Merge tag 'bcachefs-2025-07-11' of git://evilpiepirate.org/bcachefs

Pull bcachefs fixes from Kent Overstreet.

* tag 'bcachefs-2025-07-11' of git://evilpiepirate.org/bcachefs:
  bcachefs: Don't set BCH_FS_error on transaction restart
  bcachefs: Fix additional misalignment in journal space calculations
  bcachefs: Don't schedule non persistent passes persistently
  bcachefs: Fix bch2_btree_transactions_read() synchronization
  bcachefs: btree read retry fixes
  bcachefs: btree node scan no longer uses btree cache
  bcachefs: Tweak btree cache helpers for use by btree node scan
  bcachefs: Fix btree for nonexistent tree depth
  bcachefs: Fix bch2_io_failures_to_text()
  bcachefs: bch2_fpunch_snapshot()

Author: torvalds

fs/bcachefs/btree_cache.c

@@ -85,7 +85,7 @@ void bch2_btree_node_to_freelist(struct bch_fs *c, struct btree *b)
 	six_unlock_intent(&b->c.lock);
 }
 
-static void __btree_node_data_free(struct btree_cache *bc, struct btree *b)
+void __btree_node_data_free(struct btree *b)
 {
 	BUG_ON(!list_empty(&b->list));
 	BUG_ON(btree_node_hashed(b));
@@ -112,16 +112,17 @@ static void __btree_node_data_free(struct btree_cache *bc, struct btree *b)
 	munmap(b->aux_data, btree_aux_data_bytes(b));
 #endif
 	b->aux_data = NULL;
-
-	btree_node_to_freedlist(bc, b);
 }
 
 static void btree_node_data_free(struct btree_cache *bc, struct btree *b)
 {
 	BUG_ON(list_empty(&b->list));
 	list_del_init(&b->list);
+
+	__btree_node_data_free(b);
+
 	--bc->nr_freeable;
-	__btree_node_data_free(bc, b);
+	btree_node_to_freedlist(bc, b);
 }
 
 static int bch2_btree_cache_cmp_fn(struct rhashtable_compare_arg *arg,
@@ -185,10 +186,7 @@ static struct btree *__btree_node_mem_alloc(struct bch_fs *c, gfp_t gfp)
 
 struct btree *__bch2_btree_node_mem_alloc(struct bch_fs *c)
 {
-	struct btree_cache *bc = &c->btree_cache;
-	struct btree *b;
-
-	b = __btree_node_mem_alloc(c, GFP_KERNEL);
+	struct btree *b = __btree_node_mem_alloc(c, GFP_KERNEL);
 	if (!b)
 		return NULL;
 
@@ -198,8 +196,6 @@ struct btree *__bch2_btree_node_mem_alloc(struct bch_fs *c)
 	}
 
 	bch2_btree_lock_init(&b->c, 0, GFP_KERNEL);
-
-	__bch2_btree_node_to_freelist(bc, b);
 	return b;
 }
 
@@ -524,7 +520,8 @@ static unsigned long bch2_btree_cache_scan(struct shrinker *shrink,
 			--touched;;
 		} else if (!btree_node_reclaim(c, b)) {
 			__bch2_btree_node_hash_remove(bc, b);
-			__btree_node_data_free(bc, b);
+			__btree_node_data_free(b);
+			btree_node_to_freedlist(bc, b);
 
 			freed++;
 			bc->nr_freed++;
@@ -652,9 +649,12 @@ int bch2_fs_btree_cache_init(struct bch_fs *c)
 
 	bch2_recalc_btree_reserve(c);
 
-	for (i = 0; i < bc->nr_reserve; i++)
-		if (!__bch2_btree_node_mem_alloc(c))
+	for (i = 0; i < bc->nr_reserve; i++) {
+		struct btree *b = __bch2_btree_node_mem_alloc(c);
+		if (!b)
 			goto err;
+		__bch2_btree_node_to_freelist(bc, b);
+	}
 
 	list_splice_init(&bc->live[0].list, &bc->freeable);
 

fs/bcachefs/btree_cache.h

@@ -30,6 +30,7 @@ void bch2_btree_node_update_key_early(struct btree_trans *, enum btree_id, unsig
 void bch2_btree_cache_cannibalize_unlock(struct btree_trans *);
 int bch2_btree_cache_cannibalize_lock(struct btree_trans *, struct closure *);
 
+void __btree_node_data_free(struct btree *);
 struct btree *__bch2_btree_node_mem_alloc(struct bch_fs *);
 struct btree *bch2_btree_node_mem_alloc(struct btree_trans *, bool);
 

fs/bcachefs/btree_io.c

@@ -568,9 +568,9 @@ static int __btree_err(int ret,
 		bch2_mark_btree_validate_failure(failed, ca->dev_idx);
 
 		struct extent_ptr_decoded pick;
-		have_retry = !bch2_bkey_pick_read_device(c,
+		have_retry = bch2_bkey_pick_read_device(c,
 					bkey_i_to_s_c(&b->key),
-					failed, &pick, -1);
+					failed, &pick, -1) == 1;
 	}
 
 	if (!have_retry && ret == -BCH_ERR_btree_node_read_err_want_retry)
@@ -615,7 +615,6 @@ static int __btree_err(int ret,
 			goto out;
 		case -BCH_ERR_btree_node_read_err_bad_node:
 			prt_str(&out, ", ");
-			ret = __bch2_topology_error(c, &out);
 			break;
 		}
 
@@ -644,7 +643,6 @@ static int __btree_err(int ret,
 		goto out;
 	case -BCH_ERR_btree_node_read_err_bad_node:
 		prt_str(&out, ", ");
-		ret = __bch2_topology_error(c, &out);
 		break;
 	}
 print:
@@ -1408,7 +1406,7 @@ static void btree_node_read_work(struct work_struct *work)
 		ret = bch2_bkey_pick_read_device(c,
 					bkey_i_to_s_c(&b->key),
 					&failed, &rb->pick, -1);
-		if (ret) {
+		if (ret <= 0) {
 			set_btree_node_read_error(b);
 			break;
 		}

fs/bcachefs/btree_node_scan.c

@@ -75,39 +75,6 @@ static inline u64 bkey_journal_seq(struct bkey_s_c k)
 	}
 }
 
-static bool found_btree_node_is_readable(struct btree_trans *trans,
-					 struct found_btree_node *f)
-{
-	struct { __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX); } tmp;
-
-	found_btree_node_to_key(&tmp.k, f);
-
-	struct btree *b = bch2_btree_node_get_noiter(trans, &tmp.k, f->btree_id, f->level, false);
-	bool ret = !IS_ERR_OR_NULL(b);
-	if (!ret)
-		return ret;
-
-	f->sectors_written = b->written;
-	f->journal_seq = le64_to_cpu(b->data->keys.journal_seq);
-
-	struct bkey_s_c k;
-	struct bkey unpacked;
-	struct btree_node_iter iter;
-	for_each_btree_node_key_unpack(b, k, &iter, &unpacked)
-		f->journal_seq = max(f->journal_seq, bkey_journal_seq(k));
-
-	six_unlock_read(&b->c.lock);
-
-	/*
-	 * We might update this node's range; if that happens, we need the node
-	 * to be re-read so the read path can trim keys that are no longer in
-	 * this node
-	 */
-	if (b != btree_node_root(trans->c, b))
-		bch2_btree_node_evict(trans, &tmp.k);
-	return ret;
-}
-
 static int found_btree_node_cmp_cookie(const void *_l, const void *_r)
 {
 	const struct found_btree_node *l = _l;
@@ -159,17 +126,17 @@ static const struct min_heap_callbacks found_btree_node_heap_cbs = {
 };
 
 static void try_read_btree_node(struct find_btree_nodes *f, struct bch_dev *ca,
-				struct bio *bio, struct btree_node *bn, u64 offset)
+				struct btree *b, struct bio *bio, u64 offset)
 {
 	struct bch_fs *c = container_of(f, struct bch_fs, found_btree_nodes);
+	struct btree_node *bn = b->data;
 
 	bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ);
 	bio->bi_iter.bi_sector	= offset;
-	bch2_bio_map(bio, bn, PAGE_SIZE);
+	bch2_bio_map(bio, b->data, c->opts.block_size);
 
 	u64 submit_time = local_clock();
 	submit_bio_wait(bio);
-
 	bch2_account_io_completion(ca, BCH_MEMBER_ERROR_read, submit_time, !bio->bi_status);
 
 	if (bio->bi_status) {
@@ -201,6 +168,14 @@ static void try_read_btree_node(struct find_btree_nodes *f, struct bch_dev *ca,
 	if (BTREE_NODE_ID(bn) >= BTREE_ID_NR_MAX)
 		return;
 
+	bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ);
+	bio->bi_iter.bi_sector	= offset;
+	bch2_bio_map(bio, b->data, c->opts.btree_node_size);
+
+	submit_time = local_clock();
+	submit_bio_wait(bio);
+	bch2_account_io_completion(ca, BCH_MEMBER_ERROR_read, submit_time, !bio->bi_status);
+
 	rcu_read_lock();
 	struct found_btree_node n = {
 		.btree_id	= BTREE_NODE_ID(bn),
@@ -217,7 +192,20 @@ static void try_read_btree_node(struct find_btree_nodes *f, struct bch_dev *ca,
 	};
 	rcu_read_unlock();
 
-	if (bch2_trans_run(c, found_btree_node_is_readable(trans, &n))) {
+	found_btree_node_to_key(&b->key, &n);
+
+	CLASS(printbuf, buf)();
+	if (!bch2_btree_node_read_done(c, ca, b, NULL, &buf)) {
+		/* read_done will swap out b->data for another buffer */
+		bn = b->data;
+		/*
+		 * Grab journal_seq here because we want the max journal_seq of
+		 * any bset; read_done sorts down to a single set and picks the
+		 * max journal_seq
+		 */
+		n.journal_seq		= le64_to_cpu(bn->keys.journal_seq),
+		n.sectors_written	= b->written;
+
 		mutex_lock(&f->lock);
 		if (BSET_BIG_ENDIAN(&bn->keys) != CPU_BIG_ENDIAN) {
 			bch_err(c, "try_read_btree_node() can't handle endian conversion");
@@ -237,12 +225,20 @@ static int read_btree_nodes_worker(void *p)
 	struct find_btree_nodes_worker *w = p;
 	struct bch_fs *c = container_of(w->f, struct bch_fs, found_btree_nodes);
 	struct bch_dev *ca = w->ca;
-	void *buf = (void *) __get_free_page(GFP_KERNEL);
-	struct bio *bio = bio_alloc(NULL, 1, 0, GFP_KERNEL);
 	unsigned long last_print = jiffies;
+	struct btree *b = NULL;
+	struct bio *bio = NULL;
+
+	b = __bch2_btree_node_mem_alloc(c);
+	if (!b) {
+		bch_err(c, "read_btree_nodes_worker: error allocating buf");
+		w->f->ret = -ENOMEM;
+		goto err;
+	}
 
-	if (!buf || !bio) {
-		bch_err(c, "read_btree_nodes_worker: error allocating bio/buf");
+	bio = bio_alloc(NULL, buf_pages(b->data, c->opts.btree_node_size), 0, GFP_KERNEL);
+	if (!bio) {
+		bch_err(c, "read_btree_nodes_worker: error allocating bio");
 		w->f->ret = -ENOMEM;
 		goto err;
 	}
@@ -266,11 +262,13 @@ static int read_btree_nodes_worker(void *p)
 			    !bch2_dev_btree_bitmap_marked_sectors(ca, sector, btree_sectors(c)))
 				continue;
 
-			try_read_btree_node(w->f, ca, bio, buf, sector);
+			try_read_btree_node(w->f, ca, b, bio, sector);
 		}
 err:
+	if (b)
+		__btree_node_data_free(b);
+	kfree(b);
 	bio_put(bio);
-	free_page((unsigned long) buf);
 	enumerated_ref_put(&ca->io_ref[READ], BCH_DEV_READ_REF_btree_node_scan);
 	closure_put(w->cl);
 	kfree(w);

fs/bcachefs/debug.c

@@ -153,8 +153,6 @@ void __bch2_btree_verify(struct bch_fs *c, struct btree *b)
 		c->verify_data = __bch2_btree_node_mem_alloc(c);
 		if (!c->verify_data)
 			goto out;
-
-		list_del_init(&c->verify_data->list);
 	}
 
 	BUG_ON(b->nsets != 1);
@@ -586,6 +584,8 @@ static ssize_t bch2_btree_transactions_read(struct file *file, char __user *buf,
 	i->ubuf = buf;
 	i->size	= size;
 	i->ret	= 0;
+
+	int srcu_idx = srcu_read_lock(&c->btree_trans_barrier);
 restart:
 	seqmutex_lock(&c->btree_trans_lock);
 	list_sort(&c->btree_trans_list, list_ptr_order_cmp);
@@ -599,6 +599,11 @@ static ssize_t bch2_btree_transactions_read(struct file *file, char __user *buf,
 		if (!closure_get_not_zero(&trans->ref))
 			continue;
 
+		if (!trans->srcu_held) {
+			closure_put(&trans->ref);
+			continue;
+		}
+
 		u32 seq = seqmutex_unlock(&c->btree_trans_lock);
 
 		bch2_btree_trans_to_text(&i->buf, trans);
@@ -620,6 +625,8 @@ static ssize_t bch2_btree_transactions_read(struct file *file, char __user *buf,
 	}
 	seqmutex_unlock(&c->btree_trans_lock);
 unlocked:
+	srcu_read_unlock(&c->btree_trans_barrier, srcu_idx);
+
 	if (i->buf.allocation_failure)
 		ret = -ENOMEM;
 

fs/bcachefs/errcode.h

@@ -282,7 +282,6 @@
 	x(EIO,				sb_not_downgraded)			\
 	x(EIO,				btree_node_write_all_failed)		\
 	x(EIO,				btree_node_read_error)			\
-	x(EIO,				btree_node_read_validate_error)		\
 	x(EIO,				btree_need_topology_repair)		\
 	x(EIO,				bucket_ref_update)			\
 	x(EIO,				trigger_alloc)				\

fs/bcachefs/error.c

@@ -103,7 +103,7 @@ int __bch2_topology_error(struct bch_fs *c, struct printbuf *out)
 		return bch_err_throw(c, btree_need_topology_repair);
 	} else {
 		return bch2_run_explicit_recovery_pass(c, out, BCH_RECOVERY_PASS_check_topology, 0) ?:
-			bch_err_throw(c, btree_node_read_validate_error);
+			bch_err_throw(c, btree_need_topology_repair);
 	}
 }
 
@@ -633,7 +633,9 @@ int __bch2_fsck_err(struct bch_fs *c,
 	 * log_fsck_err()s: that would require us to track for every error type
 	 * which recovery pass corrects it, to get the fsck exit status correct:
 	 */
-	if (bch2_err_matches(ret, BCH_ERR_fsck_fix)) {
+	if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
+		/* nothing */
+	} else if (bch2_err_matches(ret, BCH_ERR_fsck_fix)) {
 		set_bit(BCH_FS_errors_fixed, &c->flags);
 	} else {
 		set_bit(BCH_FS_errors_not_fixed, &c->flags);

fs/bcachefs/extents.c

@@ -50,19 +50,17 @@ void bch2_io_failures_to_text(struct printbuf *out,
 			      struct bch_io_failures *failed)
 {
 	static const char * const error_types[] = {
-		"io", "checksum", "ec reconstruct", NULL
+		"btree validate", "io", "checksum", "ec reconstruct", NULL
 	};
 
 	for (struct bch_dev_io_failures *f = failed->devs;
 	     f < failed->devs + failed->nr;
 	     f++) {
 		unsigned errflags =
-			((!!f->failed_io)	<< 0) |
-			((!!f->failed_csum_nr)	<< 1) |
-			((!!f->failed_ec)	<< 2);
-
-		if (!errflags)
-			continue;
+			((!!f->failed_btree_validate)	<< 0) |
+			((!!f->failed_io)		<< 1) |
+			((!!f->failed_csum_nr)		<< 2) |
+			((!!f->failed_ec)		<< 3);
 
 		bch2_printbuf_make_room(out, 1024);
 		out->atomic++;
@@ -77,7 +75,9 @@ void bch2_io_failures_to_text(struct printbuf *out,
 
 		prt_char(out, ' ');
 
-		if (is_power_of_2(errflags)) {
+		if (!errflags) {
+			prt_str(out, "no error - confused");
+		} else if (is_power_of_2(errflags)) {
 			prt_bitflags(out, error_types, errflags);
 			prt_str(out, " error");
 		} else {