Skip to contentSkip to conversation

can: bittiming: fix two defects in the userspace bitrate conversion

PATCH New Sureshkumar S 2 participants 5 messages Aug 3 – Aug 4

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

cover can: bittiming: fix two defects in the userspace bitrate conversion

can_calc_bittiming() converts a userspace supplied bitrate into hardware timing parameters. Two defects in that conversion are fixed here, both reachable through IFLA_CAN_BITTIMING on any CAN device that provides a bittiming_const, and both caused by 32 bit arithmetic on a value that userspace fully controls.

Patch 1 fixes a divide-by-zero. The 32 bit tsegall * bt->bitrate product wraps to zero for bitrates carrying enough factors of two, and the following division faults. A bitrate of 16777216 is already enough, which is below the 20 Mbit/s CAN XL data bitrate ceiling.

Patch 2 fixes the candidate rating, which applies abs() to an unsigned subtraction. A wrapped difference is read back as a small error, so a bitrate far outside what the controller can reach passes the 5% gate and is silently accepted instead of rejected.

Both were found with a netlink fuzzer against dummy_can and verified under KASAN in QEMU. Testing covered the unpatched tree, patch 1 alone and the full series: bitrates from 125 kbit/s to 10 Mbit/s produce byte-identical bittiming at every step, patch 1 alone removes the oops, and patch 2 turns the silently accepted out-of-range bitrates into -EINVAL.

Sureshkumar S (2):
can: bittiming: fix divide-by-zero in can_calc_bittiming()
can: bittiming: fix bitrate error calculation on unsigned operands

 drivers/net/can/dev/calc_bittiming.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

base-commit: af39eb111ce6b5eba9c08513b62c4868eb7e7fd5

Signature · 1 line
2.43.0
#1 lore raw

PATCH 1/2 can: bittiming: fix divide-by-zero in can_calc_bittiming()

can_calc_bittiming() scans the possible time segment combinations and computes the prescaler for each of them as:

	brp = priv->clock.freq / (tsegall * bt->bitrate)

bt->bitrate is supplied by userspace via IFLA_CAN_BITTIMING and tsegall * bt->bitrate is a 32 bit multiplication, so the product wraps to zero as soon as bt->bitrate carries enough factors of two for the tsegall values walked by the loop. The division then faults:

  Oops: divide error: 0000 [#1] SMP KASAN NOPTI
  RIP: 0010:can_calc_bittiming+0x32e/0xcc0
  Call Trace:
   can_changelink+0x8ba/0x2060
   __rtnl_newlink+0x1013/0x18a0
   rtnl_newlink+0x6b/0xa0
   rtnetlink_rcv_msg+0x6f9/0xb70
   netlink_rcv_skb+0x11f/0x350
   netlink_unicast+0x5f5/0x860
   netlink_sendmsg+0x70a/0xba0

This does not require an absurd bitrate. With the segment limits of a typical controller tsegall reaches 256, so a bitrate of 16777216 is already enough to wrap the product, and that value is below the 20 Mbit/s CAN XL data bitrate ceiling. Any CAN driver providing a bittiming_const is affected; reproduced on dummy_can. Triggering it needs CAP_NET_ADMIN in the netns owning the device.

priv->bitrate_max cannot guard against this: it is populated from the optional "max-bitrate" device tree property, so it is zero for most drivers, and can_changelink() only consults it after can_get_bittiming() has already returned.

Compute the product with mul_u32_u32(), as can_fixup_bittiming() already does for bt->brp * NSEC_PER_SEC, and divide with div64_u64(). div_u64() cannot be used here because its divisor is a u32, which would truncate the product back to the faulting value.

Fixes 39549eef3587 ("can: CAN Network device driver and Netlink interface")
---
 drivers/net/can/dev/calc_bittiming.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
drivers/net/can/dev/calc_bittiming.c +6 -2
index 42498e9d3f38..4809f5e0c96e 100644
@@ -119,8 +119,12 @@ int can_calc_bittiming(const struct net_device *dev, struct can_bittiming *bt,
 	     tseg >= (btc->tseg1_min + btc->tseg2_min) * 2; tseg--) {
 		tsegall = CAN_SYNC_SEG + tseg / 2;
 
-		/* Compute all possible tseg choices (tseg=tseg1+tseg2) */
-		brp = priv->clock.freq / (tsegall * bt->bitrate) + tseg % 2;
+		/* Compute all possible tseg choices (tseg=tseg1+tseg2).
+		 * A 32 bit tsegall * bt->bitrate can wrap to zero for large
+		 * userspace bitrates, so compute the product in 64 bit.
+		 */
+		brp = div64_u64(priv->clock.freq,
+				mul_u32_u32(tsegall, bt->bitrate)) + tseg % 2;
 
 		/* choose brp step which is possible in system */
 		brp = (brp / btc->brp_inc) * btc->brp_inc;
Signature · 1 line
2.43.0
#2 lore raw

Re: [PATCH net 1/2] can: bittiming: fix divide-by-zero in can_calc_bittiming()

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Pre-existing issues:
- [High] Missing validation for hardware-provided bit-timing constants brp_inc and brp_min allows malicious or misbehaving CAN devices to trigger divide-by-zero kernel panics.
- [High] A 32-bit integer overflow in the brp * tsegall calculation can wrap to zero, causing a divide-by-zero kernel panic.
--

--- Patch [1]: [PATCH net 1/2] can: bittiming: fix divide-by-zero in can_calc_bittiming() --- commit 1d93cc0a0370ce5feb4636b0a8c6900dd609c628
Author: Sureshkumar S <[email protected]>

can: bittiming: fix divide-by-zero in can_calc_bittiming()

This commit patches a potential divide-by-zero in can_calc_bittiming() when calculating the prescaler. It uses 64-bit math to prevent the 32-bit integer multiplication of tsegall and bitrate from wrapping to zero for large bitrates.

Quoted 9 lines
drivers/net/can/dev/calc_bittiming.c +0 -1
index 42498e9d3f38d..4809f5e0c96e9 100644
@@ -119,8 +119,12 @@ int can_calc_bittiming(const struct net_device *dev, struct can_bittiming *bt,
 	     tseg >= (btc->tseg1_min + btc->tseg2_min) * 2; tseg--) {
 		tsegall = CAN_SYNC_SEG + tseg / 2;
 
-		/* Compute all possible tseg choices (tseg=tseg1+tseg2) */

drivers/net/can/dev/calc_bittiming.c int can_calc_bittiming(const struct net_device *dev, struct can_bittiming *bt, · line 130

Quoted 10 lines
drivers/net/can/dev/calc_bittiming.c +6 -1
-		brp = priv->clock.freq / (tsegall * bt->bitrate) + tseg % 2;
+		/* Compute all possible tseg choices (tseg=tseg1+tseg2).
+		 * A 32 bit tsegall * bt->bitrate can wrap to zero for large
+		 * userspace bitrates, so compute the product in 64 bit.
+		 */
+		brp = div64_u64(priv->clock.freq,
+				mul_u32_u32(tsegall, bt->bitrate)) + tseg % 2;
 
 		/* choose brp step which is possible in system */
 		brp = (brp / btc->brp_inc) * btc->brp_inc;

[Severity: High]
This is a pre-existing issue, but missing validation for hardware-provided bit-timing constants might allow a divide-by-zero here.

If a misbehaving or malicious USB adapter (like gs_usb) provides 0 for btc->brp_inc, won't this division directly panic the kernel when a user with CAP_NET_ADMIN configures the interface?

drivers/net/can/dev/calc_bittiming.c

Quoted 4 lines
drivers/net/can/dev/calc_bittiming.c +0 -0
 		if (brp < btc->brp_min || brp > btc->brp_max)
 			continue;
 
 		bitrate = priv->clock.freq / (brp * tsegall);

[Severity: High]
This isn't a bug introduced by this patch, but doesn't this calculation have the exact same 32-bit overflow vulnerability that the patch fixes above?

Since both brp and tsegall are 32-bit unsigned integers, couldn't their product wrap to zero for large values derived from untrusted hardware limits, causing a divide-by-zero panic on this line?

Additionally, if the hardware provides a btc->brp_min of 0, the bounds check above allows brp to evaluate to 0. Wouldn't that also result in a divide-by-zero here?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
#3 lore raw

PATCH 2/2 can: bittiming: fix bitrate error calculation on unsigned operands

In reply to Sureshkumar S

can_calc_bittiming() rates each candidate against the requested bitrate with:

	bitrate_error = abs(bt->bitrate - bitrate);

Both operands are unsigned int, so the subtraction wraps instead of becoming negative, and abs() resolves an unsigned int argument to its int branch. A wrapped difference is therefore reinterpreted as a small positive value instead of the large error it actually represents.

A requested bitrate far above anything the controller can reach then passes the CAN_CALC_MAX_ERROR gate. On a dummy_can device with a 160 MHz clock, requesting 4294967294 bps reports an error of 0.01%, configures 415584 bps and returns success to userspace, where -EINVAL is expected.

Use abs_diff(), which subtracts the smaller operand from the larger one and keeps the whole comparison unsigned.

Fixes 7da29f97d6c8 ("can: dev: can-calc-bit-timing(): better sample point calculation")
---
 drivers/net/can/dev/calc_bittiming.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/can/dev/calc_bittiming.c +1 -1
index 4809f5e0c96e..2789b99ab6a8 100644
@@ -132,7 +132,7 @@ int can_calc_bittiming(const struct net_device *dev, struct can_bittiming *bt,
 			continue;
 
 		bitrate = priv->clock.freq / (brp * tsegall);
-		bitrate_error = abs(bt->bitrate - bitrate);
+		bitrate_error = abs_diff(bt->bitrate, bitrate);
 
 		/* tseg brp biterror */
 		if (bitrate_error > best_bitrate_error)
Signature · 1 line
2.43.0
#4 lore raw

Re: [PATCH net 2/2] can: bittiming: fix bitrate error calculation on unsigned operands

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Pre-existing issues:
- [High] Unchecked bit-timing hardware constraints from USB devices can cause kernel panics due to division by zero in can_calc_bittiming().
- [High] An unchecked minimum time segment configuration from a malicious USB CAN device leads to an infinite loop, hanging the kernel. --

--- Patch [2]: [PATCH net 2/2] can: bittiming: fix bitrate error calculation on unsigned operands --- commit 93441c10d7d0dd29814ef57910bbfdffcd4e5112
Author: Sureshkumar S <[email protected]>

can: bittiming: fix bitrate error calculation on unsigned operands

This patch fixes a wrap-around issue in can_calc_bittiming() caused by subtracting unsigned operands when rating candidate bitrates. It updates the code to use abs_diff() so that large bitrate errors are correctly identified rather than being misinterpreted as small errors.

drivers/net/can/dev/calc_bittiming.c

Quoted 4 lines
drivers/net/can/dev/calc_bittiming.c +0 -0
index 4809f5e0c96e9..2789b99ab6a83 100644

[Severity: High]
This isn't a bug introduced by this patch, but could a malicious USB CAN device cause an infinite loop in this function?

Looking at can_calc_bittiming(), if a device advertises tseg1_min = 0 and tseg2_min = 0:

	for (tseg = (btc->tseg1_max + btc->tseg2_max) * 2 + 1;
	     tseg >= (btc->tseg1_min + btc->tseg2_min) * 2; tseg--) {

Since tseg is an unsigned int, the loop condition tseg >= 0 becomes a tautology. This would cause tseg to wrap around to UINT_MAX, resulting in an infinite loop that hangs the kernel.

drivers/net/can/dev/calc_bittiming.c int can_calc_bittiming(const struct net_device *dev, struct can_bittiming *bt, · line 134

Quoted 4 lines
drivers/net/can/dev/calc_bittiming.c +0 -0
@@ -132,7 +132,7 @@ int can_calc_bittiming(const struct net_device *dev, struct can_bittiming *bt,
 			continue;
 
 		bitrate = priv->clock.freq / (brp * tsegall);

[Severity: High]
This is also a pre-existing issue not introduced by this patch, but is it possible for this division to cause a kernel panic?

If a malicious USB device provides unvalidated values, could brp evaluate to 0 if btc->brp_min == 0 or priv->clock.freq == 0?

Also, earlier in the same function there is another division:

		brp = (brp / btc->brp_inc) * btc->brp_inc;

Could this divide by zero if btc->brp_inc is provided as 0 by the hardware?

Quoted 5 lines
drivers/net/can/dev/calc_bittiming.c +1 -1
-		bitrate_error = abs(bt->bitrate - bitrate);
+		bitrate_error = abs_diff(bt->bitrate, bitrate);
 
 		/* tseg brp biterror */
 		if (bitrate_error > best_bitrate_error)
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=2