[PATCH v3] powerpc/uprobes: Validation for prefixed instruction

Subsystems: linux for powerpc (32-bit and 64-bit), the rest, uprobes

STALE2003d

7 messages, 3 authors, 2021-03-04 · open the first message on its own page

[PATCH v3] powerpc/uprobes: Validation for prefixed instruction

From: Ravi Bangoria <hidden>
Date: 2021-03-04 05:09:59

As per ISA 3.1, prefixed instruction should not cross 64-byte
boundary. So don't allow Uprobe on such prefixed instruction.

There are two ways probed instruction is changed in mapped pages.
First, when Uprobe is activated, it searches for all the relevant
pages and replace instruction in them. In this case, if that probe
is on the 64-byte unaligned prefixed instruction, error out
directly. Second, when Uprobe is already active and user maps a
relevant page via mmap(), instruction is replaced via mmap() code
path. But because Uprobe is invalid, entire mmap() operation can
not be stopped. In this case just print an error and continue.

Signed-off-by: Ravi Bangoria <redacted>
---
v2: https://lore.kernel.org/r/20210204104703.273429-1-ravi.bangoria@linux.ibm.com
v2->v3:
  - Drop restriction for Uprobe on suffix of prefixed instruction.
    It needs lot of code change including generic code but what
    we get in return is not worth it.

 arch/powerpc/kernel/uprobes.c | 8 ++++++++
 1 file changed, 8 insertions(+)
diff --git a/arch/powerpc/kernel/uprobes.c b/arch/powerpc/kernel/uprobes.c
index e8a63713e655..c400971ebe70 100644
--- a/arch/powerpc/kernel/uprobes.c
+++ b/arch/powerpc/kernel/uprobes.c
@@ -41,6 +41,14 @@ int arch_uprobe_analyze_insn(struct arch_uprobe *auprobe,
 	if (addr & 0x03)
 		return -EINVAL;
 
+	if (!IS_ENABLED(CONFIG_PPC64) || !cpu_has_feature(CPU_FTR_ARCH_31))
+		return 0;
+
+	if (ppc_inst_prefixed(auprobe->insn) && (addr & 0x3F) == 0x3C) {
+		pr_info_ratelimited("Cannot register a uprobe on 64 byte unaligned prefixed instruction\n");
+		return -EINVAL;
+	}
+
 	return 0;
 }
 
-- 
2.27.0

Re: [PATCH v3] powerpc/uprobes: Validation for prefixed instruction

From: Naveen N. Rao <hidden>
Date: 2021-03-04 05:55:21

On 2021/03/04 10:35AM, Ravi Bangoria wrote:
quoted hunk
As per ISA 3.1, prefixed instruction should not cross 64-byte
boundary. So don't allow Uprobe on such prefixed instruction.

There are two ways probed instruction is changed in mapped pages.
First, when Uprobe is activated, it searches for all the relevant
pages and replace instruction in them. In this case, if that probe
is on the 64-byte unaligned prefixed instruction, error out
directly. Second, when Uprobe is already active and user maps a
relevant page via mmap(), instruction is replaced via mmap() code
path. But because Uprobe is invalid, entire mmap() operation can
not be stopped. In this case just print an error and continue.

Signed-off-by: Ravi Bangoria <redacted>
---
v2: https://lore.kernel.org/r/20210204104703.273429-1-ravi.bangoria@linux.ibm.com
v2->v3:
  - Drop restriction for Uprobe on suffix of prefixed instruction.
    It needs lot of code change including generic code but what
    we get in return is not worth it.

 arch/powerpc/kernel/uprobes.c | 8 ++++++++
 1 file changed, 8 insertions(+)
diff --git a/arch/powerpc/kernel/uprobes.c b/arch/powerpc/kernel/uprobes.c
index e8a63713e655..c400971ebe70 100644
--- a/arch/powerpc/kernel/uprobes.c
+++ b/arch/powerpc/kernel/uprobes.c
@@ -41,6 +41,14 @@ int arch_uprobe_analyze_insn(struct arch_uprobe *auprobe,
 	if (addr & 0x03)
 		return -EINVAL;
 
+	if (!IS_ENABLED(CONFIG_PPC64) || !cpu_has_feature(CPU_FTR_ARCH_31))
+		return 0;
Sorry, I missed this last time, but I think we can drop the above 
checks. ppc_inst_prefixed() already factors in the dependency on 
CONFIG_PPC64, and I don't think we need to confirm if we're running on a 
ISA V3.1 for the below check.

With that:
Acked-by: Naveen N. Rao <redacted>
+
+	if (ppc_inst_prefixed(auprobe->insn) && (addr & 0x3F) == 0x3C) {
+		pr_info_ratelimited("Cannot register a uprobe on 64 byte unaligned prefixed instruction\n");
+		return -EINVAL;
+	}
+
- Naveen

Re: [PATCH v3] powerpc/uprobes: Validation for prefixed instruction

From: Christophe Leroy <hidden>
Date: 2021-03-04 07:34:37


Le 04/03/2021 à 06:05, Ravi Bangoria a écrit :
quoted hunk
As per ISA 3.1, prefixed instruction should not cross 64-byte
boundary. So don't allow Uprobe on such prefixed instruction.

There are two ways probed instruction is changed in mapped pages.
First, when Uprobe is activated, it searches for all the relevant
pages and replace instruction in them. In this case, if that probe
is on the 64-byte unaligned prefixed instruction, error out
directly. Second, when Uprobe is already active and user maps a
relevant page via mmap(), instruction is replaced via mmap() code
path. But because Uprobe is invalid, entire mmap() operation can
not be stopped. In this case just print an error and continue.

Signed-off-by: Ravi Bangoria <redacted>
---
v2: https://lore.kernel.org/r/20210204104703.273429-1-ravi.bangoria@linux.ibm.com
v2->v3:
   - Drop restriction for Uprobe on suffix of prefixed instruction.
     It needs lot of code change including generic code but what
     we get in return is not worth it.

  arch/powerpc/kernel/uprobes.c | 8 ++++++++
  1 file changed, 8 insertions(+)
diff --git a/arch/powerpc/kernel/uprobes.c b/arch/powerpc/kernel/uprobes.c
index e8a63713e655..c400971ebe70 100644
--- a/arch/powerpc/kernel/uprobes.c
+++ b/arch/powerpc/kernel/uprobes.c
@@ -41,6 +41,14 @@ int arch_uprobe_analyze_insn(struct arch_uprobe *auprobe,
  	if (addr & 0x03)
  		return -EINVAL;
  
+	if (!IS_ENABLED(CONFIG_PPC64) || !cpu_has_feature(CPU_FTR_ARCH_31))
cpu_has_feature(CPU_FTR_ARCH_31) should return 'false' when CONFIG_PPC64 is not enabled, no need to 
double check.
+		return 0;
+
+	if (ppc_inst_prefixed(auprobe->insn) && (addr & 0x3F) == 0x3C) {
Maybe 3C instead of 4F ? : (addr & 0x3C) == 0x3C

What about

(addr & (SZ_64 - 4)) == SZ_64 - 4 to make it more explicit ?

Or ALIGN(addr, SZ_64) != ALIGN(addr + 8, SZ_64)
+		pr_info_ratelimited("Cannot register a uprobe on 64 byte unaligned prefixed instruction\n");
+		return -EINVAL;
+	}
+
  	return 0;
  }
  
Christophe

Re: [PATCH v3] powerpc/uprobes: Validation for prefixed instruction

From: Ravi Bangoria <hidden>
Date: 2021-03-04 07:42:21

quoted
@@ -41,6 +41,14 @@ int arch_uprobe_analyze_insn(struct arch_uprobe *auprobe,
  	if (addr & 0x03)
  		return -EINVAL;
  
+	if (!IS_ENABLED(CONFIG_PPC64) || !cpu_has_feature(CPU_FTR_ARCH_31))
+		return 0;
Sorry, I missed this last time, but I think we can drop the above
checks. ppc_inst_prefixed() already factors in the dependency on
CONFIG_PPC64,
Yeah, makes sense. I initially added CONFIG_PPC64 check because
I was using ppc_inst_prefix(x, y) macro which is not available
for !CONFIG_PPC64.
and I don't think we need to confirm if we're running on a
ISA V3.1 for the below check.
Prefixed instructions would be supported only when ARCH_31 is set.
(Ignoring insane scenario where user probes on prefixed instruction
on p10 predecessors). So I guess I still need ARCH_31 check?

Ravi

Re: [PATCH v3] powerpc/uprobes: Validation for prefixed instruction

From: Ravi Bangoria <hidden>
Date: 2021-03-04 10:16:31


On 3/4/21 1:02 PM, Christophe Leroy wrote:

Le 04/03/2021 à 06:05, Ravi Bangoria a écrit :
quoted
As per ISA 3.1, prefixed instruction should not cross 64-byte
boundary. So don't allow Uprobe on such prefixed instruction.

There are two ways probed instruction is changed in mapped pages.
First, when Uprobe is activated, it searches for all the relevant
pages and replace instruction in them. In this case, if that probe
is on the 64-byte unaligned prefixed instruction, error out
directly. Second, when Uprobe is already active and user maps a
relevant page via mmap(), instruction is replaced via mmap() code
path. But because Uprobe is invalid, entire mmap() operation can
not be stopped. In this case just print an error and continue.

Signed-off-by: Ravi Bangoria <redacted>
---
v2: https://lore.kernel.org/r/20210204104703.273429-1-ravi.bangoria@linux.ibm.com
v2->v3:
   - Drop restriction for Uprobe on suffix of prefixed instruction.
     It needs lot of code change including generic code but what
     we get in return is not worth it.

  arch/powerpc/kernel/uprobes.c | 8 ++++++++
  1 file changed, 8 insertions(+)
diff --git a/arch/powerpc/kernel/uprobes.c b/arch/powerpc/kernel/uprobes.c
index e8a63713e655..c400971ebe70 100644
--- a/arch/powerpc/kernel/uprobes.c
+++ b/arch/powerpc/kernel/uprobes.c
@@ -41,6 +41,14 @@ int arch_uprobe_analyze_insn(struct arch_uprobe *auprobe,
      if (addr & 0x03)
          return -EINVAL;
+    if (!IS_ENABLED(CONFIG_PPC64) || !cpu_has_feature(CPU_FTR_ARCH_31))
cpu_has_feature(CPU_FTR_ARCH_31) should return 'false' when CONFIG_PPC64 is not enabled, no need to double check.
Ok.

I'm going to drop CONFIG_PPC64 check because it's not really
required as I replied to Naveen. So, I'll keep CPU_FTR_ARCH_31
check as is.
quoted
+        return 0;
+
+    if (ppc_inst_prefixed(auprobe->insn) && (addr & 0x3F) == 0x3C) {
Maybe 3C instead of 4F ? : (addr & 0x3C) == 0x3C
Didn't follow. It's not (addr & 0x3C), it's (addr & 0x3F).
What about

(addr & (SZ_64 - 4)) == SZ_64 - 4 to make it more explicit ?
Yes this is bit better. Though, it should be:

     (addr & (SZ_64 - 1)) == SZ_64 - 4

Ravi

Re: [PATCH v3] powerpc/uprobes: Validation for prefixed instruction

From: Christophe Leroy <hidden>
Date: 2021-03-04 10:53:02


Le 04/03/2021 à 11:13, Ravi Bangoria a écrit :

On 3/4/21 1:02 PM, Christophe Leroy wrote:
quoted

Le 04/03/2021 à 06:05, Ravi Bangoria a écrit :
quoted
As per ISA 3.1, prefixed instruction should not cross 64-byte
boundary. So don't allow Uprobe on such prefixed instruction.

There are two ways probed instruction is changed in mapped pages.
First, when Uprobe is activated, it searches for all the relevant
pages and replace instruction in them. In this case, if that probe
is on the 64-byte unaligned prefixed instruction, error out
directly. Second, when Uprobe is already active and user maps a
relevant page via mmap(), instruction is replaced via mmap() code
path. But because Uprobe is invalid, entire mmap() operation can
not be stopped. In this case just print an error and continue.

Signed-off-by: Ravi Bangoria <redacted>
---
v2: https://lore.kernel.org/r/20210204104703.273429-1-ravi.bangoria@linux.ibm.com
v2->v3:
   - Drop restriction for Uprobe on suffix of prefixed instruction.
     It needs lot of code change including generic code but what
     we get in return is not worth it.

  arch/powerpc/kernel/uprobes.c | 8 ++++++++
  1 file changed, 8 insertions(+)
diff --git a/arch/powerpc/kernel/uprobes.c b/arch/powerpc/kernel/uprobes.c
index e8a63713e655..c400971ebe70 100644
--- a/arch/powerpc/kernel/uprobes.c
+++ b/arch/powerpc/kernel/uprobes.c
@@ -41,6 +41,14 @@ int arch_uprobe_analyze_insn(struct arch_uprobe *auprobe,
      if (addr & 0x03)
          return -EINVAL;
+    if (!IS_ENABLED(CONFIG_PPC64) || !cpu_has_feature(CPU_FTR_ARCH_31))
cpu_has_feature(CPU_FTR_ARCH_31) should return 'false' when CONFIG_PPC64 is not enabled, no need 
to double check.
Ok.

I'm going to drop CONFIG_PPC64 check because it's not really
required as I replied to Naveen. So, I'll keep CPU_FTR_ARCH_31
check as is.
quoted
quoted
+        return 0;
+
+    if (ppc_inst_prefixed(auprobe->insn) && (addr & 0x3F) == 0x3C) {
Maybe 3C instead of 4F ? : (addr & 0x3C) == 0x3C
Didn't follow. It's not (addr & 0x3C), it's (addr & 0x3F).
Sorry I meant 3c instead of 3f (And usually we don't use capital letters for that).
The last two bits are supposed to always be 0, so it doesn't really matter, I just thought it would 
look better having the same value both sides of the test, ie (addr & 0x3c) == 0x3c.
quoted
What about

(addr & (SZ_64 - 4)) == SZ_64 - 4 to make it more explicit ?
Yes this is bit better. Though, it should be:

     (addr & (SZ_64 - 1)) == SZ_64 - 4
-1 or -4 should give the same results as instructions are always 32 bits aligned though.

Christophe

Re: [PATCH v3] powerpc/uprobes: Validation for prefixed instruction

From: Ravi Bangoria <hidden>
Date: 2021-03-04 11:05:03


On 3/4/21 4:21 PM, Christophe Leroy wrote:

Le 04/03/2021 à 11:13, Ravi Bangoria a écrit :
quoted

On 3/4/21 1:02 PM, Christophe Leroy wrote:
quoted

Le 04/03/2021 à 06:05, Ravi Bangoria a écrit :
quoted
As per ISA 3.1, prefixed instruction should not cross 64-byte
boundary. So don't allow Uprobe on such prefixed instruction.

There are two ways probed instruction is changed in mapped pages.
First, when Uprobe is activated, it searches for all the relevant
pages and replace instruction in them. In this case, if that probe
is on the 64-byte unaligned prefixed instruction, error out
directly. Second, when Uprobe is already active and user maps a
relevant page via mmap(), instruction is replaced via mmap() code
path. But because Uprobe is invalid, entire mmap() operation can
not be stopped. In this case just print an error and continue.

Signed-off-by: Ravi Bangoria <redacted>
---
v2: https://lore.kernel.org/r/20210204104703.273429-1-ravi.bangoria@linux.ibm.com
v2->v3:
   - Drop restriction for Uprobe on suffix of prefixed instruction.
     It needs lot of code change including generic code but what
     we get in return is not worth it.

  arch/powerpc/kernel/uprobes.c | 8 ++++++++
  1 file changed, 8 insertions(+)
diff --git a/arch/powerpc/kernel/uprobes.c b/arch/powerpc/kernel/uprobes.c
index e8a63713e655..c400971ebe70 100644
--- a/arch/powerpc/kernel/uprobes.c
+++ b/arch/powerpc/kernel/uprobes.c
@@ -41,6 +41,14 @@ int arch_uprobe_analyze_insn(struct arch_uprobe *auprobe,
      if (addr & 0x03)
          return -EINVAL;
+    if (!IS_ENABLED(CONFIG_PPC64) || !cpu_has_feature(CPU_FTR_ARCH_31))
cpu_has_feature(CPU_FTR_ARCH_31) should return 'false' when CONFIG_PPC64 is not enabled, no need to double check.
Ok.

I'm going to drop CONFIG_PPC64 check because it's not really
required as I replied to Naveen. So, I'll keep CPU_FTR_ARCH_31
check as is.
quoted
quoted
+        return 0;
+
+    if (ppc_inst_prefixed(auprobe->insn) && (addr & 0x3F) == 0x3C) {
Maybe 3C instead of 4F ? : (addr & 0x3C) == 0x3C
Didn't follow. It's not (addr & 0x3C), it's (addr & 0x3F).
Sorry I meant 3c instead of 3f (And usually we don't use capital letters for that).
The last two bits are supposed to always be 0, so it doesn't really matter, I just thought it would look better having the same value both sides of the test, ie (addr & 0x3c) == 0x3c.
Ok yeah makes sense. Thanks.
quoted
quoted
What about

(addr & (SZ_64 - 4)) == SZ_64 - 4 to make it more explicit ?
Yes this is bit better. Though, it should be:

     (addr & (SZ_64 - 1)) == SZ_64 - 4
-1 or -4 should give the same results as instructions are always 32 bits aligned though.
Got it.

Ravi
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help