tls: rx: do not use the standard strparser

TLS is a relatively poor fit for strparser. We pause the input
every time a message is received, wait for a read which will
decrypt the message, start the parser, repeat. strparser is
built to delineate the messages, wrap them in individual skbs
and let them float off into the stack or a different socket.
TLS wants the data pages and nothing else. There's no need
for TLS to keep cloning (and occasionally skb_unclone()'ing)
the TCP rx queue.

This patch uses a pre-allocated skb and attaches the skbs
from the TCP rx queue to it as frags. TLS is careful never
to modify the input skb without CoW'ing / detaching it first.

Since we call TCP rx queue cleanup directly we also get back
the benefit of skb deferred free.

Overall this results in a 6% gain in my benchmarks.

Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

1 files changed, 470 insertions, 14 deletions

diff --git a/net/tls/tls_strp.c b/net/tls/tls_strp.c
index d9bb4f23f01a..b945288c312e 100644
--- a/net/tls/tls_strp.c
+++ b/net/tls/tls_strp.c
@@ -1,37 +1,493 @@
 // SPDX-License-Identifier: GPL-2.0-only
+/* Copyright (c) 2016 Tom Herbert <tom@herbertland.com> */
 
 #include <linux/skbuff.h>
+#include <linux/workqueue.h>
+#include <net/strparser.h>
+#include <net/tcp.h>
+#include <net/sock.h>
+#include <net/tls.h>
 
 #include "tls.h"
 
-struct sk_buff *tls_strp_msg_detach(struct tls_sw_context_rx *ctx)
+static struct workqueue_struct *tls_strp_wq;
+
+static void tls_strp_abort_strp(struct tls_strparser *strp, int err)
+{
+	if (strp->stopped)
+		return;
+
+	strp->stopped = 1;
+
+	/* Report an error on the lower socket */
+	strp->sk->sk_err = -err;
+	sk_error_report(strp->sk);
+}
+
+static void tls_strp_anchor_free(struct tls_strparser *strp)
 {
+	struct skb_shared_info *shinfo = skb_shinfo(strp->anchor);
+
+	DEBUG_NET_WARN_ON_ONCE(atomic_read(&shinfo->dataref) != 1);
+	shinfo->frag_list = NULL;
+	consume_skb(strp->anchor);
+	strp->anchor = NULL;
+}
+
+/* Create a new skb with the contents of input copied to its page frags */
+static struct sk_buff *tls_strp_msg_make_copy(struct tls_strparser *strp)
+{
+	struct strp_msg *rxm;
 	struct sk_buff *skb;
+	int i, err, offset;
+
+	skb = alloc_skb_with_frags(0, strp->anchor->len, TLS_PAGE_ORDER,
+				   &err, strp->sk->sk_allocation);
+	if (!skb)
+		return NULL;
+
+	offset = strp->stm.offset;
+	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
+		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 
-	skb = ctx->recv_pkt;
-	ctx->recv_pkt = NULL;
+		WARN_ON_ONCE(skb_copy_bits(strp->anchor, offset,
+					   skb_frag_address(frag),
+					   skb_frag_size(frag)));
+		offset += skb_frag_size(frag);
+	}
+
+	skb_copy_header(skb, strp->anchor);
+	rxm = strp_msg(skb);
+	rxm->offset = 0;
 	return skb;
 }
 
+/* Steal the input skb, input msg is invalid after calling this function */
+struct sk_buff *tls_strp_msg_detach(struct tls_sw_context_rx *ctx)
+{
+	struct tls_strparser *strp = &ctx->strp;
+
+#ifdef CONFIG_TLS_DEVICE
+	DEBUG_NET_WARN_ON_ONCE(!strp->anchor->decrypted);
+#else
+	/* This function turns an input into an output,
+	 * that can only happen if we have offload.
+	 */
+	WARN_ON(1);
+#endif
+
+	if (strp->copy_mode) {
+		struct sk_buff *skb;
+
+		/* Replace anchor with an empty skb, this is a little
+		 * dangerous but __tls_cur_msg() warns on empty skbs
+		 * so hopefully we'll catch abuses.
+		 */
+		skb = alloc_skb(0, strp->sk->sk_allocation);
+		if (!skb)
+			return NULL;
+
+		swap(strp->anchor, skb);
+		return skb;
+	}
+
+	return tls_strp_msg_make_copy(strp);
+}
+
+/* Force the input skb to be in copy mode. The data ownership remains
+ * with the input skb itself (meaning unpause will wipe it) but it can
+ * be modified.
+ */
 int tls_strp_msg_cow(struct tls_sw_context_rx *ctx)
 {
-	struct sk_buff *unused;
-	int nsg;
+	struct tls_strparser *strp = &ctx->strp;
+	struct sk_buff *skb;
+
+	if (strp->copy_mode)
+		return 0;
+
+	skb = tls_strp_msg_make_copy(strp);
+	if (!skb)
+		return -ENOMEM;
+
+	tls_strp_anchor_free(strp);
+	strp->anchor = skb;
+
+	tcp_read_done(strp->sk, strp->stm.full_len);
+	strp->copy_mode = 1;
+
+	return 0;
+}
+
+/* Make a clone (in the skb sense) of the input msg to keep a reference
+ * to the underlying data. The reference-holding skbs get placed on
+ * @dst.
+ */
+int tls_strp_msg_hold(struct tls_strparser *strp, struct sk_buff_head *dst)
+{
+	struct skb_shared_info *shinfo = skb_shinfo(strp->anchor);
+
+	if (strp->copy_mode) {
+		struct sk_buff *skb;
+
+		WARN_ON_ONCE(!shinfo->nr_frags);
+
+		/* We can't skb_clone() the anchor, it gets wiped by unpause */
+		skb = alloc_skb(0, strp->sk->sk_allocation);
+		if (!skb)
+			return -ENOMEM;
+
+		__skb_queue_tail(dst, strp->anchor);
+		strp->anchor = skb;
+	} else {
+		struct sk_buff *iter, *clone;
+		int chunk, len, offset;
+
+		offset = strp->stm.offset;
+		len = strp->stm.full_len;
+		iter = shinfo->frag_list;
+
+		while (len > 0) {
+			if (iter->len <= offset) {
+				offset -= iter->len;
+				goto next;
+			}
+
+			chunk = iter->len - offset;
+			offset = 0;
+
+			clone = skb_clone(iter, strp->sk->sk_allocation);
+			if (!clone)
+				return -ENOMEM;
+			__skb_queue_tail(dst, clone);
+
+			len -= chunk;
+next:
+			iter = iter->next;
+		}
+	}
+
+	return 0;
+}
+
+static void tls_strp_flush_anchor_copy(struct tls_strparser *strp)
+{
+	struct skb_shared_info *shinfo = skb_shinfo(strp->anchor);
+	int i;
+
+	DEBUG_NET_WARN_ON_ONCE(atomic_read(&shinfo->dataref) != 1);
+
+	for (i = 0; i < shinfo->nr_frags; i++)
+		__skb_frag_unref(&shinfo->frags[i], false);
+	shinfo->nr_frags = 0;
+	strp->copy_mode = 0;
+}
+
+static int tls_strp_copyin(read_descriptor_t *desc, struct sk_buff *in_skb,
+			   unsigned int offset, size_t in_len)
+{
+	struct tls_strparser *strp = (struct tls_strparser *)desc->arg.data;
+	size_t sz, len, chunk;
+	struct sk_buff *skb;
+	skb_frag_t *frag;
+
+	if (strp->msg_ready)
+		return 0;
+
+	skb = strp->anchor;
+	frag = &skb_shinfo(skb)->frags[skb->len / PAGE_SIZE];
+
+	len = in_len;
+	/* First make sure we got the header */
+	if (!strp->stm.full_len) {
+		/* Assume one page is more than enough for headers */
+		chunk =	min_t(size_t, len, PAGE_SIZE - skb_frag_size(frag));
+		WARN_ON_ONCE(skb_copy_bits(in_skb, offset,
+					   skb_frag_address(frag) +
+					   skb_frag_size(frag),
+					   chunk));
+
+		sz = tls_rx_msg_size(strp, strp->anchor);
+		if (sz < 0) {
+			desc->error = sz;
+			return 0;
+		}
+
+		/* We may have over-read, sz == 0 is guaranteed under-read */
+		if (sz > 0)
+			chunk =	min_t(size_t, chunk, sz - skb->len);
+
+		skb->len += chunk;
+		skb->data_len += chunk;
+		skb_frag_size_add(frag, chunk);
+		frag++;
+		len -= chunk;
+		offset += chunk;
+
+		strp->stm.full_len = sz;
+		if (!strp->stm.full_len)
+			goto read_done;
+	}
+
+	/* Load up more data */
+	while (len && strp->stm.full_len > skb->len) {
+		chunk =	min_t(size_t, len, strp->stm.full_len - skb->len);
+		chunk = min_t(size_t, chunk, PAGE_SIZE - skb_frag_size(frag));
+		WARN_ON_ONCE(skb_copy_bits(in_skb, offset,
+					   skb_frag_address(frag) +
+					   skb_frag_size(frag),
+					   chunk));
+
+		skb->len += chunk;
+		skb->data_len += chunk;
+		skb_frag_size_add(frag, chunk);
+		frag++;
+		len -= chunk;
+		offset += chunk;
+	}
+
+	if (strp->stm.full_len == skb->len) {
+		desc->count = 0;
+
+		strp->msg_ready = 1;
+		tls_rx_msg_ready(strp);
+	}
+
+read_done:
+	return in_len - len;
+}
+
+static int tls_strp_read_copyin(struct tls_strparser *strp)
+{
+	struct socket *sock = strp->sk->sk_socket;
+	read_descriptor_t desc;
+
+	desc.arg.data = strp;
+	desc.error = 0;
+	desc.count = 1; /* give more than one skb per call */
+
+	/* sk should be locked here, so okay to do read_sock */
+	sock->ops->read_sock(strp->sk, &desc, tls_strp_copyin);
+
+	return desc.error;
+}
+
+static int tls_strp_read_short(struct tls_strparser *strp)
+{
+	struct skb_shared_info *shinfo;
+	struct page *page;
+	int need_spc, len;
+
+	/* If the rbuf is small or rcv window has collapsed to 0 we need
+	 * to read the data out. Otherwise the connection will stall.
+	 * Without pressure threshold of INT_MAX will never be ready.
+	 */
+	if (likely(!tcp_epollin_ready(strp->sk, INT_MAX)))
+		return 0;
+
+	shinfo = skb_shinfo(strp->anchor);
+	shinfo->frag_list = NULL;
+
+	/* If we don't know the length go max plus page for cipher overhead */
+	need_spc = strp->stm.full_len ?: TLS_MAX_PAYLOAD_SIZE + PAGE_SIZE;
+
+	for (len = need_spc; len > 0; len -= PAGE_SIZE) {
+		page = alloc_page(strp->sk->sk_allocation);
+		if (!page) {
+			tls_strp_flush_anchor_copy(strp);
+			return -ENOMEM;
+		}
+
+		skb_fill_page_desc(strp->anchor, shinfo->nr_frags++,
+				   page, 0, 0);
+	}
+
+	strp->copy_mode = 1;
+	strp->stm.offset = 0;
+
+	strp->anchor->len = 0;
+	strp->anchor->data_len = 0;
+	strp->anchor->truesize = round_up(need_spc, PAGE_SIZE);
+
+	tls_strp_read_copyin(strp);
+
+	return 0;
+}
+
+static void tls_strp_load_anchor_with_queue(struct tls_strparser *strp, int len)
+{
+	struct tcp_sock *tp = tcp_sk(strp->sk);
+	struct sk_buff *first;
+	u32 offset;
+
+	first = tcp_recv_skb(strp->sk, tp->copied_seq, &offset);
+	if (WARN_ON_ONCE(!first))
+		return;
+
+	/* Bestow the state onto the anchor */
+	strp->anchor->len = offset + len;
+	strp->anchor->data_len = offset + len;
+	strp->anchor->truesize = offset + len;
+
+	skb_shinfo(strp->anchor)->frag_list = first;
+
+	skb_copy_header(strp->anchor, first);
+	strp->anchor->destructor = NULL;
+
+	strp->stm.offset = offset;
+}
+
+void tls_strp_msg_load(struct tls_strparser *strp, bool force_refresh)
+{
+	struct strp_msg *rxm;
+	struct tls_msg *tlm;
+
+	DEBUG_NET_WARN_ON_ONCE(!strp->msg_ready);
+	DEBUG_NET_WARN_ON_ONCE(!strp->stm.full_len);
+
+	if (!strp->copy_mode && force_refresh) {
+		if (WARN_ON(tcp_inq(strp->sk) < strp->stm.full_len))
+			return;
+
+		tls_strp_load_anchor_with_queue(strp, strp->stm.full_len);
+	}
+
+	rxm = strp_msg(strp->anchor);
+	rxm->full_len	= strp->stm.full_len;
+	rxm->offset	= strp->stm.offset;
+	tlm = tls_msg(strp->anchor);
+	tlm->control	= strp->mark;
+}
+
+/* Called with lock held on lower socket */
+static int tls_strp_read_sock(struct tls_strparser *strp)
+{
+	int sz, inq;
+
+	inq = tcp_inq(strp->sk);
+	if (inq < 1)
+		return 0;
+
+	if (unlikely(strp->copy_mode))
+		return tls_strp_read_copyin(strp);
+
+	if (inq < strp->stm.full_len)
+		return tls_strp_read_short(strp);
+
+	if (!strp->stm.full_len) {
+		tls_strp_load_anchor_with_queue(strp, inq);
+
+		sz = tls_rx_msg_size(strp, strp->anchor);
+		if (sz < 0) {
+			tls_strp_abort_strp(strp, sz);
+			return sz;
+		}
+
+		strp->stm.full_len = sz;
+
+		if (!strp->stm.full_len || inq < strp->stm.full_len)
+			return tls_strp_read_short(strp);
+	}
+
+	strp->msg_ready = 1;
+	tls_rx_msg_ready(strp);
+
+	return 0;
+}
+
+void tls_strp_check_rcv(struct tls_strparser *strp)
+{
+	if (unlikely(strp->stopped) || strp->msg_ready)
+		return;
+
+	if (tls_strp_read_sock(strp) == -ENOMEM)
+		queue_work(tls_strp_wq, &strp->work);
+}
+
+/* Lower sock lock held */
+void tls_strp_data_ready(struct tls_strparser *strp)
+{
+	/* This check is needed to synchronize with do_tls_strp_work.
+	 * do_tls_strp_work acquires a process lock (lock_sock) whereas
+	 * the lock held here is bh_lock_sock. The two locks can be
+	 * held by different threads at the same time, but bh_lock_sock
+	 * allows a thread in BH context to safely check if the process
+	 * lock is held. In this case, if the lock is held, queue work.
+	 */
+	if (sock_owned_by_user_nocheck(strp->sk)) {
+		queue_work(tls_strp_wq, &strp->work);
+		return;
+	}
+
+	tls_strp_check_rcv(strp);
+}
+
+static void tls_strp_work(struct work_struct *w)
+{
+	struct tls_strparser *strp =
+		container_of(w, struct tls_strparser, work);
+
+	lock_sock(strp->sk);
+	tls_strp_check_rcv(strp);
+	release_sock(strp->sk);
+}
+
+void tls_strp_msg_done(struct tls_strparser *strp)
+{
+	WARN_ON(!strp->stm.full_len);
+
+	if (likely(!strp->copy_mode))
+		tcp_read_done(strp->sk, strp->stm.full_len);
+	else
+		tls_strp_flush_anchor_copy(strp);
+
+	strp->msg_ready = 0;
+	memset(&strp->stm, 0, sizeof(strp->stm));
+
+	tls_strp_check_rcv(strp);
+}
+
+void tls_strp_stop(struct tls_strparser *strp)
+{
+	strp->stopped = 1;
+}
+
+int tls_strp_init(struct tls_strparser *strp, struct sock *sk)
+{
+	memset(strp, 0, sizeof(*strp));
+
+	strp->sk = sk;
+
+	strp->anchor = alloc_skb(0, GFP_KERNEL);
+	if (!strp->anchor)
+		return -ENOMEM;
+
+	INIT_WORK(&strp->work, tls_strp_work);
 
-	nsg = skb_cow_data(ctx->recv_pkt, 0, &unused);
-	if (nsg < 0)
-		return nsg;
 	return 0;
 }
 
-int tls_strp_msg_hold(struct sock *sk, struct sk_buff *skb,
-		      struct sk_buff_head *dst)
+/* strp must already be stopped so that tls_strp_recv will no longer be called.
+ * Note that tls_strp_done is not called with the lower socket held.
+ */
+void tls_strp_done(struct tls_strparser *strp)
 {
-	struct sk_buff *clone;
+	WARN_ON(!strp->stopped);
 
-	clone = skb_clone(skb, sk->sk_allocation);
-	if (!clone)
+	cancel_work_sync(&strp->work);
+	tls_strp_anchor_free(strp);
+}
+
+int __init tls_strp_dev_init(void)
+{
+	tls_strp_wq = create_singlethread_workqueue("kstrp");
+	if (unlikely(!tls_strp_wq))
 		return -ENOMEM;
-	__skb_queue_tail(dst, clone);
+
 	return 0;
 }
+
+void tls_strp_dev_exit(void)
+{
+	destroy_workqueue(tls_strp_wq);
+}