[PATCH] 2.6.17.9 Incorrect string length checking in param_set_copystring()

[PATCH] 2.6.17.9 Incorrect string length checking in param_set_copystring()

[Julio Auto]

As for 2.6.17.9, linux/include/linux/moduleparam.h suggests the user
of module_param_string() to set the maxlen parameter to
strlen(string), ie. '\0' excluded. However the function that actually
sets the string (param_set_copystring()), doesn't accept inputs with
maxlen-1 characters, reporting that the supplied string should fit
maxlen-1 chars.
See patch below.
Cheers,

    Julio Auto

--- linux-2.6.17.9/kernel/params.c.old  2006-08-19 20:48:30.000000000 -0700
+++ linux-2.6.17.9/kernel/params.c      2006-08-19 20:49:15.000000000 -0700
@@ -351,9 +351,9 @@ int param_set_copystring(const char *val
 {
        struct kparam_string *kps = kp->arg;

-       if (strlen(val)+1 > kps->maxlen) {
+       if (strlen(val) > kps->maxlen) {
                printk(KERN_ERR "%s: string doesn't fit in %u chars.\n",
-                      kp->name, kps->maxlen-1);
+                      kp->name, kps->maxlen);
                return -ENOSPC;
        }
        strcpy(kps->string, val);

Re: [PATCH] 2.6.17.9 Incorrect string length checking in param_set_copystring()

[Julio Auto]

> As for 2.6.17.9, linux/include/linux/moduleparam.h suggests the user
> of module_param_string() to set the maxlen parameter to
> strlen(string), ie. '\0' excluded.

Actually, sizeof(string), not strlen(string). Senseless typo here.
Sorry, my bad. :)

Cheers,

    Julio Auto

Re: [PATCH] 2.6.17.9 Incorrect string length checking in param_set_copystring()

[Jan Engelhardt]

>> As for 2.6.17.9, linux/include/linux/moduleparam.h suggests the user
>> of module_param_string() to set the maxlen parameter to
>> strlen(string), ie. '\0' excluded.
>
> Actually, sizeof(string), not strlen(string). Senseless typo here.
> Sorry, my bad. :)

With \0 excluded, you want strlen(string) or sizeof(string)-1.


Jan Engelhardt
-- 

Re: [PATCH] 2.6.17.9 Incorrect string length checking in param_set_copystring()

[Luca]

Julio Auto <mindvortex@gmail.com> ha scritto:
> As for 2.6.17.9, linux/include/linux/moduleparam.h suggests the user
> of module_param_string() to set the maxlen parameter to
> strlen(string), ie. '\0' excluded. However the function that actually
> sets the string (param_set_copystring()), doesn't accept inputs with
> maxlen-1 characters, reporting that the supplied string should fit
> maxlen-1 chars.
> See patch below.
> Cheers,
> 
>    Julio Auto
> 
> --- linux-2.6.17.9/kernel/params.c.old  2006-08-19 20:48:30.000000000 -0700
> +++ linux-2.6.17.9/kernel/params.c      2006-08-19 20:49:15.000000000 -0700
> @@ -351,9 +351,9 @@ int param_set_copystring(const char *val
> {
>        struct kparam_string *kps = kp->arg;
> 
> -       if (strlen(val)+1 > kps->maxlen) {
> +       if (strlen(val) > kps->maxlen) {
>                printk(KERN_ERR "%s: string doesn't fit in %u chars.\n",
> -                      kp->name, kps->maxlen-1);
> +                      kp->name, kps->maxlen);
>                return -ENOSPC;
>        }
>        strcpy(kps->string, val);

Hi,
I believe that the code is correct. kps->maxlen is the lenght of the
buffer; when dealing with a string of N chars you need an array of (N +
1) bytes in order to store the terminator ('\0').

Look again at the check:

        if (strlen(val) > kps->maxlen) {

Suppose that val is a string of 10 chars (strlen(val) == 10), suppose
that the buffer holds 10 bytes (kps->maxlen == 10). The expression:

        if (10 > 10) {

is false, so strcpy() ends up writing 11 bytes in a buffer of 10 bytes.

Also, for future patches see point 11 of
Documentation/SubmittingPatches.

Luca
-- 
Home: http://kronoz.cjb.net
"Su cio` di cui non si puo` parlare e` bene tacere".
 Ludwig Wittgenstein

Re: [PATCH] 2.6.17.9 Incorrect string length checking in param_set_copystring()

[Julio Auto]

First of all, thanks for the reply. I'll take the afore mentioned
point of SubmittingPatches in consideration in the future.

And yes, you're correct.
I was misled by an incorrect use of sizeof() and it made believe that
its return would strip the null terminator. Once again, my bad. :)

    Julio Auto

On 8/20/06, Luca <kronos.it@gmail.com> wrote:
> Julio Auto <mindvortex@gmail.com> ha scritto:
> > As for 2.6.17.9, linux/include/linux/moduleparam.h suggests the user
> > of module_param_string() to set the maxlen parameter to
> > strlen(string), ie. '\0' excluded. However the function that actually
> > sets the string (param_set_copystring()), doesn't accept inputs with
> > maxlen-1 characters, reporting that the supplied string should fit
> > maxlen-1 chars.
> > See patch below.
> > Cheers,
> >
> >    Julio Auto
> >
> > --- linux-2.6.17.9/kernel/params.c.old  2006-08-19 20:48:30.000000000 -0700
> > +++ linux-2.6.17.9/kernel/params.c      2006-08-19 20:49:15.000000000 -0700
> > @@ -351,9 +351,9 @@ int param_set_copystring(const char *val
> > {
> >        struct kparam_string *kps = kp->arg;
> >
> > -       if (strlen(val)+1 > kps->maxlen) {
> > +       if (strlen(val) > kps->maxlen) {
> >                printk(KERN_ERR "%s: string doesn't fit in %u chars.\n",
> > -                      kp->name, kps->maxlen-1);
> > +                      kp->name, kps->maxlen);
> >                return -ENOSPC;
> >        }
> >        strcpy(kps->string, val);
>
> Hi,
> I believe that the code is correct. kps->maxlen is the lenght of the
> buffer; when dealing with a string of N chars you need an array of (N +
> 1) bytes in order to store the terminator ('\0').
>
> Look again at the check:
>
>         if (strlen(val) > kps->maxlen) {
>
> Suppose that val is a string of 10 chars (strlen(val) == 10), suppose
> that the buffer holds 10 bytes (kps->maxlen == 10). The expression:
>
>         if (10 > 10) {
>
> is false, so strcpy() ends up writing 11 bytes in a buffer of 10 bytes.
>
> Also, for future patches see point 11 of
> Documentation/SubmittingPatches.
>
> Luca
> --
> Home: http://kronoz.cjb.net
> "Su cio` di cui non si puo` parlare e` bene tacere".
>  Ludwig Wittgenstein
>