Skip to contentSkip to conversation

bpf: Do more tight ALU bounds tracking

PATCH In review Kuee K1r0a 3 participants 14 messages Jul 29, 2022 – Aug 30, 2022 1 message not in the archive

Subsystems (4)
lore
Share & tools
lore

Kuee K1r0a

author
#0 lore raw

PATCH 1/1 bpf: Do more tight ALU bounds tracking

32bit bounds and 64bit bounds are updated separately in adjust_scalar_min_max_vals() currently, let them learn from each other to get more tight bounds tracking. Similar operation can be found in reg_set_min_max().

Before:

    func#0 @0
    0: R1=ctx(off=0,imm=0) R10=fp0
    0: (b7) r0 = 0                        ; R0_w=0
    1: (b7) r1 = 0                        ; R1_w=0
    2: (87) r1 = -r1                      ; R1_w=scalar()
    3: (87) r1 = -r1                      ; R1_w=scalar()
    4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
    5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0xffffffff))  <--- [*]
    6: (95) exit

It can be seen that even if the 64bit bounds is clear here, the 32bit bounds is still in the state of 'UNKNOWN'.

After:

    func#0 @0
    0: R1=ctx(off=0,imm=0) R10=fp0
    0: (b7) r0 = 0                        ; R0_w=0
    1: (b7) r1 = 0                        ; R1_w=0
    2: (87) r1 = -r1                      ; R1_w=scalar()
    3: (87) r1 = -r1                      ; R1_w=scalar()
    4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
    5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0x3))  <--- [*]
    6: (95) exit
Fixes 3f50f132d840 ("bpf: Verifier, do explicit ALU32 bounds tracking")
---
 kernel/bpf/verifier.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
kernel/bpf/verifier.c +6 -3
index 0efbac0fd126..888aa50fbdc0 100644
@@ -8934,10 +8934,13 @@ static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env,
 		break;
 	}
 
-	/* ALU32 ops are zero extended into 64bit register */
-	if (alu32)
+	if (alu32) {
+		/* ALU32 ops are zero extended into 64bit register */
 		zext_32_to_64(dst_reg);
-	reg_bounds_sync(dst_reg);
+		__reg_combine_32_into_64(dst_reg);
+	} else {
+		__reg_combine_64_into_32(dst_reg);
+	}
 	return 0;
 }
 
Signature · 1 line
2.25.1
#1 lore raw

Hi,

Quoted 33 lines · On Thu, Jul 28, 2022 at 8:31 PM Kuee K1r0a <[email protected]> wrote:

32bit bounds and 64bit bounds are updated separately in adjust_scalar_min_max_vals() currently, let them learn from each other to get more tight bounds tracking. Similar operation can be found in reg_set_min_max().

Before:

    func#0 @0
    0: R1=ctx(off=0,imm=0) R10=fp0
    0: (b7) r0 = 0                        ; R0_w=0
    1: (b7) r1 = 0                        ; R1_w=0
    2: (87) r1 = -r1                      ; R1_w=scalar()
    3: (87) r1 = -r1                      ; R1_w=scalar()
    4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
    5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0xffffffff))  <--- [*]
    6: (95) exit

It can be seen that even if the 64bit bounds is clear here, the 32bit bounds is still in the state of 'UNKNOWN'.

After:

    func#0 @0
    0: R1=ctx(off=0,imm=0) R10=fp0
    0: (b7) r0 = 0                        ; R0_w=0
    1: (b7) r1 = 0                        ; R1_w=0
    2: (87) r1 = -r1                      ; R1_w=scalar()
    3: (87) r1 = -r1                      ; R1_w=scalar()
    4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
    5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0x3))  <--- [*]
    6: (95) exit
Quoted 2 lines
Fixes 3f50f132d840 ("bpf: Verifier, do explicit ALU32 bounds tracking")

Please sign with your real name. Thanks.

Quoted 28 lines
---
 kernel/bpf/verifier.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
kernel/bpf/verifier.c +6 -4
index 0efbac0fd126..888aa50fbdc0 100644
@@ -8934,10 +8934,13 @@ static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env,
                break;
        }
 
-       /* ALU32 ops are zero extended into 64bit register */
-       if (alu32)
+       if (alu32) {
+               /* ALU32 ops are zero extended into 64bit register */
                zext_32_to_64(dst_reg);
-       reg_bounds_sync(dst_reg);
+               __reg_combine_32_into_64(dst_reg);
+       } else {
+               __reg_combine_64_into_32(dst_reg);
+       }
        return 0;
 }
 
--

2.25.1

author
#2 lore raw

PATCH 1/1 bpf: Do more tight ALU bounds tracking

32bit bounds and 64bit bounds are updated separately in adjust_scalar_min_max_vals() currently, let them learn from each other to get more tight bounds tracking. Similar operation can be found in reg_set_min_max().

Before:

    func#0 @0
    0: R1=ctx(off=0,imm=0) R10=fp0
    0: (b7) r0 = 0                        ; R0_w=0
    1: (b7) r1 = 0                        ; R1_w=0
    2: (87) r1 = -r1                      ; R1_w=scalar()
    3: (87) r1 = -r1                      ; R1_w=scalar()
    4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
    5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0xffffffff))  <--- [*]
    6: (95) exit

It can be seen that even if the 64bit bounds is clear here, the 32bit bounds is still in the state of 'UNKNOWN'.

After:

    func#0 @0
    0: R1=ctx(off=0,imm=0) R10=fp0
    0: (b7) r0 = 0                        ; R0_w=0
    1: (b7) r1 = 0                        ; R1_w=0
    2: (87) r1 = -r1                      ; R1_w=scalar()
    3: (87) r1 = -r1                      ; R1_w=scalar()
    4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
    5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0x3))  <--- [*]
    6: (95) exit
Fixes 3f50f132d840 ("bpf: Verifier, do explicit ALU32 bounds tracking")
---
 kernel/bpf/verifier.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
kernel/bpf/verifier.c +6 -3
index 0efbac0fd126..888aa50fbdc0 100644
@@ -8934,10 +8934,13 @@ static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env,
 		break;
 	}
 
-	/* ALU32 ops are zero extended into 64bit register */
-	if (alu32)
+	if (alu32) {
+		/* ALU32 ops are zero extended into 64bit register */
 		zext_32_to_64(dst_reg);
-	reg_bounds_sync(dst_reg);
+		__reg_combine_32_into_64(dst_reg);
+	} else {
+		__reg_combine_64_into_32(dst_reg);
+	}
 	return 0;
 }
 
Signature · 1 line
2.25.1
#3 lore raw

Hi Youlin,

Quoted 33 lines · Youlin Li

32bit bounds and 64bit bounds are updated separately in adjust_scalar_min_max_vals() currently, let them learn from each other to get more tight bounds tracking. Similar operation can be found in reg_set_min_max().

Before:

    func#0 @0
    0: R1=ctx(off=0,imm=0) R10=fp0
    0: (b7) r0 = 0                        ; R0_w=0
    1: (b7) r1 = 0                        ; R1_w=0
    2: (87) r1 = -r1                      ; R1_w=scalar()
    3: (87) r1 = -r1                      ; R1_w=scalar()
    4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
    5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0xffffffff))  <--- [*]
    6: (95) exit

It can be seen that even if the 64bit bounds is clear here, the 32bit bounds is still in the state of 'UNKNOWN'.

After:

    func#0 @0
    0: R1=ctx(off=0,imm=0) R10=fp0
    0: (b7) r0 = 0                        ; R0_w=0
    1: (b7) r1 = 0                        ; R1_w=0
    2: (87) r1 = -r1                      ; R1_w=scalar()
    3: (87) r1 = -r1                      ; R1_w=scalar()
    4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
    5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0x3))  <--- [*]
    6: (95) exit
Quoted 1 line
Fixes 3f50f132d840 ("bpf: Verifier, do explicit ALU32 bounds tracking")

This change looks to me like an improvement, rather than a bug fix. We probably don't need this tag.

Quoted 10 lines
---
 kernel/bpf/verifier.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
kernel/bpf/verifier.c +0 -0
index 0efbac0fd126..888aa50fbdc0 100644

kernel/bpf/verifier.c static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env, · line 8940

Quoted 10 lines
kernel/bpf/verifier.c +3 -3
@@ -8934,10 +8934,13 @@ static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env,
                break;
        }
 
-       /* ALU32 ops are zero extended into 64bit register */
-       if (alu32)
+       if (alu32) {
+               /* ALU32 ops are zero extended into 64bit register */
                zext_32_to_64(dst_reg);
-       reg_bounds_sync(dst_reg);
+               __reg_combine_32_into_64(dst_reg);

This __reg_combine_32_into_64 can be replaced with simply reg_bounds_sync, because the above zext_32_to_64 has already propagated 32 to 64. Using reg_bounds_sync would be more efficient.

It turns out we can now fold reg_bounds_sync into zext_32_to_64. Can you do that and resend? IMO that will make the code slightly cleaner.

Quoted 9 lines
kernel/bpf/verifier.c +3 -1
+       } else {
+               __reg_combine_64_into_32(dst_reg);
+       }
        return 0;
 }
 
--

2.25.1

author
#4 lore raw

PATCH 1/1 bpf: Do more tight ALU bounds tracking

In adjust_scalar_min_max_vals(), let 32bit bounds learn from 64bit bounds to get more tight bounds tracking. Similar operation can be found in reg_set_min_max().

Also, we can now fold reg_bounds_sync() into zext_32_to_64().

Before:

    func#0 @0
    0: R1=ctx(off=0,imm=0) R10=fp0
    0: (b7) r0 = 0                        ; R0_w=0
    1: (b7) r1 = 0                        ; R1_w=0
    2: (87) r1 = -r1                      ; R1_w=scalar()
    3: (87) r1 = -r1                      ; R1_w=scalar()
    4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
    5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0xffffffff))  <--- [*]
    6: (95) exit

It can be seen that even if the 64bit bounds is clear here, the 32bit bounds is still in the state of 'UNKNOWN'.

After:

    func#0 @0
    0: R1=ctx(off=0,imm=0) R10=fp0
    0: (b7) r0 = 0                        ; R0_w=0
    1: (b7) r1 = 0                        ; R1_w=0
    2: (87) r1 = -r1                      ; R1_w=scalar()
    3: (87) r1 = -r1                      ; R1_w=scalar()
    4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
    5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0x3))  <--- [*]
    6: (95) exit
---
 kernel/bpf/verifier.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)
kernel/bpf/verifier.c +6 -4
index 0efbac0fd126..1f5c6e3634d6 100644
@@ -4383,6 +4383,7 @@ static void zext_32_to_64(struct bpf_reg_state *reg)
 {
 	reg->var_off = tnum_subreg(reg->var_off);
 	__reg_assign_32_into_64(reg);
+	reg_bounds_sync(reg);
 }
 
 /* truncate register to smaller size (in bytes)
@@ -8934,10 +8935,12 @@ static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env,
 		break;
 	}
 
-	/* ALU32 ops are zero extended into 64bit register */
-	if (alu32)
+	if (alu32) {
+		/* ALU32 ops are zero extended into 64bit register */
 		zext_32_to_64(dst_reg);
-	reg_bounds_sync(dst_reg);
+	} else {
+		__reg_combine_64_into_32(dst_reg);
+	}
 	return 0;
 }
 
@@ -9126,7 +9129,6 @@ static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
 							 insn->dst_reg);
 				}
 				zext_32_to_64(dst_reg);
-				reg_bounds_sync(dst_reg);
 			}
 		} else {
 			/* case: R = imm
Signature · 1 line
2.25.1
Acked-by on v1 1/1
#5 lore raw
Quoted 34 lines · Youlin Li

In adjust_scalar_min_max_vals(), let 32bit bounds learn from 64bit bounds to get more tight bounds tracking. Similar operation can be found in reg_set_min_max().

Also, we can now fold reg_bounds_sync() into zext_32_to_64().

Before:

    func#0 @0
    0: R1=ctx(off=0,imm=0) R10=fp0
    0: (b7) r0 = 0                        ; R0_w=0
    1: (b7) r1 = 0                        ; R1_w=0
    2: (87) r1 = -r1                      ; R1_w=scalar()
    3: (87) r1 = -r1                      ; R1_w=scalar()
    4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
    5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0xffffffff))  <--- [*]
    6: (95) exit

It can be seen that even if the 64bit bounds is clear here, the 32bit bounds is still in the state of 'UNKNOWN'.

After:

    func#0 @0
    0: R1=ctx(off=0,imm=0) R10=fp0
    0: (b7) r0 = 0                        ; R0_w=0
    1: (b7) r1 = 0                        ; R1_w=0
    2: (87) r1 = -r1                      ; R1_w=scalar()
    3: (87) r1 = -r1                      ; R1_w=scalar()
    4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
    5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0x3))  <--- [*]
    6: (95) exit
Quoted 1 line

Looks good to me. Thanks Youlin.

Hao

maintainer
#6 lore raw
Quoted 35 lines · Hao Luo
Quoted 34 lines · Youlin Li

In adjust_scalar_min_max_vals(), let 32bit bounds learn from 64bit bounds to get more tight bounds tracking. Similar operation can be found in reg_set_min_max().

Also, we can now fold reg_bounds_sync() into zext_32_to_64().

Before:

     func#0 @0
     0: R1=ctx(off=0,imm=0) R10=fp0
     0: (b7) r0 = 0                        ; R0_w=0
     1: (b7) r1 = 0                        ; R1_w=0
     2: (87) r1 = -r1                      ; R1_w=scalar()
     3: (87) r1 = -r1                      ; R1_w=scalar()
     4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
     5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0xffffffff))  <--- [*]
     6: (95) exit

It can be seen that even if the 64bit bounds is clear here, the 32bit bounds is still in the state of 'UNKNOWN'.

After:

     func#0 @0
     0: R1=ctx(off=0,imm=0) R10=fp0
     0: (b7) r0 = 0                        ; R0_w=0
     1: (b7) r1 = 0                        ; R1_w=0
     2: (87) r1 = -r1                      ; R1_w=scalar()
     3: (87) r1 = -r1                      ; R1_w=scalar()
     4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
     5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0x3))  <--- [*]
     6: (95) exit
Quoted 5 lines
Quoted 1 line

Looks good to me. Thanks Youlin.

Thanks Youlin! Looks like the patch breaks CI [0] e.g.:

#142/p bounds check after truncation of non-boundary-crossing range FAIL
Failed to load prog 'Permission denied'!
invalid access to map value, value_size=8 off=16777215 size=1
R0 max value is outside of the allowed memory range
verification time 296 usec
stack depth 8
processed 15 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0

Please take a look. Also it would be great to add a test_verifier selftest to assert above case from commit log against future changes.

Thanks,
Daniel

[0] https://github.com/kernel-patches/bpf/runs/7696324041?check_suite_focus=true

A message that is not in the archive (sent off-list or not yet received): CANdZH3U7axKg6zDY+iswF2d1fBYY1Xo2jeVsbgMYMoJfd1AYJg@mail.gmail.com

author
#8 lore raw

PATCH 1/1 bpf: Do more tight ALU bounds tracking

---------- Forwarded message ---------
From: Kuee k1r0a <[email protected]>
Date: Mon, Aug 8, 2022 at 11:11 PM
Subject: Re: [PATCH bpf] bpf: Do more tight ALU bounds tracking
To: Daniel Borkmann <[email protected]>

Quoted 53 lines · Daniel Borkmann
Quoted 35 lines · Hao Luo
Quoted 34 lines · Youlin Li

In adjust_scalar_min_max_vals(), let 32bit bounds learn from 64bit bounds to get more tight bounds tracking. Similar operation can be found in reg_set_min_max().

Also, we can now fold reg_bounds_sync() into zext_32_to_64().

Before:

     func#0 @0
     0: R1=ctx(off=0,imm=0) R10=fp0
     0: (b7) r0 = 0                        ; R0_w=0
     1: (b7) r1 = 0                        ; R1_w=0
     2: (87) r1 = -r1                      ; R1_w=scalar()
     3: (87) r1 = -r1                      ; R1_w=scalar()
     4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
     5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0xffffffff))  <--- [*]
     6: (95) exit

It can be seen that even if the 64bit bounds is clear here, the 32bit bounds is still in the state of 'UNKNOWN'.

After:

     func#0 @0
     0: R1=ctx(off=0,imm=0) R10=fp0
     0: (b7) r0 = 0                        ; R0_w=0
     1: (b7) r1 = 0                        ; R1_w=0
     2: (87) r1 = -r1                      ; R1_w=scalar()
     3: (87) r1 = -r1                      ; R1_w=scalar()
     4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
     5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0x3))  <--- [*]
     6: (95) exit
Quoted 5 lines
Quoted 1 line

Looks good to me. Thanks Youlin.

Thanks Youlin! Looks like the patch breaks CI [0] e.g.:

#142/p bounds check after truncation of non-boundary-crossing range FAIL
Failed to load prog 'Permission denied'!
invalid access to map value, value_size=8 off=16777215 size=1
R0 max value is outside of the allowed memory range
verification time 296 usec
stack depth 8
processed 15 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0

Quoted 7 lines

Please take a look. Also it would be great to add a test_verifier selftest to assert above case from commit log against future changes.

Thanks,
Daniel

   [0] https://github.com/kernel-patches/bpf/runs/7696324041?check_suite_focus=true

This test case fails because the 32bit boundary information is lost after the 11th instruction is executed:
Before:

    11: (07) r1 += 2147483647             ;
R1_w=scalar(umin=70866960383,umax=70866960638,var_off=(0x1000000000;
0xffffffff),u32_min=2147483647,u32_max=-2147483394)
After:
    11: (07) r1 += 2147483647             ;

R1_w=scalar(umin=70866960383,umax=70866960638,var_off=(0x1000000000; 0xffffffff))

This may be because, in previous versions of the code, when __reg_combine_64_into_32() was called, the 32bit boundary was completely deduced from the 64bit boundary, so there was a call to __mark_reg32_unbounded() in __reg_combine_64_into_32().

But now, before adjust_scalar_min_max_vals() calls __reg_combine_64_into_32() , the 32bit bounds are already calculated to some extent, and __mark_reg32_unbounded() will eliminate these information.

Simply copying a code without __mark_reg32_unbounded() should work, perhaps it would be more elegant to introduce a flag into __reg_combine_64_into_32()?

Sorry for not completing the tests because I did not 'make selftests' successfully, and uploaded the code that caused the error.

maintainer
#9 lore raw
Quoted 90 lines · On 8/8/22 5:14 PM, Kuee k1r0a wrote:

---------- Forwarded message ---------
From: Kuee k1r0a <[email protected]>
Date: Mon, Aug 8, 2022 at 11:11 PM
Subject: Re: [PATCH bpf] bpf: Do more tight ALU bounds tracking
To: Daniel Borkmann <[email protected]>

Quoted 53 lines · Daniel Borkmann
Quoted 35 lines · Hao Luo
Quoted 34 lines · Youlin Li

In adjust_scalar_min_max_vals(), let 32bit bounds learn from 64bit bounds to get more tight bounds tracking. Similar operation can be found in reg_set_min_max().

Also, we can now fold reg_bounds_sync() into zext_32_to_64().

Before:

      func#0 @0
      0: R1=ctx(off=0,imm=0) R10=fp0
      0: (b7) r0 = 0                        ; R0_w=0
      1: (b7) r1 = 0                        ; R1_w=0
      2: (87) r1 = -r1                      ; R1_w=scalar()
      3: (87) r1 = -r1                      ; R1_w=scalar()
      4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
      5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0xffffffff))  <--- [*]
      6: (95) exit

It can be seen that even if the 64bit bounds is clear here, the 32bit bounds is still in the state of 'UNKNOWN'.

After:

      func#0 @0
      0: R1=ctx(off=0,imm=0) R10=fp0
      0: (b7) r0 = 0                        ; R0_w=0
      1: (b7) r1 = 0                        ; R1_w=0
      2: (87) r1 = -r1                      ; R1_w=scalar()
      3: (87) r1 = -r1                      ; R1_w=scalar()
      4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
      5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0x3))  <--- [*]
      6: (95) exit
Quoted 5 lines
Quoted 1 line

Looks good to me. Thanks Youlin.

Thanks Youlin! Looks like the patch breaks CI [0] e.g.:

#142/p bounds check after truncation of non-boundary-crossing range FAIL
Failed to load prog 'Permission denied'!
invalid access to map value, value_size=8 off=16777215 size=1
R0 max value is outside of the allowed memory range
verification time 296 usec
stack depth 8
processed 15 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0

Quoted 7 lines

Please take a look. Also it would be great to add a test_verifier selftest to assert above case from commit log against future changes.

Thanks,
Daniel

    [0] https://github.com/kernel-patches/bpf/runs/7696324041?check_suite_focus=true

This test case fails because the 32bit boundary information is lost after the 11th instruction is executed:
Before:

     11: (07) r1 += 2147483647             ;
R1_w=scalar(umin=70866960383,umax=70866960638,var_off=(0x1000000000;
0xffffffff),u32_min=2147483647,u32_max=-2147483394)
After:
     11: (07) r1 += 2147483647             ;

R1_w=scalar(umin=70866960383,umax=70866960638,var_off=(0x1000000000; 0xffffffff))

This may be because, in previous versions of the code, when __reg_combine_64_into_32() was called, the 32bit boundary was completely deduced from the 64bit boundary, so there was a call to __mark_reg32_unbounded() in __reg_combine_64_into_32().

But now, before adjust_scalar_min_max_vals() calls __reg_combine_64_into_32() , the 32bit bounds are already calculated to some extent, and __mark_reg32_unbounded() will eliminate these information.

Quoted 6 lines

Simply copying a code without __mark_reg32_unbounded() should work, perhaps it would be more elegant to introduce a flag into __reg_combine_64_into_32()?

Sorry for not completing the tests because I did not 'make selftests' successfully, and uploaded the code that caused the error.

Under tools/testing/selftests/bpf/, you can run test_progs and test_verifier through the vmtest script, e.g. ./vmtest.sh -- ./test_progs should ease running it. The whole make selftests is not necessary given here we care about BPF, CI is running these where 2 failed and need investigation:

test_progs: PASS
test_progs-no_alu32: FAIL (returned 1)
test_maps: PASS
test_verifier: FAIL (returned 1)

Fwiw, for the test_verifier failure case at least, we should then adapt it in a separate commit with an analysis explaining why it is okay to alter the test; plus a 3rd commit adding new test cases as mentioned earlier.

Thanks a lot, Kuee!
Daniel

author
#10 lore raw

PATCH 1/2 bpf: Fix 32bit bounds update in ALU64

The commit ("bpf: Do more tight ALU bounds tracking") introduces a bug that fails some selftests.

in previous versions of the code, when __reg_combine_64_into_32() was called, the 32bit boundary was completely deduced from the 64bit boundary, so there was a call to __mark_reg32_unbounded() in __reg_combine_64_into_32(). But before adjust_scalar_min_max_vals() calls __reg_combine_64_into_32() , the 32bit bounds are already calculated to some extent, and __mark_reg32_unbounded() will eliminate these information.

Simply remove the call to __reg_combine_64_into_32() and copying a code without __mark_reg32_unbounded() should work.

Before:
./test_verifier 142
#142/p bounds check after truncation of non-boundary-crossing range FAIL Failed to load prog 'Permission denied'!
invalid access to map value, value_size=8 off=16777215 size=1 R0 max value is outside of the allowed memory range verification time 149 usec
stack depth 8
processed 15 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0
Summary: 0 PASSED, 1 SKIPPED, 1 FAILED

After:
./test_verifier 142
#142/p bounds check after truncation of non-boundary-crossing range OK
Summary: 1 PASSED, 1 SKIPPED, 0 FAILED

---
 kernel/bpf/verifier.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)
kernel/bpf/verifier.c +11 -1
index 11d8bb54ba6b..7ea6e0372d62 100644
@@ -9014,7 +9014,17 @@ static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env,
 		/* ALU32 ops are zero extended into 64bit register */
 		zext_32_to_64(dst_reg);
 	} else {
-		__reg_combine_64_into_32(dst_reg);
+		if (__reg64_bound_s32(dst_reg->smin_value) &&
+		    __reg64_bound_s32(dst_reg->smax_value)) {
+			dst_reg->s32_min_value = (s32)dst_reg->smin_value;
+			dst_reg->s32_max_value = (s32)dst_reg->smax_value;
+		}
+		if (__reg64_bound_u32(dst_reg->umin_value) &&
+		    __reg64_bound_u32(dst_reg->umax_value)) {
+			dst_reg->u32_min_value = (u32)dst_reg->umin_value;
+			dst_reg->u32_max_value = (u32)dst_reg->umax_value;
+		}
+		reg_bounds_sync(dst_reg);
 	}
 	return 0;
 }
Signature · 1 line
2.25.1
maintainer
#11 lore raw

Re: [PATCH 1/2] bpf: Fix 32bit bounds update in ALU64

Quoted 47 lines · Youlin Li

The commit ("bpf: Do more tight ALU bounds tracking") introduces a bug that fails some selftests.

in previous versions of the code, when __reg_combine_64_into_32() was called, the 32bit boundary was completely deduced from the 64bit boundary, so there was a call to __mark_reg32_unbounded() in __reg_combine_64_into_32(). But before adjust_scalar_min_max_vals() calls __reg_combine_64_into_32() , the 32bit bounds are already calculated to some extent, and __mark_reg32_unbounded() will eliminate these information.

Simply remove the call to __reg_combine_64_into_32() and copying a code without __mark_reg32_unbounded() should work.

Before:
./test_verifier 142
#142/p bounds check after truncation of non-boundary-crossing range FAIL Failed to load prog 'Permission denied'!
invalid access to map value, value_size=8 off=16777215 size=1 R0 max value is outside of the allowed memory range verification time 149 usec
stack depth 8
processed 15 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0
Summary: 0 PASSED, 1 SKIPPED, 1 FAILED

After:
./test_verifier 142
#142/p bounds check after truncation of non-boundary-crossing range OK
Summary: 1 PASSED, 1 SKIPPED, 0 FAILED

---
  kernel/bpf/verifier.c | 12 +++++++++++-
  1 file changed, 11 insertions(+), 1 deletion(-)
kernel/bpf/verifier.c +1 -1
index 11d8bb54ba6b..7ea6e0372d62 100644
@@ -9014,7 +9014,17 @@ static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env,
  		/* ALU32 ops are zero extended into 64bit register */
  		zext_32_to_64(dst_reg);
  	} else {
-		__reg_combine_64_into_32(dst_reg);
+		if (__reg64_bound_s32(dst_reg->smin_value) &&

kernel/bpf/verifier.c static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env, · line 9027

Quoted 10 lines
kernel/bpf/verifier.c +10 -0
+		    __reg64_bound_s32(dst_reg->smax_value)) {
+			dst_reg->s32_min_value = (s32)dst_reg->smin_value;
+			dst_reg->s32_max_value = (s32)dst_reg->smax_value;
+		}
+		if (__reg64_bound_u32(dst_reg->umin_value) &&
+		    __reg64_bound_u32(dst_reg->umax_value)) {
+			dst_reg->u32_min_value = (u32)dst_reg->umin_value;
+			dst_reg->u32_max_value = (u32)dst_reg->umax_value;
+		}
+		reg_bounds_sync(dst_reg);

Hm, this doesn't apply to the bpf tree. Is this on top of your previous patch [0]? Please squash both together in that case and resubmit your previous one as a v2.

[0] https://lore.kernel.org/bpf/[email protected]/

Thanks,
Daniel

author
#12 lore raw

PATCH v2 1/2 bpf: Do more tight ALU bounds tracking

In adjust_scalar_min_max_vals(), let 32bit bounds learn from 64bit bounds to get more tight bounds tracking. Similar operation can be found in reg_set_min_max().

Note that we cannot simply add a call to __reg_combine_64_into_32(). In previous versions of the code, when __reg_combine_64_into_32() was called, the 32bit boundary was completely deduced from the 64bit boundary, so there was a call to __mark_reg32_unbounded() in __reg_combine_64_into_32(). But in adjust_scalar_min_max_vals(), the 32bit bounds are already calculated to some extent, and __mark_reg32_unbounded() will eliminate these information.

Simply copying a code without __mark_reg32_unbounded() should work.

Also, we can now fold reg_bounds_sync() into zext_32_to_64().

Before:

    func#0 @0
    0: R1=ctx(off=0,imm=0) R10=fp0
    0: (b7) r0 = 0                        ; R0_w=0
    1: (b7) r1 = 0                        ; R1_w=0
    2: (87) r1 = -r1                      ; R1_w=scalar()
    3: (87) r1 = -r1                      ; R1_w=scalar()
    4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
    5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0xffffffff))  <--- [*]
    6: (95) exit

It can be seen that even if the 64bit bounds is clear here, the 32bit bounds is still in the state of 'UNKNOWN'.

After:

    func#0 @0
    0: R1=ctx(off=0,imm=0) R10=fp0
    0: (b7) r0 = 0                        ; R0_w=0
    1: (b7) r1 = 0                        ; R1_w=0
    2: (87) r1 = -r1                      ; R1_w=scalar()
    3: (87) r1 = -r1                      ; R1_w=scalar()
    4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
    5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0x3))  <--- [*]
    6: (95) exit

---
v1 -> v2:

    Replaced the call to __reg_combine_64_into_32() with the code in
    __reg_combine_64_into_32(), and removed the call to
    __mark_reg32_unbounded().

Sorry for the delay, I've been busy looking for a job recently :)

 kernel/bpf/verifier.c | 20 ++++++++++++++++----
 1 file changed, 16 insertions(+), 4 deletions(-)
kernel/bpf/verifier.c +16 -4
index 3eadb14e090b..b7403773e834 100644
@@ -4383,6 +4383,7 @@ static void zext_32_to_64(struct bpf_reg_state *reg)
 {
 	reg->var_off = tnum_subreg(reg->var_off);
 	__reg_assign_32_into_64(reg);
+	reg_bounds_sync(reg);
 }
 
 /* truncate register to smaller size (in bytes)
@@ -9010,10 +9011,22 @@ static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env,
 		break;
 	}
 
-	/* ALU32 ops are zero extended into 64bit register */
-	if (alu32)
+	if (alu32) {
+		/* ALU32 ops are zero extended into 64bit register */
 		zext_32_to_64(dst_reg);
-	reg_bounds_sync(dst_reg);
+	} else {
+		if (__reg64_bound_s32(dst_reg->smin_value) &&
+		    __reg64_bound_s32(dst_reg->smax_value)) {
+			dst_reg->s32_min_value = (s32)dst_reg->smin_value;
+			dst_reg->s32_max_value = (s32)dst_reg->smax_value;
+		}
+		if (__reg64_bound_u32(dst_reg->umin_value) &&
+		    __reg64_bound_u32(dst_reg->umax_value)) {
+			dst_reg->u32_min_value = (u32)dst_reg->umin_value;
+			dst_reg->u32_max_value = (u32)dst_reg->umax_value;
+		}
+		reg_bounds_sync(dst_reg);
+	}
 	return 0;
 }
 
@@ -9202,7 +9215,6 @@ static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
 							 insn->dst_reg);
 				}
 				zext_32_to_64(dst_reg);
-				reg_bounds_sync(dst_reg);
 			}
 		} else {
 			/* case: R = imm
Signature · 1 line
2.25.1
Acked-by on v2 1/2
#13 lore raw

Hi Youlin,

Quoted 44 lines · Youlin Li

In adjust_scalar_min_max_vals(), let 32bit bounds learn from 64bit bounds to get more tight bounds tracking. Similar operation can be found in reg_set_min_max().

Note that we cannot simply add a call to __reg_combine_64_into_32(). In previous versions of the code, when __reg_combine_64_into_32() was called, the 32bit boundary was completely deduced from the 64bit boundary, so there was a call to __mark_reg32_unbounded() in __reg_combine_64_into_32(). But in adjust_scalar_min_max_vals(), the 32bit bounds are already calculated to some extent, and __mark_reg32_unbounded() will eliminate these information.

Simply copying a code without __mark_reg32_unbounded() should work.

Also, we can now fold reg_bounds_sync() into zext_32_to_64().

Before:

    func#0 @0
    0: R1=ctx(off=0,imm=0) R10=fp0
    0: (b7) r0 = 0                        ; R0_w=0
    1: (b7) r1 = 0                        ; R1_w=0
    2: (87) r1 = -r1                      ; R1_w=scalar()
    3: (87) r1 = -r1                      ; R1_w=scalar()
    4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
    5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0xffffffff))  <--- [*]
    6: (95) exit

It can be seen that even if the 64bit bounds is clear here, the 32bit bounds is still in the state of 'UNKNOWN'.

After:

    func#0 @0
    0: R1=ctx(off=0,imm=0) R10=fp0
    0: (b7) r0 = 0                        ; R0_w=0
    1: (b7) r1 = 0                        ; R1_w=0
    2: (87) r1 = -r1                      ; R1_w=scalar()
    3: (87) r1 = -r1                      ; R1_w=scalar()
    4: (c7) r1 s>>= 63                    ; R1_w=scalar(smin=-1,smax=0)
    5: (07) r1 += 2                       ; R1_w=scalar(umin=1,umax=2,var_off=(0x0; 0x3))  <--- [*]
    6: (95) exit
Quoted 2 lines

---

It might be better to put the code that performs the actual bounds deduction into a helper function. It avoids code duplication. But the current version looks fine to me. Thanks for the patch!

author
#14 lore raw

PATCH 2/2 bpf, selftests: Add verifier test case for ALU64

In reply to Daniel Borkmann

Add a test case to ensure that 32-bit bounds can be learned from 64-bit bounds when performing 64-bit ALU operations.

Make use of dead code elimination, so that we can see the verifier bailing out on unmodified kernels.

Before:
./test_verifier 165
#165/p 32-bit bounds update in ALU64 FAIL
Failed to load prog 'Permission denied'!
R2 !read_ok
verification time 49 usec
stack depth 0
processed 8 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0
Summary: 0 PASSED, 1 SKIPPED, 1 FAILED

After:

    ./test_verifier 165
    #165/p 32-bit bounds update in ALU64 OK
    Summary: 1 PASSED, 1 SKIPPED, 0 FAILED
---
 tools/testing/selftests/bpf/verifier/bounds.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
tools/testing/selftests/bpf/verifier/bounds.c +17 -0
index 33125d5f6772..b9aee2f2c66e 100644
@@ -753,3 +753,20 @@
 	.result_unpriv = REJECT,
 	.result = ACCEPT,
 },
+{
+	"32-bit bounds update in ALU64",
+	.insns = {
+	BPF_MOV64_IMM(BPF_REG_0, 0),
+	BPF_MOV64_IMM(BPF_REG_1, 0),
+	BPF_ALU64_IMM(BPF_NEG, BPF_REG_1, 0),
+	BPF_ALU64_IMM(BPF_NEG, BPF_REG_1, 0),
+	BPF_ALU64_IMM(BPF_ARSH, BPF_REG_1, 63),
+	BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, 2),
+	BPF_JMP32_IMM(BPF_JGE, BPF_REG_1, 1, 1),
+	BPF_MOV64_REG(BPF_REG_0, BPF_REG_2),
+	BPF_JMP32_IMM(BPF_JLE, BPF_REG_1, 2, 1),
+	BPF_MOV64_REG(BPF_REG_0, BPF_REG_2),
+	BPF_EXIT_INSN()
+	},
+	.result = ACCEPT,
+},
Signature · 1 line
2.25.1