| 1 | From: Geliang Tang <tanggeliang@kylinos.cn> | 1 | From: Geliang Tang <tanggeliang@kylinos.cn> |
|---|---|---|---|
| 2 | 2 | ||
| 3 | v4: | 3 | v4: |
| 4 | - CI reports the following BUILD_BUG_ON fails on i386: | 4 | - drop sock_owned_by_user_nocheck and spin_is_locked. According to |
| 5 | 5 | comments from Mat and Martin, in this set mptcp_subflow | |
| 6 | BUILD_BUG_ON(sizeof(struct bpf_iter_mptcp_subflow_kern) != | 6 | bpf_iter only used from a cg sockopt bpf prog, no need to add these |
| 7 | sizeof(struct bpf_iter_mptcp_subflow)) | 7 | check at this moment. |
| 8 | |||
| 9 | Just like in bpf_iter_task_new(), change this "!=" to ">". | ||
| 10 | 8 | ||
| 11 | v3: | 9 | v3: |
| 12 | - check sock_owned_by_user_nocheck(sk)/spin_is_locked(&sk->sk_lock.slock), | 10 | - patch 3, continue to use sock_owned_by_user_nocheck() and spin_is_locked() |
| 13 | instead of lockdep_sock_is_held(sk). | 11 | checks instead of using msk_owned_by_me(). |
| 14 | - add "sizeof" and "alignof" checks. | 12 | - patch 5, drop declaration of bpf_mptcp_subflow_tcp_sock. It's no longer |
| 15 | - drop bpf_mptcp_sk() and bpf_mptcp_subflow_tcp_sock() definitions. Use | 13 | used. |
| 16 | bpf_skc_to_mptcp_sock() and mptcp_subflow_tcp_sock() in mptcp_subflow | 14 | - patch 5, update the comment for mptcp_subflow_tcp_sock(), which is a BPF |
| 17 | bpf_iter selftests instead. | 15 | helper, not a kfunc. |
| 16 | |||
| 17 | The commit log of "bpf: Register mptcp common kfunc set" doesn't match the | ||
| 18 | code, please update it as: | ||
| 19 | |||
| 20 | ''' | ||
| 21 | bpf: Register mptcp common kfunc set | ||
| 22 | |||
| 23 | MPTCP helper mptcp_subflow_ctx() is used to convert struct sock to | ||
| 24 | struct mptcp_subflow_context. It will be used in MPTCP BPF programs. | ||
| 25 | |||
| 26 | This patch defines corresponding wrapper of this helper, and put it | ||
| 27 | into the newly defined mptcp common kfunc set and register this set | ||
| 28 | with the flag BPF_PROG_TYPE_CGROUP_SOCKOPT to let it accessible to | ||
| 29 | the 'cgroup/getsockopt' type of BPF programs. | ||
| 30 | ''' | ||
| 18 | 31 | ||
| 19 | v2: | 32 | v2: |
| 20 | - add CONFIG_LOCKDEP check in patch 2 to fix the build error reported | 33 | - Drop bpf_skc_to_mptcp_sock |
| 21 | by CI. | 34 | - Check the owner before assigning the msk as Mat suggested. |
| 35 | - Use bpf_core_cast() in mptcp_subflow bpf_iter subtest instead of | ||
| 36 | using bpf_skc_to_mptcp_sock(). | ||
| 22 | 37 | ||
| 23 | Address Martin's comments in v1. | 38 | Address Martin's suggestions for "Add mptcp_subflow bpf_iter support" v2. |
| 24 | 39 | ||
| 25 | Geliang Tang (5): | 40 | Geliang Tang (5): |
| 26 | mptcp: update bpf_mptcp_sock_from_subflow | 41 | Revert "bpf: Extend bpf_skc_to_mptcp_sock to MPTCP sock" |
| 27 | bpf: Allow use of skc_to_mptcp_sock in cg_sockopt | 42 | Revert "bpf: Allow use of skc_to_mptcp_sock in cg_sockopt" |
| 28 | Squash to "bpf: Register mptcp common kfunc set" | ||
| 29 | Squash to "bpf: Add mptcp_subflow bpf_iter" | 43 | Squash to "bpf: Add mptcp_subflow bpf_iter" |
| 44 | Revert "bpf: Acquire and release mptcp socket" | ||
| 30 | Squash to "selftests/bpf: Add mptcp_subflow bpf_iter subtest" | 45 | Squash to "selftests/bpf: Add mptcp_subflow bpf_iter subtest" |
| 31 | 46 | ||
| 32 | include/net/mptcp.h | 4 +-- | 47 | include/net/mptcp.h | 4 +- |
| 33 | kernel/bpf/cgroup.c | 2 ++ | 48 | kernel/bpf/cgroup.c | 2 - |
| 34 | net/core/filter.c | 2 +- | 49 | net/core/filter.c | 2 +- |
| 35 | net/mptcp/bpf.c | 34 +++++++++---------- | 50 | net/mptcp/bpf.c | 41 ++++--------------- |
| 36 | tools/testing/selftests/bpf/progs/mptcp_bpf.h | 1 - | 51 | .../testing/selftests/bpf/bpf_experimental.h | 2 +- |
| 37 | .../selftests/bpf/progs/mptcp_bpf_iters.c | 11 +++--- | 52 | tools/testing/selftests/bpf/progs/mptcp_bpf.h | 5 --- |
| 38 | 6 files changed, 27 insertions(+), 27 deletions(-) | 53 | .../selftests/bpf/progs/mptcp_bpf_iters.c | 10 ++--- |
| 54 | 7 files changed, 15 insertions(+), 51 deletions(-) | ||
| 39 | 55 | ||
| 40 | -- | 56 | -- |
| 41 | 2.45.2 | 57 | 2.43.0 | diff view generated by jsdifflib |
| 1 | From: Geliang Tang <tanggeliang@kylinos.cn> | 1 | From: Geliang Tang <tanggeliang@kylinos.cn> |
|---|---|---|---|
| 2 | 2 | ||
| 3 | The input parameter of bpf_mptcp_sock_from_subflow() is an msk subsocket | 3 | bpf_core_cast() can be used in mptcp_subflow bpf_iter selftests to get the |
| 4 | of type IPPROTO_TCP. This patch extends it to accept an IPPROTO_MPTCP | 4 | msk, instead of using bpf_skc_to_mptcp_sock(). No need to add this patch |
| 5 | socket. With this change, the function name bpf_mptcp_sock_from_subflow | 5 | anymore, revert it. |
| 6 | is no longer appropriate, and it is renamed to bpf_mptcp_sock_from_sock. | ||
| 7 | 6 | ||
| 8 | Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn> | 7 | Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn> |
| 9 | --- | 8 | --- |
| 10 | include/net/mptcp.h | 4 ++-- | 9 | include/net/mptcp.h | 4 ++-- |
| 11 | net/core/filter.c | 2 +- | 10 | net/core/filter.c | 2 +- |
| 12 | net/mptcp/bpf.c | 4 +++- | 11 | net/mptcp/bpf.c | 10 ++-------- |
| 13 | 3 files changed, 6 insertions(+), 4 deletions(-) | 12 | 3 files changed, 5 insertions(+), 11 deletions(-) |
| 14 | 13 | ||
| 15 | diff --git a/include/net/mptcp.h b/include/net/mptcp.h | 14 | diff --git a/include/net/mptcp.h b/include/net/mptcp.h |
| 16 | index XXXXXXX..XXXXXXX 100644 | 15 | index XXXXXXX..XXXXXXX 100644 |
| 17 | --- a/include/net/mptcp.h | 16 | --- a/include/net/mptcp.h |
| 18 | +++ b/include/net/mptcp.h | 17 | +++ b/include/net/mptcp.h |
| 19 | @@ -XXX,XX +XXX,XX @@ static inline void mptcpv6_handle_mapped(struct sock *sk, bool mapped) { } | 18 | @@ -XXX,XX +XXX,XX @@ static inline void mptcpv6_handle_mapped(struct sock *sk, bool mapped) { } |
| 20 | #endif | 19 | #endif |
| 21 | 20 | ||
| 22 | #if defined(CONFIG_MPTCP) && defined(CONFIG_BPF_SYSCALL) | 21 | #if defined(CONFIG_MPTCP) && defined(CONFIG_BPF_SYSCALL) |
| 23 | -struct mptcp_sock *bpf_mptcp_sock_from_subflow(struct sock *sk); | 22 | -struct mptcp_sock *bpf_mptcp_sock_from_sock(struct sock *sk); |
| 24 | +struct mptcp_sock *bpf_mptcp_sock_from_sock(struct sock *sk); | 23 | +struct mptcp_sock *bpf_mptcp_sock_from_subflow(struct sock *sk); |
| 25 | #else | 24 | #else |
| 26 | -static inline struct mptcp_sock *bpf_mptcp_sock_from_subflow(struct sock *sk) { return NULL; } | 25 | -static inline struct mptcp_sock *bpf_mptcp_sock_from_sock(struct sock *sk) { return NULL; } |
| 27 | +static inline struct mptcp_sock *bpf_mptcp_sock_from_sock(struct sock *sk) { return NULL; } | 26 | +static inline struct mptcp_sock *bpf_mptcp_sock_from_subflow(struct sock *sk) { return NULL; } |
| 28 | #endif | 27 | #endif |
| 29 | 28 | ||
| 30 | #if !IS_ENABLED(CONFIG_MPTCP) | 29 | #if !IS_ENABLED(CONFIG_MPTCP) |
| 31 | diff --git a/net/core/filter.c b/net/core/filter.c | 30 | diff --git a/net/core/filter.c b/net/core/filter.c |
| 32 | index XXXXXXX..XXXXXXX 100644 | 31 | index XXXXXXX..XXXXXXX 100644 |
| 33 | --- a/net/core/filter.c | 32 | --- a/net/core/filter.c |
| 34 | +++ b/net/core/filter.c | 33 | +++ b/net/core/filter.c |
| 35 | @@ -XXX,XX +XXX,XX @@ const struct bpf_func_proto bpf_skc_to_unix_sock_proto = { | 34 | @@ -XXX,XX +XXX,XX @@ const struct bpf_func_proto bpf_skc_to_unix_sock_proto = { |
| 36 | BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk) | 35 | BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk) |
| 37 | { | 36 | { |
| 38 | BTF_TYPE_EMIT(struct mptcp_sock); | 37 | BTF_TYPE_EMIT(struct mptcp_sock); |
| 39 | - return (unsigned long)bpf_mptcp_sock_from_subflow(sk); | 38 | - return (unsigned long)bpf_mptcp_sock_from_sock(sk); |
| 40 | + return (unsigned long)bpf_mptcp_sock_from_sock(sk); | 39 | + return (unsigned long)bpf_mptcp_sock_from_subflow(sk); |
| 41 | } | 40 | } |
| 42 | 41 | ||
| 43 | const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = { | 42 | const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = { |
| 44 | diff --git a/net/mptcp/bpf.c b/net/mptcp/bpf.c | 43 | diff --git a/net/mptcp/bpf.c b/net/mptcp/bpf.c |
| 45 | index XXXXXXX..XXXXXXX 100644 | 44 | index XXXXXXX..XXXXXXX 100644 |
| 46 | --- a/net/mptcp/bpf.c | 45 | --- a/net/mptcp/bpf.c |
| 47 | +++ b/net/mptcp/bpf.c | 46 | +++ b/net/mptcp/bpf.c |
| 48 | @@ -XXX,XX +XXX,XX @@ static struct bpf_struct_ops bpf_mptcp_sched_ops = { | 47 | @@ -XXX,XX +XXX,XX @@ static struct bpf_struct_ops bpf_mptcp_sched_ops = { |
| 49 | }; | 48 | }; |
| 50 | #endif /* CONFIG_BPF_JIT */ | 49 | #endif /* CONFIG_BPF_JIT */ |
| 51 | 50 | ||
| 52 | -struct mptcp_sock *bpf_mptcp_sock_from_subflow(struct sock *sk) | 51 | -struct mptcp_sock *bpf_mptcp_sock_from_sock(struct sock *sk) |
| 53 | +struct mptcp_sock *bpf_mptcp_sock_from_sock(struct sock *sk) | 52 | +struct mptcp_sock *bpf_mptcp_sock_from_subflow(struct sock *sk) |
| 54 | { | 53 | { |
| 55 | + if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_MPTCP) | 54 | - if (unlikely(!sk || !sk_fullsock(sk))) |
| 56 | + return mptcp_sk(sk); | 55 | - return NULL; |
| 57 | if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP && sk_is_mptcp(sk)) | 56 | - |
| 57 | - if (sk->sk_protocol == IPPROTO_MPTCP) | ||
| 58 | - return mptcp_sk(sk); | ||
| 59 | - | ||
| 60 | - if (sk->sk_protocol == IPPROTO_TCP && sk_is_mptcp(sk)) | ||
| 61 | + if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP && sk_is_mptcp(sk)) | ||
| 58 | return mptcp_sk(mptcp_subflow_ctx(sk)->conn); | 62 | return mptcp_sk(mptcp_subflow_ctx(sk)->conn); |
| 59 | 63 | ||
| 64 | return NULL; | ||
| 60 | -- | 65 | -- |
| 61 | 2.45.2 | 66 | 2.43.0 | diff view generated by jsdifflib |
| 1 | From: Geliang Tang <tanggeliang@kylinos.cn> | 1 | From: Geliang Tang <tanggeliang@kylinos.cn> |
|---|---|---|---|
| 2 | 2 | ||
| 3 | Currently, bpf_skc_to_mptcp_sock() helper is not allowed to be used | 3 | bpf_core_cast() can be used in mptcp_subflow bpf_iter selftests to get the |
| 4 | in cg_sockopt. This patch adds this permission. | 4 | msk, instead of using bpf_skc_to_mptcp_sock(). No need to add this patch |
| 5 | anymore, revert it. | ||
| 5 | 6 | ||
| 6 | Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn> | 7 | This reverts commit d8d42f5a4542c2dc31e4e00c58508a821c6ab789. |
| 7 | --- | 8 | --- |
| 8 | kernel/bpf/cgroup.c | 2 ++ | 9 | kernel/bpf/cgroup.c | 2 -- |
| 9 | 1 file changed, 2 insertions(+) | 10 | 1 file changed, 2 deletions(-) |
| 10 | 11 | ||
| 11 | diff --git a/kernel/bpf/cgroup.c b/kernel/bpf/cgroup.c | 12 | diff --git a/kernel/bpf/cgroup.c b/kernel/bpf/cgroup.c |
| 12 | index XXXXXXX..XXXXXXX 100644 | 13 | index XXXXXXX..XXXXXXX 100644 |
| 13 | --- a/kernel/bpf/cgroup.c | 14 | --- a/kernel/bpf/cgroup.c |
| 14 | +++ b/kernel/bpf/cgroup.c | 15 | +++ b/kernel/bpf/cgroup.c |
| 15 | @@ -XXX,XX +XXX,XX @@ cg_sockopt_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) | 16 | @@ -XXX,XX +XXX,XX @@ cg_sockopt_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) |
| 16 | #ifdef CONFIG_INET | 17 | #ifdef CONFIG_INET |
| 17 | case BPF_FUNC_tcp_sock: | 18 | case BPF_FUNC_tcp_sock: |
| 18 | return &bpf_tcp_sock_proto; | 19 | return &bpf_tcp_sock_proto; |
| 19 | + case BPF_FUNC_skc_to_mptcp_sock: | 20 | - case BPF_FUNC_skc_to_mptcp_sock: |
| 20 | + return &bpf_skc_to_mptcp_sock_proto; | 21 | - return &bpf_skc_to_mptcp_sock_proto; |
| 21 | #endif | 22 | #endif |
| 22 | case BPF_FUNC_perf_event_output: | 23 | case BPF_FUNC_perf_event_output: |
| 23 | return &bpf_event_output_data_proto; | 24 | return &bpf_event_output_data_proto; |
| 24 | -- | 25 | -- |
| 25 | 2.45.2 | 26 | 2.43.0 | diff view generated by jsdifflib |
| 1 | From: Geliang Tang <tanggeliang@kylinos.cn> | 1 | From: Geliang Tang <tanggeliang@kylinos.cn> |
|---|---|---|---|
| 2 | 2 | ||
| 3 | Add "sizeof" and "alignof" checks. | 3 | Drop the NULL check for 'msk' as Martin suggested, add more checks |
| 4 | for 'sk'. | ||
| 4 | 5 | ||
| 5 | Address Martin's comments in v1: | 6 | Use the "struct sock *sk" instead of "struct mptcp-sock *msk" as the |
| 7 | argument in the bpf_iter_mptcp_subflow_new as Martin suggested. | ||
| 6 | 8 | ||
| 7 | - bpf_iter_mptcp_subflow_new returns -EINVAL when msk socket lock isn't | 9 | v4: |
| 8 | held. | 10 | - drop sock_owned_by_user_nocheck and spin_is_locked. According to |
| 11 | comments from Mat [2] and Martin [1], in this set mptcp_subflow | ||
| 12 | bpf_iter only used from a cg sockopt bpf prog, no need to add these | ||
| 13 | check at this moment. | ||
| 14 | |||
| 15 | [1] | ||
| 16 | https://lore.kernel.org/all/fdf0ddbe-e007-4a5f-bbdf-9a144e8fbe35@linux.dev/ | ||
| 17 | [2] | ||
| 18 | https://patchwork.kernel.org/project/mptcp/patch/f6469225598beecbf0bda12a4c33fafa86c0ff15.1739787744.git.tanggeliang@kylinos.cn/ | ||
| 19 | |||
| 20 | v3: | ||
| 21 | - continue to use sock_owned_by_user_nocheck and spin_is_locked | ||
| 22 | checks instead of using msk_owned_by_me(). | ||
| 23 | |||
| 24 | v2: | ||
| 25 | - check the owner before assigning the msk as Mat suggested. | ||
| 9 | 26 | ||
| 10 | Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn> | 27 | Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn> |
| 11 | --- | 28 | --- |
| 12 | net/mptcp/bpf.c | 10 +++++++++- | 29 | net/mptcp/bpf.c | 12 ++++++------ |
| 13 | 1 file changed, 9 insertions(+), 1 deletion(-) | 30 | 1 file changed, 6 insertions(+), 6 deletions(-) |
| 14 | 31 | ||
| 15 | diff --git a/net/mptcp/bpf.c b/net/mptcp/bpf.c | 32 | diff --git a/net/mptcp/bpf.c b/net/mptcp/bpf.c |
| 16 | index XXXXXXX..XXXXXXX 100644 | 33 | index XXXXXXX..XXXXXXX 100644 |
| 17 | --- a/net/mptcp/bpf.c | 34 | --- a/net/mptcp/bpf.c |
| 18 | +++ b/net/mptcp/bpf.c | 35 | +++ b/net/mptcp/bpf.c |
| 19 | @@ -XXX,XX +XXX,XX @@ bpf_iter_mptcp_subflow_new(struct bpf_iter_mptcp_subflow *it, | 36 | @@ -XXX,XX +XXX,XX @@ bpf_mptcp_subflow_ctx(const struct sock *sk) |
| 20 | struct mptcp_sock *msk) | 37 | |
| 38 | __bpf_kfunc static int | ||
| 39 | bpf_iter_mptcp_subflow_new(struct bpf_iter_mptcp_subflow *it, | ||
| 40 | - struct mptcp_sock *msk) | ||
| 41 | + struct sock *sk) | ||
| 21 | { | 42 | { |
| 22 | struct bpf_iter_mptcp_subflow_kern *kit = (void *)it; | 43 | struct bpf_iter_mptcp_subflow_kern *kit = (void *)it; |
| 23 | + struct sock *sk = (struct sock *)msk; | 44 | - struct sock *sk = (struct sock *)msk; |
| 24 | + | 45 | + struct mptcp_sock *msk; |
| 25 | + BUILD_BUG_ON(sizeof(struct bpf_iter_mptcp_subflow_kern) > | 46 | |
| 26 | + sizeof(struct bpf_iter_mptcp_subflow)); | 47 | BUILD_BUG_ON(sizeof(struct bpf_iter_mptcp_subflow_kern) > |
| 27 | + BUILD_BUG_ON(__alignof__(struct bpf_iter_mptcp_subflow_kern) != | 48 | sizeof(struct bpf_iter_mptcp_subflow)); |
| 28 | + __alignof__(struct bpf_iter_mptcp_subflow)); | 49 | BUILD_BUG_ON(__alignof__(struct bpf_iter_mptcp_subflow_kern) != |
| 29 | 50 | __alignof__(struct bpf_iter_mptcp_subflow)); | |
| 30 | kit->msk = msk; | 51 | |
| 31 | if (!msk) | 52 | - kit->msk = msk; |
| 53 | - if (!msk) | ||
| 54 | + if (unlikely(!sk || !sk_fullsock(sk))) | ||
| 32 | return -EINVAL; | 55 | return -EINVAL; |
| 33 | 56 | ||
| 34 | - msk_owned_by_me(msk); | 57 | - if (!sock_owned_by_user_nocheck(sk) && |
| 35 | + if (!sock_owned_by_user_nocheck(sk) && | 58 | - !spin_is_locked(&sk->sk_lock.slock)) |
| 36 | + !spin_is_locked(&sk->sk_lock.slock)) | 59 | + if (sk->sk_protocol != IPPROTO_MPTCP) |
| 37 | + return -EINVAL; | 60 | return -EINVAL; |
| 38 | 61 | ||
| 62 | + msk = mptcp_sk(sk); | ||
| 63 | + kit->msk = msk; | ||
| 39 | kit->pos = &msk->conn_list; | 64 | kit->pos = &msk->conn_list; |
| 40 | return 0; | 65 | return 0; |
| 66 | } | ||
| 41 | -- | 67 | -- |
| 42 | 2.45.2 | 68 | 2.43.0 | diff view generated by jsdifflib |
| 1 | From: Geliang Tang <tanggeliang@kylinos.cn> | 1 | From: Geliang Tang <tanggeliang@kylinos.cn> |
|---|---|---|---|
| 2 | 2 | ||
| 3 | Drop bpf_mptcp_sk() and bpf_mptcp_subflow_tcp_sock() definitions. Use | 3 | Drop this patch as Martin suggested. |
| 4 | bpf_skc_to_mptcp_sock() and mptcp_subflow_tcp_sock() in mptcp_subflow | ||
| 5 | bpf_iter selftests instead. | ||
| 6 | 4 | ||
| 7 | Address Martin's comments in v1: | 5 | From Martin's review [1], this mptcp_sock_acquire() helper was a workaround |
| 6 | only to please the verifier, but they were not needed. | ||
| 8 | 7 | ||
| 9 | - add null-check for bpf_mptcp_subflow_ctx. | 8 | [1] |
| 10 | - add KF_RET_NULL flags for bpf_mptcp_subflow_ctx. | 9 | https://lore.kernel.org/9b373a23-c093-42d8-b4ae-99f2e62e7681@linux.dev |
| 11 | - register this kfunc set to BPF_PROG_TYPE_CGROUP_SOCKOPT only, | ||
| 12 | not BPF_PROG_TYPE_UNSPEC. | ||
| 13 | 10 | ||
| 14 | Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn> | 11 | Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn> |
| 15 | --- | 12 | --- |
| 16 | net/mptcp/bpf.c | 20 +++++--------------- | 13 | net/mptcp/bpf.c | 19 ------------------- |
| 17 | 1 file changed, 5 insertions(+), 15 deletions(-) | 14 | 1 file changed, 19 deletions(-) |
| 18 | 15 | ||
| 19 | diff --git a/net/mptcp/bpf.c b/net/mptcp/bpf.c | 16 | diff --git a/net/mptcp/bpf.c b/net/mptcp/bpf.c |
| 20 | index XXXXXXX..XXXXXXX 100644 | 17 | index XXXXXXX..XXXXXXX 100644 |
| 21 | --- a/net/mptcp/bpf.c | 18 | --- a/net/mptcp/bpf.c |
| 22 | +++ b/net/mptcp/bpf.c | 19 | +++ b/net/mptcp/bpf.c |
| 23 | @@ -XXX,XX +XXX,XX @@ struct bpf_iter_mptcp_subflow_kern { | 20 | @@ -XXX,XX +XXX,XX @@ bpf_iter_mptcp_subflow_destroy(struct bpf_iter_mptcp_subflow *it) |
| 24 | 21 | { | |
| 25 | __bpf_kfunc_start_defs(); | 22 | } |
| 26 | 23 | ||
| 27 | -__bpf_kfunc static struct mptcp_sock *bpf_mptcp_sk(struct sock *sk) | 24 | -__bpf_kfunc static struct |
| 25 | -mptcp_sock *bpf_mptcp_sock_acquire(struct mptcp_sock *msk) | ||
| 28 | -{ | 26 | -{ |
| 29 | - return mptcp_sk(sk); | 27 | - struct sock *sk = (struct sock *)msk; |
| 28 | - | ||
| 29 | - if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) | ||
| 30 | - return msk; | ||
| 31 | - return NULL; | ||
| 30 | -} | 32 | -} |
| 31 | - | 33 | - |
| 32 | __bpf_kfunc static struct mptcp_subflow_context * | 34 | -__bpf_kfunc static void bpf_mptcp_sock_release(struct mptcp_sock *msk) |
| 33 | bpf_mptcp_subflow_ctx(const struct sock *sk) | 35 | -{ |
| 36 | - struct sock *sk = (struct sock *)msk; | ||
| 37 | - | ||
| 38 | - WARN_ON_ONCE(!sk || !refcount_dec_not_one(&sk->sk_refcnt)); | ||
| 39 | -} | ||
| 40 | - | ||
| 41 | __bpf_kfunc struct mptcp_subflow_context * | ||
| 42 | bpf_mptcp_subflow_ctx_by_pos(const struct mptcp_sched_data *data, unsigned int pos) | ||
| 34 | { | 43 | { |
| 35 | - return mptcp_subflow_ctx(sk); | 44 | @@ -XXX,XX +XXX,XX @@ BTF_ID_FLAGS(func, bpf_mptcp_subflow_ctx, KF_RET_NULL) |
| 36 | -} | ||
| 37 | + if (!sk) | ||
| 38 | + return NULL; | ||
| 39 | |||
| 40 | -__bpf_kfunc static struct sock * | ||
| 41 | -bpf_mptcp_subflow_tcp_sock(const struct mptcp_subflow_context *subflow) | ||
| 42 | -{ | ||
| 43 | - return mptcp_subflow_tcp_sock(subflow); | ||
| 44 | + return mptcp_subflow_ctx(sk); | ||
| 45 | } | ||
| 46 | |||
| 47 | __bpf_kfunc static int | ||
| 48 | @@ -XXX,XX +XXX,XX @@ __bpf_kfunc static bool bpf_mptcp_subflow_queues_empty(struct sock *sk) | ||
| 49 | __bpf_kfunc_end_defs(); | ||
| 50 | |||
| 51 | BTF_KFUNCS_START(bpf_mptcp_common_kfunc_ids) | ||
| 52 | -BTF_ID_FLAGS(func, bpf_mptcp_sk) | ||
| 53 | -BTF_ID_FLAGS(func, bpf_mptcp_subflow_ctx) | ||
| 54 | -BTF_ID_FLAGS(func, bpf_mptcp_subflow_tcp_sock) | ||
| 55 | +BTF_ID_FLAGS(func, bpf_mptcp_subflow_ctx, KF_RET_NULL) | ||
| 56 | BTF_ID_FLAGS(func, bpf_iter_mptcp_subflow_new, KF_ITER_NEW | KF_TRUSTED_ARGS) | 45 | BTF_ID_FLAGS(func, bpf_iter_mptcp_subflow_new, KF_ITER_NEW | KF_TRUSTED_ARGS) |
| 57 | BTF_ID_FLAGS(func, bpf_iter_mptcp_subflow_next, KF_ITER_NEXT | KF_RET_NULL) | 46 | BTF_ID_FLAGS(func, bpf_iter_mptcp_subflow_next, KF_ITER_NEXT | KF_RET_NULL) |
| 58 | BTF_ID_FLAGS(func, bpf_iter_mptcp_subflow_destroy, KF_ITER_DESTROY) | 47 | BTF_ID_FLAGS(func, bpf_iter_mptcp_subflow_destroy, KF_ITER_DESTROY) |
| 59 | @@ -XXX,XX +XXX,XX @@ static int __init bpf_mptcp_kfunc_init(void) | 48 | -BTF_ID_FLAGS(func, bpf_mptcp_sock_acquire, KF_ACQUIRE | KF_RET_NULL) |
| 60 | int ret; | 49 | -BTF_ID_FLAGS(func, bpf_mptcp_sock_release, KF_RELEASE) |
| 61 | 50 | BTF_KFUNCS_END(bpf_mptcp_common_kfunc_ids) | |
| 62 | ret = register_btf_fmodret_id_set(&bpf_mptcp_fmodret_set); | 51 | |
| 63 | - ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_UNSPEC, | 52 | static const struct btf_kfunc_id_set bpf_mptcp_common_kfunc_set = { |
| 64 | + ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SOCKOPT, | ||
| 65 | &bpf_mptcp_common_kfunc_set); | ||
| 66 | ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_STRUCT_OPS, | ||
| 67 | &bpf_mptcp_sched_kfunc_set); | ||
| 68 | -- | 53 | -- |
| 69 | 2.45.2 | 54 | 2.43.0 | diff view generated by jsdifflib |
| 1 | From: Geliang Tang <tanggeliang@kylinos.cn> | 1 | From: Geliang Tang <tanggeliang@kylinos.cn> |
|---|---|---|---|
| 2 | 2 | ||
| 3 | Use bpf_skc_to_mptcp_sock() and mptcp_subflow_tcp_sock() instead of | 3 | Use bpf_core_cast() instead of bpf_skc_to_mptcp_sock(). |
| 4 | bpf_mptcp_sk() and bpf_mptcp_subflow_tcp_sock(). | 4 | Change the 2nd parameter type of bpf_for_each() as 'struct sock'. |
| 5 | Drop use of bpf_mptcp_sock_acquire/release. | ||
| 6 | Drop declaration of bpf_mptcp_subflow_tcp_sock. It's no longer used. | ||
| 7 | Update the comment for mptcp_subflow_tcp_sock(), which is a BPF helper, | ||
| 8 | not a kfunc. | ||
| 5 | 9 | ||
| 6 | IPPROTO_MPTCP is checked in bpf_skc_to_mptcp_sock(), no need to check | 10 | Please update the commit log as: |
| 7 | it in BPF program. | ||
| 8 | 11 | ||
| 9 | bpf_skc_to_mptcp_sock() and bpf_mptcp_subflow_ctx() may return NULL, | 12 | ''' |
| 10 | need to check the return values. | 13 | This patch adds a "cgroup/getsockopt" program "iters_subflow" to test the |
| 14 | newly added mptcp_subflow bpf_iter. | ||
| 15 | |||
| 16 | Export mptcp_subflow helpers bpf_iter_mptcp_subflow_new/_next/_destroy | ||
| 17 | and other helpers into bpf_experimental.h. | ||
| 18 | |||
| 19 | Use bpf_for_each() to walk the subflow list of this msk. MPTCP-specific | ||
| 20 | packet scheduler kfunc can be called in the loop. In this test, just | ||
| 21 | add all subflow ids to local variable local_ids, then invoke the helper | ||
| 22 | mptcp_subflow_tcp_sock() in the loop to pick a subsocket. | ||
| 23 | |||
| 24 | Out of the loop, use bpf_mptcp_subflow_ctx() to get the subflow context | ||
| 25 | of the picked subsocket and do some verification. Finally, assign | ||
| 26 | local_ids to global variable ids so that the application can obtain this | ||
| 27 | value. | ||
| 28 | |||
| 29 | Add a subtest named test_iters_subflow to load and verify the newly added | ||
| 30 | mptcp_subflow type bpf_iter example in test_mptcp. Use the helper | ||
| 31 | endpoint_init() to add 3 new subflow endpoints. Send a byte of message | ||
| 32 | to start the mptcp connection, and wait for new subflows to be added. | ||
| 33 | getsockopt() is invoked to trigger the "cgroup/getsockopt" test program | ||
| 34 | "iters_subflow". Check if skel->bss->ids equals 10 to verify whether this | ||
| 35 | mptcp_subflow bpf_iter loops correctly as expected. | ||
| 36 | ''' | ||
| 11 | 37 | ||
| 12 | Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn> | 38 | Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn> |
| 13 | --- | 39 | --- |
| 14 | tools/testing/selftests/bpf/progs/mptcp_bpf.h | 1 - | 40 | tools/testing/selftests/bpf/bpf_experimental.h | 2 +- |
| 15 | tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c | 11 +++++------ | 41 | tools/testing/selftests/bpf/progs/mptcp_bpf.h | 5 ----- |
| 16 | 2 files changed, 5 insertions(+), 7 deletions(-) | 42 | tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c | 10 +++------- |
| 43 | 3 files changed, 4 insertions(+), 13 deletions(-) | ||
| 17 | 44 | ||
| 45 | diff --git a/tools/testing/selftests/bpf/bpf_experimental.h b/tools/testing/selftests/bpf/bpf_experimental.h | ||
| 46 | index XXXXXXX..XXXXXXX 100644 | ||
| 47 | --- a/tools/testing/selftests/bpf/bpf_experimental.h | ||
| 48 | +++ b/tools/testing/selftests/bpf/bpf_experimental.h | ||
| 49 | @@ -XXX,XX +XXX,XX @@ extern void bpf_iter_css_destroy(struct bpf_iter_css *it) __weak __ksym; | ||
| 50 | |||
| 51 | struct bpf_iter_mptcp_subflow; | ||
| 52 | extern int bpf_iter_mptcp_subflow_new(struct bpf_iter_mptcp_subflow *it, | ||
| 53 | - struct mptcp_sock *msk) __weak __ksym; | ||
| 54 | + struct sock *sk) __weak __ksym; | ||
| 55 | extern struct mptcp_subflow_context * | ||
| 56 | bpf_iter_mptcp_subflow_next(struct bpf_iter_mptcp_subflow *it) __weak __ksym; | ||
| 57 | extern void | ||
| 18 | diff --git a/tools/testing/selftests/bpf/progs/mptcp_bpf.h b/tools/testing/selftests/bpf/progs/mptcp_bpf.h | 58 | diff --git a/tools/testing/selftests/bpf/progs/mptcp_bpf.h b/tools/testing/selftests/bpf/progs/mptcp_bpf.h |
| 19 | index XXXXXXX..XXXXXXX 100644 | 59 | index XXXXXXX..XXXXXXX 100644 |
| 20 | --- a/tools/testing/selftests/bpf/progs/mptcp_bpf.h | 60 | --- a/tools/testing/selftests/bpf/progs/mptcp_bpf.h |
| 21 | +++ b/tools/testing/selftests/bpf/progs/mptcp_bpf.h | 61 | +++ b/tools/testing/selftests/bpf/progs/mptcp_bpf.h |
| 22 | @@ -XXX,XX +XXX,XX @@ mptcp_subflow_tcp_sock(const struct mptcp_subflow_context *subflow) | 62 | @@ -XXX,XX +XXX,XX @@ mptcp_subflow_tcp_sock(const struct mptcp_subflow_context *subflow) |
| 23 | extern struct mptcp_sock *bpf_mptcp_sock_acquire(struct mptcp_sock *msk) __ksym; | 63 | } |
| 24 | extern void bpf_mptcp_sock_release(struct mptcp_sock *msk) __ksym; | 64 | |
| 25 | 65 | /* ksym */ | |
| 26 | -extern struct mptcp_sock *bpf_mptcp_sk(struct sock *sk) __ksym; | 66 | -extern struct mptcp_sock *bpf_mptcp_sock_acquire(struct mptcp_sock *msk) __ksym; |
| 67 | -extern void bpf_mptcp_sock_release(struct mptcp_sock *msk) __ksym; | ||
| 68 | - | ||
| 27 | extern struct mptcp_subflow_context * | 69 | extern struct mptcp_subflow_context * |
| 28 | bpf_mptcp_subflow_ctx(const struct sock *sk) __ksym; | 70 | bpf_mptcp_subflow_ctx(const struct sock *sk) __ksym; |
| 29 | extern struct sock * | 71 | -extern struct sock * |
| 72 | -bpf_mptcp_subflow_tcp_sock(const struct mptcp_subflow_context *subflow) __ksym; | ||
| 73 | |||
| 74 | extern void mptcp_subflow_set_scheduled(struct mptcp_subflow_context *subflow, | ||
| 75 | bool scheduled) __ksym; | ||
| 30 | diff --git a/tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c b/tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c | 76 | diff --git a/tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c b/tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c |
| 31 | index XXXXXXX..XXXXXXX 100644 | 77 | index XXXXXXX..XXXXXXX 100644 |
| 32 | --- a/tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c | 78 | --- a/tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c |
| 33 | +++ b/tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c | 79 | +++ b/tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c |
| 34 | @@ -XXX,XX +XXX,XX @@ int iters_subflow(struct bpf_sockopt *ctx) | 80 | @@ -XXX,XX +XXX,XX @@ int iters_subflow(struct bpf_sockopt *ctx) |
| 35 | struct mptcp_sock *msk; | 81 | if (ctx->level != SOL_TCP || ctx->optname != TCP_IS_MPTCP) |
| 36 | int local_ids = 0; | ||
| 37 | |||
| 38 | - if (!sk || sk->protocol != IPPROTO_MPTCP || | ||
| 39 | - ctx->level != SOL_TCP || ctx->optname != TCP_IS_MPTCP) | ||
| 40 | + if (ctx->level != SOL_TCP || ctx->optname != TCP_IS_MPTCP) | ||
| 41 | return 1; | 82 | return 1; |
| 42 | 83 | ||
| 43 | - msk = bpf_mptcp_sk((struct sock *)sk); | 84 | - msk = bpf_skc_to_mptcp_sock(sk); |
| 44 | - if (msk->pm.server_side || !msk->pm.subflows) | 85 | + msk = bpf_core_cast(sk, struct mptcp_sock); |
| 45 | + msk = bpf_skc_to_mptcp_sock(sk); | 86 | if (!msk || msk->pm.server_side || !msk->pm.subflows) |
| 46 | + if (!msk || msk->pm.server_side || !msk->pm.subflows) | ||
| 47 | return 1; | 87 | return 1; |
| 48 | 88 | ||
| 49 | msk = bpf_mptcp_sock_acquire(msk); | 89 | - msk = bpf_mptcp_sock_acquire(msk); |
| 90 | - if (!msk) | ||
| 91 | - return 1; | ||
| 92 | - bpf_for_each(mptcp_subflow, subflow, msk) { | ||
| 93 | + bpf_for_each(mptcp_subflow, subflow, (struct sock *)sk) { | ||
| 94 | /* Here MPTCP-specific packet scheduler kfunc can be called: | ||
| 95 | * this test is not doing anything really useful, only to | ||
| 96 | * verify the iteration works. | ||
| 50 | @@ -XXX,XX +XXX,XX @@ int iters_subflow(struct bpf_sockopt *ctx) | 97 | @@ -XXX,XX +XXX,XX @@ int iters_subflow(struct bpf_sockopt *ctx) |
| 98 | |||
| 51 | local_ids += subflow->subflow_id; | 99 | local_ids += subflow->subflow_id; |
| 52 | 100 | ||
| 53 | /* only to check the following kfunc works */ | 101 | - /* only to check the following kfunc works */ |
| 54 | - ssk = bpf_mptcp_subflow_tcp_sock(subflow); | 102 | + /* only to check the following helper works */ |
| 55 | + ssk = mptcp_subflow_tcp_sock(subflow); | 103 | ssk = mptcp_subflow_tcp_sock(subflow); |
| 56 | } | 104 | } |
| 57 | 105 | ||
| 58 | if (!ssk) | ||
| 59 | @@ -XXX,XX +XXX,XX @@ int iters_subflow(struct bpf_sockopt *ctx) | 106 | @@ -XXX,XX +XXX,XX @@ int iters_subflow(struct bpf_sockopt *ctx) |
| 60 | |||
| 61 | /* only to check the following kfunc works */ | ||
| 62 | subflow = bpf_mptcp_subflow_ctx(ssk); | ||
| 63 | - if (subflow->token != msk->token) | ||
| 64 | + if (!subflow || subflow->token != msk->token) | ||
| 65 | goto out; | ||
| 66 | |||
| 67 | ids = local_ids; | 107 | ids = local_ids; |
| 108 | |||
| 109 | out: | ||
| 110 | - bpf_mptcp_sock_release(msk); | ||
| 111 | return 1; | ||
| 112 | } | ||
| 68 | -- | 113 | -- |
| 69 | 2.45.2 | 114 | 2.43.0 | diff view generated by jsdifflib |