[PATCH v2 bpf-next 0/2] Track read-only map contents as known scalars in BPF verifiers

[PATCH v2 bpf-next 0/2] Track read-only map contents as known scalars in BPF verifiers

[Andrii Nakryiko]

With BPF maps supporting direct map access (currently, array_map w/ single
element, used for global data) that are read-only both from system call and
BPF side, it's possible for BPF verifier to track its contents as known
constants.

Now it's possible for user-space control app to pre-initialize read-only map
(e.g., for .rodata section) with user-provided flags and parameters and rely
on BPF verifier to detect and eliminate dead code resulting from specific
combination of input parameters.

v1->v2:
- BPF_F_RDONLY means nothing, stick to just map->frozen (Daniel);
- stick to passing just offset into map_direct_value_addr (Martin).

Andrii Nakryiko (2):
  bpf: track contents of read-only maps as scalars
  selftests/bpf: add read-only map values propagation tests

 kernel/bpf/verifier.c                         | 57 ++++++++++-
 .../selftests/bpf/prog_tests/rdonly_maps.c    | 99 +++++++++++++++++++
 .../selftests/bpf/progs/test_rdonly_maps.c    | 83 ++++++++++++++++
 3 files changed, 237 insertions(+), 2 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/rdonly_maps.c
 create mode 100644 tools/testing/selftests/bpf/progs/test_rdonly_maps.c

-- 
2.17.1

[PATCH v2 bpf-next 1/2] bpf: track contents of read-only maps as scalars

[Andrii Nakryiko]

Maps that are read-only both from BPF program side and user space side
have their contents constant, so verifier can track referenced values
precisely and use that knowledge for dead code elimination, branch
pruning, etc. This patch teaches BPF verifier how to do this.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
---
 kernel/bpf/verifier.c | 57 +++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 55 insertions(+), 2 deletions(-)

diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index ffc3e53f5300..b818fed3208d 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -2739,6 +2739,41 @@ static void coerce_reg_to_size(struct bpf_reg_state *reg, int size)
 	reg->smax_value = reg->umax_value;
 }
 
+static bool bpf_map_is_rdonly(const struct bpf_map *map)
+{
+	return (map->map_flags & BPF_F_RDONLY_PROG) && map->frozen;
+}
+
+static int bpf_map_direct_read(struct bpf_map *map, int off, int size, u64 *val)
+{
+	void *ptr;
+	u64 addr;
+	int err;
+
+	err = map->ops->map_direct_value_addr(map, &addr, off);
+	if (err)
+		return err;
+	ptr = (void *)addr + off;
+
+	switch (size) {
+	case sizeof(u8):
+		*val = (u64)*(u8 *)ptr;
+		break;
+	case sizeof(u16):
+		*val = (u64)*(u16 *)ptr;
+		break;
+	case sizeof(u32):
+		*val = (u64)*(u32 *)ptr;
+		break;
+	case sizeof(u64):
+		*val = *(u64 *)ptr;
+		break;
+	default:
+		return -EINVAL;
+	}
+	return 0;
+}
+
 /* check whether memory at (regno + off) is accessible for t = (read | write)
  * if t==write, value_regno is a register which value is stored into memory
  * if t==read, value_regno is a register which will receive the value from memory
@@ -2776,9 +2811,27 @@ static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regn
 		if (err)
 			return err;
 		err = check_map_access(env, regno, off, size, false);
-		if (!err && t == BPF_READ && value_regno >= 0)
-			mark_reg_unknown(env, regs, value_regno);
+		if (!err && t == BPF_READ && value_regno >= 0) {
+			struct bpf_map *map = reg->map_ptr;
+
+			/* if map is read-only, track its contents as scalars */
+			if (tnum_is_const(reg->var_off) &&
+			    bpf_map_is_rdonly(map) &&
+			    map->ops->map_direct_value_addr) {
+				int map_off = off + reg->var_off.value;
+				u64 val = 0;
 
+				err = bpf_map_direct_read(map, map_off, size,
+							  &val);
+				if (err)
+					return err;
+
+				regs[value_regno].type = SCALAR_VALUE;
+				__mark_reg_known(&regs[value_regno], val);
+			} else {
+				mark_reg_unknown(env, regs, value_regno);
+			}
+		}
 	} else if (reg->type == PTR_TO_CTX) {
 		enum bpf_reg_type reg_type = SCALAR_VALUE;
 
-- 
2.17.1

[PATCH v2 bpf-next 2/2] selftests/bpf: add read-only map values propagation tests

[Andrii Nakryiko]

Add tests checking that verifier does proper constant propagation for
read-only maps. If constant propagation didn't work, skipp_loop and
part_loop BPF programs would be rejected due to BPF verifier otherwise
not being able to prove they ever complete. With constant propagation,
though, they are succesfully validated as properly terminating loops.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
---
 .../selftests/bpf/prog_tests/rdonly_maps.c    | 99 +++++++++++++++++++
 .../selftests/bpf/progs/test_rdonly_maps.c    | 83 ++++++++++++++++
 2 files changed, 182 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/rdonly_maps.c
 create mode 100644 tools/testing/selftests/bpf/progs/test_rdonly_maps.c

diff --git a/tools/testing/selftests/bpf/prog_tests/rdonly_maps.c b/tools/testing/selftests/bpf/prog_tests/rdonly_maps.c
new file mode 100644
index 000000000000..9bf9de0aaeea
--- /dev/null
+++ b/tools/testing/selftests/bpf/prog_tests/rdonly_maps.c
@@ -0,0 +1,99 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <test_progs.h>
+
+struct bss {
+	unsigned did_run;
+	unsigned iters;
+	unsigned sum;
+};
+
+struct rdonly_map_subtest {
+	const char *subtest_name;
+	const char *prog_name;
+	unsigned exp_iters;
+	unsigned exp_sum;
+};
+
+void test_rdonly_maps(void)
+{
+	const char *prog_name_skip_loop = "raw_tracepoint/sys_enter:skip_loop";
+	const char *prog_name_part_loop = "raw_tracepoint/sys_enter:part_loop";
+	const char *prog_name_full_loop = "raw_tracepoint/sys_enter:full_loop";
+	const char *file = "test_rdonly_maps.o";
+	struct rdonly_map_subtest subtests[] = {
+		{ "skip loop", prog_name_skip_loop, 0, 0 },
+		{ "part loop", prog_name_part_loop, 3, 2 + 3 + 4 },
+		{ "full loop", prog_name_full_loop, 4, 2 + 3 + 4 + 5 },
+	};
+	int i, err, zero = 0, duration = 0;
+	struct bpf_link *link = NULL;
+	struct bpf_program *prog;
+	struct bpf_map *bss_map;
+	struct bpf_object *obj;
+	struct bss bss;
+
+	obj = bpf_object__open_file(file, NULL);
+	if (CHECK(IS_ERR(obj), "obj_open", "err %ld\n", PTR_ERR(obj)))
+		return;
+
+	bpf_object__for_each_program(prog, obj) {
+		bpf_program__set_raw_tracepoint(prog);
+	}
+
+	err = bpf_object__load(obj);
+	if (CHECK(err, "obj_load", "err %d errno %d\n", err, errno))
+		goto cleanup;
+
+	bss_map = bpf_object__find_map_by_name(obj, "test_rdo.bss");
+	if (CHECK(!bss_map, "find_bss_map", "failed\n"))
+		goto cleanup;
+
+	for (i = 0; i < ARRAY_SIZE(subtests); i++) {
+		const struct rdonly_map_subtest *t = &subtests[i];
+
+		if (!test__start_subtest(t->subtest_name))
+			continue;
+
+		prog = bpf_object__find_program_by_title(obj, t->prog_name);
+		if (CHECK(!prog, "find_prog", "prog '%s' not found\n",
+			  t->prog_name))
+			goto cleanup;
+
+		memset(&bss, 0, sizeof(bss));
+		err = bpf_map_update_elem(bpf_map__fd(bss_map), &zero, &bss, 0);
+		if (CHECK(err, "set_bss", "failed to set bss data: %d\n", err))
+			goto cleanup;
+
+		link = bpf_program__attach_raw_tracepoint(prog, "sys_enter");
+		if (CHECK(IS_ERR(link), "attach_prog", "prog '%s', err %ld\n",
+			  t->prog_name, PTR_ERR(link))) {
+			link = NULL;
+			goto cleanup;
+		}
+
+		/* trigger probe */
+		usleep(1);
+
+		bpf_link__destroy(link);
+		link = NULL;
+
+		err = bpf_map_lookup_elem(bpf_map__fd(bss_map), &zero, &bss);
+		if (CHECK(err, "get_bss", "failed to get bss data: %d\n", err))
+			goto cleanup;
+		if (CHECK(bss.did_run == 0, "check_run",
+			  "prog '%s' didn't run?\n", t->prog_name))
+			goto cleanup;
+		if (CHECK(bss.iters != t->exp_iters, "check_iters",
+			  "prog '%s' iters: %d, expected: %d\n",
+			  t->prog_name, bss.iters, t->exp_iters))
+			goto cleanup;
+		if (CHECK(bss.sum != t->exp_sum, "check_sum",
+			  "prog '%s' sum: %d, expected: %d\n",
+			  t->prog_name, bss.sum, t->exp_sum))
+			goto cleanup;
+	}
+
+cleanup:
+	bpf_link__destroy(link);
+	bpf_object__close(obj);
+}
diff --git a/tools/testing/selftests/bpf/progs/test_rdonly_maps.c b/tools/testing/selftests/bpf/progs/test_rdonly_maps.c
new file mode 100644
index 000000000000..52d94e8b214d
--- /dev/null
+++ b/tools/testing/selftests/bpf/progs/test_rdonly_maps.c
@@ -0,0 +1,83 @@
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (c) 2019 Facebook
+
+#include <linux/ptrace.h>
+#include <linux/bpf.h>
+#include "bpf_helpers.h"
+
+static volatile const struct {
+	unsigned a[4];
+	/*
+	 * if the struct's size is multiple of 16, compiler will put it into
+	 * .rodata.cst16 section, which is not recognized by libbpf; work
+	 * around this by ensuring we don't have 16-aligned struct
+	 */
+	char _y;
+} rdonly_values = { .a = {2, 3, 4, 5} };
+
+static volatile struct {
+	unsigned did_run;
+	unsigned iters;
+	unsigned sum;
+} res;
+
+SEC("raw_tracepoint/sys_enter:skip_loop")
+int skip_loop(struct pt_regs *ctx)
+{
+	/* prevent compiler to optimize everything out */
+	unsigned * volatile p = (void *)&rdonly_values.a;
+	unsigned iters = 0, sum = 0;
+
+	/* we should never enter this loop */
+	while (*p & 1) {
+		iters++;
+		sum += *p;
+		p++;
+	}
+	res.did_run = 1;
+	res.iters = iters;
+	res.sum = sum;
+	return 0;
+}
+
+SEC("raw_tracepoint/sys_enter:part_loop")
+int part_loop(struct pt_regs *ctx)
+{
+	/* prevent compiler to optimize everything out */
+	unsigned * volatile p = (void *)&rdonly_values.a;
+	unsigned iters = 0, sum = 0;
+
+	/* validate verifier can derive loop termination */
+	while (*p < 5) {
+		iters++;
+		sum += *p;
+		p++;
+	}
+	res.did_run = 1;
+	res.iters = iters;
+	res.sum = sum;
+	return 0;
+}
+
+SEC("raw_tracepoint/sys_enter:full_loop")
+int full_loop(struct pt_regs *ctx)
+{
+	/* prevent compiler to optimize everything out */
+	unsigned * volatile p = (void *)&rdonly_values.a;
+	int i = sizeof(rdonly_values.a) / sizeof(rdonly_values.a[0]);
+	unsigned iters = 0, sum = 0;
+
+	/* validate verifier can allow full loop as well */
+	while (i > 0 ) {
+		iters++;
+		sum += *p;
+		p++;
+		i--;
+	}
+	res.did_run = 1;
+	res.iters = iters;
+	res.sum = sum;
+	return 0;
+}
+
+char _license[] SEC("license") = "GPL";
-- 
2.17.1

Re: [PATCH v2 bpf-next 0/2] Track read-only map contents as known scalars in BPF verifiers

[Daniel Borkmann]

On Wed, Oct 09, 2019 at 01:14:56PM -0700, Andrii Nakryiko wrote:
> With BPF maps supporting direct map access (currently, array_map w/ single
> element, used for global data) that are read-only both from system call and
> BPF side, it's possible for BPF verifier to track its contents as known
> constants.
> 
> Now it's possible for user-space control app to pre-initialize read-only map
> (e.g., for .rodata section) with user-provided flags and parameters and rely
> on BPF verifier to detect and eliminate dead code resulting from specific
> combination of input parameters.
> 
> v1->v2:
> - BPF_F_RDONLY means nothing, stick to just map->frozen (Daniel);
> - stick to passing just offset into map_direct_value_addr (Martin).
> 
> Andrii Nakryiko (2):
>   bpf: track contents of read-only maps as scalars
>   selftests/bpf: add read-only map values propagation tests
> 
>  kernel/bpf/verifier.c                         | 57 ++++++++++-
>  .../selftests/bpf/prog_tests/rdonly_maps.c    | 99 +++++++++++++++++++
>  .../selftests/bpf/progs/test_rdonly_maps.c    | 83 ++++++++++++++++
>  3 files changed, 237 insertions(+), 2 deletions(-)
>  create mode 100644 tools/testing/selftests/bpf/prog_tests/rdonly_maps.c
>  create mode 100644 tools/testing/selftests/bpf/progs/test_rdonly_maps.c

Applied, thanks!