Skip to contentSkip to conversation

Insert statement before every return point

Cocci Diego Nieto Cid 3 participants 5 messages Apr 3, 2016 – Apr 6, 2016

lore
Share & tools
lore
#0 lore raw

Hello,

I'm new to Coccinelle and there's a patch I wrote that doesn't work as I expected.

The purpose of the patch is to implement a toy stack smashing protector by inserting a local variable with a unique name, by using 'fresh identifier', in every function and checking its value before returning.

The patch is the followig:
----8<----8<----8<----8<----8<----8<----
@@
identifier fn;
fresh identifier canary = "canary_";
@@
fn (...)
{
+ int canary = 0xdeadbeef;
...
+ if ((canary ^ 0xdeadbeef) != 0)
+ DIVIDE_BY_ZERO;
return ...;
}
---->8---->8---->8---->8---->8---->8----

I tried it on a simple function, like the one below:

----8<----8<----8<----8<----8<----8<----
int fn (int a, int b)

{
	if (a > b)
		return a - b;
	return 0;
}

---->8---->8---->8---->8---->8---->8----

and got the expected result. The generated patch follows.

----8<----8<----8<----8<----8<----8<----
init_defs_builtins: /usr/lib/coccinelle/standard.h
HANDLING: branch.c

tmp/cocci-output-3058-7f8e45-branch.c renamed from branch.c +7 -1
diff = 
@@ -1,6 +1,12 @@
 int fn (int a, int b)
 {
-	if (a > b)
+	int canary_0 = 0xdeadbeef;
+	if (a > b) {
+		if ((canary_0 ^ 0xdeadbeef) != 0)
+			DIVIDE_BY_ZERO;
 		return a - b;
+	}
+	if ((canary_0 ^ 0xdeadbeef) != 0)
+		DIVIDE_BY_ZERO;
 	return 0;
 }

---->8---->8---->8---->8---->8---->8----

But as soon as I add an elseif branch the patch stops working. Here are the code

----8<----8<----8<----8<----8<----8<----
int fn (int a, int b)

{
	if (a > b)
		return a - b;
	else if (b > a)
		return b - a;
	return 0;
}

---->8---->8---->8---->8---->8---->8----

and 'spatch --debug' output:

----8<----8<----8<----8<----8<----8<----
init_defs_builtins: /usr/lib/coccinelle/standard.h
-----------------------------------------------------------------------
processing semantic patch file: ssp.cocci with isos from: /usr/lib/coccinelle/standard.iso
-----------------------------------------------------------------------
@@
identifier fn;
fresh identifier canary = "canary_";
@@
fn (...)
{
+ int canary = 0xdeadbeef;
...
+ if ((canary ^ 0xdeadbeef) != 0)
+ DIVIDE_BY_ZERO;
return ...;
}

HANDLING: elseif.c
-----------------------------------------------------------------------
let's go
-----------------------------------------------------------------------
-----------------------------------------------------------------------
-----------------------------------------------------------------------
rule starting on line 1 =
-----------------------------------------------------------------------
dependencies for rule rule starting on line 1 satisfied:
binding in = []
binding relevant in = []
-----------------------------------------------------------------------
Finished
-----------------------------------------------------------------------
Check duplication for 1 files ---->8---->8---->8---->8---->8---->8----

I don't understand why introducing an 'else if' clause is causing the semantic patch to skip over the function.

Anybody knows what's happenig and how to fix it?

Thanks,
Diego

#1 lore raw
Quoted 18 lines · Diego Nieto Cid

Hello,

I'm new to Coccinelle and there's a patch I wrote that doesn't work as I expected.

The purpose of the patch is to implement a toy stack smashing protector by inserting a local variable with a unique name, by using 'fresh identifier', in every function and checking its value before returning.

The patch is the followig:
----8<----8<----8<----8<----8<----8<----
@@
identifier fn;
fresh identifier canary = "canary_";
@@
fn (...)
{

Review

Quoted 6 lines
(quoted diff) +3 -0
+  int canary = 0xdeadbeef;
...
+  if ((canary ^ 0xdeadbeef) != 0)
+      DIVIDE_BY_ZERO;
  return ...;
 }

I'm not sure why it doesn't work on your example. However it sems to work better like this:

@@
identifier fn;
fresh identifier canary = "canary_";
@@
fn (...)
{
+ int canary = 0xdeadbeef;
<...
+ if ((canary ^ 0xdeadbeef) != 0)
+ DIVIDE_BY_ZERO;
return ...;
...>
}

This just makes the change whereever it is relevant, regardless of control-flow. It seems to even work in the case of an implicit return at the end of the function, when the function has a void return type.

julia

Quoted 101 lines

---->8---->8---->8---->8---->8---->8----

I tried it on a simple function, like the one below:

----8<----8<----8<----8<----8<----8<----
int fn (int a, int b)

{
	if (a > b)
		return a - b;
	return 0;
}

---->8---->8---->8---->8---->8---->8----

and got the expected result. The generated patch follows.

----8<----8<----8<----8<----8<----8<----
init_defs_builtins: /usr/lib/coccinelle/standard.h
HANDLING: branch.c
diff =

tmp/cocci-output-3058-7f8e45-branch.c renamed from branch.c +7 -2
@@ -1,6 +1,12 @@
 int fn (int a, int b)
 {
-	if (a > b)
+	int canary_0 = 0xdeadbeef;
+	if (a > b) {
+		if ((canary_0 ^ 0xdeadbeef) != 0)
+			DIVIDE_BY_ZERO;
 		return a - b;
+	}
+	if ((canary_0 ^ 0xdeadbeef) != 0)
+		DIVIDE_BY_ZERO;
 	return 0;
 }
---->8---->8---->8---->8---->8---->8----

But as soon as I add an elseif branch the patch stops working. Here are the code

----8<----8<----8<----8<----8<----8<----
int fn (int a, int b)

{
	if (a > b)
		return a - b;
	else if (b > a)
		return b - a;
	return 0;
}

---->8---->8---->8---->8---->8---->8----

and 'spatch --debug' output:

----8<----8<----8<----8<----8<----8<----
init_defs_builtins: /usr/lib/coccinelle/standard.h
-----------------------------------------------------------------------
processing semantic patch file: ssp.cocci with isos from: /usr/lib/coccinelle/standard.iso
-----------------------------------------------------------------------
@@
identifier fn;
fresh identifier canary = "canary_";
@@
fn (...)
{

tmp/cocci-output-3058-7f8e45-branch.c renamed from branch.c +3 -0
+  int canary = 0xdeadbeef;
...
+  if ((canary ^ 0xdeadbeef) != 0)
+      DIVIDE_BY_ZERO;
   return ...;
 }

HANDLING: elseif.c
-----------------------------------------------------------------------
let's go
-----------------------------------------------------------------------
-----------------------------------------------------------------------
-----------------------------------------------------------------------
rule starting on line 1 =
-----------------------------------------------------------------------
dependencies for rule rule starting on line 1 satisfied:
binding in = []
binding relevant in = []
-----------------------------------------------------------------------
Finished
-----------------------------------------------------------------------
Check duplication for 1 files ---->8---->8---->8---->8---->8---->8----

I don't understand why introducing an 'else if' clause is causing the semantic patch to skip over the function.

Anybody knows what's happenig and how to fix it?

Thanks,
Diego
_______________________________________________
Cocci mailing list
Cocci at systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

#2 lore raw
Quoted 5 lines · Julia Lawall

I'm not sure why it doesn't work on your example. However it sems to work better like this:
...

Indeed, your version works great. Thank you very much!

#3 lore raw
In reply to Julia Lawall
Quoted 46 lines · Julia Lawall
Quoted 18 lines · Diego Nieto Cid

Hello,

I'm new to Coccinelle and there's a patch I wrote that doesn't work as I expected.

The purpose of the patch is to implement a toy stack smashing protector by inserting a local variable with a unique name, by using 'fresh identifier', in every function and checking its value before returning.

The patch is the followig:
----8<----8<----8<----8<----8<----8<----
@@
identifier fn;
fresh identifier canary = "canary_";
@@
fn (...)
{

Review

Quoted 6 lines
(quoted diff) +3 -0
+  int canary = 0xdeadbeef;
...
+  if ((canary ^ 0xdeadbeef) != 0)
+      DIVIDE_BY_ZERO;
  return ...;
 }

I'm not sure why it doesn't work on your example. However it sems to work better like this:

@@
identifier fn;
fresh identifier canary = "canary_";
@@
fn (...)
{

(quoted diff) +3 -0
+  int canary = 0xdeadbeef;
   <...
+  if ((canary ^ 0xdeadbeef) != 0)
+      DIVIDE_BY_ZERO;
  return ...;
  ...>
 }
Quoted 4 lines

This just makes the change whereever it is relevant, regardless of control-flow. It seems to even work in the case of an implicit return at the end of the function, when the function has a void return type.

the canary would need to be declared volatile or gcc will optimize it out.

thx!
hofrat

#4 lore raw

El 6 abr 2016, a las 03:02, Nicholas Mc Guire <[email protected]> escribi?:

Quoted 3 lines

the canary would need to be declared
volatile or gcc will optimize it out.

Yes, indeed. I discovered it the hard way after applying Julia's suggestion :( Thanks.

BTW, I used the DIVIDE_BY_ZERO macro because Coccinelle was failing when inline assembly was used, like this:

--------8<-------------8<----------8<------------8<-------
init_defs_builtins: /usr/lib/coccinelle/standard.h
-----------------------------------------------------------------------
processing semantic patch file: ssp3.cocci
with isos from: /usr/lib/coccinelle/standard.iso
-----------------------------------------------------------------------
@@
identifier fn;
fresh identifier canary = "canary_";
@@
fn (...)
{
+ volatile int canary = 0xdeadbeef;
<...
+ if ((canary ^ 0xdeadbeef) != 0)
+ asm volatile ("divl %0"::"r"(0):"eax","edx");
return ...;
...>
}

162 170
Fatal error: exception Failure("plus: parse error: \n = File \"ssp3.cocci\", line 10, column 11, charpos = 162\n around = 'volatile', whole content = + asm volatile (\"divl %0\"::\"r\"(0):\"eax\",\"edx\");\n")
--------8<-------------8<----------8<------------8<-------

Is it because I'm using GCC extensions?

Regards,
Diego.