bcachefs: Don't issue btree writes that weren't journalled

If we have an error in the btree interior update path that prevents us
from journalling the update, we can't issue the corresponding btree node
write - we didn't get a journal sequence number that would cause it to
be ignored in recovery.

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

diff --git a/fs/bcachefs/btree_io.c b/fs/bcachefs/btree_io.c
index 520eef531d397ac454cda72ea7d57896689f503a..b2ffdff48637ccaf7562c7ff6f7080e5ddfb3934 100644 (file)
--- a/fs/bcachefs/btree_io.c
+++ b/fs/bcachefs/btree_io.c
@@ -1499,6 +1499,9 @@ void __bch2_btree_node_write(struct bch_fs *c, struct btree *b,
                if (!btree_node_may_write(b))
                        return;
 
+               if (old & (1 << BTREE_NODE_never_write))
+                       return;
+
                if (old & (1 << BTREE_NODE_write_in_flight)) {
                        btree_node_wait_on_io(b);
                        continue;
@@ -1545,6 +1548,8 @@ void __bch2_btree_node_write(struct bch_fs *c, struct btree *b,
                seq = max(seq, le64_to_cpu(i->journal_seq));
        }
 
+       BUG_ON(b->written && !seq);
+
        /* bch2_varint_decode may read up to 7 bytes past the end of the buffer: */
        bytes += 8;
 

diff --git a/fs/bcachefs/btree_types.h b/fs/bcachefs/btree_types.h
index d30c31f0f11fdae4c6c06ca010118fb036bd4ad8..51ad87abc763852efeeec6d7cb1285341f3dad5b 100644 (file)
--- a/fs/bcachefs/btree_types.h
+++ b/fs/bcachefs/btree_types.h
@@ -415,6 +415,7 @@ enum btree_flags {
        BTREE_NODE_fake,
        BTREE_NODE_old_extent_overwrite,
        BTREE_NODE_need_rewrite,
+       BTREE_NODE_never_write,
 };
 
 BTREE_FLAG(read_in_flight);
@@ -429,6 +430,7 @@ BTREE_FLAG(dying);
 BTREE_FLAG(fake);
 BTREE_FLAG(old_extent_overwrite);
 BTREE_FLAG(need_rewrite);
+BTREE_FLAG(never_write);
 
 static inline struct btree_write *btree_current_write(struct btree *b)
 {

diff --git a/fs/bcachefs/btree_update_interior.c b/fs/bcachefs/btree_update_interior.c
index aed54d5b5251e6c2a7b9467ec11d83fbc5e5d870..594bcd7975169f581043d4375f55af92a10af6f9 100644 (file)
--- a/fs/bcachefs/btree_update_interior.c
+++ b/fs/bcachefs/btree_update_interior.c
@@ -603,17 +603,30 @@ err:
 
                list_del(&as->write_blocked_list);
 
-               if (!ret && as->b == b) {
+               /*
+                * Node might have been freed, recheck under
+                * btree_interior_update_lock:
+                */
+               if (as->b == b) {
                        struct bset *i = btree_bset_last(b);
 
                        BUG_ON(!b->c.level);
                        BUG_ON(!btree_node_dirty(b));
 
-                       i->journal_seq = cpu_to_le64(
-                               max(journal_seq,
-                                   le64_to_cpu(i->journal_seq)));
-
-                       bch2_btree_add_journal_pin(c, b, journal_seq);
+                       if (!ret) {
+                               i->journal_seq = cpu_to_le64(
+                                       max(journal_seq,
+                                           le64_to_cpu(i->journal_seq)));
+
+                               bch2_btree_add_journal_pin(c, b, journal_seq);
+                       } else {
+                               /*
+                                * If we didn't get a journal sequence number we
+                                * can't write this btree node, because recovery
+                                * won't know to ignore this write:
+                                */
+                               set_btree_node_never_write(b);
+                       }
                }
 
                mutex_unlock(&c->btree_interior_update_lock);