Re: [PATCH] net/failsafe: fix parameters parsing

Hi Gaetan

-----Original Message-----
From: Gaëtan Rivet [mailto:gaetan.rivet@6wind.com]
Sent: Thursday, August 17, 2017 7:25 PM
To: Matan Azrad <redacted>
Cc: dev@dpdk.org; Ori Kam <redacted>; stable@dpdk.org
Subject: Re: [PATCH] net/failsafe: fix parameters parsing

On Thu, Aug 17, 2017 at 03:54:23PM +0000, Matan Azrad wrote:

quoted

quoted

-----Original Message-----
From: Gaëtan Rivet [mailto:gaetan.rivet@6wind.com]
Sent: Thursday, August 17, 2017 6:26 PM
To: Matan Azrad <redacted>
Cc: dev@dpdk.org; Ori Kam <redacted>; stable@dpdk.org
Subject: Re: [PATCH] net/failsafe: fix parameters parsing

Hi Matan,

Thanks for spotting the bug.

On Thu, Aug 17, 2017 at 05:19:41PM +0300, Matan Azrad wrote:

quoted

The corrupted code used wrongly snprintf return value as the
number of characters actually copied, in spite of the meanning is
the number of characters which would be generated for the given

input.

quoted

quoted

quoted

It caused to remain zerod bytes between the failsafe command line
non sub device parameters indicates end of string.

Hence, when rte_kvargs_parse tried to parse all parameters, it got
end of string after the first one and the others weren't parsed.

So, if the mac parameters was the first in command line it was
taken while hotplug_poll was left default, and vice versa.

The fix updates the buffer index by dedicated variable contains
the copy size, by the way uses memcpy instead of snprintf which is
good enouth for this copy scenario.

Actually snprintf should still be used.

Why?
We just want to copy from buffer to buffer, no needs snprintf
overheads If actually we are not using complicated format.

snprintf plays nice with strings and ensure that it will be terminated by a '\0'.
It is generally preferable, particularly considering that the parameter string
we are building here is meant to be parsed by another lib (rte_kvargs).

In this specific scenario it is not needed because we for sure know that the length doesn't pass after '\0'.
By using memcpy we can prevent format parsing and other checks done by snprintf.

quoted

quoted

quoted

Fixes: a46f8d584eb8 ("net/failsafe: add fail-safe PMD")
Cc: stable@dpdk.org

Signed-off-by: Matan Azrad <redacted>
---
 drivers/net/failsafe/failsafe_args.c | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)

diff --git a/drivers/net/failsafe/failsafe_args.c

b/drivers/net/failsafe/failsafe_args.c
index 1f22416..0a98b04 100644

--- a/drivers/net/failsafe/failsafe_args.c
+++ b/drivers/net/failsafe/failsafe_args.c

@@ -271,7 +271,7 @@ static int
 fs_remove_sub_devices_definition(char params[DEVARGS_MAXLEN])

{

quoted

quoted

quoted

 	char buffer[DEVARGS_MAXLEN] = {0};
-	size_t a, b;
+	size_t a, b, temp;

temp is not descriptive enough.

What are about a, b, i ?

quoted

You are declaring this variable here because you want to re-use it
instead of `start`. This is an overreach however, this fix must be
restricted to the actual bug.

temp is helping to address the original bug, don't we want to reuse
variable Instead of 2  if statement variables, maybe other name for all?

The compiler should be able to see that their use does not overlap.

This is new for me, thanks for the lesson.
I will update it to len & start as you suggested.

The issue is that temp is meant to describe a length limit in one branch and
be a marker of a starting point in another. Thus both variable should be
named differently to make the intent clear.

When declaring temp (or saved_val) in the higher-scope, you are thus forced
to use generic, non-descriptive name precisely because you are sharing the
variable between two different uses.

quoted

Something like:
a=> start
b=> end
i => next
temp=> saved_val

You didn't answer what are about my suggestions for new names instead of 
a,b and I which are not descriptive too (like temp :))

quoted

quoted

quoted

 	int i;

 	a = 0;

@@ -286,12 +286,14 @@ fs_remove_sub_devices_definition(char

params[DEVARGS_MAXLEN])

quoted

 			ERROR("Invalid parameter");
 			return -EINVAL;
 		}
-		if (params[b] == ',' || params[b] == '\0')

Declare the temporary variable in this scope, with a descriptive
name such as "len", "length", "param_len" or something close.

quoted

-			i += snprintf(&buffer[i], b - a + 1, "%s", &params[a]);
-		if (params[b] == '(') {
-			size_t start = b;
+		if (params[b] == ',' || params[b] == '\0') {
+			temp = b - a + 1;

The value here should be "b - a".
If a != 0 however, then we are parsing a new parameter and buffer
already contains at least one. A comma should be added to separate the

two.

quoted

quoted

An example might clarify what I mean:

                if (params[b] == ',' || params[b] == '\0') {
                        size_t param_len = b - a;

                        if (a)
                                param_len += 1;
                        snprintf(&buffer[i], param_len + 1, "%s%s",
                                 a ? "," : "", &params[a]);
                        i += param_len;
                }

The conditionals about `a` are ugly however, if you find a better
way to write those you are most welcome :).

quoted

+			memcpy(&buffer[i], &params[a], temp);
+			i += temp;
+		} else if (params[b] == '(') {
+			temp = b;
 			b += closing_paren(&params[b]);
-			if (b == start)

I think the last comma is harmless for next parse But I can just
change the last coma to '\0' in the end of function(if exists).
But, This solves another issue, don't it?  Maybe in different patch?

It might be harmless, but I prefer having a clean output.
Which other issue? Handling the commas seems in line with properly copying
regular parameters.

Yes, but it is another issue in this specific copy (don't refer to snprintf return value) 
But, it's OK, I think them both can be handle by this patch :)

You could certainly check for the presence of a comma, but you already see
how ugly the code flow would be ("if it exists"), and you will keep a dirty
string between two loops, with possibly other branches wanting to return on
error.

I prefer having a compact, self-contained logic addressing kvargs which
should not leak conceptual gotcha that the next programmer would need to
be wary of when trying to touch the code in other parts of the function.

If we leave always b-a+1 we potentially can get the next buffer strings:
p1,p2...pN,
or 
p1,p2,pN(\0)
 while i is the index of the character after a comma or '\0'

so, for prevent the last comma we can override always in the end of function(not between 2 loops) '\0' like this:
if (i >0 )
/* For prevent last comma. */
	buffer[i-1] = '\0';
instead of using 2 conditions per non sub device param - 'if(a)' and 'a ? "," : ""'
I think each future programmer can easily understand it.

 

quoted

quoted

The changes should be limited to the actual bug. No need to change this.

quoted

+			if (b == temp)
 				return -EINVAL;
 			b += 1;
 			if (params[b] == '\0')
--
2.7.4

Thanks again for the debug and sorry for being nitpicky.

--
Gaëtan Rivet
6WIND

Regards,
Matan Azrad.

--
Gaëtan Rivet
6WIND

Regards
Matan Azrad.