sock_map: some bug fixes and improvements
Subsystems (8)
v3 cover sock_map: some bug fixes and improvements
From: Cong Wang <[email protected]>
This patchset contains a few bug fixes and improvements for sock_map.
Patch 1 improves recvmsg() accuracy for UDP, patch 2 improves UDP non-blocking read() by retrying on EAGAIN. With both of them, the failure rate of the UDP test case goes down from 10% to 1%.
Patch 3 is memory leak fix I posted, no change since v1. The rest patches address similar memory leaks or improve error handling, including one increases sk_drops counter for error cases. Please check each patch description for more details.
---
v3: add another bug fix as patch 4
update patch 5 accordingly
address John's review on the last patch
fix a few typos in patch descriptions
v2: group all patches together
set max for retries of EAGAIN
Cong Wang (8):
skmsg: improve udp_bpf_recvmsg() accuracy selftests/bpf: Retry for EAGAIN in udp_redir_to_connected()
udp: fix a memory leak in udp_read_sock()
skmsg: clear skb redirect pointer before dropping it
skmsg: fix a memory leak in sk_psock_verdict_apply()
skmsg: teach sk_psock_verdict_apply() to return errors
skmsg: pass source psock to sk_psock_skb_redirect()
skmsg: increase sk->sk_drops when dropping packets
include/linux/skmsg.h | 2 - net/core/skmsg.c | 82 +++++++++---------- net/ipv4/tcp_bpf.c | 24 +++++- net/ipv4/udp.c | 2 + net/ipv4/udp_bpf.c | 47 +++++++++-- .../selftests/bpf/prog_tests/sockmap_listen.c | 7 +- 6 files changed, 112 insertions(+), 52 deletions(-)
Signature · 1 line
2.25.1
PATCH v3 1/8 skmsg: improve udp_bpf_recvmsg() accuracy
From: Cong Wang <[email protected]>
I tried to reuse sk_msg_wait_data() for different protocols, but it turns out it can not be simply reused. For example, UDP actually uses two queues to receive skb:
udp_sk(sk)->reader_queue and sk->sk_receive_queue. So we have to check both of them to know whether we have received any packet.
Also, UDP does not lock the sock during BH Rx path, it makes no sense for its ->recvmsg() to lock the sock. It is always possible for ->recvmsg() to be called before packets actually arrive in the receive queue, we just use best effort to make it accurate here.
--- include/linux/skmsg.h | 2 -- net/core/skmsg.c | 23 --------------------- net/ipv4/tcp_bpf.c | 24 +++++++++++++++++++++- net/ipv4/udp_bpf.c | 47 ++++++++++++++++++++++++++++++++++++++----- 4 files changed, 65 insertions(+), 31 deletions(-)
include/linux/skmsg.h +0 -2
@@ -126,8 +126,6 @@ int sk_msg_zerocopy_from_iter(struct sock *sk, struct iov_iter *from, struct sk_msg *msg, u32 bytes);int sk_msg_memcopy_from_iter (struct sock *sk,struct iov_iter *from,struct sk_msg *msg,u32 bytes); -int sk_msg_wait_data (struct sock *sk,struct sk_psock *psock,int flags, -long timeo,int *err);int sk_msg_recvmsg (struct sock *sk,struct sk_psock *psock,struct msghdr *msg,int len,int flags);
net/core/skmsg.c +0 -23
@@ -399,29 +399,6 @@ int sk_msg_memcopy_from_iter(struct sock *sk, struct iov_iter *from, }EXPORT_SYMBOL_GPL (sk_msg_memcopy_from_iter); -int sk_msg_wait_data (struct sock *sk,struct sk_psock *psock,int flags, -long timeo,int *err) -{ -DEFINE_WAIT_FUNC (wait, woken_wake_function); -int ret =0 ; - -if (sk->sk_shutdown &RCV_SHUTDOWN ) -return 1 ; - -if (!timeo) -return ret; - -add_wait_queue (sk_sleep (sk), &wait); -sk_set_bit (SOCKWQ_ASYNC_WAITDATA , sk); - ret =sk_wait_event (sk, &timeo, - !list_empty (&psock->ingress_msg) || - !skb_queue_empty (&sk->sk_receive_queue), &wait); -sk_clear_bit (SOCKWQ_ASYNC_WAITDATA , sk); -remove_wait_queue (sk_sleep (sk), &wait); -return ret; -} -EXPORT_SYMBOL_GPL (sk_msg_wait_data); -/* Receive sk_msg from psock->ingress_msg to @msg. */ int sk_msg_recvmsg (struct sock *sk,struct sk_psock *psock,struct msghdr *msg,int len,int flags)
net/ipv4/tcp_bpf.c +23 -1
@@ -163,6 +163,28 @@ static bool tcp_bpf_stream_read(const struct sock *sk)return !empty; } +static int tcp_msg_wait_data (struct sock *sk,struct sk_psock *psock,int flags, +long timeo,int *err) +{ +DEFINE_WAIT_FUNC (wait, woken_wake_function); +int ret =0 ; + +if (sk->sk_shutdown &RCV_SHUTDOWN ) +return 1 ; + +if (!timeo) +return ret; + +add_wait_queue (sk_sleep (sk), &wait); +sk_set_bit (SOCKWQ_ASYNC_WAITDATA , sk); + ret =sk_wait_event (sk, &timeo, + !list_empty (&psock->ingress_msg) || + !skb_queue_empty (&sk->sk_receive_queue), &wait); +sk_clear_bit (SOCKWQ_ASYNC_WAITDATA , sk); +remove_wait_queue (sk_sleep (sk), &wait); +return ret; +} +static int tcp_bpf_recvmsg (struct sock *sk,struct msghdr *msg,size_t len,int nonblock,int flags,int *addr_len) {
@@ -188,7 +210,7 @@ static int tcp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,long timeo; timeo =sock_rcvtimeo (sk, nonblock); - data =sk_msg_wait_data (sk, psock, flags, timeo, &err); + data =tcp_msg_wait_data (sk, psock, flags, timeo, &err);if (data) {if (!sk_psock_queue_empty (psock)) goto msg_bytes_ready;
net/ipv4/udp_bpf.c +42 -5
72 lines not on this page · raw message
Signature · 1 line
2.25.1
PATCH v3 2/8 selftests/bpf: Retry for EAGAIN in udp_redir_to_connected()
In reply to Cong WangFrom: Cong Wang <[email protected]>
We use non-blocking sockets for testing sockmap redirections, and got some random EAGAIN errors from UDP tests.
There is no guarantee the packet would be immediately available to receive as soon as it is sent out, even on the local host. For UDP, this is especially true because it does not lock the sock during BH (unlike the TCP path). This is probably why we only saw this error in UDP cases.
No matter how hard we try to make the queue empty check accurate, it is always possible for recvmsg() to beat ->sk_data_ready(). Therefore, we should just retry in case of EAGAIN.
--- tools/testing/selftests/bpf/prog_tests/sockmap_listen.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)
tools/testing/selftests/bpf/prog_tests/sockmap_listen.c +6 -1
@@ -1610,6 +1610,7 @@ static void udp_redir_to_connected(int family, int sotype, int sock_mapfd,struct sockaddr_storage addr;int c0, c1, p0, p1;unsigned int pass; +int retries =100 ;socklen_t len;int err, n;u64 value;
@@ -1686,9 +1687,13 @@ static void udp_redir_to_connected(int family, int sotype, int sock_mapfd,if (pass !=1 )FAIL ("%s: want pass count 1, have %d" , log_prefix, pass); +again: n =read (mode ==REDIR_INGRESS ? p0 : c0, &b,1 ); -if (n <0 ) +if (n <0 ) { +if (errno ==EAGAIN && retries--) + goto again;FAIL_ERRNO ("%s: read" , log_prefix); + }if (n ==0 )FAIL ("%s: incomplete read" , log_prefix);
Signature · 1 line
2.25.1
PATCH v3 3/8 udp: fix a memory leak in udp_read_sock()
In reply to Cong WangFrom: Cong Wang <[email protected]>
sk_psock_verdict_recv() clones the skb and uses the clone afterward, so udp_read_sock() should free the skb after using it, regardless of error or not.
This fixes a real kmemleak.
--- net/ipv4/udp.c | 2 ++ 1 file changed, 2 insertions(+)
net/ipv4/udp.c +2 -0
@@ -1798,11 +1798,13 @@ int udp_read_sock(struct sock *sk, read_descriptor_t *desc,if (used <=0 ) {if (!copied) copied = used; +kfree_skb (skb);break ; }else if (used <= skb->len) { copied += used; } +kfree_skb (skb);if (!desc->count)break ; }
Signature · 1 line
2.25.1
PATCH v3 4/8 skmsg: clear skb redirect pointer before dropping it
In reply to Cong WangFrom: Cong Wang <[email protected]>
When we drop skb inside sk_psock_skb_redirect(), we have to clear its skb->_sk_redir pointer too, otherwise kfree_skb() would misinterpret it as a valid skb->_skb_refdst and dst_release() would eventually complain.
--- net/core/skmsg.c | 2 ++ 1 file changed, 2 insertions(+)
net/core/skmsg.c +2 -0
@@ -843,12 +843,14 @@ static void sk_psock_skb_redirect(struct sk_buff *skb)* a socket that is in this state so we drop the skb. */ if (!psock_other ||sock_flag (sk_other,SOCK_DEAD )) { +skb_bpf_redirect_clear (skb);kfree_skb (skb);return ; }spin_lock_bh (&psock_other->ingress_lock);if (!sk_psock_test_state (psock_other,SK_PSOCK_TX_ENABLED )) {spin_unlock_bh (&psock_other->ingress_lock); +skb_bpf_redirect_clear (skb);kfree_skb (skb);return ; }
Signature · 1 line
2.25.1
PATCH v3 5/8 skmsg: fix a memory leak in sk_psock_verdict_apply()
In reply to Cong WangFrom: Cong Wang <[email protected]>
If the dest psock does not set SK_PSOCK_TX_ENABLED, the skb can't be queued anywhere so must be dropped.
This one is found during code review.
--- net/core/skmsg.c | 5 +++++ 1 file changed, 5 insertions(+)
net/core/skmsg.c +5 -0
@@ -924,8 +924,13 @@ static void sk_psock_verdict_apply(struct sk_psock *psock,if (sk_psock_test_state (psock,SK_PSOCK_TX_ENABLED )) {skb_queue_tail (&psock->ingress_skb, skb);schedule_work (&psock->work); + err =0 ; }spin_unlock_bh (&psock->ingress_lock); +if (err <0 ) { +skb_bpf_redirect_clear (skb); + goto out_free; + } }break ;case __SK_REDIRECT:
Signature · 1 line
2.25.1
PATCH v3 6/8 skmsg: teach sk_psock_verdict_apply() to return errors
In reply to Cong WangFrom: Cong Wang <[email protected]>
Currently sk_psock_verdict_apply() is void, but it handles some error conditions too. Its caller is impossible to learn whether it succeeds or fails, especially sk_psock_verdict_recv().
Make it return int to indicate error cases and propagate errors to callers properly.
--- net/core/skmsg.c | 23 ++++++++++++++--------- 1 file changed, 14 insertions(+), 9 deletions(-)
net/core/skmsg.c +14 -9
@@ -824,7 +824,7 @@ int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock, }EXPORT_SYMBOL_GPL (sk_psock_msg_verdict); -static void sk_psock_skb_redirect (struct sk_buff *skb) +static int sk_psock_skb_redirect (struct sk_buff *skb) {struct sk_psock *psock_other;struct sock *sk_other;
@@ -835,7 +835,7 @@ static void sk_psock_skb_redirect(struct sk_buff *skb)*/ if (unlikely (!sk_other)) {kfree_skb (skb); -return ; +return -EIO ; } psock_other =sk_psock (sk_other);/* This error indicates the socket is being torn down or had another
@@ -845,19 +845,20 @@ static void sk_psock_skb_redirect(struct sk_buff *skb)if (!psock_other ||sock_flag (sk_other,SOCK_DEAD )) {skb_bpf_redirect_clear (skb);kfree_skb (skb); -return ; +return -EIO ; }spin_lock_bh (&psock_other->ingress_lock);if (!sk_psock_test_state (psock_other,SK_PSOCK_TX_ENABLED )) {spin_unlock_bh (&psock_other->ingress_lock);skb_bpf_redirect_clear (skb);kfree_skb (skb); -return ; +return -EIO ; }skb_queue_tail (&psock_other->ingress_skb, skb);schedule_work (&psock_other->work);spin_unlock_bh (&psock_other->ingress_lock); +return 0 ; }static void sk_psock_tls_verdict_apply (struct sk_buff *skb,struct sock *sk,int verdict)
@@ -894,14 +895,15 @@ int sk_psock_tls_strp_read(struct sk_psock *psock, struct sk_buff *skb) }EXPORT_SYMBOL_GPL (sk_psock_tls_strp_read); -static void sk_psock_verdict_apply (struct sk_psock *psock, -struct sk_buff *skb,int verdict) +static int sk_psock_verdict_apply (struct sk_psock *psock,struct sk_buff *skb, +int verdict) {struct sock *sk_other; -int err = -EIO ; +int err =0 ;switch (verdict) {case __SK_PASS: + err = -EIO ; sk_other = psock->sk;if (sock_flag (sk_other,SOCK_DEAD ) || !sk_psock_test_state (psock,SK_PSOCK_TX_ENABLED )) {
@@ -934,13 +936,15 @@ static void sk_psock_verdict_apply(struct sk_psock *psock, }break ;case __SK_REDIRECT: -sk_psock_skb_redirect (skb); + err =sk_psock_skb_redirect (skb);break ;case __SK_DROP:default : out_free:kfree_skb (skb); } + +return err; }static void sk_psock_write_space (struct sock *sk)
@@ -1107,7 +1111,8 @@ static int sk_psock_verdict_recv(read_descriptor_t *desc, struct sk_buff *skb, ret =sk_psock_map_verd (ret,skb_bpf_redirect_fetch (skb)); skb->sk =NULL ; } -sk_psock_verdict_apply (psock, skb, ret); +if (sk_psock_verdict_apply (psock, skb, ret) <0 ) + len =0 ; out:rcu_read_unlock ();return len;
Signature · 1 line
2.25.1
PATCH v3 7/8 skmsg: pass source psock to sk_psock_skb_redirect()
In reply to Cong WangFrom: Cong Wang <[email protected]>
sk_psock_skb_redirect() only takes skb as a parameter, we will need to know where this skb is from, so just pass the source psock to this function as a new parameter. This patch prepares for the next one.
--- net/core/skmsg.c | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-)
net/core/skmsg.c +6 -5
@@ -824,7 +824,7 @@ int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock, }EXPORT_SYMBOL_GPL (sk_psock_msg_verdict); -static int sk_psock_skb_redirect (struct sk_buff *skb) +static int sk_psock_skb_redirect (struct sk_psock *from,struct sk_buff *skb) {struct sk_psock *psock_other;struct sock *sk_other;
@@ -861,11 +861,12 @@ static int sk_psock_skb_redirect(struct sk_buff *skb)return 0 ; } -static void sk_psock_tls_verdict_apply (struct sk_buff *skb,struct sock *sk,int verdict) +static void sk_psock_tls_verdict_apply (struct sk_buff *skb, +struct sk_psock *from,int verdict) {switch (verdict) {case __SK_REDIRECT: -sk_psock_skb_redirect (skb); +sk_psock_skb_redirect (from, skb);break ;case __SK_PASS:case __SK_DROP:
@@ -889,7 +890,7 @@ int sk_psock_tls_strp_read(struct sk_psock *psock, struct sk_buff *skb) ret =sk_psock_map_verd (ret,skb_bpf_redirect_fetch (skb)); skb->sk =NULL ; } -sk_psock_tls_verdict_apply (skb, psock->sk, ret); +sk_psock_tls_verdict_apply (skb, psock, ret);rcu_read_unlock ();return ret; }
@@ -936,7 +937,7 @@ static int sk_psock_verdict_apply(struct sk_psock *psock, struct sk_buff *skb, }break ;case __SK_REDIRECT: - err =sk_psock_skb_redirect (skb); + err =sk_psock_skb_redirect (psock, skb);break ;case __SK_DROP:default :
Signature · 1 line
2.25.1
PATCH v3 8/8 skmsg: increase sk->sk_drops when dropping packets
In reply to Cong WangFrom: Cong Wang <[email protected]>
It is hard to observe packet drops without increasing relevant drop counters, here we should increase sk->sk_drops which is a protocol-independent counter. Fortunately psock is always associated with a struct sock, we can just use psock->sk.
--- net/core/skmsg.c | 22 ++++++++++++++-------- 1 file changed, 14 insertions(+), 8 deletions(-)
net/core/skmsg.c +14 -8
@@ -578,6 +578,12 @@ static int sk_psock_handle_skb(struct sk_psock *psock, struct sk_buff *skb,return sk_psock_skb_ingress (psock, skb); } +static void sock_drop (struct sock *sk,struct sk_buff *skb) +{ +sk_drops_add (sk, skb); +kfree_skb (skb); +} +static void sk_psock_backlog (struct work_struct *work) { struct sk_psock *psock = container_of(work, struct sk_psock, work);
@@ -617,7 +623,7 @@ static void sk_psock_backlog(struct work_struct *work)/* Hard errors break pipe and stop xmit. */ sk_psock_report_error (psock, ret ? -ret :EPIPE );sk_psock_clear_state (psock,SK_PSOCK_TX_ENABLED ); -kfree_skb (skb); +sock_drop (psock->sk, skb); goto end; } off += ret;
@@ -708,7 +714,7 @@ static void __sk_psock_zap_ingress(struct sk_psock *psock)while ((skb =skb_dequeue (&psock->ingress_skb)) !=NULL ) {skb_bpf_redirect_clear (skb); -kfree_skb (skb); +sock_drop (psock->sk, skb); }__sk_psock_purge_ingress_msg (psock); }
@@ -834,7 +840,7 @@ static int sk_psock_skb_redirect(struct sk_psock *from, struct sk_buff *skb)* return code, but then didn't set a redirect interface. */ if (unlikely (!sk_other)) { -kfree_skb (skb); +sock_drop (from->sk, skb);return -EIO ; } psock_other =sk_psock (sk_other);
@@ -844,14 +850,14 @@ static int sk_psock_skb_redirect(struct sk_psock *from, struct sk_buff *skb)*/ if (!psock_other ||sock_flag (sk_other,SOCK_DEAD )) {skb_bpf_redirect_clear (skb); -kfree_skb (skb); +sock_drop (from->sk, skb);return -EIO ; }spin_lock_bh (&psock_other->ingress_lock);if (!sk_psock_test_state (psock_other,SK_PSOCK_TX_ENABLED )) {spin_unlock_bh (&psock_other->ingress_lock);skb_bpf_redirect_clear (skb); -kfree_skb (skb); +sock_drop (from->sk, skb);return -EIO ; }
@@ -942,7 +948,7 @@ static int sk_psock_verdict_apply(struct sk_psock *psock, struct sk_buff *skb,case __SK_DROP:default : out_free: -kfree_skb (skb); +sock_drop (psock->sk, skb); }return err;
@@ -977,7 +983,7 @@ static void sk_psock_strp_read(struct strparser *strp, struct sk_buff *skb) sk = strp->sk; psock =sk_psock (sk);if (unlikely (!psock)) { -kfree_skb (skb); +sock_drop (sk, skb); goto out; } prog =READ_ONCE (psock->progs.stream_verdict);
@@ -1098,7 +1104,7 @@ static int sk_psock_verdict_recv(read_descriptor_t *desc, struct sk_buff *skb, psock =sk_psock (sk);if (unlikely (!psock)) { len =0 ; -kfree_skb (skb); +sock_drop (sk, skb); goto out; } prog =READ_ONCE (psock->progs.stream_verdict);
Signature · 1 line
2.25.1
John Fastabend
maintainerRE: [Patch bpf v3 8/8] skmsg: increase sk->sk_drops when dropping packets
Quoted 16 lines · Cong Wang
From: Cong Wang <[email protected]>
It is hard to observe packet drops without increasing relevant drop counters, here we should increase
sk->sk_dropswhich is a protocol-independent counter. Fortunately psock is always associated with a struct sock, we can just usepsock->sk.Suggested-by John Fastabend <[email protected]>Cc Daniel Borkmann <[email protected]>Cc Jakub Sitnicki <[email protected]>Cc Lorenz Bauer <[email protected]>Signed-off-by Cong Wang <[email protected]>--- net/core/skmsg.c | 22 ++++++++++++++-------- 1 file changed, 14 insertions(+), 8 deletions(-)
[...]
net/core/skmsg.c static int sk_psock_verdict_apply(struct sk_psock *psock, struct sk_buff *skb, · line 951
Quoted 6 lines
net/core/skmsg.c +1
-1@@ -942,7 +948,7 @@ static int sk_psock_verdict_apply(struct sk_psock *psock, struct sk_buff *skb,case __SK_DROP:default : out_free: -kfree_skb (skb); +sock_drop (psock->sk, skb);
I must have missed this on first review.
Why should we mark a packet we intentionally drop as sk_drops? I think we should leave it as just kfree_skb() this way sk_drops is just the error cases and if users want this counter they can always add it to the bpf prog itself.
Quoted 3 lines
net/core/skmsg.c +0
-0}return err;
Re: [Patch bpf v3 8/8] skmsg: increase sk->sk_drops when dropping packets
Quoted 30 lines · John Fastabend
Quoted 16 lines · Cong Wang
From: Cong Wang <[email protected]>
It is hard to observe packet drops without increasing relevant drop counters, here we should increase
sk->sk_dropswhich is a protocol-independent counter. Fortunately psock is always associated with a struct sock, we can just usepsock->sk.Suggested-by John Fastabend <[email protected]>Cc Daniel Borkmann <[email protected]>Cc Jakub Sitnicki <[email protected]>Cc Lorenz Bauer <[email protected]>Signed-off-by Cong Wang <[email protected]>--- net/core/skmsg.c | 22 ++++++++++++++-------- 1 file changed, 14 insertions(+), 8 deletions(-)[...]
net/core/skmsg.c static int sk_psock_verdict_apply(struct sk_psock *psock, struct sk_buff *skb, · line 951
Quoted 6 lines
net/core/skmsg.c +1
-1@@ -942,7 +948,7 @@ static int sk_psock_verdict_apply(struct sk_psock *psock, struct sk_buff *skb,case __SK_DROP:default : out_free: -kfree_skb (skb); +sock_drop (psock->sk, skb);I must have missed this on first review.
Quoted 4 lines
Why should we mark a packet we intentionally drop as
sk_drops? I think we should leave it as justkfree_skb()this waysk_dropsis just the error cases and if users want this counter they can always add it to the bpf prog itself.
This is actually a mixed case of error and non-error drops, because bpf_sk_redirect_map() could return SK_DROP in error cases. And of course users could want to drop packets in whatever cases.
But if you look at packet filter cases, for example UDP one, it increases drop counters too when user-defined rules drop them:
2182 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2183 goto drop;
2184
...
2192 drop:
2193 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2194 atomic_inc(&sk->sk_drops); 2195 kfree_skb(skb); 2196 return -1;
Thanks.
John Fastabend
maintainerRe: [Patch bpf v3 8/8] skmsg: increase sk->sk_drops when dropping packets
Quoted 51 lines · Cong Wang
Quoted 30 lines · John Fastabend
Quoted 16 lines · Cong Wang
From: Cong Wang <[email protected]>
It is hard to observe packet drops without increasing relevant drop counters, here we should increase
sk->sk_dropswhich is a protocol-independent counter. Fortunately psock is always associated with a struct sock, we can just usepsock->sk.Suggested-by John Fastabend <[email protected]>Cc Daniel Borkmann <[email protected]>Cc Jakub Sitnicki <[email protected]>Cc Lorenz Bauer <[email protected]>Signed-off-by Cong Wang <[email protected]>--- net/core/skmsg.c | 22 ++++++++++++++-------- 1 file changed, 14 insertions(+), 8 deletions(-)[...]
net/core/skmsg.c static int sk_psock_verdict_apply(struct sk_psock *psock, struct sk_buff *skb, · line 951
Quoted 6 lines
net/core/skmsg.c +1
-1@@ -942,7 +948,7 @@ static int sk_psock_verdict_apply(struct sk_psock *psock, struct sk_buff *skb,case __SK_DROP:default : out_free: -kfree_skb (skb); +sock_drop (psock->sk, skb);I must have missed this on first review.
Quoted 4 lines
Why should we mark a packet we intentionally drop as
sk_drops? I think we should leave it as justkfree_skb()this waysk_dropsis just the error cases and if users want this counter they can always add it to the bpf prog itself.This is actually a mixed case of error and non-error drops, because
bpf_sk_redirect_map()could returnSK_DROPin error cases. And of course users could want to drop packets in whatever cases.But if you look at packet filter cases, for example UDP one, it increases drop counters too when user-defined rules drop them:
2182 if (
sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2183 goto drop;
2184
...
2192 drop:
Quoted 7 lines
2193 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2194 atomic_inc(&sk->sk_drops); 2195 kfree_skb(skb); 2196 return -1;Thanks.
OK same for TCP side for sk_filter_trim_cap() case. Works for me.
Quoted 4 lines · Cong Wang
From: Cong Wang <[email protected]>
This patchset contains a few bug fixes and improvements for
sock_map.
Quoted 8 lines
Patch 1 improves
recvmsg()accuracy for UDP, patch 2 improves UDP non-blockingread()by retrying on EAGAIN. With both of them, the failure rate of the UDP test case goes down from 10% to 1%.Patch 3 is memory leak fix I posted, no change since v1. The rest patches address similar memory leaks or improve error handling, including one increases
sk_dropscounter for error cases. Please check each patch description for more details.
For the series. My initial concern about marking BPF drops as socket drops appears to be common so I guess lets do it.
Quoted 31 lines
---
v3: add another bug fix as patch 4
update patch 5 accordingly
address John's review on the last patch
fix a few typos in patch descriptionsv2: group all patches together
set max for retries of EAGAINCong Wang (8):
skmsg: improveudp_bpf_recvmsg()accuracy selftests/bpf: Retry for EAGAIN inudp_redir_to_connected()
udp: fix a memory leak inudp_read_sock()
skmsg: clear skb redirect pointer before dropping it
skmsg: fix a memory leak insk_psock_verdict_apply()
skmsg: teachsk_psock_verdict_apply()to return errors
skmsg: pass source psock tosk_psock_skb_redirect()
skmsg: increasesk->sk_dropswhen dropping packetsinclude/linux/skmsg.h | 2 - net/core/skmsg.c | 82 +++++++++---------- net/ipv4/tcp_bpf.c | 24 +++++- net/ipv4/udp.c | 2 + net/ipv4/udp_bpf.c | 47 +++++++++-- .../selftests/bpf/prog_tests/sockmap_listen.c | 7 +- 6 files changed, 112 insertions(+), 52 deletions(-)Signature · 1 line
2.25.1