6 +57,13 @@ struct btree_update {
 	unsigned			took_gc_lock:1;
 
 	enum btree_id			btree_id;
+	struct bpos			node_start;
+	struct bpos			node_end;
+	enum btree_node_rewrite_reason	node_needed_rewrite;
+	u16				node_written;
+	u16				node_sectors;
+	u16				node_remaining;
+
 	unsigned			update_level_start;
 	unsigned			update_level_end;
 

fs/bcachefs/chardev.c

@@ -399,7 +399,7 @@ static long bch2_ioctl_data(struct bch_fs *c,
 	return ret;
 }
 
-static long bch2_ioctl_fs_usage(struct bch_fs *c,
+static noinline_for_stack long bch2_ioctl_fs_usage(struct bch_fs *c,
 				struct bch_ioctl_fs_usage __user *user_arg)
 {
 	struct bch_ioctl_fs_usage arg = {};
@@ -469,7 +469,7 @@ static long bch2_ioctl_query_accounting(struct bch_fs *c,
 }
 
 /* obsolete, didn't allow for new data types: */
-static long bch2_ioctl_dev_usage(struct bch_fs *c,
+static noinline_for_stack long bch2_ioctl_dev_usage(struct bch_fs *c,
 				 struct bch_ioctl_dev_usage __user *user_arg)
 {
 	struct bch_ioctl_dev_usage arg;

fs/bcachefs/disk_accounting.c

@@ -618,7 +618,9 @@ int bch2_gc_accounting_done(struct bch_fs *c)
 			for (unsigned j = 0; j < nr; j++)
 				src_v[j] -= dst_v[j];
 
-			if (fsck_err(trans, accounting_mismatch, "%s", buf.buf)) {
+			bch2_trans_unlock_long(trans);
+
+			if (fsck_err(c, accounting_mismatch, "%s", buf.buf)) {
 				percpu_up_write(&c->mark_lock);
 				ret = commit_do(trans, NULL, NULL, 0,
 						bch2_disk_accounting_mod(trans, &acc_k, src_v, nr, false));