add bpf_stale scheduler
Subsystems (4)
v3 cover add bpf_stale scheduler
v3:
- init sk_storage in .init, delete it in .release.
We invoke bpf_sk_storage_get() to get the sk_storage map many times. Only the first call is the slow path (see bpf_local_storage_lookup in kernel/bpf/bpf_local_storage.c), alloc the map and cache it. The subsequent calls are all in fast path, the cache hits. So we should first call bpf_sk_storage_get in .init, then call it many times in .get_subflow.
- if no subflow is scheduled, it should return '-1' to indicate an error.
v2:
- store subflow ids instead of storing subflow pointers in sk_storage.
v1:
- This patchset adds the new bpf_stale scheduler. Use sk_storage to save the stale map instead of using subflow->stale flag to manage it.
Geliang Tang (5):
Squash to "selftests/bpf: Add bpf_bkup scheduler" Squash to "selftests/bpf: Add bpf_rr scheduler" Squash to "selftests/bpf: Add bpf_burst scheduler" selftests/bpf: Add bpf_stale scheduler selftests/bpf: Add bpf_stale test
tools/testing/selftests/bpf/bpf_tcp_helpers.h | 1 + .../testing/selftests/bpf/prog_tests/mptcp.c | 38 +++++ .../selftests/bpf/progs/mptcp_bpf_bkup.c | 4 +- .../selftests/bpf/progs/mptcp_bpf_burst.c | 2 + .../selftests/bpf/progs/mptcp_bpf_rr.c | 2 + .../selftests/bpf/progs/mptcp_bpf_stale.c | 152 ++++++++++++++++++ 6 files changed, 198 insertions(+), 1 deletion(-) create mode 100644 tools/testing/selftests/bpf/progs/mptcp_bpf_stale.c
Signature · 1 line
2.35.3
PATCH v3 1/5 Squash to "selftests/bpf: Add bpf_bkup scheduler"
Return -1 when no subflow is selected.
--- tools/testing/selftests/bpf/progs/mptcp_bpf_bkup.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
tools/testing/selftests/bpf/progs/mptcp_bpf_bkup.c +3 -1
@@ -36,8 +36,10 @@ int BPF_STRUCT_OPS(bpf_bkup_get_subflow, struct mptcp_sock *msk, } } -if (nr !=-1 ) +if (nr !=-1 ) {mptcp_subflow_set_scheduled (mptcp_subflow_ctx_by_pos (data, nr), true); +return -1 ; + }return 0 ; }
Signature · 1 line
2.35.3
PATCH v3 2/5 Squash to "selftests/bpf: Add bpf_rr scheduler"
In reply to Geliang TangAlloc and init mptcp_rr_map first in mptcp_sched_rr_init().
--- tools/testing/selftests/bpf/progs/mptcp_bpf_rr.c | 2 ++ 1 file changed, 2 insertions(+)
tools/testing/selftests/bpf/progs/mptcp_bpf_rr.c +2 -0
@@ -20,6 +20,8 @@ struct { SEC("struct_ops/mptcp_sched_rr_init")void BPF_PROG (mptcp_sched_rr_init ,struct mptcp_sock *msk) { +bpf_sk_storage_get (&mptcp_rr_map, msk,0 , +BPF_LOCAL_STORAGE_GET_F_CREATE ); }SEC ("struct_ops/mptcp_sched_rr_release" )
Signature · 1 line
2.35.3
PATCH v3 3/5 Squash to "selftests/bpf: Add bpf_burst scheduler"
In reply to Geliang TangAlloc and init sk_storage in data_init().
--- tools/testing/selftests/bpf/progs/mptcp_bpf_burst.c | 2 ++ 1 file changed, 2 insertions(+)
tools/testing/selftests/bpf/progs/mptcp_bpf_burst.c +2 -0
@@ -69,6 +69,8 @@ static __always_inline bool sk_stream_memory_free(const struct sock *sk) SEC("struct_ops/mptcp_sched_burst_init")void BPF_PROG (mptcp_sched_burst_init ,struct mptcp_sock *msk) { +bpf_sk_storage_get (&mptcp_burst_map, msk,0 , +BPF_LOCAL_STORAGE_GET_F_CREATE ); }SEC ("struct_ops/mptcp_sched_burst_release" )
Signature · 1 line
2.35.3
PATCH v3 4/5 selftests/bpf: Add bpf_stale scheduler
In reply to Geliang TangThis patch implements the setting a subflow as stale/unstale in BPF MPTCP scheduler, named bpf_stale. The staled subflow id will be added into a map in sk_storage.
Two helper mptcp_subflow_set_stale() and mptcp_subflow_clear_stale() are added.
In this test, subflow 1 is set as stale in bpf_stale_data_init(). Each subflow is checked whether it's a stale one in bpf_stale_get_subflow() to select a unstale subflow to send data.
--- tools/testing/selftests/bpf/bpf_tcp_helpers.h | 1 + .../selftests/bpf/progs/mptcp_bpf_stale.c | 152 ++++++++++++++++++ 2 files changed, 153 insertions(+) create mode 100644 tools/testing/selftests/bpf/progs/mptcp_bpf_stale.c
tools/testing/selftests/bpf/bpf_tcp_helpers.h +1 -0
@@ -239,6 +239,7 @@ struct mptcp_subflow_context {unsigned long avg_pacing_rate; __u32 backup : 1;__u8 stale_count; +__u32 subflow_id;struct sock *tcp_sock;/* tcp sk backpointer */ } __attribute__((preserve_access_index));
tools/testing/selftests/bpf/progs/mptcp_bpf_stale.c new file +152 -0
152 lines not on this page · raw message
Signature · 1 line
2.35.3
Mat Martineau
maintainerRe: [PATCH mptcp-next v3 4/5] selftests/bpf: Add bpf_stale scheduler
Quoted 140 lines · Geliang Tang
This patch implements the setting a subflow as stale/unstale in BPF MPTCP scheduler, named
bpf_stale. The staled subflow id will be added into a map insk_storage.Two helper
mptcp_subflow_set_stale()andmptcp_subflow_clear_stale()are added.In this test, subflow 1 is set as stale in
bpf_stale_data_init(). Each subflow is checked whether it's a stale one inbpf_stale_get_subflow()to select a unstale subflow to send data.Signed-off-by Geliang Tang <[email protected]>---
tools/testing/selftests/bpf/bpf_tcp_helpers.h| 1 +.../selftests/bpf/progs/mptcp_bpf_stale.c| 152 ++++++++++++++++++2 files changed, 153 insertions(+)create mode 100644
tools/testing/selftests/bpf/progs/mptcp_bpf_stale.ctools/testing/selftests/bpf/bpf_tcp_helpers.h +1
-0@@ -239,6 +239,7 @@ struct mptcp_subflow_context {unsigned long avg_pacing_rate; __u32 backup : 1;__u8 stale_count; +__u32 subflow_id;struct sock *tcp_sock;/* tcp sk backpointer */ } __attribute__((preserve_access_index));tools/testing/selftests/bpf/progs/mptcp_bpf_stale.c new file +104
-0@@ -0,0 +1,152 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright (c) 2023, SUSE. */ + +#include <linux/bpf.h> +#include "bpf_tcp_helpers.h" + +char _license[] SEC("license") = "GPL"; + +struct mptcp_stale_storage { +__u8 nr; +__u32 ids[MPTCP_SUBFLOWS_MAX ]; +}; + +struct { + __uint(type, BPF_MAP_TYPE_SK_STORAGE); + __uint(map_flags, BPF_F_NO_PREALLOC); + __type(key, int); + __type(value, struct mptcp_stale_storage); +} mptcp_stale_map SEC(".maps"); + +static void mptcp_subflow_set_stale (struct mptcp_stale_storage *storage, +__u32 subflow_id) +{ +if (!subflow_id) +return ; + +for (int i =0 ; i < storage->nr && i <MPTCP_SUBFLOWS_MAX ; i++) { +if (storage->ids[i] == subflow_id) +return ; + } + +if (storage->nr <MPTCP_SUBFLOWS_MAX -1 ) + storage->ids[storage->nr++] = subflow_id; +} + +static void mptcp_subflow_clear_stale (struct mptcp_stale_storage *storage, +__u32 subflow_id) +{ +if (!subflow_id) +return ; + +for (int i =0 ; i < storage->nr && i <MPTCP_SUBFLOWS_MAX ; i++) { +if (storage->ids[i] == subflow_id) { +for (int j = i; j <MPTCP_SUBFLOWS_MAX -1 ; j++) { +if (!storage->ids[j +1 ]) +break ; + storage->ids[j] = storage->ids[j +1 ]; + storage->ids[j +1 ] =0 ; + } + storage->nr--; +return ; + } + } +} + +static bool mptcp_subflow_is_stale (struct mptcp_stale_storage *storage, +__u32 subflow_id) +{ +for (int i =0 ; i < storage->nr && i <MPTCP_SUBFLOWS_MAX ; i++) { +if (storage->ids[i] == subflow_id) +return true; + } + +return false; +} + +static bool mptcp_subflow_is_active (struct mptcp_sched_data *data, +__u32 subflow_id) +{ +for (int i =0 ; i < data->subflows && i <MPTCP_SUBFLOWS_MAX ; i++) { +struct mptcp_subflow_context *subflow; + + subflow =mptcp_subflow_ctx_by_pos (data, i); +if (!subflow) +break ; +if (subflow->subflow_id == subflow_id) +return true; + } + +return false; +} + +SEC("struct_ops/mptcp_sched_stale_init") +void BPF_PROG (mptcp_sched_stale_init ,struct mptcp_sock *msk) +{ +struct mptcp_stale_storage *storage; + + storage =bpf_sk_storage_get (&mptcp_stale_map, msk,0 , +BPF_LOCAL_STORAGE_GET_F_CREATE ); +if (!storage) +return ; + + storage->nr =0 ; +} + +SEC("struct_ops/mptcp_sched_stale_release") +void BPF_PROG (mptcp_sched_stale_release ,struct mptcp_sock *msk) +{ +bpf_sk_storage_delete (&mptcp_stale_map, msk); +} + +int BPF_STRUCT_OPS (bpf_stale_get_subflow ,struct mptcp_sock *msk, +struct mptcp_sched_data *data) +{
tools/testing/selftests/bpf/progs/mptcp_bpf_stale.c line 114
Quoted 10 lines
tools/testing/selftests/bpf/progs/mptcp_bpf_stale.c new file +10
-0+struct mptcp_subflow_context *subflow; +struct mptcp_stale_storage *storage; +int nr =-1 ; + + storage =bpf_sk_storage_get (&mptcp_stale_map, msk,0 , +BPF_LOCAL_STORAGE_GET_F_CREATE ); +if (!storage) +return -1 ; + +mptcp_sched_data_set_contexts (msk, data);
Should this call be moved to sched.c, right before the calls to msk->sched->get_subflow() in the mptcp_sched_get functions?
It looks like every BPF scheduler example has to call this, so it would be more efficient to set up the data before calling the BPF functions.
tools/testing/selftests/bpf/progs/mptcp_bpf_stale.c line 124
Quoted 10 lines
tools/testing/selftests/bpf/progs/mptcp_bpf_stale.c new file +10
-0+ +/* Handle invalid subflow ids for subflows that have been closed */ +for (int i =0 ; i < storage->nr && i <MPTCP_SUBFLOWS_MAX ; i++) { +if (!mptcp_subflow_is_active (data, storage->ids[i])) +mptcp_subflow_clear_stale (storage, storage->ids[i]); + } + + subflow =mptcp_subflow_ctx_by_pos (data,1 ); +if (subflow) +mptcp_subflow_set_stale (storage, subflow->subflow_id);
Since this is always marking the subflow in position 1 as stale...
tools/testing/selftests/bpf/progs/mptcp_bpf_stale.c line 133
Quoted 9 lines
tools/testing/selftests/bpf/progs/mptcp_bpf_stale.c new file +9
-0+ +for (int i =0 ; i < data->subflows && i <MPTCP_SUBFLOWS_MAX ; i++) { +struct mptcp_subflow_context *subflow; + + subflow =mptcp_subflow_ctx_by_pos (data, i); +if (!subflow) +break ; + +if (mptcp_subflow_is_stale (storage, subflow->subflow_id))
...and then looking up that subflow again, this scheduler is not doing anything with actual stale subflows. The other BPF schedulers are already exercising everything in the BPF infrastructure that is done here (with sk_storage, etc.).
If there's a way to check for real stuck/stale subflows, then I think it would be good to have a sample scheduler that does that. But for now, I don't think this code adds test coverage or helps demonstrate the BPF scheduler.
Quoted 24 lines
tools/testing/selftests/bpf/progs/mptcp_bpf_stale.c new file +19
-0+continue ; + + nr = i; + } + +if (nr !=-1 ) { +mptcp_subflow_set_scheduled (mptcp_subflow_ctx_by_pos (data, nr), true); +return -1 ; + } +return 0 ; +} + +SEC(".struct_ops") +struct mptcp_sched_ops stale = { + .init = (void *)mptcp_sched_stale_init, + .release = (void *)mptcp_sched_stale_release, + .get_subflow = (void *)bpf_stale_get_subflow, + .name ="bpf_stale" , +};Signature · 1 line
2.35.3
PATCH v3 5/5 selftests/bpf: Add bpf_stale test
In reply to Geliang TangThis patch adds the bpf_stale scheduler test: test_stale(). Use sysctl to set net.mptcp.scheduler to use this sched. Add two veth net devices to simulate the multiple addresses case. Use 'ip mptcp endpoint' command to add the new endpoint ADDR_2 to PM netlink. Send data and check bytes_sent of 'ss' output after it to make sure the data has been only sent on ADDR_1 since ADDR_2 is set as stale.
--- .../testing/selftests/bpf/prog_tests/mptcp.c | 38 +++++++++++++++++++ 1 file changed, 38 insertions(+)
tools/testing/selftests/bpf/prog_tests/mptcp.c +38 -0
@@ -15,6 +15,7 @@#include "mptcp_bpf_rr.skel.h" #include "mptcp_bpf_red.skel.h" #include "mptcp_bpf_burst.skel.h" +#include "mptcp_bpf_stale.skel.h" #define NS_TEST "mptcp_ns"
@@ -651,6 +652,41 @@ static void test_burst(void)mptcp_bpf_burst__destroy (burst_skel); } +static void test_stale (void ) +{ +struct mptcp_bpf_stale *stale_skel; +int server_fd, client_fd; +struct nstoken *nstoken; +struct bpf_link *link; + + stale_skel =mptcp_bpf_stale__open_and_load (); +if (!ASSERT_OK_PTR (stale_skel,"bpf_stale__open_and_load" )) +return ; + + link =bpf_map__attach_struct_ops (stale_skel->maps.stale); +if (!ASSERT_OK_PTR (link,"bpf_map__attach_struct_ops" )) { +mptcp_bpf_stale__destroy (stale_skel); +return ; + } + + nstoken =sched_init ("subflow" ,"bpf_stale" ); +if (!ASSERT_OK_PTR (nstoken,"sched_init:bpf_stale" )) + goto fail; + server_fd =start_mptcp_server (AF_INET ,ADDR_1 ,PORT_1 ,0 ); + client_fd =connect_to_fd (server_fd,0 ); + +send_data (server_fd, client_fd,"bpf_stale" ); +ASSERT_OK (has_bytes_sent (ADDR_1 ),"has_bytes_sent addr_1" ); +ASSERT_GT (has_bytes_sent (ADDR_2 ),0 ,"has_bytes_sent addr_2" ); + +close (client_fd); +close (server_fd); +fail: +cleanup_netns (nstoken); +bpf_link__destroy (link); +mptcp_bpf_stale__destroy (stale_skel); +} +void test_mptcp (void ) {if (test__start_subtest ("base" ))
@@ -669,4 +705,6 @@ void test_mptcp(void)test_red ();if (test__start_subtest ("burst" ))test_burst (); +if (test__start_subtest ("stale" )) +test_stale (); }
Signature · 1 line
2.35.3
Mat Martineau
maintainerQuoted 9 lines · Geliang Tang
v3:
- initsk_storagein .init, delete it in .release.
We invokebpf_sk_storage_get()to get thesk_storagemap many times. Only the first call is the slow path (seebpf_local_storage_lookupinkernel/bpf/bpf_local_storage.c), alloc the map and cache it. The subsequent calls are all in fast path, the cache hits. So we should first callbpf_sk_storage_getin .init, then call it many times in .get_subflow.
- if no subflow is scheduled, it should return '-1' to indicate an error.
Thanks Geliang, patches 1-3 are ok to squash.
For 4-5, see my reply to patch 4.
- Mat
Quoted 31 lines
v2:
- store subflow ids instead of storing subflow pointers insk_storage.v1:
- This patchset adds the newbpf_stalescheduler. Usesk_storageto save the stale map instead of usingsubflow->staleflag to manage it.Geliang Tang (5):
Squash to "selftests/bpf: Addbpf_bkupscheduler" Squash to "selftests/bpf: Addbpf_rrscheduler" Squash to "selftests/bpf: Addbpf_burstscheduler" selftests/bpf: Addbpf_stalescheduler selftests/bpf: Addbpf_staletesttools/testing/selftests/bpf/bpf_tcp_helpers.h | 1 + .../testing/selftests/bpf/prog_tests/mptcp.c | 38 +++++ .../selftests/bpf/progs/mptcp_bpf_bkup.c | 4 +- .../selftests/bpf/progs/mptcp_bpf_burst.c | 2 + .../selftests/bpf/progs/mptcp_bpf_rr.c | 2 + .../selftests/bpf/progs/mptcp_bpf_stale.c | 152 ++++++++++++++++++ 6 files changed, 198 insertions(+), 1 deletion(-)create mode 100644
tools/testing/selftests/bpf/progs/mptcp_bpf_stale.cSignature · 1 line
2.35.3
Hi Geliang, Mat,
Quoted 12 lines · Mat Martineau
Quoted 9 lines · Geliang Tang
v3:
- initsk_storagein .init, delete it in .release.
We invokebpf_sk_storage_get()to get thesk_storagemap many times. Only the first call is the slow path (seebpf_local_storage_lookupinkernel/bpf/bpf_local_storage.c), alloc the map and cache it. The subsequent calls are all in fast path, the cache hits. So we should first callbpf_sk_storage_getin .init, then call it many times in .get_subflow.
- if no subflow is scheduled, it should return '-1' to indicate an error.
Quoted 1 line
Thanks Geliang, patches 1-3 are ok to squash.
Thank you for the patches and your patience!
Now in our tree:
New patches for t/upstream:
- b2afd09b16a1: "squashed" patch 1/5 in "selftests/bpf: Add bpf_bkup scheduler"
- ff989d90d2ea: "squashed" patch 2/5 in "selftests/bpf: Add bpf_rr scheduler"
- ff0afb0f9763: "squashed" patch 3/5 in "selftests/bpf: Add bpf_burst scheduler"
- Results: 86fa15fc9c4f..07a689bb4cbf (export)
Tests are now in progress:
https://cirrus-ci.com/github/multipath-tcp/mptcp_net-next/export/20230909T152638
Cheers,
Matt
Signature · 2 lines
Tessares | Belgium | Hybrid Access Solutions www.tessares.net