Skip to contentSkip to conversation

bpf: fix accesses to uninit stack slots

PATCH Superseded Andrei Matei 3 participants 14 messages Nov 26, 2023 – Dec 2, 2023

Subsystems (4)

v2 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

v2 cover bpf: fix accesses to uninit stack slots

Fix two related issues issues around verifying stack accesses:
1. accesses to uninitialized stack memory was allowed inconsistently
2. the maximum stack depth needed for a program was not always maintained correctly

The two issues are fixed together in one commit because the code for one affects the other.

The second patch is tests only. It was split for review purposes; it can be squashed when merging if it looks good.

Andrei Matei (2):
bpf: fix accesses to uninit stack slots
bpf: new verifier tests for stack access

 include/linux/bpf_verifier.h                  |  4 ++
 kernel/bpf/verifier.c                         | 70 ++++++++-----------
 .../selftests/bpf/progs/test_global_func16.c  |  2 +-
 .../bpf/progs/verifier_basic_stack.c          |  6 +-
 .../selftests/bpf/progs/verifier_int_ptr.c    |  2 +-
 .../selftests/bpf/progs/verifier_raw_stack.c  |  2 +-
 .../selftests/bpf/progs/verifier_var_off.c    |  4 +-
 tools/testing/selftests/bpf/test_verifier.c   | 24 +++++++
 .../selftests/bpf/verifier/atomic_cmpxchg.c   | 11 ---
 tools/testing/selftests/bpf/verifier/calls.c  |  2 +-
 tools/testing/selftests/bpf/verifier/stack.c  | 40 +++++++++++
 11 files changed, 106 insertions(+), 61 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/verifier/stack.c
Signature · 1 line
2.40.1
#1 lore raw

PATCH v2 1/2 bpf: fix accesses to uninit stack slots

Privileged programs are supposed to be able to read uninitialized stack memory (ever since 6715df8d5) but, before this patch, these accesses were permitted inconsistently. In particular, accesses were permitted above state->allocated_stack, but not below it. In other words, if the stack was already "large enough", the access was permitted, but otherwise the access was rejected instead of being allowed to "grow the stack". This undesired rejection was happening in two places:
- in check_stack_slot_within_bounds()
- in check_stack_range_initialized()
This patch arranges for these accesses to be permitted.

This patch also fixes the tracking of the stack size for variable-offset reads. This second fix is bundled in the same commit as the first one because they're inter-related. Before this patch, writes to the stack using registers containing a variable offset (as opposed to registers with fixed, known values) were not properly contributing to the function's needed stack size. As a result, it was possible for a program to verify, but then to attempt to read out-of-bounds data at runtime because a too small stack had been allocated for it.

Each function tracks the size of the stack it needs in bpf_subprog_info.stack_depth, which is maintained by update_stack_depth(). For regular memory accesses, check_mem_access() was calling update_state_depth() but it was passing in only the fixed part of the offset register, ignoring the variable offset. This was incorrect; the minimum possible value of that register should be used instead.

This tracking is now fixed by centralizing the tracking of stack size in grow_stack_state(), and by lifting the calls to grow_stack_state() to check_stack_access_within_bounds() as suggested by Andrii. The code is now simpler and more convincingly tracks the correct maximum stack size. check_stack_range_initialized() can now rely on enough stack having been allocated for the access; this helps with the fix for the first issue.

---
 include/linux/bpf_verifier.h                  |  4 ++
 kernel/bpf/verifier.c                         | 70 ++++++++-----------
 .../selftests/bpf/progs/test_global_func16.c  |  2 +-
 .../bpf/progs/verifier_basic_stack.c          |  6 +-
 .../selftests/bpf/progs/verifier_int_ptr.c    |  2 +-
 .../selftests/bpf/progs/verifier_raw_stack.c  |  2 +-
 .../selftests/bpf/progs/verifier_var_off.c    |  4 +-
 .../selftests/bpf/verifier/atomic_cmpxchg.c   | 11 ---
 tools/testing/selftests/bpf/verifier/calls.c  |  2 +-
 9 files changed, 42 insertions(+), 61 deletions(-)
include/linux/bpf_verifier.h +4 -0
index aa4d19d0bc94..5fc389e8be35 100644
@@ -630,6 +630,10 @@ struct bpf_verifier_env {
 	int exception_callback_subprog;
 	bool explore_alu_limits;
 	bool allow_ptr_leaks;
+	/* Allow access to uninitialized stack memory. Writes with fixed offset are
+	 * always allowed, so this refers to reads (with fixed or variable offset),
+	 * to writes with variable offset and to indirect (helper) accesses.
+	 */
 	bool allow_uninit_stack;
 	bool bpf_capable;
 	bool bypass_spec_v1;
kernel/bpf/verifier.c +29 -41

161 lines not on this page · raw message

tools/testing/selftests/bpf/progs/test_global_func16.c +1 -1
index e7206304632e..e3e64bc472cd 100644
@@ -13,7 +13,7 @@ __noinline int foo(int (*arr)[10])
 }
 
 SEC("cgroup_skb/ingress")
-__failure __msg("invalid indirect read from stack")
+__success
 int global_func16(struct __sk_buff *skb)
 {
 	int array[10];
tools/testing/selftests/bpf/progs/verifier_basic_stack.c +3 -3
index 359df865a8f3..069c3f91705c 100644
@@ -27,8 +27,8 @@ __naked void stack_out_of_bounds(void)
 
 SEC("socket")
 __description("uninitialized stack1")
-__failure __msg("invalid indirect read from stack")
-__failure_unpriv
+__success
+__failure_unpriv __msg_unpriv("invalid indirect read from stack")
 __naked void uninitialized_stack1(void)
 {
 	asm volatile ("					\
@@ -45,7 +45,7 @@ __naked void uninitialized_stack1(void)
 
 SEC("socket")
 __description("uninitialized stack2")
-__failure __msg("invalid read from stack")
+__success
 __failure_unpriv
 __naked void uninitialized_stack2(void)
 {
tools/testing/selftests/bpf/progs/verifier_int_ptr.c +1 -1
index b054f9c48143..b5cedc0d23c1 100644
@@ -7,7 +7,7 @@
 
 SEC("cgroup/sysctl")
 __description("ARG_PTR_TO_LONG uninitialized")
-__failure __msg("invalid indirect read from stack R4 off -16+0 size 8")
+__success
 __naked void arg_ptr_to_long_uninitialized(void)
 {
 	asm volatile ("					\
tools/testing/selftests/bpf/progs/verifier_raw_stack.c +1 -1
index efbfc3a4ad6a..5468c5302495 100644
@@ -7,7 +7,7 @@
 
 SEC("tc")
 __description("raw_stack: no skb_load_bytes")
-__failure __msg("invalid read from stack R6 off=-8 size=8")
+__success
 __naked void stack_no_skb_load_bytes(void)
 {
 	asm volatile ("					\
tools/testing/selftests/bpf/progs/verifier_var_off.c +2 -2
index 83a90afba785..bbf3628c625a 100644
@@ -61,7 +61,7 @@ __naked void stack_read_priv_vs_unpriv(void)
 
 SEC("lwt_in")
 __description("variable-offset stack read, uninitialized")
-__failure __msg("invalid variable-offset read from stack R2")
+__success
 __naked void variable_offset_stack_read_uninitialized(void)
 {
 	asm volatile ("					\
@@ -255,7 +255,7 @@ __naked void access_min_out_of_bound(void)
 
 SEC("lwt_in")
 __description("indirect variable-offset stack access, min_off < min_initialized")
-__failure __msg("invalid indirect read from stack R2 var_off")
+__success
 __naked void access_min_off_min_initialized(void)
 {
 	asm volatile ("					\
tools/testing/selftests/bpf/verifier/atomic_cmpxchg.c +0 -11
index 319337bdcfc8..9a7b1106fda8 100644
@@ -83,17 +83,6 @@
 	.result = REJECT,
 	.errstr = "!read_ok",
 },
-{
-	"Can't use cmpxchg on uninit memory",
-	.insns = {
-		BPF_MOV64_IMM(BPF_REG_0, 3),
-		BPF_MOV64_IMM(BPF_REG_2, 4),
-		BPF_ATOMIC_OP(BPF_DW, BPF_CMPXCHG, BPF_REG_10, BPF_REG_2, -8),
-		BPF_EXIT_INSN(),
-	},
-	.result = REJECT,
-	.errstr = "invalid read from stack",
-},
 {
 	"BPF_W cmpxchg should zero top 32 bits",
 	.insns = {
tools/testing/selftests/bpf/verifier/calls.c +1 -1
index 3d5cd51071f0..89b79997c1b4 100644
@@ -1505,7 +1505,7 @@
 	.prog_type = BPF_PROG_TYPE_XDP,
 	.fixup_map_hash_8b = { 23 },
 	.result = REJECT,
-	.errstr = "invalid read from stack R7 off=-16 size=8",
+	.errstr = "R0 invalid mem access 'scalar'",
 },
 {
 	"calls: two calls that receive map_value via arg=ptr_stack_of_caller. test1",
Signature · 1 line
2.40.1
#9 lore raw

PATCH v2 2/2 bpf: new verifier tests for stack access

In reply to Andrei Matei

This patch adds tests for the previous patch, checking the tracking of the maximum stack size and checking that accesses to uninit stack memory are allowed.

They are a separate patch for review purposes; whoever merges them can consider squashing.

---
 tools/testing/selftests/bpf/test_verifier.c  | 24 ++++++++++++
 tools/testing/selftests/bpf/verifier/stack.c | 40 ++++++++++++++++++++
 2 files changed, 64 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/verifier/stack.c
tools/testing/selftests/bpf/test_verifier.c +24 -0
index 98107e0452d3..a62610585ee4 100644
@@ -135,6 +135,10 @@ struct bpf_test {
 	const char *errstr;
 	const char *errstr_unpriv;
 	uint32_t insn_processed;
+	/* Expected maximum stack depth for the main subprogram. Not checked if 0.
+	 * Only checked if the program is accepted.
+	 */
+	uint16_t max_stack_depth;
 	int prog_len;
 	enum {
 		UNDEF,
@@ -1703,6 +1707,26 @@ static void do_test_single(struct bpf_test *test, bool unpriv,
 		}
 	}
 
+	/* Check the stack size if the test configured an expecation and the program
+	 * was loaded successfully.
+	 */
+	if (test->max_stack_depth && fd_prog >= 0) {
+		uint32_t max_stack;
+		char *s;
+
+		s = strstr(bpf_vlog, "stack depth ");
+		if (s == NULL) {
+			printf("FAIL\nstack depth result not found in verifier output\n");
+			goto fail_log;
+		}
+		max_stack = atoi(s + 12);
+		if (test->max_stack_depth != max_stack) {
+			printf("FAIL\nUnexpected max stack %u vs %u\n",
+			       max_stack, test->max_stack_depth);
+			goto fail_log;
+		}
+	}
+
 	if (verbose)
 		printf(", verifier log:\n%s", bpf_vlog);
 
tools/testing/selftests/bpf/verifier/stack.c new file +40 -0
new file mode 100644
index 000000000000..ac571783c05e
@@ -0,0 +1,40 @@
+{
+	/* Check that reading unitialized stack memory is allowed only in privileged
+	 * mode. Also check that such reads maintain the max stack depth.
+	 */
+	"read uninit stack",
+	.insns = {
+		BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_10, -504),
+		/* exit(0); */
+		BPF_MOV32_IMM(BPF_REG_0, 0),
+		BPF_EXIT_INSN(),
+	},
+	.result = ACCEPT,
+	.result_unpriv = REJECT,
+	.errstr_unpriv = "invalid read from stack",
+    .max_stack_depth = 504,
+},
+{
+    /* Check that indirect accesses to stack maintain the max stack depth. */
+	"read (indirect) uninit stack",
+	.insns = {
+		/* We'll use probe_read_user as an arbitrary helper that can access the
+		 * stack. We're going to read into *(fp-104).
+		 */
+		BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
+		BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -104),
+		BPF_MOV32_IMM(BPF_REG_2, 8),
+        /* read from a random address */
+		BPF_MOV32_IMM(BPF_REG_3, 0x4242),
+        BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_probe_read_user),
+        BPF_JMP_IMM(BPF_JNE, BPF_REG_0, 0, 1),
+	    BPF_EXIT_INSN(),
+		/* exit(0); */
+		BPF_MOV32_IMM(BPF_REG_0, 0),
+		BPF_EXIT_INSN(),
+	},
+	.result = ACCEPT,
+	.result_unpriv = REJECT,
+	.errstr_unpriv = "",
+    .max_stack_depth = 104,
+},
\ No newline at end of file
Signature · 1 line
2.40.1
maintainer
#10 lore raw

Re: [PATCH bpf v2 2/2] bpf: new verifier tests for stack access

Quoted 9 lines · Andrei Matei

This patch adds tests for the previous patch, checking the tracking of the maximum stack size and checking that accesses to uninit stack memory are allowed.

They are a separate patch for review purposes; whoever merges them can consider squashing.

---

I think the strategy now is to add new tests using inline assembly, e.g. as a part of verifier_* tests in test_progs.

maintainer Acked-by on v2 1/2
#2 lore raw
In reply to Andrei Matei
Quoted 35 lines · Andrei Matei

Privileged programs are supposed to be able to read uninitialized stack memory (ever since 6715df8d5) but, before this patch, these accesses were permitted inconsistently. In particular, accesses were permitted above state->allocated_stack, but not below it. In other words, if the stack was already "large enough", the access was permitted, but otherwise the access was rejected instead of being allowed to "grow the stack". This undesired rejection was happening in two places:
- in check_stack_slot_within_bounds()
- in check_stack_range_initialized()
This patch arranges for these accesses to be permitted.

This patch also fixes the tracking of the stack size for variable-offset reads. This second fix is bundled in the same commit as the first one because they're inter-related. Before this patch, writes to the stack using registers containing a variable offset (as opposed to registers with fixed, known values) were not properly contributing to the function's needed stack size. As a result, it was possible for a program to verify, but then to attempt to read out-of-bounds data at runtime because a too small stack had been allocated for it.

Each function tracks the size of the stack it needs in bpf_subprog_info.stack_depth, which is maintained by update_stack_depth(). For regular memory accesses, check_mem_access() was calling update_state_depth() but it was passing in only the fixed part of the offset register, ignoring the variable offset. This was incorrect; the minimum possible value of that register should be used instead.

This tracking is now fixed by centralizing the tracking of stack size in grow_stack_state(), and by lifting the calls to grow_stack_state() to check_stack_access_within_bounds() as suggested by Andrii. The code is now simpler and more convincingly tracks the correct maximum stack size. check_stack_range_initialized() can now rely on enough stack having been allocated for the access; this helps with the fix for the first issue.

Quoted 5 lines

---

I think these changes make sense.
Question: would it be possible to recover some of the tests (those converted from failure to success) by changing execution mode from priv to unpriv?

Also, I think there are some tests that do oob stack read in branches that should be proven unreachable, with expectation that if certain verifier logic does not work as expected stack access would serve as a canary. Have no idea how to identify these tests, though.

[...]

kernel/bpf/verifier.c static int grow_stack_state(struct bpf_func_state *state, int size) · line 1704

Quoted 7 lines
kernel/bpf/verifier.c +3 -0
@@ -1697,6 +1699,12 @@ static int grow_stack_state(struct bpf_func_state *state, int size)
 		return -ENOMEM;
 
 	state->allocated_stack = size;
+
+	/* update known max for given subprogram */
+	u16 stack = env->subprog_info[state->subprogno].stack_depth;

Nit: 'u16 stack;' should be at the top of the function.

Quoted 5 lines
kernel/bpf/verifier.c +3 -0
+	if (stack < size)
+		env->subprog_info[state->subprogno].stack_depth = size;
+
 	return 0;
 }

[...]

maintainer
#3 lore raw
Quoted 1 line · Eduard Zingerman

I think these changes make sense.

(Under assumption that access to uninitialized stack should be allowed,
although I implemented the patch-set that made such behavior legal I still don't really like this idea).

#11 lore raw

Re: [PATCH bpf v2 2/2] bpf: new verifier tests for stack access

In reply to Eduard Zingerman
Quoted 6 lines · Eduard Zingerman
Quoted 4 lines · Andrei Matei

This patch adds tests for the previous patch, checking the tracking of the maximum stack size and checking that accesses to uninit stack memory are allowed.

Quoted 8 lines
Quoted 5 lines

They are a separate patch for review purposes; whoever merges them can consider squashing.

---

I think the strategy now is to add new tests using inline assembly, e.g. as a part of verifier_* tests in test_progs.

Thanks, I'll try that. I see in some of the verifier tests that you have converted to test_progs hints that they were converted "automatically". I'm curious what tool you've used, if any. Do you have any thoughts on how a test could assert the maximum stack depth? test_verifier has access to the verifier verifier log and parses it out of there; do you know how I could achieve the same in test_progs?

For the curiosity of someone who doesn't know much about this code base - how come we moved from test_verifier, which seems a bit more unit-test-y, do the higher level test_progs? Is it because of the nicer assembly syntax?

maintainer
#12 lore raw

Re: [PATCH bpf v2 2/2] bpf: new verifier tests for stack access

Quoted 13 lines · Andrei Matei
Quoted 12 lines · Eduard Zingerman
Quoted 9 lines · Andrei Matei

This patch adds tests for the previous patch, checking the tracking of the maximum stack size and checking that accesses to uninit stack memory are allowed.

They are a separate patch for review purposes; whoever merges them can consider squashing.

---

Quoted 6 lines
Quoted 2 lines

I think the strategy now is to add new tests using inline assembly, e.g. as a part of verifier_* tests in test_progs.

Thanks, I'll try that. I see in some of the verifier tests that you have converted to test_progs hints that they were converted "automatically". I'm curious what tool you've used, if any.

I wrote a small tree-sitter based tool for this:
https://github.com/eddyz87/verifier-tests-migrator

Quoted 4 lines

Do you have any thoughts on how a test could assert the maximum stack depth? test_verifier has access to the verifier verifier log and parses it out of there; do you know how I could achieve the same in test_progs?

Could be done like in the patch below (set log level so that stack depth is printed, match stack depth message):

tools/testing/selftests/bpf/progs/verifier_basic_stack.c +1 -1
index 069c3f91705c..e85ac95c8dd3 100644
@@ -27,7 +27,7 @@ __naked void stack_out_of_bounds(void)
 
 SEC("socket")
 __description("uninitialized stack1")
-__success
+__success __log_level(4) __msg("stack depth 8")
 __failure_unpriv __msg_unpriv("invalid indirect read from stack")
 __naked void uninitialized_stack1(void)
 {
Quoted 4 lines

For the curiosity of someone who doesn't know much about this code base - how come we moved from test_verifier, which seems a bit more unit-test-y, do the higher level test_progs? Is it because of the nicer assembly syntax?

Yes, ability to use assembly syntax is the main (sole?) driver. In fact, tests that use annotations from bpf_misc.h and are registered in tools/testing/selftests/bpf/prog_tests/verifier.c provide almost the same functionality as test_verifier:
- interface:
- select test to run using filter, e.g.:
./test_progs -a 'verifier_basic_stack/uninitialized stack1' ./test_progs -a 'verifier_basic_stack/uninitialized stack1 @unpriv'
- see verifier log for the test, e.g.:
./test_progs -vvv -a 'verifier_basic_stack/uninitialized stack1'
- test expectations:
- use __success / __failure to mark expected test outcome;
- use __msg to search for expected messages in the log;
- use __log_level to specify log level when program is loaded;
- use __flags to enable additional knobs, e.g. BPF_F_TEST_STATE_FREQ;
- use __retval if test program has to be executed;
- use _unpriv suffix to specify any of the above for when test is executed in unprivileged mode.
See comments in tools/testing/selftests/bpf/progs/bpf_misc.h for details.
- plus clang generates BTF and CO-RE relocations for us,
with test_verifier one has to write these things "by hand" (which is only truly needed in tests for CO-RE/BTF).

The only drawback is compilation time, because all test_progs/*.c files depend on all *.bpf.o files. Locally I mitigate this by using
ccache: make CC='ccache clang' LLVM=1 -j14 test_progs .
I probably need to look at what could be done in the makefile one day.

Unfortunately neither test_verifier nor test_progs are true unit tests.

maintainer
#4 lore raw
In reply to Eduard Zingerman

On Tue, 2023-11-28 at 03:33 +0200, Eduard Zingerman wrote:
[...]

Quoted 4 lines

Also, I think there are some tests that do oob stack read in branches that should be proven unreachable, with expectation that if certain verifier logic does not work as expected stack access would serve as a canary. Have no idea how to identify these tests, though.

I looked through all test cases I ever added (not so many) and it seems that only one test case should be updated:

tools/testing/selftests/bpf/progs/iters.c +1 -1
index b2181f850d3e..3aca3dc145b5 100644
@@ -846,7 +846,7 @@ __naked int delayed_precision_mark(void)
                "call %[bpf_iter_num_next];"
                "if r0 == 0 goto 2f;"
                "if r6 != 42 goto 3f;"
-               "r7 = -32;"
+               "r7 = -33;"
                "call %[bpf_get_prandom_u32];"
                "r6 = r0;"
                "goto 1b;\n"

Here oob access is replaced by unaligned, this does not affect error message, but should be future proof in case if widening logic would get smarter.

#5 lore raw
In reply to Andrei Matei
Quoted 72 lines · Andrei Matei

Privileged programs are supposed to be able to read uninitialized stack memory (ever since 6715df8d5) but, before this patch, these accesses were permitted inconsistently. In particular, accesses were permitted above state->allocated_stack, but not below it. In other words, if the stack was already "large enough", the access was permitted, but otherwise the access was rejected instead of being allowed to "grow the stack". This undesired rejection was happening in two places:
- in check_stack_slot_within_bounds()
- in check_stack_range_initialized()
This patch arranges for these accesses to be permitted.

This patch also fixes the tracking of the stack size for variable-offset reads. This second fix is bundled in the same commit as the first one because they're inter-related. Before this patch, writes to the stack using registers containing a variable offset (as opposed to registers with fixed, known values) were not properly contributing to the function's needed stack size. As a result, it was possible for a program to verify, but then to attempt to read out-of-bounds data at runtime because a too small stack had been allocated for it.

Each function tracks the size of the stack it needs in bpf_subprog_info.stack_depth, which is maintained by update_stack_depth(). For regular memory accesses, check_mem_access() was calling update_state_depth() but it was passing in only the fixed part of the offset register, ignoring the variable offset. This was incorrect; the minimum possible value of that register should be used instead.

This tracking is now fixed by centralizing the tracking of stack size in grow_stack_state(), and by lifting the calls to grow_stack_state() to check_stack_access_within_bounds() as suggested by Andrii. The code is now simpler and more convincingly tracks the correct maximum stack size. check_stack_range_initialized() can now rely on enough stack having been allocated for the access; this helps with the fix for the first issue.

---
 include/linux/bpf_verifier.h                  |  4 ++
 kernel/bpf/verifier.c                         | 70 ++++++++-----------
 .../selftests/bpf/progs/test_global_func16.c  |  2 +-
 .../bpf/progs/verifier_basic_stack.c          |  6 +-
 .../selftests/bpf/progs/verifier_int_ptr.c    |  2 +-
 .../selftests/bpf/progs/verifier_raw_stack.c  |  2 +-
 .../selftests/bpf/progs/verifier_var_off.c    |  4 +-
 .../selftests/bpf/verifier/atomic_cmpxchg.c   | 11 ---
 tools/testing/selftests/bpf/verifier/calls.c  |  2 +-
 9 files changed, 42 insertions(+), 61 deletions(-)
include/linux/bpf_verifier.h +4 -0
index aa4d19d0bc94..5fc389e8be35 100644
@@ -630,6 +630,10 @@ struct bpf_verifier_env {
        int exception_callback_subprog;
        bool explore_alu_limits;
        bool allow_ptr_leaks;
+       /* Allow access to uninitialized stack memory. Writes with fixed offset are
+        * always allowed, so this refers to reads (with fixed or variable offset),
+        * to writes with variable offset and to indirect (helper) accesses.
+        */
        bool allow_uninit_stack;
        bool bpf_capable;
        bool bypass_spec_v1;
kernel/bpf/verifier.c +0 -0
index af2819d5c8ee..f9546dd73f3c 100644

kernel/bpf/verifier.c static int resize_reference_state(struct bpf_func_state *state, size_t n) · line 1693

Quoted 10 lines
kernel/bpf/verifier.c +4 -1
@@ -1685,10 +1685,12 @@ static int resize_reference_state(struct bpf_func_state *state, size_t n)
        return 0;
 }
 
-static int grow_stack_state(struct bpf_func_state *state, int size)
+/* Possibly update state->allocated_stack to be at least size bytes. Also
+ * possibly update the function's high-water mark in its bpf_subprog_info.
+ */
+static int grow_stack_state(struct bpf_verifier_env *env, struct bpf_func_state *state, int size)
 {
        size_t old_n = state->allocated_stack / BPF_REG_SIZE, n = size / BPF_REG_SIZE;

shouldn't this be rounding up? (size + BPF_REG_SIZE - 1) / BPF_REG_SIZE?

kernel/bpf/verifier.c static int resize_reference_state(struct bpf_func_state *state, size_t n) · line 1691

Quoted 1 line
kernel/bpf/verifier.c +0 -1
-

According to kernel code style, there should be an empty line between variable declaration and other statements. Please keep this empty line.

Quoted 3 lines
kernel/bpf/verifier.c +0 -0
        if (old_n >= n)
                return 0;
 

[...]

kernel/bpf/verifier.c static int check_ptr_to_map_access(struct bpf_verifier_env *env, · line 6755

Quoted 12 lines
kernel/bpf/verifier.c +5 -3
@@ -6761,13 +6748,15 @@ static int check_ptr_to_map_access(struct bpf_verifier_env *env,
  * The minimum valid offset is -MAX_BPF_STACK for writes, and
  * -state->allocated_stack for reads.
  */
-static int check_stack_slot_within_bounds(int off,
-                                         struct bpf_func_state *state,
-                                         enum bpf_access_type t)
+static int check_stack_slot_within_bounds(
+               struct bpf_verifier_env *env,
+               int off,
+               struct bpf_func_state *state,
+               enum bpf_access_type t)

nit: please keep code style, first argument is normally on the same line as opening parenthesis

Quoted 8 lines
kernel/bpf/verifier.c +1 -1
 {
        int min_valid_off;
 
-       if (t == BPF_WRITE)
+       if (t == BPF_WRITE || env->allow_uninit_stack)
                min_valid_off = -MAX_BPF_STACK;
        else
                min_valid_off = -state->allocated_stack;

[...]

Quoted 14 lines
tools/testing/selftests/bpf/progs/verifier_var_off.c +1 -1
index 83a90afba785..bbf3628c625a 100644
@@ -61,7 +61,7 @@ __naked void stack_read_priv_vs_unpriv(void)
 
 SEC("lwt_in")
 __description("variable-offset stack read, uninitialized")
-__failure __msg("invalid variable-offset read from stack R2")
+__success
 __naked void variable_offset_stack_read_uninitialized(void)
 {
        asm volatile ("                                 \

tools/testing/selftests/bpf/progs/verifier_var_off.c __naked void access_min_out_of_bound(void) · line 258

Quoted 5 lines
tools/testing/selftests/bpf/progs/verifier_var_off.c +1 -1
@@ -255,7 +255,7 @@ __naked void access_min_out_of_bound(void)
 
 SEC("lwt_in")
 __description("indirect variable-offset stack access, min_off < min_initialized")
-__failure __msg("invalid indirect read from stack R2 var_off")
+__success

as Eduard mentioned, can you please try updating program type to something that is allowed in unprivileged mode, e.g., SEC("socket") and make sure that it still fails in unpriv mode

Quoted 3 lines
tools/testing/selftests/bpf/progs/verifier_var_off.c +0 -0
 __naked void access_min_off_min_initialized(void)
 {
        asm volatile ("                                 \

[...]

#13 lore raw

Re: [PATCH bpf v2 2/2] bpf: new verifier tests for stack access

In reply to Eduard Zingerman
Quoted 47 lines · Eduard Zingerman
Quoted 13 lines · Andrei Matei
Quoted 12 lines · Eduard Zingerman
Quoted 9 lines · Andrei Matei

This patch adds tests for the previous patch, checking the tracking of the maximum stack size and checking that accesses to uninit stack memory are allowed.

They are a separate patch for review purposes; whoever merges them can consider squashing.

---

Quoted 6 lines
Quoted 2 lines

I think the strategy now is to add new tests using inline assembly, e.g. as a part of verifier_* tests in test_progs.

Thanks, I'll try that. I see in some of the verifier tests that you have converted to test_progs hints that they were converted "automatically". I'm curious what tool you've used, if any.

I wrote a small tree-sitter based tool for this:
https://github.com/eddyz87/verifier-tests-migrator

Quoted 4 lines

Do you have any thoughts on how a test could assert the maximum stack depth? test_verifier has access to the verifier verifier log and parses it out of there; do you know how I could achieve the same in test_progs?

Could be done like in the patch below (set log level so that stack depth is printed, match stack depth message):

tools/testing/selftests/bpf/progs/verifier_basic_stack.c +1 -1
index 069c3f91705c..e85ac95c8dd3 100644
@@ -27,7 +27,7 @@ __naked void stack_out_of_bounds(void)
 
 SEC("socket")
 __description("uninitialized stack1")
-__success
+__success __log_level(4) __msg("stack depth 8")
 __failure_unpriv __msg_unpriv("invalid indirect read from stack")
 __naked void uninitialized_stack1(void)
 {
Quoted 6 lines
Quoted 4 lines

For the curiosity of someone who doesn't know much about this code base - how come we moved from test_verifier, which seems a bit more unit-test-y, do the higher level test_progs? Is it because of the nicer assembly syntax?

Yes, ability to use assembly syntax is the main (sole?) driver.

Not just that, test_progs by itself is a much nicer test runner for debugging tests. It's easier to get log_level 2 logs, for example. It's much easier to select a subset of tests to run. Plus by having these BPF programs compiled as stand-alone .bpf.o objects, we can use veristat to test them, bypassing test_progs altogether.

Quoted 22 lines

In fact, tests that use annotations from bpf_misc.h and are registered in tools/testing/selftests/bpf/prog_tests/verifier.c provide almost the same functionality as test_verifier:
- interface:
- select test to run using filter, e.g.:
./test_progs -a 'verifier_basic_stack/uninitialized stack1' ./test_progs -a 'verifier_basic_stack/uninitialized stack1 @unpriv'
- see verifier log for the test, e.g.:
./test_progs -vvv -a 'verifier_basic_stack/uninitialized stack1'
- test expectations:
- use __success / __failure to mark expected test outcome;
- use __msg to search for expected messages in the log;
- use __log_level to specify log level when program is loaded;
- use __flags to enable additional knobs, e.g. BPF_F_TEST_STATE_FREQ;
- use __retval if test program has to be executed;
- use _unpriv suffix to specify any of the above for when test is executed in unprivileged mode.
See comments in tools/testing/selftests/bpf/progs/bpf_misc.h for details.
- plus clang generates BTF and CO-RE relocations for us,
with test_verifier one has to write these things "by hand" (which is only truly needed in tests for CO-RE/BTF).

Quoted 4 lines

The only drawback is compilation time, because all test_progs/*.c files depend on all *.bpf.o files. Locally I mitigate this by using
ccache: make CC='ccache clang' LLVM=1 -j14 test_progs .
I probably need to look at what could be done in the makefile one day.

I think if we make a push to use BPF skeletons in all tests, we should be able to get proper dependencies between prog_tests/xxx.c and its corresponding progs/yyy.bpf.o file(s). And at which point we should be able to teach Makefile to only recompile relevant test files.

Quoted 2 lines

Unfortunately neither test_verifier nor test_progs are true unit tests.

#6 lore raw
In reply to Andrii Nakryiko

[...]

Quoted 8 lines
Quoted 8 lines
kernel/bpf/verifier.c +0 -0
index af2819d5c8ee..f9546dd73f3c 100644
@@ -1685,10 +1685,12 @@ static int resize_reference_state(struct bpf_func_state *state, size_t n)
        return 0;
 }
 
Quoted 9 lines

kernel/bpf/verifier.c static int resize_reference_state(struct bpf_func_state *state, size_t n) · line 1693

Quoted 7 lines
kernel/bpf/verifier.c +4 -1
-static int grow_stack_state(struct bpf_func_state *state, int size)
+/* Possibly update state->allocated_stack to be at least size bytes. Also
+ * possibly update the function's high-water mark in its bpf_subprog_info.
+ */
+static int grow_stack_state(struct bpf_verifier_env *env, struct bpf_func_state *state, int size)
 {
        size_t old_n = state->allocated_stack / BPF_REG_SIZE, n = size / BPF_REG_SIZE;

shouldn't this be rounding up? (size + BPF_REG_SIZE - 1) / BPF_REG_SIZE?

You're saying this was always broken, regardless of the current patch, right? I think you're right, but that seems like a bug that should have been caught somehow; I'm surprised no programs crashed the verifier. Perhaps in practice all stack accesses are 8-byte aligned, so the rounding doesn't matter?

I'll spend a bit of time reading code and come back.

[...]

#7 lore raw
Quoted 21 lines · Andrei Matei

[...]

Quoted 8 lines
Quoted 8 lines
kernel/bpf/verifier.c +0 -0
index af2819d5c8ee..f9546dd73f3c 100644
@@ -1685,10 +1685,12 @@ static int resize_reference_state(struct bpf_func_state *state, size_t n)
        return 0;
 }
 
Quoted 9 lines

kernel/bpf/verifier.c static int resize_reference_state(struct bpf_func_state *state, size_t n) · line 1693

Quoted 7 lines
kernel/bpf/verifier.c +4 -1
-static int grow_stack_state(struct bpf_func_state *state, int size)
+/* Possibly update state->allocated_stack to be at least size bytes. Also
+ * possibly update the function's high-water mark in its bpf_subprog_info.
+ */
+static int grow_stack_state(struct bpf_verifier_env *env, struct bpf_func_state *state, int size)
 {
        size_t old_n = state->allocated_stack / BPF_REG_SIZE, n = size / BPF_REG_SIZE;

shouldn't this be rounding up? (size + BPF_REG_SIZE - 1) / BPF_REG_SIZE?

Quoted 1 line

You're saying this was always broken, regardless of the current patch, right? I

I think so, yes...

Quoted 5 lines

think you're right, but that seems like a bug that should have been caught somehow; I'm surprised no programs crashed the verifier. Perhaps in practice all stack accesses are 8-byte aligned, so the rounding doesn't matter?

I'll spend a bit of time reading code and come back.

Thanks!

Quoted 2 lines

[...]

#8 lore raw
Quoted 23 lines · Andrii Nakryiko
Quoted 21 lines · Andrei Matei

[...]

Quoted 17 lines
Quoted 5 lines
kernel/bpf/verifier.c +0 -0
index af2819d5c8ee..f9546dd73f3c 100644

kernel/bpf/verifier.c static int resize_reference_state(struct bpf_func_state *state, size_t n) · line 1693

Quoted 10 lines
kernel/bpf/verifier.c +4 -1
@@ -1685,10 +1685,12 @@ static int resize_reference_state(struct bpf_func_state *state, size_t n)
        return 0;
 }
 
-static int grow_stack_state(struct bpf_func_state *state, int size)
+/* Possibly update state->allocated_stack to be at least size bytes. Also
+ * possibly update the function's high-water mark in its bpf_subprog_info.
+ */
+static int grow_stack_state(struct bpf_verifier_env *env, struct bpf_func_state *state, int size)
 {
        size_t old_n = state->allocated_stack / BPF_REG_SIZE, n = size / BPF_REG_SIZE;

shouldn't this be rounding up? (size + BPF_REG_SIZE - 1) / BPF_REG_SIZE?

Quoted 3 lines
Quoted 1 line

You're saying this was always broken, regardless of the current patch, right? I

I think so, yes...

I believe that this code was OK after all because all the callers to grow_stack_state() do the rounding. This seems fragile though; I'll include a patch to push the rounding inside grow_stack_state() if it's OK to you.

While reading the code around how the stack is accessed, something else caught my eye. I'm not sure if there's a problem or not; perhaps you could take a look. Around here in check_stack_write_fixed_off() [1], the code that deals with register spills has a couple of cases, including:

[1] https://github.com/torvalds/linux/blob/815fb87b753055df2d9e50f6cd80eb10235fe3e9/kernel/bpf/verifier.c#L4733-L4744
} else if (reg && is_spillable_regtype(reg->type)) {
/* register containing pointer is being spilled into stack */

  if (size != BPF_REG_SIZE) {
    verbose_linfo(env, insn_idx, "; ");
    verbose(env, "invalid size of register spill\n");
    return -EACCES;
  }
  if (state != cur && reg->type == PTR_TO_STACK) {
    verbose(env, "cannot spill pointers to stack into stack frame of
the caller\n");
    return -EINVAL;

}
save_register_state(state, spi, reg, size);
}

This branch, as opposed to all the others, calls save_register_state() without checking that !(off % BPF_REG_SIZE). I believe save_register_spill() implicitly assumes that the access is aligned, so it'd be bad if off was not aligned. Is it possible for the offset to not be aligned? I'm not sure... The higher-level check_mem_access() starts with a check_ptr_alignment(), which I think does the required check at least on the code paths that I've tried. But then why do all the other branches in check_stack_write_fixed_off() check for the alignment explicitly? FWIW, the branch in question also has a is_spillable_regtype() check which is perhaps relevant.

Would an assertion that off is indeed aligned in the branch I'm talking about (or elsewhere around) be welcomed?

Apologies if I'm paranoid for no reason.

Quoted 11 lines
Quoted 6 lines

think you're right, but that seems like a bug that should have been caught somehow; I'm surprised no programs crashed the verifier. Perhaps in practice all stack accesses are 8-byte aligned, so the rounding doesn't matter?

I'll spend a bit of time reading code and come back.

Thanks!

Quoted 2 lines

[...]