Skip to contentSkip to conversation

IB selinux related fixes

PATCH Applied Leon Romanovsky 3 participants 7 messages Feb 2, 2019 – Feb 8, 2019

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover IB selinux related fixes

From: Leon Romanovsky <[email protected]>

Hi,

This is second version of previously sent patch "IB/core: Don't register MAD agents for LSM notifications", due to the fact that we found extra things to fix during the review, I grouped that patch together with other fixes in that area.

Thanks

Daniel Jurgens (4):
IB/core: Unregister notifier before freeing MAD security IB/core: Fix potential memory leak while creating MAD agents IB/core: Eliminate a hole in MAD agent struct IB/core: Don't register each MAD agent for LSM notifier

 drivers/infiniband/core/core_priv.h |  5 +++
 drivers/infiniband/core/device.c    |  1 +
 drivers/infiniband/core/security.c  | 57 +++++++++++++++++------------
 include/rdma/ib_mad.h               |  5 +--
 4 files changed, 42 insertions(+), 26 deletions(-)

--
2.19.1

#1 lore raw

PATCH 1/4 IB/core: Unregister notifier before freeing MAD security

From: Daniel Jurgens <[email protected]>

If the notifier runs after the security context is freed an access of freed memory can occur.

Fixes 47a2b338fe63 ("IB/core: Enforce security on management datagrams")
---
 drivers/infiniband/core/security.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/infiniband/core/security.c +2 -1
index 1efadbccf394..402449d4a888 100644
@@ -727,9 +727,10 @@ void ib_mad_agent_security_cleanup(struct ib_mad_agent *agent)
 	if (!rdma_protocol_ib(agent->device, agent->port_num))
 		return;
 
-	security_ib_free_security(agent->security);
 	if (agent->lsm_nb_reg)
 		unregister_lsm_notifier(&agent->lsm_nb);
+
+	security_ib_free_security(agent->security);
 }
 
 int ib_mad_enforce_security(struct ib_mad_agent_private *map, u16 pkey_index)

--
2.19.1

#2 lore raw

PATCH 2/4 IB/core: Fix potential memory leak while creating MAD agents

In reply to Leon Romanovsky

From: Daniel Jurgens <[email protected]>

If the MAD agents isn't allowed to manage the subnet, or fails to register for the LSM notifier, the security context is leaked. Free the context in these cases.

Fixes 47a2b338fe63 ("IB/core: Enforce security on management datagrams")
---
 drivers/infiniband/core/security.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
drivers/infiniband/core/security.c +6 -2
index 402449d4a888..7662e9347238 100644
@@ -710,16 +710,20 @@ int ib_mad_agent_security_setup(struct ib_mad_agent *agent,
 						dev_name(&agent->device->dev),
 						agent->port_num);
 	if (ret)
-		return ret;
+		goto free_security;
 
 	agent->lsm_nb.notifier_call = ib_mad_agent_security_change;
 	ret = register_lsm_notifier(&agent->lsm_nb);
 	if (ret)
-		return ret;
+		goto free_security;
 
 	agent->smp_allowed = true;
 	agent->lsm_nb_reg = true;
 	return 0;
+
+free_security:
+	security_ib_free_security(agent->security);
+	return ret;
 }
 
 void ib_mad_agent_security_cleanup(struct ib_mad_agent *agent)

--
2.19.1

#3 lore raw

PATCH 3/4 IB/core: Eliminate a hole in MAD agent struct

In reply to Leon Romanovsky

From: Daniel Jurgens <[email protected]>

Move the security related fields above the u8s to eliminate a hole in the struct.

pahole before:
struct ib_mad_agent {
...

u32                        hi_tid;               /*    48     4 */
u32                        flags;                /*    52     4 */
u8                         port_num;             /*    56     1 */
u8                         rmpp_version;         /*    57     1 */

/* XXX 6 bytes hole, try to pack */

/* --- cacheline 1 boundary (64 bytes) --- */

void *                     security;             /*    64     8 */
bool                       smp_allowed;          /*    72     1 */
bool                       lsm_nb_reg;           /*    73     1 */

/* XXX 6 bytes hole, try to pack */

struct notifier_block lsm_nb; /* 80 24 */

/* XXX last struct has 4 bytes of padding */

/* size: 104, cachelines: 2, members: 14 */
...
};

pahole after:
struct ib_mad_agent {
...

u32                        hi_tid;               /*    48     4 */
u32                        flags;                /*    52     4 */
void *                     security;             /*    56     8 */

/* --- cacheline 1 boundary (64 bytes) --- */
struct notifier_block lsm_nb; /* 64 24 */

/* XXX last struct has 4 bytes of padding */

u8                         port_num;             /*    88     1 */
u8                         rmpp_version;         /*    89     1 */
bool                       smp_allowed;          /*    90     1 */
bool                       lsm_nb_reg;           /*    91     1 */

/* size: 96, cachelines: 2, members: 14 */
...
};

---
 include/rdma/ib_mad.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
include/rdma/ib_mad.h +2 -2
index fdef558e3a2d..1c0b914f199d 100644
@@ -616,12 +616,12 @@ struct ib_mad_agent {
 	void			*context;
 	u32			hi_tid;
 	u32			flags;
+	void			*security;
+	struct notifier_block   lsm_nb;
 	u8			port_num;
 	u8			rmpp_version;
-	void			*security;
 	bool			smp_allowed;
 	bool			lsm_nb_reg;
-	struct notifier_block   lsm_nb;
 };
 
 /**

--
2.19.1

#4 lore raw

PATCH 4/4 IB/core: Don't register each MAD agent for LSM notifier

In reply to Leon Romanovsky

From: Daniel Jurgens <[email protected]>

When creating many MAD agents in a short period of time, receive packet processing can be delayed long enough to cause timeouts while new agents are being added to the atomic notifier chain with IRQs disabled. Notifier chain registration and unregstration is an O(n) operation. With large numbers of MAD agents being created and destroyed simultaneously the CPUs spend too much time with interrupts disabled.

Instead of each MAD agent registering for it's own LSM notification, maintain a list of agents internally and register once, this registration already existed for handling the PKeys. This list is write mostly, so a normal spin lock is used vs a read/write lock. All MAD agents must be checked, so a single list is used instead of breaking them down per device.

Notifier calls are done under rcu_read_lock, so there isn't a risk of similar packet timeouts while checking the MAD agents security settings when notified.

---
 drivers/infiniband/core/core_priv.h |  5 +++
 drivers/infiniband/core/device.c    |  1 +
 drivers/infiniband/core/security.c  | 50 ++++++++++++++++-------------
 include/rdma/ib_mad.h               |  3 +-
 4 files changed, 35 insertions(+), 24 deletions(-)
drivers/infiniband/core/core_priv.h +5 -0
index 616734313f0c..6fd3d3d06d18 100644
@@ -199,6 +199,7 @@ int ib_mad_agent_security_setup(struct ib_mad_agent *agent,
 				enum ib_qp_type qp_type);
 void ib_mad_agent_security_cleanup(struct ib_mad_agent *agent);
 int ib_mad_enforce_security(struct ib_mad_agent_private *map, u16 pkey_index);
+void ib_mad_agent_security_change(void);
 #else
 static inline void ib_security_destroy_port_pkey_list(struct ib_device *device)
 {
@@ -264,6 +265,10 @@ static inline int ib_mad_enforce_security(struct ib_mad_agent_private *map,
 {
 	return 0;
 }
+
+static inline void ib_mad_agent_security_change(void)
+{
+}
 #endif
 
 struct ib_device *ib_device_get_by_index(u32 ifindex);
drivers/infiniband/core/device.c +1 -0
index 238ec42778ef..8c1d8ffb09e2 100644
@@ -455,6 +455,7 @@ static int ib_security_change(struct notifier_block *nb, unsigned long event,
 		return NOTIFY_DONE;
 
 	schedule_work(&ib_policy_change_work);
+	ib_mad_agent_security_change();
 
 	return NOTIFY_OK;
 }
drivers/infiniband/core/security.c +28 -22

96 lines not on this page · raw message

include/rdma/ib_mad.h +1 -2
index 1c0b914f199d..79ba8219e7dc 100644
@@ -617,11 +617,10 @@ struct ib_mad_agent {
 	u32			hi_tid;
 	u32			flags;
 	void			*security;
-	struct notifier_block   lsm_nb;
+	struct list_head	mad_agent_sec_list;
 	u8			port_num;
 	u8			rmpp_version;
 	bool			smp_allowed;
-	bool			lsm_nb_reg;
 };
 
 /**

--
2.19.1

maintainer Acked-by on v1 4/4
#5 lore raw

Re: [PATCH rdma-rc v1 4/4] IB/core: Don't register each MAD agent for LSM notifier

Quoted 26 lines · Leon Romanovsky

From: Daniel Jurgens <[email protected]>

When creating many MAD agents in a short period of time, receive packet processing can be delayed long enough to cause timeouts while new agents are being added to the atomic notifier chain with IRQs disabled. Notifier chain registration and unregstration is an O(n) operation. With large numbers of MAD agents being created and destroyed simultaneously the CPUs spend too much time with interrupts disabled.

Instead of each MAD agent registering for it's own LSM notification, maintain a list of agents internally and register once, this registration already existed for handling the PKeys. This list is write mostly, so a normal spin lock is used vs a read/write lock. All MAD agents must be checked, so a single list is used instead of breaking them down per device.

Notifier calls are done under rcu_read_lock, so there isn't a risk of similar packet timeouts while checking the MAD agents security settings when notified.

Quoted 6 lines
---
 drivers/infiniband/core/core_priv.h |  5 +++
 drivers/infiniband/core/device.c    |  1 +
 drivers/infiniband/core/security.c  | 50 ++++++++++++++++-------------
 include/rdma/ib_mad.h               |  3 +-
 4 files changed, 35 insertions(+), 24 deletions(-)

This looks reasonable from a SELinux perspective. I would still be curious to see how this would compare to calling security_ib_endport_manage_subnet() directly in ip_mad_enforce_security(), but at least this patch should fix the problem you're seeing.

Quoted 161 lines
drivers/infiniband/core/core_priv.h +5 -0
index 616734313f0c..6fd3d3d06d18 100644
@@ -199,6 +199,7 @@ int ib_mad_agent_security_setup(struct ib_mad_agent *agent,
                                enum ib_qp_type qp_type);
 void ib_mad_agent_security_cleanup(struct ib_mad_agent *agent);
 int ib_mad_enforce_security(struct ib_mad_agent_private *map, u16 pkey_index);
+void ib_mad_agent_security_change(void);
 #else
 static inline void ib_security_destroy_port_pkey_list(struct ib_device *device)
 {
@@ -264,6 +265,10 @@ static inline int ib_mad_enforce_security(struct ib_mad_agent_private *map,
 {
        return 0;
 }
+
+static inline void ib_mad_agent_security_change(void)
+{
+}
 #endif
 
 struct ib_device *ib_device_get_by_index(u32 ifindex);
drivers/infiniband/core/device.c +1 -0
index 238ec42778ef..8c1d8ffb09e2 100644
@@ -455,6 +455,7 @@ static int ib_security_change(struct notifier_block *nb, unsigned long event,
                return NOTIFY_DONE;
 
        schedule_work(&ib_policy_change_work);
+       ib_mad_agent_security_change();
 
        return NOTIFY_OK;
 }
drivers/infiniband/core/security.c +28 -22
index 7662e9347238..a70d2ba312ed 100644
@@ -39,6 +39,10 @@
 #include "core_priv.h"
 #include "mad_priv.h"
 
+static LIST_HEAD(mad_agent_list);
+/* Lock to protect mad_agent_list */
+static DEFINE_SPINLOCK(mad_agent_list_lock);
+
 static struct pkey_index_qp_list *get_pkey_idx_qp_list(struct ib_port_pkey *pp)
 {
        struct pkey_index_qp_list *pkey = NULL;
@@ -676,19 +680,18 @@ static int ib_security_pkey_access(struct ib_device *dev,
        return security_ib_pkey_access(sec, subnet_prefix, pkey);
 }
 
-static int ib_mad_agent_security_change(struct notifier_block *nb,
-                                       unsigned long event,
-                                       void *data)
+void ib_mad_agent_security_change(void)
 {
-       struct ib_mad_agent *ag = container_of(nb, struct ib_mad_agent, lsm_nb);
-
-       if (event != LSM_POLICY_CHANGE)
-               return NOTIFY_DONE;
-
-       ag->smp_allowed = !security_ib_endport_manage_subnet(
-               ag->security, dev_name(&ag->device->dev), ag->port_num);
-
-       return NOTIFY_OK;
+       struct ib_mad_agent *ag;
+
+       spin_lock(&mad_agent_list_lock);
+       list_for_each_entry(ag,
+                           &mad_agent_list,
+                           mad_agent_sec_list)
+               WRITE_ONCE(ag->smp_allowed,
+                          !security_ib_endport_manage_subnet(ag->security,
+                               dev_name(&ag->device->dev), ag->port_num));
+       spin_unlock(&mad_agent_list_lock);
 }
 
 int ib_mad_agent_security_setup(struct ib_mad_agent *agent,
@@ -699,6 +702,8 @@ int ib_mad_agent_security_setup(struct ib_mad_agent *agent,
        if (!rdma_protocol_ib(agent->device, agent->port_num))
                return 0;
 
+       INIT_LIST_HEAD(&agent->mad_agent_sec_list);
+
        ret = security_ib_alloc_security(&agent->security);
        if (ret)
                return ret;
@@ -706,22 +711,20 @@ int ib_mad_agent_security_setup(struct ib_mad_agent *agent,
        if (qp_type != IB_QPT_SMI)
                return 0;
 
+       spin_lock(&mad_agent_list_lock);
        ret = security_ib_endport_manage_subnet(agent->security,
                                                dev_name(&agent->device->dev),
                                                agent->port_num);
        if (ret)
                goto free_security;
 
-       agent->lsm_nb.notifier_call = ib_mad_agent_security_change;
-       ret = register_lsm_notifier(&agent->lsm_nb);
-       if (ret)
-               goto free_security;
-
-       agent->smp_allowed = true;
-       agent->lsm_nb_reg = true;
+       WRITE_ONCE(agent->smp_allowed, true);
+       list_add(&agent->mad_agent_sec_list, &mad_agent_list);
+       spin_unlock(&mad_agent_list_lock);
        return 0;
 
 free_security:
+       spin_unlock(&mad_agent_list_lock);
        security_ib_free_security(agent->security);
        return ret;
 }
@@ -731,8 +734,11 @@ void ib_mad_agent_security_cleanup(struct ib_mad_agent *agent)
        if (!rdma_protocol_ib(agent->device, agent->port_num))
                return;
 
-       if (agent->lsm_nb_reg)
-               unregister_lsm_notifier(&agent->lsm_nb);
+       if (agent->qp->qp_type == IB_QPT_SMI) {
+               spin_lock(&mad_agent_list_lock);
+               list_del(&agent->mad_agent_sec_list);
+               spin_unlock(&mad_agent_list_lock);
+       }
 
        security_ib_free_security(agent->security);
 }
@@ -743,7 +749,7 @@ int ib_mad_enforce_security(struct ib_mad_agent_private *map, u16 pkey_index)
                return 0;
 
        if (map->agent.qp->qp_type == IB_QPT_SMI) {
-               if (!map->agent.smp_allowed)
+               if (!READ_ONCE(map->agent.smp_allowed))
                        return -EACCES;
                return 0;
        }
include/rdma/ib_mad.h +1 -3
index 1c0b914f199d..79ba8219e7dc 100644
@@ -617,11 +617,10 @@ struct ib_mad_agent {
        u32                     hi_tid;
        u32                     flags;
        void                    *security;
-       struct notifier_block   lsm_nb;
+       struct list_head        mad_agent_sec_list;
        u8                      port_num;
        u8                      rmpp_version;
        bool                    smp_allowed;
-       bool                    lsm_nb_reg;
 };
 
 /**
--

2.19.1

Signature · 2 lines
paul moore
www.paul-moore.com
#6 lore raw
In reply to Leon Romanovsky
Quoted 9 lines · Leon Romanovsky

From: Leon Romanovsky <[email protected]>

Hi,

This is second version of previously sent patch "IB/core: Don't register MAD agents for LSM notifications", due to the fact that we found extra things to fix during the review, I grouped that patch together with other fixes in that area.

Quoted 7 lines

Thanks

Daniel Jurgens (4):
IB/core: Unregister notifier before freeing MAD security IB/core: Fix potential memory leak while creating MAD agents IB/core: Eliminate a hole in MAD agent struct IB/core: Don't register each MAD agent for LSM notifier

Applied to for-next

Thanks,
Jason