Skip to contentSkip to conversation

Sleepable local storage

PATCH Applied KP Singh 3 participants 5 messages Dec 24, 2021 – Dec 30, 2021

Subsystems (5)
lore
Share & tools
lore

KP Singh

author
#0 lore raw

v3 cover Sleepable local storage

Local storage is currently unusable in sleepable helpers. One of the important use cases of local_storage is to attach security (or performance) contextual information to kernel objects in LSM / tracing programs to be used later in the life-cyle of the object.

Sometimes this context can only be gathered from sleepable programs (because it needs accesing __user pointers or helpers like bpf_ima_inode_hash). Allowing local storage to be used from sleepable programs allows such context to be managed with the benefits of local_storage.

# v2 -> v3

* Fixed some RCU issues pointed by Martin
* Added Martin's ack

# v1 -> v2

* Generalize RCU checks (will send a separate patch for updating non local storage code where this can be used).
* Add missing RCU lock checks from v1

KP Singh (2):
bpf: Allow bpf_local_storage to be used by sleepable programs bpf/selftests: Update local storage selftest for sleepable programs

 include/linux/bpf_local_storage.h             |  5 ++
 kernel/bpf/bpf_inode_storage.c                |  6 ++-
 kernel/bpf/bpf_local_storage.c                | 50 ++++++++++++++-----
 kernel/bpf/bpf_task_storage.c                 |  6 ++-
 kernel/bpf/verifier.c                         |  3 ++
 net/core/bpf_sk_storage.c                     |  8 ++-
 .../bpf/prog_tests/test_local_storage.c       | 20 +++-----
 .../selftests/bpf/progs/local_storage.c       | 24 ++-------
 8 files changed, 73 insertions(+), 49 deletions(-)
Signature · 1 line
2.34.1.448.ga2b2bfdf31-goog
author
#1 lore raw

PATCH v3 1/2 bpf: Allow bpf_local_storage to be used by sleepable programs

Other maps like hashmaps are already available to sleepable programs. Sleepable BPF programs run under trace RCU. Allow task, sk and inode storage to be used from sleepable programs. This allows sleepable and non-sleepable programs to provide shareable annotations on kernel objects.

Sleepable programs run in trace RCU where as non-sleepable programs run in a normal RCU critical section i.e. __bpf_prog_enter{_sleepable} and __bpf_prog_exit{_sleepable}) (rcu_read_lock or rcu_read_lock_trace).

In order to make the local storage maps accessible to both sleepable and non-sleepable programs, one needs to call both call_rcu_tasks_trace and call_rcu to wait for both trace and classical RCU grace periods to expire before freeing memory.

Paul's work on call_rcu_tasks_trace allows us to have per CPU queueing for call_rcu_tasks_trace. This behaviour can be achieved by setting rcupdate.rcu_task_enqueue_lim=<num_cpus> boot parameter.

In light of these new performance changes and to keep the local storage code simple, avoid adding a new flag for sleepable maps / local storage to select the RCU synchronization (trace / classical).

Also, update the dereferencing of the pointers to use rcu_derference_check (with either the trace or normal RCU locks held) with a common bpf_rcu_lock_held helper method.

---
 include/linux/bpf_local_storage.h |  5 ++++
 kernel/bpf/bpf_inode_storage.c    |  6 +++-
 kernel/bpf/bpf_local_storage.c    | 50 +++++++++++++++++++++++--------
 kernel/bpf/bpf_task_storage.c     |  6 +++-
 kernel/bpf/verifier.c             |  3 ++
 net/core/bpf_sk_storage.c         |  8 ++++-
 6 files changed, 62 insertions(+), 16 deletions(-)
include/linux/bpf_local_storage.h +5 -0
index 24496bc28e7b..475431a36a62 100644
@@ -16,6 +16,9 @@
 
 #define BPF_LOCAL_STORAGE_CACHE_SIZE	16
 
+#define bpf_rcu_lock_held()                                                    \
+	(rcu_read_lock_held() || rcu_read_lock_trace_held() ||                 \
+	 rcu_read_lock_bh_held())
 struct bpf_local_storage_map_bucket {
 	struct hlist_head list;
 	raw_spinlock_t lock;
@@ -161,4 +164,6 @@ struct bpf_local_storage_data *
 bpf_local_storage_update(void *owner, struct bpf_local_storage_map *smap,
 			 void *value, u64 map_flags);
 
+void bpf_local_storage_free_rcu(struct rcu_head *rcu);
+
 #endif /* _BPF_LOCAL_STORAGE_H */
kernel/bpf/bpf_inode_storage.c +5 -1
index 96ceed0e0fb5..e29d9e3d853e 100644
@@ -17,6 +17,7 @@
 #include <linux/bpf_lsm.h>
 #include <linux/btf_ids.h>
 #include <linux/fdtable.h>
+#include <linux/rcupdate_trace.h>
 
 DEFINE_BPF_STORAGE_CACHE(inode_cache);
 
@@ -44,7 +45,8 @@ static struct bpf_local_storage_data *inode_storage_lookup(struct inode *inode,
 	if (!bsb)
 		return NULL;
 
-	inode_storage = rcu_dereference(bsb->storage);
+	inode_storage =
+		rcu_dereference_check(bsb->storage, bpf_rcu_lock_held());
 	if (!inode_storage)
 		return NULL;
 
@@ -172,6 +174,7 @@ BPF_CALL_4(bpf_inode_storage_get, struct bpf_map *, map, struct inode *, inode,
 {
 	struct bpf_local_storage_data *sdata;
 
+	WARN_ON_ONCE(!bpf_rcu_lock_held());
 	if (flags & ~(BPF_LOCAL_STORAGE_GET_F_CREATE))
 		return (unsigned long)NULL;
 
@@ -204,6 +207,7 @@ BPF_CALL_4(bpf_inode_storage_get, struct bpf_map *, map, struct inode *, inode,
 BPF_CALL_2(bpf_inode_storage_delete,
 	   struct bpf_map *, map, struct inode *, inode)
 {
+	WARN_ON_ONCE(!bpf_rcu_lock_held());
 	if (!inode)
 		return -EINVAL;
 
kernel/bpf/bpf_local_storage.c +37 -13

123 lines not on this page · raw message

kernel/bpf/bpf_task_storage.c +5 -1
index bb69aea1a777..5da7bed0f5f6 100644
@@ -17,6 +17,7 @@
 #include <uapi/linux/btf.h>
 #include <linux/btf_ids.h>
 #include <linux/fdtable.h>
+#include <linux/rcupdate_trace.h>
 
 DEFINE_BPF_STORAGE_CACHE(task_cache);
 
@@ -59,7 +60,8 @@ task_storage_lookup(struct task_struct *task, struct bpf_map *map,
 	struct bpf_local_storage *task_storage;
 	struct bpf_local_storage_map *smap;
 
-	task_storage = rcu_dereference(task->bpf_storage);
+	task_storage =
+		rcu_dereference_check(task->bpf_storage, bpf_rcu_lock_held());
 	if (!task_storage)
 		return NULL;
 
@@ -229,6 +231,7 @@ BPF_CALL_4(bpf_task_storage_get, struct bpf_map *, map, struct task_struct *,
 {
 	struct bpf_local_storage_data *sdata;
 
+	WARN_ON_ONCE(!bpf_rcu_lock_held());
 	if (flags & ~(BPF_LOCAL_STORAGE_GET_F_CREATE))
 		return (unsigned long)NULL;
 
@@ -260,6 +263,7 @@ BPF_CALL_2(bpf_task_storage_delete, struct bpf_map *, map, struct task_struct *,
 {
 	int ret;
 
+	WARN_ON_ONCE(!bpf_rcu_lock_held());
 	if (!task)
 		return -EINVAL;
 
kernel/bpf/verifier.c +3 -0
index ca5cd0de804c..133599dfe2a2 100644
@@ -11874,6 +11874,9 @@ static int check_map_prog_compatibility(struct bpf_verifier_env *env,
 			}
 			break;
 		case BPF_MAP_TYPE_RINGBUF:
+		case BPF_MAP_TYPE_INODE_STORAGE:
+		case BPF_MAP_TYPE_SK_STORAGE:
+		case BPF_MAP_TYPE_TASK_STORAGE:
 			break;
 		default:
 			verbose(env,
net/core/bpf_sk_storage.c +7 -1
index ea61dfe19c86..d9c37fd10809 100644
@@ -13,6 +13,7 @@
 #include <net/sock.h>
 #include <uapi/linux/sock_diag.h>
 #include <uapi/linux/btf.h>
+#include <linux/rcupdate_trace.h>
 
 DEFINE_BPF_STORAGE_CACHE(sk_cache);
 
@@ -22,7 +23,8 @@ bpf_sk_storage_lookup(struct sock *sk, struct bpf_map *map, bool cacheit_lockit)
 	struct bpf_local_storage *sk_storage;
 	struct bpf_local_storage_map *smap;
 
-	sk_storage = rcu_dereference(sk->sk_bpf_storage);
+	sk_storage =
+		rcu_dereference_check(sk->sk_bpf_storage, bpf_rcu_lock_held());
 	if (!sk_storage)
 		return NULL;
 
@@ -258,6 +260,7 @@ BPF_CALL_4(bpf_sk_storage_get, struct bpf_map *, map, struct sock *, sk,
 {
 	struct bpf_local_storage_data *sdata;
 
+	WARN_ON_ONCE(!bpf_rcu_lock_held());
 	if (!sk || !sk_fullsock(sk) || flags > BPF_SK_STORAGE_GET_F_CREATE)
 		return (unsigned long)NULL;
 
@@ -288,6 +291,7 @@ BPF_CALL_4(bpf_sk_storage_get, struct bpf_map *, map, struct sock *, sk,
 
 BPF_CALL_2(bpf_sk_storage_delete, struct bpf_map *, map, struct sock *, sk)
 {
+	WARN_ON_ONCE(!bpf_rcu_lock_held());
 	if (!sk || !sk_fullsock(sk))
 		return -EINVAL;
 
@@ -416,6 +420,7 @@ static bool bpf_sk_storage_tracing_allowed(const struct bpf_prog *prog)
 BPF_CALL_4(bpf_sk_storage_get_tracing, struct bpf_map *, map, struct sock *, sk,
 	   void *, value, u64, flags)
 {
+	WARN_ON_ONCE(!bpf_rcu_lock_held());
 	if (in_hardirq() || in_nmi())
 		return (unsigned long)NULL;
 
@@ -425,6 +430,7 @@ BPF_CALL_4(bpf_sk_storage_get_tracing, struct bpf_map *, map, struct sock *, sk,
 BPF_CALL_2(bpf_sk_storage_delete_tracing, struct bpf_map *, map,
 	   struct sock *, sk)
 {
+	WARN_ON_ONCE(!bpf_rcu_lock_held());
 	if (in_hardirq() || in_nmi())
 		return -EPERM;
 
Signature · 1 line
2.34.1.448.ga2b2bfdf31-goog
Acked-by on v3 1/2
#2 lore raw

Re: [PATCH v3 bpf-next 1/2] bpf: Allow bpf_local_storage to be used by sleepable programs

Quoted 23 lines · KP Singh

Other maps like hashmaps are already available to sleepable programs. Sleepable BPF programs run under trace RCU. Allow task, sk and inode storage to be used from sleepable programs. This allows sleepable and non-sleepable programs to provide shareable annotations on kernel objects.

Sleepable programs run in trace RCU where as non-sleepable programs run in a normal RCU critical section i.e. __bpf_prog_enter{_sleepable} and __bpf_prog_exit{_sleepable}) (rcu_read_lock or rcu_read_lock_trace).

In order to make the local storage maps accessible to both sleepable and non-sleepable programs, one needs to call both call_rcu_tasks_trace and call_rcu to wait for both trace and classical RCU grace periods to expire before freeing memory.

Paul's work on call_rcu_tasks_trace allows us to have per CPU queueing for call_rcu_tasks_trace. This behaviour can be achieved by setting rcupdate.rcu_task_enqueue_lim=<num_cpus> boot parameter.

In light of these new performance changes and to keep the local storage code simple, avoid adding a new flag for sleepable maps / local storage to select the RCU synchronization (trace / classical).

Quoted 5 lines

Also, update the dereferencing of the pointers to use rcu_derference_check (with either the trace or normal RCU locks held) with a common bpf_rcu_lock_held helper method.

Thanks for the patches. One minor comment which is not very related to this set.

We can land this set first and then use a follow up.

kernel/bpf/bpf_local_storage.c int bpf_local_storage_alloc(void *owner, · line 335

Quoted 10 lines
kernel/bpf/bpf_local_storage.c +2 -1
@@ -306,7 +328,8 @@ int bpf_local_storage_alloc(void *owner,
 		 * bucket->list, first_selem can be freed immediately
 		 * (instead of kfree_rcu) because
 		 * bpf_local_storage_map_free() does a
-		 * synchronize_rcu() before walking the bucket->list.
+		 * synchronize_rcu_mult (waiting for both sleepable and
+		 * normal programs) before walking the bucket->list.
 		 * Hence, no one is accessing selem from the
 		 * bucket->list under rcu_read_lock().
 		 */

This whole comment section is outdated and can be removed because the bpf_free_used_maps(bpf_prog->aux) is now only called after call_rcu_tasks_trace() or call_rcu().

Meaning bpf_local_storage_map_free() cannot be running in parallel with bpf_local_storage_alloc() because the running bpf prog holds a refcnt of the smap here.

author
#3 lore raw

PATCH v3 2/2 bpf/selftests: Update local storage selftest for sleepable programs

In reply to KP Singh

Remove the spin lock logic and update the selftests to use sleepable programs to use a mix of sleepable and non-sleepable programs. It's more useful to test the sleepable programs since the tests don't really need spinlocks.

---
 .../bpf/prog_tests/test_local_storage.c       | 20 +++++-----------
 .../selftests/bpf/progs/local_storage.c       | 24 ++++---------------
 2 files changed, 11 insertions(+), 33 deletions(-)
tools/testing/selftests/bpf/prog_tests/test_local_storage.c +6 -14
index d2c16eaae367..26ac26a88026 100644
@@ -28,10 +28,6 @@ static unsigned int duration;
 struct storage {
 	void *inode;
 	unsigned int value;
-	/* Lock ensures that spin locked versions of local stoage operations
-	 * also work, most operations in this tests are still single threaded
-	 */
-	struct bpf_spin_lock lock;
 };
 
 /* Fork and exec the provided rm binary and return the exit code of the
@@ -66,27 +62,24 @@ static int run_self_unlink(int *monitored_pid, const char *rm_path)
 
 static bool check_syscall_operations(int map_fd, int obj_fd)
 {
-	struct storage val = { .value = TEST_STORAGE_VALUE, .lock = { 0 } },
-		       lookup_val = { .value = 0, .lock = { 0 } };
+	struct storage val = { .value = TEST_STORAGE_VALUE },
+		       lookup_val = { .value = 0 };
 	int err;
 
 	/* Looking up an existing element should fail initially */
-	err = bpf_map_lookup_elem_flags(map_fd, &obj_fd, &lookup_val,
-					BPF_F_LOCK);
+	err = bpf_map_lookup_elem_flags(map_fd, &obj_fd, &lookup_val, 0);
 	if (CHECK(!err || errno != ENOENT, "bpf_map_lookup_elem",
 		  "err:%d errno:%d\n", err, errno))
 		return false;
 
 	/* Create a new element */
-	err = bpf_map_update_elem(map_fd, &obj_fd, &val,
-				  BPF_NOEXIST | BPF_F_LOCK);
+	err = bpf_map_update_elem(map_fd, &obj_fd, &val, BPF_NOEXIST);
 	if (CHECK(err < 0, "bpf_map_update_elem", "err:%d errno:%d\n", err,
 		  errno))
 		return false;
 
 	/* Lookup the newly created element */
-	err = bpf_map_lookup_elem_flags(map_fd, &obj_fd, &lookup_val,
-					BPF_F_LOCK);
+	err = bpf_map_lookup_elem_flags(map_fd, &obj_fd, &lookup_val, 0);
 	if (CHECK(err < 0, "bpf_map_lookup_elem", "err:%d errno:%d", err,
 		  errno))
 		return false;
@@ -102,8 +95,7 @@ static bool check_syscall_operations(int map_fd, int obj_fd)
 		return false;
 
 	/* The lookup should fail, now that the element has been deleted */
-	err = bpf_map_lookup_elem_flags(map_fd, &obj_fd, &lookup_val,
-					BPF_F_LOCK);
+	err = bpf_map_lookup_elem_flags(map_fd, &obj_fd, &lookup_val, 0);
 	if (CHECK(!err || errno != ENOENT, "bpf_map_lookup_elem",
 		  "err:%d errno:%d\n", err, errno))
 		return false;
tools/testing/selftests/bpf/progs/local_storage.c +5 -19
index 95868bc7ada9..9b1f9b75d5c2 100644
@@ -20,7 +20,6 @@ int sk_storage_result = -1;
 struct local_storage {
 	struct inode *exec_inode;
 	__u32 value;
-	struct bpf_spin_lock lock;
 };
 
 struct {
@@ -58,9 +57,7 @@ int BPF_PROG(unlink_hook, struct inode *dir, struct dentry *victim)
 				       bpf_get_current_task_btf(), 0, 0);
 	if (storage) {
 		/* Don't let an executable delete itself */
-		bpf_spin_lock(&storage->lock);
 		is_self_unlink = storage->exec_inode == victim->d_inode;
-		bpf_spin_unlock(&storage->lock);
 		if (is_self_unlink)
 			return -EPERM;
 	}
@@ -68,7 +65,7 @@ int BPF_PROG(unlink_hook, struct inode *dir, struct dentry *victim)
 	return 0;
 }
 
-SEC("lsm/inode_rename")
+SEC("lsm.s/inode_rename")
 int BPF_PROG(inode_rename, struct inode *old_dir, struct dentry *old_dentry,
 	     struct inode *new_dir, struct dentry *new_dentry,
 	     unsigned int flags)
@@ -89,10 +86,8 @@ int BPF_PROG(inode_rename, struct inode *old_dir, struct dentry *old_dentry,
 	if (!storage)
 		return 0;
 
-	bpf_spin_lock(&storage->lock);
 	if (storage->value != DUMMY_STORAGE_VALUE)
 		inode_storage_result = -1;
-	bpf_spin_unlock(&storage->lock);
 
 	err = bpf_inode_storage_delete(&inode_storage_map, old_dentry->d_inode);
 	if (!err)
@@ -101,7 +96,7 @@ int BPF_PROG(inode_rename, struct inode *old_dir, struct dentry *old_dentry,
 	return 0;
 }
 
-SEC("lsm/socket_bind")
+SEC("lsm.s/socket_bind")
 int BPF_PROG(socket_bind, struct socket *sock, struct sockaddr *address,
 	     int addrlen)
 {
@@ -117,10 +112,8 @@ int BPF_PROG(socket_bind, struct socket *sock, struct sockaddr *address,
 	if (!storage)
 		return 0;
 
-	bpf_spin_lock(&storage->lock);
 	if (storage->value != DUMMY_STORAGE_VALUE)
 		sk_storage_result = -1;
-	bpf_spin_unlock(&storage->lock);
 
 	err = bpf_sk_storage_delete(&sk_storage_map, sock->sk);
 	if (!err)
@@ -129,7 +122,7 @@ int BPF_PROG(socket_bind, struct socket *sock, struct sockaddr *address,
 	return 0;
 }
 
-SEC("lsm/socket_post_create")
+SEC("lsm.s/socket_post_create")
 int BPF_PROG(socket_post_create, struct socket *sock, int family, int type,
 	     int protocol, int kern)
 {
@@ -144,9 +137,7 @@ int BPF_PROG(socket_post_create, struct socket *sock, int family, int type,
 	if (!storage)
 		return 0;
 
-	bpf_spin_lock(&storage->lock);
 	storage->value = DUMMY_STORAGE_VALUE;
-	bpf_spin_unlock(&storage->lock);
 
 	return 0;
 }
@@ -154,7 +145,7 @@ int BPF_PROG(socket_post_create, struct socket *sock, int family, int type,
 /* This uses the local storage to remember the inode of the binary that a
  * process was originally executing.
  */
-SEC("lsm/bprm_committed_creds")
+SEC("lsm.s/bprm_committed_creds")
 void BPF_PROG(exec, struct linux_binprm *bprm)
 {
 	__u32 pid = bpf_get_current_pid_tgid() >> 32;
@@ -166,18 +157,13 @@ void BPF_PROG(exec, struct linux_binprm *bprm)
 	storage = bpf_task_storage_get(&task_storage_map,
 				       bpf_get_current_task_btf(), 0,
 				       BPF_LOCAL_STORAGE_GET_F_CREATE);
-	if (storage) {
-		bpf_spin_lock(&storage->lock);
+	if (storage)
 		storage->exec_inode = bprm->file->f_inode;
-		bpf_spin_unlock(&storage->lock);
-	}
 
 	storage = bpf_inode_storage_get(&inode_storage_map, bprm->file->f_inode,
 					0, BPF_LOCAL_STORAGE_GET_F_CREATE);
 	if (!storage)
 		return;
 
-	bpf_spin_lock(&storage->lock);
 	storage->value = DUMMY_STORAGE_VALUE;
-	bpf_spin_unlock(&storage->lock);
 }
Signature · 1 line
2.34.1.448.ga2b2bfdf31-goog
#4 lore raw
In reply to KP Singh
Quoted 11 lines · KP Singh

Local storage is currently unusable in sleepable helpers. One of the important use cases of local_storage is to attach security (or performance) contextual information to kernel objects in LSM / tracing programs to be used later in the life-cyle of the object.

Sometimes this context can only be gathered from sleepable programs (because it needs accesing __user pointers or helpers like bpf_ima_inode_hash). Allowing local storage to be used from sleepable programs allows such context to be managed with the benefits of local_storage.

Quoted 4 lines
# v2 -> v3

* Fixed some RCU issues pointed by Martin
* Added Martin's ack

Applied, Thanks