[PATCH] vsprintf: Make sure argument to %*ph/%pv is valid

[PATCH] vsprintf: Make sure argument to %*ph/%pv is valid

[Boris Ostrovsky]

If NULL pointer is passed for these specifiers then print '-'
or 'd-v-'.

Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
---
 xen/common/vsprintf.c |   19 +++++++++++++++++++
 1 files changed, 19 insertions(+), 0 deletions(-)

diff --git a/xen/common/vsprintf.c b/xen/common/vsprintf.c
index 065cc42..84cfa85 100644
--- a/xen/common/vsprintf.c
+++ b/xen/common/vsprintf.c
@@ -280,6 +280,13 @@ static char *pointer(char *str, char *end, const char **fmt_ptr,
         /* Consumed 'h' from the format string. */
         ++*fmt_ptr;
 
+        if ( hex_buffer == NULL )
+        {
+            if ( str < end )
+                *str++ = '-';
+            return str;
+        }
+
         /* Bound user count from %* to between 0 and 64 bytes. */
         if ( field_width <= 0 )
             return str;
@@ -338,6 +345,18 @@ static char *pointer(char *str, char *end, const char **fmt_ptr,
         ++*fmt_ptr;
         if ( str < end )
             *str = 'd';
+        if ( v == NULL )
+        {
+            /* If v is NULL then print 'd-v-' */
+            str++;
+            if ( str < end )
+                *str++ = '-';
+            if ( str < end )
+                *str++ = 'v';
+            if ( str < end )
+                *str++ = '-';
+            return str;
+        }
         str = number(str + 1, end, v->domain->domain_id, 10, -1, -1, 0);
         if ( str < end )
             *str = 'v';
-- 
1.7.1

Re: [PATCH] vsprintf: Make sure argument to %*ph/%pv is valid

[Andrew Cooper]

On 10/02/2015 21:44, Boris Ostrovsky wrote:
> If NULL pointer is passed for these specifiers then print '-'
> or 'd-v-'.
>
> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>

Instead of special casing each custom format, I would be tempted just to
put the switch() in an if(arg) conditional, and fall back to the regular
number for a NULL pointer, or go along the glibc route and print "(NULL)".

I presume you found this by falling over a NULL  pointer while
debugging?  I can't see a legitimate reason for formally supporting NULL
pointers in each context, but crashing is certainly better avoided.

~Andrew

> ---
>  xen/common/vsprintf.c |   19 +++++++++++++++++++
>  1 files changed, 19 insertions(+), 0 deletions(-)
>
> diff --git a/xen/common/vsprintf.c b/xen/common/vsprintf.c
> index 065cc42..84cfa85 100644
> --- a/xen/common/vsprintf.c
> +++ b/xen/common/vsprintf.c
> @@ -280,6 +280,13 @@ static char *pointer(char *str, char *end, const char **fmt_ptr,
>          /* Consumed 'h' from the format string. */
>          ++*fmt_ptr;
>  
> +        if ( hex_buffer == NULL )
> +        {
> +            if ( str < end )
> +                *str++ = '-';
> +            return str;
> +        }
> +
>          /* Bound user count from %* to between 0 and 64 bytes. */
>          if ( field_width <= 0 )
>              return str;
> @@ -338,6 +345,18 @@ static char *pointer(char *str, char *end, const char **fmt_ptr,
>          ++*fmt_ptr;
>          if ( str < end )
>              *str = 'd';
> +        if ( v == NULL )
> +        {
> +            /* If v is NULL then print 'd-v-' */
> +            str++;
> +            if ( str < end )
> +                *str++ = '-';
> +            if ( str < end )
> +                *str++ = 'v';
> +            if ( str < end )
> +                *str++ = '-';
> +            return str;
> +        }
>          str = number(str + 1, end, v->domain->domain_id, 10, -1, -1, 0);
>          if ( str < end )
>              *str = 'v';

Re: [PATCH] vsprintf: Make sure argument to %*ph/%pv is valid

[Boris Ostrovsky]

On 02/10/2015 04:49 PM, Andrew Cooper wrote:
> On 10/02/2015 21:44, Boris Ostrovsky wrote:
>> If NULL pointer is passed for these specifiers then print '-'
>> or 'd-v-'.
>>
>> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
> Instead of special casing each custom format, I would be tempted just to
> put the switch() in an if(arg) conditional, and fall back to the regular
> number for a NULL pointer, or go along the glibc route and print "(NULL)".

OK, I could do that.

>
> I presume you found this by falling over a NULL  pointer while
> debugging?  I can't see a legitimate reason for formally supporting NULL
> pointers in each context, but crashing is certainly better avoided.

Yes, I hit %pv case when the vcpu I was tracking turned NULL. And yes, 
we then crash.

-boris

>
> ~Andrew
>
>> ---
>>   xen/common/vsprintf.c |   19 +++++++++++++++++++
>>   1 files changed, 19 insertions(+), 0 deletions(-)
>>
>> diff --git a/xen/common/vsprintf.c b/xen/common/vsprintf.c
>> index 065cc42..84cfa85 100644
>> --- a/xen/common/vsprintf.c
>> +++ b/xen/common/vsprintf.c
>> @@ -280,6 +280,13 @@ static char *pointer(char *str, char *end, const char **fmt_ptr,
>>           /* Consumed 'h' from the format string. */
>>           ++*fmt_ptr;
>>   
>> +        if ( hex_buffer == NULL )
>> +        {
>> +            if ( str < end )
>> +                *str++ = '-';
>> +            return str;
>> +        }
>> +
>>           /* Bound user count from %* to between 0 and 64 bytes. */
>>           if ( field_width <= 0 )
>>               return str;
>> @@ -338,6 +345,18 @@ static char *pointer(char *str, char *end, const char **fmt_ptr,
>>           ++*fmt_ptr;
>>           if ( str < end )
>>               *str = 'd';
>> +        if ( v == NULL )
>> +        {
>> +            /* If v is NULL then print 'd-v-' */
>> +            str++;
>> +            if ( str < end )
>> +                *str++ = '-';
>> +            if ( str < end )
>> +                *str++ = 'v';
>> +            if ( str < end )
>> +                *str++ = '-';
>> +            return str;
>> +        }
>>           str = number(str + 1, end, v->domain->domain_id, 10, -1, -1, 0);
>>           if ( str < end )
>>               *str = 'v';

Re: [PATCH] vsprintf: Make sure argument to %*ph/%pv is valid

[Jan Beulich]

>>> On 10.02.15 at 22:49, <andrew.cooper3@citrix.com> wrote:
> On 10/02/2015 21:44, Boris Ostrovsky wrote:
>> If NULL pointer is passed for these specifiers then print '-'
>> or 'd-v-'.
>>
>> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
> 
> Instead of special casing each custom format, I would be tempted just to
> put the switch() in an if(arg) conditional, and fall back to the regular
> number for a NULL pointer, or go along the glibc route and print "(NULL)".
> 
> I presume you found this by falling over a NULL  pointer while
> debugging?  I can't see a legitimate reason for formally supporting NULL
> pointers in each context, but crashing is certainly better avoided.

I tend to disagree - lots of functions get passed pointers that
they dereference _assuming_ they're not NULL (or otherwise
invalid). vsprintf() doesn't need to be a special case - it's the
caller's responsibility to ensure it doesn't pass NULL in such
cases.

That said, I wouldn't object to a patch doing what is outlined in
the first paragraph of Andrew's reply above, but extended to
cover pointing anywhere into known bad ranges (i.e. at least
for x86 in particular anywhere below HYPERVISOR_VIRT_START).

Jan