Merge tag 'bcachefs-2024-04-22' of https://evilpiepirate.org/git/bcachefs

Pull bcachefs fixes from Kent Overstreet:
 "Nothing too crazy in this one, and it looks like (fingers crossed) the
  recovery and repair issues are settling down - although there's going
  to be a long tail there, as we've still yet to really ramp up on error
  injection or syzbot.

   - fix a few more deadlocks in recovery

   - fix u32/u64 issues in mi_btree_bitmap

   - btree key cache shrinker now actually frees, with more
     instrumentation coming so we can verify that it's working
     correctly more easily in the future"

* tag 'bcachefs-2024-04-22' of https://evilpiepirate.org/git/bcachefs:
  bcachefs: If we run merges at a lower watermark, they must be nonblocking
  bcachefs: Fix inode early destruction path
  bcachefs: Fix deadlock in journal write path
  bcachefs: Tweak btree key cache shrinker so it actually frees
  bcachefs: bkey_cached.btree_trans_barrier_seq needs to be a ulong
  bcachefs: Fix missing call to bch2_fs_allocator_background_exit()
  bcachefs: Check for journal entries overruning end of sb clean section
  bcachefs: Fix bio alloc in check_extent_checksum()
  bcachefs: fix leak in bch2_gc_write_reflink_key
  bcachefs: KEY_TYPE_error is allowed for reflink
  bcachefs: Fix bch2_dev_btree_bitmap_marked_sectors() shift
  bcachefs: make sure to release last journal pin in replay
  bcachefs: node scan: ignore multiple nodes with same seq if interior
  bcachefs: Fix format specifier in validate_bset_keys()
  bcachefs: Fix null ptr deref in twf from BCH_IOCTL_FSCK_OFFLINE

fs/bcachefs/backpointers.c

@@ -470,7 +470,7 @@ static int check_extent_checksum(struct btree_trans *trans,
 		goto err;
 	}
 
-	bio = bio_alloc(ca->disk_sb.bdev, 1, REQ_OP_READ, GFP_KERNEL);
+	bio = bio_alloc(ca->disk_sb.bdev, buf_pages(data_buf, bytes), 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);

fs/bcachefs/bcachefs_format.h

@@ -1504,7 +1504,8 @@ enum btree_id_flags {
 	  BIT_ULL(KEY_TYPE_stripe))						\
 	x(reflink,		7,	BTREE_ID_EXTENTS|BTREE_ID_DATA,		\
 	  BIT_ULL(KEY_TYPE_reflink_v)|						\
-	  BIT_ULL(KEY_TYPE_indirect_inline_data))				\
+	  BIT_ULL(KEY_TYPE_indirect_inline_data)|				\
+	  BIT_ULL(KEY_TYPE_error))						\
 	x(subvolumes,		8,	0,					\
 	  BIT_ULL(KEY_TYPE_subvolume))						\
 	x(snapshots,		9,	0,					\

fs/bcachefs/btree_gc.c

@@ -1587,7 +1587,7 @@ static int bch2_gc_write_reflink_key(struct btree_trans *trans,
 		struct bkey_i *new = bch2_bkey_make_mut_noupdate(trans, k);
 		ret = PTR_ERR_OR_ZERO(new);
 		if (ret)
-			return ret;
+			goto out;
 
 		if (!r->refcount)
 			new->k.type = KEY_TYPE_deleted;
@@ -1595,6 +1595,7 @@ static int bch2_gc_write_reflink_key(struct btree_trans *trans,
 			*bkey_refcount(bkey_i_to_s(new)) = cpu_to_le64(r->refcount);
 		ret = bch2_trans_update(trans, iter, new, 0);
 	}
+out:
 fsck_err:
 	printbuf_exit(&buf);
 	return ret;

fs/bcachefs/btree_io.c

@@ -888,7 +888,7 @@ static int validate_bset_keys(struct bch_fs *c, struct btree *b,
 				 -BCH_ERR_btree_node_read_err_fixable,
 				 c, NULL, b, i,
 				 btree_node_bkey_bad_u64s,
-				 "bad k->u64s %u (min %u max %lu)", k->u64s,
+				 "bad k->u64s %u (min %u max %zu)", k->u64s,
 				 bkeyp_key_u64s(&b->format, k),
 				 U8_MAX - BKEY_U64s + bkeyp_key_u64s(&b->format, k)))
 			goto drop_this_key;

fs/bcachefs/btree_key_cache.c

@@ -842,8 +842,6 @@ static unsigned long bch2_btree_key_cache_scan(struct shrinker *shrink,
 	 * Newest freed entries are at the end of the list - once we hit one
 	 * that's too new to be freed, we can bail out:
 	 */
-	scanned += bc->nr_freed_nonpcpu;
-
 	list_for_each_entry_safe(ck, t, &bc->freed_nonpcpu, list) {
 		if (!poll_state_synchronize_srcu(&c->btree_trans_barrier,
 						 ck->btree_trans_barrier_seq))
@@ -857,11 +855,6 @@ static unsigned long bch2_btree_key_cache_scan(struct shrinker *shrink,
 		bc->nr_freed_nonpcpu--;
 	}
 
-	if (scanned >= nr)
-		goto out;
-
-	scanned += bc->nr_freed_pcpu;
-
 	list_for_each_entry_safe(ck, t, &bc->freed_pcpu, list) {
 		if (!poll_state_synchronize_srcu(&c->btree_trans_barrier,
 						 ck->btree_trans_barrier_seq))
@@ -875,9 +868,6 @@ static unsigned long bch2_btree_key_cache_scan(struct shrinker *shrink,
 		bc->nr_freed_pcpu--;
 	}
 
-	if (scanned >= nr)
-		goto out;
-
 	rcu_read_lock();
 	tbl = rht_dereference_rcu(bc->table.tbl, &bc->table);
 	if (bc->shrink_iter >= tbl->size)
@@ -893,12 +883,12 @@ static unsigned long bch2_btree_key_cache_scan(struct shrinker *shrink,
 			next = rht_dereference_bucket_rcu(pos->next, tbl, bc->shrink_iter);
 			ck = container_of(pos, struct bkey_cached, hash);
 
-			if (test_bit(BKEY_CACHED_DIRTY, &ck->flags))
+			if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
 				goto next;
-
-			if (test_bit(BKEY_CACHED_ACCESSED, &ck->flags))
+			} else if (test_bit(BKEY_CACHED_ACCESSED, &ck->flags)) {
 				clear_bit(BKEY_CACHED_ACCESSED, &ck->flags);
-			else if (bkey_cached_lock_for_evict(ck)) {
+				goto next;
+			} else if (bkey_cached_lock_for_evict(ck)) {
 				bkey_cached_evict(bc, ck);
 				bkey_cached_free(bc, ck);
 			}
@@ -916,7 +906,6 @@ static unsigned long bch2_btree_key_cache_scan(struct shrinker *shrink,
 	} while (scanned < nr && bc->shrink_iter != start);
 
 	rcu_read_unlock();
-out:
 	memalloc_nofs_restore(flags);
 	srcu_read_unlock(&c->btree_trans_barrier, srcu_idx);
 	mutex_unlock(&bc->lock);

fs/bcachefs/btree_node_scan.c

@@ -302,6 +302,8 @@ static int handle_overwrites(struct bch_fs *c,
 
 			start->max_key = bpos_predecessor(n->min_key);
 			start->range_updated = true;
+		} else if (n->level) {
+			n->overwritten = true;
 		} else {
 			struct printbuf buf = PRINTBUF;
 

fs/bcachefs/btree_types.h

@@ -321,9 +321,9 @@ struct bkey_cached {
 	struct btree_bkey_cached_common c;
 
 	unsigned long		flags;
+	unsigned long		btree_trans_barrier_seq;
 	u16			u64s;
 	bool			valid;
-	u32			btree_trans_barrier_seq;
 	struct bkey_cached_key	key;
 
 	struct rhash_head	hash;

fs/bcachefs/btree_update_interior.c

@@ -1960,7 +1960,11 @@ int __bch2_foreground_maybe_merge(struct btree_trans *trans,
 	if ((flags & BCH_WATERMARK_MASK) == BCH_WATERMARK_interior_updates)
 		return 0;
 
+	if ((flags & BCH_WATERMARK_MASK) <= BCH_WATERMARK_reclaim) {
 		flags &= ~BCH_WATERMARK_MASK;
+		flags |= BCH_WATERMARK_btree;
+		flags |= BCH_TRANS_COMMIT_journal_reclaim;
+	}
 
 	b = trans->paths[path].l[level].b;
 

fs/bcachefs/chardev.c

@@ -232,13 +232,15 @@ static long bch2_ioctl_fsck_offline(struct bch_ioctl_fsck_offline __user *user_a
 	/* We need request_key() to be called before we punt to kthread: */
 	opt_set(thr->opts, nostart, true);
 
+	bch2_thread_with_stdio_init(&thr->thr, &bch2_offline_fsck_ops);
+
 	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);
+	ret = __bch2_run_thread_with_stdio(&thr->thr);
 out:
 	darray_for_each(devs, i)
 		kfree(*i);

fs/bcachefs/fs.c

@@ -188,7 +188,8 @@ static struct bch_inode_info *bch2_inode_insert(struct bch_fs *c, struct bch_ino
 	BUG_ON(!old);
 
 	if (unlikely(old != inode)) {
-		discard_new_inode(&inode->v);
+		__destroy_inode(&inode->v);
+		kmem_cache_free(bch2_inode_cache, inode);
 		inode = old;
 	} else {
 		mutex_lock(&c->vfs_inodes_lock);
@@ -225,8 +226,10 @@ static struct bch_inode_info *bch2_new_inode(struct btree_trans *trans)
 
 	if (unlikely(!inode)) {
 		int ret = drop_locks_do(trans, (inode = to_bch_ei(new_inode(c->vfs_sb))) ? 0 : -ENOMEM);
-		if (ret && inode)
-			discard_new_inode(&inode->v);
+		if (ret && inode) {
+			__destroy_inode(&inode->v);
+			kmem_cache_free(bch2_inode_cache, inode);
+		}
 		if (ret)
 			return ERR_PTR(ret);
 	}

fs/bcachefs/journal_io.c

@@ -1723,7 +1723,7 @@ static void journal_write_endio(struct bio *bio)
 	percpu_ref_put(&ca->io_ref);
 }
 
-static CLOSURE_CALLBACK(do_journal_write)
+static CLOSURE_CALLBACK(journal_write_submit)
 {
 	closure_type(w, struct journal_buf, io);
 	struct journal *j = container_of(w, struct journal, buf[w->idx]);
@@ -1768,6 +1768,44 @@ static CLOSURE_CALLBACK(do_journal_write)
 	continue_at(cl, journal_write_done, j->wq);
 }
 
+static CLOSURE_CALLBACK(journal_write_preflush)
+{
+	closure_type(w, struct journal_buf, io);
+	struct journal *j = container_of(w, struct journal, buf[w->idx]);
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
+
+	if (j->seq_ondisk + 1 != le64_to_cpu(w->data->seq)) {
+		spin_lock(&j->lock);
+		closure_wait(&j->async_wait, cl);
+		spin_unlock(&j->lock);
+
+		continue_at(cl, journal_write_preflush, j->wq);
+		return;
+	}
+
+	if (w->separate_flush) {
+		for_each_rw_member(c, ca) {
+			percpu_ref_get(&ca->io_ref);
+
+			struct journal_device *ja = &ca->journal;
+			struct bio *bio = &ja->bio[w->idx]->bio;
+			bio_reset(bio, ca->disk_sb.bdev,
+				  REQ_OP_WRITE|REQ_SYNC|REQ_META|REQ_PREFLUSH);
+			bio->bi_end_io		= journal_write_endio;
+			bio->bi_private		= ca;
+			closure_bio_submit(bio, cl);
+		}
+
+		continue_at(cl, journal_write_submit, j->wq);
+	} else {
+		/*
+		 * no need to punt to another work item if we're not waiting on
+		 * preflushes
+		 */
+		journal_write_submit(&cl->work);
+	}
+}
+
 static int bch2_journal_write_prep(struct journal *j, struct journal_buf *w)
 {
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
@@ -2033,23 +2071,9 @@ CLOSURE_CALLBACK(bch2_journal_write)
 		goto err;
 
 	if (!JSET_NO_FLUSH(w->data))
-		closure_wait_event(&j->async_wait, j->seq_ondisk + 1 == le64_to_cpu(w->data->seq));
-
-	if (!JSET_NO_FLUSH(w->data) && w->separate_flush) {
-		for_each_rw_member(c, ca) {
-			percpu_ref_get(&ca->io_ref);
-
-			struct journal_device *ja = &ca->journal;
-			struct bio *bio = &ja->bio[w->idx]->bio;
-			bio_reset(bio, ca->disk_sb.bdev,
-				  REQ_OP_WRITE|REQ_SYNC|REQ_META|REQ_PREFLUSH);
-			bio->bi_end_io		= journal_write_endio;
-			bio->bi_private		= ca;
-			closure_bio_submit(bio, cl);
-		}
-	}
-
-	continue_at(cl, do_journal_write, j->wq);
+		continue_at(cl, journal_write_preflush, j->wq);
+	else
+		continue_at(cl, journal_write_submit, j->wq);
 	return;
 no_io:
 	continue_at(cl, journal_write_done, j->wq);

fs/bcachefs/recovery.c

@@ -249,7 +249,10 @@ int bch2_journal_replay(struct bch_fs *c)
 
 		struct journal_key *k = *kp;
 
+		if (k->journal_seq)
 			replay_now_at(j, k->journal_seq);
+		else
+			replay_now_at(j, j->replay_journal_seq_end);
 
 		ret = commit_do(trans, NULL, NULL,
 				BCH_TRANS_COMMIT_no_enospc|

fs/bcachefs/sb-clean.c

@@ -29,6 +29,14 @@ int bch2_sb_clean_validate_late(struct bch_fs *c, struct bch_sb_field_clean *cle
 	for (entry = clean->start;
 	     entry < (struct jset_entry *) vstruct_end(&clean->field);
 	     entry = vstruct_next(entry)) {
+		if (vstruct_end(entry) > vstruct_end(&clean->field)) {
+			bch_err(c, "journal entry (u64s %u) overran end of superblock clean section (u64s %u) by %zu",
+				le16_to_cpu(entry->u64s), le32_to_cpu(clean->field.u64s),
+				(u64 *) vstruct_end(entry) - (u64 *) vstruct_end(&clean->field));
+			bch2_sb_error_count(c, BCH_FSCK_ERR_sb_clean_entry_overrun);
+			return -BCH_ERR_fsck_repair_unimplemented;
+		}
+
 		ret = bch2_journal_entry_validate(c, NULL, entry,
 						  le16_to_cpu(c->disk_sb.sb->version),
 						  BCH_SB_BIG_ENDIAN(c->disk_sb.sb),

fs/bcachefs/sb-errors_types.h

@@ -271,7 +271,8 @@
 	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)
+	x(btree_bitmap_not_marked,				266)	\
+	x(sb_clean_entry_overrun,				267)
 
 enum bch_sb_error_id {
 #define x(t, n) BCH_FSCK_ERR_##t = n,

fs/bcachefs/sb-members.c

@@ -463,8 +463,8 @@ static void __bch2_dev_btree_bitmap_mark(struct bch_sb_field_members_v2 *mi, uns
 		m->btree_bitmap_shift += resize;
 	}
 
-	for (unsigned bit = sectors >> m->btree_bitmap_shift;
-	     bit << m->btree_bitmap_shift < end;
+	for (unsigned bit = start >> m->btree_bitmap_shift;
+	     (u64) bit << m->btree_bitmap_shift < end;
 	     bit++)
 		bitmap |= BIT_ULL(bit);
 

fs/bcachefs/sb-members.h

@@ -235,11 +235,11 @@ static inline bool bch2_dev_btree_bitmap_marked_sectors(struct bch_dev *ca, u64
 {
 	u64 end = start + sectors;
 
-	if (end > 64 << ca->mi.btree_bitmap_shift)
+	if (end > 64ULL << ca->mi.btree_bitmap_shift)
 		return false;
 
-	for (unsigned bit = sectors >> ca->mi.btree_bitmap_shift;
-	     bit << ca->mi.btree_bitmap_shift < end;
+	for (unsigned bit = start >> ca->mi.btree_bitmap_shift;
+	     (u64) bit << ca->mi.btree_bitmap_shift < end;
 	     bit++)
 		if (!(ca->mi.btree_allocated_bitmap & BIT_ULL(bit)))
 			return false;

fs/bcachefs/super.c

@@ -544,6 +544,7 @@ static void __bch2_fs_free(struct bch_fs *c)
 
 	bch2_find_btree_nodes_exit(&c->found_btree_nodes);
 	bch2_free_pending_node_rewrites(c);
+	bch2_fs_allocator_background_exit(c);
 	bch2_fs_sb_errors_exit(c);
 	bch2_fs_counters_exit(c);
 	bch2_fs_snapshots_exit(c);

fs/bcachefs/thread_with_file.c

@@ -294,16 +294,27 @@ static int thread_with_stdio_fn(void *arg)
 	return 0;
 }
 
-int bch2_run_thread_with_stdio(struct thread_with_stdio *thr,
+void bch2_thread_with_stdio_init(struct thread_with_stdio *thr,
 				 const struct thread_with_stdio_ops *ops)
 {
 	stdio_buf_init(&thr->stdio.input);
 	stdio_buf_init(&thr->stdio.output);
 	thr->ops = ops;
+}
 
+int __bch2_run_thread_with_stdio(struct thread_with_stdio *thr)
+{
 	return bch2_run_thread_with_file(&thr->thr, &thread_with_stdio_fops, thread_with_stdio_fn);
 }
 
+int bch2_run_thread_with_stdio(struct thread_with_stdio *thr,
+			       const struct thread_with_stdio_ops *ops)
+{
+	bch2_thread_with_stdio_init(thr, ops);
+
+	return __bch2_run_thread_with_stdio(thr);
+}
+
 int bch2_run_thread_with_stdout(struct thread_with_stdio *thr,
 				const struct thread_with_stdio_ops *ops)
 {

fs/bcachefs/thread_with_file.h

@@ -63,6 +63,9 @@ struct thread_with_stdio {
 	const struct thread_with_stdio_ops	*ops;
 };
 
+void bch2_thread_with_stdio_init(struct thread_with_stdio *,
+				 const struct thread_with_stdio_ops *);
+int __bch2_run_thread_with_stdio(struct thread_with_stdio *);
 int bch2_run_thread_with_stdio(struct thread_with_stdio *,
 			       const struct thread_with_stdio_ops *);
 int bch2_run_thread_with_stdout(struct thread_with_stdio *,