selftests/bpf: Reject graph kptr use after RCU unlockgrafted

Add a sleepable verifier test that loads a graph-node local kptr in an
explicit RCU read-side critical section, then passes its node to
bpf_rbtree_remove() after the section ends.

Before the verifier fix, the stale NON_OWN_REF flag makes the node look like
a live borrowed reference and the program is accepted. After the fix, the
pointer is demoted without NON_OWN_REF and the graph kfunc argument is
rejected.

Also exercise a graph kptr loaded while a spin lock provides implicit RCU
protection. The pointer must be invalidated when the lock is released, which
guards the required ordering between non-owning-reference invalidation and
RCU demotion.

Update the existing fault-protected load test state description. The
post-unlock pointer no longer carries NON_OWN_REF, but remains readable
because the load is rewritten to use BPF_PROBE_MEM.

Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260904084325.52250-7-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

1 files changed, 59 insertions, 0 deletions

diff --git a/drivers/md/dm-vdo/cpu.h b/drivers/md/dm-vdo/cpu.h
new file mode 100644
index 000000000..d6a2615ba
--- /dev/null
+++ b/drivers/md/dm-vdo/cpu.h
@@ -0,0 +1,59 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright 2023 Red Hat
+ */
+
+#ifndef UDS_CPU_H
+#define UDS_CPU_H
+
+#include <linux/cache.h>
+
+/**
+ * uds_prefetch_address() - Minimize cache-miss latency by attempting to move data into a CPU cache
+ *                          before it is accessed.
+ *
+ * @address: the address to fetch (may be invalid)
+ * @for_write: must be constant at compile time--false if for reading, true if for writing
+ */
+static inline void uds_prefetch_address(const void *address, bool for_write)
+{
+	/*
+	 * for_write won't be a constant if we are compiled with optimization turned off, in which
+	 * case prefetching really doesn't matter. clang can't figure out that if for_write is a
+	 * constant, it can be passed as the second, mandatorily constant argument to prefetch(),
+	 * at least currently on llvm 12.
+	 */
+	if (__builtin_constant_p(for_write)) {
+		if (for_write)
+			__builtin_prefetch(address, true);
+		else
+			__builtin_prefetch(address, false);
+	}
+}
+
+/**
+ * uds_prefetch_range() - Minimize cache-miss latency by attempting to move a range of addresses
+ *                        into a CPU cache before they are accessed.
+ *
+ * @start: the starting address to fetch (may be invalid)
+ * @size: the number of bytes in the address range
+ * @for_write: must be constant at compile time--false if for reading, true if for writing
+ */
+static inline void uds_prefetch_range(const void *start, unsigned int size,
+				      bool for_write)
+{
+	/*
+	 * Count the number of cache lines to fetch, allowing for the address range to span an
+	 * extra cache line boundary due to address alignment.
+	 */
+	const char *address = (const char *) start;
+	unsigned int offset = ((uintptr_t) address % L1_CACHE_BYTES);
+	unsigned int cache_lines = (1 + ((size + offset) / L1_CACHE_BYTES));
+
+	while (cache_lines-- > 0) {
+		uds_prefetch_address(address, for_write);
+		address += L1_CACHE_BYTES;
+	}
+}
+
+#endif /* UDS_CPU_H */