broken memsets in net/sk_mca.c (multicast)
[Please CC: me on reply, I'm not subscribed (yet)]
Hi all,
I just found two very weird memsets in drivers/net/sk_mca.c. I am not working on that driver at all, but some grepping of the kernel source pointed me to it so I thought I wouldn't keep quiet about it.
Here is the offending code:
static void skmca_set_multicast_list(struct net_device *dev)
{
(...)
if (dev->flags & IFF_ALLMULTI) { /* get all multicasts */
memset(block.LAdrF, 8, 0xff);
} else { /* get selected/no multicasts */
(...)
memset(block.LAdrF, 8, 0x00);
(...)
}
(...)
}Is it just me, or are these two memsets just plain broken? Not only the size is hardcoded, but the parameters are swapped! I guess that this driver never worked in multicast mode. The correct memsets are obviously:
memset(block.LAdrF, 0xff, sizeof(block.LAdrF)); and memset(block.LAdrF, 0x00, sizeof(block.LAdrF));
respectively.
The odd thing is that the bug is there since the driver was introduced in the 2.2.10 kernel. So, 2.2, 2.4 and 2.6 kernels are all affected.
I admit I'm a bit surprized that this could survive until today, so just tell me if it's me missing the point ;)
Thanks.
Signature · 2 lines
Jean Delvare http://www.ensicaen.ismra.fr/~delvare/
Quoted 2 lines · Jean Delvare
I just found two very weird memsets in
drivers/net/sk_mca.c.
yup.
drivers/net/sk_mca.c renamed from drivers/net/sk_mca.c~sk_mca-multicast-fix +3 -5
@@ -996,14 +996,12 @@ static void skmca_set_multicast_list(strelse block.Mode &= ~LANCE_INIT_PROM ; -if (dev->flags &IFF_ALLMULTI ) {/* get all multicasts */ -memset (block.LAdrF,8 ,0xff ); - }else {/* get selected/no multicasts */ - +memset (block.LAdrF,0xff ,sizeof (block.LAdrF)); +if (!(dev->flags &IFF_ALLMULTI )) { +/* get selected/no multicasts */ struct dev_mc_list *mptr;int code; -memset (block.LAdrF,8 ,0x00 );for (mptr = dev->mc_list; mptr !=NULL ; mptr = mptr->next) { code =GetHash (mptr->dmi_addr); block.LAdrF[(code >>3 ) &7 ] |=1 << (code &7 );
_
Quoted 4 lines · Andrew Morton
Quoted 1 line · Jean Delvare
I just found two very weird memsets in
drivers/net/sk_mca.c.yup.
Erm...
drivers/net/sk_mca.c static void skmca_set_multicast_list(str · line 999
Quoted 11 lines
drivers/net/sk_mca.c renamed from drivers/net/sk_mca.c~sk_mca-multicast-fix +1
-4@@ -996,14 +996,12 @@ static void skmca_set_multicast_list(strelse block.Mode &= ~LANCE_INIT_PROM ; -if (dev->flags &IFF_ALLMULTI ) {/* get all multicasts */ -memset (block.LAdrF,8 ,0xff ); - }else {/* get selected/no multicasts */ - +memset (block.LAdrF,0xff ,sizeof (block.LAdrF));
Initialise the whole of block.LAdrF to all-bits-set, and then...
drivers/net/sk_mca.c static void skmca_set_multicast_list(str · line 1007
Quoted 9 lines
drivers/net/sk_mca.c renamed from drivers/net/sk_mca.c~sk_mca-multicast-fix +2
-1+if (!(dev->flags &IFF_ALLMULTI )) { +/* get selected/no multicasts */ struct dev_mc_list *mptr;int code; -memset (block.LAdrF,8 ,0x00 );for (mptr = dev->mc_list; mptr !=NULL ; mptr = mptr->next) { code =GetHash (mptr->dmi_addr); block.LAdrF[(code >>3 ) &7 ] |=1 << (code &7 );
Set bits, which are already set from the previous memset.
Surely this can't be right?
Signature · 4 lines
Russell King
Linux kernel 2.6 ARM Linux - http://www.arm.linux.org.uk/
maintainer of: 2.6 PCMCIA - http://pcmcia.arm.linux.org.uk/
2.6 Serial coreQuoted 2 lines · Russell King
Erm...
drivers/net/sk_mca.c renamed from drivers/net/sk_mca.c~sk_mca-multicast-fix +2 -2
@@ -997,13 +997,13 @@ static void skmca_set_multicast_list(str block.Mode &= ~LANCE_INIT_PROM ;if (dev->flags &IFF_ALLMULTI ) {/* get all multicasts */ -memset (block.LAdrF,8 ,0xff ); +memset (block.LAdrF,0xff ,sizeof (block.LAdrF)); }else {/* get selected/no multicasts */ struct dev_mc_list *mptr;int code; -memset (block.LAdrF,8 ,0x00 ); +memset (block.LAdrF,0 ,sizeof (block.LAdrF));for (mptr = dev->mc_list; mptr !=NULL ; mptr = mptr->next) { code =GetHash (mptr->dmi_addr); block.LAdrF[(code >>3 ) &7 ] |=1 << (code &7 );
_
Quoted 9 lines
drivers/net/sk_mca.c renamed from drivers/net/sk_mca.c~sk_mca-multicast-fix +1
-1@@ -997,13 +997,13 @@ static void skmca_set_multicast_list(str block.Mode &= ~LANCE_INIT_PROM ;if (dev->flags &IFF_ALLMULTI ) {/* get all multicasts */ -memset (block.LAdrF,8 ,0xff ); +memset (block.LAdrF,0xff ,sizeof (block.LAdrF));
drivers/net/sk_mca.c static void skmca_set_multicast_list(str · line 1009
Quoted 10 lines
drivers/net/sk_mca.c renamed from drivers/net/sk_mca.c~sk_mca-multicast-fix +1
-1}else {/* get selected/no multicasts */ struct dev_mc_list *mptr;int code; -memset (block.LAdrF,8 ,0x00 ); +memset (block.LAdrF,0 ,sizeof (block.LAdrF));for (mptr = dev->mc_list; mptr !=NULL ; mptr = mptr->next) { code =GetHash (mptr->dmi_addr); block.LAdrF[(code >>3 ) &7 ] |=1 << (code &7 );
Looks better ;)
While you're at it, please consider the following misc changes:
1* In net/sk_mca.h:
#define CSR3_BSWAP_OFF 0 /* Bit 2 = 0 -> no byte swap */ #define CSR3_BSWAP_ON 0 /* Bit 2 = 1 -> byte swap */
The second define should obviously read 4, not 0. It's harmess because the define isn't used anywhere, but still...
2* Alfred Arnold's alternate e-mail address, aarnold at elsa.de, looks dead. Maybe you could remove it from the two drivers it appears in:net/ibmlana.c and net/sk_mca.c. The first address still works. Quoting
Alfred: "ELSA (my old employer) went bankrupt about two years ago, you may remove this address or replace it with my current work address:
alfred.arnold at lancom.de".
Thanks.
Signature · 2 lines
Jean Delvare http://www.ensicaen.ismra.fr/~delvare/
Quoted 30 lines · Jean Delvare
[Please CC: me on reply, I'm not subscribed (yet)]
Hi all,
I just found two very weird memsets in
drivers/net/sk_mca.c. I am not working on that driver at all, but some grepping of the kernel source pointed me to it so I thought I wouldn't keep quiet about it.Here is the offending code:
static void
skmca_set_multicast_list(struct net_device *dev){ (...) if (dev->flags & IFF_ALLMULTI) { /* get all multicasts */ memset(block.LAdrF, 8, 0xff); } else { /* get selected/no multicasts */ (...) memset(block.LAdrF, 8, 0x00); (...) } (...) }Is it just me, or are these two memsets just plain broken? Not only the size is hardcoded, but the parameters are swapped! I guess that this driver never worked in multicast mode. The correct memsets are obviously:
memset(block.LAdrF, 0xff, sizeof(block.LAdrF)); and memset(block.LAdrF, 0x00, sizeof(block.LAdrF));
Quoted 7 lines
respectively.
The odd thing is that the bug is there since the driver was introduced in the 2.2.10 kernel. So, 2.2, 2.4 and 2.6 kernels are all affected.
I admit I'm a bit surprized that this could survive until today, so just tell me if it's me missing the point ;)
Good catch!
No, I don't think you're missing the point. There are lots of drivers, and unlike core kernel code, drivers have corner cases which are never tested. Work on drivers is always welcome.
Feel free to make patch and submit it to Jeff Garzik <[email protected]>. For 2.2 and 2.4, I think you can CC kernel maintainers too. --
vda