In e1000_tbi_should_accept() we read the last byte of the frame via 'data[length - 1]' to evaluate the TBI workaround. If the descriptor- reported length is zero or larger than the actual RX buffer size, this read goes out of bounds and can hit unrelated slab objects. The issue is observed from the NAPI receive path (e1000_clean_rx_irq):
================================================================== BUG: KASAN: slab-out-of-bounds in e1000_tbi_should_accept+0x610/0x790 Read of size 1 at addr ffff888014114e54 by task sshd/363
CPU: 0 PID: 363 Comm: sshd Not tainted 5.18.0-rc1 #1 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.0-59-gc9ba5276e321-prebuilt.qemu.org 04/01/2014 Call Trace: <IRQ> dump_stack_lvl+0x5a/0x74 print_address_description+0x7b/0x440 print_report+0x101/0x200 kasan_report+0xc1/0xf0 e1000_tbi_should_accept+0x610/0x790 e1000_clean_rx_irq+0xa8c/0x1110 e1000_clean+0xde2/0x3c10 __napi_poll+0x98/0x380 net_rx_action+0x491/0xa20 __do_softirq+0x2c9/0x61d do_softirq+0xd1/0x120 </IRQ> <TASK> __local_bh_enable_ip+0xfe/0x130 ip_finish_output2+0x7d5/0xb00 __ip_queue_xmit+0xe24/0x1ab0 __tcp_transmit_skb+0x1bcb/0x3340 tcp_write_xmit+0x175d/0x6bd0 __tcp_push_pending_frames+0x7b/0x280 tcp_sendmsg_locked+0x2e4f/0x32d0 tcp_sendmsg+0x24/0x40 sock_write_iter+0x322/0x430 vfs_write+0x56c/0xa60 ksys_write+0xd1/0x190 do_syscall_64+0x43/0x90 entry_SYSCALL_64_after_hwframe+0x44/0xae RIP: 0033:0x7f511b476b10 Code: 73 01 c3 48 8b 0d 88 d3 2b 00 f7 d8 64 89 01 48 83 c8 ff c3 66 0f 1f 44 00 00 83 3d f9 2b 2c 00 00 75 10 b8 01 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 31 c3 48 83 ec 08 e8 8e 9b 01 00 48 89 04 24 RSP: 002b:00007ffc9211d4e8 EFLAGS: 00000246 ORIG_RAX: 0000000000000001 RAX: ffffffffffffffda RBX: 0000000000004024 RCX: 00007f511b476b10 RDX: 0000000000004024 RSI: 0000559a9385962c RDI: 0000000000000003 RBP: 0000559a9383a400 R08: fffffffffffffff0 R09: 0000000000004f00 R10: 0000000000000070 R11: 0000000000000246 R12: 0000000000000000 R13: 00007ffc9211d57f R14: 0000559a9347bde7 R15: 0000000000000003 </TASK> Allocated by task 1: __kasan_krealloc+0x131/0x1c0 krealloc+0x90/0xc0 add_sysfs_param+0xcb/0x8a0 kernel_add_sysfs_param+0x81/0xd4 param_sysfs_builtin+0x138/0x1a6 param_sysfs_init+0x57/0x5b do_one_initcall+0x104/0x250 do_initcall_level+0x102/0x132 do_initcalls+0x46/0x74 kernel_init_freeable+0x28f/0x393 kernel_init+0x14/0x1a0 ret_from_fork+0x22/0x30 The buggy address belongs to the object at ffff888014114000 which belongs to the cache kmalloc-2k of size 2048 The buggy address is located 1620 bytes to the right of 2048-byte region [ffff888014114000, ffff888014114800] The buggy address belongs to the physical page: page:ffffea0000504400 refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x14110 head:ffffea0000504400 order:3 compound_mapcount:0 compound_pincount:0 flags: 0x100000000010200(slab|head|node=0|zone=1) raw: 0100000000010200 0000000000000000 dead000000000001 ffff888013442000 raw: 0000000000000000 0000000000080008 00000001ffffffff 0000000000000000 page dumped because: kasan: bad access detected ==================================================================
This happens because the TBI check unconditionally dereferences the last byte without validating the reported length first:
u8 last_byte = *(data + length - 1);
Fix by rejecting the frame early if the length is zero, or if it exceeds adapter->rx_buffer_len. This preserves the TBI workaround semantics for valid frames and prevents touching memory beyond the RX buffer.
Fixes: 2037110c96d5 ("e1000: move tbi workaround code into helper function") Cc: stable@vger.kernel.org Suggested-by: Tony Nguyen tony.nguyen@intel.com Signed-off-by: Guangshuo Li lgs201920130244@gmail.com --- changelog: v2: - Keep declarations at the beginning of e1000_tbi_should_accept(). - Move the last_byte assignment after the length bounds checks (suggested by Tony Nguyen) --- drivers/net/ethernet/intel/e1000/e1000_main.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/drivers/net/ethernet/intel/e1000/e1000_main.c b/drivers/net/ethernet/intel/e1000/e1000_main.c index 3f5feb55cfba..cb49ec49f836 100644 --- a/drivers/net/ethernet/intel/e1000/e1000_main.c +++ b/drivers/net/ethernet/intel/e1000/e1000_main.c @@ -4091,7 +4091,14 @@ static bool e1000_tbi_should_accept(struct e1000_adapter *adapter, u32 length, const u8 *data) { struct e1000_hw *hw = &adapter->hw; - u8 last_byte = *(data + length - 1); + u8 last_byte; + /* Guard against OOB on data[length - 1] */ + if (unlikely(!length)) + return false; + /* Upper bound: length must not exceed rx_buffer_len */ + if (unlikely(length > adapter->rx_buffer_len)) + return false; + last_byte = *(data + length - 1);
if (TBI_ACCEPT(hw, status, errors, length, last_byte)) { unsigned long irq_flags;
+ Aleksandr
On Mon, Dec 01, 2025 at 11:40:58AM +0800, Guangshuo Li wrote:
In e1000_tbi_should_accept() we read the last byte of the frame via 'data[length - 1]' to evaluate the TBI workaround. If the descriptor- reported length is zero or larger than the actual RX buffer size, this read goes out of bounds and can hit unrelated slab objects. The issue is observed from the NAPI receive path (e1000_clean_rx_irq):
...
Fixes: 2037110c96d5 ("e1000: move tbi workaround code into helper function") Cc: stable@vger.kernel.org Suggested-by: Tony Nguyen tony.nguyen@intel.com Signed-off-by: Guangshuo Li lgs201920130244@gmail.com
changelog: v2:
- Keep declarations at the beginning of e1000_tbi_should_accept().
- Move the last_byte assignment after the length bounds checks (suggested by Tony Nguyen)
I'm not sure that Tony's suggestions warrant a Suggested-by tag. And perhaps Aleksandr's Reviewed-by tag should have been carried over from v1: IMHO, I don't think the changes between v1 and v2 materially effect the review. But overall this looks good to me.
Reviewed-by: Simon Horman horms@kernel.org
Dear Guangshuo,
Thank you for your patch.
Am 01.12.25 um 04:40 schrieb Guangshuo Li:
In e1000_tbi_should_accept() we read the last byte of the frame via 'data[length - 1]' to evaluate the TBI workaround. If the descriptor- reported length is zero or larger than the actual RX buffer size, this read goes out of bounds and can hit unrelated slab objects. The issue is observed from the NAPI receive path (e1000_clean_rx_irq):
================================================================== BUG: KASAN: slab-out-of-bounds in e1000_tbi_should_accept+0x610/0x790 Read of size 1 at addr ffff888014114e54 by task sshd/363
CPU: 0 PID: 363 Comm: sshd Not tainted 5.18.0-rc1 #1 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.0-59-gc9ba5276e321-prebuilt.qemu.org 04/01/2014 Call Trace:
<IRQ> dump_stack_lvl+0x5a/0x74 print_address_description+0x7b/0x440 print_report+0x101/0x200 kasan_report+0xc1/0xf0 e1000_tbi_should_accept+0x610/0x790 e1000_clean_rx_irq+0xa8c/0x1110 e1000_clean+0xde2/0x3c10 __napi_poll+0x98/0x380 net_rx_action+0x491/0xa20 __do_softirq+0x2c9/0x61d do_softirq+0xd1/0x120 </IRQ> <TASK> __local_bh_enable_ip+0xfe/0x130 ip_finish_output2+0x7d5/0xb00 __ip_queue_xmit+0xe24/0x1ab0 __tcp_transmit_skb+0x1bcb/0x3340 tcp_write_xmit+0x175d/0x6bd0 __tcp_push_pending_frames+0x7b/0x280 tcp_sendmsg_locked+0x2e4f/0x32d0 tcp_sendmsg+0x24/0x40 sock_write_iter+0x322/0x430 vfs_write+0x56c/0xa60 ksys_write+0xd1/0x190 do_syscall_64+0x43/0x90 entry_SYSCALL_64_after_hwframe+0x44/0xae RIP: 0033:0x7f511b476b10 Code: 73 01 c3 48 8b 0d 88 d3 2b 00 f7 d8 64 89 01 48 83 c8 ff c3 66 0f 1f 44 00 00 83 3d f9 2b 2c 00 00 75 10 b8 01 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 31 c3 48 83 ec 08 e8 8e 9b 01 00 48 89 04 24 RSP: 002b:00007ffc9211d4e8 EFLAGS: 00000246 ORIG_RAX: 0000000000000001 RAX: ffffffffffffffda RBX: 0000000000004024 RCX: 00007f511b476b10 RDX: 0000000000004024 RSI: 0000559a9385962c RDI: 0000000000000003 RBP: 0000559a9383a400 R08: fffffffffffffff0 R09: 0000000000004f00 R10: 0000000000000070 R11: 0000000000000246 R12: 0000000000000000 R13: 00007ffc9211d57f R14: 0000559a9347bde7 R15: 0000000000000003 </TASK> Allocated by task 1: __kasan_krealloc+0x131/0x1c0 krealloc+0x90/0xc0 add_sysfs_param+0xcb/0x8a0 kernel_add_sysfs_param+0x81/0xd4 param_sysfs_builtin+0x138/0x1a6 param_sysfs_init+0x57/0x5b do_one_initcall+0x104/0x250 do_initcall_level+0x102/0x132 do_initcalls+0x46/0x74 kernel_init_freeable+0x28f/0x393 kernel_init+0x14/0x1a0 ret_from_fork+0x22/0x30 The buggy address belongs to the object at ffff888014114000 which belongs to the cache kmalloc-2k of size 2048 The buggy address is located 1620 bytes to the right of 2048-byte region [ffff888014114000, ffff888014114800] The buggy address belongs to the physical page: page:ffffea0000504400 refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x14110 head:ffffea0000504400 order:3 compound_mapcount:0 compound_pincount:0 flags: 0x100000000010200(slab|head|node=0|zone=1) raw: 0100000000010200 0000000000000000 dead000000000001 ffff888013442000 raw: 0000000000000000 0000000000080008 00000001ffffffff 0000000000000000 page dumped because: kasan: bad access detected ==================================================================
This happens because the TBI check unconditionally dereferences the last byte without validating the reported length first:
u8 last_byte = *(data + length - 1);
Fix by rejecting the frame early if the length is zero, or if it exceeds adapter->rx_buffer_len. This preserves the TBI workaround semantics for valid frames and prevents touching memory beyond the RX buffer.
Do you have reproducer to forth an invalid length?
Fixes: 2037110c96d5 ("e1000: move tbi workaround code into helper function") Cc: stable@vger.kernel.org Suggested-by: Tony Nguyen tony.nguyen@intel.com Signed-off-by: Guangshuo Li lgs201920130244@gmail.com
changelog: v2:
- Keep declarations at the beginning of e1000_tbi_should_accept().
- Move the last_byte assignment after the length bounds checks (suggested by Tony Nguyen)
drivers/net/ethernet/intel/e1000/e1000_main.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/drivers/net/ethernet/intel/e1000/e1000_main.c b/drivers/net/ethernet/intel/e1000/e1000_main.c index 3f5feb55cfba..cb49ec49f836 100644 --- a/drivers/net/ethernet/intel/e1000/e1000_main.c +++ b/drivers/net/ethernet/intel/e1000/e1000_main.c @@ -4091,7 +4091,14 @@ static bool e1000_tbi_should_accept(struct e1000_adapter *adapter, u32 length, const u8 *data) { struct e1000_hw *hw = &adapter->hw;
- u8 last_byte = *(data + length - 1);
- u8 last_byte;
- /* Guard against OOB on data[length - 1] */
- if (unlikely(!length))
return false;- /* Upper bound: length must not exceed rx_buffer_len */
- if (unlikely(length > adapter->rx_buffer_len))
Should an error be logged, or is it a common scenario, that such traffic exists?
return false;- last_byte = *(data + length - 1);
if (TBI_ACCEPT(hw, status, errors, length, last_byte)) { unsigned long irq_flags;
Kind regards,
Paul
[Cc: Remove bouncing address of Jeffrey and update Jesse’s]
Am 01.12.25 um 14:18 schrieb Paul Menzel:
Dear Guangshuo,
Thank you for your patch.
Am 01.12.25 um 04:40 schrieb Guangshuo Li:
In e1000_tbi_should_accept() we read the last byte of the frame via 'data[length - 1]' to evaluate the TBI workaround. If the descriptor- reported length is zero or larger than the actual RX buffer size, this read goes out of bounds and can hit unrelated slab objects. The issue is observed from the NAPI receive path (e1000_clean_rx_irq):
================================================================== BUG: KASAN: slab-out-of-bounds in e1000_tbi_should_accept+0x610/0x790 Read of size 1 at addr ffff888014114e54 by task sshd/363
CPU: 0 PID: 363 Comm: sshd Not tainted 5.18.0-rc1 #1 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.0-59-gc9ba5276e321-prebuilt.qemu.org 04/01/2014 Call Trace: <IRQ> dump_stack_lvl+0x5a/0x74 print_address_description+0x7b/0x440 print_report+0x101/0x200 kasan_report+0xc1/0xf0 e1000_tbi_should_accept+0x610/0x790 e1000_clean_rx_irq+0xa8c/0x1110 e1000_clean+0xde2/0x3c10 __napi_poll+0x98/0x380 net_rx_action+0x491/0xa20 __do_softirq+0x2c9/0x61d do_softirq+0xd1/0x120 </IRQ> <TASK> __local_bh_enable_ip+0xfe/0x130 ip_finish_output2+0x7d5/0xb00 __ip_queue_xmit+0xe24/0x1ab0 __tcp_transmit_skb+0x1bcb/0x3340 tcp_write_xmit+0x175d/0x6bd0 __tcp_push_pending_frames+0x7b/0x280 tcp_sendmsg_locked+0x2e4f/0x32d0 tcp_sendmsg+0x24/0x40 sock_write_iter+0x322/0x430 vfs_write+0x56c/0xa60 ksys_write+0xd1/0x190 do_syscall_64+0x43/0x90 entry_SYSCALL_64_after_hwframe+0x44/0xae RIP: 0033:0x7f511b476b10 Code: 73 01 c3 48 8b 0d 88 d3 2b 00 f7 d8 64 89 01 48 83 c8 ff c3 66 0f 1f 44 00 00 83 3d f9 2b 2c 00 00 75 10 b8 01 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 31 c3 48 83 ec 08 e8 8e 9b 01 00 48 89 04 24 RSP: 002b:00007ffc9211d4e8 EFLAGS: 00000246 ORIG_RAX: 0000000000000001 RAX: ffffffffffffffda RBX: 0000000000004024 RCX: 00007f511b476b10 RDX: 0000000000004024 RSI: 0000559a9385962c RDI: 0000000000000003 RBP: 0000559a9383a400 R08: fffffffffffffff0 R09: 0000000000004f00 R10: 0000000000000070 R11: 0000000000000246 R12: 0000000000000000 R13: 00007ffc9211d57f R14: 0000559a9347bde7 R15: 0000000000000003 </TASK> Allocated by task 1: __kasan_krealloc+0x131/0x1c0 krealloc+0x90/0xc0 add_sysfs_param+0xcb/0x8a0 kernel_add_sysfs_param+0x81/0xd4 param_sysfs_builtin+0x138/0x1a6 param_sysfs_init+0x57/0x5b do_one_initcall+0x104/0x250 do_initcall_level+0x102/0x132 do_initcalls+0x46/0x74 kernel_init_freeable+0x28f/0x393 kernel_init+0x14/0x1a0 ret_from_fork+0x22/0x30 The buggy address belongs to the object at ffff888014114000 which belongs to the cache kmalloc-2k of size 2048 The buggy address is located 1620 bytes to the right of 2048-byte region [ffff888014114000, ffff888014114800] The buggy address belongs to the physical page: page:ffffea0000504400 refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x14110 head:ffffea0000504400 order:3 compound_mapcount:0 compound_pincount:0 flags: 0x100000000010200(slab|head|node=0|zone=1) raw: 0100000000010200 0000000000000000 dead000000000001 ffff888013442000 raw: 0000000000000000 0000000000080008 00000001ffffffff 0000000000000000 page dumped because: kasan: bad access detected ==================================================================
This happens because the TBI check unconditionally dereferences the last byte without validating the reported length first:
u8 last_byte = *(data + length - 1);
Fix by rejecting the frame early if the length is zero, or if it exceeds adapter->rx_buffer_len. This preserves the TBI workaround semantics for valid frames and prevents touching memory beyond the RX buffer.
Do you have reproducer to forth an invalid length?
Fixes: 2037110c96d5 ("e1000: move tbi workaround code into helper function") Cc: stable@vger.kernel.org Suggested-by: Tony Nguyen tony.nguyen@intel.com Signed-off-by: Guangshuo Li lgs201920130244@gmail.com
changelog: v2:
- Keep declarations at the beginning of e1000_tbi_should_accept().
- Move the last_byte assignment after the length bounds checks
(suggested by Tony Nguyen)
drivers/net/ethernet/intel/e1000/e1000_main.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/drivers/net/ethernet/intel/e1000/e1000_main.c b/drivers/net/ethernet/intel/e1000/e1000_main.c index 3f5feb55cfba..cb49ec49f836 100644 --- a/drivers/net/ethernet/intel/e1000/e1000_main.c +++ b/drivers/net/ethernet/intel/e1000/e1000_main.c @@ -4091,7 +4091,14 @@ static bool e1000_tbi_should_accept(struct e1000_adapter *adapter, u32 length, const u8 *data) { struct e1000_hw *hw = &adapter->hw; - u8 last_byte = *(data + length - 1); + u8 last_byte; + /* Guard against OOB on data[length - 1] */ + if (unlikely(!length)) + return false; + /* Upper bound: length must not exceed rx_buffer_len */ + if (unlikely(length > adapter->rx_buffer_len))
Should an error be logged, or is it a common scenario, that such traffic exists?
+ return false; + last_byte = *(data + length - 1); if (TBI_ACCEPT(hw, status, errors, length, last_byte)) { unsigned long irq_flags;
Kind regards,
Paul
-----Original Message----- From: Intel-wired-lan intel-wired-lan-bounces@osuosl.org On Behalf Of Guangshuo Li Sent: Monday, December 1, 2025 4:41 AM To: Jesse Brandeburg jesse.brandeburg@intel.com; Nguyen, Anthony L anthony.l.nguyen@intel.com; David S. Miller davem@davemloft.net; Jakub Kicinski kuba@kernel.org; Paolo Abeni pabeni@redhat.com; Florian Westphal fw@strlen.de; Jeff Kirsher jeffrey.t.kirsher@intel.com; intel-wired-lan@lists.osuosl.org; netdev@vger.kernel.org; linux-kernel@vger.kernel.org Cc: Guangshuo Li lgs201920130244@gmail.com; stable@vger.kernel.org; Nguyen, Tony tony.nguyen@intel.com Subject: [Intel-wired-lan] [PATCH v2] e1000: fix OOB in e1000_tbi_should_accept()
In e1000_tbi_should_accept() we read the last byte of the frame via 'data[length - 1]' to evaluate the TBI workaround. If the descriptor- reported length is zero or larger than the actual RX buffer size, this read goes out of bounds and can hit unrelated slab objects. The issue is observed from the NAPI receive path (e1000_clean_rx_irq):
================================================================== BUG: KASAN: slab-out-of-bounds in e1000_tbi_should_accept+0x610/0x790 Read of size 1 at addr ffff888014114e54 by task sshd/363
CPU: 0 PID: 363 Comm: sshd Not tainted 5.18.0-rc1 #1 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.0-59- gc9ba5276e321-prebuilt.qemu.org 04/01/2014 Call Trace:
<IRQ> dump_stack_lvl+0x5a/0x74 print_address_description+0x7b/0x440 print_report+0x101/0x200 kasan_report+0xc1/0xf0 e1000_tbi_should_accept+0x610/0x790 e1000_clean_rx_irq+0xa8c/0x1110 e1000_clean+0xde2/0x3c10 __napi_poll+0x98/0x380 net_rx_action+0x491/0xa20 __do_softirq+0x2c9/0x61d do_softirq+0xd1/0x120 </IRQ> <TASK> __local_bh_enable_ip+0xfe/0x130 ip_finish_output2+0x7d5/0xb00 __ip_queue_xmit+0xe24/0x1ab0 __tcp_transmit_skb+0x1bcb/0x3340 tcp_write_xmit+0x175d/0x6bd0 __tcp_push_pending_frames+0x7b/0x280 tcp_sendmsg_locked+0x2e4f/0x32d0 tcp_sendmsg+0x24/0x40 sock_write_iter+0x322/0x430 vfs_write+0x56c/0xa60 ksys_write+0xd1/0x190 do_syscall_64+0x43/0x90 entry_SYSCALL_64_after_hwframe+0x44/0xae RIP: 0033:0x7f511b476b10 Code: 73 01 c3 48 8b 0d 88 d3 2b 00 f7 d8 64 89 01 48 83 c8 ff c3 66 0f 1f 44 00 00 83 3d f9 2b 2c 00 00 75 10 b8 01 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 31 c3 48 83 ec 08 e8 8e 9b 01 00 48 89 04 24 RSP: 002b:00007ffc9211d4e8 EFLAGS: 00000246 ORIG_RAX: 0000000000000001 RAX: ffffffffffffffda RBX: 0000000000004024 RCX: 00007f511b476b10 RDX: 0000000000004024 RSI: 0000559a9385962c RDI: 0000000000000003 RBP: 0000559a9383a400 R08: fffffffffffffff0 R09: 0000000000004f00 R10: 0000000000000070 R11: 0000000000000246 R12: 0000000000000000 R13: 00007ffc9211d57f R14: 0000559a9347bde7 R15: 0000000000000003 </TASK> Allocated by task 1: __kasan_krealloc+0x131/0x1c0 krealloc+0x90/0xc0 add_sysfs_param+0xcb/0x8a0 kernel_add_sysfs_param+0x81/0xd4 param_sysfs_builtin+0x138/0x1a6 param_sysfs_init+0x57/0x5b do_one_initcall+0x104/0x250 do_initcall_level+0x102/0x132 do_initcalls+0x46/0x74 kernel_init_freeable+0x28f/0x393 kernel_init+0x14/0x1a0 ret_from_fork+0x22/0x30 The buggy address belongs to the object at ffff888014114000 which belongs to the cache kmalloc-2k of size 2048 The buggy address is located 1620 bytes to the right of 2048-byte region [ffff888014114000, ffff888014114800] The buggy address belongs to the physical page: page:ffffea0000504400 refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x14110 head:ffffea0000504400 order:3 compound_mapcount:0 compound_pincount:0 flags: 0x100000000010200(slab|head|node=0|zone=1) raw: 0100000000010200 0000000000000000 dead000000000001 ffff888013442000 raw: 0000000000000000 0000000000080008 00000001ffffffff 0000000000000000 page dumped because: kasan: bad access detected ==================================================================
This happens because the TBI check unconditionally dereferences the last byte without validating the reported length first:
u8 last_byte = *(data + length - 1);
Fix by rejecting the frame early if the length is zero, or if it exceeds adapter->rx_buffer_len. This preserves the TBI workaround semantics for valid frames and prevents touching memory beyond the RX buffer.
Fixes: 2037110c96d5 ("e1000: move tbi workaround code into helper function") Cc: stable@vger.kernel.org Suggested-by: Tony Nguyen tony.nguyen@intel.com Signed-off-by: Guangshuo Li lgs201920130244@gmail.com
changelog: v2:
- Keep declarations at the beginning of e1000_tbi_should_accept().
- Move the last_byte assignment after the length bounds checks
(suggested by Tony Nguyen)
drivers/net/ethernet/intel/e1000/e1000_main.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/drivers/net/ethernet/intel/e1000/e1000_main.c b/drivers/net/ethernet/intel/e1000/e1000_main.c index 3f5feb55cfba..cb49ec49f836 100644 --- a/drivers/net/ethernet/intel/e1000/e1000_main.c +++ b/drivers/net/ethernet/intel/e1000/e1000_main.c @@ -4091,7 +4091,14 @@ static bool e1000_tbi_should_accept(struct e1000_adapter *adapter, u32 length, const u8 *data) { struct e1000_hw *hw = &adapter->hw;
- u8 last_byte = *(data + length - 1);
u8 last_byte;
/* Guard against OOB on data[length - 1] */
if (unlikely(!length))
return false;/* Upper bound: length must not exceed rx_buffer_len */
if (unlikely(length > adapter->rx_buffer_len))
return false;last_byte = *(data + length - 1);
if (TBI_ACCEPT(hw, status, errors, length, last_byte)) { unsigned long irq_flags;
-- 2.43.0
Reviewed-by: Aleksandr Loktionov aleksandr.loktionov@intel.com
Hi Paul,
thanks for your comments.
----Do you have reproducer to forth an invalid length?
Yes. The issue is reproducible with a concrete system call sequence.
I am running on top of a fuzzer called PrIntFuzz, which is built on syzkaller. PrIntFuzz adds a custom syscall syz_prepare_data() that is used to simulate device input. In other words, the device side traffic is not coming from a real hardware device, but is deliberately constructed by the fuzzer through syz_prepare_data().
The exact reproducer is provided in the attached syzkaller program (system call sequence) generated by PrIntFuzz, which consistently triggers the invalid length and the crash on my setup.
(The attached program is exactly the sequence I am running to reproduce the problem.)
----Should an error be logged, or is it a common scenario, that such traffic exists?
In normal deployments, I don’t expect such traffic from a well-behaved I2C device. In my case, the malformed length only appears because PrIntFuzz is intentionally crafting invalid inputs and feeding them to the driver via syz_prepare_data(). So this is not a “common” or expected scenario in real-world use, but it is a realistic attacker/fuzzer scenario, since the length field can be controlled by an external peer/device.
Given that, I think the driver should treat an invalid length as an error and fail the request instead of trusting it and risking memory corruption.
Regarding logging, I’m fully open to your preference. From my point of view, logging this as an error seems reasonable, because it indicates malformed or buggy input from the device side. However, if you expect this condition might occur more frequently in practice and would prefer to reduce noise, I can switch it to dev_dbg() or even drop the log entirely.
Please let me know which logging level you would prefer, and I will update the patch accordingly.
Best regards, Guangshuo Li
Dear Guangshuo,
Thank you for your quick and insightful reply. (No need to resend this often.)
Am 02.12.25 um 12:34 schrieb Guangshuo Li:
thanks for your comments.
----Do you have reproducer to forth an invalid length?
Yes. The issue is reproducible with a concrete system call sequence.
I am running on top of a fuzzer called PrIntFuzz, which is built on syzkaller. PrIntFuzz adds a custom syscall syz_prepare_data() that is used to simulate device input. In other words, the device side traffic is not coming from a real hardware device, but is deliberately constructed by the fuzzer through syz_prepare_data().
The exact reproducer is provided in the attached syzkaller program (system call sequence) generated by PrIntFuzz, which consistently triggers the invalid length and the crash on my setup.
(The attached program is exactly the sequence I am running to reproduce the problem.)
Thank you for attaching it. Excuse my ignorance, but how do I run it?
----Should an error be logged, or is it a common scenario, that such traffic exists?
In normal deployments, I don’t expect such traffic from a well-behaved I2C device. In my case, the malformed length only appears because PrIntFuzz is intentionally crafting invalid inputs and feeding them to the driver via syz_prepare_data(). So this is not a “common” or expected scenario in real-world use, but it is a realistic attacker/fuzzer scenario, since the length field can be controlled by an external peer/device.
Given that, I think the driver should treat an invalid length as an error and fail the request instead of trusting it and risking memory corruption.
Regarding logging, I’m fully open to your preference. From my point of view, logging this as an error seems reasonable, because it indicates malformed or buggy input from the device side. However, if you expect this condition might occur more frequently in practice and would prefer to reduce noise, I can switch it to dev_dbg() or even drop the log entirely.
Please let me know which logging level you would prefer, and I will update the patch accordingly.
Then I’d suggest to add an error message with error level so people notice and can take a look.
Kind regards,
Paul
Hi Paul,
Thanks for your reply.
To run the reproducer, you'll first need to download the PrIntFuzz project from https://github.com/vul337/PrIntFuzz and set up the environment. Once that is done, you should be able to run the attached syzkaller program to reproduce the issue.
Kind regards, Guangshuo
Paul Menzel pmenzel@molgen.mpg.de 于2025年12月2日周二 19:53写道：
Dear Guangshuo,
Thank you for your quick and insightful reply. (No need to resend this often.)
Am 02.12.25 um 12:34 schrieb Guangshuo Li:
thanks for your comments.
----Do you have reproducer to forth an invalid length?
Yes. The issue is reproducible with a concrete system call sequence.
I am running on top of a fuzzer called PrIntFuzz, which is built on syzkaller. PrIntFuzz adds a custom syscall syz_prepare_data() that is used to simulate device input. In other words, the device side traffic is not coming from a real hardware device, but is deliberately constructed by the fuzzer through syz_prepare_data().
The exact reproducer is provided in the attached syzkaller program (system call sequence) generated by PrIntFuzz, which consistently triggers the invalid length and the crash on my setup.
(The attached program is exactly the sequence I am running to reproduce the problem.)
Thank you for attaching it. Excuse my ignorance, but how do I run it?
----Should an error be logged, or is it a common scenario, that such traffic exists?
In normal deployments, I don’t expect such traffic from a well-behaved I2C device. In my case, the malformed length only appears because PrIntFuzz is intentionally crafting invalid inputs and feeding them to the driver via syz_prepare_data(). So this is not a “common” or expected scenario in real-world use, but it is a realistic attacker/fuzzer scenario, since the length field can be controlled by an external peer/device.
Given that, I think the driver should treat an invalid length as an error and fail the request instead of trusting it and risking memory corruption.
Regarding logging, I’m fully open to your preference. From my point of view, logging this as an error seems reasonable, because it indicates malformed or buggy input from the device side. However, if you expect this condition might occur more frequently in practice and would prefer to reduce noise, I can switch it to dev_dbg() or even drop the log entirely.
Please let me know which logging level you would prefer, and I will update the patch accordingly.
Then I’d suggest to add an error message with error level so people notice and can take a look.
Kind regards,
Paul
linux-stable-mirror@lists.linaro.org
-
Guangshuo Li -
Loktionov, Aleksandr -
Paul Menzel -
Simon Horman