[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <20250826162500.34292-1-qianjiaru77@gmail.com>
Date: Wed, 27 Aug 2025 00:25:00 +0800
From: qianjiaru77@...il.com
To: steffen.klassert@...unet.com,
herbert@...dor.apana.org.au,
davem@...emloft.net,
edumazet@...gle.com,
kuba@...nel.org,
pabeni@...hat.com
Cc: horms@...nel.org,
netdev@...r.kernel.org,
linux-kernel@...r.kernel.org,
qianjiaru <qianjiaru77@...il.com>
Subject: [PATCH 1/1] Reference Counting Vulnerability in Linux XFRM PolicyManagement
From: qianjiaru <qianjiaru77@...il.com>
A reference counting management vulnerability exists in the
Linux kernel's XFRM (IPsec Transform) policy subsystem.
Based on variant analysis of CVE-2022-36879, this
vulnerability involves improper policy object lifecycle
management in the `__xfrm_policy_check()` function,
potentially leading to double free conditions
and system instability.
## Vulnerability Mechanism
The issue follows the same pattern as CVE-2022-36879 but in a different code path:
1. **Policy Reference Acquired**:
`pols[0]` contains a valid policy with acquired reference
2. **Secondary Lookup Fails**:
`xfrm_policy_lookup_bytype()` returns error for `pols[1]`
3. **Partial Cleanup**:
Error path calls `xfrm_pol_put(pols[0])` but doesn't clear policy array state
4. **Caller Confusion**:
Calling function may not be aware that `pols[0]` has already been released
5. **Double Release Risk**:
If caller attempts cleanup based on incomplete state information
## Comparison with CVE-2022-36879
**CVE-2022-36879 (Fixed)**:
`xfrm_expand_policies()` had double reference counting issues
```c
// Original vulnerability pattern (now fixed):
if (IS_ERR(pols[1])) {
xfrm_pols_put(pols, *num_pols); // Released references
return PTR_ERR(pols[1]); // But didn't set *num_pols = 0
}
```
**This Variant**:
Similar reference management issues in different function
```c
// Current potential issue:
if (IS_ERR(pols[1])) {
xfrm_pol_put(pols[0]); // Releases pols[0]
return 0; // But pols[0] pointer remains unchanged
}
```
## Attack Scenario
1. **Policy Lookup**:
Network packet triggers `__xfrm_policy_check()`
with sub-policy configuration
2. **Primary Policy Found**:
`pols[0]` gets a valid policy reference
3. **Secondary Lookup Fails**:
`xfrm_policy_lookup_bytype()` fails, returns error in `pols[1]`
4. **Partial Cleanup**:
Function releases `pols[0]` reference but doesn't clear the pointer
5. **Caller Misunderstanding**:
Calling function may attempt to use or release `pols[0]` again
6. **Memory Corruption**:
Double free or use-after-free leads to system instability
## Proposed Fix
The vulnerability should be fixed by ensuring
complete state cleanup in error paths:
```c
// Current potentially vulnerable code:
if (IS_ERR(pols[1])) {
XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
xfrm_pol_put(pols[0]);
return 0;
}
// Proposed secure fix:
if (IS_ERR(pols[1])) {
XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
xfrm_pol_put(pols[0]);
pols[0] = NULL; // Clear pointer to prevent reuse
return 0;
}
```
### Alternative Comprehensive Fix
If the calling context requires more extensive cleanup:
```c
if (IS_ERR(pols[1])) {
XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
xfrm_pol_put(pols[0]);
memset(pols, 0, sizeof(pols)); // Clear entire policy array
npols = 0; // Reset policy count
return 0;
}
```
## References
- **Original CVE**:
CVE-2022-36879 (xfrm_expand_policies double free)
- **Linux XFRM Documentation**:
`Documentation/networking/xfrm_*.txt`
- **XFRM Source**:
`net/xfrm/xfrm_policy.c`
- **IPsec RFCs**:
RFC 4301, RFC 4306
Signed-off-by: qianjiaru <qianjiaru77@...il.com>
---
net/xfrm/xfrm_policy.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/net/xfrm/xfrm_policy.c b/net/xfrm/xfrm_policy.c
index c5035a9bc..50943fa4e 100644
--- a/net/xfrm/xfrm_policy.c
+++ b/net/xfrm/xfrm_policy.c
@@ -3786,6 +3786,7 @@ int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
if (IS_ERR(pols[1])) {
XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
xfrm_pol_put(pols[0]);
+ pols[0] = NULL; // Clear pointer to prevent reuse
return 0;
}
/* This write can happen from different cpus. */
--
2.34.1
Powered by blists - more mailing lists