[PATCH bpf-next v5 0/9] Introduce arena library and runtime

[PATCH bpf-next v5 0/9] Introduce arena library and runtime

[Emil Tsalapatis]

Add a new subdirectory to tools/testing/selftests/bpf called libarena,
along with programs useful for writing arena-based BPF code. This
patchset adds the following:

1) libarena, a subdirectory where arena BPF code that is generally useful
to BPF arena programs can be easily added and tested.

2) An ASAN runtime for BPF arena programs. BPF arenas allow for accessing
memory after it has been freed or if it is out of bounds, making it more
difficult to triage bugs combined to regular BPF. Use LLVM's recently added
support for address-space based sanitization to selectively sanitize just
the arena accesses.

3) A buddy memory allocator that can be reused by BPF programs to handle
memory allocation/deletion. The allocator uses the ASAN runtime to add
address sanitization if requested.

The patch includes testing for the new allocators and ASAN features that
can be built from the top directory using "make libarena_test" and
"make libarena_test_asan". The generated binaries reside in libarena/.
The patch also adds test-progs-based selftests to the codebase for the
libarena code, so the new tests are run by ./test_progs.

The patchset has the following stucture:

1-2: Addresses a verification failure triggered by SCALAR += PTR_TO_ARENA
instructions.

3-4: Minor changes to selftest headers to prepare for the introduction
of libarena.

5-7: Add the libarena directory and testing scaffolding, and introduce
the ASAN runtime.

8-9: Add the new buddy memory allocator along with self-contained and
prog-tests-based selftests.

Signed-off-by: Emil Tsalapatis <emil@etsalapatis.com>

HISTORY
=======

v4->v5 (https://lore.kernel.org/bpf/20260407045730.13359-1-emil@etsalapatis.com)
Omitting various nits and fixups.
- Properly adjust subreg_def for scalar += ptr_to_arena calls (Sashiko)
- Remove extraneous definition from prog_tests/arena_spin_lock.c (Song)
- Trim extraneous comments from ASAN and buddy (Alexei)
- Remove asan_dummy call and replace with function pointer array (Alexei)
- Remove usersapi.h header and merge it into common.h (Alexei)
- Replace ASAN macros with function calls (Alexei)
- Embed buddy lock into the struct and move the buddy allocator to __arena_global
  (Alexei)
- Add commenting for buddy allocator constants (Alexei)
- Add default buddy allocator directly in common.bpf.c, so that the user does
  not need to define it.
- Expand test harnesses to dynamically find individual selftests. Now the
  selftests also reports each test individually (e.g., 5 entries for the
  buddy allocator instead of 1). This brings them to par with the rest of
  the test_progs.

v3->v4 (https://lore.kernel.org/bpf/20260403042720.18862-1-emil@etsalapatis.com)
- Add Acks by Song to patches 1-4.
- Expand the verifier's handling of scalar/arena operations to
  include all 3-operand operations in Patch 1 (Alexei)
- Add additional tests for arena/arena (allowed) and arena/pointer (not allowed)
operations in Patch 2
- Remove ASAN version of the library from default compilation since it requires
LLVM 22 and up (CI)
- Rework buddy allocator locking for clarity and add comments
- Fix From: email to be consistent with SOB
- Address (most) Sashiko comments

v2->v3 (https://lore.kernel.org/bpf/20260127181610.86376-1-emil@etsalapatis.com)
Nonexhaustive due to significant patch rework.
- Do not duplicate WRITE_ONCE macro (Mykyta, Kumar)
- Add SPDX headers (Alexei)
- Remove bump/stack allocators (Alexei)
- Integrate testing with test_progs (Kumar)
- Add short description of ASAN algorithm at the top of the file (Alexei)

v1->v2 (https://lore.kernel.org/bpf/20260122160131.2238331-1-etsal@meta.com/)

- Added missing format string argument (AI)
- Fix outdated selftests prog name check (AI)
- Fixed stack allocation check for segment creation (AI)
- Fix errors in non-ASAN bump allocator selftests (AI)
- Propagate error value from individual selftests in selftest.c
- Removed embedded metadata from bump allocator as it was needlessly
  complicating its behavior


Emil Tsalapatis (9):
  bpf: Upgrade scalar to PTR_TO_ARENA on arena pointer addition
  selftests/bpf: Add test for scalar/arena pointer addition
  selftests/bpf: Move bpf_arena_spin_lock.h to the top level
  selftests/bpf: Deduplicate WRITE_ONCE macro between headers
  selftests/bpf: Add basic libarena scaffolding
  selftests/bpf: Add arena ASAN runtime to libarena
  selftests/bpf: Add ASAN support for libarena selftests
  selftests/bpf: Add buddy allocator for libarena
  selftests/bpf: Add selftests for libarena buddy allocator

 kernel/bpf/verifier.c                         |  25 +-
 tools/testing/selftests/bpf/.gitignore        |   2 +
 tools/testing/selftests/bpf/Makefile          |  28 +-
 .../testing/selftests/bpf/bpf_arena_common.h  |   4 -
 .../bpf/{progs => }/bpf_arena_spin_lock.h     |   4 +-
 tools/testing/selftests/bpf/bpf_atomic.h      |   4 -
 .../testing/selftests/bpf/bpf_experimental.h  |   3 +
 tools/testing/selftests/bpf/libarena/Makefile |  77 ++
 .../selftests/bpf/libarena/include/asan.h     | 105 +++
 .../selftests/bpf/libarena/include/buddy.h    |  92 ++
 .../selftests/bpf/libarena/include/common.h   |  68 ++
 .../bpf/libarena/include/selftest_helpers.h   | 121 +++
 .../bpf/libarena/selftests/selftest.c         | 184 ++++
 .../libarena/selftests/st_asan_buddy.bpf.c    | 243 +++++
 .../bpf/libarena/selftests/st_asan_common.h   |  47 +
 .../bpf/libarena/selftests/st_buddy.bpf.c     | 211 +++++
 .../selftests/bpf/libarena/src/asan.bpf.c     | 550 +++++++++++
 .../selftests/bpf/libarena/src/buddy.bpf.c    | 879 ++++++++++++++++++
 .../selftests/bpf/libarena/src/common.bpf.c   |  71 ++
 .../bpf/prog_tests/arena_spin_lock.c          |   7 -
 .../selftests/bpf/prog_tests/libarena.c       |  61 ++
 .../selftests/bpf/progs/arena_spin_lock.c     |   2 +-
 .../selftests/bpf/progs/verifier_arena.c      |  85 ++
 23 files changed, 2852 insertions(+), 21 deletions(-)
 rename tools/testing/selftests/bpf/{progs => }/bpf_arena_spin_lock.h (99%)
 create mode 100644 tools/testing/selftests/bpf/libarena/Makefile
 create mode 100644 tools/testing/selftests/bpf/libarena/include/asan.h
 create mode 100644 tools/testing/selftests/bpf/libarena/include/buddy.h
 create mode 100644 tools/testing/selftests/bpf/libarena/include/common.h
 create mode 100644 tools/testing/selftests/bpf/libarena/include/selftest_helpers.h
 create mode 100644 tools/testing/selftests/bpf/libarena/selftests/selftest.c
 create mode 100644 tools/testing/selftests/bpf/libarena/selftests/st_asan_buddy.bpf.c
 create mode 100644 tools/testing/selftests/bpf/libarena/selftests/st_asan_common.h
 create mode 100644 tools/testing/selftests/bpf/libarena/selftests/st_buddy.bpf.c
 create mode 100644 tools/testing/selftests/bpf/libarena/src/asan.bpf.c
 create mode 100644 tools/testing/selftests/bpf/libarena/src/buddy.bpf.c
 create mode 100644 tools/testing/selftests/bpf/libarena/src/common.bpf.c
 create mode 100644 tools/testing/selftests/bpf/prog_tests/libarena.c

-- 
2.53.0

[PATCH bpf-next v5 1/9] bpf: Upgrade scalar to PTR_TO_ARENA on arena pointer addition

[Emil Tsalapatis]

The compiler sometimes stores the result of a PTR_TO_ARENA + SCALAR
addition into the scalar register rather than the pointer register.
Handle this case by upgrading the destination scalar register to
PTR_TO_ARENA, matching the existing handling when the destination is
already PTR_TO_ARENA.

Signed-off-by: Emil Tsalapatis <emil@etsalapatis.com>
Acked-by: Song Liu <song@kernel.org>
---
 kernel/bpf/verifier.c | 25 +++++++++++++++++++++++--
 1 file changed, 23 insertions(+), 2 deletions(-)

diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index 594260c1f382..30c3e26aae96 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -16641,11 +16641,32 @@ static int adjust_reg_min_max_vals(struct bpf_verifier_env *env,
 	int err;
 
 	dst_reg = &regs[insn->dst_reg];
-	src_reg = NULL;
+	if (BPF_SRC(insn->code) == BPF_X)
+		src_reg = &regs[insn->src_reg];
+	else
+		src_reg = NULL;
 
-	if (dst_reg->type == PTR_TO_ARENA) {
+	/* Case where at least one operand is an arena. */
+	if (dst_reg->type == PTR_TO_ARENA || (src_reg && src_reg->type == PTR_TO_ARENA)) {
 		struct bpf_insn_aux_data *aux = cur_aux(env);
 
+		if (dst_reg->type != PTR_TO_ARENA) {
+			/* Can't do arena arithmetic with non-scalars. */
+			if (dst_reg->type != SCALAR_VALUE) {
+				verbose(env, "R%d %s R%d: Invalid operation between "
+						"bpf_reg_state types %s and %s\n",
+					insn->dst_reg,
+					bpf_alu_string[opcode >> 4],
+					insn->src_reg,
+					reg_type_str(env, dst_reg->type),
+					reg_type_str(env, src_reg->type));
+				return -EACCES;
+			}
+
+			*dst_reg = *src_reg;
+			dst_reg->subreg_def = env->insn_idx + 1;
+		}
+
 		if (BPF_CLASS(insn->code) == BPF_ALU64)
 			/*
 			 * 32-bit operations zero upper bits automatically.
-- 
2.53.0

[PATCH bpf-next v5 2/9] selftests/bpf: Add test for scalar/arena pointer addition

[Emil Tsalapatis]

Add a selftest that ensures scalar += ptr_to_arena instructions
are accepted by the verifier.

Signed-off-by: Emil Tsalapatis <emil@etsalapatis.com>
Acked-by: Song Liu <song@kernel.org>
---
 .../selftests/bpf/progs/verifier_arena.c      | 85 +++++++++++++++++++
 1 file changed, 85 insertions(+)

diff --git a/tools/testing/selftests/bpf/progs/verifier_arena.c b/tools/testing/selftests/bpf/progs/verifier_arena.c
index c4b8daac4388..df5632ea777f 100644
--- a/tools/testing/selftests/bpf/progs/verifier_arena.c
+++ b/tools/testing/selftests/bpf/progs/verifier_arena.c
@@ -477,4 +477,89 @@ int arena_kfuncs_under_bpf_lock(void *ctx)
 
 	return 0;
 }
+
+#if defined(__BPF_FEATURE_ADDR_SPACE_CAST)
+
+/*
+ * Test that scalar += PTR_TO_ARENA correctly upgrades the
+ * destination register to a PTR_TO_ARENA.
+ */
+SEC("syscall")
+__success __retval(0)
+int scalar_add_arena_ptr(void *ctx)
+{
+	int __arena *scalar, *arena_ptr;
+
+	volatile char __arena *base = arena_base(&arena);
+
+	asm volatile (
+		"%[arena_ptr] = 8192;"
+		"%[arena_ptr] = addr_space_cast(%[arena_ptr], 0x0, 0x1);"
+		"%[scalar] = 12;"
+		"%[scalar] += %[arena_ptr];"
+		: [scalar] "=r"(scalar),
+		  [arena_ptr] "=&r"(arena_ptr)
+		: "r"(base)
+		:
+	);
+	return 0;
+}
+
+/*
+ * Test that PTR_TO_ARENA + PTR_TO_ARENA is allowed.
+ */
+SEC("syscall")
+__success __retval(0)
+int arena_ptr_add_arena_ptr(void *ctx)
+{
+	int __arena *arena_ptr2, *arena_ptr1;
+
+	/* Needed for the verifier to link the arena to the subprog. */
+	volatile char __arena *base = arena_base(&arena);
+
+	asm volatile (
+		"%[arena_ptr1] = 8192;"
+		"%[arena_ptr1] = addr_space_cast(%[arena_ptr1], 0x0, 0x1);"
+		"%[arena_ptr2] = 4096;"
+		"%[arena_ptr2] = addr_space_cast(%[arena_ptr2], 0x0, 0x1);"
+		"%[arena_ptr2] += %[arena_ptr1];"
+		: [arena_ptr2] "=r"(arena_ptr2),
+		  [arena_ptr1] "=&r"(arena_ptr1)
+		: "r"(base)
+		:
+	);
+	return 0;
+}
+
+/*
+ * Test that PTR_TO_ARENA + non-arena pointer is rejected.
+ */
+SEC("syscall")
+__failure __msg("R4 += R3: Invalid operation between bpf_reg_state types")
+int arena_ptr_add_to_non_arena_ptr(void *ctx)
+{
+	register int __arena *arena_ptr asm("r3");
+	register void *dst asm("r4");
+
+	/* Needed for the verifier to link the arena to the subprog. */
+	volatile char __arena *base = arena_base(&arena);
+
+	asm volatile (
+		"%[arena_ptr] = 8192;"
+		"%[arena_ptr] = addr_space_cast(%[arena_ptr], 0x0, 0x1);"
+		"%[dst] = %[ctx];"
+		"%[dst] += %[arena_ptr];"
+		: [arena_ptr] "=&r"(arena_ptr),
+		  [dst] "=&r"(dst)
+		: [ctx] "r"(ctx), "r"(base)
+		:
+	);
+
+	(void)ctx;
+
+	return 0;
+}
+
+#endif
+
 char _license[] SEC("license") = "GPL";
-- 
2.53.0

[PATCH bpf-next v5 3/9] selftests/bpf: Move bpf_arena_spin_lock.h to the top level

[Emil Tsalapatis]

The bpf_arena_spin_lock.h header is useful for all programs and not
just the selftests. Move it to the top level of the BPF selftests
to make it more readily accessible.

Signed-off-by: Emil Tsalapatis <emil@etsalapatis.com>
Acked-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Acked-by: Song Liu <song@kernel.org>
---
 .../selftests/bpf/{progs => }/bpf_arena_spin_lock.h        | 4 ++--
 tools/testing/selftests/bpf/prog_tests/arena_spin_lock.c   | 7 -------
 tools/testing/selftests/bpf/progs/arena_spin_lock.c        | 2 +-
 3 files changed, 3 insertions(+), 10 deletions(-)
 rename tools/testing/selftests/bpf/{progs => }/bpf_arena_spin_lock.h (99%)

diff --git a/tools/testing/selftests/bpf/progs/bpf_arena_spin_lock.h b/tools/testing/selftests/bpf/bpf_arena_spin_lock.h
similarity index 99%
rename from tools/testing/selftests/bpf/progs/bpf_arena_spin_lock.h
rename to tools/testing/selftests/bpf/bpf_arena_spin_lock.h
index f90531cf3ee5..680c9e6cb35d 100644
--- a/tools/testing/selftests/bpf/progs/bpf_arena_spin_lock.h
+++ b/tools/testing/selftests/bpf/bpf_arena_spin_lock.h
@@ -107,7 +107,7 @@ struct arena_qnode {
 #define _Q_LOCKED_VAL		(1U << _Q_LOCKED_OFFSET)
 #define _Q_PENDING_VAL		(1U << _Q_PENDING_OFFSET)
 
-struct arena_qnode __arena qnodes[_Q_MAX_CPUS][_Q_MAX_NODES];
+struct arena_qnode __weak __arena __hidden qnodes[_Q_MAX_CPUS][_Q_MAX_NODES];
 
 static inline u32 encode_tail(int cpu, int idx)
 {
@@ -240,7 +240,7 @@ static __always_inline int arena_spin_trylock(arena_spinlock_t __arena *lock)
 	return likely(atomic_try_cmpxchg_acquire(&lock->val, &val, _Q_LOCKED_VAL));
 }
 
-__noinline
+__noinline __weak
 int arena_spin_lock_slowpath(arena_spinlock_t __arena __arg_arena *lock, u32 val)
 {
 	struct arena_mcs_spinlock __arena *prev, *next, *node0, *node;
diff --git a/tools/testing/selftests/bpf/prog_tests/arena_spin_lock.c b/tools/testing/selftests/bpf/prog_tests/arena_spin_lock.c
index 693fd86fbde6..acb9d53b5973 100644
--- a/tools/testing/selftests/bpf/prog_tests/arena_spin_lock.c
+++ b/tools/testing/selftests/bpf/prog_tests/arena_spin_lock.c
@@ -5,13 +5,6 @@
 #include <sys/sysinfo.h>
 
 struct __qspinlock { int val; };
-typedef struct __qspinlock arena_spinlock_t;
-
-struct arena_qnode {
-	unsigned long next;
-	int count;
-	int locked;
-};
 
 #include "arena_spin_lock.skel.h"
 
diff --git a/tools/testing/selftests/bpf/progs/arena_spin_lock.c b/tools/testing/selftests/bpf/progs/arena_spin_lock.c
index 086b57a426cf..6c04e7707644 100644
--- a/tools/testing/selftests/bpf/progs/arena_spin_lock.c
+++ b/tools/testing/selftests/bpf/progs/arena_spin_lock.c
@@ -4,7 +4,7 @@
 #include <bpf/bpf_tracing.h>
 #include <bpf/bpf_helpers.h>
 #include "bpf_misc.h"
-#include "bpf_arena_spin_lock.h"
+#include "../bpf_arena_spin_lock.h"
 
 struct {
 	__uint(type, BPF_MAP_TYPE_ARENA);
-- 
2.53.0

[PATCH bpf-next v5 4/9] selftests/bpf: Deduplicate WRITE_ONCE macro between headers

[Emil Tsalapatis]

The WRITE_ONCE macro is identically defined both in bpf_atomic.h
and in bpf_arena_common.h. Deduplicate the definitions by moving
the macro to the commonly used bpf_experimental.h header.

Signed-off-by: Emil Tsalapatis <emil@etsalapatis.com>
Acked-by: Song Liu <song@kernel.org>
---
 tools/testing/selftests/bpf/bpf_arena_common.h | 4 ----
 tools/testing/selftests/bpf/bpf_atomic.h       | 4 ----
 tools/testing/selftests/bpf/bpf_experimental.h | 3 +++
 3 files changed, 3 insertions(+), 8 deletions(-)

diff --git a/tools/testing/selftests/bpf/bpf_arena_common.h b/tools/testing/selftests/bpf/bpf_arena_common.h
index 16f8ce832004..a0503ecbc315 100644
--- a/tools/testing/selftests/bpf/bpf_arena_common.h
+++ b/tools/testing/selftests/bpf/bpf_arena_common.h
@@ -2,10 +2,6 @@
 /* Copyright (c) 2024 Meta Platforms, Inc. and affiliates. */
 #pragma once
 
-#ifndef WRITE_ONCE
-#define WRITE_ONCE(x, val) ((*(volatile typeof(x) *) &(x)) = (val))
-#endif
-
 #ifndef NUMA_NO_NODE
 #define	NUMA_NO_NODE	(-1)
 #endif
diff --git a/tools/testing/selftests/bpf/bpf_atomic.h b/tools/testing/selftests/bpf/bpf_atomic.h
index c550e5711967..d77338ac7e6b 100644
--- a/tools/testing/selftests/bpf/bpf_atomic.h
+++ b/tools/testing/selftests/bpf/bpf_atomic.h
@@ -40,10 +40,6 @@ extern bool CONFIG_X86_64 __kconfig __weak;
 /* No-op for BPF */
 #define cpu_relax() ({})
 
-#define READ_ONCE(x) (*(volatile typeof(x) *)&(x))
-
-#define WRITE_ONCE(x, val) ((*(volatile typeof(x) *)&(x)) = (val))
-
 #define cmpxchg(p, old, new) __sync_val_compare_and_swap((p), old, new)
 
 #define try_cmpxchg(p, pold, new)                                 \
diff --git a/tools/testing/selftests/bpf/bpf_experimental.h b/tools/testing/selftests/bpf/bpf_experimental.h
index 2234bd6bc9d3..56214f7051c6 100644
--- a/tools/testing/selftests/bpf/bpf_experimental.h
+++ b/tools/testing/selftests/bpf/bpf_experimental.h
@@ -11,6 +11,9 @@
 /* Convenience macro to wrap over bpf_obj_new */
 #define bpf_obj_new(type) ((type *)bpf_obj_new(bpf_core_type_id_local(type)))
 
+#define READ_ONCE(x) (*(volatile typeof(x) *)&(x))
+#define WRITE_ONCE(x, val) ((*(volatile typeof(x) *)&(x)) = (val))
+
 /* Convenience macro to wrap over bpf_percpu_obj_new */
 #define bpf_percpu_obj_new(type) ((type __percpu_kptr *)bpf_percpu_obj_new(bpf_core_type_id_local(type)))
 
-- 
2.53.0

[PATCH bpf-next v5 5/9] selftests/bpf: Add basic libarena scaffolding

[Emil Tsalapatis]

Add initial code for an arena-based BPF library. The current commit
introduces a test runner and Makefile for the library. Library code
can be added just by including the source file in the library's src/
subdirectory. Future commits will introduce the library code itself.

The current commit also includes a standalone test runner. This is
to keep libarena self-contained and testable even when copied out
of the kernel tree. Subsequent commits add integration with test_progs.

Signed-off-by: Emil Tsalapatis <emil@etsalapatis.com>
---
 tools/testing/selftests/bpf/.gitignore        |   1 +
 tools/testing/selftests/bpf/Makefile          |  15 ++
 tools/testing/selftests/bpf/libarena/Makefile |  50 ++++++
 .../selftests/bpf/libarena/include/common.h   |  67 +++++++++
 .../bpf/libarena/include/selftest_helpers.h   |  86 +++++++++++
 .../bpf/libarena/selftests/selftest.c         | 142 ++++++++++++++++++
 .../selftests/bpf/libarena/src/common.bpf.c   |  59 ++++++++
 7 files changed, 420 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/libarena/Makefile
 create mode 100644 tools/testing/selftests/bpf/libarena/include/common.h
 create mode 100644 tools/testing/selftests/bpf/libarena/include/selftest_helpers.h
 create mode 100644 tools/testing/selftests/bpf/libarena/selftests/selftest.c
 create mode 100644 tools/testing/selftests/bpf/libarena/src/common.bpf.c

diff --git a/tools/testing/selftests/bpf/.gitignore b/tools/testing/selftests/bpf/.gitignore
index bfdc5518ecc8..7f1960d6b59e 100644
--- a/tools/testing/selftests/bpf/.gitignore
+++ b/tools/testing/selftests/bpf/.gitignore
@@ -49,3 +49,4 @@ verification_cert.h
 *.BTF.base
 usdt_1
 usdt_2
+libarena/test_libarena
diff --git a/tools/testing/selftests/bpf/Makefile b/tools/testing/selftests/bpf/Makefile
index f75c4f52c028..0d332c991023 100644
--- a/tools/testing/selftests/bpf/Makefile
+++ b/tools/testing/selftests/bpf/Makefile
@@ -153,6 +153,7 @@ override define CLEAN
 	$(Q)$(RM) -r $(TEST_KMODS)
 	$(Q)$(RM) -r $(EXTRA_CLEAN)
 	$(Q)$(MAKE) -C test_kmods clean
+	$(Q)$(MAKE) -C libarena clean
 	$(Q)$(MAKE) docs-clean
 endef
 
@@ -739,6 +740,17 @@ $(VERIFY_SIG_HDR): $(VERIFICATION_CERT)
 	 echo "};"; \
 	 echo "unsigned int test_progs_verification_cert_len = $$(wc -c < $<);") > $@
 
+LIBARENA_MAKE_ARGS = \
+		BPFTOOL="$(BPFTOOL)" \
+		INCLUDE_DIR="$(HOST_INCLUDE_DIR)" \
+		LIBBPF_INCLUDE="$(HOST_INCLUDE_DIR)" \
+		BPFOBJ="$(BPFOBJ)" \
+		LDLIBS="$(LDLIBS) -lzstd" \
+		CLANG="$(CLANG)" \
+		BPF_CFLAGS="$(BPF_CFLAGS) $(CLANG_CFLAGS)" \
+		BPF_TARGET_ENDIAN="$(BPF_TARGET_ENDIAN)" \
+		Q="$(Q)"
+
 # Define test_progs test runner.
 TRUNNER_TESTS_DIR := prog_tests
 TRUNNER_BPF_PROGS_DIR := progs
@@ -931,3 +943,6 @@ override define INSTALL_RULE
 		rsync -a $(OUTPUT)/$$DIR/*.bpf.o $(INSTALL_PATH)/$$DIR;\
 	done
 endef
+
+test_libarena: $(INCLUDE_DIR)/vmlinux.h $(BPFOBJ)
+	+$(MAKE) -C libarena $@ $(LIBARENA_MAKE_ARGS)
diff --git a/tools/testing/selftests/bpf/libarena/Makefile b/tools/testing/selftests/bpf/libarena/Makefile
new file mode 100644
index 000000000000..151ab6d0509c
--- /dev/null
+++ b/tools/testing/selftests/bpf/libarena/Makefile
@@ -0,0 +1,50 @@
+# SPDX-License-Identifier: LGPL-2.1 OR BSD-2-Clause
+# Copyright (c) 2026 Meta Platforms, Inc. and affiliates.
+
+.PHONY: clean
+
+LIBARENA=$(abspath .)
+
+
+LIBARENA_SOURCES = $(wildcard $(LIBARENA)/src/*.bpf.c) $(wildcard $(LIBARENA)/selftests/*.bpf.c)
+LIBARENA_OBJECTS = $(notdir $(LIBARENA_SOURCES:.bpf.c=.bpf.o))
+
+INCLUDES = -I$(LIBARENA)/include -I$(LIBARENA)/..
+ifneq ($(INCLUDE_DIR),)
+INCLUDES += -I$(INCLUDE_DIR)
+endif
+ifneq ($(LIBBPF_INCLUDE),)
+INCLUDES += -I$(LIBBPF_INCLUDE)
+endif
+
+# ENABLE_ATOMICS_TESTS required because we use arena spinlocks
+override BPF_CFLAGS += -DENABLE_ATOMICS_TESTS
+override BPF_CFLAGS += -O2 -Wno-incompatible-pointer-types-discards-qualifiers
+override BPF_CFLAGS += $(INCLUDES)
+
+CFLAGS = -O2 -no-pie
+CFLAGS += $(INCLUDES)
+
+vpath %.bpf.c $(LIBARENA)/src $(LIBARENA)/selftests
+vpath %.c $(LIBARENA)/src $(LIBARENA)/selftests
+
+all: test_libarena
+
+test_libarena: selftest.c $(BPFOBJ) libarena.skel.h
+	$(call msg,BINARY,libarena,$@)
+	$(Q)$(CLANG) $(CFLAGS) $< $(BPFOBJ) $(LDLIBS) -o $@
+
+libarena.skel.h: main.bpf.o
+	$(call msg,GEN-SKEL,libarena,$@)
+	$(Q)$(BPFTOOL) gen skeleton $< name "libarena" > $@
+
+main.bpf.o: $(LIBARENA_OBJECTS)
+	$(call msg,GEN-OBJ,libarena,$@)
+	$(Q)$(BPFTOOL) gen object $@ $^
+
+%.bpf.o: %.bpf.c
+	$(call msg,CLNG-BPF,libarena,$@)
+	$(Q)$(CLANG) $(BPF_CFLAGS) $(BPF_TARGET_ENDIAN) -c $< -o $@
+
+clean:
+	$(Q)rm -f *.skel.h *.bpf.o test_libarena
diff --git a/tools/testing/selftests/bpf/libarena/include/common.h b/tools/testing/selftests/bpf/libarena/include/common.h
new file mode 100644
index 000000000000..5acf9b994943
--- /dev/null
+++ b/tools/testing/selftests/bpf/libarena/include/common.h
@@ -0,0 +1,67 @@
+// SPDX-License-Identifier: LGPL-2.1 OR BSD-2-Clause
+/* Copyright (c) 2026 Meta Platforms, Inc. and affiliates. */
+#pragma once
+
+#ifdef __BPF__
+
+#include <vmlinux.h>
+
+#include "bpf_experimental.h"
+#include "bpf_arena_common.h"
+#include "bpf_arena_spin_lock.h"
+
+#include <asm-generic/errno.h>
+
+#ifndef __BPF_FEATURE_ADDR_SPACE_CAST
+#error "Arena allocators require bpf_addr_space_cast feature"
+#endif
+
+#define arena_stdout(fmt, ...) bpf_stream_printk(1, (fmt), ##__VA_ARGS__)
+#define arena_stderr(fmt, ...) bpf_stream_printk(2, (fmt), ##__VA_ARGS__)
+
+#ifndef __maybe_unused
+#define __maybe_unused __attribute__((__unused__))
+#endif
+
+#define private(name) SEC(".data." #name) __hidden __attribute__((aligned(8)))
+
+#define ARENA_PAGES (1UL << (32 - __builtin_ffs(__PAGE_SIZE) + 1))
+
+struct {
+	__uint(type, BPF_MAP_TYPE_ARENA);
+	__uint(map_flags, BPF_F_MMAPABLE);
+	__uint(max_entries, ARENA_PAGES); /* number of pages */
+#if defined(__TARGET_ARCH_arm64) || defined(__aarch64__)
+	__ulong(map_extra, (1ull << 32)); /* start of mmap() region */
+#else
+	__ulong(map_extra, (1ull << 44)); /* start of mmap() region */
+#endif
+} arena __weak SEC(".maps");
+
+extern const volatile u32 zero;
+
+int arena_fls(__u64 word);
+
+#else /* ! __BPF__ */
+
+#include <stdint.h>
+
+#define __arena
+
+typedef uint8_t u8;
+typedef uint16_t u16;
+typedef uint32_t u32;
+typedef uint64_t u64;
+typedef int8_t s8;
+typedef int16_t s16;
+typedef int32_t s32;
+typedef int64_t s64;
+
+/* Dummy "definition" for userspace. */
+#define arena_spinlock_t int
+
+#endif /* __BPF__ */
+
+struct arena_get_base_args {
+	void __arena *arena_base;
+};
diff --git a/tools/testing/selftests/bpf/libarena/include/selftest_helpers.h b/tools/testing/selftests/bpf/libarena/include/selftest_helpers.h
new file mode 100644
index 000000000000..78db8bc0fc67
--- /dev/null
+++ b/tools/testing/selftests/bpf/libarena/include/selftest_helpers.h
@@ -0,0 +1,86 @@
+// SPDX-License-Identifier: LGPL-2.1 OR BSD-2-Clause
+/* Copyright (c) 2026 Meta Platforms, Inc. and affiliates. */
+#pragma once
+
+#include <errno.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <sys/mman.h>
+
+#include <bpf/libbpf.h>
+#include <bpf/bpf.h>
+
+static inline int libarena_run_prog(int prog_fd)
+{
+	LIBBPF_OPTS(bpf_test_run_opts, opts);
+	int ret;
+
+	ret = bpf_prog_test_run_opts(prog_fd, &opts);
+	if (ret)
+		return ret;
+
+	return opts.retval;
+}
+
+static inline bool libarena_is_test_prog(const char *name)
+{
+	return strstr(name, "test_") != NULL;
+}
+
+static inline int libarena_get_arena_base(int arena_get_base_fd,
+					  void **arena_base)
+{
+	LIBBPF_OPTS(bpf_test_run_opts, opts);
+	struct arena_get_base_args args = { .arena_base = NULL };
+	int ret;
+
+	opts.ctx_in = &args;
+	opts.ctx_size_in = sizeof(args);
+
+	ret = bpf_prog_test_run_opts(arena_get_base_fd, &opts);
+	if (ret)
+		return ret;
+	if (opts.retval)
+		return opts.retval;
+
+	*arena_base = args.arena_base;
+	return 0;
+}
+
+static inline int libarena_get_globals_pages(int arena_get_base_fd,
+					     size_t arena_all_pages,
+					     u64 *globals_pages)
+{
+	size_t pgsize = sysconf(_SC_PAGESIZE);
+	void *arena_base;
+	ssize_t i;
+	u8 *vec;
+	int ret;
+
+	ret = libarena_get_arena_base(arena_get_base_fd, &arena_base);
+	if (ret)
+		return ret;
+
+	if (!arena_base)
+		return -EINVAL;
+
+	vec = calloc(arena_all_pages, sizeof(*vec));
+	if (!vec)
+		return -ENOMEM;
+
+	if (mincore(arena_base, arena_all_pages * pgsize, vec) < 0) {
+		ret = -errno;
+		free(vec);
+		return ret;
+	}
+
+	*globals_pages = 0;
+	for (i = arena_all_pages - 1; i >= 0; i--) {
+		if (!(vec[i] & 0x1))
+			break;
+		*globals_pages += 1;
+	}
+
+	free(vec);
+	return 0;
+}
diff --git a/tools/testing/selftests/bpf/libarena/selftests/selftest.c b/tools/testing/selftests/bpf/libarena/selftests/selftest.c
new file mode 100644
index 000000000000..da70648968d4
--- /dev/null
+++ b/tools/testing/selftests/bpf/libarena/selftests/selftest.c
@@ -0,0 +1,142 @@
+// SPDX-License-Identifier: LGPL-2.1 OR BSD-2-Clause
+/* Copyright (c) 2026 Meta Platforms, Inc. and affiliates. */
+
+#define _GNU_SOURCE
+#include <assert.h>
+#include <errno.h>
+#include <stdio.h>
+#include <string.h>
+#include <unistd.h>
+
+#include <bpf/libbpf.h>
+#include <bpf/bpf.h>
+
+#include <sys/mman.h>
+#include <sys/resource.h>
+#include <sys/sysinfo.h>
+
+#include <common.h>
+#include <selftest_helpers.h>
+
+#include "../libarena.skel.h"
+typedef struct libarena selftest;
+#define selftest__open libarena__open
+#define selftest__open_and_load libarena__open_and_load
+#define selftest__load libarena__load
+#define selftest__attach libarena__attach
+#define selftest__destroy libarena__destroy
+
+static bool verbose;
+static int testno = 1;
+
+static int
+run_prog_verbose(int prog_fd)
+{
+	char buf[1024];
+	int ret, err;
+
+	ret = libarena_run_prog(prog_fd);
+
+	if (ret)
+		fprintf(stderr, "error %d in %s\n", ret, __func__);
+
+	if (verbose) {
+		printf("BPF stdout:\n");
+		while ((err = bpf_prog_stream_read(prog_fd, 1, buf, 1024, NULL)) > 0)
+			printf("%.*s", err, buf);
+
+		if (err)
+			return err;
+
+		printf("BPF stderr:\n");
+		while ((err = bpf_prog_stream_read(prog_fd, 2, buf, 1024, NULL)) > 0)
+			printf("%.*s", err, buf);
+
+		if (err)
+			return err;
+	}
+
+	return ret;
+}
+
+static int libbpf_print_fn(enum libbpf_print_level level,
+		const char *format, va_list args)
+{
+	if (level == LIBBPF_DEBUG)
+		return 0;
+	return vfprintf(stderr, format, args);
+}
+
+static int run_test(selftest *skel, struct bpf_program *prog)
+{
+	int prog_fd;
+
+	prog_fd = bpf_program__fd(prog);
+	if (prog_fd < 0)
+		return prog_fd;
+
+	return run_prog_verbose(prog_fd);
+}
+
+static void
+banner(const char *progpath)
+{
+	char *name = basename(progpath);
+
+	printf("%s\n", name);
+
+	printf("=== %s ===\n", "libarena selftests");
+}
+
+int main(int argc, char *argv[])
+{
+	struct bpf_program *prog;
+	selftest *skel;
+	int ret;
+
+	banner(argv[0]);
+
+	for (int i = 1; i < argc; i++) {
+		if (strcmp(argv[i], "-v") == 0 || strcmp(argv[i], "--verbose") == 0)
+			verbose = true;
+	}
+
+	libbpf_set_print(libbpf_print_fn);
+
+	skel = selftest__open_and_load();
+	if (!skel) {
+		fprintf(stderr, "Failed to open and load skeleton\n");
+		return 1;
+	}
+
+	ret = selftest__attach(skel);
+	if (ret) {
+		fprintf(stderr, "Failed to attach skeleton\n");
+		selftest__destroy(skel);
+		return 1;
+	}
+
+	ret = libarena_run_prog(bpf_program__fd(skel->progs.arena_alloc_reserve));
+	if (ret) {
+		fprintf(stderr, "Failed to initialize arena: %d\n", ret);
+		selftest__destroy(skel);
+		return 1;
+	}
+
+	bpf_object__for_each_program(prog, skel->obj) {
+		const char *name = bpf_program__name(prog);
+
+		if (!libarena_is_test_prog(name))
+			continue;
+
+		ret = run_test(skel, prog);
+		if (ret)
+			printf("not ok %d - %s\n", testno++, name);
+		else
+			printf("ok %d - %s\n", testno++, name);
+	}
+
+	selftest__destroy(skel);
+
+	return 0;
+}
diff --git a/tools/testing/selftests/bpf/libarena/src/common.bpf.c b/tools/testing/selftests/bpf/libarena/src/common.bpf.c
new file mode 100644
index 000000000000..4f9e95aa6cd5
--- /dev/null
+++ b/tools/testing/selftests/bpf/libarena/src/common.bpf.c
@@ -0,0 +1,59 @@
+// SPDX-License-Identifier: LGPL-2.1 OR BSD-2-Clause
+/* Copyright (c) 2026 Meta Platforms, Inc. and affiliates. */
+#include <common.h>
+
+const volatile u32 zero = 0;
+
+/* How many pages do we reserve at the beginning of the arena segment? */
+#define RESERVE_ALLOC (8)
+
+int arena_fls(__u64 word)
+{
+	unsigned int num = 0;
+
+	if (word & 0xffffffff00000000ULL) {
+		num += 32;
+		word >>= 32;
+	}
+
+	if (word & 0xffff0000) {
+		num += 16;
+		word >>= 16;
+	}
+
+	if (word & 0xff00) {
+		num += 8;
+		word >>= 8;
+	}
+
+	if (word & 0xf0) {
+		num += 4;
+		word >>= 4;
+	}
+
+	if (word & 0xc) {
+		num += 2;
+		word >>= 2;
+	}
+
+	if (word & 0x2)
+		num += 1;
+
+	return num;
+}
+
+SEC("syscall")
+__weak int arena_get_base(struct arena_get_base_args *args)
+{
+	args->arena_base = arena_base(&arena);
+
+	return 0;
+}
+
+SEC("syscall")
+__weak int arena_alloc_reserve(void)
+{
+	return bpf_arena_reserve_pages(&arena, NULL, RESERVE_ALLOC);
+}
+
+char _license[] SEC("license") = "GPL";
-- 
2.53.0

[PATCH bpf-next v5 6/9] selftests/bpf: Add arena ASAN runtime to libarena

[Emil Tsalapatis]

Add an address sanitizer (ASAN) runtime to the arena library. The
ASAN runtime implements the functions injected into BPF binaries
by LLVM sanitization when ASAN is enabled during compilation.

The runtime also includes functions called explicitly by memory
allocation code to mark memory as poisoned/unpoisoned to ASAN.
This code is a no-op when sanitization is turned off.

Signed-off-by: Emil Tsalapatis <emil@etsalapatis.com>
---
 .../selftests/bpf/libarena/include/asan.h     | 105 ++++
 .../selftests/bpf/libarena/include/common.h   |   1 +
 .../selftests/bpf/libarena/src/asan.bpf.c     | 550 ++++++++++++++++++
 3 files changed, 656 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/libarena/include/asan.h
 create mode 100644 tools/testing/selftests/bpf/libarena/src/asan.bpf.c

diff --git a/tools/testing/selftests/bpf/libarena/include/asan.h b/tools/testing/selftests/bpf/libarena/include/asan.h
new file mode 100644
index 000000000000..c02f632f4b59
--- /dev/null
+++ b/tools/testing/selftests/bpf/libarena/include/asan.h
@@ -0,0 +1,105 @@
+// SPDX-License-Identifier: LGPL-2.1 OR BSD-2-Clause
+/* Copyright (c) 2026 Meta Platforms, Inc. and affiliates. */
+#pragma once
+
+struct asan_init_args {
+	u64 arena_all_pages;
+	u64 arena_globals_pages;
+};
+
+int asan_init(struct asan_init_args *args);
+
+extern volatile u64 __asan_shadow_memory_dynamic_address;
+extern volatile u32 asan_reported;
+extern volatile bool asan_inited;
+extern volatile bool asan_report_once;
+extern volatile bool asan_emit_stack;
+
+#ifdef __BPF__
+
+#define ASAN_SHADOW_SHIFT 3
+#define ASAN_SHADOW_SCALE (1ULL << ASAN_SHADOW_SHIFT)
+#define ASAN_GRANULE_MASK ((1ULL << ASAN_SHADOW_SHIFT) - 1)
+#define ASAN_GRANULE(addr) ((s8)((u32)(u64)((addr)) & ASAN_GRANULE_MASK))
+
+#define __noasan __attribute__((no_sanitize("address")))
+
+#ifdef BPF_ARENA_ASAN
+
+typedef s8 __arena s8a;
+
+static inline
+s8a *mem_to_shadow(void __arena __arg_arena *addr)
+{
+	return (s8a *)(((u32)(u64)addr >> ASAN_SHADOW_SHIFT) +
+			__asan_shadow_memory_dynamic_address);
+}
+
+__weak __noasan
+bool asan_ready(void)
+{
+	return __asan_shadow_memory_dynamic_address;
+}
+
+int asan_poison(void __arena *addr, s8 val, size_t size);
+int asan_unpoison(void __arena *addr, size_t size);
+bool asan_shadow_set(void __arena *addr);
+
+/*
+ * Dummy calls to ensure the ASAN runtime's BTF information is present
+ * in every object file when compiling the runtime and local BPF code
+ * separately. The runtime calls are injected into the LLVM IR file
+ */
+#define DECLARE_ASAN_LOAD_STORE_SIZE(size)				\
+	void __asan_store##size(void *addr);				\
+	void __asan_store##size##_noabort(void *addr);	\
+	void __asan_load##size(void *addr);				\
+	void __asan_load##size##_noabort(void *addr);	\
+	void __asan_report_store##size(void *addr);			\
+	void __asan_report_store##size##_noabort(void *addr);		\
+	void __asan_report_load##size(void *addr);			\
+	void __asan_report_load##size##_noabort(void *addr);
+
+DECLARE_ASAN_LOAD_STORE_SIZE(1);
+DECLARE_ASAN_LOAD_STORE_SIZE(2);
+DECLARE_ASAN_LOAD_STORE_SIZE(4);
+DECLARE_ASAN_LOAD_STORE_SIZE(8);
+
+/*
+ * Force LLVM to emit BTF information for the stubs,
+ * because the ASAN pass in LLVM by itself doesn't.
+ */
+#define ASAN_LOAD_STORE_SIZE(size)		\
+	__asan_store##size,			\
+	__asan_store##size##_noabort,		\
+	__asan_load##size,			\
+	__asan_load##size##_noabort,		\
+	__asan_report_store##size,		\
+	__asan_report_store##size##_noabort,	\
+	__asan_report_load##size,		\
+	__asan_report_load##size##_noabort
+
+/*
+ * Force LLVM to generate BTF information for the ASAN function
+ * declarations without invoking them. The used attribute prevents
+ * the function pointer table from being optimized out by the
+ * compiler.
+ */
+__attribute__((used))
+static void (*__asan_btf_anchors[])(void *) = {
+	ASAN_LOAD_STORE_SIZE(1),
+	ASAN_LOAD_STORE_SIZE(2),
+	ASAN_LOAD_STORE_SIZE(4),
+	ASAN_LOAD_STORE_SIZE(8)
+};
+
+#else /* BPF_ARENA_ASAN */
+
+static inline int asan_poison(void __arena *addr, s8 val, size_t size) { return 0; }
+static inline int asan_unpoison(void __arena *addr, size_t size) { return 0; }
+static inline bool asan_shadow_set(void __arena *addr) { return 0; }
+__weak bool asan_ready(void) { return true; }
+
+#endif /* BPF_ARENA_ASAN */
+
+#endif /* __BPF__ */
diff --git a/tools/testing/selftests/bpf/libarena/include/common.h b/tools/testing/selftests/bpf/libarena/include/common.h
index 5acf9b994943..881b6d06369d 100644
--- a/tools/testing/selftests/bpf/libarena/include/common.h
+++ b/tools/testing/selftests/bpf/libarena/include/common.h
@@ -39,6 +39,7 @@ struct {
 } arena __weak SEC(".maps");
 
 extern const volatile u32 zero;
+extern volatile u64 asan_violated;
 
 int arena_fls(__u64 word);
 
diff --git a/tools/testing/selftests/bpf/libarena/src/asan.bpf.c b/tools/testing/selftests/bpf/libarena/src/asan.bpf.c
new file mode 100644
index 000000000000..14b6e62383d1
--- /dev/null
+++ b/tools/testing/selftests/bpf/libarena/src/asan.bpf.c
@@ -0,0 +1,550 @@
+// SPDX-License-Identifier: LGPL-2.1 OR BSD-2-Clause
+/* Copyright (c) 2026 Meta Platforms, Inc. and affiliates. */
+#include <vmlinux.h>
+#include <common.h>
+#include <asan.h>
+
+
+enum {
+	/*
+	 * Is the access checked by check_region_inline
+	 * a read or a write?
+	 */
+	ASAN_READ		= 0x0U,
+	ASAN_WRITE		= 0x1U,
+};
+
+/*
+ * Address sanitizer (ASAN) for arena-based BPF programs, inspired
+ * by KASAN.
+ *
+ * The API
+ * -------
+ *
+ * The implementation includes two kinds of components: Implementation
+ * of ASAN hooks injected by LLVM into the program, and API calls that
+ * allocators use to mark memory as valid or invalid. The full list is:
+ *
+ * LLVM stubs:
+ *
+ * void __asan_{load, store}<size>(void *addr)
+ *	Checks whether an access is valid. All variations covered
+ *	by check_region_inline().
+ *
+ * void __asan_{store, load}((void *addr, ssize_t size)
+ *
+ * void __asan_report_{load, store}<size>(void *addr)
+ *	Report an access violation for the program. Used when LLVM
+ *	uses direct code generation for shadow map checks.
+ *
+ * void *__asan_memcpy(void *d, const void *s, size_t n)
+ * void *__asan_memmove(void *d, const void *s, size_t n)
+ * void *__asan_memset(void *p, int c, size_t n)
+ *	Hooks for ASAN instrumentation of the LLVM mem* builtins.
+ *	Currently unimplemented just like the builtins themselves.
+ *
+ * API methods:
+ *
+ * asan_init()
+ *	Initialize the ASAN map for the arena.
+ *
+ * asan_poison()
+ *	Mark a region of memory as poisoned. Accessing poisoned memory
+ *	causes asan_report() to fire. Invoked during free().
+ *
+ * asan_unpoison()
+ *	Mark a region as unpoisoned after alloc().
+ *
+ * asan_shadow_set()
+ *	Check a byte's validity directly.
+ *
+ * The Algorithm In Brief
+ * ----------------------
+ * Each group of 8 bytes is mapped to a "granule" in the shadow map. This
+ * granule is the size of the byte and describes which bytes are valid.
+ * Possible values are:
+ *
+ * 0: All bytes are valid. Makes checks in the middle of an allocated region
+ * (most of them) fast.
+ * (0, 7]: How many consecutive bytes are valid, starting from the lowest one.
+ * The tradeoff is that we can't poison individual bytes in the middle of a
+ * valid region.
+ * [0x80, 0xff]: Special poison values, can be used to denote specific error
+ * modes (e.g., recently freed vs uninitialized memory).
+ *
+ * The mapping between a memory location and its shadow is:
+ * shadow_addr = shadow_base + (addr >> 3). We retain the 8:1 data:shadow
+ * ratio of existing ASAN implementations as a compromise between tracking
+ * granularity and space usage/scan overhead.
+ */
+
+#ifdef BPF_ARENA_ASAN
+
+#pragma clang attribute push(__attribute__((no_sanitize("address"))), \
+			     apply_to = function)
+
+#define SHADOW_ALL_ZEROES ((u64)-1)
+
+/*
+ * Canary variable for ASAN violations. Set to the offending address.
+ */
+volatile u64 asan_violated = 0;
+
+/*
+ * Shadow map occupancy map.
+ */
+volatile u64 __asan_shadow_memory_dynamic_address;
+
+volatile u32 asan_reported = false;
+volatile bool asan_inited = false;
+
+/*
+ * Set during program load.
+ */
+volatile bool asan_report_once = false;
+volatile bool asan_emit_stack = false;
+
+/*
+ * BPF does not currently support the memset/memcpy/memcmp intrinsics.
+ * For large sequential copies, or assignments of large data structures,
+ * the frontend will generate an instrisic that causes the BPF backend
+ * to exit due to a missing implementation. Provide a simple implementation
+ * just for memset to use it for poisoning/unpoisoning the map.
+ */
+__weak int asan_memset(s8a __arg_arena *dst, s8 val, size_t size)
+{
+	size_t i;
+
+	for (i = zero; i < size && can_loop; i++)
+		dst[i] = val;
+
+	return 0;
+}
+
+/* Validate a 1-byte access, always within a single byte. */
+static __always_inline bool memory_is_poisoned_1(s8a *addr)
+{
+	s8 shadow_value = *(s8a *)mem_to_shadow(addr);
+
+	/* Byte is 0, access is valid. */
+	if (likely(!shadow_value))
+		return false;
+
+	/*
+	 * Byte is non-zero. Access is valid if granule offset in [0, shadow_value),
+	 * so the memory is poisoned if shadow_value is negative or smaller than
+	 * the granule's value.
+	 */
+
+	return ASAN_GRANULE(addr) >= shadow_value;
+}
+
+/* Validate a 2- 4-, 8-byte access, shadow spans up to 2 bytes. */
+static __always_inline bool memory_is_poisoned_2_4_8(s8a *addr, u64 size)
+{
+	u64 end = (u64)addr + size - 1;
+
+	/*
+	 * Region fully within a single byte (addition didn't
+	 * overflow above ASAN_GRANULE).
+	 */
+	if (likely(ASAN_GRANULE(end) >= size - 1))
+		return memory_is_poisoned_1((s8a *)end);
+
+	/*
+	 * Otherwise first byte must be fully unpoisoned, and second byte
+	 * must be unpoisoned up to the end of the accessed region.
+	 */
+
+	return *(s8a *)mem_to_shadow(addr) || memory_is_poisoned_1((s8a *)end);
+}
+
+__weak bool asan_shadow_set(void __arena __arg_arena *addr)
+{
+	return memory_is_poisoned_1(addr);
+}
+
+static __always_inline u64 first_nonzero_byte(u64 addr, size_t size)
+{
+	while (size && can_loop) {
+		if (unlikely(*(s8a *)addr))
+			return addr;
+		addr += 1;
+		size -= 1;
+	}
+
+	return SHADOW_ALL_ZEROES;
+}
+
+static __always_inline bool memory_is_poisoned_n(s8a *addr, u64 size)
+{
+	u64 ret;
+	u64 start;
+	u64 end;
+
+	/* Size of [start, end] is end - start + 1. */
+	start = (u64)mem_to_shadow(addr);
+	end = (u64)mem_to_shadow(addr + size - 1);
+
+	ret = first_nonzero_byte(start, (end - start) + 1);
+	if (likely(ret == SHADOW_ALL_ZEROES))
+		return false;
+
+	return unlikely(ret != end || ASAN_GRANULE(addr + size - 1) >= *(s8a *)end);
+}
+
+__weak int asan_report(s8a __arg_arena *addr, size_t sz, u32 flags)
+{
+	u32 reported = __sync_val_compare_and_swap(&asan_reported, false, true);
+
+	/* Only report the first ASAN violation. */
+	if (reported && asan_report_once)
+		return 0;
+
+	asan_violated = (u64)addr;
+
+	if (asan_emit_stack) {
+		arena_stderr("Memory violation for address %p (0x%lx) for %s of size %ld",
+				addr, (u64)addr,
+				(flags & ASAN_WRITE) ? "write" : "read",
+				sz);
+		bpf_stream_print_stack(BPF_STDERR);
+	}
+
+	return 0;
+}
+
+static __always_inline bool check_asan_args(s8a *addr, size_t size,
+					    bool *result)
+{
+	bool valid = true;
+
+	/* Size 0 accesses are valid even if the address is invalid. */
+	if (unlikely(size == 0))
+		goto confirmed_valid;
+
+	/*
+	 * Wraparound is possible for values close to the the edge of the
+	 * 4GiB boundary of the arena (last valid address is 1UL << 32 - 1).
+	 *
+	 *
+	 * The wraparound detection below works for small sizes.check_asan_args is
+	 * always called from the builtin ASAN checks, so 1 <= size <= 64. We do
+	 * not implement storeN/loadN, so size is guaranteed to be in that range.
+	 * Even if we did, sane storeN/loadN intrinsics are not expected to have
+	 * a large enough size that
+	 *
+	 * - addr + size  > MAX_U32
+	 * - (u32)(addr + size) > (u32) addr
+	 *
+	 * which would defeat wraparound detection.
+	 */
+	if (unlikely((u32)(u64)(addr + size) < (u32)(u64)addr))
+		goto confirmed_invalid;
+
+	return false;
+
+confirmed_invalid:
+	valid = false;
+
+	/* FALLTHROUGH */
+confirmed_valid:
+	*result = valid;
+
+	return true;
+}
+
+static __always_inline bool check_region_inline(void *ptr, size_t size,
+						u32 flags)
+{
+	s8a *addr = (s8a *)(u64)ptr;
+	bool is_poisoned, is_valid;
+
+	if (check_asan_args(addr, size, &is_valid)) {
+		if (!is_valid)
+			asan_report(addr, size, flags);
+		return is_valid;
+	}
+
+	switch (size) {
+	case 1:
+		is_poisoned = memory_is_poisoned_1(addr);
+		break;
+	case 2:
+	case 4:
+	case 8:
+		is_poisoned = memory_is_poisoned_2_4_8(addr, size);
+		break;
+	default:
+		is_poisoned = memory_is_poisoned_n(addr, size);
+	}
+
+	if (is_poisoned) {
+		asan_report(addr, size, flags);
+		return false;
+	}
+
+	return true;
+}
+
+/*
+ * __alias is not supported for BPF so define *__noabort() variants as wrappers.
+ */
+#define DEFINE_ASAN_LOAD_STORE(size)                                  \
+	__hidden void __asan_store##size(void *addr)                  \
+	{                                                             \
+		check_region_inline(addr, size, ASAN_WRITE);          \
+	}                                                             \
+	__hidden void __asan_store##size##_noabort(void *addr)        \
+	{                                                             \
+		check_region_inline(addr, size, ASAN_WRITE);          \
+	}                                                             \
+	__hidden void __asan_load##size(void *addr)                   \
+	{                                                             \
+		check_region_inline(addr, size, ASAN_READ);           \
+	}                                                             \
+	__hidden void __asan_load##size##_noabort(void *addr)         \
+	{                                                             \
+		check_region_inline(addr, size, ASAN_READ);           \
+	}                                                             \
+	__hidden void __asan_report_store##size(void *addr)           \
+	{                                                             \
+		asan_report((s8a *)addr, size, ASAN_WRITE);           \
+	}                                                             \
+	__hidden void __asan_report_store##size##_noabort(void *addr) \
+	{                                                             \
+		asan_report((s8a *)addr, size, ASAN_WRITE);           \
+	}                                                             \
+	__hidden void __asan_report_load##size(void *addr)            \
+	{                                                             \
+		asan_report((s8a *)addr, size, ASAN_READ);            \
+	}                                                             \
+	__hidden void __asan_report_load##size##_noabort(void *addr)  \
+	{                                                             \
+		asan_report((s8a *)addr, size, ASAN_READ);            \
+	}
+
+DEFINE_ASAN_LOAD_STORE(1);
+DEFINE_ASAN_LOAD_STORE(2);
+DEFINE_ASAN_LOAD_STORE(4);
+DEFINE_ASAN_LOAD_STORE(8);
+
+void __asan_storeN(void *addr, ssize_t size)
+{
+	check_region_inline(addr, size, ASAN_WRITE);
+}
+
+void __asan_loadN(void *addr, ssize_t size)
+{
+	check_region_inline(addr, size, ASAN_READ);
+}
+
+/*
+ * We currently do not sanitize globals.
+ */
+void __asan_register_globals(void *globals, size_t n)
+{
+}
+
+void __asan_unregister_globals(void *globals, size_t n)
+{
+}
+
+/*
+ * We do not currently have memcpy/memmove/memset intrinsics
+ * in LLVM. Do not implement sanitization.
+ */
+void *__asan_memcpy(void *d, const void *s, size_t n)
+{
+	arena_stderr("ASAN: Unexpected %s call", __func__);
+	return NULL;
+}
+
+void *__asan_memmove(void *d, const void *s, size_t n)
+{
+	arena_stderr("ASAN: Unexpected %s call", __func__);
+	return NULL;
+}
+
+void *__asan_memset(void *p, int c, size_t n)
+{
+	arena_stderr("ASAN: Unexpected %s call", __func__);
+	return NULL;
+}
+
+/*
+ * Poisoning code, used when we add more freed memory to the allocator by:
+ * 	a) pulling memory from the arena segment using bpf_arena_alloc_pages()
+ * 	b) freeing memory from application code
+ */
+__hidden __noasan int asan_poison(void __arena *addr, s8 val, size_t size)
+{
+	s8a *shadow;
+	size_t len;
+
+	/*
+	 * Poisoning from a non-granule address makes no sense: We can only allocate
+	 * memory to the application that has a granule-aligned starting address,
+	 * and bpf_arena_alloc_pages returns page-aligned memory. A non-aligned
+	 * addr then implies we're freeing a different address than the one we
+	 * allocated.
+	 */
+	if (unlikely((u64)addr & ASAN_GRANULE_MASK))
+		return -EINVAL;
+
+	/*
+	 * We cannot free an unaligned region because it'd be possible that we
+	 * cannot describe the resulting poisoning state of the granule in
+	 * the ASAN encoding.
+	 *
+	 * Every granule represents a region of memory that looks like the
+	 * following (P for poisoned bytes, C for clear):
+	 *
+	 * <Clear>  <Poisoned>
+	 * [ C C C ... P P ]
+	 *
+	 * The value of the granule's shadow map is the number of clear bytes in
+	 * it. We cannot represent granules with the following state:
+	 *
+	 * [ P P ... C C ... P P ]
+	 *
+	 * That would be possible if we could free unaligned regions, so prevent that.
+	 */
+	if (unlikely(size & ASAN_GRANULE_MASK))
+		return -EINVAL;
+
+	shadow = mem_to_shadow(addr);
+	len = size >> ASAN_SHADOW_SHIFT;
+
+	asan_memset(shadow, val, len);
+
+	return 0;
+}
+
+/*
+ * Unpoisoning code for marking memory as valid during allocation calls.
+ *
+ * Very similar to asan_poison, except we need to round up instead of
+ * down, then partially poison the last granule if necessary.
+ *
+ * Partial poisoning is useful for keeping the padding poisoned. Allocations
+ * are granule-aligned, so we we're reserving granule-aligned sizes for the
+ * allocation. However, we want to still treat accesses to the padding as
+ * invalid. Partial poisoning takes care of that. Freeing and poisoning the
+ * memory is still done in granule-aligned sizes and repoisons the already
+ * poisoned padding.
+ */
+__hidden __noasan int asan_unpoison(void __arena *addr, size_t size)
+{
+	size_t partial = size & ASAN_GRANULE_MASK;
+	s8a *shadow;
+	size_t len;
+
+	/*
+	 * We cannot allocate in the middle of the granule. The ASAN shadow
+	 * map encoding only describes regions of memory where every granule
+	 * follows this format (P for poisoned, C for clear):
+	 *
+	 * <Clear>  <Poisoned>
+	 * [ C C C ... P P ]
+	 *
+	 * This is so we can use a single number in [0, ASAN_SHADOW_SCALE)
+	 * to represent the poison state of the granule.
+	 */
+	if (unlikely((u64)addr & ASAN_GRANULE_MASK))
+		return -EINVAL;
+
+	shadow = mem_to_shadow(addr);
+	len = size >> ASAN_SHADOW_SHIFT;
+
+	asan_memset(shadow, 0, len);
+
+	/*
+	 * If we are allocating a non-granule aligned region, we need to adjust
+	 * the last byte of the shadow map to list how many bytes in the granule
+	 * are unpoisoned. If the region is aligned, then the memset call above
+	 * was enough.
+	 */
+	if (partial)
+		shadow[len] = partial;
+
+	return 0;
+}
+
+/*
+ * Initialize ASAN state when necessary. Triggered from userspace before
+ * allocator startup.
+ */
+SEC("syscall")
+__weak __noasan int asan_init(struct asan_init_args *args)
+{
+	u64 globals_pages = args->arena_globals_pages;
+	u64 all_pages = args->arena_all_pages;
+	u64 shadow_map, shadow_pgoff;
+	u64 shadow_pages;
+
+	if (asan_inited)
+		return 0;
+
+	/*
+	 * Round up the shadow map size to the nearest page.
+	 */
+	shadow_pages = all_pages >> ASAN_SHADOW_SHIFT;
+	if ((all_pages & ((1 << ASAN_SHADOW_SHIFT) - 1)))
+		shadow_pages += 1;
+
+	/*
+	 * Make sure the numbers provided by userspace are sane.
+	 */
+	if (all_pages > (1ULL << 32) / __PAGE_SIZE) {
+		arena_stderr("error: arena size %lx too large", all_pages);
+		return -EINVAL;
+	}
+
+	if (globals_pages > all_pages) {
+		arena_stderr("error: globals %lx do not fit in arena %lx",
+				globals_pages, all_pages);
+		return -EINVAL;
+	}
+
+	if (globals_pages + shadow_pages >= all_pages) {
+		arena_stderr("error: globals %lx do not leave room for shadow map %lx "
+				"(arena pages %lx)",
+				globals_pages, shadow_pages, all_pages);
+		return -EINVAL;
+	}
+
+	shadow_pgoff = all_pages - shadow_pages - globals_pages;
+	__asan_shadow_memory_dynamic_address = shadow_pgoff * __PAGE_SIZE;
+
+	/*
+	 * Allocate the last (1/ASAN_SHADOW_SCALE)th of an arena's pages for the map
+	 * We find the offset and size from the arena map.
+	 *
+	 * The allocated map pages are zeroed out, meaning all memory is marked as valid
+	 * even if it's not allocated already. This is expected: Since the actual memory
+	 * pages are not allocated, accesses to it will trigger page faults and will be
+	 * reported through BPF streams. Any pages allocated through bpf_arena_alloc_pages
+	 * should be poisoned by the allocator right after the call succeeds.
+	 */
+	shadow_map = (u64)bpf_arena_alloc_pages(
+		&arena, (void __arena *)__asan_shadow_memory_dynamic_address,
+		shadow_pages, NUMA_NO_NODE, 0);
+	if (!shadow_map) {
+		arena_stderr("Could not allocate shadow map\n");
+
+		__asan_shadow_memory_dynamic_address = 0;
+
+		return -ENOMEM;
+	}
+
+	asan_inited = true;
+
+	return 0;
+}
+
+#pragma clang attribute pop
+
+#endif /* BPF_ARENA_ASAN */
+
+__weak char _license[] SEC("license") = "GPL";
-- 
2.53.0

[PATCH bpf-next v5 7/9] selftests/bpf: Add ASAN support for libarena selftests

[Emil Tsalapatis]

Expand the arena library selftest infrastructure to support
address sanitization. Add the compiler flags necessary to
compile the library under ASAN when supported.

Signed-off-by: Emil Tsalapatis <emil@etsalapatis.com>
---
 tools/testing/selftests/bpf/.gitignore        |  1 +
 tools/testing/selftests/bpf/Makefile          | 13 ++++-
 tools/testing/selftests/bpf/libarena/Makefile | 31 +++++++++++-
 .../bpf/libarena/include/selftest_helpers.h   | 28 +++++++++++
 .../bpf/libarena/selftests/selftest.c         | 38 ++++++++++++++-
 .../bpf/libarena/selftests/st_asan_common.h   | 47 +++++++++++++++++++
 6 files changed, 153 insertions(+), 5 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/libarena/selftests/st_asan_common.h

diff --git a/tools/testing/selftests/bpf/.gitignore b/tools/testing/selftests/bpf/.gitignore
index 7f1960d6b59e..50fa79b2daac 100644
--- a/tools/testing/selftests/bpf/.gitignore
+++ b/tools/testing/selftests/bpf/.gitignore
@@ -50,3 +50,4 @@ verification_cert.h
 usdt_1
 usdt_2
 libarena/test_libarena
+libarena/test_libarena_asan
diff --git a/tools/testing/selftests/bpf/Makefile b/tools/testing/selftests/bpf/Makefile
index 0d332c991023..f10e865338f6 100644
--- a/tools/testing/selftests/bpf/Makefile
+++ b/tools/testing/selftests/bpf/Makefile
@@ -751,6 +751,13 @@ LIBARENA_MAKE_ARGS = \
 		BPF_TARGET_ENDIAN="$(BPF_TARGET_ENDIAN)" \
 		Q="$(Q)"
 
+# libarena_asan.skel.h currently requires LLVM 22. For now,
+# only integrate the non-ASAN version with test_progs.
+LIBARENA_SKEL := libarena/libarena.skel.h
+
+$(LIBARENA_SKEL): $(INCLUDE_DIR)/vmlinux.h $(BPFOBJ)
+	+$(MAKE) -C libarena libarena.skel.h $(LIBARENA_MAKE_ARGS)
+
 # Define test_progs test runner.
 TRUNNER_TESTS_DIR := prog_tests
 TRUNNER_BPF_PROGS_DIR := progs
@@ -774,7 +781,8 @@ TRUNNER_EXTRA_SOURCES := test_progs.c		\
 			 flow_dissector_load.h	\
 			 ip_check_defrag_frags.h	\
 			 bpftool_helpers.c	\
-			 usdt_1.c usdt_2.c
+			 usdt_1.c usdt_2.c	\
+			 $(LIBARENA_SKEL)
 TRUNNER_LIB_SOURCES := find_bit.c
 TRUNNER_EXTRA_FILES := $(OUTPUT)/urandom_read				\
 		       $(OUTPUT)/liburandom_read.so			\
@@ -946,3 +954,6 @@ endef
 
 test_libarena: $(INCLUDE_DIR)/vmlinux.h $(BPFOBJ)
 	+$(MAKE) -C libarena $@ $(LIBARENA_MAKE_ARGS)
+
+test_libarena_asan: $(INCLUDE_DIR)/vmlinux.h $(BPFOBJ)
+	+$(MAKE) -C libarena $@ $(LIBARENA_MAKE_ARGS)
diff --git a/tools/testing/selftests/bpf/libarena/Makefile b/tools/testing/selftests/bpf/libarena/Makefile
index 151ab6d0509c..c988da56c863 100644
--- a/tools/testing/selftests/bpf/libarena/Makefile
+++ b/tools/testing/selftests/bpf/libarena/Makefile
@@ -8,6 +8,7 @@ LIBARENA=$(abspath .)
 
 LIBARENA_SOURCES = $(wildcard $(LIBARENA)/src/*.bpf.c) $(wildcard $(LIBARENA)/selftests/*.bpf.c)
 LIBARENA_OBJECTS = $(notdir $(LIBARENA_SOURCES:.bpf.c=.bpf.o))
+LIBARENA_OBJECTS_ASAN = $(notdir $(LIBARENA_SOURCES:.bpf.c=_asan.bpf.o))
 
 INCLUDES = -I$(LIBARENA)/include -I$(LIBARENA)/..
 ifneq ($(INCLUDE_DIR),)
@@ -17,6 +18,13 @@ ifneq ($(LIBBPF_INCLUDE),)
 INCLUDES += -I$(LIBBPF_INCLUDE)
 endif
 
+ASAN_FLAGS = -fsanitize=kernel-address -fno-stack-protector -fno-builtin
+ASAN_FLAGS += -mllvm -asan-instrument-address-spaces=1 -mllvm -asan-shadow-addr-space=1
+ASAN_FLAGS += -mllvm -asan-use-stack-safety=0 -mllvm -asan-stack=0
+ASAN_FLAGS += -mllvm -asan-kernel=1
+ASAN_FLAGS += -mllvm -asan-constructor-kind=none
+ASAN_FLAGS += -mllvm -asan-destructor-kind=none
+
 # ENABLE_ATOMICS_TESTS required because we use arena spinlocks
 override BPF_CFLAGS += -DENABLE_ATOMICS_TESTS
 override BPF_CFLAGS += -O2 -Wno-incompatible-pointer-types-discards-qualifiers
@@ -28,23 +36,42 @@ CFLAGS += $(INCLUDES)
 vpath %.bpf.c $(LIBARENA)/src $(LIBARENA)/selftests
 vpath %.c $(LIBARENA)/src $(LIBARENA)/selftests
 
-all: test_libarena
+all: test_libarena test_libarena_asan
+
+test_libarena_asan: selftest.c $(BPFOBJ) libarena_asan.skel.h
+	$(call msg,BINARY,libarena,$@)
+	$(Q)$(CLANG) $(CFLAGS) -DBPF_ARENA_ASAN $< $(BPFOBJ) $(LDLIBS) -o $@
 
 test_libarena: selftest.c $(BPFOBJ) libarena.skel.h
 	$(call msg,BINARY,libarena,$@)
 	$(Q)$(CLANG) $(CFLAGS) $< $(BPFOBJ) $(LDLIBS) -o $@
 
+skeletons: libarena.skel.h libarena_asan.skel.h
+.PHONY: skeletons
+
+libarena_asan.skel.h: main_asan.bpf.o
+	$(call msg,GEN-SKEL,libarena,$@)
+	$(Q)$(BPFTOOL) gen skeleton $< name "libarena_asan" > $@
+
 libarena.skel.h: main.bpf.o
 	$(call msg,GEN-SKEL,libarena,$@)
 	$(Q)$(BPFTOOL) gen skeleton $< name "libarena" > $@
 
+main_asan.bpf.o: $(LIBARENA_OBJECTS_ASAN)
+	$(call msg,GEN-OBJ,libarena,$@)
+	$(Q)$(BPFTOOL) gen object $@ $^
+
 main.bpf.o: $(LIBARENA_OBJECTS)
 	$(call msg,GEN-OBJ,libarena,$@)
 	$(Q)$(BPFTOOL) gen object $@ $^
 
+%_asan.bpf.o: %.bpf.c
+	$(call msg,CLNG-BPF,libarena,$@)
+	$(Q)$(CLANG) $(BPF_CFLAGS) $(ASAN_FLAGS) -DBPF_ARENA_ASAN $(BPF_TARGET_ENDIAN) -c $< -o $@
+
 %.bpf.o: %.bpf.c
 	$(call msg,CLNG-BPF,libarena,$@)
 	$(Q)$(CLANG) $(BPF_CFLAGS) $(BPF_TARGET_ENDIAN) -c $< -o $@
 
 clean:
-	$(Q)rm -f *.skel.h *.bpf.o test_libarena
+	$(Q)rm -f *.skel.h *.bpf.o *.linked*.o test_libarena test_libarena_asan
diff --git a/tools/testing/selftests/bpf/libarena/include/selftest_helpers.h b/tools/testing/selftests/bpf/libarena/include/selftest_helpers.h
index 78db8bc0fc67..36e4f36ae8e6 100644
--- a/tools/testing/selftests/bpf/libarena/include/selftest_helpers.h
+++ b/tools/testing/selftests/bpf/libarena/include/selftest_helpers.h
@@ -84,3 +84,31 @@ static inline int libarena_get_globals_pages(int arena_get_base_fd,
 	free(vec);
 	return 0;
 }
+
+static inline int libarena_asan_init(int arena_get_base_fd,
+				     int asan_init_fd,
+				     size_t arena_all_pages)
+{
+	LIBBPF_OPTS(bpf_test_run_opts, opts);
+	struct asan_init_args args;
+	u64 globals_pages;
+	int ret;
+
+	ret = libarena_get_globals_pages(arena_get_base_fd,
+					 arena_all_pages, &globals_pages);
+	if (ret)
+		return ret;
+
+	args = (struct asan_init_args){
+		.arena_all_pages = arena_all_pages,
+		.arena_globals_pages = globals_pages,
+	};
+
+	opts.ctx_in = &args;
+	opts.ctx_size_in = sizeof(args);
+
+	ret = bpf_prog_test_run_opts(asan_init_fd, &opts);
+	if (ret)
+		return ret;
+	return opts.retval;
+}
diff --git a/tools/testing/selftests/bpf/libarena/selftests/selftest.c b/tools/testing/selftests/bpf/libarena/selftests/selftest.c
index da70648968d4..45ab1ac0ef78 100644
--- a/tools/testing/selftests/bpf/libarena/selftests/selftest.c
+++ b/tools/testing/selftests/bpf/libarena/selftests/selftest.c
@@ -16,8 +16,18 @@
 #include <sys/sysinfo.h>
 
 #include <common.h>
+#include <asan.h>
 #include <selftest_helpers.h>
 
+#ifdef BPF_ARENA_ASAN
+#include "../libarena_asan.skel.h"
+typedef struct libarena_asan selftest;
+#define selftest__open libarena_asan__open
+#define selftest__open_and_load libarena_asan__open_and_load
+#define selftest__load libarena_asan__load
+#define selftest__attach libarena_asan__attach
+#define selftest__destroy libarena_asan__destroy
+#else
 #include "../libarena.skel.h"
 typedef struct libarena selftest;
 #define selftest__open libarena__open
@@ -25,6 +35,7 @@ typedef struct libarena selftest;
 #define selftest__load libarena__load
 #define selftest__attach libarena__attach
 #define selftest__destroy libarena__destroy
+#endif
 
 static bool verbose;
 static int testno = 1;
@@ -67,6 +78,26 @@ static int libbpf_print_fn(enum libbpf_print_level level,
 	return vfprintf(stderr, format, args);
 }
 
+static int init_arena(selftest *skel)
+{
+	int ret;
+
+	ret = libarena_run_prog(bpf_program__fd(skel->progs.arena_alloc_reserve));
+	if (ret)
+		return ret;
+
+#ifdef BPF_ARENA_ASAN
+	ret = libarena_asan_init(
+		bpf_program__fd(skel->progs.arena_get_base),
+		bpf_program__fd(skel->progs.asan_init),
+		(1ULL << 32) / sysconf(_SC_PAGESIZE));
+	if (ret)
+		return ret;
+#endif
+
+	return 0;
+}
+
 static int run_test(selftest *skel, struct bpf_program *prog)
 {
 	int prog_fd;
@@ -82,10 +113,13 @@ static void
 banner(const char *progpath)
 {
 	char *name = basename(progpath);
+	bool is_asan;
 
+	/* Check if our BPF programs are ASAN-capable using strstr on the prog name. */
 	printf("%s\n", name);
+	is_asan = strstr(name, "_asan");
 
-	printf("=== %s ===\n", "libarena selftests");
+	printf("=== %s %s===\n", "libarena selftests", is_asan ? "(asan) " : "");
 }
 
 int main(int argc, char *argv[])
@@ -116,7 +150,7 @@ int main(int argc, char *argv[])
 		return 1;
 	}
 
-	ret = libarena_run_prog(bpf_program__fd(skel->progs.arena_alloc_reserve));
+	ret = init_arena(skel);
 	if (ret) {
 		fprintf(stderr, "Failed to initialize arena: %d\n", ret);
 		selftest__destroy(skel);
diff --git a/tools/testing/selftests/bpf/libarena/selftests/st_asan_common.h b/tools/testing/selftests/bpf/libarena/selftests/st_asan_common.h
new file mode 100644
index 000000000000..346e89dce172
--- /dev/null
+++ b/tools/testing/selftests/bpf/libarena/selftests/st_asan_common.h
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: LGPL-2.1 OR BSD-2-Clause
+/* Copyright (c) 2026 Meta Platforms, Inc. and affiliates. */
+
+#pragma once
+
+#define ST_PAGES 64
+
+static inline void print_asan_map_state(void __arena *addr)
+{
+	arena_stdout("%s:%d ASAN %lx -> (val: %x gran: %x set: [%s])",
+			__func__, __LINE__, addr,
+			*(s8a *)(addr), ASAN_GRANULE(addr),
+			asan_shadow_set(addr) ? "yes" : "no");
+}
+
+/*
+ * Emit an error and force the current function to exit if the ASAN
+ * violation state is unexpected. Reset the violation state after.
+ */
+static inline int asan_validate_addr(bool cond, void __arena *addr)
+{
+	if ((asan_violated != 0) == cond) {
+		asan_violated = 0;
+		return 0;
+	}
+
+	arena_stdout("%s:%d ASAN asan_violated %lx", __func__, __LINE__,
+			(u64)asan_violated);
+	print_asan_map_state(addr);
+	return -EINVAL;
+}
+
+static inline int asan_validate(void)
+{
+	if (!asan_violated)
+		return 0;
+
+	arena_stdout("%s:%d Found ASAN violation at %lx", __func__, __LINE__,
+			asan_violated);
+
+	return -EINVAL;
+}
+
+struct blob {
+	volatile u8 mem[59];
+	u8 oob;
+};
-- 
2.53.0

[PATCH bpf-next v5 8/9] selftests/bpf: Add buddy allocator for libarena

[Emil Tsalapatis]

Add a byte-oriented buddy allocator for libarena. The buddy
allocator provides an alloc/free interface for small arena allocations
ranging from 16 bytes to 512 KiB. Lower allocations values are rounded
up to 16 bytes. The buddy allocator does not handle larger allocations
that can instead use the existing bpf_arena_{alloc, free}_pages() kfunc.

Signed-off-by: Emil Tsalapatis <emil@etsalapatis.com>
---
 .../selftests/bpf/libarena/include/buddy.h    |  92 ++
 .../selftests/bpf/libarena/src/buddy.bpf.c    | 879 ++++++++++++++++++
 .../selftests/bpf/libarena/src/common.bpf.c   |  12 +
 3 files changed, 983 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/libarena/include/buddy.h
 create mode 100644 tools/testing/selftests/bpf/libarena/src/buddy.bpf.c

diff --git a/tools/testing/selftests/bpf/libarena/include/buddy.h b/tools/testing/selftests/bpf/libarena/include/buddy.h
new file mode 100644
index 000000000000..2302d36c88c2
--- /dev/null
+++ b/tools/testing/selftests/bpf/libarena/include/buddy.h
@@ -0,0 +1,92 @@
+// SPDX-License-Identifier: LGPL-2.1 OR BSD-2-Clause
+/* Copyright (c) 2026 Meta Platforms, Inc. and affiliates. */
+#pragma once
+
+struct buddy_chunk;
+typedef struct buddy_chunk __arena buddy_chunk_t;
+
+struct buddy_header;
+typedef struct buddy_header __arena buddy_header_t;
+
+enum buddy_consts {
+	/*
+	 * Minimum allocation is 1 << BUDDY_MIN_ALLOC_SHIFT.
+	 * Larger sizes increase internal fragmentation, but smaller
+	 * sizes increase the space overhead of the block metadata.
+	 */
+	BUDDY_MIN_ALLOC_SHIFT	= 4,
+	BUDDY_MIN_ALLOC_BYTES	= 1 << BUDDY_MIN_ALLOC_SHIFT,
+
+	/*
+	 * How many orders the buddy allocator can serve. Minimum block
+	 * size is 1 << BUDDY_MIN_ALLOC_SHIFT, maximum block size is
+	 * 1 << (BUDDY_MIN_ALLOC_SHIFT + BUDDY_CHUNK_NUM_ORDERS - 1):
+	 * Each block has size 1 << BUDDY_MIN_ALLOC_SHIFT, and the
+	 * allocation orders are in [0, BUDDY_CHUNK_NUM_ORDERS).
+	 * We keep two blocks of the maximum size to retain the
+	 * property in the code that all blocks have a buddy.
+	 * values increase the maximum allocation size, but
+	 * also the size of the metadata for each block.
+	 */
+	BUDDY_CHUNK_NUM_ORDERS	= 1 << 4,
+	BUDDY_CHUNK_BYTES	= BUDDY_MIN_ALLOC_BYTES << (BUDDY_CHUNK_NUM_ORDERS),
+
+	/* Offset off the buddy header within a free block, see buddy.bpf.c for details */
+	BUDDY_HEADER_OFF	= 8,
+
+	/* The maximum number of blocks a chunk may have to track. */
+	BUDDY_CHUNK_ITEMS	= 1 << (BUDDY_CHUNK_NUM_ORDERS),
+	BUDDY_CHUNK_OFFSET_MASK	= BUDDY_CHUNK_BYTES - 1,
+
+	/*
+	 * Alignment for chunk allocations based on bpf_arena_alloc_pages.
+	 * The arena allocation kfunc does not ahave an alignment argument,
+	 * but that is required all the block calculations in the chunk to
+	 * work.
+	 */
+	BUDDY_VADDR_OFFSET	= BUDDY_CHUNK_BYTES,
+
+	/* Total arena virtual address space the allocator can consume. */
+	BUDDY_VADDR_SIZE	= BUDDY_CHUNK_BYTES << 10
+};
+
+struct buddy_header {
+	u32 prev_index;	/* "Pointer" to the previous available allocation of the same size. */
+	u32 next_index; /* Same for the next allocation. */
+};
+
+/*
+ * We bring memory into the allocator 1 MiB at a time.
+ */
+struct buddy_chunk {
+	/* The order of the current allocation for a item. 4 bits per order. */
+	u8		orders[BUDDY_CHUNK_ITEMS / 2];
+	/*
+	 * Bit to denote whether chunk is allocated. Size of the allocated/free
+	 * chunk found from the orders array.
+	 */
+	u8		allocated[BUDDY_CHUNK_ITEMS / 8];
+	/* Freelists for O(1) allocation. */
+	u64		freelists[BUDDY_CHUNK_NUM_ORDERS];
+	buddy_chunk_t	*next;
+};
+
+struct buddy {
+	buddy_chunk_t *first_chunk;		/* Pointer to the chunk linked list. */
+	arena_spinlock_t lock;			/* Allocator lock */
+	u64 vaddr;				/* Allocation into reserved vaddr */
+};
+
+typedef struct buddy __arena buddy_t;
+
+#ifdef __BPF__
+
+int buddy_init(buddy_t *buddy);
+int buddy_destroy(buddy_t *buddy);
+int buddy_free_internal(buddy_t *buddy, u64 free);
+#define buddy_free(buddy, ptr) buddy_free_internal((buddy), (u64)(ptr))
+u64 buddy_alloc_internal(buddy_t *buddy, size_t size);
+#define buddy_alloc(alloc, size) ((void __arena *)buddy_alloc_internal((alloc), (size)))
+
+
+#endif /* __BPF__  */
diff --git a/tools/testing/selftests/bpf/libarena/src/buddy.bpf.c b/tools/testing/selftests/bpf/libarena/src/buddy.bpf.c
new file mode 100644
index 000000000000..ea3d26a0d88b
--- /dev/null
+++ b/tools/testing/selftests/bpf/libarena/src/buddy.bpf.c
@@ -0,0 +1,879 @@
+// SPDX-License-Identifier: LGPL-2.1 OR BSD-2-Clause
+/* Copyright (c) 2026 Meta Platforms, Inc. and affiliates. */
+
+#include <common.h>
+#include <asan.h>
+#include <buddy.h>
+
+/*
+ * Buddy allocator arena-based implementation.
+ *
+ * Memory is organized into chunks. These chunks
+ * cannot be coalesced or split. Allocating
+ * chunks allocates their memory eagerly.
+ *
+ * Internally, each chunk is organized into blocks.
+ * Blocks _can_ be coalesced/split, but only inside
+ * the chunk. Each block can be allocated or
+ * unallocated. If allocated, the entire block holds
+ * user data. If unallocated, the block is mostly
+ * invalid memory, with the exception of a header
+ * used for freelist tracking.
+ *
+ * The header is placed at an offset inside the block
+ * to prevent off-by-one errors from the previous block
+ * from trivially overwriting the header. Such an error
+ * is also not catchable by ASAN, since the header remains
+ * valid memory even after the block is freed. It is still
+ * theoretically possible for the header to be corrupted
+ * without being caught by ASAN, but harder.
+ *
+ * Since the allocator needs to track order information for
+ * both allocated and free blocks, and allocated blocks cannot
+ * store a header, the allocator also stores per-chunk order
+ * information in a reserved region at the beginning of the
+ * chunk. The header includes a bitmap with the order of blocks
+ * and their allocation state. It also includes the freelist
+ * heads for the allocation itself.
+ */
+
+
+enum {
+	BUDDY_POISONED = (s8)0xef,
+
+	/* Number of pages to be allocated per chunk. */
+	BUDDY_CHUNK_PAGES	= BUDDY_CHUNK_BYTES / __PAGE_SIZE
+};
+
+static inline int buddy_lock(buddy_t *buddy)
+{
+	return arena_spin_lock(&buddy->lock);
+}
+
+static inline void buddy_unlock(buddy_t *buddy)
+{
+	arena_spin_unlock(&buddy->lock);
+}
+
+/*
+ * Reserve part of the arena address space for the allocator. We use
+ * this to get aligned addresses for the chunks, since the arena
+ * page alloc kfuncs do not support aligning to a boundary (in this
+ * case 1 MiB, see buddy.h on how this is derived).
+ */
+static int buddy_reserve_arena_vaddr(buddy_t *buddy)
+{
+	buddy->vaddr = 0;
+
+	return bpf_arena_reserve_pages(&arena,
+				       (void __arena *)BUDDY_VADDR_OFFSET,
+				       BUDDY_VADDR_SIZE / __PAGE_SIZE);
+}
+
+/*
+ * Free up any unused address space. Used only during teardown.
+ */
+static void buddy_unreserve_arena_vaddr(buddy_t *buddy)
+{
+	bpf_arena_free_pages(
+		&arena, (void __arena *)(BUDDY_VADDR_OFFSET + buddy->vaddr),
+		(BUDDY_VADDR_SIZE - buddy->vaddr) / __PAGE_SIZE);
+
+	buddy->vaddr = 0;
+}
+
+/*
+ * Carve out part of the reserved address space and hand it over
+ * to the buddy allocator.
+ *
+ * We are assuming the buddy allocator is the only allocator in the
+ * system, so there is no race between this function reserving a
+ * page range and some other allocator actually making the BPF call
+ * to really create and reserve it.
+ *
+ * However, bump allocation must still be atomic because this function
+ * is called without the buddy lock from multiple threads concurrently.
+ */
+__weak int buddy_alloc_arena_vaddr(buddy_t __arg_arena *buddy, u64 *vaddrp)
+{
+	u64 vaddr, old, new;
+
+	if (!buddy || !vaddrp)
+		return -EINVAL;
+
+	do {
+		vaddr = buddy->vaddr;
+		new = vaddr + BUDDY_CHUNK_BYTES;
+
+		if (new > BUDDY_VADDR_SIZE)
+			return -EINVAL;
+
+		old = __sync_val_compare_and_swap(&buddy->vaddr, vaddr, new);
+	} while (old != vaddr && can_loop);
+
+	if (old != vaddr)
+		return -EINVAL;
+
+	*vaddrp = BUDDY_VADDR_OFFSET + vaddr;
+
+	return 0;
+}
+
+static u64 arena_next_pow2(__u64 n)
+{
+	n--;
+	n |= n >> 1;
+	n |= n >> 2;
+	n |= n >> 4;
+	n |= n >> 8;
+	n |= n >> 16;
+	n |= n >> 32;
+	n++;
+
+	return n;
+}
+
+__weak
+int idx_set_allocated(buddy_chunk_t __arg_arena *chunk, u64 idx, bool allocated)
+{
+	if (unlikely(idx >= BUDDY_CHUNK_ITEMS)) {
+		arena_stderr("setting state of invalid idx (%d, max %d)\n", idx,
+			     BUDDY_CHUNK_ITEMS);
+		return -EINVAL;
+	}
+
+	if (allocated)
+		chunk->allocated[idx / 8] |= 1 << (idx % 8);
+	else
+		chunk->allocated[idx / 8] &= ~(1 << (idx % 8));
+
+	return 0;
+}
+
+static int idx_is_allocated(buddy_chunk_t *chunk, u64 idx, bool *allocated)
+{
+	if (unlikely(idx >= BUDDY_CHUNK_ITEMS)) {
+		arena_stderr("getting state of invalid idx (%llu, max %d)\n", idx,
+			     BUDDY_CHUNK_ITEMS);
+		return -EINVAL;
+	}
+
+	*allocated = chunk->allocated[idx / 8] & (1 << (idx % 8));
+	return 0;
+}
+
+__weak
+int idx_set_order(buddy_chunk_t __arg_arena *chunk, u64 idx, u8 order)
+{
+	u8 prev_order;
+
+	if (unlikely(order >= BUDDY_CHUNK_NUM_ORDERS)) {
+		arena_stderr("setting invalid order %u\n", order);
+		return -EINVAL;
+	}
+
+	if (unlikely(idx >= BUDDY_CHUNK_ITEMS)) {
+		arena_stderr("setting order of invalid idx (%d, max %d)\n", idx,
+			     BUDDY_CHUNK_ITEMS);
+		return -EINVAL;
+	}
+
+	/*
+	 * We store two order instances per byte, one per nibble.
+	 * Retain the existing nibble.
+	 */
+	prev_order = chunk->orders[idx / 2];
+	if (idx & 0x1) {
+		order &= 0xf;
+		order |= (prev_order & 0xf0);
+	} else {
+		order <<= 4;
+		order |= (prev_order & 0xf);
+	}
+
+	chunk->orders[idx / 2] = order;
+
+	return 0;
+}
+
+static u8 idx_get_order(buddy_chunk_t *chunk, u64 idx)
+{
+	u8 result;
+
+	_Static_assert(BUDDY_CHUNK_NUM_ORDERS <= 16,
+		       "order must fit in 4 bits");
+
+	if (unlikely(idx >= BUDDY_CHUNK_ITEMS)) {
+		arena_stderr("getting order of invalid idx %u\n", idx);
+		return BUDDY_CHUNK_NUM_ORDERS;
+	}
+
+	result = chunk->orders[idx / 2];
+
+	return (idx & 0x1) ? (result & 0xf) : (result >> 4);
+}
+
+static void __arena *idx_to_addr(buddy_chunk_t *chunk, size_t idx)
+{
+	u64 address;
+
+	if (unlikely(idx >= BUDDY_CHUNK_ITEMS)) {
+		arena_stderr("translating invalid idx %u\n", idx);
+		return NULL;
+	}
+
+	/*
+	 * The data blocks start in the chunk after the metadata block.
+	 * We find the actual address by indexing into the region at an
+	 * BUDDY_MIN_ALLOC_BYTES granularity, the minimum allowed.
+	 * The index number already accounts for the fact that the first
+	 * blocks in the chunk are occupied by the metadata, so we do
+	 * not need to offset it.
+	 */
+
+	address = (u64)chunk + (idx * BUDDY_MIN_ALLOC_BYTES);
+
+	return (void __arena *)address;
+}
+
+static buddy_header_t *idx_to_header(buddy_chunk_t *chunk, size_t idx)
+{
+	bool allocated;
+	u64 address;
+
+	if (unlikely(idx_is_allocated(chunk, idx, &allocated))) {
+		arena_stderr("accessing invalid idx 0x%lx\n", idx);
+		return NULL;
+	}
+
+	if (unlikely(allocated)) {
+		arena_stderr("accessing allocated idx 0x%lx as header\n", idx);
+		return NULL;
+	}
+
+	address = (u64)idx_to_addr(chunk, idx);
+
+	/*
+	 * Offset the header within the block. This avoids accidental overwrites
+	 * to the header because of off-by-one errors when using adjacent blocks.
+	 *
+	 * The offset has been chosen as a compromise between ASAN effectiveness
+	 * and allocator granularity:
+	 * 1) ASAN dictates valid data runs are 8-byte aligned.
+	 * 2) We want to keep a low minimum allocation size (currently 16).
+	 *
+	 * As a result, we have only two possible positions for the header: Bytes
+	 * 0 and 8. Keeping the header in byte 0 means off-by-ones from the previous
+	 * block touch the header, and, since the header must be accessible, ASAN
+	 * will not trigger. Keeping the header on byte 8 means off-by-one errors from
+	 * the previous block are caught by ASAN. Negative offsets are rarer, so
+	 * while accesses into the block from the next block are possible, they are
+	 * less probable.
+	 */
+
+	return (buddy_header_t *)(address + BUDDY_HEADER_OFF);
+}
+
+static void header_add_freelist(buddy_chunk_t *chunk, buddy_header_t *header,
+		u64 idx, u8 order)
+{
+	buddy_header_t *tmp_header;
+
+	idx_set_order(chunk, idx, order);
+
+	header->next_index = chunk->freelists[order];
+	header->prev_index = BUDDY_CHUNK_ITEMS;
+
+	if (header->next_index != BUDDY_CHUNK_ITEMS) {
+		tmp_header = idx_to_header(chunk, header->next_index);
+		tmp_header->prev_index = idx;
+	}
+
+	chunk->freelists[order] = idx;
+}
+
+static void header_remove_freelist(buddy_chunk_t  *chunk,
+				   buddy_header_t *header, u8 order)
+{
+	buddy_header_t *tmp_header;
+
+	if (header->prev_index != BUDDY_CHUNK_ITEMS) {
+		tmp_header = idx_to_header(chunk, header->prev_index);
+		tmp_header->next_index = header->next_index;
+	}
+
+	if (header->next_index != BUDDY_CHUNK_ITEMS) {
+		tmp_header = idx_to_header(chunk, header->next_index);
+		tmp_header->prev_index = header->prev_index;
+	}
+
+	/* Pop off the list head if necessary. */
+	if (idx_to_header(chunk, chunk->freelists[order]) == header)
+		chunk->freelists[order] = header->next_index;
+
+	header->prev_index = BUDDY_CHUNK_ITEMS;
+	header->next_index = BUDDY_CHUNK_ITEMS;
+}
+
+static u64 size_to_order(size_t size)
+{
+	u64 order;
+
+	/*
+	 * Legal sizes are [1, 4GiB] (the biggest possible arena).
+	 * Of course, sizes close to GiB are practically impossible
+	 * to fulfill and allocation will fail, but that's taken care
+	 * of by the caller.
+	 */
+
+	if (unlikely(size == 0 || size > (1UL << 32))) {
+		arena_stderr("illegal size request %lu\n", size);
+		return 64;
+	}
+	/*
+	 * To find the order of the allocation we find the first power of two
+	 * >= the requested size, take the log2, then adjust it for the minimum
+	 * allocation size by removing the minimum shift from it. Requests
+	 * smaller than the minimum allocation size are rounded up.
+	 */
+	order = arena_fls(arena_next_pow2(size));
+	if (order < BUDDY_MIN_ALLOC_SHIFT)
+		return 0;
+
+	return order - BUDDY_MIN_ALLOC_SHIFT;
+}
+
+__weak
+int add_leftovers_to_freelist(buddy_chunk_t __arg_arena *chunk, u32 cur_idx,
+		u64 min_order, u64 max_order)
+{
+	buddy_header_t *header;
+	u64 ord;
+	u32 idx;
+
+	for (ord = min_order; ord < max_order && can_loop; ord++) {
+		/* Mark the buddy as free and add it to the freelists. */
+		idx = cur_idx + (1 << ord);
+
+		header = idx_to_header(chunk, idx);
+		if (unlikely(!header)) {
+			arena_stderr("idx %u has no header", idx);
+			return -EINVAL;
+		}
+
+		asan_unpoison(header, sizeof(*header));
+
+		idx_set_allocated(chunk, idx, false);
+		header_add_freelist(chunk, header, idx, ord);
+	}
+
+	return 0;
+}
+
+static buddy_chunk_t *buddy_chunk_get(buddy_t *buddy)
+{
+	u64 order, ord, min_order, max_order;
+	buddy_chunk_t  *chunk;
+	size_t left;
+	int power2;
+	u64 vaddr;
+	u32 idx;
+	int ret;
+
+	/*
+	 * Step 1:  Allocate a properly aligned chunk, and
+	 * prep it for insertion into the buddy allocator.
+	 * We don't need the allocator lock until step 2.
+	 */
+
+	ret = buddy_alloc_arena_vaddr(buddy, &vaddr);
+	if (ret)
+		return NULL;
+
+	/* Addresses must be aligned to the chunk boundary. */
+	if (vaddr % BUDDY_CHUNK_BYTES)
+		return NULL;
+
+	/* Unreserve the address space. */
+	bpf_arena_free_pages(&arena, (void __arena *)vaddr,
+			     BUDDY_CHUNK_PAGES);
+
+	chunk = bpf_arena_alloc_pages(&arena, (void __arena *)vaddr,
+				      BUDDY_CHUNK_PAGES, NUMA_NO_NODE, 0);
+	if (!chunk) {
+		arena_stderr("[ALLOC FAILED]");
+		return NULL;
+	}
+
+	if (buddy_lock(buddy)) {
+		/*
+		 * We cannot reclaim the vaddr space, but that is ok - this
+		 * operation should always succeed. The error path is to catch
+		 * accidental deadlocks that will cause -ENOMEMs to the program as
+		 * the allocator fails to refill itself, in which case vaddr usage
+		 * is the least of our worries.
+		 */
+		bpf_arena_free_pages(&arena, (void __arena *)vaddr, BUDDY_CHUNK_PAGES);
+		return NULL;
+	}
+
+	asan_poison(chunk, BUDDY_POISONED, BUDDY_CHUNK_PAGES * __PAGE_SIZE);
+
+	/* Unpoison the chunk itself. */
+	asan_unpoison(chunk, sizeof(*chunk));
+
+	/* Mark all freelists as empty. */
+	for (ord = zero; ord < BUDDY_CHUNK_NUM_ORDERS && can_loop; ord++)
+		chunk->freelists[ord] = BUDDY_CHUNK_ITEMS;
+
+	/*
+	 * Initialize the chunk by carving out a page range to hold the metadata
+	 * struct above, then dumping the rest of the pages into the allocator.
+	 */
+
+	_Static_assert(BUDDY_CHUNK_PAGES * __PAGE_SIZE >=
+			       BUDDY_MIN_ALLOC_BYTES *
+				       BUDDY_CHUNK_ITEMS,
+		       "chunk must fit within the allocation");
+
+	/*
+	 * Step 2: Reserve a chunk for the chunk metadata, then breaks
+	 * the rest of the full allocation into the different buckets.
+	 * We allocating the memory by grabbing blocks of progressively
+	 * smaller sizes from the allocator, which are guaranteed to be
+	 * continuous.
+	 *
+	 * This operation also populates the allocator.
+	 *
+	 * Algorithm:
+	 *
+	 * - max_order: The last order allocation we made
+	 * - left: How many bytes are left to allocate
+	 * - cur_index: Current index into the top-level block we are
+	 * allocating from.
+	 *
+	 * Step 3:
+	 * - Find the largest power-of-2 allocation still smaller than left (infimum)
+	 * - Reserve a chunk of that size, along with its buddy
+	 * - For every order from [infimum + 1, last order), carve out a block
+	 *   and put it into the allocator.
+	 *
+	 *  Example: Chunk size 0b1010000 (80 bytes)
+	 *
+	 *  Step 1:
+	 *
+	 *   idx  infimum                             1 << max_order
+	 *   0        64        128                    1 << 20
+	 *   |________|_________|______________________|
+	 *
+	 *   Blocks set aside:
+	 *   	[0, 64)         - Completely allocated
+	 *   	[64, 128)       - Will be further split in the next iteration
+	 *
+	 *   Blocks added to the allocator:
+	 *   	[128, 256)
+	 *   	[256, 512)
+	 *   	...
+	 *   	[1 << 18, 1 << 19)
+	 *   	[1 << 19, 1 << 20)
+	 *
+	 *  Step 2:
+	 *
+	 *   idx  infimum			   idx + 1 << max_order
+	 *   64	      80	96		   	64 + 1 << 6 = 128
+	 *   |________|_________|______________________|
+	 *
+	 *   Blocks set aside:
+	 *   	[64, 80)	- Completely allocated
+	 *
+	 *   Blocks added to the allocator:
+	 *      [80, 96) - left == 0 so the buddy is unused and marked as freed
+	 *   	[96, 128)
+	 */
+	 max_order = BUDDY_CHUNK_NUM_ORDERS;
+	left = sizeof(*chunk);
+	idx = 0;
+	while (left && can_loop) {
+		power2 = arena_fls(left);
+		/*
+		 * Note: The condition below only triggers to catch serious bugs
+		 * early. There is no sane way to undo any block insertions from
+		 * the allocated chunk, so just leak any leftover allocations,
+		 * emit a diagnostic, unlock and exit.
+		 *
+		 */
+		if (unlikely(power2 >= BUDDY_CHUNK_NUM_ORDERS)) {
+			arena_stderr(
+				"buddy chunk metadata require allocation of order %d\n",
+				power2);
+			arena_stderr(
+				"chunk has size of 0x%lx bytes (left %lx bytes)\n",
+				sizeof(*chunk), left);
+			buddy_unlock(buddy);
+
+			return NULL;
+		}
+
+		/* Round up allocations that are too small. */
+
+		left -= (power2 >= BUDDY_MIN_ALLOC_SHIFT) ? 1 << power2 : left;
+		order = (power2 >= BUDDY_MIN_ALLOC_SHIFT) ? power2 - BUDDY_MIN_ALLOC_SHIFT : 0;
+
+		idx_set_allocated(chunk, idx, true);
+
+		/*
+		 * Starting an order above the one we allocated, populate
+		 * the allocator with free blocks. If this is the last
+		 * allocation (left == 0), also mark the buddy as free.
+		 *
+		 * See comment above about error handling: The error path
+		 * is only there as a way to mitigate deeply buggy allocator
+		 * states by emitting a diagnostic in add_leftovers_to_freelist()
+		 * and leaking any memory not added in the freelists.
+		 */
+		min_order = left ? order + 1 : order;
+		if (add_leftovers_to_freelist(chunk, idx, min_order, max_order)) {
+			buddy_unlock(buddy);
+			return NULL;
+		}
+
+		/* Adjust the index. */
+		idx += 1 << order;
+		max_order = order;
+	}
+
+	buddy_unlock(buddy);
+
+	return chunk;
+}
+
+__weak int buddy_init(buddy_t __arg_arena *buddy)
+{
+	buddy_chunk_t *chunk;
+	int ret;
+
+	if (!asan_ready())
+		return -EINVAL;
+
+	/* Reserve enough address space to ensure allocations are aligned. */
+	ret = buddy_reserve_arena_vaddr(buddy);
+	if (ret)
+		return ret;
+
+	_Static_assert(BUDDY_CHUNK_PAGES > 0,
+		       "chunk must use one or more pages");
+
+	chunk = buddy_chunk_get(buddy);
+
+	if (buddy_lock(buddy)) {
+		bpf_arena_free_pages(&arena, chunk, BUDDY_CHUNK_PAGES);
+		return -EINVAL;
+	}
+
+	/* Chunk is already properly unpoisoned if allocated. */
+	if (chunk)
+		chunk->next = buddy->first_chunk;
+
+	/* Put the chunk at the beginning of the list. */
+	buddy->first_chunk = chunk;
+
+	buddy_unlock(buddy);
+
+	return chunk ? 0 : -ENOMEM;
+}
+
+/*
+ * Destroy the allocator. This does not check whether there are any allocations
+ * currently in use, so any pages being accessed will start taking arena faults.
+ * We do not take a lock because we are freeing arena pages, and nobody should
+ * be using the allocator at that point in the execution.
+ */
+__weak int buddy_destroy(buddy_t __arg_arena *buddy)
+{
+	buddy_chunk_t *chunk, *next;
+
+	if (!buddy)
+		return -EINVAL;
+
+	/*
+	 * Traverse all buddy chunks and free them back to the arena
+	 * with the same granularity they were allocated with.
+	 */
+	for (chunk = buddy->first_chunk; chunk && can_loop; chunk = next) {
+		next = chunk->next;
+
+		/* Wholesale poison the entire block. */
+		asan_poison(chunk, BUDDY_POISONED,
+			    BUDDY_CHUNK_PAGES * __PAGE_SIZE);
+		bpf_arena_free_pages(&arena, chunk, BUDDY_CHUNK_PAGES);
+	}
+
+	/* Free up any part of the address space that did not get used. */
+	buddy_unreserve_arena_vaddr(buddy);
+
+	/* Clear all fields. */
+	buddy->first_chunk = NULL;
+
+	return 0;
+}
+
+__weak u64 buddy_chunk_alloc(buddy_chunk_t __arg_arena *chunk, int order_req)
+{
+	buddy_header_t *header, *tmp_header, *next_header;
+	u32 idx, tmpidx, retidx;
+	u64 address;
+	u64 order = 0;
+	u64 i;
+
+	for (order = order_req; order < BUDDY_CHUNK_NUM_ORDERS && can_loop; order++) {
+		if (chunk->freelists[order] != BUDDY_CHUNK_ITEMS)
+			break;
+	}
+
+	if (order >= BUDDY_CHUNK_NUM_ORDERS)
+		return (u64)NULL;
+
+	retidx = chunk->freelists[order];
+	header = idx_to_header(chunk, retidx);
+	chunk->freelists[order] = header->next_index;
+
+	if (header->next_index != BUDDY_CHUNK_ITEMS) {
+		next_header = idx_to_header(chunk, header->next_index);
+		next_header->prev_index = BUDDY_CHUNK_ITEMS;
+	}
+
+	header->prev_index = BUDDY_CHUNK_ITEMS;
+	header->next_index = BUDDY_CHUNK_ITEMS;
+	if (idx_set_order(chunk, retidx, order_req))
+		return (u64)NULL;
+
+	if (idx_set_allocated(chunk, retidx, true))
+		return (u64)NULL;
+
+	/*
+	 * Do not unpoison the address yet, will be done by the caller
+	 * because the caller has the exact allocation size requested.
+	 */
+	address = (u64)idx_to_addr(chunk, retidx);
+
+	/* If we allocated from a larger-order chunk, split the buddies. */
+	for (i = order_req; i < order && can_loop; i++) {
+		/*
+		 * Flip the bit for the current order (the bit is guaranteed
+		 * to be 0, so just add 1 << i).
+		 */
+		idx = retidx + (1 << i);
+
+		/* Add the buddy of the allocation to the free list. */
+		header = idx_to_header(chunk, idx);
+		/* Unpoison the buddy header */
+		asan_unpoison(header, sizeof(*header));
+		if (idx_set_allocated(chunk, idx, false))
+			return (u64)NULL;
+
+		if (idx_set_order(chunk, idx, i))
+			return (u64)NULL;
+
+		/* Push the header to the beginning of the freelists list. */
+		tmpidx = chunk->freelists[i];
+
+		header->prev_index = BUDDY_CHUNK_ITEMS;
+		header->next_index = tmpidx;
+
+		if (tmpidx != BUDDY_CHUNK_ITEMS) {
+			tmp_header = idx_to_header(chunk, tmpidx);
+			tmp_header->prev_index = idx;
+		}
+
+		chunk->freelists[i] = idx;
+	}
+
+	return address;
+}
+
+/* Scan the existing chunks for available memory. */
+static u64 buddy_alloc_from_existing_chunks(buddy_t *buddy, int order)
+{
+	buddy_chunk_t *chunk;
+	u64 address;
+
+	for (chunk = buddy->first_chunk; chunk != NULL && can_loop;
+	     chunk = chunk->next) {
+		address = buddy_chunk_alloc(chunk, order);
+		if (address)
+			return address;
+	}
+
+	return (u64)NULL;
+}
+
+/*
+ * Try an allocation from a newly allocated chunk. Also
+ * incorporate the chunk into the linked list.
+ */
+static u64 buddy_alloc_from_new_chunk(buddy_t *buddy, buddy_chunk_t *chunk, int order)
+{
+	u64 address;
+
+	if (buddy_lock(buddy))
+		return (u64)NULL;
+
+
+	/*
+	 * Add the chunk into the allocator and try
+	 * to allocate specifically from that chunk.
+	 */
+	chunk->next = buddy->first_chunk;
+	buddy->first_chunk = chunk;
+
+	address = buddy_chunk_alloc(buddy->first_chunk, order);
+
+	buddy_unlock(buddy);
+
+	return (u64)address;
+}
+__weak
+u64 buddy_alloc_internal(buddy_t __arg_arena *buddy, size_t size)
+{
+	buddy_chunk_t *chunk;
+	u64 address = (u64)NULL;
+	int order;
+
+	if (!buddy)
+		return (u64)NULL;
+
+	order = size_to_order(size);
+	if (order >= BUDDY_CHUNK_NUM_ORDERS || order < 0) {
+		arena_stderr("invalid order %d (sz %lu)\n", order, size);
+		return (u64)NULL;
+	}
+
+	if (buddy_lock(buddy))
+		return (u64)NULL;
+
+	address = buddy_alloc_from_existing_chunks(buddy, order);
+	buddy_unlock(buddy);
+	if (address)
+		goto done;
+
+	/* Get a new chunk. */
+	chunk = buddy_chunk_get(buddy);
+	if (chunk)
+		address = buddy_alloc_from_new_chunk(buddy, chunk, order);
+
+done:
+	/* If we failed to allocate memory, return NULL. */
+	if (!address)
+		return (u64)NULL;
+
+	/*
+	 * Unpoison exactly the amount of bytes requested. If the
+	 * data is smaller than the header, we must poison any
+	 * unused bytes that were part of the header.
+	 */
+	if (size < BUDDY_HEADER_OFF + sizeof(buddy_header_t))
+		asan_poison((u8 __arena *)address + BUDDY_HEADER_OFF,
+			    BUDDY_POISONED, sizeof(buddy_header_t));
+
+	asan_unpoison((u8 __arena *)address, size);
+
+	return address;
+}
+
+static __always_inline int buddy_free_unlocked(buddy_t *buddy, u64 addr)
+{
+	buddy_header_t *header, *buddy_header;
+	u64 idx, buddy_idx, tmp_idx;
+	buddy_chunk_t *chunk;
+	bool allocated;
+	u8 order;
+
+	if (!buddy)
+		return -EINVAL;
+
+	if (addr & (BUDDY_MIN_ALLOC_BYTES - 1)) {
+		arena_stderr("Freeing unaligned address %llx\n", addr);
+		return -EINVAL;
+	}
+
+	/* Get (chunk, idx) out of the address. */
+	chunk = (void __arena *)(addr & ~BUDDY_CHUNK_OFFSET_MASK);
+	idx = (addr & BUDDY_CHUNK_OFFSET_MASK) / BUDDY_MIN_ALLOC_BYTES;
+
+	/* Mark the block as unallocated so we can access the header. */
+	idx_set_allocated(chunk, idx, false);
+
+	order  = idx_get_order(chunk, idx);
+	header = idx_to_header(chunk, idx);
+
+	/* The header is in the block itself, keep it unpoisoned. */
+	asan_poison((u8 __arena *)addr, BUDDY_POISONED,
+		    BUDDY_MIN_ALLOC_BYTES << order);
+	asan_unpoison(header, sizeof(*header));
+
+	/*
+	 * Coalescing loop. Merge with free buddies of equal order.
+	 * For every coalescing step, keep the left buddy and
+	 * drop the right buddy's header.
+	 */
+	for (; order < BUDDY_CHUNK_NUM_ORDERS && can_loop; order++) {
+		buddy_idx = idx ^ (1 << order);
+
+		/* Check if the buddy is actually free. */
+		idx_is_allocated(chunk, buddy_idx, &allocated);
+		if (allocated)
+			break;
+
+		/*
+		 * If buddy is not the same order as the chunk
+		 * being freed, then we're done coalescing.
+		 */
+		if (idx_get_order(chunk, buddy_idx) != order)
+			break;
+
+		buddy_header = idx_to_header(chunk, buddy_idx);
+		header_remove_freelist(chunk, buddy_header, order);
+
+		/* Keep the left header out of the two buddies, drop the other one. */
+		if (buddy_idx < idx) {
+			tmp_idx = idx;
+			idx = buddy_idx;
+			buddy_idx = tmp_idx;
+		}
+
+		/* Remove the buddy from the freelists so that we can merge it. */
+		idx_set_order(chunk, buddy_idx, order);
+
+		buddy_header = idx_to_header(chunk, buddy_idx);
+		asan_poison(buddy_header, BUDDY_POISONED,
+			    sizeof(*buddy_header));
+	}
+
+	/* Header properly freed but not in any freelists yet .*/
+	idx_set_order(chunk, idx, order);
+
+	header = idx_to_header(chunk, idx);
+	header_add_freelist(chunk, header, idx, order);
+
+	return 0;
+}
+
+__weak int buddy_free_internal(buddy_t __arg_arena *buddy, u64 addr)
+{
+	int ret;
+
+	if (!buddy)
+		return -EINVAL;
+
+	ret = buddy_lock(buddy);
+	if (ret)
+		return ret;
+
+	buddy_free_unlocked(buddy, addr);
+
+	buddy_unlock(buddy);
+
+	return 0;
+}
+
+__weak char _license[] SEC("license") = "GPL";
diff --git a/tools/testing/selftests/bpf/libarena/src/common.bpf.c b/tools/testing/selftests/bpf/libarena/src/common.bpf.c
index 4f9e95aa6cd5..7296d5a851f4 100644
--- a/tools/testing/selftests/bpf/libarena/src/common.bpf.c
+++ b/tools/testing/selftests/bpf/libarena/src/common.bpf.c
@@ -1,9 +1,13 @@
 // SPDX-License-Identifier: LGPL-2.1 OR BSD-2-Clause
 /* Copyright (c) 2026 Meta Platforms, Inc. and affiliates. */
 #include <common.h>
+#include <asan.h>
+#include <buddy.h>
 
 const volatile u32 zero = 0;
 
+buddy_t buddy;
+
 /* How many pages do we reserve at the beginning of the arena segment? */
 #define RESERVE_ALLOC (8)
 
@@ -56,4 +60,12 @@ __weak int arena_alloc_reserve(void)
 	return bpf_arena_reserve_pages(&arena, NULL, RESERVE_ALLOC);
 }
 
+SEC("syscall")
+__weak int arena_buddy_reset(void)
+{
+	buddy_destroy(&buddy);
+
+	return buddy_init(&buddy);
+}
+
 char _license[] SEC("license") = "GPL";
-- 
2.53.0

[PATCH bpf-next v5 9/9] selftests/bpf: Add selftests for libarena buddy allocator

[Emil Tsalapatis]

Introduce selftests for the buddy allocator without ASAN.
Add the libarena selftests both to the libarena test
runner and to test_progs, so that they are a) available when
libarena is pulled as a standalone library, and b) exercised
along with all other test programs in this directory.

Currently ASAN for libarena requires LLVM 22, which is not
available in the CI. Delay testing the ASAN case directly
through test_progs until that is the case. The ASAN version
is available for testing on demand through the library's own
harness.

Signed-off-by: Emil Tsalapatis <emil@etsalapatis.com>
---
 .../bpf/libarena/include/selftest_helpers.h   |   7 +
 .../bpf/libarena/selftests/selftest.c         |   8 +
 .../libarena/selftests/st_asan_buddy.bpf.c    | 243 ++++++++++++++++++
 .../bpf/libarena/selftests/st_buddy.bpf.c     | 211 +++++++++++++++
 .../selftests/bpf/prog_tests/libarena.c       |  61 +++++
 5 files changed, 530 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/libarena/selftests/st_asan_buddy.bpf.c
 create mode 100644 tools/testing/selftests/bpf/libarena/selftests/st_buddy.bpf.c
 create mode 100644 tools/testing/selftests/bpf/prog_tests/libarena.c

diff --git a/tools/testing/selftests/bpf/libarena/include/selftest_helpers.h b/tools/testing/selftests/bpf/libarena/include/selftest_helpers.h
index 36e4f36ae8e6..cb8bf9b7cd71 100644
--- a/tools/testing/selftests/bpf/libarena/include/selftest_helpers.h
+++ b/tools/testing/selftests/bpf/libarena/include/selftest_helpers.h
@@ -112,3 +112,10 @@ static inline int libarena_asan_init(int arena_get_base_fd,
 		return ret;
 	return opts.retval;
 }
+
+static inline bool libarena_must_setup_alloc(struct bpf_program *prog)
+{
+	const char *name = bpf_program__name(prog);
+
+	return !strstr(name, "test_buddy");
+}
diff --git a/tools/testing/selftests/bpf/libarena/selftests/selftest.c b/tools/testing/selftests/bpf/libarena/selftests/selftest.c
index 45ab1ac0ef78..824e5f6b4b26 100644
--- a/tools/testing/selftests/bpf/libarena/selftests/selftest.c
+++ b/tools/testing/selftests/bpf/libarena/selftests/selftest.c
@@ -17,6 +17,7 @@
 
 #include <common.h>
 #include <asan.h>
+#include <buddy.h>
 #include <selftest_helpers.h>
 
 #ifdef BPF_ARENA_ASAN
@@ -101,6 +102,13 @@ static int init_arena(selftest *skel)
 static int run_test(selftest *skel, struct bpf_program *prog)
 {
 	int prog_fd;
+	int ret;
+
+	if (libarena_must_setup_alloc(prog)) {
+		ret = libarena_run_prog(bpf_program__fd(skel->progs.arena_buddy_reset));
+		if (ret)
+			return ret;
+	}
 
 	prog_fd = bpf_program__fd(prog);
 	if (prog_fd < 0)
diff --git a/tools/testing/selftests/bpf/libarena/selftests/st_asan_buddy.bpf.c b/tools/testing/selftests/bpf/libarena/selftests/st_asan_buddy.bpf.c
new file mode 100644
index 000000000000..21e811f60003
--- /dev/null
+++ b/tools/testing/selftests/bpf/libarena/selftests/st_asan_buddy.bpf.c
@@ -0,0 +1,243 @@
+// SPDX-License-Identifier: LGPL-2.1 OR BSD-2-Clause
+/* Copyright (c) 2026 Meta Platforms, Inc. and affiliates. */
+
+#include <common.h>
+
+#include <asan.h>
+#include <buddy.h>
+
+extern buddy_t buddy;
+
+#ifdef BPF_ARENA_ASAN
+
+#include "st_asan_common.h"
+
+u64 __arena st_asan_buddy_lock;
+
+static __always_inline int asan_test_buddy_oob_single(size_t alloc_size)
+{
+	u8 __arena *mem;
+	int ret, i;
+
+	ret = asan_validate();
+	if (ret < 0)
+		return ret;
+
+	mem = buddy_alloc(&buddy, alloc_size);
+	if (!mem) {
+		arena_stdout("buddy_alloc failed for size %lu", alloc_size);
+		return -ENOMEM;
+	}
+
+	ret = asan_validate();
+	if (ret < 0)
+		return ret;
+
+	for (i = zero; i < alloc_size && can_loop; i++) {
+		mem[i] = 0xba;
+		ret = asan_validate_addr(false, &mem[i]);
+		if (ret < 0)
+			return ret;
+	}
+
+	mem[alloc_size] = 0xba;
+	ret = asan_validate_addr(true, &mem[alloc_size]);
+	if (ret < 0)
+		return ret;
+
+	buddy_free(&buddy, mem);
+
+	return 0;
+}
+
+/*
+ * Factored out because asan_validate_addr is complex enough to cause
+ * verification failures if verified with the rest of asan_test_buddy_uaf_single.
+ */
+__weak int asan_test_buddy_byte(u8 __arena __arg_arena *mem, int i, bool freed)
+{
+	int ret;
+
+	/* The header in freed blocks doesn't get poisoned. */
+	if (freed && BUDDY_HEADER_OFF <= i &&
+		i < BUDDY_HEADER_OFF + sizeof(struct buddy_header))
+		return 0;
+
+	mem[i] = 0xba;
+	ret = asan_validate_addr(freed, &mem[i]);
+	if (ret < 0)
+		return ret;
+
+	return 0;
+}
+
+__weak int asan_test_buddy_uaf_single(size_t alloc_size)
+{
+	u8 __arena *mem;
+	int ret;
+	int i;
+
+	mem = buddy_alloc(&buddy, alloc_size);
+	if (!mem) {
+		arena_stdout("buddy_alloc failed for size %lu", alloc_size);
+		return -ENOMEM;
+	}
+
+	ret = asan_validate();
+	if (ret < 0)
+		return ret;
+
+	for (i = zero; i < alloc_size && can_loop; i++) {
+		ret = asan_test_buddy_byte(mem, i, false);
+		if (ret)
+			return ret;
+	}
+
+	ret = asan_validate();
+	if (ret < 0)
+		return ret;
+
+	buddy_free(&buddy, mem);
+
+	for (i = zero; i < alloc_size && can_loop; i++) {
+		ret = asan_test_buddy_byte(mem, i, true);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+struct buddy_blob {
+	volatile u8 mem[48];
+	u8 oob;
+};
+
+static __always_inline int asan_test_buddy_blob_single(void)
+{
+	volatile struct buddy_blob __arena *blob;
+	const size_t alloc_size = sizeof(struct buddy_blob) - 1;
+	int ret;
+
+	blob = buddy_alloc(&buddy, alloc_size);
+	if (!blob)
+		return -ENOMEM;
+
+	blob->mem[0] = 0xba;
+	ret = asan_validate_addr(false, &blob->mem[0]);
+	if (ret < 0)
+		return ret;
+
+	blob->mem[47] = 0xba;
+	ret = asan_validate_addr(false, &blob->mem[47]);
+	if (ret < 0)
+		return ret;
+
+	blob->oob = 0;
+	ret = asan_validate_addr(true, &blob->oob);
+	if (ret < 0)
+		return ret;
+
+	buddy_free(&buddy, (void __arena *)blob);
+
+	return 0;
+}
+
+SEC("syscall")
+__weak int asan_test_buddy_oob(void)
+{
+	size_t sizes[] = {
+		7, 8, 17, 18, 64, 256, 317, 512, 1024,
+	};
+	int ret, i;
+
+	ret = buddy_init(&buddy);
+	if (ret) {
+		arena_stdout("buddy_init failed with %d", ret);
+		return ret;
+	}
+
+	for (i = zero; i < sizeof(sizes) / sizeof(sizes[0]) && can_loop; i++) {
+		ret = asan_test_buddy_oob_single(sizes[i]);
+		if (ret) {
+			arena_stdout("%s:%d Failed for size %lu", __func__,
+				   __LINE__, sizes[i]);
+			buddy_destroy(&buddy);
+			return ret;
+		}
+	}
+
+	buddy_destroy(&buddy);
+
+	ret = asan_validate();
+	if (ret < 0)
+		return ret;
+
+	return 0;
+}
+
+SEC("syscall")
+__weak int asan_test_buddy_uaf(void)
+{
+	size_t sizes[] = { 16, 32, 64, 128, 256, 512, 128, 1024, 16384 };
+	int ret, i;
+
+	ret = buddy_init(&buddy);
+	if (ret) {
+		arena_stdout("buddy_init failed with %d", ret);
+		return ret;
+	}
+
+	for (i = zero; i < sizeof(sizes) / sizeof(sizes[0]) && can_loop; i++) {
+		ret = asan_test_buddy_uaf_single(sizes[i]);
+		if (ret) {
+			arena_stdout("%s:%d Failed for size %lu", __func__,
+				   __LINE__, sizes[i]);
+			buddy_destroy(&buddy);
+			return ret;
+		}
+	}
+
+	buddy_destroy(&buddy);
+
+	ret = asan_validate();
+	if (ret < 0)
+		return ret;
+
+	return 0;
+}
+
+SEC("syscall")
+__weak int asan_test_buddy_blob(void)
+{
+	const int iters = 10;
+	int ret, i;
+
+	ret = buddy_init(&buddy);
+	if (ret) {
+		arena_stdout("buddy_init failed with %d", ret);
+		return ret;
+	}
+
+	for (i = zero; i < iters && can_loop; i++) {
+		ret = asan_test_buddy_blob_single();
+		if (ret) {
+			arena_stdout("%s:%d Failed on iteration %d", __func__,
+				   __LINE__, i);
+			buddy_destroy(&buddy);
+			return ret;
+		}
+	}
+
+	buddy_destroy(&buddy);
+
+	ret = asan_validate();
+	if (ret < 0)
+		return ret;
+
+	return 0;
+}
+
+#endif
+
+__weak char _license[] SEC("license") = "GPL";
diff --git a/tools/testing/selftests/bpf/libarena/selftests/st_buddy.bpf.c b/tools/testing/selftests/bpf/libarena/selftests/st_buddy.bpf.c
new file mode 100644
index 000000000000..82afdeb49496
--- /dev/null
+++ b/tools/testing/selftests/bpf/libarena/selftests/st_buddy.bpf.c
@@ -0,0 +1,211 @@
+// SPDX-License-Identifier: LGPL-2.1 OR BSD-2-Clause
+/* Copyright (c) 2026 Meta Platforms, Inc. and affiliates. */
+
+#include <common.h>
+
+#include <asan.h>
+#include <buddy.h>
+
+extern buddy_t buddy;
+
+struct segarr_entry {
+	u8 __arena *block;
+	size_t sz;
+	u8 poison;
+};
+
+typedef struct segarr_entry __arena segarr_entry_t;
+
+#define SEGARRLEN (512)
+static struct segarr_entry __arena segarr[SEGARRLEN];
+static void __arena *ptrs[17];
+size_t __arena alloc_sizes[] = { 3, 17, 1025, 129, 16350, 333, 9, 517 };
+size_t __arena alloc_multiple_sizes[] = { 3, 17, 1025, 129, 16350, 333, 9, 517, 2099 };
+size_t __arena alloc_free_sizes[] = { 3, 17, 64, 129, 256, 333, 512, 517 };
+size_t __arena alignment_sizes[] = { 1, 3, 7, 8, 9, 15, 16, 17, 31,
+				     32, 64, 100, 128, 255, 256, 512, 1000 };
+
+SEC("syscall")
+__weak int test_buddy_create(void)
+{
+	const int iters = 10;
+	int ret, i;
+
+	for (i = zero; i < iters && can_loop; i++) {
+		ret = buddy_init(&buddy);
+		if (ret)
+			return ret;
+
+		ret = buddy_destroy(&buddy);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+SEC("syscall")
+__weak int test_buddy_alloc(void)
+{
+	void __arena *mem;
+	int ret, i;
+
+	for (i = zero; i < 8 && can_loop; i++) {
+		ret = buddy_init(&buddy);
+		if (ret)
+			return ret;
+
+		mem = buddy_alloc(&buddy, alloc_sizes[i]);
+		if (!mem) {
+			buddy_destroy(&buddy);
+			return -ENOMEM;
+		}
+
+		buddy_destroy(&buddy);
+	}
+
+	return 0;
+}
+
+SEC("syscall")
+__weak int test_buddy_alloc_free(void)
+{
+	const int iters = 800;
+	void __arena *mem;
+	int ret, i;
+
+	ret = buddy_init(&buddy);
+	if (ret)
+		return ret;
+
+	for (i = zero; i < iters && can_loop; i++) {
+		mem = buddy_alloc(&buddy, alloc_free_sizes[(i * 5) % 8]);
+		if (!mem) {
+			buddy_destroy(&buddy);
+			return -ENOMEM;
+		}
+
+		buddy_free(&buddy, mem);
+	}
+
+	buddy_destroy(&buddy);
+
+	return 0;
+}
+
+SEC("syscall")
+__weak int test_buddy_alloc_multiple(void)
+{
+	int ret, j;
+	u32 i, idx;
+	u8 __arena *mem;
+	size_t sz;
+	u8 poison;
+
+	ret = buddy_init(&buddy);
+	if (ret)
+		return ret;
+
+	/*
+	 * Cycle through each size, allocating an entry in the
+	 * segarr. Continue for SEGARRLEN iterations. For every
+	 * allocation write down the size, use the current index
+	 * as a poison value, and log it with the pointer in the
+	 * segarr entry. Use the poison value to poison the entire
+	 * allocated memory according to the size given.
+	 */
+	idx = 0;
+	for (i = zero; i < SEGARRLEN && can_loop; i++) {
+		sz = alloc_multiple_sizes[i % 9];
+		poison = (u8)i;
+
+		mem = buddy_alloc(&buddy, sz);
+		if (!mem) {
+			buddy_destroy(&buddy);
+			arena_stdout("%s:%d", __func__, __LINE__);
+			return -ENOMEM;
+		}
+
+		segarr[i].block = mem;
+		segarr[i].sz = sz;
+		segarr[i].poison = poison;
+
+		for (j = zero; j < sz && can_loop; j++) {
+			mem[j] = poison;
+			if (mem[j] != poison) {
+				buddy_destroy(&buddy);
+				return -EINVAL;
+			}
+		}
+	}
+
+	/*
+	 * Go to (i * 17) % SEGARRLEN, and free the block pointed to.
+	 * Before freeing, check all bytes have the poisoned value
+	 * corresponding to the element. If any values are unexpected,
+	 * return an error. Skip some elements to test destroying the
+	 * buddy allocator while data is still allocated.
+	 */
+	for (i = 10; i < SEGARRLEN && can_loop; i++) {
+		idx = (i * 17) % SEGARRLEN;
+
+		mem = segarr[idx].block;
+		sz = segarr[idx].sz;
+		poison = segarr[idx].poison;
+
+		for (j = zero; j < sz && can_loop; j++) {
+			if (mem[j] != poison) {
+				buddy_destroy(&buddy);
+				arena_stdout("%s:%d %lx %u vs %u", __func__,
+					   __LINE__, &mem[j], mem[j], poison);
+				return -EINVAL;
+			}
+		}
+
+		buddy_free(&buddy, mem);
+	}
+
+	buddy_destroy(&buddy);
+
+	return 0;
+}
+
+SEC("syscall")
+__weak int test_buddy_alignment(void)
+{
+	int ret, i;
+
+	ret = buddy_init(&buddy);
+	if (ret)
+		return ret;
+
+	/* Allocate various sizes and check alignment */
+	for (i = zero; i < 17 && can_loop; i++) {
+		ptrs[i] = buddy_alloc(&buddy, alignment_sizes[i]);
+		if (!ptrs[i]) {
+			arena_stdout("alignment test: alloc failed for size %lu",
+				   alignment_sizes[i]);
+			buddy_destroy(&buddy);
+			return -ENOMEM;
+		}
+
+		/* Check 8-byte alignment */
+		if ((u64)ptrs[i] & 0x7) {
+			arena_stdout(
+				"alignment test: ptr %llx not 8-byte aligned (size %lu)",
+				(u64)ptrs[i], alignment_sizes[i]);
+			buddy_destroy(&buddy);
+			return -EINVAL;
+		}
+	}
+
+	/* Free all allocations */
+	for (i = zero; i < 17 && can_loop; i++)
+		buddy_free(&buddy, ptrs[i]);
+
+	buddy_destroy(&buddy);
+
+	return 0;
+}
+
+__weak char _license[] SEC("license") = "GPL";
diff --git a/tools/testing/selftests/bpf/prog_tests/libarena.c b/tools/testing/selftests/bpf/prog_tests/libarena.c
new file mode 100644
index 000000000000..548e76360e22
--- /dev/null
+++ b/tools/testing/selftests/bpf/prog_tests/libarena.c
@@ -0,0 +1,61 @@
+// SPDX-License-Identifier: LGPL-2.1 OR BSD-2-Clause
+/* Copyright (c) 2026 Meta Platforms, Inc. and affiliates. */
+#include <test_progs.h>
+#include <unistd.h>
+
+#include "libarena/include/common.h"
+#include "libarena/include/asan.h"
+#include "libarena/include/buddy.h"
+#include "libarena/include/selftest_helpers.h"
+
+#include "libarena/libarena.skel.h"
+
+static void run_libarena_test(struct libarena *skel, struct bpf_program *prog,
+		const char *name)
+{
+	int ret;
+
+	if (libarena_must_setup_alloc(prog)) {
+		ret = libarena_run_prog(bpf_program__fd(skel->progs.arena_buddy_reset));
+		ASSERT_OK(ret, "arena_buddy_reset");
+	}
+
+	ret = libarena_run_prog(bpf_program__fd(prog));
+
+	ASSERT_OK(ret, name);
+
+}
+
+void test_libarena(void)
+{
+	struct libarena *skel;
+	struct bpf_program *prog;
+	int ret;
+
+	skel = libarena__open_and_load();
+	if (!ASSERT_OK_PTR(skel, "open_and_load"))
+		return;
+
+	ret = libarena__attach(skel);
+	if (!ASSERT_OK(ret, "attach"))
+		goto out;
+
+	ret = libarena_run_prog(bpf_program__fd(skel->progs.arena_alloc_reserve));
+	if (!ASSERT_OK(ret, "arena_alloc_reserve"))
+		goto out;
+
+	bpf_object__for_each_program(prog, skel->obj) {
+		const char *name = bpf_program__name(prog);
+
+		if (!libarena_is_test_prog(name))
+			continue;
+
+		if (!test__start_subtest(name))
+			continue;
+
+		run_libarena_test(skel, prog, name);
+	}
+
+out:
+	libarena__destroy(skel);
+}
-- 
2.53.0

Re: [PATCH bpf-next v5 1/9] bpf: Upgrade scalar to PTR_TO_ARENA on arena pointer addition

[bot+bpf-ci]

[-- Attachment #1: Type: text/plain, Size: 2140 bytes --]

> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> index 9c1135d373e24..6c5721434bc4b 100644
> --- a/kernel/bpf/verifier.c
> +++ b/kernel/bpf/verifier.c
> @@ -16655,11 +16655,32 @@ static int adjust_reg_min_max_vals(struct bpf_verifier_env *env,

[ ... ]

> +		if (dst_reg->type != PTR_TO_ARENA) {
> +			/* Can't do arena arithmetic with non-scalars. */
> +			if (dst_reg->type != SCALAR_VALUE) {
> +				verbose(env, "R%d %s R%d: Invalid operation between "
> +						"bpf_reg_state types %s and %s\n",
> +					insn->dst_reg,
> +					bpf_alu_string[opcode >> 4],
> +					insn->src_reg,
> +					reg_type_str(env, dst_reg->type),
> +					reg_type_str(env, src_reg->type));

reg_type_str() returns a pointer to the shared buffer env->tmp_str_buf.
Its own comment in log.c says:

    Note that reg_type_str() can not appear more than once in a
    single verbose() statement.

Calling it twice in the same verbose() here means both %s arguments
point to the same buffer, so the error message will display one of the
two type names twice and the other not at all.

Would it work to split this into two verbose() calls, or save one
result to a local buffer before printing?

> +				return -EACCES;
> +			}
> +
> +			*dst_reg = *src_reg;
> +			dst_reg->subreg_def = env->insn_idx + 1;
> +		}

[ ... ]

This commit fixes a bug where the verifier rejects valid BPF programs
when the compiler places the result of a PTR_TO_ARENA + SCALAR addition
into the scalar register. The bug was introduced by commit 6082b6c328b5
("bpf: Recognize addr_space_cast instruction in the verifier.") which
added the original PTR_TO_ARENA handling in the verifier.

This was also noted during review of the v3 series on lore, and the
Fixes: tag was acknowledged but appears to be missing from the final
version. Should this carry:

Fixes: 6082b6c328b5 ("bpf: Recognize addr_space_cast instruction in the verifier.")


---
AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md

CI run summary: https://github.com/kernel-patches/bpf/actions/runs/24254055723

Re: [PATCH bpf-next v5 1/9] bpf: Upgrade scalar to PTR_TO_ARENA on arena pointer addition

[Emil Tsalapatis]

On Fri Apr 10, 2026 at 1:34 PM EDT, bot+bpf-ci wrote:
>> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
>> index 9c1135d373e24..6c5721434bc4b 100644
>> --- a/kernel/bpf/verifier.c
>> +++ b/kernel/bpf/verifier.c
>> @@ -16655,11 +16655,32 @@ static int adjust_reg_min_max_vals(struct bpf_verifier_env *env,
>
> [ ... ]
>
>> +		if (dst_reg->type != PTR_TO_ARENA) {
>> +			/* Can't do arena arithmetic with non-scalars. */
>> +			if (dst_reg->type != SCALAR_VALUE) {
>> +				verbose(env, "R%d %s R%d: Invalid operation between "
>> +						"bpf_reg_state types %s and %s\n",
>> +					insn->dst_reg,
>> +					bpf_alu_string[opcode >> 4],
>> +					insn->src_reg,
>> +					reg_type_str(env, dst_reg->type),
>> +					reg_type_str(env, src_reg->type));
>
> reg_type_str() returns a pointer to the shared buffer env->tmp_str_buf.
> Its own comment in log.c says:
>
>     Note that reg_type_str() can not appear more than once in a
>     single verbose() statement.
>
> Calling it twice in the same verbose() here means both %s arguments
> point to the same buffer, so the error message will display one of the
> two type names twice and the other not at all.
>
> Would it work to split this into two verbose() calls, or save one
> result to a local buffer before printing?
>

Ack. We can just split this into two verbose() statements.

>> +				return -EACCES;
>> +			}
>> +
>> +			*dst_reg = *src_reg;
>> +			dst_reg->subreg_def = env->insn_idx + 1;
>> +		}
>
> [ ... ]
>
> This commit fixes a bug where the verifier rejects valid BPF programs
> when the compiler places the result of a PTR_TO_ARENA + SCALAR addition
> into the scalar register. The bug was introduced by commit 6082b6c328b5
> ("bpf: Recognize addr_space_cast instruction in the verifier.") which
> added the original PTR_TO_ARENA handling in the verifier.
>
> This was also noted during review of the v3 series on lore, and the
> Fixes: tag was acknowledged but appears to be missing from the final
> version. Should this carry:
>
> Fixes: 6082b6c328b5 ("bpf: Recognize addr_space_cast instruction in the verifier.")
>

Not sure actually. The original patch logic was overly restrictive, but
it's not like it was broken. This code is more of an expansion than a
fix imo.

>
> ---
> AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
> See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md
>
> CI run summary: https://github.com/kernel-patches/bpf/actions/runs/24254055723

Re: [PATCH bpf-next v5 5/9] selftests/bpf: Add basic libarena scaffolding

[Song Liu]

On Fri, Apr 10, 2026 at 9:31 AM Emil Tsalapatis <emil@etsalapatis.com> wrote:
[...]
> diff --git a/tools/testing/selftests/bpf/libarena/include/common.h b/tools/testing/selftests/bpf/libarena/include/common.h
> new file mode 100644
> index 000000000000..5acf9b994943
> --- /dev/null
> +++ b/tools/testing/selftests/bpf/libarena/include/common.h
> @@ -0,0 +1,67 @@
> +// SPDX-License-Identifier: LGPL-2.1 OR BSD-2-Clause
> +/* Copyright (c) 2026 Meta Platforms, Inc. and affiliates. */
> +#pragma once
> +
> +#ifdef __BPF__
> +
> +#include <vmlinux.h>
> +
> +#include "bpf_experimental.h"

bpf_experimental.h is not needed here.

> +#include "bpf_arena_common.h"
> +#include "bpf_arena_spin_lock.h"

These two headers are indeed needed. However, what's the plan
with them? They are not part of libarena at the moment. How should
we include them in the future?

Thanks,
Song

> +#include <asm-generic/errno.h>
> +
> +#ifndef __BPF_FEATURE_ADDR_SPACE_CAST
> +#error "Arena allocators require bpf_addr_space_cast feature"
> +#endif
> +
[...]

Re: [PATCH bpf-next v5 5/9] selftests/bpf: Add basic libarena scaffolding

[Kumar Kartikeya Dwivedi]

On Fri, 10 Apr 2026 at 20:39, Song Liu <song@kernel.org> wrote:
>
> On Fri, Apr 10, 2026 at 9:31 AM Emil Tsalapatis <emil@etsalapatis.com> wrote:
> [...]
> > diff --git a/tools/testing/selftests/bpf/libarena/include/common.h b/tools/testing/selftests/bpf/libarena/include/common.h
> > new file mode 100644
> > index 000000000000..5acf9b994943
> > --- /dev/null
> > +++ b/tools/testing/selftests/bpf/libarena/include/common.h
> > @@ -0,0 +1,67 @@
> > +// SPDX-License-Identifier: LGPL-2.1 OR BSD-2-Clause
> > +/* Copyright (c) 2026 Meta Platforms, Inc. and affiliates. */
> > +#pragma once
> > +
> > +#ifdef __BPF__
> > +
> > +#include <vmlinux.h>
> > +
> > +#include "bpf_experimental.h"
>
> bpf_experimental.h is not needed here.
>
> > +#include "bpf_arena_common.h"
> > +#include "bpf_arena_spin_lock.h"
>
> These two headers are indeed needed. However, what's the plan
> with them? They are not part of libarena at the moment. How should
> we include them in the future?

I would propose making them part of libarena, and adding libarena's
header directory to -I for all progs in selftest to simplify
inclusion.

>
> Thanks,
> Song
>
> > +#include <asm-generic/errno.h>
> > +
> > +#ifndef __BPF_FEATURE_ADDR_SPACE_CAST
> > +#error "Arena allocators require bpf_addr_space_cast feature"
> > +#endif
> > +
> [...]

Re: [PATCH bpf-next v5 5/9] selftests/bpf: Add basic libarena scaffolding

[Emil Tsalapatis]

On Fri Apr 10, 2026 at 3:44 PM EDT, Kumar Kartikeya Dwivedi wrote:
> On Fri, 10 Apr 2026 at 20:39, Song Liu <song@kernel.org> wrote:
>>
>> On Fri, Apr 10, 2026 at 9:31 AM Emil Tsalapatis <emil@etsalapatis.com> wrote:
>> [...]
>> > diff --git a/tools/testing/selftests/bpf/libarena/include/common.h b/tools/testing/selftests/bpf/libarena/include/common.h
>> > new file mode 100644
>> > index 000000000000..5acf9b994943
>> > --- /dev/null
>> > +++ b/tools/testing/selftests/bpf/libarena/include/common.h
>> > @@ -0,0 +1,67 @@
>> > +// SPDX-License-Identifier: LGPL-2.1 OR BSD-2-Clause
>> > +/* Copyright (c) 2026 Meta Platforms, Inc. and affiliates. */
>> > +#pragma once
>> > +
>> > +#ifdef __BPF__
>> > +
>> > +#include <vmlinux.h>
>> > +
>> > +#include "bpf_experimental.h"
>>
>> bpf_experimental.h is not needed here.
>>
>> > +#include "bpf_arena_common.h"
>> > +#include "bpf_arena_spin_lock.h"
>>
>> These two headers are indeed needed. However, what's the plan
>> with them? They are not part of libarena at the moment. How should
>> we include them in the future?
>
> I would propose making them part of libarena, and adding libarena's
> header directory to -I for all progs in selftest to simplify
> inclusion.
>

+1 on this. We should have all arena related functionality in a single
directory.

>>
>> Thanks,
>> Song
>>
>> > +#include <asm-generic/errno.h>
>> > +
>> > +#ifndef __BPF_FEATURE_ADDR_SPACE_CAST
>> > +#error "Arena allocators require bpf_addr_space_cast feature"
>> > +#endif
>> > +
>> [...]