On 01/28/2013 06:52 PM, Eric Blake wrote:
On 01/22/2013 12:28 PM, John Ferlan wrote:
>>> @@ -1795,8 +1795,11 @@ virXen_setvcpumap(int handle, int id, unsigned int
vcpu,
>>> return -1;
>>>
>>> memset(pm, 0, sizeof(cpumap_t));
>>> - for (j = 0; j < maplen; j++)
>>> + for (j = 0; j < maplen; j++) {
>>> + /* coverity[ptr_arith] */
>>> + /* coverity[sign_extension] */
>>> *(pm + (j / 8)) |= cpumap[j] << (8 * (j & 7));
>>> + }
>>
>> Having to add two comments to shut up Coverity feels awkward. Would it
>> also work to do 'uint64_t j' instead of the current 'int j' in
the
>> declaration a few lines earlier? Not only would it be a smaller diff,
>> but the fewer Coverity comments we have to use, the better I feel.
>>
>> I know this has already been pushed, but it is still worth seeing if a
>> followup patch can clean things further.
Ouch, we really DO have a bug, not to mention some very horrible code
trying to do nasty aliasing that is not very portable. I'm surprised we
don't have alignment complaints, by trying to treat cpumap_t as an array
of 64-bit integers.
>>
>
> Nope, just tried using uint64_t on 'j' without any luck. I also tried
putting the comments on the same line without the desired effect. Here's data on the
two reported defects (I turned OFF line wrap for this - the line numbers are from an older
analysis):
>
> Error: ARRAY_VS_SINGLETON (CWE-119): [#def1]
> libvirt-1.0.0/src/xen/xen_hypervisor.c:1751: cond_false: Condition
"hv_versions.hypervisor > 1", taking false branch
> libvirt-1.0.0/src/xen/xen_hypervisor.c:1790: else_branch: Reached else branch
> libvirt-1.0.0/src/xen/xen_hypervisor.c:1792: address_of: Taking address with
"&xen_cpumap" yields a singleton pointer.
> libvirt-1.0.0/src/xen/xen_hypervisor.c:1792: assign: Assigning: "pm" =
"&xen_cpumap".
> libvirt-1.0.0/src/xen/xen_hypervisor.c:1795: cond_false: Condition "maplen >
8 /* (int)sizeof (cpumap_t) */", taking false branch
> libvirt-1.0.0/src/xen/xen_hypervisor.c:1795: cond_false: Condition "0UL /*
sizeof (cpumap_t) & 7 */", taking false branch
> libvirt-1.0.0/src/xen/xen_hypervisor.c:1799: cond_true: Condition "j <
maplen", taking true branch
> libvirt-1.0.0/src/xen/xen_hypervisor.c:1800: ptr_arith: Using "pm" as an
array. This might corrupt or misinterpret adjacent memory locations.
This one, I don't know if we can silence without a coverity comment.
Basically, it boils down to whether cpumap_t is typedef'd to something
that can possibly be larger than 64 bits (it isn't - Coverity just
confirmed that sizeof(cpumap_t) is 8 bytes). Since we just ensured that
maplen will not go beyond the bounds of a 64-bit int array that overlays
the same memory space, I'm okay with the /* coverity[ptr_arith] */
comment, but see below...
>
> AND
>
> Error: SIGN_EXTENSION (CWE-194): [#def245]
> libvirt-1.0.0/src/xen/xen_hypervisor.c:1800: sign_extension: Suspicious implicit sign
extension: "cpumap[j]" with type "unsigned char" (8 bits, unsigned) is
promoted in "cpumap[j] << 8 * (j & 7)" to type "int" (32
bits, signed), then sign-extended to type "unsigned long" (64 bits, unsigned).
If "cpumap[j] << 8 * (j & 7)" is greater than 0x7FFFFFFF, the upper
bits of the result will all be 1.
Here is the real bug (but I'm surprised why it didn't go away when you
changed j from int to int64_t). When j==4, you are attempting to do
'int << (8*4)'; but you _can't_ portably shift a 32-bit integer by any
more than 31 bits. We _have_ to add in a type conversion to force this
shift to occur in 64-bit math, such as:
*(pm + (j / 8)) |= cpumap[j] << (8ULL * (j & 7));
Or better yet, why even futz around with 64-bit aliasing? It looks like
this code is trying to take endian-independent input and force it into
an endian-dependent xen_cpumap variable. I think it might be cleaner as:
} else {
cpumap_t xen_cpumap; /* limited to 64 CPUs in old hypervisors */
uint64_t val = 0;
int j;
if ((maplen > (int)sizeof(cpumap_t)) || (sizeof(cpumap_t) & 7))
return -1;
memset(&xen_cpumap, 0, sizeof(*xen_cpumap));
for (j = 0; j < maplen; j++) {
val |= cpumap[j] << (8ULL * (j & 7));
if (j % 7 == 7) {
memcpy(((char *)&xen_cpumap) + j, &val, sizeof(val));
val = 0;
}
}
and see if that shuts up Coverity.
FWIW: This path of code only occurs when "(hv_versions.hypervisor <=
1)" - IOW really old code. I checked history and the code was added by
commit id '86247f2c'. Also since the target of the "|" operation is a
'uint64_t' (e.g. *(pm + (j / 8)), wouldn't the shift from 0->56 be OK
(e.g. (8 * (j & 7)))? That is it's not an 'int << (8*4)' it's
a
'uint64_t << (8*4)'.
When first approaching this I figured I didn't want to introduce a bug
into code that's been around a long time and that may not have any one
using it. I agree the line looks ugly and it did take me a bit to think
about it.
Mathematically what you propose with the memcpy() works; however, I come
from an architecture where a memcpy to an unaligned address causes a
fault of which there'd be many. I wrote a little sample loop that went
from 0->63 and printed out values just to see what one would get. The
memcpy value is the newer algorithm and the *pm value is the former
algorithm.
j=0 j%7=0 j&7=0 memcpy=0x7fffde5b4828 *pm=0x7fffde5b4828 shift=0
j=1 j%7=1 j&7=1 memcpy=0x7fffde5b4829 *pm=0x7fffde5b4828 shift=8
j=2 j%7=2 j&7=2 memcpy=0x7fffde5b482a *pm=0x7fffde5b4828 shift=16
j=3 j%7=3 j&7=3 memcpy=0x7fffde5b482b *pm=0x7fffde5b4828 shift=24
j=4 j%7=4 j&7=4 memcpy=0x7fffde5b482c *pm=0x7fffde5b4828 shift=32
j=5 j%7=5 j&7=5 memcpy=0x7fffde5b482d *pm=0x7fffde5b4828 shift=40
j=6 j%7=6 j&7=6 memcpy=0x7fffde5b482e *pm=0x7fffde5b4828 shift=48
j=7 j%7=0 j&7=7 memcpy=0x7fffde5b482f *pm=0x7fffde5b4828 shift=56
j=8 j%7=1 j&7=0 memcpy=0x7fffde5b4830 *pm=0x7fffde5b4830 shift=0
j=9 j%7=2 j&7=1 memcpy=0x7fffde5b4831 *pm=0x7fffde5b4830 shift=8
j=10 j%7=3 j&7=2 memcpy=0x7fffde5b4832 *pm=0x7fffde5b4830 shift=16
j=11 j%7=4 j&7=3 memcpy=0x7fffde5b4833 *pm=0x7fffde5b4830 shift=24
j=12 j%7=5 j&7=4 memcpy=0x7fffde5b4834 *pm=0x7fffde5b4830 shift=32
j=13 j%7=6 j&7=5 memcpy=0x7fffde5b4835 *pm=0x7fffde5b4830 shift=40
j=14 j%7=0 j&7=6 memcpy=0x7fffde5b4836 *pm=0x7fffde5b4830 shift=48
j=15 j%7=1 j&7=7 memcpy=0x7fffde5b4837 *pm=0x7fffde5b4830 shift=56
...
j=56 j%7=0 j&7=0 memcpy=0x7fffde5b4860 *pm=0x7fffde5b4860 shift=0
j=57 j%7=1 j&7=1 memcpy=0x7fffde5b4861 *pm=0x7fffde5b4860 shift=8
j=58 j%7=2 j&7=2 memcpy=0x7fffde5b4862 *pm=0x7fffde5b4860 shift=16
j=59 j%7=3 j&7=3 memcpy=0x7fffde5b4863 *pm=0x7fffde5b4860 shift=24
j=60 j%7=4 j&7=4 memcpy=0x7fffde5b4864 *pm=0x7fffde5b4860 shift=32
j=61 j%7=5 j&7=5 memcpy=0x7fffde5b4865 *pm=0x7fffde5b4860 shift=40
j=62 j%7=6 j&7=6 memcpy=0x7fffde5b4866 *pm=0x7fffde5b4860 shift=48
j=63 j%7=0 j&7=7 memcpy=0x7fffde5b4867 *pm=0x7fffde5b4860 shift=56
John