Skip to contentSkip to conversation

IB/core: Only maintain real QPs in the security lists

PATCH Applied Leon Romanovsky 5 participants 10 messages Nov 7, 2017 – Nov 13, 2017 5 messages not in the archive

Subsystems (2)
Stable branch InfiniBand
lore
Share & tools
lore

Leon Romanovsky

author
#0 lore raw

PATCH 1/1 IB/core: Only maintain real QPs in the security lists

From: Daniel Jurgens <[email protected]>

When modify QP is called on a shared QP update the security context for the real QP. When security is subsequently enforced the shared QP handles will be checked as well.

Without this change shared QP handles get added to the port/pkey lists, which is a bug, because not all shared QP handles will be checked for access. Also the shared QP security context wouldn't get removed from the port/pkey lists causing access to free memory and list corruption when they are destroyed.

Cc [email protected]
Fixes d291f1a65232 ("IB/core: Enforce PKey security on QPs")
Signed-off-by Daniel Jurgens <[email protected]>
Reviewed-by Parav Pandit <[email protected]>
Signed-off-by Leon Romanovsky <[email protected]>
---
 drivers/infiniband/core/security.c | 51 +++++++++++++++++++++-----------------
 1 file changed, 28 insertions(+), 23 deletions(-)
drivers/infiniband/core/security.c +28 -23
index 455cba7f9640..23278ed5be45 100644
@@ -87,16 +87,14 @@ static int enforce_qp_pkey_security(u16 pkey,
 	if (ret)
 		return ret;
 
-	if (qp_sec->qp == qp_sec->qp->real_qp) {
-		list_for_each_entry(shared_qp_sec,
-				    &qp_sec->shared_qp_list,
-				    shared_qp_list) {
-			ret = security_ib_pkey_access(shared_qp_sec->security,
-						      subnet_prefix,
-						      pkey);
-			if (ret)
-				return ret;
-		}
+	list_for_each_entry(shared_qp_sec,
+			    &qp_sec->shared_qp_list,
+			    shared_qp_list) {
+		ret = security_ib_pkey_access(shared_qp_sec->security,
+					      subnet_prefix,
+					      pkey);
+		if (ret)
+			return ret;
 	}
 	return 0;
 }

@@ -560,15 +558,22 @@ int ib_security_modify_qp(struct ib_qp *qp,
 	int ret = 0;
 	struct ib_ports_pkeys *tmp_pps;
 	struct ib_ports_pkeys *new_pps;
-	bool special_qp = (qp->qp_type == IB_QPT_SMI ||
-			   qp->qp_type == IB_QPT_GSI ||
-			   qp->qp_type >= IB_QPT_RESERVED1);
+	struct ib_qp *real_qp = qp->real_qp;
+	bool special_qp = (real_qp->qp_type == IB_QPT_SMI ||
+			   real_qp->qp_type == IB_QPT_GSI ||
+			   real_qp->qp_type >= IB_QPT_RESERVED1);
 	bool pps_change = ((qp_attr_mask & (IB_QP_PKEY_INDEX | IB_QP_PORT)) ||
 			   (qp_attr_mask & IB_QP_ALT_PATH));
 
+	/* The port/pkey settings are maintained only for the real QP. Open
+	 * handles on the real QP will be in the shared_qp_list. When
+	 * enforcing security on the real QP all the shared QPs will be
+	 * checked as well.
+	 */
+
 	if (pps_change && !special_qp) {
-		mutex_lock(&qp->qp_sec->mutex);
-		new_pps = get_new_pps(qp,
+		mutex_lock(&real_qp->qp_sec->mutex);
+		new_pps = get_new_pps(real_qp,
 				      qp_attr,
 				      qp_attr_mask);
 

@@ -586,14 +591,14 @@ int ib_security_modify_qp(struct ib_qp *qp,
 
 		if (!ret)
 			ret = check_qp_port_pkey_settings(new_pps,
-							  qp->qp_sec);
+							  real_qp->qp_sec);
 	}
 
 	if (!ret)
-		ret = qp->device->modify_qp(qp->real_qp,
-					    qp_attr,
-					    qp_attr_mask,
-					    udata);
+		ret = real_qp->device->modify_qp(real_qp,
+						 qp_attr,
+						 qp_attr_mask,
+						 udata);
 
 	if (pps_change && !special_qp) {
 		/* Clean up the lists and free the appropriate

@@ -602,8 +607,8 @@ int ib_security_modify_qp(struct ib_qp *qp,
 		if (ret) {
 			tmp_pps = new_pps;
 		} else {
-			tmp_pps = qp->qp_sec->ports_pkeys;
-			qp->qp_sec->ports_pkeys = new_pps;
+			tmp_pps = real_qp->qp_sec->ports_pkeys;
+			real_qp->qp_sec->ports_pkeys = new_pps;
 		}
 
 		if (tmp_pps) {

@@ -611,7 +616,7 @@ int ib_security_modify_qp(struct ib_qp *qp,
 			port_pkey_list_remove(&tmp_pps->alt);
 		}
 		kfree(tmp_pps);
-		mutex_unlock(&qp->qp_sec->mutex);
+		mutex_unlock(&real_qp->qp_sec->mutex);
 	}
 	return ret;
 }

A message that is not in the archive (sent off-list or not yet received): 20171107163326.23881-1-leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org

#2 lore raw

drivers/infiniband/core/security.c int ib_security_modify_qp(struct ib_qp *qp, · line 564

Quoted 7 lines · Leon Romanovsky
drivers/infiniband/core/security.c +4 -3
-	bool special_qp = (qp->qp_type == IB_QPT_SMI ||
-			   qp->qp_type == IB_QPT_GSI ||
-			   qp->qp_type >= IB_QPT_RESERVED1);
+	struct ib_qp *real_qp = qp->real_qp;
+	bool special_qp = (real_qp->qp_type == IB_QPT_SMI ||
+			   real_qp->qp_type == IB_QPT_GSI ||
+			   real_qp->qp_type >= IB_QPT_RESERVED1);

This QPT_RESERVED stuff was not supposed to be visible to the core layer, so why are we adding checks in security????

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

#3 lore raw
Quoted 12 lines · Jason Gunthorpe

drivers/infiniband/core/security.c int ib_security_modify_qp(struct ib_qp *qp, · line 564

Quoted 7 lines · Leon Romanovsky
drivers/infiniband/core/security.c +4 -3
-	bool special_qp = (qp->qp_type == IB_QPT_SMI ||
-			   qp->qp_type == IB_QPT_GSI ||
-			   qp->qp_type >= IB_QPT_RESERVED1);
+	struct ib_qp *real_qp = qp->real_qp;
+	bool special_qp = (real_qp->qp_type == IB_QPT_SMI ||
+			   real_qp->qp_type == IB_QPT_GSI ||
+			   real_qp->qp_type >= IB_QPT_RESERVED1);

This QPT_RESERVED stuff was not supposed to be visible to the core layer, so why are we adding checks in security????

Quoted 1 line

Jason

The checks exclude those QPs from security enforcement.  They've been there the whole time, you reviewed this previously right?

A message that is not in the archive (sent off-list or not yet received): 065e66ff-e901-bd5c-674d-173eab720989-VPRAkNaXOzVWk0Htik3J/[email protected]

#5 lore raw
Quoted 13 lines · Daniel Jurgens
Quoted 11 lines · Jason Gunthorpe

drivers/infiniband/core/security.c int ib_security_modify_qp(struct ib_qp *qp, · line 564

Quoted 7 lines · Leon Romanovsky
drivers/infiniband/core/security.c +4 -3
-	bool special_qp = (qp->qp_type == IB_QPT_SMI ||
-			   qp->qp_type == IB_QPT_GSI ||
-			   qp->qp_type >= IB_QPT_RESERVED1);
+	struct ib_qp *real_qp = qp->real_qp;
+	bool special_qp = (real_qp->qp_type == IB_QPT_SMI ||
+			   real_qp->qp_type == IB_QPT_GSI ||
+			   real_qp->qp_type >= IB_QPT_RESERVED1);

This QPT_RESERVED stuff was not supposed to be visible to the core layer, so why are we adding checks in security????

Quoted 2 lines

The checks exclude those QPs from security enforcement.  They've been there the whole time, you reviewed this previously right?

Nope

.. and they shouldn't be there, those reserved QPTs are totally banned from user space so it shouldn't matter for security

Jason

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): 20171107165539.GB7063-uk2M96/[email protected]

#7 lore raw
Quoted 14 lines · Jason Gunthorpe
Quoted 13 lines · Daniel Jurgens
Quoted 11 lines · Jason Gunthorpe

Review

Quoted 7 lines · Leon Romanovsky
(quoted diff) +4 -3
-	bool special_qp = (qp->qp_type == IB_QPT_SMI ||
-			   qp->qp_type == IB_QPT_GSI ||
-			   qp->qp_type >= IB_QPT_RESERVED1);
+	struct ib_qp *real_qp = qp->real_qp;
+	bool special_qp = (real_qp->qp_type == IB_QPT_SMI ||
+			   real_qp->qp_type == IB_QPT_GSI ||
+			   real_qp->qp_type >= IB_QPT_RESERVED1);

This QPT_RESERVED stuff was not supposed to be visible to the core layer, so why are we adding checks in security????

Quoted 7 lines
Quoted 2 lines

The checks exclude those QPs from security enforcement.  They've been there the whole time, you reviewed this previously right?

Nope

.. and they shouldn't be there, those reserved QPTs are totally banned from user space so it shouldn't matter for security

We will send followup patch to clean it, but for this patch I prefer do not change the logic here because it is supposed to go to stable@.

Thanks

Quoted 3 lines

Jason

A message that is not in the archive (sent off-list or not yet received): 20171107171026.GN18825-U/[email protected]

#10 lore raw
In reply to Jason Gunthorpe
Quoted 13 lines · Jason Gunthorpe
Quoted 12 lines · Daniel Jurgens
Quoted 11 lines · Jason Gunthorpe

drivers/infiniband/core/security.c int ib_security_modify_qp(struct ib_qp *qp, · line 564

Quoted 7 lines · Leon Romanovsky
drivers/infiniband/core/security.c +4 -3
-	bool special_qp = (qp->qp_type == IB_QPT_SMI ||
-			   qp->qp_type == IB_QPT_GSI ||
-			   qp->qp_type >= IB_QPT_RESERVED1);
+	struct ib_qp *real_qp = qp->real_qp;
+	bool special_qp = (real_qp->qp_type == IB_QPT_SMI ||
+			   real_qp->qp_type == IB_QPT_GSI ||
+			   real_qp->qp_type >= IB_QPT_RESERVED1);

This QPT_RESERVED stuff was not supposed to be visible to the core layer, so why are we adding checks in security????

Quoted 9 lines
Quoted 2 lines

The checks exclude those QPs from security enforcement.  They've been there the whole time, you reviewed this previously right?

Nope

.. and they shouldn't be there, those reserved QPTs are totally banned from user space so it shouldn't matter for security

Jason

This flow is through ib_modify_qp, it's not user space specific.  If it really pains you that much it can be changed an inclusive compare instead of exclusive. This discussion isn't relevant  to the content of the patch in question though.

A message that is not in the archive (sent off-list or not yet received): 87a902ae-351b-43c4-4a42-b362c981a767-VPRAkNaXOzVWk0Htik3J/[email protected]

#12 lore raw
Quoted 3 lines · Daniel Jurgens

This flow is through ib_modify_qp, it's not user space specific. If it really pains you that much it can be changed an inclusive compare instead of exclusive.

There was already a RFC on re-purpsing those constants and exposing them to user space.

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

author
#13 lore raw
Quoted 8 lines · Jason Gunthorpe
Quoted 3 lines · Daniel Jurgens

This flow is through ib_modify_qp, it's not user space specific. If it really pains you that much it can be changed an inclusive compare instead of exclusive.

There was already a RFC on re-purpsing those constants and exposing them to user space.

It pains me, the flows of create_qp/modify_qp are completely unmaintainable mess and I would like to see more explicit checks, so refactoring will be easy. We will reach that step much sooner than I expected.

Thanks

BTW, I spent last day to annotate all different paths of create_qp for the resource tracking information.

Quoted 2 lines

Jason

Applied
#14 lore raw
In reply to Leon Romanovsky
Quoted 12 lines · Leon Romanovsky

From: Daniel Jurgens <[email protected]>

When modify QP is called on a shared QP update the security context for the real QP. When security is subsequently enforced the shared QP handles will be checked as well.

Without this change shared QP handles get added to the port/pkey lists, which is a bug, because not all shared QP handles will be checked for access. Also the shared QP security context wouldn't get removed from the port/pkey lists causing access to free memory and list corruption when they are destroyed.

Quoted 5 lines
Cc [email protected]
Fixes d291f1a65232 ("IB/core: Enforce PKey security on QPs")
Signed-off-by Daniel Jurgens <[email protected]>
Reviewed-by Parav Pandit <[email protected]>
Signed-off-by Leon Romanovsky <[email protected]>

Thanks, applied.

Signature · 3 lines
Doug Ledford <[email protected]>
    GPG KeyID: B826A3330E572FDD
    Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57 2FDD