[PATCH net] net/tls: Don't recursively call push_record during tls_write_space callbacks

[PATCH net] net/tls: Don't recursively call push_record during tls_write_space callbacks

[Dave Watson]

It is reported that in some cases, write_space may be called in
do_tcp_sendpages, such that we recursively invoke do_tcp_sendpages again:

[  660.468802]  ? do_tcp_sendpages+0x8d/0x580
[  660.468826]  ? tls_push_sg+0x74/0x130 [tls]
[  660.468852]  ? tls_push_record+0x24a/0x390 [tls]
[  660.468880]  ? tls_write_space+0x6a/0x80 [tls]
...

tls_push_sg already does a loop over all sending sg's, so ignore
any tls_write_space notifications until we are done sending.
We then have to call the previous write_space to wake up
poll() waiters after we are done with the send loop.

Reported-by: Andre Tomt <andre@tomt.net>
Signed-off-by: Dave Watson <davejwatson@fb.com>
---
 include/net/tls.h  | 1 +
 net/tls/tls_main.c | 7 +++++++
 2 files changed, 8 insertions(+)

diff --git a/include/net/tls.h b/include/net/tls.h
index 3da8e13..b400d0bb 100644
--- a/include/net/tls.h
+++ b/include/net/tls.h
@@ -148,6 +148,7 @@ struct tls_context {
 	struct scatterlist *partially_sent_record;
 	u16 partially_sent_offset;
 	unsigned long flags;
+	bool in_tcp_sendpages;
 
 	u16 pending_open_record_frags;
 	int (*push_pending_record)(struct sock *sk, int flags);
diff --git a/net/tls/tls_main.c b/net/tls/tls_main.c
index 0d37997..cc03e00 100644
--- a/net/tls/tls_main.c
+++ b/net/tls/tls_main.c
@@ -114,6 +114,7 @@ int tls_push_sg(struct sock *sk,
 	size = sg->length - offset;
 	offset += sg->offset;
 
+	ctx->in_tcp_sendpages = true;
 	while (1) {
 		if (sg_is_last(sg))
 			sendpage_flags = flags;
@@ -148,6 +149,8 @@ int tls_push_sg(struct sock *sk,
 	}
 
 	clear_bit(TLS_PENDING_CLOSED_RECORD, &ctx->flags);
+	ctx->in_tcp_sendpages = false;
+	ctx->sk_write_space(sk);
 
 	return 0;
 }
@@ -217,6 +220,10 @@ static void tls_write_space(struct sock *sk)
 {
 	struct tls_context *ctx = tls_get_ctx(sk);
 
+	/* We are already sending pages, ignore notification */
+	if (ctx->in_tcp_sendpages)
+		return;
+
 	if (!sk->sk_write_pending && tls_is_pending_closed_record(ctx)) {
 		gfp_t sk_allocation = sk->sk_allocation;
 		int rc;
-- 
2.9.5

Re: [PATCH net] net/tls: Don't recursively call push_record during tls_write_space callbacks

[David Miller]

From: Dave Watson <davejwatson@fb.com>
Date: Tue, 1 May 2018 13:05:39 -0700

> It is reported that in some cases, write_space may be called in
> do_tcp_sendpages, such that we recursively invoke do_tcp_sendpages again:
> 
> [  660.468802]  ? do_tcp_sendpages+0x8d/0x580
> [  660.468826]  ? tls_push_sg+0x74/0x130 [tls]
> [  660.468852]  ? tls_push_record+0x24a/0x390 [tls]
> [  660.468880]  ? tls_write_space+0x6a/0x80 [tls]
> ...
> 
> tls_push_sg already does a loop over all sending sg's, so ignore
> any tls_write_space notifications until we are done sending.
> We then have to call the previous write_space to wake up
> poll() waiters after we are done with the send loop.
> 
> Reported-by: Andre Tomt <andre@tomt.net>
> Signed-off-by: Dave Watson <davejwatson@fb.com>

Applied.

Re: [PATCH net] net/tls: Don't recursively call push_record during tls_write_space callbacks

[Andre Tomt]

On 01. mai 2018 22:05, Dave Watson wrote:
> It is reported that in some cases, write_space may be called in
> do_tcp_sendpages, such that we recursively invoke do_tcp_sendpages again:
> 
> [  660.468802]  ? do_tcp_sendpages+0x8d/0x580
> [  660.468826]  ? tls_push_sg+0x74/0x130 [tls]
> [  660.468852]  ? tls_push_record+0x24a/0x390 [tls]
> [  660.468880]  ? tls_write_space+0x6a/0x80 [tls]
> ...
> 
> tls_push_sg already does a loop over all sending sg's, so ignore
> any tls_write_space notifications until we are done sending.
> We then have to call the previous write_space to wake up
> poll() waiters after we are done with the send loop.
> 
> Reported-by: Andre Tomt <andre@tomt.net>
> Signed-off-by: Dave Watson <davejwatson@fb.com>

Unfortunately it seems like this patch has a bug, while it fixed the 
kernel crashing it is causing some connections in my testbed to stall.

Making sure ctx->in_tcp_sendpages is also cleared before the return ret 
within the while(1) loop seems to fix it for me.


diff -Naurp a/net/tls/tls_main.c b/net/tls/tls_main.c
--- a/net/tls/tls_main.c	2018-05-06 02:23:41.638597066 +0200
+++ b/net/tls/tls_main.c	2018-05-06 01:59:14.378568139 +0200
@@ -135,6 +135,7 @@ retry:
  			offset -= sg->offset;
  			ctx->partially_sent_offset = offset;
  			ctx->partially_sent_record = (void *)sg;
+			ctx->in_tcp_sendpages = false;
  			return ret;
  		}