[PATCH bpf v3 1/2] bpf, sockmap: settle sk_forward_alloc for strparser SK_PASS
From: Junseo Lim <hidden>
Date: 2026-08-17 15:50:23
Also in:
bpf, lkml
Subsystem:
bpf [l7 framework] (sockmap), networking [general], the rest · Maintainers:
John Fastabend, Jakub Sitnicki, Jiayuan Chen, "David S. Miller", Eric Dumazet, Jakub Kicinski, Paolo Abeni, Linus Torvalds
The strparser SK_PASS path can queue cloned skbs back to the same
socket. A single TCP receive skb may be split into multiple strparser
messages. The strparser clones are unowned, but keep the original
truesize.
sk_psock_skb_ingress_self() assigns receive ownership with
skb_set_owner_r(). That charges each clone to the socket. When this is
repeated for strparser clones, sk_forward_alloc can already be in
deficit before the next owner assignment. Releasing the queued skbs can
then uncharge more memcg pages than were reserved and trigger a
page_counter underflow.
Fix by making same-socket ingress preserve existing receive
ownership and only assign ownership to unowned self-pass skbs. For
strparser clones, use a zero-sized sk_rmem_schedule() before
skb_set_owner_r() to settle any sk_forward_alloc deficit without
reserving the skb's full truesize again.
When the skb is retried from the psock backlog, preserve the original
_sk_redir value across skb_bpf_redirect_clear() so the deferred path
keeps the same ingress and strparser state. Perform the deferred owner
assignment under the socket lock because psock backlog work only holds
psock->work_mutex.
Fixes: 144748eb0c44 ("bpf, sockmap: Fix incorrect fwd_alloc accounting")
Reported-by: Sechang Lim <redacted>
Suggested-by: Emil Tsalapatis <emil@etsalapatis.com>
Assisted-by: Codex:gpt-5.5
Signed-off-by: Junseo Lim <redacted>
---
net/core/skmsg.c | 125 +++++++++++++++++++++++++++++++++++++++--------
1 file changed, 105 insertions(+), 20 deletions(-)
diff --git a/net/core/skmsg.c b/net/core/skmsg.c
index 2521b643fa05..347cb168f749 100644
--- a/net/core/skmsg.c
+++ b/net/core/skmsg.c@@ -586,21 +586,24 @@ static int sk_psock_skb_ingress_enqueue(struct sk_buff *skb, } static int sk_psock_skb_ingress_self(struct sk_psock *psock, struct sk_buff *skb, - u32 off, u32 len, bool take_ref); + u32 off, u32 len, bool take_ref, + bool settle_fwd_alloc); +static int sk_psock_skb_ingress_self_backlog(struct sk_psock *psock, + struct sk_buff *skb, + u32 off, u32 len, bool take_ref, + bool settle_fwd_alloc); static int sk_psock_skb_ingress(struct sk_psock *psock, struct sk_buff *skb, - u32 off, u32 len) + u32 off, u32 len, bool settle_fwd_alloc) { struct sock *sk = psock->sk; struct sk_msg *msg; int err; - /* If we are receiving on the same sock skb->sk is already assigned, - * skip memory accounting and owner transition seeing it already set - * correctly. - */ if (unlikely(skb->sk == sk)) - return sk_psock_skb_ingress_self(psock, skb, off, len, true); + return sk_psock_skb_ingress_self_backlog(psock, skb, off, + len, true, + settle_fwd_alloc); msg = sk_psock_create_ingress_msg(sk, skb); if (!msg) return -EAGAIN;
@@ -618,34 +621,100 @@ static int sk_psock_skb_ingress(struct sk_psock *psock, struct sk_buff *skb, return err; } -/* Puts an skb on the ingress queue of the socket already assigned to the - * skb. In this case we do not need to check memory limits or skb_set_owner_r - * because the skb is already accounted for here. +static int sk_psock_skb_ingress_self_assign(struct sock *sk, + struct sk_buff *skb, + bool settle_fwd_alloc) +{ + /* Leave skbs already receive-accounted to sk untouched. */ + if (skb->sk == sk && skb->destructor == sock_rfree) + return 0; + + if (settle_fwd_alloc) { + sock_owned_by_me(sk); + + if (!sk_rmem_schedule(sk, skb, 0)) + return -EAGAIN; + } + + skb_set_owner_r(skb, sk); + return 0; +} + +/* Puts an skb on the ingress queue for psock->sk. + * + * If the skb already has receive ownership for this socket, leave socket + * memory accounting untouched. Otherwise, before assigning receive ownership + * to an unowned strparser SK_PASS skb, settle any existing sk_forward_alloc + * deficit from earlier clone charges. */ static int sk_psock_skb_ingress_self(struct sk_psock *psock, struct sk_buff *skb, - u32 off, u32 len, bool take_ref) + u32 off, u32 len, bool take_ref, + bool settle_fwd_alloc) { + struct sock *sk = psock->sk; struct sk_msg *msg = alloc_sk_msg(GFP_ATOMIC); + int err; + + if (unlikely(!msg)) + return -EAGAIN; + + err = sk_psock_skb_ingress_self_assign(sk, skb, settle_fwd_alloc); + if (err) + goto free; + + /* This is used in tcp_bpf_recvmsg_parser() to determine whether the + * data originates from the socket's own protocol stack. No need to + * refcount sk because msg's lifetime is bound to sk via the ingress_msg. + */ + msg->sk = sk; + err = sk_psock_skb_ingress_enqueue(skb, off, len, psock, sk, msg, + take_ref); + if (err < 0) + goto free; + + return err; +free: + kfree(msg); + return err; +} + +static int sk_psock_skb_ingress_self_backlog(struct sk_psock *psock, + struct sk_buff *skb, + u32 off, u32 len, bool take_ref, + bool settle_fwd_alloc) +{ struct sock *sk = psock->sk; + struct sk_msg *msg = alloc_sk_msg(GFP_ATOMIC); int err; if (unlikely(!msg)) return -EAGAIN; - skb_set_owner_r(skb, sk); + + lock_sock(sk); + err = sk_psock_skb_ingress_self_assign(sk, skb, settle_fwd_alloc); + release_sock(sk); + if (err) + goto free; /* This is used in tcp_bpf_recvmsg_parser() to determine whether the * data originates from the socket's own protocol stack. No need to * refcount sk because msg's lifetime is bound to sk via the ingress_msg. */ msg->sk = sk; - err = sk_psock_skb_ingress_enqueue(skb, off, len, psock, sk, msg, take_ref); + err = sk_psock_skb_ingress_enqueue(skb, off, len, psock, sk, msg, + take_ref); if (err < 0) - kfree(msg); + goto free; + + return err; +free: + kfree(msg); return err; } static int sk_psock_handle_skb(struct sk_psock *psock, struct sk_buff *skb, - u32 off, u32 len, bool ingress) + u32 off, u32 len, bool ingress, + bool self_pass, bool strparser) { if (!ingress) { if (!sock_writeable(psock->sk))
@@ -653,7 +722,11 @@ static int sk_psock_handle_skb(struct sk_psock *psock, struct sk_buff *skb, return skb_send_sock(psock->sk, skb, off, len); } - return sk_psock_skb_ingress(psock, skb, off, len); + if (self_pass) + return sk_psock_skb_ingress_self_backlog(psock, skb, off, + len, true, strparser); + + return sk_psock_skb_ingress(psock, skb, off, len, strparser); } static void sk_psock_skb_state(struct sk_psock *psock,
@@ -694,9 +767,14 @@ static void sk_psock_backlog(struct work_struct *work) return; mutex_lock(&psock->work_mutex); while ((skb = skb_peek(&psock->ingress_skb))) { + unsigned long saved_redir; + bool strparser; + bool self_pass; + len = skb->len; off = 0; - if (skb_bpf_strparser(skb)) { + strparser = skb_bpf_strparser(skb); + if (strparser) { struct strp_msg *stm = strp_msg(skb); off = stm->offset;
@@ -710,17 +788,20 @@ static void sk_psock_backlog(struct work_struct *work) } ingress = skb_bpf_ingress(skb); + self_pass = ingress && !skb_bpf_redirect_fetch(skb); + saved_redir = skb->_sk_redir; skb_bpf_redirect_clear(skb); do { ret = -EIO; if (!sock_flag(psock->sk, SOCK_DEAD)) ret = sk_psock_handle_skb(psock, skb, off, - len, ingress); + len, ingress, + self_pass, strparser); if (ret <= 0) { if (ret == -EAGAIN) { sk_psock_skb_state(psock, state, len, off); /* Restore redir info we cleared before */ - skb_bpf_set_redir(skb, psock->sk, ingress); + skb->_sk_redir = saved_redir; /* Delay slightly to prioritize any * other work that might be here. */
@@ -1017,6 +1098,8 @@ static int sk_psock_verdict_apply(struct sk_psock *psock, struct sk_buff *skb, * retrying later from workqueue. */ if (skb_queue_empty(&psock->ingress_skb)) { + bool settle_fwd_alloc = false; + len = skb->len; off = 0; if (skb_bpf_strparser(skb)) {
@@ -1024,8 +1107,10 @@ static int sk_psock_verdict_apply(struct sk_psock *psock, struct sk_buff *skb, off = stm->offset; len = stm->full_len; + settle_fwd_alloc = true; } - err = sk_psock_skb_ingress_self(psock, skb, off, len, false); + err = sk_psock_skb_ingress_self(psock, skb, off, len, + false, settle_fwd_alloc); } if (err < 0) { spin_lock_bh(&psock->ingress_lock);
--
2.55.0