Skip to contentSkip to conversation

rt2x00: fix memory corruption caused by eeprom buffer overflow

PATCH In review Pavel Roskin 2 participants 5 messages Feb 19, 2007 Networking drivers (wireless)

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 rt2x00: fix memory corruption caused by eeprom buffer overflow

eeprom_93cx6_multiread() expects the last argument to be the buffer length in words, but kzalloc() expects the length in bytes. This results in dangerous kernel memory corruption.

Since there are already occurrences of "EEPROM_SIZE * sizeof(u16)" in the driver, I'm assuming that EEPROM_SIZE is in words, so the driver needs to allocate more memory.

---

 drivers/net/wireless/d80211/rt2x00/rt2400pci.c |    2 +-
 drivers/net/wireless/d80211/rt2x00/rt2500pci.c |    2 +-
 drivers/net/wireless/d80211/rt2x00/rt2500usb.c |    2 +-
 drivers/net/wireless/d80211/rt2x00/rt61pci.c   |    2 +-
 drivers/net/wireless/d80211/rt2x00/rt73usb.c   |    2 +-
 5 files changed, 5 insertions(+), 5 deletions(-)
drivers/net/wireless/d80211/rt2x00/rt2400pci.c +1 -1
index 2e3a514..ef47554 100644
@@ -2370,7 +2370,7 @@ static int rt2400pci_alloc_eeprom(struct rt2x00_dev *rt2x00dev)
 	 * Allocate the eeprom memory, check the eeprom width
 	 * and copy the entire eeprom into this allocated memory.
 	 */
-	rt2x00dev->eeprom = kzalloc(EEPROM_SIZE, GFP_KERNEL);
+	rt2x00dev->eeprom = kzalloc(EEPROM_SIZE * sizeof(u16), GFP_KERNEL);
 	if (!rt2x00dev->eeprom)
 		return -ENOMEM;
 
drivers/net/wireless/d80211/rt2x00/rt2500pci.c +1 -1
index 305cff6..1085978 100644
@@ -2526,7 +2526,7 @@ static int rt2500pci_alloc_eeprom(struct rt2x00_dev *rt2x00dev)
 	 * Allocate the eeprom memory, check the eeprom width
 	 * and copy the entire eeprom into this allocated memory.
 	 */
-	rt2x00dev->eeprom = kzalloc(EEPROM_SIZE, GFP_KERNEL);
+	rt2x00dev->eeprom = kzalloc(EEPROM_SIZE * sizeof(u16), GFP_KERNEL);
 	if (!rt2x00dev->eeprom)
 		return -ENOMEM;
 
drivers/net/wireless/d80211/rt2x00/rt2500usb.c +1 -1
index 0976c98..0ace302 100644
@@ -2386,7 +2386,7 @@ static int rt2500usb_alloc_eeprom(struct rt2x00_dev *rt2x00dev)
 	 * Allocate the eeprom memory, check the eeprom width
 	 * and copy the entire eeprom into this allocated memory.
 	 */
-	rt2x00dev->eeprom = kzalloc(EEPROM_SIZE, GFP_KERNEL);
+	rt2x00dev->eeprom = kzalloc(EEPROM_SIZE * sizeof(u16), GFP_KERNEL);
 	if (!rt2x00dev->eeprom)
 		return -ENOMEM;
 
drivers/net/wireless/d80211/rt2x00/rt61pci.c +1 -1
index 547c660..46c552f 100644
@@ -3015,7 +3015,7 @@ static int rt61pci_alloc_eeprom(struct rt2x00_dev *rt2x00dev)
 	 * Allocate the eeprom memory, check the eeprom width
 	 * and copy the entire eeprom into this allocated memory.
 	 */
-	rt2x00dev->eeprom = kzalloc(EEPROM_SIZE, GFP_KERNEL);
+	rt2x00dev->eeprom = kzalloc(EEPROM_SIZE * sizeof(u16), GFP_KERNEL);
 	if (!rt2x00dev->eeprom)
 		return -ENOMEM;
 
drivers/net/wireless/d80211/rt2x00/rt73usb.c +1 -1
index be66e2f..6fa5a7e 100644
@@ -2705,7 +2705,7 @@ static int rt73usb_alloc_eeprom(struct rt2x00_dev *rt2x00dev)
 	 * Allocate the eeprom memory, check the eeprom width
 	 * and copy the entire eeprom into this allocated memory.
 	 */
-	rt2x00dev->eeprom = kzalloc(EEPROM_SIZE, GFP_KERNEL);
+	rt2x00dev->eeprom = kzalloc(EEPROM_SIZE, GFP_KERNEL * sizeof(u16));
 	if (!rt2x00dev->eeprom)
 		return -ENOMEM;
 
#1 lore raw
Quoted 9 lines

eeprom_93cx6_multiread() expects the last argument to be the buffer length in words, but kzalloc() expects the length in bytes. This results in dangerous kernel memory corruption.

Since there are already occurrences of "EEPROM_SIZE * sizeof(u16)" in the driver, I'm assuming that EEPROM_SIZE is in words, so the driver needs to allocate more memory.

ACK, this fix has been in CVS already, but I hadn't send the patch yet.

Ivo

#2 lore raw
In reply to Pavel Roskin

hi

drivers/net/wireless/d80211/rt2x00/rt73usb.c static int rt73usb_alloc_eeprom(struct rt2x00_dev *rt2x00dev) · line 2710

Quoted 12 lines
drivers/net/wireless/d80211/rt2x00/rt73usb.c +1 -1
index be66e2f..6fa5a7e 100644
@@ -2705,7 +2705,7 @@ static int rt73usb_alloc_eeprom(struct rt2x00_dev *rt2x00dev)
         * Allocate the eeprom memory, check the eeprom width
         * and copy the entire eeprom into this allocated memory.
         */
-       rt2x00dev->eeprom = kzalloc(EEPROM_SIZE, GFP_KERNEL);
+       rt2x00dev->eeprom = kzalloc(EEPROM_SIZE, GFP_KERNEL * sizeof(u16));
        if (!rt2x00dev->eeprom)
                return -ENOMEM;

you might want to change this to EEPROM_SIZE * sizeof(u16). ;)

Ivo

#3 lore raw
Quoted 16 lines · Ivo Van Doorn

hi

drivers/net/wireless/d80211/rt2x00/rt73usb.c static int rt73usb_alloc_eeprom(struct rt2x00_dev *rt2x00dev) · line 2710

Quoted 12 lines
drivers/net/wireless/d80211/rt2x00/rt73usb.c +1 -1
index be66e2f..6fa5a7e 100644
@@ -2705,7 +2705,7 @@ static int rt73usb_alloc_eeprom(struct rt2x00_dev *rt2x00dev)
         * Allocate the eeprom memory, check the eeprom width
         * and copy the entire eeprom into this allocated memory.
         */
-       rt2x00dev->eeprom = kzalloc(EEPROM_SIZE, GFP_KERNEL);
+       rt2x00dev->eeprom = kzalloc(EEPROM_SIZE, GFP_KERNEL * sizeof(u16));
        if (!rt2x00dev->eeprom)
                return -ENOMEM;

you might want to change this to EEPROM_SIZE * sizeof(u16). ;)

Shame on me. I wrote a patch that worked on rt61, and then I discarded it to make a "clean" fix for all drivers.

But anyway, I'm seeing now that CVS on sourceforge.net is not abandoned, so I think the best approach would be to integrate it into the kernel. It would be great if you remove dead code from CVS, all those "experimental" branches that haven't been touched for months if not years. It was hard for me to find the actual sources. I would have spent last Sunday evening doing something more interesting than fixing fixed bugs :)

Finally, I'll appreciate if the driver is fixed to compile against wireless-dev kernels. You can take an approach similar to iwlwifi and even rely on the same d80211 package.

Signature · 2 lines
Regards,
Pavel Roskin
#4 lore raw

Hi,

Quoted 10 lines

Shame on me. I wrote a patch that worked on rt61, and then I discarded it to make a "clean" fix for all drivers.

But anyway, I'm seeing now that CVS on sourceforge.net is not abandoned, so I think the best approach would be to integrate it into the kernel. It would be great if you remove dead code from CVS, all those "experimental" branches that haven't been touched for months if not years. It was hard for me to find the actual sources. I would have spent last Sunday evening doing something more interesting than fixing fixed bugs :)

True true, everything useful is in the "source" module of the CVS tree. Along with the legacy drivers for each chipset.
I thought the other older tree was already removed... :(

Quoted 3 lines

Finally, I'll appreciate if the driver is fixed to compile against wireless-dev kernels. You can take an approach similar to iwlwifi and even rely on the same d80211 package.

Well the d80211 stack inside rt2x00 is slightly modified to contain compatibility
fixes for regular kernels. That way regular users can test rt2x00, which is extremely helpful.
Other than that the stack also contains small patches that are in testing and I am sending
to the wireless list when it is stable. At this moment I have for example the sequence
counter patch that is currently under discussion on this list, as well as the currently
antenna patch that will split antenna_sel into a RX and TX antenna selection.

Ivo