Skip to contentSkip to conversation

selftests/bpf: Use the last page in test_snprintf_btf on s390

PATCH Superseded Ilya Leoshkevich 3 participants 4 messages Feb 26, 2021 – Feb 28, 2021

Subsystems (3)

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 selftests/bpf: Use the last page in test_snprintf_btf on s390

test_snprintf_btf fails on s390, because NULL points to a readable struct lowcore there. Fix by using the last page instead.

Error message example:

    printing 0000000000000000 should generate error, got (361)

---

v1: https://lore.kernel.org/bpf/[email protected]/
v1 -> v2: Yonghong suggested to add the pointer value to the error
message.
I've noticed that I've been passing BADPTR as flags, therefore the fix worked only by accident. Put it into p.ptr where it belongs.

v2: https://lore.kernel.org/bpf/[email protected]/
v2 -> v3: Heiko mentioned that using _REGION1_SIZE is not future-proof.
We had a private discussion and came to the conclusion that the the last page is good enough.

 .../testing/selftests/bpf/progs/netif_receive_skb.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)
tools/testing/selftests/bpf/progs/netif_receive_skb.c +10 -3
index 6b670039ea67..c3669967067e 100644
@@ -16,6 +16,13 @@ bool skip = false;
 #define STRSIZE			2048
 #define EXPECTED_STRSIZE	256
 
+#if defined(bpf_target_s390)
+/* NULL points to a readable struct lowcore on s390, so take the last page */
+#define BADPTR			((void *)0xFFFFFFFFFFFFF000ULL)
+#else
+#define BADPTR			0
+#endif
+
 #ifndef ARRAY_SIZE
 #define ARRAY_SIZE(x)	(sizeof(x) / sizeof((x)[0]))
 #endif
@@ -113,11 +120,11 @@ int BPF_PROG(trace_netif_receive_skb, struct sk_buff *skb)
 	}
 
 	/* Check invalid ptr value */
-	p.ptr = 0;
+	p.ptr = BADPTR;
 	__ret = bpf_snprintf_btf(str, STRSIZE, &p, sizeof(p), 0);
 	if (__ret >= 0) {
-		bpf_printk("printing NULL should generate error, got (%d)",
-			   __ret);
+		bpf_printk("printing %p should generate error, got (%d)",
+			   BADPTR, __ret);
 		ret = -ERANGE;
 	}
 
Signature · 1 line
2.29.2
#1 lore raw
Quoted 17 lines · Ilya Leoshkevich

test_snprintf_btf fails on s390, because NULL points to a readable struct lowcore there. Fix by using the last page instead.

Error message example:

     printing 0000000000000000 should generate error, got (361)

---

v1: https://lore.kernel.org/bpf/[email protected]/
v1 -> v2: Yonghong suggested to add the pointer value to the error
message.
I've noticed that I've been passing BADPTR as flags, therefore the fix worked only by accident. Put it into p.ptr where it belongs.

Quoted 4 lines

v2: https://lore.kernel.org/bpf/[email protected]/
v2 -> v3: Heiko mentioned that using _REGION1_SIZE is not future-proof.
We had a private discussion and came to the conclusion that the the last page is good enough.

Heiko, could you ack the patch if it is okay? Thanks!

Quoted 24 lines
  .../testing/selftests/bpf/progs/netif_receive_skb.c | 13 ++++++++++---
  1 file changed, 10 insertions(+), 3 deletions(-)
tools/testing/selftests/bpf/progs/netif_receive_skb.c +7 -0
index 6b670039ea67..c3669967067e 100644
@@ -16,6 +16,13 @@ bool skip = false;
  #define STRSIZE			2048
  #define EXPECTED_STRSIZE	256
 
+#if defined(bpf_target_s390)
+/* NULL points to a readable struct lowcore on s390, so take the last page */
+#define BADPTR			((void *)0xFFFFFFFFFFFFF000ULL)
+#else
+#define BADPTR			0
+#endif
+
  #ifndef ARRAY_SIZE
  #define ARRAY_SIZE(x)	(sizeof(x) / sizeof((x)[0]))
  #endif
@@ -113,11 +120,11 @@ int BPF_PROG(trace_netif_receive_skb, struct sk_buff *skb)
  	}

tools/testing/selftests/bpf/progs/netif_receive_skb.c int BPF_PROG(trace_netif_receive_skb, struct sk_buff *skb) · line 127

Quoted 10 lines
tools/testing/selftests/bpf/progs/netif_receive_skb.c +3 -3
 
  	/* Check invalid ptr value */
-	p.ptr = 0;
+	p.ptr = BADPTR;
  	__ret = bpf_snprintf_btf(str, STRSIZE, &p, sizeof(p), 0);
  	if (__ret >= 0) {
-		bpf_printk("printing NULL should generate error, got (%d)",
-			   __ret);
+		bpf_printk("printing %p should generate error, got (%d)",
+			   BADPTR, __ret);

From https://www.kernel.org/doc/Documentation/printk-formats.txt:

Pointers printed without a specifier extension (i.e unadorned %p) are hashed to give a unique identifier without leaking kernel addresses to user space. On 64 bit machines the first 32 bits are zeroed. If you _really_ want the address see %px below.

I think it is okay to use %px here.

Quoted 4 lines
tools/testing/selftests/bpf/progs/netif_receive_skb.c +0 -0
  		ret = -ERANGE;
  	}
 
#2 lore raw
Quoted 76 lines · Yonghong Song
Quoted 20 lines · Ilya Leoshkevich

test_snprintf_btf fails on s390, because NULL points to a readable struct lowcore there. Fix by using the last page instead.

Error message example:

     printing 0000000000000000 should generate error, got (361)

---

v1:
https://lore.kernel.org/bpf/[email protected]/
v1 -> v2: Yonghong suggested to add the pointer value to the error message.
           I've noticed that I've been passing BADPTR as flags, therefore
           the fix worked only by accident. Put it into p.ptr where it
           belongs.

Quoted 7 lines

v2:
https://lore.kernel.org/bpf/[email protected]/
v2 -> v3: Heiko mentioned that using _REGION1_SIZE is not future- proof.
           We had a private discussion and came to the conclusion that
           the the last page is good enough.

Heiko, could you ack the patch if it is okay? Thanks!

Quoted 32 lines

  .../testing/selftests/bpf/progs/netif_receive_skb.c | 13
++++++++++---

  1 file changed, 10 insertions(+), 3 deletions(-)
tools/testing/selftests/bpf/progs/netif_receive_skb.c +0 -0
index 6b670039ea67..c3669967067e 100644
@@ -16,6 +16,13 @@ bool skip = false;
   #define STRSIZE                       2048
   #define EXPECTED_STRSIZE      256

tools/testing/selftests/bpf/progs/netif_receive_skb.c +2 -0
+#if defined(bpf_target_s390)
+/* NULL points to a readable struct lowcore on s390, so take the

last page */

tools/testing/selftests/bpf/progs/netif_receive_skb.c +5 -0
+#define BADPTR                 ((void *)0xFFFFFFFFFFFFF000ULL)
+#else
+#define BADPTR                 0
+#endif
+

  #ifndef ARRAY_SIZE
  #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0])) #endif

tools/testing/selftests/bpf/progs/netif_receive_skb.c +0 -0
@@ -113,11 +120,11 @@ int BPF_PROG(trace_netif_receive_skb, struct
 sk_buff *skb)
         }


        /* Check invalid ptr value */
-       p.ptr = 0;
+       p.ptr = BADPTR;

Quoted 8 lines

        __ret = bpf_snprintf_btf(str, STRSIZE, &p, sizeof(p), 0); if (__ret >= 0) {
-               bpf_printk("printing NULL should generate error, got (%d)",
-                          __ret);
+               bpf_printk("printing %p should generate error, got (%d)",
+                          BADPTR, __ret);

 From https://www.kernel.org/doc/Documentation/printk-formats.txt:

Quoted 8 lines

Pointers printed without a specifier extension (i.e unadorned %p) are hashed to give a unique identifier without leaking kernel addresses to user
space. On 64 bit machines the first 32 bits are zeroed. If you _really_
want the address see %px below.

I think it is okay to use %px here.

I don't think bpf_trace_printk supports it, but I'll use %llx instead.

[...]

#3 lore raw
In reply to Yonghong Song
Quoted 7 lines · Yonghong Song
Quoted 7 lines

test_snprintf_btf fails on s390, because NULL points to a readable struct lowcore there. Fix by using the last page instead.

Error message example:

     printing 0000000000000000 should generate error, got (361)
Quoted 7 lines
Quoted 5 lines

---
v2 -> v3: Heiko mentioned that using _REGION1_SIZE is not future-proof.
We had a private discussion and came to the conclusion that the the last page is good enough.

Heiko, could you ack the patch if it is okay? Thanks!

Yes, sure.