Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit 50100a5e authored by Jon Paul Maloy's avatar Jon Paul Maloy Committed by David S. Miller
Browse files

tipc: use pseudo message to wake up sockets after link congestion 



The current link implementation keeps a linked list of blocked ports/
sockets that is populated when there is link congestion. The purpose
of this is to let the link know which users to wake up when the
congestion abates.

This adds unnecessary complexity to the data structure and the code,
since it forces us to involve the link each time we want to delete
a socket. It also forces us to grab the spinlock port_lock within
the scope of node_lock. We want to get rid of this direct dependence,
as well as the deadlock hazard resulting from the usage of port_lock.

In this commit, we instead let the link keep list of a "wakeup" pseudo
messages for use in such situations. Those messages are sent to the
pending sockets via the ordinary message reception path, and wake up
the socket's owner when they are received.

This enables us to get rid of the 'waiting_ports' linked lists in struct
tipc_port that manifest this direct reference. As a consequence, we can
eliminate another BH entry into the socket, and hence the need to grab
port_lock. This is a further step in our effort to remove port_lock
altogether.

Signed-off-by: default avatarJon Maloy <jon.maloy@ericsson.com>
Reviewed-by: default avatarErik Hugne <erik.hugne@ericsson.com>
Reviewed-by: default avatarYing Xue <ying.xue@windriver.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
parent 1dd0bd2b

net/tipc/bcast.c

+4 −3
Original line number Diff line number Diff line
@@ -300,8 +300,8 @@ void tipc_bclink_acknowledge(struct tipc_node *n_ptr, u32 acked) 
		tipc_link_push_queue(bcl);

		bclink_set_last_sent();

	}

	if (unlikely(released && !list_empty(&bcl->waiting_ports)))

		tipc_link_wakeup_ports(bcl, 0);

	if (unlikely(released && !skb_queue_empty(&bcl->waiting_sks)))

		bclink->node.action_flags |= TIPC_WAKEUP_USERS;

exit:

	tipc_bclink_unlock();

}
@@ -840,9 +840,10 @@ int tipc_bclink_init(void) 
	sprintf(bcbearer->media.name, "tipc-broadcast");



	spin_lock_init(&bclink->lock);

	INIT_LIST_HEAD(&bcl->waiting_ports);

	__skb_queue_head_init(&bcl->waiting_sks);

	bcl->next_out_no = 1;

	spin_lock_init(&bclink->node.lock);

	__skb_queue_head_init(&bclink->node.waiting_sks);

	bcl->owner = &bclink->node;

	bcl->max_pkt = MAX_PKT_DEFAULT_MCAST;

	tipc_link_set_queue_limits(bcl, BCLINK_WIN_DEFAULT);

net/tipc/core.h

+4 −1
Original line number Diff line number Diff line
@@ -187,8 +187,11 @@ static inline void k_term_timer(struct timer_list *timer) 


struct tipc_skb_cb {

	void *handle;

	bool deferred;

	struct sk_buff *tail;

	bool deferred;

	bool wakeup_pending;

	u16 chain_sz;

	u16 chain_imp;

};



#define TIPC_SKB_CB(__skb) ((struct tipc_skb_cb *)&((__skb)->cb[0]))

net/tipc/link.c

+54 −65
Original line number Diff line number Diff line
@@ -275,7 +275,7 @@ struct tipc_link *tipc_link_create(struct tipc_node *n_ptr, 
	link_init_max_pkt(l_ptr);



	l_ptr->next_out_no = 1;

	INIT_LIST_HEAD(&l_ptr->waiting_ports);

	__skb_queue_head_init(&l_ptr->waiting_sks);



	link_reset_statistics(l_ptr);
@@ -322,66 +322,47 @@ void tipc_link_delete_list(unsigned int bearer_id, bool shutting_down) 
}



/**

 * link_schedule_port - schedule port for deferred sending

 * @l_ptr: pointer to link

 * @origport: reference to sending port

 * @sz: amount of data to be sent

 *

 * Schedules port for renewed sending of messages after link congestion

 * has abated.

 * link_schedule_user - schedule user for wakeup after congestion

 * @link: congested link

 * @oport: sending port

 * @chain_sz: size of buffer chain that was attempted sent

 * @imp: importance of message attempted sent

 * Create pseudo msg to send back to user when congestion abates

 */

static int link_schedule_port(struct tipc_link *l_ptr, u32 origport, u32 sz)

static bool link_schedule_user(struct tipc_link *link, u32 oport,

			       uint chain_sz, uint imp)

{

	struct tipc_port *p_ptr;

	struct tipc_sock *tsk;

	struct sk_buff *buf;



	spin_lock_bh(&tipc_port_list_lock);

	p_ptr = tipc_port_lock(origport);

	if (p_ptr) {

		if (!list_empty(&p_ptr->wait_list))

			goto exit;

		tsk = tipc_port_to_sock(p_ptr);

		tsk->link_cong = 1;

		p_ptr->waiting_pkts = 1 + ((sz - 1) / l_ptr->max_pkt);

		list_add_tail(&p_ptr->wait_list, &l_ptr->waiting_ports);

		l_ptr->stats.link_congs++;

exit:

		tipc_port_unlock(p_ptr);

	}

	spin_unlock_bh(&tipc_port_list_lock);

	return -ELINKCONG;

	buf = tipc_msg_create(SOCK_WAKEUP, 0, INT_H_SIZE, 0, tipc_own_addr,

			      tipc_own_addr, oport, 0, 0);

	if (!buf)

		return false;

	TIPC_SKB_CB(buf)->chain_sz = chain_sz;

	TIPC_SKB_CB(buf)->chain_imp = imp;

	__skb_queue_tail(&link->waiting_sks, buf);

	link->stats.link_congs++;

	return true;

}



void tipc_link_wakeup_ports(struct tipc_link *l_ptr, int all)

/**

 * link_prepare_wakeup - prepare users for wakeup after congestion

 * @link: congested link

 * Move a number of waiting users, as permitted by available space in

 * the send queue, from link wait queue to node wait queue for wakeup

 */

static void link_prepare_wakeup(struct tipc_link *link)

{

	struct tipc_port *p_ptr;

	struct tipc_sock *tsk;

	struct tipc_port *temp_p_ptr;

	int win = l_ptr->queue_limit[0] - l_ptr->out_queue_size;



	if (all)

		win = 100000;

	if (win <= 0)

		return;

	if (!spin_trylock_bh(&tipc_port_list_lock))

		return;

	if (link_congested(l_ptr))

		goto exit;

	list_for_each_entry_safe(p_ptr, temp_p_ptr, &l_ptr->waiting_ports,

				 wait_list) {

		if (win <= 0)

	struct sk_buff_head *wq = &link->waiting_sks;

	struct sk_buff *buf;

	uint pend_qsz = link->out_queue_size;



	for (buf = skb_peek(wq); buf; buf = skb_peek(wq)) {

		if (pend_qsz >= link->queue_limit[TIPC_SKB_CB(buf)->chain_imp])

			break;

		tsk = tipc_port_to_sock(p_ptr);

		list_del_init(&p_ptr->wait_list);

		spin_lock_bh(p_ptr->lock);

		tsk->link_cong = 0;

		tipc_sock_wakeup(tsk);

		win -= p_ptr->waiting_pkts;

		spin_unlock_bh(p_ptr->lock);

		pend_qsz += TIPC_SKB_CB(buf)->chain_sz;

		__skb_queue_tail(&link->owner->waiting_sks, __skb_dequeue(wq));

	}



exit:

	spin_unlock_bh(&tipc_port_list_lock);

}



/**
@@ -423,6 +404,7 @@ void tipc_link_reset(struct tipc_link *l_ptr) 
	u32 prev_state = l_ptr->state;

	u32 checkpoint = l_ptr->next_in_no;

	int was_active_link = tipc_link_is_active(l_ptr);

	struct tipc_node *owner = l_ptr->owner;



	msg_set_session(l_ptr->pmsg, ((msg_session(l_ptr->pmsg) + 1) & 0xffff));
@@ -450,9 +432,10 @@ void tipc_link_reset(struct tipc_link *l_ptr) 
	kfree_skb(l_ptr->proto_msg_queue);

	l_ptr->proto_msg_queue = NULL;

	kfree_skb_list(l_ptr->oldest_deferred_in);

	if (!list_empty(&l_ptr->waiting_ports))

		tipc_link_wakeup_ports(l_ptr, 1);



	if (!skb_queue_empty(&l_ptr->waiting_sks)) {

		skb_queue_splice_init(&l_ptr->waiting_sks, &owner->waiting_sks);

		owner->action_flags |= TIPC_WAKEUP_USERS;

	}

	l_ptr->retransm_queue_head = 0;

	l_ptr->retransm_queue_size = 0;

	l_ptr->last_out = NULL;
@@ -688,19 +671,23 @@ static void link_state_event(struct tipc_link *l_ptr, unsigned int event) 
static int tipc_link_cong(struct tipc_link *link, struct sk_buff *buf)

{

	struct tipc_msg *msg = buf_msg(buf);

	uint psz = msg_size(msg);

	uint imp = tipc_msg_tot_importance(msg);

	u32 oport = msg_tot_origport(msg);



	if (likely(imp <= TIPC_CRITICAL_IMPORTANCE)) {

		if (!msg_errcode(msg) && !msg_reroute_cnt(msg)) {

			link_schedule_port(link, oport, psz);

			return -ELINKCONG;

		}

	} else {

	if (unlikely(imp > TIPC_CRITICAL_IMPORTANCE)) {

		pr_warn("%s<%s>, send queue full", link_rst_msg, link->name);

		tipc_link_reset(link);

		goto drop;

	}

	if (unlikely(msg_errcode(msg)))

		goto drop;

	if (unlikely(msg_reroute_cnt(msg)))

		goto drop;

	if (TIPC_SKB_CB(buf)->wakeup_pending)

		return -ELINKCONG;

	if (link_schedule_user(link, oport, TIPC_SKB_CB(buf)->chain_sz, imp))

		return -ELINKCONG;

drop:

	kfree_skb_list(buf);

	return -EHOSTUNREACH;

}
@@ -1202,8 +1189,10 @@ void tipc_rcv(struct sk_buff *head, struct tipc_bearer *b_ptr) 
		if (unlikely(l_ptr->next_out))

			tipc_link_push_queue(l_ptr);



		if (unlikely(!list_empty(&l_ptr->waiting_ports)))

			tipc_link_wakeup_ports(l_ptr, 0);

		if (released && !skb_queue_empty(&l_ptr->waiting_sks)) {

			link_prepare_wakeup(l_ptr);

			l_ptr->owner->action_flags |= TIPC_WAKEUP_USERS;

		}



		/* Process the incoming packet */

		if (unlikely(!link_working_working(l_ptr))) {

net/tipc/link.h

+3 −4
Original line number Diff line number Diff line 
/*

 * net/tipc/link.h: Include file for TIPC link code

 *

 * Copyright (c) 1995-2006, 2013, Ericsson AB

 * Copyright (c) 1995-2006, 2013-2014, Ericsson AB

 * Copyright (c) 2004-2005, 2010-2011, Wind River Systems

 * All rights reserved.

 *
@@ -133,7 +133,7 @@ struct tipc_stats { 
 * @retransm_queue_size: number of messages to retransmit

 * @retransm_queue_head: sequence number of first message to retransmit

 * @next_out: ptr to first unsent outbound message in queue

 * @waiting_ports: linked list of ports waiting for link congestion to abate

 * @waiting_sks: linked list of sockets waiting for link congestion to abate

 * @long_msg_seq_no: next identifier to use for outbound fragmented messages

 * @reasm_buf: head of partially reassembled inbound message fragments

 * @stats: collects statistics regarding link activity
@@ -194,7 +194,7 @@ struct tipc_link { 
	u32 retransm_queue_size;

	u32 retransm_queue_head;

	struct sk_buff *next_out;

	struct list_head waiting_ports;

	struct sk_buff_head waiting_sks;



	/* Fragmentation/reassembly */

	u32 long_msg_seq_no;
@@ -235,7 +235,6 @@ void tipc_link_proto_xmit(struct tipc_link *l_ptr, u32 msg_typ, int prob, 
void tipc_link_push_queue(struct tipc_link *l_ptr);

u32 tipc_link_defer_pkt(struct sk_buff **head, struct sk_buff **tail,

			struct sk_buff *buf);

void tipc_link_wakeup_ports(struct tipc_link *l_ptr, int all);

void tipc_link_set_queue_limits(struct tipc_link *l_ptr, u32 window);

void tipc_link_retransmit(struct tipc_link *l_ptr,

			  struct sk_buff *start, u32 retransmits);

net/tipc/msg.c

+5 −2
Original line number Diff line number Diff line
@@ -182,7 +182,7 @@ int tipc_msg_build(struct tipc_msg *mhdr, struct iovec const *iov, 
	struct sk_buff *buf, *prev;

	char *pktpos;

	int rc;



	uint chain_sz = 0;

	msg_set_size(mhdr, msz);



	/* No fragmentation needed? */
@@ -193,6 +193,7 @@ int tipc_msg_build(struct tipc_msg *mhdr, struct iovec const *iov, 
			return -ENOMEM;

		skb_copy_to_linear_data(buf, mhdr, mhsz);

		pktpos = buf->data + mhsz;

		TIPC_SKB_CB(buf)->chain_sz = 1;

		if (!dsz || !memcpy_fromiovecend(pktpos, iov, offset, dsz))

			return dsz;

		rc = -EFAULT;
@@ -209,6 +210,7 @@ int tipc_msg_build(struct tipc_msg *mhdr, struct iovec const *iov, 
	*chain = buf = tipc_buf_acquire(pktmax);

	if (!buf)

		return -ENOMEM;

	chain_sz = 1;

	pktpos = buf->data;

	skb_copy_to_linear_data(buf, &pkthdr, INT_H_SIZE);

	pktpos += INT_H_SIZE;
@@ -242,6 +244,7 @@ int tipc_msg_build(struct tipc_msg *mhdr, struct iovec const *iov, 
			rc = -ENOMEM;

			goto error;

		}

		chain_sz++;

		prev->next = buf;

		msg_set_type(&pkthdr, FRAGMENT);

		msg_set_size(&pkthdr, pktsz);
@@ -251,7 +254,7 @@ int tipc_msg_build(struct tipc_msg *mhdr, struct iovec const *iov, 
		pktrem = pktsz - INT_H_SIZE;



	} while (1);



	TIPC_SKB_CB(*chain)->chain_sz = chain_sz;

	msg_set_type(buf_msg(buf), LAST_FRAGMENT);

	return dsz;

error: