bpf: Propagate scalar ranges through register assignments.

The llvm register allocator may use two different registers representing the
same virtual register. In such case the following pattern can be observed:
1047: (bf) r9 = r6
1048: (a5) if r6 < 0x1000 goto pc+1
1050: ...
1051: (a5) if r9 < 0x2 goto pc+66
1052: ...
1053: (bf) r2 = r9 /* r2 needs to have upper and lower bounds */

This is normal behavior of greedy register allocator.
The slides 137+ explain why regalloc introduces such register copy:
http://llvm.org/devmtg/2018-04/slides/Yatsina-LLVM%20Greedy%20Register%20Allocator.pdf
There is no way to tell llvm 'not to do this'.
Hence the verifier has to recognize such patterns.

In order to track this information without backtracking allocate ID
for scalars in a similar way as it's done for find_good_pkt_pointers().

When the verifier encounters r9 = r6 assignment it will assign the same ID
to both registers. Later if either register range is narrowed via conditional
jump propagate the register state into the other register.

Clear register ID in adjust_reg_min_max_vals() for any alu instruction. The
register ID is ignored for scalars in regsafe() and doesn't affect state
pruning. mark_reg_unknown() clears the ID. It's used to process call, endian
and other instructions. Hence ID is explicitly cleared only in
adjust_reg_min_max_vals() and in 32-bit mov.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20201009011240.48506-2-alexei.starovoitov@gmail.com

2 files changed, 9 insertions, 9 deletions

diff --git a/tools/testing/selftests/bpf/prog_tests/align.c b/tools/testing/selftests/bpf/prog_tests/align.c
index c548aded6585..52414058a627 100644
--- a/tools/testing/selftests/bpf/prog_tests/align.c
+++ b/tools/testing/selftests/bpf/prog_tests/align.c
@@ -195,13 +195,13 @@ static struct bpf_align_test tests[] = {
 		.prog_type = BPF_PROG_TYPE_SCHED_CLS,
 		.matches = {
 			{7, "R3_w=inv(id=0,umax_value=255,var_off=(0x0; 0xff))"},
-			{8, "R4_w=inv(id=0,umax_value=255,var_off=(0x0; 0xff))"},
+			{8, "R4_w=inv(id=1,umax_value=255,var_off=(0x0; 0xff))"},
 			{9, "R4_w=inv(id=0,umax_value=255,var_off=(0x0; 0xff))"},
-			{10, "R4_w=inv(id=0,umax_value=255,var_off=(0x0; 0xff))"},
+			{10, "R4_w=inv(id=1,umax_value=255,var_off=(0x0; 0xff))"},
 			{11, "R4_w=inv(id=0,umax_value=510,var_off=(0x0; 0x1fe))"},
-			{12, "R4_w=inv(id=0,umax_value=255,var_off=(0x0; 0xff))"},
+			{12, "R4_w=inv(id=1,umax_value=255,var_off=(0x0; 0xff))"},
 			{13, "R4_w=inv(id=0,umax_value=1020,var_off=(0x0; 0x3fc))"},
-			{14, "R4_w=inv(id=0,umax_value=255,var_off=(0x0; 0xff))"},
+			{14, "R4_w=inv(id=1,umax_value=255,var_off=(0x0; 0xff))"},
 			{15, "R4_w=inv(id=0,umax_value=2040,var_off=(0x0; 0x7f8))"},
 			{16, "R4_w=inv(id=0,umax_value=4080,var_off=(0x0; 0xff0))"},
 		},
@@ -518,7 +518,7 @@ static struct bpf_align_test tests[] = {
 			 * the total offset is 4-byte aligned and meets the
 			 * load's requirements.
 			 */
-			{20, "R5=pkt(id=1,off=0,r=4,umin_value=2,umax_value=1034,var_off=(0x2; 0x7fc)"},
+			{20, "R5=pkt(id=2,off=0,r=4,umin_value=2,umax_value=1034,var_off=(0x2; 0x7fc)"},
 
 		},
 	},
@@ -561,18 +561,18 @@ static struct bpf_align_test tests[] = {
 			/* Adding 14 makes R6 be (4n+2) */
 			{11, "R6_w=inv(id=0,umin_value=14,umax_value=74,var_off=(0x2; 0x7c))"},
 			/* Subtracting from packet pointer overflows ubounds */
-			{13, "R5_w=pkt(id=1,off=0,r=8,umin_value=18446744073709551542,umax_value=18446744073709551602,var_off=(0xffffffffffffff82; 0x7c)"},
+			{13, "R5_w=pkt(id=2,off=0,r=8,umin_value=18446744073709551542,umax_value=18446744073709551602,var_off=(0xffffffffffffff82; 0x7c)"},
 			/* New unknown value in R7 is (4n), >= 76 */
 			{15, "R7_w=inv(id=0,umin_value=76,umax_value=1096,var_off=(0x0; 0x7fc))"},
 			/* Adding it to packet pointer gives nice bounds again */
-			{16, "R5_w=pkt(id=2,off=0,r=0,umin_value=2,umax_value=1082,var_off=(0x2; 0xfffffffc)"},
+			{16, "R5_w=pkt(id=3,off=0,r=0,umin_value=2,umax_value=1082,var_off=(0x2; 0xfffffffc)"},
 			/* At the time the word size load is performed from R5,
 			 * its total fixed offset is NET_IP_ALIGN + reg->off (0)
 			 * which is 2.  Then the variable offset is (4n+2), so
 			 * the total offset is 4-byte aligned and meets the
 			 * load's requirements.
 			 */
-			{20, "R5=pkt(id=2,off=0,r=4,umin_value=2,umax_value=1082,var_off=(0x2; 0xfffffffc)"},
+			{20, "R5=pkt(id=3,off=0,r=4,umin_value=2,umax_value=1082,var_off=(0x2; 0xfffffffc)"},
 		},
 	},
 };
diff --git a/tools/testing/selftests/bpf/verifier/direct_packet_access.c b/tools/testing/selftests/bpf/verifier/direct_packet_access.c
index 2c5fbe7bcd27..ae72536603fe 100644
--- a/tools/testing/selftests/bpf/verifier/direct_packet_access.c
+++ b/tools/testing/selftests/bpf/verifier/direct_packet_access.c
@@ -529,7 +529,7 @@
 	},
 	.prog_type = BPF_PROG_TYPE_SCHED_CLS,
 	.result = REJECT,
-	.errstr = "invalid access to packet, off=0 size=8, R5(id=1,off=0,r=0)",
+	.errstr = "invalid access to packet, off=0 size=8, R5(id=2,off=0,r=0)",
 	.flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS,
 },
 {