mptcp: fix stall because of data_ready
Subsystems (3)
v4 cover mptcp: fix stall because of data_ready
From: Gang Yan <[email protected]>
Changelog:
v4:
- Pass 'delta' to 'move_skbs_to_msk' to fix the potential problems, and add a helper named 'move_skbs_from_backlog' to achieve this.
- Add a test case in mptcp_join.sh which can reproduce such stall prolems.
- Fix a dead_lock problems which is imported in the 1d77124a6322 ('mptcp: fix the stall problems with data_ready.').
v3:
- Replace backlog list with backlog_queue(rb-tree) to solve the stall problems.
v2:
Link: https://patchwork.kernel.org/project/mptcp/patch/[email protected]/ v1: Link: https://patchwork.kernel.org/project/mptcp/cover/[email protected]/
Gang Yan (5):
mptcp: replace backlog_list with backlog_queue
mptcp: fix the stall problems using backlog_queue
mptcp: fix the stall problems with data_ready
mptcp: fix the dead_loack in mptcp_data_ready
selftests: mptcp: test transmission with small rcvbuf
net/mptcp/protocol.c | 112 +++++++++++++++--- net/mptcp/protocol.h | 2 +- .../testing/selftests/net/mptcp/mptcp_join.sh | 20 ++++ 3 files changed, 117 insertions(+), 17 deletions(-)
Signature · 1 line
2.43.0
PATCH v4 1/5 mptcp: replace backlog_list with backlog_queue
From: Gang Yan <[email protected]>
This patch replaces the original list-based backlog_list with a red-black tree (RB-tree) based backlog_queue for MPTCP.
Add key helper functions:
- mptcp_queue_backlog: Insert skb into backlog_queue in order of map_seq via RB-tree
- mptcp_backlog_queue_to_list: Convert RB-tree based backlog_queue to list_head
- mptcp_backlog_list_to_queue: Convert list_head back to RB-tree based backlog_queue
Adapt existing backlog operation logic:
- Update mptcp_can_spool_backlog to splice RB-tree backlog to list via new helper
- Adjust mptcp_backlog_spooled to restore list skbs back to RB-tree backlog_queue
- Modify mptcp_close_ssk and mptcp_recv_skb to check RB-tree emptiness instead of list
- Update mptcp_backlog_purge to use RB-tree to list conversion for backlog cleanup
Furthermore, this patch also initialize the msk->backlog_unaccounted in '__mptcp_init_sock'.
--- net/mptcp/protocol.c | 72 +++++++++++++++++++++++++++++++++++++------- net/mptcp/protocol.h | 2 +- 2 files changed, 62 insertions(+), 12 deletions(-)
net/mptcp/protocol.c +61 -11
@@ -653,6 +653,33 @@ static void mptcp_dss_corruption(struct mptcp_sock *msk, struct sock *ssk) } } +static int mptcp_queue_backlog (struct mptcp_sock *msk,struct sk_buff *skb) +{ +u64 seq =MPTCP_SKB_CB (skb)->map_seq; +struct rb_node **p, *parent =NULL ; + + p = &msk->backlog_queue.rb_node; +if (RB_EMPTY_ROOT (&msk->backlog_queue)) + goto insert; + +while (*p) { +struct sk_buff *s; + + parent = *p; + s =rb_to_skb (parent); + +if (before64 (seq,MPTCP_SKB_CB (s)->map_seq)) + p = &parent->rb_left; +else + p = &parent->rb_right; + } + +insert: +rb_link_node (&skb->rbnode, parent, p); +rb_insert_color (&skb->rbnode, &msk->backlog_queue); +return 0 ; +} +static void __mptcp_add_backlog (struct sock *sk,struct mptcp_subflow_context *subflow,struct sk_buff *skb)
@@ -669,8 +696,8 @@ static void __mptcp_add_backlog(struct sock *sk, }/* Try to coalesce with the last skb in our backlog */ -if (!list_empty (&msk->backlog_list)) - tail = list_last_entry(&msk->backlog_list, struct sk_buff, list); +if (!RB_EMPTY_ROOT (&msk->backlog_queue)) + tail =skb_rb_last (&msk->backlog_queue); if (tail && MPTCP_SKB_CB(skb)->map_seq == MPTCP_SKB_CB(tail)->end_seq && ssk == tail->sk &&
@@ -681,7 +708,7 @@ static void __mptcp_add_backlog(struct sock *sk, goto account; } -list_add_tail (&skb->list, &msk->backlog_list); +mptcp_queue_backlog (msk, skb);mptcp_subflow_lend_fwdmem (subflow, skb); delta = skb->truesize;
@@ -2199,6 +2226,29 @@ static bool __mptcp_move_skbs(struct sock *sk, struct list_head *skbs, u32 *deltreturn moved; } +static void mptcp_backlog_queue_to_list (struct mptcp_sock *msk, +struct list_head *list) +{ +struct sk_buff *skb; + +while ((skb =skb_rb_first (&msk->backlog_queue)) !=NULL ) { +rb_erase (&skb->rbnode, &msk->backlog_queue); +RB_CLEAR_NODE (&skb->rbnode); +list_add_tail (&skb->list, list); + } +} + +static void mptcp_backlog_list_to_queue (struct mptcp_sock *msk, +struct list_head *list) +{ +struct sk_buff *skb, *tmp; + +list_for_each_entry_safe (skb, tmp, list, list) { +list_del (&skb->list); +mptcp_queue_backlog (msk, skb); + } +} +static bool mptcp_can_spool_backlog (struct sock *sk,struct list_head *skbs) {struct mptcp_sock *msk =mptcp_sk (sk);
@@ -2210,12 +2260,12 @@ static bool mptcp_can_spool_backlog(struct sock *sk, struct list_head *skbs) mem_cgroup_from_sk(sk));/* Don't spool the backlog if the rcvbuf is full. */ -if (list_empty (&msk->backlog_list) || +if (RB_EMPTY_ROOT (&msk->backlog_queue) ||sk_rmem_alloc_get (sk) > sk->sk_rcvbuf)return false;INIT_LIST_HEAD (skbs); -list_splice_init (&msk->backlog_list, skbs); +mptcp_backlog_queue_to_list (msk, skbs);return true; }
@@ -2225,7 +2275,7 @@ static void mptcp_backlog_spooled(struct sock *sk, u32 moved,struct mptcp_sock *msk =mptcp_sk (sk);WRITE_ONCE (msk->backlog_len, msk->backlog_len - moved); -list_splice (skbs, &msk->backlog_list); +mptcp_backlog_list_to_queue (msk, skbs); }static bool mptcp_move_skbs (struct sock *sk)
@@ -2311,7 +2361,7 @@ static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, copied += bytes_read; -if (!list_empty (&msk->backlog_list) &&mptcp_move_skbs (sk)) +if (!RB_EMPTY_ROOT (&msk->backlog_queue) &&mptcp_move_skbs (sk))continue ;/* only the MPTCP socket status is relevant here. The exit
@@ -2640,7 +2690,7 @@ void mptcp_close_ssk(struct sock *sk, struct sock *ssk,/* Remove any reference from the backlog to this ssk; backlog skbs consume * space in the msk receive queue, no need to touch sk->sk_rmem_alloc */ -list_for_each_entry (skb, &msk->backlog_list, list) { + skb_rbtree_walk(skb, &msk->backlog_queue) {if (skb->sk != ssk)continue ;
@@ -2896,7 +2946,7 @@ static void mptcp_backlog_purge(struct sock *sk)LIST_HEAD (backlog);mptcp_data_lock (sk); -list_splice_init (&msk->backlog_list, &backlog); +mptcp_backlog_queue_to_list (msk, &backlog); msk->backlog_len =0 ;mptcp_data_unlock (sk);
@@ -2999,7 +3049,7 @@ static void __mptcp_init_sock(struct sock *sk)INIT_LIST_HEAD (&msk->conn_list);INIT_LIST_HEAD (&msk->join_list);INIT_LIST_HEAD (&msk->rtx_queue); -INIT_LIST_HEAD (&msk->backlog_list); + msk->backlog_queue =RB_ROOT ;INIT_WORK (&msk->work, mptcp_worker); msk->out_of_order_queue =RB_ROOT ; msk->first_pending =NULL ;
@@ -4335,7 +4385,7 @@ static struct sk_buff *mptcp_recv_skb(struct sock *sk, u32 *off)struct sk_buff *skb;u32 offset; -if (!list_empty (&msk->backlog_list)) +if (!RB_EMPTY_ROOT (&msk->backlog_queue))mptcp_move_skbs (sk);while ((skb =skb_peek (&sk->sk_receive_queue)) !=NULL ) {
net/mptcp/protocol.h +1 -1
@@ -372,7 +372,7 @@ struct mptcp_sock {* allow_join */ -struct list_head backlog_list;/* protected by the data lock */ +struct rb_root backlog_queue;/* protected by the data lock */ u32 backlog_len;u32 backlog_unaccounted; };
Signature · 1 line
2.43.0
PATCH v4 2/5 mptcp: fix the stall problems using backlog_queue
In reply to Gang YanFrom: Gang Yan <[email protected]>
The original condition would stop moving skbs or spooling backlog even when the receive queue is empty, leading to receive stall.
Modify the condition in __mptcp_move_skbs() and mptcp_can_spool_backlog() to only treat rcvbuf as full when:
sk_rmem_alloc_get(sk) > sk->sk_rcvbuf && !skb_queue_empty(&sk->sk_receive_queue)
This ensures the backlog can still be moved to the receive queue when the queue is empty, avoiding stall problems.
--- net/mptcp/protocol.c | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-)
net/mptcp/protocol.c +8 -2
@@ -2197,6 +2197,12 @@ static void mptcp_rcv_space_adjust(struct mptcp_sock *msk, int copied) msk->rcvq_space.time = mstamp; } +static bool mptcp_can_dequeue_from_backlog (struct sock *sk) +{ +return sk_rmem_alloc_get (sk) <= sk->sk_rcvbuf || +skb_queue_empty (&sk->sk_receive_queue); +} +static bool __mptcp_move_skbs (struct sock *sk,struct list_head *skbs,u32 *delta) { struct sk_buff *skb = list_first_entry(skbs, struct sk_buff, list);
@@ -2206,7 +2212,7 @@ static bool __mptcp_move_skbs(struct sock *sk, struct list_head *skbs, u32 *delt *delta =0 ;while (1 ) {/* If the msk recvbuf is full stop, don't drop */ -if (sk_rmem_alloc_get (sk) > sk->sk_rcvbuf) +if (!mptcp_can_dequeue_from_backlog (sk))break ;prefetch (skb->next);
@@ -2261,7 +2267,7 @@ static bool mptcp_can_spool_backlog(struct sock *sk, struct list_head *skbs)/* Don't spool the backlog if the rcvbuf is full. */ if (RB_EMPTY_ROOT (&msk->backlog_queue) || -sk_rmem_alloc_get (sk) > sk->sk_rcvbuf) + !mptcp_can_dequeue_from_backlog (sk))return false;INIT_LIST_HEAD (skbs);
Signature · 1 line
2.43.0
PATCH v4 3/5 mptcp: fix the stall problems with data_ready
In reply to Gang YanFrom: Gang Yan <[email protected]>
There exists a stall caused by unprocessed backlog_queue in 'move_skbs_to_msk'.
This patch adds a check for backlog_queue and move skbs to receive queue when no skbs were moved from subflow but backlog_queue is not empty.
--- net/mptcp/protocol.c | 22 +++++++++++++++++++++- 1 file changed, 21 insertions(+), 1 deletion(-)
net/mptcp/protocol.c +21 -1
@@ -49,6 +49,11 @@ static struct percpu_counter mptcp_sockets_allocated ____cacheline_aligned_in_smstatic void __mptcp_destroy_sock (struct sock *sk);static void mptcp_check_send_data_fin (struct sock *sk); +static bool mptcp_can_spool_backlog (struct sock *sk,struct list_head *skbs); +static void mptcp_backlog_spooled (struct sock *sk,u32 moved, +struct list_head *skbs); +static bool __mptcp_move_skbs (struct sock *sk,struct list_head *skbs, +u32 *delta); DEFINE_PER_CPU(struct mptcp_delegated_action, mptcp_delegated_actions) = { .bh_lock =INIT_LOCAL_LOCK (bh_lock),
@@ -906,6 +911,19 @@ static bool move_skbs_to_msk(struct mptcp_sock *msk, struct sock *ssk)return moved; } +static bool move_skbs_from_backlog (struct sock *sk) +{ +struct list_head skbs; +bool enqueued = false; +u32 moved; + +while (mptcp_can_spool_backlog (sk, &skbs)) { + enqueued |=__mptcp_move_skbs (sk, &skbs, &moved); +mptcp_backlog_spooled (sk, moved, &skbs); + } +return enqueued; +} +static void mptcp_rcv_rtt_update (struct mptcp_sock *msk,struct mptcp_subflow_context *subflow) {
@@ -948,7 +966,9 @@ void mptcp_data_ready(struct sock *sk, struct sock *ssk)mptcp_rcv_rtt_update (msk, subflow);if (!sock_owned_by_user (sk)) {/* Wake-up the reader only for in-sequence data */ -if (move_skbs_to_msk (msk, ssk) &&mptcp_epollin_ready (sk)) +if ((move_skbs_to_msk (msk, ssk) || +move_skbs_from_backlog (sk)) && +mptcp_epollin_ready (sk)) sk->sk_data_ready (sk); }else {__mptcp_move_skbs_from_subflow (msk, ssk, false);
Signature · 1 line
2.43.0
PATCH v4 4/5 mptcp: fix the dead_loack in mptcp_data_ready
In reply to Gang YanFrom: Gang Yan <[email protected]>
This patch defers mptcp_check_data_fin from __mptcp_move_skbs to avoid deadlock.
When processing backlogged data in softirq context, __mptcp_move_skbs directly calls mptcp_check_data_fin, which can lead to a deadlock if the msk socket is in TCP_FIN_WAIT2 state. In that case, mptcp_check_data_fin calls mptcp_shutdown_subflows, which attempts to lock the subflow socket with lock_sock_fast() - a sleeping function that cannot be called in atomic context (softirq).
The correct pattern is already used in move_skbs_to_msk: if a data fin is pending, schedule the work to be processed later in process context rather than handling it directly.
--- net/mptcp/protocol.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
net/mptcp/protocol.c +6 -2
@@ -2247,8 +2247,12 @@ static bool __mptcp_move_skbs(struct sock *sk, struct list_head *skbs, u32 *delt }__mptcp_ofo_queue (msk); -if (moved) -mptcp_check_data_fin ((struct sock *)msk); +if (moved) { +if (mptcp_pending_data_fin (sk,NULL )) +mptcp_schedule_work (sk); +else +mptcp_check_data_fin (sk); + }return moved; }
Signature · 1 line
2.43.0
PATCH v4 5/5 selftests: mptcp: test transmission with small rcvbuf
In reply to Gang YanFrom: Gang Yan <[email protected]>
This patch adds a test for transmission with a limited receive buffer.
Without the fixes in this set, this test fails with errors:
# 001 4 subflows with reduced rcvbuf
# net.ipv4.tcp_rmem = 4096 4096 4096 # Info: Test file (size 1024 KB) for client # Info: Test file (size 1024 KB) for server # copyfd_io_poll: poll timed out (events: POLLIN 1, POLLOUT 0) # copyfd_io_poll: poll timed out (events: POLLIN 0, POLLOUT 4) # [FAIL] client exit code 2, server 2 # Server ns stats (ns2-qg1Hqo)
--- .../testing/selftests/net/mptcp/mptcp_join.sh | 20 +++++++++++++++++++ 1 file changed, 20 insertions(+)
tools/testing/selftests/net/mptcp/mptcp_join.sh +20 -0
@@ -4417,6 +4417,25 @@ endpoint_tests() fi } +rcvbuf_tests() +{ + if reset "4 subflows with reduced rcvbuf"; then + local tcp_rmem=$(ip netns exec $ns2 sysctl -n net.ipv4.tcp_rmem) + ip netns exec $ns2 sysctl -w net.ipv4.tcp_rmem="4096 4096 4096" + + pm_nl_set_limits $ns1 0 4 + pm_nl_set_limits $ns2 0 4 + pm_nl_add_endpoint $ns2 10.0.2.2 flags subflow + pm_nl_add_endpoint $ns2 10.0.3.2 flags subflow + pm_nl_add_endpoint $ns2 10.0.4.2 flags subflow + speed=fast test_linkfail=1024 \ + run_tests $ns1 $ns2 10.0.1.1 + chk_join_nr 3 3 3 + + ip netns exec $ns2 sysctl -w net.ipv4.tcp_rmem="$tcp_rmem" + fi +} + # [$1: error message] usage() {
@@ -4467,6 +4486,7 @@ all_tests_sorted=( F@fail_tests u@userspace_tests I@endpoint_tests + R@rcvbuf_tests ) all_tests_args=""
Signature · 1 line
2.43.0
Hi Gang,
Thank you for your modifications, that's great!
Our CI did some validations and here is its report:
- KVM Validation: normal (except selftest_mptcp_join): Success! ✅
- KVM Validation: normal (only selftest_mptcp_join): Success! ✅
- KVM Validation: debug (except selftest_mptcp_join): Unstable: 2 failed test(s): packetdrill_dss packetdrill_fastopen 🔴
- KVM Validation: debug (only selftest_mptcp_join): Success! ✅
- KVM Validation: btf-normal (only bpftest_all): Success! ✅
- KVM Validation: btf-debug (only bpftest_all): Success! ✅
- Task: https://github.com/multipath-tcp/mptcp_net-next/actions/runs/23729981055
Initiator: Patchew Applier
Commits: https://github.com/multipath-tcp/mptcp_net-next/commits/723fa97ad85d
Patchwork: https://patchwork.kernel.org/project/mptcp/list/?series=1074352
If there are some issues, you can reproduce them using the same environment as the one used by the CI thanks to a docker image, e.g.:
$ cd [kernel source code]
$ docker run -v "${PWD}:${PWD}:rw" -w "${PWD}" --privileged --rm -it \
--pull always mptcp/mptcp-upstream-virtme-docker:latest \
auto-normalFor more details:
https://github.com/multipath-tcp/mptcp-upstream-virtme-docker
Please note that despite all the efforts that have been already done to have a stable tests suite when executed on a public CI like here, it is possible some reported issues are not due to your modifications. Still, do not hesitate to help us improve that ;-)
Cheers,
MPTCP GH Action bot
Bot operated by Matthieu Baerts (NGI0 Core)