[PATCH bpf-next v4 0/7] Notify user space when a struct_ops object is detached/unregistered

[PATCH bpf-next v4 0/7] Notify user space when a struct_ops object is detached/unregistered

[Kui-Feng Lee]

The subsystems managing struct_ops objects may need to detach a
struct_ops object due to errors or other reasons. It would be useful
to notify user space programs so that error recovery or logging can be
carried out.

This patch set enables the detach feature for struct_ops links and
send an event to epoll when a link is detached.  Subsystems could call
link->ops->detach() to detach a link and notify user space programs
through epoll.

The signatures of callback functions in "struct bpf_struct_ops" have
been changed as well to pass an extra link argument to
subsystems. Subsystems could detach the links received from reg() and
update() callbacks if there is. This also provides a way that
subsystems can distinguish registrations for an object that has been
registered multiple times for several links.

However, bpf struct_ops maps without BPF_F_LINK have no any link.
Subsystems will receive NULL link pointer for this case.

---
Changes from v3:

 - Add a comment to explain why holding update_mutex is not necessary
   in bpf_struct_ops_link_create()

 - Use rcu_access_pointer() in bpf_struct_ops_map_link_poll().

Changes from v2:

 - Rephrased commit logs and comments.

 - Addressed some mistakes from patch splitting.

 - Replace mutex with spinlock in bpf_testmod.c to address lockdep
   Splat and simplify the implementation.

 - Fix an argument passing to rcu_dereference_protected().

Changes from v1:

 - Pass a link to reg, unreg, and update callbacks.

 - Provide a function to detach a link from underlying subsystems.

 - Add a kfunc to mimic detachments from subsystems, and provide a
   flexible way to control when to do detachments.

 - Add two tests to detach a link from the subsystem after the refcount
   of the link drops to zero.

v3: https://lore.kernel.org/all/20240510002942.1253354-1-thinker.li@gmail.com/
v2: https://lore.kernel.org/all/20240507055600.2382627-1-thinker.li@gmail.com/
v1: https://lore.kernel.org/all/20240429213609.487820-1-thinker.li@gmail.com/

Kui-Feng Lee (7):
  bpf: pass bpf_struct_ops_link to callbacks in bpf_struct_ops.
  bpf: enable detaching links of struct_ops objects.
  bpf: support epoll from bpf struct_ops links.
  bpf: export bpf_link_inc_not_zero.
  selftests/bpf: test struct_ops with epoll
  selftests/bpf: detach a struct_ops link from the subsystem managing
    it.
  selftests/bpf: make sure bpf_testmod handling racing link destroying
    well.

 include/linux/bpf.h                           |  13 +-
 kernel/bpf/bpf_struct_ops.c                   |  90 ++++++++--
 kernel/bpf/syscall.c                          |  14 +-
 net/bpf/bpf_dummy_struct_ops.c                |   4 +-
 net/ipv4/bpf_tcp_ca.c                         |   6 +-
 .../bpf/bpf_test_no_cfi/bpf_test_no_cfi.c     |   4 +-
 .../selftests/bpf/bpf_testmod/bpf_testmod.c   |  48 ++++-
 .../bpf/bpf_testmod/bpf_testmod_kfunc.h       |   1 +
 .../bpf/prog_tests/test_struct_ops_module.c   | 168 ++++++++++++++++++
 .../selftests/bpf/progs/struct_ops_detach.c   |  16 ++
 10 files changed, 339 insertions(+), 25 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/progs/struct_ops_detach.c

-- 
2.34.1

[PATCH bpf-next v4 1/7] bpf: pass bpf_struct_ops_link to callbacks in bpf_struct_ops.

[Kui-Feng Lee]

Pass an additional pointer of bpf_struct_ops_link to callback function reg,
unreg, and update provided by subsystems defined in bpf_struct_ops. A
bpf_struct_ops_map can be registered for multiple links. Passing a pointer
of bpf_struct_ops_link helps subsystems to distinguish them.

This pointer will be used in the later patches to let the subsystem
initiate a detachment on a link that was registered to it previously.

Signed-off-by: Kui-Feng Lee <thinker.li@gmail.com>
---
 include/linux/bpf.h                                    |  6 +++---
 kernel/bpf/bpf_struct_ops.c                            | 10 +++++-----
 net/bpf/bpf_dummy_struct_ops.c                         |  4 ++--
 net/ipv4/bpf_tcp_ca.c                                  |  6 +++---
 .../selftests/bpf/bpf_test_no_cfi/bpf_test_no_cfi.c    |  4 ++--
 tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c  |  6 +++---
 6 files changed, 18 insertions(+), 18 deletions(-)

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index 90094400cc63..b600767ebe02 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -1730,9 +1730,9 @@ struct bpf_struct_ops {
 	int (*init_member)(const struct btf_type *t,
 			   const struct btf_member *member,
 			   void *kdata, const void *udata);
-	int (*reg)(void *kdata);
-	void (*unreg)(void *kdata);
-	int (*update)(void *kdata, void *old_kdata);
+	int (*reg)(void *kdata, struct bpf_link *link);
+	void (*unreg)(void *kdata, struct bpf_link *link);
+	int (*update)(void *kdata, void *old_kdata, struct bpf_link *link);
 	int (*validate)(void *kdata);
 	void *cfi_stubs;
 	struct module *owner;
diff --git a/kernel/bpf/bpf_struct_ops.c b/kernel/bpf/bpf_struct_ops.c
index 86c7884abaf8..1542dded7489 100644
--- a/kernel/bpf/bpf_struct_ops.c
+++ b/kernel/bpf/bpf_struct_ops.c
@@ -757,7 +757,7 @@ static long bpf_struct_ops_map_update_elem(struct bpf_map *map, void *key,
 		goto unlock;
 	}
 
-	err = st_ops->reg(kdata);
+	err = st_ops->reg(kdata, NULL);
 	if (likely(!err)) {
 		/* This refcnt increment on the map here after
 		 * 'st_ops->reg()' is secure since the state of the
@@ -805,7 +805,7 @@ static long bpf_struct_ops_map_delete_elem(struct bpf_map *map, void *key)
 			     BPF_STRUCT_OPS_STATE_TOBEFREE);
 	switch (prev_state) {
 	case BPF_STRUCT_OPS_STATE_INUSE:
-		st_map->st_ops_desc->st_ops->unreg(&st_map->kvalue.data);
+		st_map->st_ops_desc->st_ops->unreg(&st_map->kvalue.data, NULL);
 		bpf_map_put(map);
 		return 0;
 	case BPF_STRUCT_OPS_STATE_TOBEFREE:
@@ -1060,7 +1060,7 @@ static void bpf_struct_ops_map_link_dealloc(struct bpf_link *link)
 		/* st_link->map can be NULL if
 		 * bpf_struct_ops_link_create() fails to register.
 		 */
-		st_map->st_ops_desc->st_ops->unreg(&st_map->kvalue.data);
+		st_map->st_ops_desc->st_ops->unreg(&st_map->kvalue.data, link);
 		bpf_map_put(&st_map->map);
 	}
 	kfree(st_link);
@@ -1125,7 +1125,7 @@ static int bpf_struct_ops_map_link_update(struct bpf_link *link, struct bpf_map
 		goto err_out;
 	}
 
-	err = st_map->st_ops_desc->st_ops->update(st_map->kvalue.data, old_st_map->kvalue.data);
+	err = st_map->st_ops_desc->st_ops->update(st_map->kvalue.data, old_st_map->kvalue.data, link);
 	if (err)
 		goto err_out;
 
@@ -1176,7 +1176,7 @@ int bpf_struct_ops_link_create(union bpf_attr *attr)
 	if (err)
 		goto err_out;
 
-	err = st_map->st_ops_desc->st_ops->reg(st_map->kvalue.data);
+	err = st_map->st_ops_desc->st_ops->reg(st_map->kvalue.data, &link->link);
 	if (err) {
 		bpf_link_cleanup(&link_primer);
 		link = NULL;
diff --git a/net/bpf/bpf_dummy_struct_ops.c b/net/bpf/bpf_dummy_struct_ops.c
index 891cdf61c65a..3ea52b05adfb 100644
--- a/net/bpf/bpf_dummy_struct_ops.c
+++ b/net/bpf/bpf_dummy_struct_ops.c
@@ -272,12 +272,12 @@ static int bpf_dummy_init_member(const struct btf_type *t,
 	return -EOPNOTSUPP;
 }
 
-static int bpf_dummy_reg(void *kdata)
+static int bpf_dummy_reg(void *kdata, struct bpf_link *link)
 {
 	return -EOPNOTSUPP;
 }
 
-static void bpf_dummy_unreg(void *kdata)
+static void bpf_dummy_unreg(void *kdata, struct bpf_link *link)
 {
 }
 
diff --git a/net/ipv4/bpf_tcp_ca.c b/net/ipv4/bpf_tcp_ca.c
index 18227757ec0c..3f88d0961e5b 100644
--- a/net/ipv4/bpf_tcp_ca.c
+++ b/net/ipv4/bpf_tcp_ca.c
@@ -260,17 +260,17 @@ static int bpf_tcp_ca_check_member(const struct btf_type *t,
 	return 0;
 }
 
-static int bpf_tcp_ca_reg(void *kdata)
+static int bpf_tcp_ca_reg(void *kdata, struct bpf_link *link)
 {
 	return tcp_register_congestion_control(kdata);
 }
 
-static void bpf_tcp_ca_unreg(void *kdata)
+static void bpf_tcp_ca_unreg(void *kdata, struct bpf_link *link)
 {
 	tcp_unregister_congestion_control(kdata);
 }
 
-static int bpf_tcp_ca_update(void *kdata, void *old_kdata)
+static int bpf_tcp_ca_update(void *kdata, void *old_kdata, struct bpf_link *link)
 {
 	return tcp_update_congestion_control(kdata, old_kdata);
 }
diff --git a/tools/testing/selftests/bpf/bpf_test_no_cfi/bpf_test_no_cfi.c b/tools/testing/selftests/bpf/bpf_test_no_cfi/bpf_test_no_cfi.c
index b1dd889d5d7d..948eb3962732 100644
--- a/tools/testing/selftests/bpf/bpf_test_no_cfi/bpf_test_no_cfi.c
+++ b/tools/testing/selftests/bpf/bpf_test_no_cfi/bpf_test_no_cfi.c
@@ -22,12 +22,12 @@ static int dummy_init_member(const struct btf_type *t,
 	return 0;
 }
 
-static int dummy_reg(void *kdata)
+static int dummy_reg(void *kdata, struct bpf_link *link)
 {
 	return 0;
 }
 
-static void dummy_unreg(void *kdata)
+static void dummy_unreg(void *kdata, struct bpf_link *link)
 {
 }
 
diff --git a/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c b/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c
index 2a18bd320e92..0a09732cde4b 100644
--- a/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c
+++ b/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c
@@ -820,7 +820,7 @@ static const struct bpf_verifier_ops bpf_testmod_verifier_ops = {
 	.is_valid_access = bpf_testmod_ops_is_valid_access,
 };
 
-static int bpf_dummy_reg(void *kdata)
+static int bpf_dummy_reg(void *kdata, struct bpf_link *link)
 {
 	struct bpf_testmod_ops *ops = kdata;
 
@@ -835,7 +835,7 @@ static int bpf_dummy_reg(void *kdata)
 	return 0;
 }
 
-static void bpf_dummy_unreg(void *kdata)
+static void bpf_dummy_unreg(void *kdata, struct bpf_link *link)
 {
 }
 
@@ -871,7 +871,7 @@ struct bpf_struct_ops bpf_bpf_testmod_ops = {
 	.owner = THIS_MODULE,
 };
 
-static int bpf_dummy_reg2(void *kdata)
+static int bpf_dummy_reg2(void *kdata, struct bpf_link *link)
 {
 	struct bpf_testmod_ops2 *ops = kdata;
 
-- 
2.34.1

[PATCH bpf-next v4 2/7] bpf: enable detaching links of struct_ops objects.

[Kui-Feng Lee]

Implement the detach callback in bpf_link_ops for struct_ops so that user
programs can detach a struct_ops link. The subsystems that struct_ops
objects are registered to can also use this callback to detach the links
being passed to them.

Signed-off-by: Kui-Feng Lee <thinker.li@gmail.com>
---
 kernel/bpf/bpf_struct_ops.c | 63 +++++++++++++++++++++++++++++++++----
 1 file changed, 57 insertions(+), 6 deletions(-)

diff --git a/kernel/bpf/bpf_struct_ops.c b/kernel/bpf/bpf_struct_ops.c
index 1542dded7489..fb6e8a3190ef 100644
--- a/kernel/bpf/bpf_struct_ops.c
+++ b/kernel/bpf/bpf_struct_ops.c
@@ -1057,9 +1057,6 @@ static void bpf_struct_ops_map_link_dealloc(struct bpf_link *link)
 	st_map = (struct bpf_struct_ops_map *)
 		rcu_dereference_protected(st_link->map, true);
 	if (st_map) {
-		/* st_link->map can be NULL if
-		 * bpf_struct_ops_link_create() fails to register.
-		 */
 		st_map->st_ops_desc->st_ops->unreg(&st_map->kvalue.data, link);
 		bpf_map_put(&st_map->map);
 	}
@@ -1075,7 +1072,8 @@ static void bpf_struct_ops_map_link_show_fdinfo(const struct bpf_link *link,
 	st_link = container_of(link, struct bpf_struct_ops_link, link);
 	rcu_read_lock();
 	map = rcu_dereference(st_link->map);
-	seq_printf(seq, "map_id:\t%d\n", map->id);
+	if (map)
+		seq_printf(seq, "map_id:\t%d\n", map->id);
 	rcu_read_unlock();
 }
 
@@ -1088,7 +1086,8 @@ static int bpf_struct_ops_map_link_fill_link_info(const struct bpf_link *link,
 	st_link = container_of(link, struct bpf_struct_ops_link, link);
 	rcu_read_lock();
 	map = rcu_dereference(st_link->map);
-	info->struct_ops.map_id = map->id;
+	if (map)
+		info->struct_ops.map_id = map->id;
 	rcu_read_unlock();
 	return 0;
 }
@@ -1113,6 +1112,10 @@ static int bpf_struct_ops_map_link_update(struct bpf_link *link, struct bpf_map
 	mutex_lock(&update_mutex);
 
 	old_map = rcu_dereference_protected(st_link->map, lockdep_is_held(&update_mutex));
+	if (!old_map) {
+		err = -EINVAL;
+		goto err_out;
+	}
 	if (expected_old_map && old_map != expected_old_map) {
 		err = -EPERM;
 		goto err_out;
@@ -1139,8 +1142,37 @@ static int bpf_struct_ops_map_link_update(struct bpf_link *link, struct bpf_map
 	return err;
 }
 
+static int bpf_struct_ops_map_link_detach(struct bpf_link *link)
+{
+	struct bpf_struct_ops_link *st_link = container_of(link, struct bpf_struct_ops_link, link);
+	struct bpf_struct_ops_map *st_map;
+	struct bpf_map *map;
+
+	mutex_lock(&update_mutex);
+
+	map = rcu_dereference_protected(st_link->map, lockdep_is_held(&update_mutex));
+	if (!map) {
+		mutex_unlock(&update_mutex);
+		return -EINVAL;
+	}
+	st_map = container_of(map, struct bpf_struct_ops_map, map);
+
+	st_map->st_ops_desc->st_ops->unreg(&st_map->kvalue.data, link);
+
+	rcu_assign_pointer(st_link->map, NULL);
+	/* Pair with bpf_map_get() in bpf_struct_ops_link_create() or
+	 * bpf_map_inc() in bpf_struct_ops_map_link_update().
+	 */
+	bpf_map_put(&st_map->map);
+
+	mutex_unlock(&update_mutex);
+
+	return 0;
+}
+
 static const struct bpf_link_ops bpf_struct_ops_map_lops = {
 	.dealloc = bpf_struct_ops_map_link_dealloc,
+	.detach = bpf_struct_ops_map_link_detach,
 	.show_fdinfo = bpf_struct_ops_map_link_show_fdinfo,
 	.fill_link_info = bpf_struct_ops_map_link_fill_link_info,
 	.update_map = bpf_struct_ops_map_link_update,
@@ -1176,13 +1208,32 @@ int bpf_struct_ops_link_create(union bpf_attr *attr)
 	if (err)
 		goto err_out;
 
+	/* Init link->map before calling reg() in case being detached
+	 * immediately.
+	 */
+	RCU_INIT_POINTER(link->map, map);
+
+	/* Once reg() is called, the object and link is already available
+	 * to the subsystem, and it can call
+	 * bpf_struct_ops_map_link_detach() to unreg() it. However, it is
+	 * sfae not holding update_mutex here.
+	 *
+	 * In the case of failure in reg(), the subsystem has no reason to
+	 * call bpf_struct_ops_map_link_detach() since the object is not
+	 * accepted by it. In the case of success, the subsystem may call
+	 * bpf_struct_ops_map_link_detach() to unreg() it, but we don't
+	 * change the content of the link anymore except changing link->id
+	 * in bpf_link_settle(). So, it is safe to not hold update_mutex
+	 * here.
+	 */
 	err = st_map->st_ops_desc->st_ops->reg(st_map->kvalue.data, &link->link);
 	if (err) {
+		RCU_INIT_POINTER(link->map, NULL);
 		bpf_link_cleanup(&link_primer);
+		/* The link has been free by bpf_link_cleanup() */
 		link = NULL;
 		goto err_out;
 	}
-	RCU_INIT_POINTER(link->map, map);
 
 	return bpf_link_settle(&link_primer);
 
-- 
2.34.1

Re: [PATCH bpf-next v4 2/7] bpf: enable detaching links of struct_ops objects.

[Martin KaFai Lau]

On 5/21/24 3:51 PM, Kui-Feng Lee wrote:
> Implement the detach callback in bpf_link_ops for struct_ops so that user
> programs can detach a struct_ops link. The subsystems that struct_ops
> objects are registered to can also use this callback to detach the links
> being passed to them.
> 
> Signed-off-by: Kui-Feng Lee <thinker.li@gmail.com>
> ---
>   kernel/bpf/bpf_struct_ops.c | 63 +++++++++++++++++++++++++++++++++----
>   1 file changed, 57 insertions(+), 6 deletions(-)
> 
> diff --git a/kernel/bpf/bpf_struct_ops.c b/kernel/bpf/bpf_struct_ops.c
> index 1542dded7489..fb6e8a3190ef 100644
> --- a/kernel/bpf/bpf_struct_ops.c
> +++ b/kernel/bpf/bpf_struct_ops.c
> @@ -1057,9 +1057,6 @@ static void bpf_struct_ops_map_link_dealloc(struct bpf_link *link)
>   	st_map = (struct bpf_struct_ops_map *)
>   		rcu_dereference_protected(st_link->map, true);
>   	if (st_map) {
> -		/* st_link->map can be NULL if
> -		 * bpf_struct_ops_link_create() fails to register.
> -		 */
>   		st_map->st_ops_desc->st_ops->unreg(&st_map->kvalue.data, link);
>   		bpf_map_put(&st_map->map);
>   	}
> @@ -1075,7 +1072,8 @@ static void bpf_struct_ops_map_link_show_fdinfo(const struct bpf_link *link,
>   	st_link = container_of(link, struct bpf_struct_ops_link, link);
>   	rcu_read_lock();
>   	map = rcu_dereference(st_link->map);
> -	seq_printf(seq, "map_id:\t%d\n", map->id);
> +	if (map)
> +		seq_printf(seq, "map_id:\t%d\n", map->id);
>   	rcu_read_unlock();
>   }
>   
> @@ -1088,7 +1086,8 @@ static int bpf_struct_ops_map_link_fill_link_info(const struct bpf_link *link,
>   	st_link = container_of(link, struct bpf_struct_ops_link, link);
>   	rcu_read_lock();
>   	map = rcu_dereference(st_link->map);
> -	info->struct_ops.map_id = map->id;
> +	if (map)
> +		info->struct_ops.map_id = map->id;
>   	rcu_read_unlock();
>   	return 0;
>   }
> @@ -1113,6 +1112,10 @@ static int bpf_struct_ops_map_link_update(struct bpf_link *link, struct bpf_map
>   	mutex_lock(&update_mutex);
>   
>   	old_map = rcu_dereference_protected(st_link->map, lockdep_is_held(&update_mutex));
> +	if (!old_map) {
> +		err = -EINVAL;

Just noticed this while checking the return value in patch 3.

This should be -ENOLINK such that it is consistent to the other links' 
.update_prog (e.g. cgroup, tcx, net_namespace...).

> +		goto err_out;
> +	}
>   	if (expected_old_map && old_map != expected_old_map) {
>   		err = -EPERM;
>   		goto err_out;
> @@ -1139,8 +1142,37 @@ static int bpf_struct_ops_map_link_update(struct bpf_link *link, struct bpf_map
>   	return err;
>   }
>   
> +static int bpf_struct_ops_map_link_detach(struct bpf_link *link)
> +{
> +	struct bpf_struct_ops_link *st_link = container_of(link, struct bpf_struct_ops_link, link);
> +	struct bpf_struct_ops_map *st_map;
> +	struct bpf_map *map;
> +
> +	mutex_lock(&update_mutex);
> +
> +	map = rcu_dereference_protected(st_link->map, lockdep_is_held(&update_mutex));
> +	if (!map) {
> +		mutex_unlock(&update_mutex);
> +		return -EINVAL;

Same here but should be always 0 (detach always succeeds).

> +	}
> +	st_map = container_of(map, struct bpf_struct_ops_map, map);
> +
> +	st_map->st_ops_desc->st_ops->unreg(&st_map->kvalue.data, link);
> +
> +	rcu_assign_pointer(st_link->map, NULL);
> +	/* Pair with bpf_map_get() in bpf_struct_ops_link_create() or
> +	 * bpf_map_inc() in bpf_struct_ops_map_link_update().
> +	 */
> +	bpf_map_put(&st_map->map);
> +
> +	mutex_unlock(&update_mutex);
> +
> +	return 0;
> +}
> +
>   static const struct bpf_link_ops bpf_struct_ops_map_lops = {
>   	.dealloc = bpf_struct_ops_map_link_dealloc,
> +	.detach = bpf_struct_ops_map_link_detach,
>   	.show_fdinfo = bpf_struct_ops_map_link_show_fdinfo,
>   	.fill_link_info = bpf_struct_ops_map_link_fill_link_info,
>   	.update_map = bpf_struct_ops_map_link_update,
> @@ -1176,13 +1208,32 @@ int bpf_struct_ops_link_create(union bpf_attr *attr)
>   	if (err)
>   		goto err_out;
>   
> +	/* Init link->map before calling reg() in case being detached
> +	 * immediately.
> +	 */
> +	RCU_INIT_POINTER(link->map, map);
> +
> +	/* Once reg() is called, the object and link is already available
> +	 * to the subsystem, and it can call
> +	 * bpf_struct_ops_map_link_detach() to unreg() it. However, it is
> +	 * sfae not holding update_mutex here.
> +	 *
> +	 * In the case of failure in reg(), the subsystem has no reason to
> +	 * call bpf_struct_ops_map_link_detach() since the object is not
> +	 * accepted by it. In the case of success, the subsystem may call
> +	 * bpf_struct_ops_map_link_detach() to unreg() it, but we don't
> +	 * change the content of the link anymore except changing link->id
> +	 * in bpf_link_settle(). So, it is safe to not hold update_mutex
> +	 * here.

After sleeping on the RCU_INIT_POINTER dance and re-reading this comment, I need 
to walk back my early reply.

Instead of having comment to explain the RCU_INIT_POINTER dance (resetting it to 
NULL on reg() err because bpf_struct_ops_map_link_dealloc is not supposed to 
unreg when the reg did fail), how about simplifying it and just take the 
update_mutex here such that the subsystem cannot detach until the 
RCU_INIT_POINTER(link->map, map) is done. Performance is not a concern here, so 
I would prefer simplicity.

> +	 */
>   	err = st_map->st_ops_desc->st_ops->reg(st_map->kvalue.data, &link->link);
>   	if (err) {
> +		RCU_INIT_POINTER(link->map, NULL);
>   		bpf_link_cleanup(&link_primer);
> +		/* The link has been free by bpf_link_cleanup() */
>   		link = NULL;
>   		goto err_out;
>   	}
> -	RCU_INIT_POINTER(link->map, map);
>   
>   	return bpf_link_settle(&link_primer);
>   

Re: [PATCH bpf-next v4 2/7] bpf: enable detaching links of struct_ops objects.

[Kui-Feng Lee]

On 5/23/24 11:09, Martin KaFai Lau wrote:
> On 5/21/24 3:51 PM, Kui-Feng Lee wrote:
>> Implement the detach callback in bpf_link_ops for struct_ops so that user
>> programs can detach a struct_ops link. The subsystems that struct_ops
>> objects are registered to can also use this callback to detach the links
>> being passed to them.
>>
>> Signed-off-by: Kui-Feng Lee <thinker.li@gmail.com>
>> ---
>>   kernel/bpf/bpf_struct_ops.c | 63 +++++++++++++++++++++++++++++++++----
>>   1 file changed, 57 insertions(+), 6 deletions(-)
>>
>> diff --git a/kernel/bpf/bpf_struct_ops.c b/kernel/bpf/bpf_struct_ops.c
>> index 1542dded7489..fb6e8a3190ef 100644
>> --- a/kernel/bpf/bpf_struct_ops.c
>> +++ b/kernel/bpf/bpf_struct_ops.c
>> @@ -1057,9 +1057,6 @@ static void 
>> bpf_struct_ops_map_link_dealloc(struct bpf_link *link)
>>       st_map = (struct bpf_struct_ops_map *)
>>           rcu_dereference_protected(st_link->map, true);
>>       if (st_map) {
>> -        /* st_link->map can be NULL if
>> -         * bpf_struct_ops_link_create() fails to register.
>> -         */
>>           st_map->st_ops_desc->st_ops->unreg(&st_map->kvalue.data, link);
>>           bpf_map_put(&st_map->map);
>>       }
>> @@ -1075,7 +1072,8 @@ static void 
>> bpf_struct_ops_map_link_show_fdinfo(const struct bpf_link *link,
>>       st_link = container_of(link, struct bpf_struct_ops_link, link);
>>       rcu_read_lock();
>>       map = rcu_dereference(st_link->map);
>> -    seq_printf(seq, "map_id:\t%d\n", map->id);
>> +    if (map)
>> +        seq_printf(seq, "map_id:\t%d\n", map->id);
>>       rcu_read_unlock();
>>   }
>> @@ -1088,7 +1086,8 @@ static int 
>> bpf_struct_ops_map_link_fill_link_info(const struct bpf_link *link,
>>       st_link = container_of(link, struct bpf_struct_ops_link, link);
>>       rcu_read_lock();
>>       map = rcu_dereference(st_link->map);
>> -    info->struct_ops.map_id = map->id;
>> +    if (map)
>> +        info->struct_ops.map_id = map->id;
>>       rcu_read_unlock();
>>       return 0;
>>   }
>> @@ -1113,6 +1112,10 @@ static int 
>> bpf_struct_ops_map_link_update(struct bpf_link *link, struct bpf_map
>>       mutex_lock(&update_mutex);
>>       old_map = rcu_dereference_protected(st_link->map, 
>> lockdep_is_held(&update_mutex));
>> +    if (!old_map) {
>> +        err = -EINVAL;
> 
> Just noticed this while checking the return value in patch 3.
> 
> This should be -ENOLINK such that it is consistent to the other links' 
> .update_prog (e.g. cgroup, tcx, net_namespace...).

Sure

> 
>> +        goto err_out;
>> +    }
>>       if (expected_old_map && old_map != expected_old_map) {
>>           err = -EPERM;
>>           goto err_out;
>> @@ -1139,8 +1142,37 @@ static int 
>> bpf_struct_ops_map_link_update(struct bpf_link *link, struct bpf_map
>>       return err;
>>   }
>> +static int bpf_struct_ops_map_link_detach(struct bpf_link *link)
>> +{
>> +    struct bpf_struct_ops_link *st_link = container_of(link, struct 
>> bpf_struct_ops_link, link);
>> +    struct bpf_struct_ops_map *st_map;
>> +    struct bpf_map *map;
>> +
>> +    mutex_lock(&update_mutex);
>> +
>> +    map = rcu_dereference_protected(st_link->map, 
>> lockdep_is_held(&update_mutex));
>> +    if (!map) {
>> +        mutex_unlock(&update_mutex);
>> +        return -EINVAL;
> 
> Same here but should be always 0 (detach always succeeds).

Got it.

> 
>> +    }
>> +    st_map = container_of(map, struct bpf_struct_ops_map, map);
>> +
>> +    st_map->st_ops_desc->st_ops->unreg(&st_map->kvalue.data, link);
>> +
>> +    rcu_assign_pointer(st_link->map, NULL);
>> +    /* Pair with bpf_map_get() in bpf_struct_ops_link_create() or
>> +     * bpf_map_inc() in bpf_struct_ops_map_link_update().
>> +     */
>> +    bpf_map_put(&st_map->map);
>> +
>> +    mutex_unlock(&update_mutex);
>> +
>> +    return 0;
>> +}
>> +
>>   static const struct bpf_link_ops bpf_struct_ops_map_lops = {
>>       .dealloc = bpf_struct_ops_map_link_dealloc,
>> +    .detach = bpf_struct_ops_map_link_detach,
>>       .show_fdinfo = bpf_struct_ops_map_link_show_fdinfo,
>>       .fill_link_info = bpf_struct_ops_map_link_fill_link_info,
>>       .update_map = bpf_struct_ops_map_link_update,
>> @@ -1176,13 +1208,32 @@ int bpf_struct_ops_link_create(union bpf_attr 
>> *attr)
>>       if (err)
>>           goto err_out;
>> +    /* Init link->map before calling reg() in case being detached
>> +     * immediately.
>> +     */
>> +    RCU_INIT_POINTER(link->map, map);
>> +
>> +    /* Once reg() is called, the object and link is already available
>> +     * to the subsystem, and it can call
>> +     * bpf_struct_ops_map_link_detach() to unreg() it. However, it is
>> +     * sfae not holding update_mutex here.
>> +     *
>> +     * In the case of failure in reg(), the subsystem has no reason to
>> +     * call bpf_struct_ops_map_link_detach() since the object is not
>> +     * accepted by it. In the case of success, the subsystem may call
>> +     * bpf_struct_ops_map_link_detach() to unreg() it, but we don't
>> +     * change the content of the link anymore except changing link->id
>> +     * in bpf_link_settle(). So, it is safe to not hold update_mutex
>> +     * here.
> 
> After sleeping on the RCU_INIT_POINTER dance and re-reading this 
> comment, I need to walk back my early reply.
> 
> Instead of having comment to explain the RCU_INIT_POINTER dance 
> (resetting it to NULL on reg() err because 
> bpf_struct_ops_map_link_dealloc is not supposed to unreg when the reg 
> did fail), how about simplifying it and just take the update_mutex here 
> such that the subsystem cannot detach until the 
> RCU_INIT_POINTER(link->map, map) is done. Performance is not a concern 
> here, so I would prefer simplicity.

sure!

> 
>> +     */
>>       err = st_map->st_ops_desc->st_ops->reg(st_map->kvalue.data, 
>> &link->link);
>>       if (err) {
>> +        RCU_INIT_POINTER(link->map, NULL);
>>           bpf_link_cleanup(&link_primer);
>> +        /* The link has been free by bpf_link_cleanup() */
>>           link = NULL;
>>           goto err_out;
>>       }
>> -    RCU_INIT_POINTER(link->map, map);
>>       return bpf_link_settle(&link_primer);
> 

[PATCH bpf-next v4 3/7] bpf: support epoll from bpf struct_ops links.

[Kui-Feng Lee]

Add epoll support to bpf struct_ops links to trigger EPOLLHUP event upon
detachment.

This patch implements the "poll" of the "struct file_operations" for BPF
links and introduces a new "poll" operator in the "struct bpf_link_ops". By
implementing "poll" of "struct bpf_link_ops" for the links of struct_ops,
the file descriptor of a struct_ops link can be added to an epoll file
descriptor to receive EPOLLHUP events.

Signed-off-by: Kui-Feng Lee <thinker.li@gmail.com>
---
 include/linux/bpf.h         |  1 +
 kernel/bpf/bpf_struct_ops.c | 17 +++++++++++++++++
 kernel/bpf/syscall.c        | 11 +++++++++++
 3 files changed, 29 insertions(+)

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index b600767ebe02..5f7496ef8b7c 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -1612,6 +1612,7 @@ struct bpf_link_ops {
 			      struct bpf_link_info *info);
 	int (*update_map)(struct bpf_link *link, struct bpf_map *new_map,
 			  struct bpf_map *old_map);
+	__poll_t (*poll)(struct file *file, struct poll_table_struct *pts);
 };
 
 struct bpf_tramp_link {
diff --git a/kernel/bpf/bpf_struct_ops.c b/kernel/bpf/bpf_struct_ops.c
index fb6e8a3190ef..794549dc9f4b 100644
--- a/kernel/bpf/bpf_struct_ops.c
+++ b/kernel/bpf/bpf_struct_ops.c
@@ -12,6 +12,7 @@
 #include <linux/mutex.h>
 #include <linux/btf_ids.h>
 #include <linux/rcupdate_wait.h>
+#include <linux/poll.h>
 
 struct bpf_struct_ops_value {
 	struct bpf_struct_ops_common_value common;
@@ -56,6 +57,7 @@ struct bpf_struct_ops_map {
 struct bpf_struct_ops_link {
 	struct bpf_link link;
 	struct bpf_map __rcu *map;
+	wait_queue_head_t wait_hup;
 };
 
 static DEFINE_MUTEX(update_mutex);
@@ -1167,15 +1169,28 @@ static int bpf_struct_ops_map_link_detach(struct bpf_link *link)
 
 	mutex_unlock(&update_mutex);
 
+	wake_up_interruptible_poll(&st_link->wait_hup, EPOLLHUP);
+
 	return 0;
 }
 
+static __poll_t bpf_struct_ops_map_link_poll(struct file *file,
+					     struct poll_table_struct *pts)
+{
+	struct bpf_struct_ops_link *st_link = file->private_data;
+
+	poll_wait(file, &st_link->wait_hup, pts);
+
+	return rcu_access_pointer(st_link->map) ? 0 : EPOLLHUP;
+}
+
 static const struct bpf_link_ops bpf_struct_ops_map_lops = {
 	.dealloc = bpf_struct_ops_map_link_dealloc,
 	.detach = bpf_struct_ops_map_link_detach,
 	.show_fdinfo = bpf_struct_ops_map_link_show_fdinfo,
 	.fill_link_info = bpf_struct_ops_map_link_fill_link_info,
 	.update_map = bpf_struct_ops_map_link_update,
+	.poll = bpf_struct_ops_map_link_poll,
 };
 
 int bpf_struct_ops_link_create(union bpf_attr *attr)
@@ -1213,6 +1228,8 @@ int bpf_struct_ops_link_create(union bpf_attr *attr)
 	 */
 	RCU_INIT_POINTER(link->map, map);
 
+	init_waitqueue_head(&link->wait_hup);
+
 	/* Once reg() is called, the object and link is already available
 	 * to the subsystem, and it can call
 	 * bpf_struct_ops_map_link_detach() to unreg() it. However, it is
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
index 13ad74ecf2cd..ad4f81ed27f0 100644
--- a/kernel/bpf/syscall.c
+++ b/kernel/bpf/syscall.c
@@ -3150,6 +3150,16 @@ static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
 }
 #endif
 
+static __poll_t bpf_link_poll(struct file *file, struct poll_table_struct *pts)
+{
+	struct bpf_link *link = file->private_data;
+
+	if (link->ops->poll)
+		return link->ops->poll(file, pts);
+
+	return 0;
+}
+
 static const struct file_operations bpf_link_fops = {
 #ifdef CONFIG_PROC_FS
 	.show_fdinfo	= bpf_link_show_fdinfo,
@@ -3157,6 +3167,7 @@ static const struct file_operations bpf_link_fops = {
 	.release	= bpf_link_release,
 	.read		= bpf_dummy_read,
 	.write		= bpf_dummy_write,
+	.poll		= bpf_link_poll,
 };
 
 static int bpf_link_alloc_id(struct bpf_link *link)
-- 
2.34.1

Re: [PATCH bpf-next v4 3/7] bpf: support epoll from bpf struct_ops links.

[Martin KaFai Lau]

On 5/21/24 3:51 PM, Kui-Feng Lee wrote:
> +static __poll_t bpf_link_poll(struct file *file, struct poll_table_struct *pts)
> +{
> +	struct bpf_link *link = file->private_data;
> +
> +	if (link->ops->poll)
> +		return link->ops->poll(file, pts);
> +
> +	return 0;

The current bpf_link_fops.poll is NULL before this patch. From vfs_poll, it 
seems to be DEFAULT_POLLMASK for this case. Please double check.

> +}
> +
>   static const struct file_operations bpf_link_fops = {
>   #ifdef CONFIG_PROC_FS
>   	.show_fdinfo	= bpf_link_show_fdinfo,
> @@ -3157,6 +3167,7 @@ static const struct file_operations bpf_link_fops = {
>   	.release	= bpf_link_release,
>   	.read		= bpf_dummy_read,
>   	.write		= bpf_dummy_write,
> +	.poll		= bpf_link_poll,

Same here. What does the epoll_ctl(EPOLL_CTL_ADD) currently expect for link 
(e.g. cgroup) that does not support poll?

>   };
>   

Re: [PATCH bpf-next v4 3/7] bpf: support epoll from bpf struct_ops links.

[Kui-Feng Lee]

On 5/23/24 10:23, Martin KaFai Lau wrote:
> On 5/21/24 3:51 PM, Kui-Feng Lee wrote:
>> +static __poll_t bpf_link_poll(struct file *file, struct 
>> poll_table_struct *pts)
>> +{
>> +    struct bpf_link *link = file->private_data;
>> +
>> +    if (link->ops->poll)
>> +        return link->ops->poll(file, pts);
>> +
>> +    return 0;
> 
> The current bpf_link_fops.poll is NULL before this patch. From vfs_poll, 
> it seems to be DEFAULT_POLLMASK for this case. Please double check.


Yes, it returns DEFAULT_POLLMASK if file->f_op->epoll is NULL. But,
before this patch, link can not be added to an epoll. See the
explanation below.

> 
>> +}
>> +
>>   static const struct file_operations bpf_link_fops = {
>>   #ifdef CONFIG_PROC_FS
>>       .show_fdinfo    = bpf_link_show_fdinfo,
>> @@ -3157,6 +3167,7 @@ static const struct file_operations 
>> bpf_link_fops = {
>>       .release    = bpf_link_release,
>>       .read        = bpf_dummy_read,
>>       .write        = bpf_dummy_write,
>> +    .poll        = bpf_link_poll,
> 
> Same here. What does the epoll_ctl(EPOLL_CTL_ADD) currently expect for 
> link (e.g. cgroup) that does not support poll?
> 

epoll_ctl() always returns -EPERM for files not supporting poll.
Should I add another instance of struct file_operations to keep the
consistency for other types of links?

>>   };
> 

Re: [PATCH bpf-next v4 3/7] bpf: support epoll from bpf struct_ops links.

[Martin KaFai Lau]

On 5/23/24 11:24 AM, Kui-Feng Lee wrote:
> 
> 
> On 5/23/24 10:23, Martin KaFai Lau wrote:
>> On 5/21/24 3:51 PM, Kui-Feng Lee wrote:
>>> +static __poll_t bpf_link_poll(struct file *file, struct poll_table_struct *pts)
>>> +{
>>> +    struct bpf_link *link = file->private_data;
>>> +
>>> +    if (link->ops->poll)
>>> +        return link->ops->poll(file, pts);
>>> +
>>> +    return 0;
>>
>> The current bpf_link_fops.poll is NULL before this patch. From vfs_poll, it 
>> seems to be DEFAULT_POLLMASK for this case. Please double check.
> 
> 
> Yes, it returns DEFAULT_POLLMASK if file->f_op->epoll is NULL. But,
> before this patch, link can not be added to an epoll. See the
> explanation below.

How about select() and poll() that do not need epoll_ctl() setup?

> 
>>
>>> +}
>>> +
>>>   static const struct file_operations bpf_link_fops = {
>>>   #ifdef CONFIG_PROC_FS
>>>       .show_fdinfo    = bpf_link_show_fdinfo,
>>> @@ -3157,6 +3167,7 @@ static const struct file_operations bpf_link_fops = {
>>>       .release    = bpf_link_release,
>>>       .read        = bpf_dummy_read,
>>>       .write        = bpf_dummy_write,
>>> +    .poll        = bpf_link_poll,
>>
>> Same here. What does the epoll_ctl(EPOLL_CTL_ADD) currently expect for link 
>> (e.g. cgroup) that does not support poll?
>>
> 
> epoll_ctl() always returns -EPERM for files not supporting poll.
> Should I add another instance of struct file_operations to keep the
> consistency for other types of links?

imo, it makes sense to have another instance for link that supports poll such 
that epoll_ctl(EPOLL_CTL_ADD) can fail early for the unsupported links.

> 
>>>   };
>>

Re: [PATCH bpf-next v4 3/7] bpf: support epoll from bpf struct_ops links.

[Kui-Feng Lee]

On 5/23/24 11:34, Martin KaFai Lau wrote:
> On 5/23/24 11:24 AM, Kui-Feng Lee wrote:
>>
>>
>> On 5/23/24 10:23, Martin KaFai Lau wrote:
>>> On 5/21/24 3:51 PM, Kui-Feng Lee wrote:
>>>> +static __poll_t bpf_link_poll(struct file *file, struct 
>>>> poll_table_struct *pts)
>>>> +{
>>>> +    struct bpf_link *link = file->private_data;
>>>> +
>>>> +    if (link->ops->poll)
>>>> +        return link->ops->poll(file, pts);
>>>> +
>>>> +    return 0;
>>>
>>> The current bpf_link_fops.poll is NULL before this patch. From 
>>> vfs_poll, it seems to be DEFAULT_POLLMASK for this case. Please 
>>> double check.
>>
>>
>> Yes, it returns DEFAULT_POLLMASK if file->f_op->epoll is NULL. But,
>> before this patch, link can not be added to an epoll. See the
>> explanation below.
> 
> How about select() and poll() that do not need epoll_ctl() setup?

AFAIK, they just don't check it at all, calling vfs_poll() directly.

> 
>>
>>>
>>>> +}
>>>> +
>>>>   static const struct file_operations bpf_link_fops = {
>>>>   #ifdef CONFIG_PROC_FS
>>>>       .show_fdinfo    = bpf_link_show_fdinfo,
>>>> @@ -3157,6 +3167,7 @@ static const struct file_operations 
>>>> bpf_link_fops = {
>>>>       .release    = bpf_link_release,
>>>>       .read        = bpf_dummy_read,
>>>>       .write        = bpf_dummy_write,
>>>> +    .poll        = bpf_link_poll,
>>>
>>> Same here. What does the epoll_ctl(EPOLL_CTL_ADD) currently expect 
>>> for link (e.g. cgroup) that does not support poll?
>>>
>>
>> epoll_ctl() always returns -EPERM for files not supporting poll.
>> Should I add another instance of struct file_operations to keep the
>> consistency for other types of links?
> 
> imo, it makes sense to have another instance for link that supports poll 
> such that epoll_ctl(EPOLL_CTL_ADD) can fail early for the unsupported 
> links.

Ok! I will add another instance.

> 
>>
>>>>   };
>>>
> 

Re: [PATCH bpf-next v4 3/7] bpf: support epoll from bpf struct_ops links.

[Martin KaFai Lau]

On 5/23/24 12:03 PM, Kui-Feng Lee wrote:
> 
> 
> On 5/23/24 11:34, Martin KaFai Lau wrote:
>> On 5/23/24 11:24 AM, Kui-Feng Lee wrote:
>>>
>>>
>>> On 5/23/24 10:23, Martin KaFai Lau wrote:
>>>> On 5/21/24 3:51 PM, Kui-Feng Lee wrote:
>>>>> +static __poll_t bpf_link_poll(struct file *file, struct poll_table_struct 
>>>>> *pts)
>>>>> +{
>>>>> +    struct bpf_link *link = file->private_data;
>>>>> +
>>>>> +    if (link->ops->poll)
>>>>> +        return link->ops->poll(file, pts);
>>>>> +
>>>>> +    return 0;
>>>>
>>>> The current bpf_link_fops.poll is NULL before this patch. From vfs_poll, it 
>>>> seems to be DEFAULT_POLLMASK for this case. Please double check.
>>>
>>>
>>> Yes, it returns DEFAULT_POLLMASK if file->f_op->epoll is NULL. But,
>>> before this patch, link can not be added to an epoll. See the
>>> explanation below.
>>
>> How about select() and poll() that do not need epoll_ctl() setup?
> 
> AFAIK, they just don't check it at all, calling vfs_poll() directly.

right, vfs_poll returns DEFAULT_POLLMASK which is not 0.

#define DEFAULT_POLLMASK (EPOLLIN | EPOLLOUT | EPOLLRDNORM | EPOLLWRNORM)

static inline __poll_t vfs_poll(struct file *file, struct poll_table_struct *pt)
{
	if (unlikely(!file->f_op->poll))
		return DEFAULT_POLLMASK;
	return file->f_op->poll(file, pt);
}

but this discussion is moot if another file_operations instance is used.

> 
>>
>>>
>>>>
>>>>> +}
>>>>> +
>>>>>   static const struct file_operations bpf_link_fops = {
>>>>>   #ifdef CONFIG_PROC_FS
>>>>>       .show_fdinfo    = bpf_link_show_fdinfo,
>>>>> @@ -3157,6 +3167,7 @@ static const struct file_operations bpf_link_fops = {
>>>>>       .release    = bpf_link_release,
>>>>>       .read        = bpf_dummy_read,
>>>>>       .write        = bpf_dummy_write,
>>>>> +    .poll        = bpf_link_poll,
>>>>
>>>> Same here. What does the epoll_ctl(EPOLL_CTL_ADD) currently expect for link 
>>>> (e.g. cgroup) that does not support poll?
>>>>
>>>
>>> epoll_ctl() always returns -EPERM for files not supporting poll.
>>> Should I add another instance of struct file_operations to keep the
>>> consistency for other types of links?
>>
>> imo, it makes sense to have another instance for link that supports poll such 
>> that epoll_ctl(EPOLL_CTL_ADD) can fail early for the unsupported links.
> 
> Ok! I will add another instance.

Re: [PATCH bpf-next v4 3/7] bpf: support epoll from bpf struct_ops links.

[Kui-Feng Lee]

On 5/23/24 12:10, Martin KaFai Lau wrote:
> On 5/23/24 12:03 PM, Kui-Feng Lee wrote:
>>
>>
>> On 5/23/24 11:34, Martin KaFai Lau wrote:
>>> On 5/23/24 11:24 AM, Kui-Feng Lee wrote:
>>>>
>>>>
>>>> On 5/23/24 10:23, Martin KaFai Lau wrote:
>>>>> On 5/21/24 3:51 PM, Kui-Feng Lee wrote:
>>>>>> +static __poll_t bpf_link_poll(struct file *file, struct 
>>>>>> poll_table_struct *pts)
>>>>>> +{
>>>>>> +    struct bpf_link *link = file->private_data;
>>>>>> +
>>>>>> +    if (link->ops->poll)
>>>>>> +        return link->ops->poll(file, pts);
>>>>>> +
>>>>>> +    return 0;
>>>>>
>>>>> The current bpf_link_fops.poll is NULL before this patch. From 
>>>>> vfs_poll, it seems to be DEFAULT_POLLMASK for this case. Please 
>>>>> double check.
>>>>
>>>>
>>>> Yes, it returns DEFAULT_POLLMASK if file->f_op->epoll is NULL. But,
>>>> before this patch, link can not be added to an epoll. See the
>>>> explanation below.
>>>
>>> How about select() and poll() that do not need epoll_ctl() setup?
>>
>> AFAIK, they just don't check it at all, calling vfs_poll() directly.
> 
> right, vfs_poll returns DEFAULT_POLLMASK which is not 0.
> 
> #define DEFAULT_POLLMASK (EPOLLIN | EPOLLOUT | EPOLLRDNORM | EPOLLWRNORM)
> 
> static inline __poll_t vfs_poll(struct file *file, struct 
> poll_table_struct *pt)
> {
>      if (unlikely(!file->f_op->poll))
>          return DEFAULT_POLLMASK;
>      return file->f_op->poll(file, pt);
> }
> 
> but this discussion is moot if another file_operations instance is used.

Sure! I am adding another instance.

> 
>>
>>>
>>>>
>>>>>
>>>>>> +}
>>>>>> +
>>>>>>   static const struct file_operations bpf_link_fops = {
>>>>>>   #ifdef CONFIG_PROC_FS
>>>>>>       .show_fdinfo    = bpf_link_show_fdinfo,
>>>>>> @@ -3157,6 +3167,7 @@ static const struct file_operations 
>>>>>> bpf_link_fops = {
>>>>>>       .release    = bpf_link_release,
>>>>>>       .read        = bpf_dummy_read,
>>>>>>       .write        = bpf_dummy_write,
>>>>>> +    .poll        = bpf_link_poll,
>>>>>
>>>>> Same here. What does the epoll_ctl(EPOLL_CTL_ADD) currently expect 
>>>>> for link (e.g. cgroup) that does not support poll?
>>>>>
>>>>
>>>> epoll_ctl() always returns -EPERM for files not supporting poll.
>>>> Should I add another instance of struct file_operations to keep the
>>>> consistency for other types of links?
>>>
>>> imo, it makes sense to have another instance for link that supports 
>>> poll such that epoll_ctl(EPOLL_CTL_ADD) can fail early for the 
>>> unsupported links.
>>
>> Ok! I will add another instance.
> 

[PATCH bpf-next v4 4/7] bpf: export bpf_link_inc_not_zero.

[Kui-Feng Lee]

bpf_link_inc_not_zero() will be used by kernel modules.  We will use it in
bpf_testmod.c later.

Signed-off-by: Kui-Feng Lee <thinker.li@gmail.com>
---
 include/linux/bpf.h  | 6 ++++++
 kernel/bpf/syscall.c | 3 ++-
 2 files changed, 8 insertions(+), 1 deletion(-)

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index 5f7496ef8b7c..6b592094f9b4 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -2351,6 +2351,7 @@ int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer);
 int bpf_link_settle(struct bpf_link_primer *primer);
 void bpf_link_cleanup(struct bpf_link_primer *primer);
 void bpf_link_inc(struct bpf_link *link);
+struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link);
 void bpf_link_put(struct bpf_link *link);
 int bpf_link_new_fd(struct bpf_link *link);
 struct bpf_link *bpf_link_get_from_fd(u32 ufd);
@@ -2722,6 +2723,11 @@ static inline void bpf_link_inc(struct bpf_link *link)
 {
 }
 
+static inline struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
+{
+	return NULL;
+}
+
 static inline void bpf_link_put(struct bpf_link *link)
 {
 }
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
index ad4f81ed27f0..31fabe26371d 100644
--- a/kernel/bpf/syscall.c
+++ b/kernel/bpf/syscall.c
@@ -5422,10 +5422,11 @@ static int link_detach(union bpf_attr *attr)
 	return ret;
 }
 
-static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
+struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
 {
 	return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
 }
+EXPORT_SYMBOL(bpf_link_inc_not_zero);
 
 struct bpf_link *bpf_link_by_id(u32 id)
 {
-- 
2.34.1

[PATCH bpf-next v4 5/7] selftests/bpf: test struct_ops with epoll

[Kui-Feng Lee]

Verify whether a user space program is informed through epoll with EPOLLHUP
when a struct_ops object is detached.

The BPF code in selftests/bpf/progs/struct_ops_module.c has become
complex. Therefore, struct_ops_detach.c has been added to segregate the BPF
code for detachment tests from the BPF code for other tests based on the
recommendation of Andrii Nakryiko.

Suggested-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Kui-Feng Lee <thinker.li@gmail.com>
---
 .../bpf/prog_tests/test_struct_ops_module.c   | 57 +++++++++++++++++++
 .../selftests/bpf/progs/struct_ops_detach.c   |  9 +++
 2 files changed, 66 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/progs/struct_ops_detach.c

diff --git a/tools/testing/selftests/bpf/prog_tests/test_struct_ops_module.c b/tools/testing/selftests/bpf/prog_tests/test_struct_ops_module.c
index 29e183a80f49..bbcf12696a6b 100644
--- a/tools/testing/selftests/bpf/prog_tests/test_struct_ops_module.c
+++ b/tools/testing/selftests/bpf/prog_tests/test_struct_ops_module.c
@@ -3,9 +3,12 @@
 #include <test_progs.h>
 #include <time.h>
 
+#include <sys/epoll.h>
+
 #include "struct_ops_module.skel.h"
 #include "struct_ops_nulled_out_cb.skel.h"
 #include "struct_ops_forgotten_cb.skel.h"
+#include "struct_ops_detach.skel.h"
 
 static void check_map_info(struct bpf_map_info *info)
 {
@@ -242,6 +245,58 @@ static void test_struct_ops_forgotten_cb(void)
 	struct_ops_forgotten_cb__destroy(skel);
 }
 
+/* Detach a link from a user space program */
+static void test_detach_link(void)
+{
+	struct epoll_event ev, events[2];
+	struct struct_ops_detach *skel;
+	struct bpf_link *link = NULL;
+	int fd, epollfd = -1, nfds;
+	int err;
+
+	skel = struct_ops_detach__open_and_load();
+	if (!ASSERT_OK_PTR(skel, "struct_ops_detach__open_and_load"))
+		return;
+
+	link = bpf_map__attach_struct_ops(skel->maps.testmod_do_detach);
+	if (!ASSERT_OK_PTR(link, "attach_struct_ops"))
+		goto cleanup;
+
+	fd = bpf_link__fd(link);
+	if (!ASSERT_GE(fd, 0, "link_fd"))
+		goto cleanup;
+
+	epollfd = epoll_create1(0);
+	if (!ASSERT_GE(epollfd, 0, "epoll_create1"))
+		goto cleanup;
+
+	ev.events = EPOLLHUP;
+	ev.data.fd = fd;
+	err = epoll_ctl(epollfd, EPOLL_CTL_ADD, fd, &ev);
+	if (!ASSERT_OK(err, "epoll_ctl"))
+		goto cleanup;
+
+	err = bpf_link__detach(link);
+	if (!ASSERT_OK(err, "detach_link"))
+		goto cleanup;
+
+	/* Wait for EPOLLHUP */
+	nfds = epoll_wait(epollfd, events, 2, 500);
+	if (!ASSERT_EQ(nfds, 1, "epoll_wait"))
+		goto cleanup;
+
+	if (!ASSERT_EQ(events[0].data.fd, fd, "epoll_wait_fd"))
+		goto cleanup;
+	if (!ASSERT_TRUE(events[0].events & EPOLLHUP, "events[0].events"))
+		goto cleanup;
+
+cleanup:
+	if (epollfd >= 0)
+		close(epollfd);
+	bpf_link__destroy(link);
+	struct_ops_detach__destroy(skel);
+}
+
 void serial_test_struct_ops_module(void)
 {
 	if (test__start_subtest("struct_ops_load"))
@@ -254,5 +309,7 @@ void serial_test_struct_ops_module(void)
 		test_struct_ops_nulled_out_cb();
 	if (test__start_subtest("struct_ops_forgotten_cb"))
 		test_struct_ops_forgotten_cb();
+	if (test__start_subtest("test_detach_link"))
+		test_detach_link();
 }
 
diff --git a/tools/testing/selftests/bpf/progs/struct_ops_detach.c b/tools/testing/selftests/bpf/progs/struct_ops_detach.c
new file mode 100644
index 000000000000..45eacc2ca657
--- /dev/null
+++ b/tools/testing/selftests/bpf/progs/struct_ops_detach.c
@@ -0,0 +1,9 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2024 Meta Platforms, Inc. and affiliates. */
+#include <vmlinux.h>
+#include "../bpf_testmod/bpf_testmod.h"
+
+char _license[] SEC("license") = "GPL";
+
+SEC(".struct_ops.link")
+struct bpf_testmod_ops testmod_do_detach;
-- 
2.34.1

[PATCH bpf-next v4 6/7] selftests/bpf: detach a struct_ops link from the subsystem managing it.

[Kui-Feng Lee]

Not only a user space program can detach a struct_ops link, the subsystem
managing a link can also detach the link. This patch adds a kfunc to
simulate detaching a link by the subsystem managing it and makes sure user
space programs get notified through epoll.

Signed-off-by: Kui-Feng Lee <thinker.li@gmail.com>
---
 .../selftests/bpf/bpf_testmod/bpf_testmod.c   | 42 ++++++++++++
 .../bpf/bpf_testmod/bpf_testmod_kfunc.h       |  1 +
 .../bpf/prog_tests/test_struct_ops_module.c   | 67 +++++++++++++++++++
 .../selftests/bpf/progs/struct_ops_detach.c   |  7 ++
 4 files changed, 117 insertions(+)

diff --git a/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c b/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c
index 0a09732cde4b..2b3a89609b7e 100644
--- a/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c
+++ b/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c
@@ -744,6 +744,38 @@ __bpf_kfunc int bpf_kfunc_call_kernel_getpeername(struct addr_args *args)
 	return err;
 }
 
+static DEFINE_SPINLOCK(detach_lock);
+static struct bpf_link *link_to_detach;
+
+__bpf_kfunc int bpf_dummy_do_link_detach(void)
+{
+	struct bpf_link *link;
+	int ret = -ENOENT;
+
+	/* A subsystem must ensure that a link is valid when detaching the
+	 * link. In order to achieve that, the subsystem may need to obtain
+	 * a lock to safeguard a table that holds the pointer to the link
+	 * being detached. However, the subsystem cannot invoke
+	 * link->ops->detach() while holding the lock because other tasks
+	 * may be in the process of unregistering, which could lead to
+	 * acquiring the same lock and causing a deadlock. This is why
+	 * bpf_link_inc_not_zero() is used to maintain the link's validity.
+	 */
+	spin_lock(&detach_lock);
+	link = link_to_detach;
+	/* Make sure the link is still valid by increasing its refcnt */
+	if (link && IS_ERR(bpf_link_inc_not_zero(link)))
+		link = NULL;
+	spin_unlock(&detach_lock);
+
+	if (link) {
+		ret = link->ops->detach(link);
+		bpf_link_put(link);
+	}
+
+	return ret;
+}
+
 BTF_KFUNCS_START(bpf_testmod_check_kfunc_ids)
 BTF_ID_FLAGS(func, bpf_testmod_test_mod_kfunc)
 BTF_ID_FLAGS(func, bpf_kfunc_call_test1)
@@ -780,6 +812,7 @@ BTF_ID_FLAGS(func, bpf_kfunc_call_kernel_sendmsg, KF_SLEEPABLE)
 BTF_ID_FLAGS(func, bpf_kfunc_call_sock_sendmsg, KF_SLEEPABLE)
 BTF_ID_FLAGS(func, bpf_kfunc_call_kernel_getsockname, KF_SLEEPABLE)
 BTF_ID_FLAGS(func, bpf_kfunc_call_kernel_getpeername, KF_SLEEPABLE)
+BTF_ID_FLAGS(func, bpf_dummy_do_link_detach)
 BTF_KFUNCS_END(bpf_testmod_check_kfunc_ids)
 
 static int bpf_testmod_ops_init(struct btf *btf)
@@ -832,11 +865,20 @@ static int bpf_dummy_reg(void *kdata, struct bpf_link *link)
 	if (ops->test_2)
 		ops->test_2(4, ops->data);
 
+	spin_lock(&detach_lock);
+	if (!link_to_detach)
+		link_to_detach = link;
+	spin_unlock(&detach_lock);
+
 	return 0;
 }
 
 static void bpf_dummy_unreg(void *kdata, struct bpf_link *link)
 {
+	spin_lock(&detach_lock);
+	if (link == link_to_detach)
+		link_to_detach = NULL;
+	spin_unlock(&detach_lock);
 }
 
 static int bpf_testmod_test_1(void)
diff --git a/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod_kfunc.h b/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod_kfunc.h
index b0d586a6751f..19131baf4a9e 100644
--- a/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod_kfunc.h
+++ b/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod_kfunc.h
@@ -121,6 +121,7 @@ void bpf_kfunc_call_test_fail1(struct prog_test_fail1 *p);
 void bpf_kfunc_call_test_fail2(struct prog_test_fail2 *p);
 void bpf_kfunc_call_test_fail3(struct prog_test_fail3 *p);
 void bpf_kfunc_call_test_mem_len_fail1(void *mem, int len);
+int bpf_dummy_do_link_detach(void) __ksym;
 
 void bpf_kfunc_common_test(void) __ksym;
 
diff --git a/tools/testing/selftests/bpf/prog_tests/test_struct_ops_module.c b/tools/testing/selftests/bpf/prog_tests/test_struct_ops_module.c
index bbcf12696a6b..f4000bf04752 100644
--- a/tools/testing/selftests/bpf/prog_tests/test_struct_ops_module.c
+++ b/tools/testing/selftests/bpf/prog_tests/test_struct_ops_module.c
@@ -2,6 +2,7 @@
 /* Copyright (c) 2024 Meta Platforms, Inc. and affiliates. */
 #include <test_progs.h>
 #include <time.h>
+#include <network_helpers.h>
 
 #include <sys/epoll.h>
 
@@ -297,6 +298,70 @@ static void test_detach_link(void)
 	struct_ops_detach__destroy(skel);
 }
 
+/* Detach a link from the subsystem that the link was registered to */
+static void test_subsystem_detach(void)
+{
+	LIBBPF_OPTS(bpf_test_run_opts, topts,
+		    .data_in = &pkt_v4,
+		    .data_size_in = sizeof(pkt_v4));
+	struct epoll_event ev, events[2];
+	struct struct_ops_detach *skel;
+	struct bpf_link *link = NULL;
+	int fd, epollfd = -1, nfds;
+	int prog_fd;
+	int err;
+
+	skel = struct_ops_detach__open_and_load();
+	if (!ASSERT_OK_PTR(skel, "struct_ops_detach_open_and_load"))
+		return;
+
+	link = bpf_map__attach_struct_ops(skel->maps.testmod_do_detach);
+	if (!ASSERT_OK_PTR(link, "attach_struct_ops"))
+		goto cleanup;
+
+	fd = bpf_link__fd(link);
+	if (!ASSERT_GE(fd, 0, "link_fd"))
+		goto cleanup;
+
+	prog_fd = bpf_program__fd(skel->progs.start_detach);
+	if (!ASSERT_GE(prog_fd, 0, "start_detach_fd"))
+		goto cleanup;
+
+	/* Do detachment from the registered subsystem */
+	err = bpf_prog_test_run_opts(prog_fd, &topts);
+	if (!ASSERT_OK(err, "start_detach_run"))
+		goto cleanup;
+
+	if (!ASSERT_EQ(topts.retval, 0, "start_detach_run_retval"))
+		goto cleanup;
+
+	epollfd = epoll_create1(0);
+	if (!ASSERT_GE(epollfd, 0, "epoll_create1"))
+		goto cleanup;
+
+	ev.events = EPOLLHUP;
+	ev.data.fd = fd;
+	err = epoll_ctl(epollfd, EPOLL_CTL_ADD, fd, &ev);
+	if (!ASSERT_OK(err, "epoll_ctl"))
+		goto cleanup;
+
+	/* Wait for EPOLLHUP */
+	nfds = epoll_wait(epollfd, events, 2, 5000);
+	if (!ASSERT_EQ(nfds, 1, "epoll_wait"))
+		goto cleanup;
+
+	if (!ASSERT_EQ(events[0].data.fd, fd, "epoll_wait_fd"))
+		goto cleanup;
+	if (!ASSERT_TRUE(events[0].events & EPOLLHUP, "events[0].events"))
+		goto cleanup;
+
+cleanup:
+	if (epollfd >= 0)
+		close(epollfd);
+	bpf_link__destroy(link);
+	struct_ops_detach__destroy(skel);
+}
+
 void serial_test_struct_ops_module(void)
 {
 	if (test__start_subtest("struct_ops_load"))
@@ -311,5 +376,7 @@ void serial_test_struct_ops_module(void)
 		test_struct_ops_forgotten_cb();
 	if (test__start_subtest("test_detach_link"))
 		test_detach_link();
+	if (test__start_subtest("test_subsystem_detach"))
+		test_subsystem_detach();
 }
 
diff --git a/tools/testing/selftests/bpf/progs/struct_ops_detach.c b/tools/testing/selftests/bpf/progs/struct_ops_detach.c
index 45eacc2ca657..5c742b0df04d 100644
--- a/tools/testing/selftests/bpf/progs/struct_ops_detach.c
+++ b/tools/testing/selftests/bpf/progs/struct_ops_detach.c
@@ -2,8 +2,15 @@
 /* Copyright (c) 2024 Meta Platforms, Inc. and affiliates. */
 #include <vmlinux.h>
 #include "../bpf_testmod/bpf_testmod.h"
+#include "../bpf_testmod/bpf_testmod_kfunc.h"
 
 char _license[] SEC("license") = "GPL";
 
 SEC(".struct_ops.link")
 struct bpf_testmod_ops testmod_do_detach;
+
+SEC("tc")
+int start_detach(void *skb)
+{
+	return bpf_dummy_do_link_detach();
+}
-- 
2.34.1

[PATCH bpf-next v4 7/7] selftests/bpf: make sure bpf_testmod handling racing link destroying well.

[Kui-Feng Lee]

Do detachment from the subsystem after a link being closed/freed.  This
test make sure the pattern implemented by bpf_dummy_do_link_detach() works
correctly.

Refer to bpf_dummy_do_link_detach() in bpf_testmod.c for more details.

Signed-off-by: Kui-Feng Lee <thinker.li@gmail.com>
---
 .../bpf/prog_tests/test_struct_ops_module.c   | 44 +++++++++++++++++++
 1 file changed, 44 insertions(+)

diff --git a/tools/testing/selftests/bpf/prog_tests/test_struct_ops_module.c b/tools/testing/selftests/bpf/prog_tests/test_struct_ops_module.c
index f4000bf04752..3a8cdf440edd 100644
--- a/tools/testing/selftests/bpf/prog_tests/test_struct_ops_module.c
+++ b/tools/testing/selftests/bpf/prog_tests/test_struct_ops_module.c
@@ -362,6 +362,48 @@ static void test_subsystem_detach(void)
 	struct_ops_detach__destroy(skel);
 }
 
+/* A subsystem detaches a link while the link is going to be free. */
+static void test_subsystem_detach_free(void)
+{
+	LIBBPF_OPTS(bpf_test_run_opts, topts,
+		    .data_in = &pkt_v4,
+		    .data_size_in = sizeof(pkt_v4));
+	struct struct_ops_detach *skel;
+	struct bpf_link *link = NULL;
+	int prog_fd;
+	int err;
+
+	skel = struct_ops_detach__open_and_load();
+	if (!ASSERT_OK_PTR(skel, "struct_ops_detach_open_and_load"))
+		return;
+
+	link = bpf_map__attach_struct_ops(skel->maps.testmod_do_detach);
+	if (!ASSERT_OK_PTR(link, "attach_struct_ops"))
+		goto cleanup;
+
+	bpf_link__destroy(link);
+
+	prog_fd = bpf_program__fd(skel->progs.start_detach);
+	if (!ASSERT_GE(prog_fd, 0, "start_detach_fd"))
+		goto cleanup;
+
+	/* Do detachment from the registered subsystem */
+	err = bpf_prog_test_run_opts(prog_fd, &topts);
+	if (!ASSERT_OK(err, "start_detach_run"))
+		goto cleanup;
+
+	/* The link has zeroed refcount value or even has been
+	 * unregistered, so the detachment from the subsystem should fail.
+	 */
+	ASSERT_EQ(topts.retval, (u32)-ENOENT, "start_detach_run_retval");
+
+	/* Sync RCU to make sure the link is freed without any crash */
+	ASSERT_OK(kern_sync_rcu(), "kern_sync_rcu");
+
+cleanup:
+	struct_ops_detach__destroy(skel);
+}
+
 void serial_test_struct_ops_module(void)
 {
 	if (test__start_subtest("struct_ops_load"))
@@ -378,5 +420,7 @@ void serial_test_struct_ops_module(void)
 		test_detach_link();
 	if (test__start_subtest("test_subsystem_detach"))
 		test_subsystem_detach();
+	if (test__start_subtest("test_subsystem_detach_free"))
+		test_subsystem_detach_free();
 }
 
-- 
2.34.1