| 1 | From: Geliang Tang <tanggeliang@kylinos.cn> | 1 | From: Geliang Tang <tanggeliang@kylinos.cn> |
|---|---|---|---|
| 2 | 2 | ||
| 3 | v5: | ||
| 4 | - update subjects and commit logs as Matt suggested (thanks). | ||
| 5 | - avoid repeating checks in patch 1. | ||
| 6 | - do stricter checks on subflows in patch 3. | ||
| 7 | |||
| 3 | v4: | 8 | v4: |
| 4 | - drop sock_owned_by_user_nocheck and spin_is_locked. According to | 9 | - CI reports the following BUILD_BUG_ON fails on i386: |
| 5 | comments from Mat and Martin, in this set mptcp_subflow | 10 | |
| 6 | bpf_iter only used from a cg sockopt bpf prog, no need to add these | 11 | BUILD_BUG_ON(sizeof(struct bpf_iter_mptcp_subflow_kern) != |
| 7 | check at this moment. | 12 | sizeof(struct bpf_iter_mptcp_subflow)) |
| 13 | |||
| 14 | Just like in bpf_iter_task_new(), change this "!=" to ">". | ||
| 8 | 15 | ||
| 9 | v3: | 16 | v3: |
| 10 | - patch 3, continue to use sock_owned_by_user_nocheck() and spin_is_locked() | 17 | - check sock_owned_by_user_nocheck(sk)/spin_is_locked(&sk->sk_lock.slock), |
| 11 | checks instead of using msk_owned_by_me(). | 18 | instead of lockdep_sock_is_held(sk). |
| 12 | - patch 5, drop declaration of bpf_mptcp_subflow_tcp_sock. It's no longer | 19 | - add "sizeof" and "alignof" checks. |
| 13 | used. | 20 | - drop bpf_mptcp_sk() and bpf_mptcp_subflow_tcp_sock() definitions. Use |
| 14 | - patch 5, update the comment for mptcp_subflow_tcp_sock(), which is a BPF | 21 | bpf_skc_to_mptcp_sock() and mptcp_subflow_tcp_sock() in mptcp_subflow |
| 15 | helper, not a kfunc. | 22 | bpf_iter selftests instead. |
| 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 | ''' | ||
| 31 | 23 | ||
| 32 | v2: | 24 | v2: |
| 33 | - Drop bpf_skc_to_mptcp_sock | 25 | - add CONFIG_LOCKDEP check in patch 2 to fix the build error reported |
| 34 | - Check the owner before assigning the msk as Mat suggested. | 26 | by CI. |
| 35 | - Use bpf_core_cast() in mptcp_subflow bpf_iter subtest instead of | ||
| 36 | using bpf_skc_to_mptcp_sock(). | ||
| 37 | 27 | ||
| 38 | Address Martin's suggestions for "Add mptcp_subflow bpf_iter support" v2. | 28 | Address Martin's comments in v1. |
| 39 | 29 | ||
| 40 | Geliang Tang (5): | 30 | Geliang Tang (5): |
| 41 | Revert "bpf: Extend bpf_skc_to_mptcp_sock to MPTCP sock" | 31 | bpf: Extend bpf_skc_to_mptcp_sock to MPTCP sock |
| 42 | Revert "bpf: Allow use of skc_to_mptcp_sock in cg_sockopt" | 32 | bpf: Allow use of skc_to_mptcp_sock in cg_sockopt |
| 33 | Squash to "bpf: Register mptcp common kfunc set" | ||
| 43 | Squash to "bpf: Add mptcp_subflow bpf_iter" | 34 | Squash to "bpf: Add mptcp_subflow bpf_iter" |
| 44 | Revert "bpf: Acquire and release mptcp socket" | ||
| 45 | Squash to "selftests/bpf: Add mptcp_subflow bpf_iter subtest" | 35 | Squash to "selftests/bpf: Add mptcp_subflow bpf_iter subtest" |
| 46 | 36 | ||
| 47 | include/net/mptcp.h | 4 +- | 37 | include/net/mptcp.h | 4 +- |
| 48 | kernel/bpf/cgroup.c | 2 - | 38 | kernel/bpf/cgroup.c | 2 + |
| 49 | net/core/filter.c | 2 +- | 39 | net/core/filter.c | 2 +- |
| 50 | net/mptcp/bpf.c | 41 ++++--------------- | 40 | net/mptcp/bpf.c | 41 +++++++++++-------- |
| 51 | .../testing/selftests/bpf/bpf_experimental.h | 2 +- | 41 | tools/testing/selftests/bpf/progs/mptcp_bpf.h | 1 - |
| 52 | tools/testing/selftests/bpf/progs/mptcp_bpf.h | 5 --- | 42 | .../selftests/bpf/progs/mptcp_bpf_iters.c | 11 +++-- |
| 53 | .../selftests/bpf/progs/mptcp_bpf_iters.c | 10 ++--- | 43 | 6 files changed, 33 insertions(+), 28 deletions(-) |
| 54 | 7 files changed, 15 insertions(+), 51 deletions(-) | ||
| 55 | 44 | ||
| 56 | -- | 45 | -- |
| 57 | 2.43.0 | 46 | 2.45.2 | diff view generated by jsdifflib |
| 1 | From: Geliang Tang <tanggeliang@kylinos.cn> | 1 | From: Geliang Tang <tanggeliang@kylinos.cn> |
|---|---|---|---|
| 2 | 2 | ||
| 3 | bpf_core_cast() can be used in mptcp_subflow bpf_iter selftests to get the | 3 | Currently, bpf_skc_to_mptcp_sock() can only be used with sockets that |
| 4 | msk, instead of using bpf_skc_to_mptcp_sock(). No need to add this patch | 4 | are MPTCP subflows: TCP sockets with tp->is_mptcp, created by the kernel |
| 5 | anymore, revert it. | 5 | from an MPTCP socket (IPPROTO_MPTCP). Typically used with BPF sock_ops |
| 6 | operators. | ||
| 7 | |||
| 8 | Here, this helper is extended to support MPTCP sockets, the ones created | ||
| 9 | by the userspace (IPPROTO_MPTCP). This is useful for BPF hooks involving | ||
| 10 | these sockets, e.g. [gs]etsocktopt. | ||
| 11 | |||
| 12 | bpf_skc_to_mptcp_sock() uses bpf_mptcp_sock_from_subflow(). The former | ||
| 13 | suggests any MPTCP type/subtype can be used, but the latter only accepts | ||
| 14 | subflow ones. So bpf_mptcp_sock_from_subflow is modified here to support | ||
| 15 | MPTCP socket, and renamed to avoid confusions. | ||
| 6 | 16 | ||
| 7 | Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn> | 17 | Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn> |
| 8 | --- | 18 | --- |
| 9 | include/net/mptcp.h | 4 ++-- | 19 | include/net/mptcp.h | 4 ++-- |
| 10 | net/core/filter.c | 2 +- | 20 | net/core/filter.c | 2 +- |
| 11 | net/mptcp/bpf.c | 10 ++-------- | 21 | net/mptcp/bpf.c | 10 ++++++++-- |
| 12 | 3 files changed, 5 insertions(+), 11 deletions(-) | 22 | 3 files changed, 11 insertions(+), 5 deletions(-) |
| 13 | 23 | ||
| 14 | diff --git a/include/net/mptcp.h b/include/net/mptcp.h | 24 | diff --git a/include/net/mptcp.h b/include/net/mptcp.h |
| 15 | index XXXXXXX..XXXXXXX 100644 | 25 | index XXXXXXX..XXXXXXX 100644 |
| 16 | --- a/include/net/mptcp.h | 26 | --- a/include/net/mptcp.h |
| 17 | +++ b/include/net/mptcp.h | 27 | +++ b/include/net/mptcp.h |
| 18 | @@ -XXX,XX +XXX,XX @@ static inline void mptcpv6_handle_mapped(struct sock *sk, bool mapped) { } | 28 | @@ -XXX,XX +XXX,XX @@ static inline void mptcpv6_handle_mapped(struct sock *sk, bool mapped) { } |
| 19 | #endif | 29 | #endif |
| 20 | 30 | ||
| 21 | #if defined(CONFIG_MPTCP) && defined(CONFIG_BPF_SYSCALL) | 31 | #if defined(CONFIG_MPTCP) && defined(CONFIG_BPF_SYSCALL) |
| 22 | -struct mptcp_sock *bpf_mptcp_sock_from_sock(struct sock *sk); | 32 | -struct mptcp_sock *bpf_mptcp_sock_from_subflow(struct sock *sk); |
| 23 | +struct mptcp_sock *bpf_mptcp_sock_from_subflow(struct sock *sk); | 33 | +struct mptcp_sock *bpf_mptcp_sock_from_sock(struct sock *sk); |
| 24 | #else | 34 | #else |
| 25 | -static inline struct mptcp_sock *bpf_mptcp_sock_from_sock(struct sock *sk) { return NULL; } | 35 | -static inline struct mptcp_sock *bpf_mptcp_sock_from_subflow(struct sock *sk) { return NULL; } |
| 26 | +static inline struct mptcp_sock *bpf_mptcp_sock_from_subflow(struct sock *sk) { return NULL; } | 36 | +static inline struct mptcp_sock *bpf_mptcp_sock_from_sock(struct sock *sk) { return NULL; } |
| 27 | #endif | 37 | #endif |
| 28 | 38 | ||
| 29 | #if !IS_ENABLED(CONFIG_MPTCP) | 39 | #if !IS_ENABLED(CONFIG_MPTCP) |
| 30 | diff --git a/net/core/filter.c b/net/core/filter.c | 40 | diff --git a/net/core/filter.c b/net/core/filter.c |
| 31 | index XXXXXXX..XXXXXXX 100644 | 41 | index XXXXXXX..XXXXXXX 100644 |
| 32 | --- a/net/core/filter.c | 42 | --- a/net/core/filter.c |
| 33 | +++ b/net/core/filter.c | 43 | +++ b/net/core/filter.c |
| 34 | @@ -XXX,XX +XXX,XX @@ const struct bpf_func_proto bpf_skc_to_unix_sock_proto = { | 44 | @@ -XXX,XX +XXX,XX @@ const struct bpf_func_proto bpf_skc_to_unix_sock_proto = { |
| 35 | BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk) | 45 | BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk) |
| 36 | { | 46 | { |
| 37 | BTF_TYPE_EMIT(struct mptcp_sock); | 47 | BTF_TYPE_EMIT(struct mptcp_sock); |
| 38 | - return (unsigned long)bpf_mptcp_sock_from_sock(sk); | 48 | - return (unsigned long)bpf_mptcp_sock_from_subflow(sk); |
| 39 | + return (unsigned long)bpf_mptcp_sock_from_subflow(sk); | 49 | + return (unsigned long)bpf_mptcp_sock_from_sock(sk); |
| 40 | } | 50 | } |
| 41 | 51 | ||
| 42 | const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = { | 52 | const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = { |
| 43 | diff --git a/net/mptcp/bpf.c b/net/mptcp/bpf.c | 53 | diff --git a/net/mptcp/bpf.c b/net/mptcp/bpf.c |
| 44 | index XXXXXXX..XXXXXXX 100644 | 54 | index XXXXXXX..XXXXXXX 100644 |
| 45 | --- a/net/mptcp/bpf.c | 55 | --- a/net/mptcp/bpf.c |
| 46 | +++ b/net/mptcp/bpf.c | 56 | +++ b/net/mptcp/bpf.c |
| 47 | @@ -XXX,XX +XXX,XX @@ static struct bpf_struct_ops bpf_mptcp_sched_ops = { | 57 | @@ -XXX,XX +XXX,XX @@ static struct bpf_struct_ops bpf_mptcp_sched_ops = { |
| 48 | }; | 58 | }; |
| 49 | #endif /* CONFIG_BPF_JIT */ | 59 | #endif /* CONFIG_BPF_JIT */ |
| 50 | 60 | ||
| 51 | -struct mptcp_sock *bpf_mptcp_sock_from_sock(struct sock *sk) | 61 | -struct mptcp_sock *bpf_mptcp_sock_from_subflow(struct sock *sk) |
| 52 | +struct mptcp_sock *bpf_mptcp_sock_from_subflow(struct sock *sk) | 62 | +struct mptcp_sock *bpf_mptcp_sock_from_sock(struct sock *sk) |
| 53 | { | 63 | { |
| 54 | - if (unlikely(!sk || !sk_fullsock(sk))) | 64 | - if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP && sk_is_mptcp(sk)) |
| 55 | - return NULL; | 65 | + if (unlikely(!sk || !sk_fullsock(sk))) |
| 56 | - | 66 | + return NULL; |
| 57 | - if (sk->sk_protocol == IPPROTO_MPTCP) | 67 | + |
| 58 | - return mptcp_sk(sk); | 68 | + if (sk->sk_protocol == IPPROTO_MPTCP) |
| 59 | - | 69 | + return mptcp_sk(sk); |
| 60 | - if (sk->sk_protocol == IPPROTO_TCP && sk_is_mptcp(sk)) | 70 | + |
| 61 | + if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP && sk_is_mptcp(sk)) | 71 | + if (sk->sk_protocol == IPPROTO_TCP && sk_is_mptcp(sk)) |
| 62 | return mptcp_sk(mptcp_subflow_ctx(sk)->conn); | 72 | return mptcp_sk(mptcp_subflow_ctx(sk)->conn); |
| 63 | 73 | ||
| 64 | return NULL; | 74 | return NULL; |
| 65 | -- | 75 | -- |
| 66 | 2.43.0 | 76 | 2.45.2 | diff view generated by jsdifflib |
| 1 | From: Geliang Tang <tanggeliang@kylinos.cn> | 1 | From: Geliang Tang <tanggeliang@kylinos.cn> |
|---|---|---|---|
| 2 | 2 | ||
| 3 | bpf_core_cast() can be used in mptcp_subflow bpf_iter selftests to get the | 3 | Currently, bpf_skc_to_mptcp_sock() helper is not allowed to be used |
| 4 | msk, instead of using bpf_skc_to_mptcp_sock(). No need to add this patch | 4 | in cg_sockopt. This patch adds this permission. |
| 5 | anymore, revert it. | ||
| 6 | 5 | ||
| 7 | This reverts commit d8d42f5a4542c2dc31e4e00c58508a821c6ab789. | 6 | Thanks to the previous patch allowing skc_to_mptcp_sock() to be used |
| 7 | with MPTCP sockets, this permission allows this helper to be use it in | ||
| 8 | CGroup BPF hooks, e.g. [gs]etsocktopt. | ||
| 9 | |||
| 10 | Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn> | ||
| 8 | --- | 11 | --- |
| 9 | kernel/bpf/cgroup.c | 2 -- | 12 | kernel/bpf/cgroup.c | 2 ++ |
| 10 | 1 file changed, 2 deletions(-) | 13 | 1 file changed, 2 insertions(+) |
| 11 | 14 | ||
| 12 | diff --git a/kernel/bpf/cgroup.c b/kernel/bpf/cgroup.c | 15 | diff --git a/kernel/bpf/cgroup.c b/kernel/bpf/cgroup.c |
| 13 | index XXXXXXX..XXXXXXX 100644 | 16 | index XXXXXXX..XXXXXXX 100644 |
| 14 | --- a/kernel/bpf/cgroup.c | 17 | --- a/kernel/bpf/cgroup.c |
| 15 | +++ b/kernel/bpf/cgroup.c | 18 | +++ b/kernel/bpf/cgroup.c |
| 16 | @@ -XXX,XX +XXX,XX @@ cg_sockopt_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) | 19 | @@ -XXX,XX +XXX,XX @@ cg_sockopt_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) |
| 17 | #ifdef CONFIG_INET | 20 | #ifdef CONFIG_INET |
| 18 | case BPF_FUNC_tcp_sock: | 21 | case BPF_FUNC_tcp_sock: |
| 19 | return &bpf_tcp_sock_proto; | 22 | return &bpf_tcp_sock_proto; |
| 20 | - case BPF_FUNC_skc_to_mptcp_sock: | 23 | + case BPF_FUNC_skc_to_mptcp_sock: |
| 21 | - return &bpf_skc_to_mptcp_sock_proto; | 24 | + return &bpf_skc_to_mptcp_sock_proto; |
| 22 | #endif | 25 | #endif |
| 23 | case BPF_FUNC_perf_event_output: | 26 | case BPF_FUNC_perf_event_output: |
| 24 | return &bpf_event_output_data_proto; | 27 | return &bpf_event_output_data_proto; |
| 25 | -- | 28 | -- |
| 26 | 2.43.0 | 29 | 2.45.2 | diff view generated by jsdifflib |
| 1 | From: Geliang Tang <tanggeliang@kylinos.cn> | 1 | From: Geliang Tang <tanggeliang@kylinos.cn> |
|---|---|---|---|
| 2 | 2 | ||
| 3 | Drop this patch as Martin suggested. | 3 | Drop bpf_mptcp_sk() and bpf_mptcp_subflow_tcp_sock() definitions. Use |
| 4 | bpf_skc_to_mptcp_sock() and mptcp_subflow_tcp_sock() in mptcp_subflow | ||
| 5 | bpf_iter selftests instead. | ||
| 4 | 6 | ||
| 5 | From Martin's review [1], this mptcp_sock_acquire() helper was a workaround | 7 | Address Martin's comments in v1: |
| 6 | only to please the verifier, but they were not needed. | ||
| 7 | 8 | ||
| 8 | [1] | 9 | - add null-check for bpf_mptcp_subflow_ctx. |
| 9 | https://lore.kernel.org/9b373a23-c093-42d8-b4ae-99f2e62e7681@linux.dev | 10 | - add KF_RET_NULL flags for bpf_mptcp_subflow_ctx. |
| 11 | - register this kfunc set to BPF_PROG_TYPE_CGROUP_SOCKOPT only, | ||
| 12 | not BPF_PROG_TYPE_UNSPEC. | ||
| 10 | 13 | ||
| 11 | Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn> | 14 | Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn> |
| 12 | --- | 15 | --- |
| 13 | net/mptcp/bpf.c | 19 ------------------- | 16 | net/mptcp/bpf.c | 21 ++++++--------------- |
| 14 | 1 file changed, 19 deletions(-) | 17 | 1 file changed, 6 insertions(+), 15 deletions(-) |
| 15 | 18 | ||
| 16 | diff --git a/net/mptcp/bpf.c b/net/mptcp/bpf.c | 19 | diff --git a/net/mptcp/bpf.c b/net/mptcp/bpf.c |
| 17 | index XXXXXXX..XXXXXXX 100644 | 20 | index XXXXXXX..XXXXXXX 100644 |
| 18 | --- a/net/mptcp/bpf.c | 21 | --- a/net/mptcp/bpf.c |
| 19 | +++ b/net/mptcp/bpf.c | 22 | +++ b/net/mptcp/bpf.c |
| 20 | @@ -XXX,XX +XXX,XX @@ bpf_iter_mptcp_subflow_destroy(struct bpf_iter_mptcp_subflow *it) | 23 | @@ -XXX,XX +XXX,XX @@ struct bpf_iter_mptcp_subflow_kern { |
| 21 | { | 24 | |
| 22 | } | 25 | __bpf_kfunc_start_defs(); |
| 23 | 26 | ||
| 24 | -__bpf_kfunc static struct | 27 | -__bpf_kfunc static struct mptcp_sock *bpf_mptcp_sk(struct sock *sk) |
| 25 | -mptcp_sock *bpf_mptcp_sock_acquire(struct mptcp_sock *msk) | ||
| 26 | -{ | 28 | -{ |
| 27 | - struct sock *sk = (struct sock *)msk; | 29 | - return mptcp_sk(sk); |
| 28 | - | ||
| 29 | - if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) | ||
| 30 | - return msk; | ||
| 31 | - return NULL; | ||
| 32 | -} | 30 | -} |
| 33 | - | 31 | - |
| 34 | -__bpf_kfunc static void bpf_mptcp_sock_release(struct mptcp_sock *msk) | 32 | __bpf_kfunc static struct mptcp_subflow_context * |
| 33 | bpf_mptcp_subflow_ctx(const struct sock *sk) | ||
| 34 | { | ||
| 35 | - return mptcp_subflow_ctx(sk); | ||
| 36 | -} | ||
| 37 | + if (sk && sk_fullsock(sk) && | ||
| 38 | + sk->sk_protocol == IPPROTO_TCP && sk_is_mptcp(sk)) | ||
| 39 | + return mptcp_subflow_ctx(sk); | ||
| 40 | |||
| 41 | -__bpf_kfunc static struct sock * | ||
| 42 | -bpf_mptcp_subflow_tcp_sock(const struct mptcp_subflow_context *subflow) | ||
| 35 | -{ | 43 | -{ |
| 36 | - struct sock *sk = (struct sock *)msk; | 44 | - return mptcp_subflow_tcp_sock(subflow); |
| 37 | - | 45 | + return NULL; |
| 38 | - WARN_ON_ONCE(!sk || !refcount_dec_not_one(&sk->sk_refcnt)); | 46 | } |
| 39 | -} | 47 | |
| 40 | - | 48 | __bpf_kfunc static int |
| 41 | __bpf_kfunc struct mptcp_subflow_context * | 49 | @@ -XXX,XX +XXX,XX @@ __bpf_kfunc static bool bpf_mptcp_subflow_queues_empty(struct sock *sk) |
| 42 | bpf_mptcp_subflow_ctx_by_pos(const struct mptcp_sched_data *data, unsigned int pos) | 50 | __bpf_kfunc_end_defs(); |
| 43 | { | 51 | |
| 44 | @@ -XXX,XX +XXX,XX @@ BTF_ID_FLAGS(func, bpf_mptcp_subflow_ctx, KF_RET_NULL) | 52 | BTF_KFUNCS_START(bpf_mptcp_common_kfunc_ids) |
| 53 | -BTF_ID_FLAGS(func, bpf_mptcp_sk) | ||
| 54 | -BTF_ID_FLAGS(func, bpf_mptcp_subflow_ctx) | ||
| 55 | -BTF_ID_FLAGS(func, bpf_mptcp_subflow_tcp_sock) | ||
| 56 | +BTF_ID_FLAGS(func, bpf_mptcp_subflow_ctx, KF_RET_NULL) | ||
| 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_new, KF_ITER_NEW | KF_TRUSTED_ARGS) |
| 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_next, KF_ITER_NEXT | KF_RET_NULL) |
| 47 | BTF_ID_FLAGS(func, bpf_iter_mptcp_subflow_destroy, KF_ITER_DESTROY) | 59 | BTF_ID_FLAGS(func, bpf_iter_mptcp_subflow_destroy, KF_ITER_DESTROY) |
| 48 | -BTF_ID_FLAGS(func, bpf_mptcp_sock_acquire, KF_ACQUIRE | KF_RET_NULL) | 60 | @@ -XXX,XX +XXX,XX @@ static int __init bpf_mptcp_kfunc_init(void) |
| 49 | -BTF_ID_FLAGS(func, bpf_mptcp_sock_release, KF_RELEASE) | 61 | int ret; |
| 50 | BTF_KFUNCS_END(bpf_mptcp_common_kfunc_ids) | 62 | |
| 51 | 63 | ret = register_btf_fmodret_id_set(&bpf_mptcp_fmodret_set); | |
| 52 | static const struct btf_kfunc_id_set bpf_mptcp_common_kfunc_set = { | 64 | - ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_UNSPEC, |
| 65 | + ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SOCKOPT, | ||
| 66 | &bpf_mptcp_common_kfunc_set); | ||
| 67 | ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_STRUCT_OPS, | ||
| 68 | &bpf_mptcp_sched_kfunc_set); | ||
| 53 | -- | 69 | -- |
| 54 | 2.43.0 | 70 | 2.45.2 | diff view generated by jsdifflib |
| 1 | From: Geliang Tang <tanggeliang@kylinos.cn> | 1 | From: Geliang Tang <tanggeliang@kylinos.cn> |
|---|---|---|---|
| 2 | 2 | ||
| 3 | Drop the NULL check for 'msk' as Martin suggested, add more checks | 3 | Add "sizeof" and "alignof" checks. |
| 4 | for 'sk'. | ||
| 5 | 4 | ||
| 6 | Use the "struct sock *sk" instead of "struct mptcp-sock *msk" as the | 5 | Address Martin's comments in v1: |
| 7 | argument in the bpf_iter_mptcp_subflow_new as Martin suggested. | ||
| 8 | 6 | ||
| 9 | v4: | 7 | - bpf_iter_mptcp_subflow_new returns -EINVAL when msk socket lock isn't |
| 10 | - drop sock_owned_by_user_nocheck and spin_is_locked. According to | 8 | held. |
| 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. | ||
| 26 | 9 | ||
| 27 | Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn> | 10 | Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn> |
| 28 | --- | 11 | --- |
| 29 | net/mptcp/bpf.c | 12 ++++++------ | 12 | net/mptcp/bpf.c | 10 +++++++++- |
| 30 | 1 file changed, 6 insertions(+), 6 deletions(-) | 13 | 1 file changed, 9 insertions(+), 1 deletion(-) |
| 31 | 14 | ||
| 32 | diff --git a/net/mptcp/bpf.c b/net/mptcp/bpf.c | 15 | diff --git a/net/mptcp/bpf.c b/net/mptcp/bpf.c |
| 33 | index XXXXXXX..XXXXXXX 100644 | 16 | index XXXXXXX..XXXXXXX 100644 |
| 34 | --- a/net/mptcp/bpf.c | 17 | --- a/net/mptcp/bpf.c |
| 35 | +++ b/net/mptcp/bpf.c | 18 | +++ b/net/mptcp/bpf.c |
| 36 | @@ -XXX,XX +XXX,XX @@ bpf_mptcp_subflow_ctx(const struct sock *sk) | 19 | @@ -XXX,XX +XXX,XX @@ bpf_iter_mptcp_subflow_new(struct bpf_iter_mptcp_subflow *it, |
| 37 | 20 | struct mptcp_sock *msk) | |
| 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) | ||
| 42 | { | 21 | { |
| 43 | struct bpf_iter_mptcp_subflow_kern *kit = (void *)it; | 22 | struct bpf_iter_mptcp_subflow_kern *kit = (void *)it; |
| 44 | - struct sock *sk = (struct sock *)msk; | 23 | + struct sock *sk = (struct sock *)msk; |
| 45 | + struct mptcp_sock *msk; | 24 | + |
| 46 | 25 | + BUILD_BUG_ON(sizeof(struct bpf_iter_mptcp_subflow_kern) > | |
| 47 | BUILD_BUG_ON(sizeof(struct bpf_iter_mptcp_subflow_kern) > | 26 | + sizeof(struct bpf_iter_mptcp_subflow)); |
| 48 | sizeof(struct bpf_iter_mptcp_subflow)); | 27 | + BUILD_BUG_ON(__alignof__(struct bpf_iter_mptcp_subflow_kern) != |
| 49 | BUILD_BUG_ON(__alignof__(struct bpf_iter_mptcp_subflow_kern) != | 28 | + __alignof__(struct bpf_iter_mptcp_subflow)); |
| 50 | __alignof__(struct bpf_iter_mptcp_subflow)); | 29 | |
| 51 | 30 | kit->msk = msk; | |
| 52 | - kit->msk = msk; | 31 | if (!msk) |
| 53 | - if (!msk) | ||
| 54 | + if (unlikely(!sk || !sk_fullsock(sk))) | ||
| 55 | return -EINVAL; | 32 | return -EINVAL; |
| 56 | 33 | ||
| 57 | - if (!sock_owned_by_user_nocheck(sk) && | 34 | - msk_owned_by_me(msk); |
| 58 | - !spin_is_locked(&sk->sk_lock.slock)) | 35 | + if (!sock_owned_by_user_nocheck(sk) && |
| 59 | + if (sk->sk_protocol != IPPROTO_MPTCP) | 36 | + !spin_is_locked(&sk->sk_lock.slock)) |
| 60 | return -EINVAL; | 37 | + return -EINVAL; |
| 61 | 38 | ||
| 62 | + msk = mptcp_sk(sk); | ||
| 63 | + kit->msk = msk; | ||
| 64 | kit->pos = &msk->conn_list; | 39 | kit->pos = &msk->conn_list; |
| 65 | return 0; | 40 | return 0; |
| 66 | } | ||
| 67 | -- | 41 | -- |
| 68 | 2.43.0 | 42 | 2.45.2 | diff view generated by jsdifflib |
| 1 | From: Geliang Tang <tanggeliang@kylinos.cn> | 1 | From: Geliang Tang <tanggeliang@kylinos.cn> |
|---|---|---|---|
| 2 | 2 | ||
| 3 | Use bpf_core_cast() instead of bpf_skc_to_mptcp_sock(). | 3 | Use bpf_skc_to_mptcp_sock() and mptcp_subflow_tcp_sock() instead of |
| 4 | Change the 2nd parameter type of bpf_for_each() as 'struct sock'. | 4 | bpf_mptcp_sk() and bpf_mptcp_subflow_tcp_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. | ||
| 9 | 5 | ||
| 10 | Please update the commit log as: | 6 | IPPROTO_MPTCP is checked in bpf_skc_to_mptcp_sock(), no need to check |
| 7 | it in BPF program. | ||
| 11 | 8 | ||
| 12 | ''' | 9 | bpf_skc_to_mptcp_sock() and bpf_mptcp_subflow_ctx() may return NULL, |
| 13 | This patch adds a "cgroup/getsockopt" program "iters_subflow" to test the | 10 | need to check the return values. |
| 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 | ''' | ||
| 37 | 11 | ||
| 38 | Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn> | 12 | Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn> |
| 39 | --- | 13 | --- |
| 40 | tools/testing/selftests/bpf/bpf_experimental.h | 2 +- | 14 | tools/testing/selftests/bpf/progs/mptcp_bpf.h | 1 - |
| 41 | tools/testing/selftests/bpf/progs/mptcp_bpf.h | 5 ----- | 15 | tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c | 11 +++++------ |
| 42 | tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c | 10 +++------- | 16 | 2 files changed, 5 insertions(+), 7 deletions(-) |
| 43 | 3 files changed, 4 insertions(+), 13 deletions(-) | ||
| 44 | 17 | ||
| 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 | ||
| 58 | diff --git a/tools/testing/selftests/bpf/progs/mptcp_bpf.h b/tools/testing/selftests/bpf/progs/mptcp_bpf.h | 18 | diff --git a/tools/testing/selftests/bpf/progs/mptcp_bpf.h b/tools/testing/selftests/bpf/progs/mptcp_bpf.h |
| 59 | index XXXXXXX..XXXXXXX 100644 | 19 | index XXXXXXX..XXXXXXX 100644 |
| 60 | --- a/tools/testing/selftests/bpf/progs/mptcp_bpf.h | 20 | --- a/tools/testing/selftests/bpf/progs/mptcp_bpf.h |
| 61 | +++ b/tools/testing/selftests/bpf/progs/mptcp_bpf.h | 21 | +++ b/tools/testing/selftests/bpf/progs/mptcp_bpf.h |
| 62 | @@ -XXX,XX +XXX,XX @@ mptcp_subflow_tcp_sock(const struct mptcp_subflow_context *subflow) | 22 | @@ -XXX,XX +XXX,XX @@ mptcp_subflow_tcp_sock(const struct mptcp_subflow_context *subflow) |
| 63 | } | 23 | extern struct mptcp_sock *bpf_mptcp_sock_acquire(struct mptcp_sock *msk) __ksym; |
| 64 | 24 | extern void bpf_mptcp_sock_release(struct mptcp_sock *msk) __ksym; | |
| 65 | /* ksym */ | 25 | |
| 66 | -extern struct mptcp_sock *bpf_mptcp_sock_acquire(struct mptcp_sock *msk) __ksym; | 26 | -extern struct mptcp_sock *bpf_mptcp_sk(struct sock *sk) __ksym; |
| 67 | -extern void bpf_mptcp_sock_release(struct mptcp_sock *msk) __ksym; | ||
| 68 | - | ||
| 69 | extern struct mptcp_subflow_context * | 27 | extern struct mptcp_subflow_context * |
| 70 | bpf_mptcp_subflow_ctx(const struct sock *sk) __ksym; | 28 | bpf_mptcp_subflow_ctx(const struct sock *sk) __ksym; |
| 71 | -extern struct sock * | 29 | 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; | ||
| 76 | diff --git a/tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c b/tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c | 30 | diff --git a/tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c b/tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c |
| 77 | index XXXXXXX..XXXXXXX 100644 | 31 | index XXXXXXX..XXXXXXX 100644 |
| 78 | --- a/tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c | 32 | --- a/tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c |
| 79 | +++ b/tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c | 33 | +++ b/tools/testing/selftests/bpf/progs/mptcp_bpf_iters.c |
| 80 | @@ -XXX,XX +XXX,XX @@ int iters_subflow(struct bpf_sockopt *ctx) | 34 | @@ -XXX,XX +XXX,XX @@ int iters_subflow(struct bpf_sockopt *ctx) |
| 81 | if (ctx->level != SOL_TCP || ctx->optname != TCP_IS_MPTCP) | 35 | struct mptcp_sock *msk; |
| 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) | ||
| 82 | return 1; | 41 | return 1; |
| 83 | 42 | ||
| 84 | - msk = bpf_skc_to_mptcp_sock(sk); | 43 | - msk = bpf_mptcp_sk((struct sock *)sk); |
| 85 | + msk = bpf_core_cast(sk, struct mptcp_sock); | 44 | - if (msk->pm.server_side || !msk->pm.subflows) |
| 86 | if (!msk || msk->pm.server_side || !msk->pm.subflows) | 45 | + msk = bpf_skc_to_mptcp_sock(sk); |
| 46 | + if (!msk || msk->pm.server_side || !msk->pm.subflows) | ||
| 87 | return 1; | 47 | return 1; |
| 88 | 48 | ||
| 89 | - msk = bpf_mptcp_sock_acquire(msk); | 49 | 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. | ||
| 97 | @@ -XXX,XX +XXX,XX @@ int iters_subflow(struct bpf_sockopt *ctx) | 50 | @@ -XXX,XX +XXX,XX @@ int iters_subflow(struct bpf_sockopt *ctx) |
| 98 | |||
| 99 | local_ids += subflow->subflow_id; | 51 | local_ids += subflow->subflow_id; |
| 100 | 52 | ||
| 101 | - /* only to check the following kfunc works */ | 53 | /* only to check the following kfunc works */ |
| 102 | + /* only to check the following helper works */ | 54 | - ssk = bpf_mptcp_subflow_tcp_sock(subflow); |
| 103 | ssk = mptcp_subflow_tcp_sock(subflow); | 55 | + ssk = mptcp_subflow_tcp_sock(subflow); |
| 104 | } | 56 | } |
| 105 | 57 | ||
| 58 | if (!ssk) | ||
| 106 | @@ -XXX,XX +XXX,XX @@ int iters_subflow(struct bpf_sockopt *ctx) | 59 | @@ -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 | |||
| 107 | ids = local_ids; | 67 | ids = local_ids; |
| 108 | |||
| 109 | out: | ||
| 110 | - bpf_mptcp_sock_release(msk); | ||
| 111 | return 1; | ||
| 112 | } | ||
| 113 | -- | 68 | -- |
| 114 | 2.43.0 | 69 | 2.45.2 | diff view generated by jsdifflib |