Skip to content

Commit

Permalink
Merge branch 'tipc-next'
Browse files Browse the repository at this point in the history 
Jon Maloy says:

====================
tipc: some improvements and fixes

We introduce a better algorithm for selecting when and which
users should be subject to link congestion control, plus clean
up some code for that mechanism.
Commit #3 fixes another rare race condition during packet reception.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
  • Loading branch information
David S. Miller committed Mar 25, 2015
2 parents b06b107 + 8b4ed86 commit 8fa38a3
Show file tree
Hide file tree
Showing 4 changed files with 157 additions and 97 deletions.
2 changes: 1 addition & 1 deletion net/tipc/bcast.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -831,7 +831,7 @@ int tipc_nl_add_bc_link(struct net *net, struct tipc_nl_msg *msg)
prop = nla_nest_start(msg->skb, TIPC_NLA_LINK_PROP);
if (!prop)
goto attr_msg_full;
if (nla_put_u32(msg->skb, TIPC_NLA_PROP_WIN, bcl->queue_limit[0]))
if (nla_put_u32(msg->skb, TIPC_NLA_PROP_WIN, bcl->window))
goto prop_msg_full;
nla_nest_end(msg->skb, prop);

Expand Down
207 changes: 130 additions & 77 deletions net/tipc/link.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -139,6 +139,13 @@ static void tipc_link_put(struct tipc_link *l_ptr)
kref_put(&l_ptr->ref, tipc_link_release);
}

static struct tipc_link *tipc_parallel_link(struct tipc_link *l)
{
if (l->owner->active_links[0] != l)
return l->owner->active_links[0];
return l->owner->active_links[1];
}

static void link_init_max_pkt(struct tipc_link *l_ptr)
{
struct tipc_node *node = l_ptr->owner;
Expand Down Expand Up @@ -310,7 +317,6 @@ struct tipc_link *tipc_link_create(struct tipc_node *n_ptr,
link_init_max_pkt(l_ptr);
l_ptr->priority = b_ptr->priority;
tipc_link_set_queue_limits(l_ptr, b_ptr->window);

l_ptr->next_out_no = 1;
__skb_queue_head_init(&l_ptr->transmq);
__skb_queue_head_init(&l_ptr->backlogq);
Expand Down Expand Up @@ -368,28 +374,43 @@ void tipc_link_delete_list(struct net *net, unsigned int bearer_id,
}

/**
* link_schedule_user - schedule user for wakeup after congestion
* link_schedule_user - schedule a message sender 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
* @list: message that was attempted sent
* Create pseudo msg to send back to user when congestion abates
* Only consumes message if there is an error
*/
static bool link_schedule_user(struct tipc_link *link, u32 oport,
uint chain_sz, uint imp)
static int link_schedule_user(struct tipc_link *link, struct sk_buff_head *list)
{
struct sk_buff *buf;
struct tipc_msg *msg = buf_msg(skb_peek(list));
int imp = msg_importance(msg);
u32 oport = msg_origport(msg);
u32 addr = link_own_addr(link);
struct sk_buff *skb;

buf = tipc_msg_create(SOCK_WAKEUP, 0, INT_H_SIZE, 0,
link_own_addr(link), link_own_addr(link),
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->wakeupq, buf);
/* This really cannot happen... */
if (unlikely(imp > TIPC_CRITICAL_IMPORTANCE)) {
pr_warn("%s<%s>, send queue full", link_rst_msg, link->name);
tipc_link_reset(link);
goto err;
}
/* Non-blocking sender: */
if (TIPC_SKB_CB(skb_peek(list))->wakeup_pending)
return -ELINKCONG;

/* Create and schedule wakeup pseudo message */
skb = tipc_msg_create(SOCK_WAKEUP, 0, INT_H_SIZE, 0,
addr, addr, oport, 0, 0);
if (!skb)
goto err;
TIPC_SKB_CB(skb)->chain_sz = skb_queue_len(list);
TIPC_SKB_CB(skb)->chain_imp = imp;
skb_queue_tail(&link->wakeupq, skb);
link->stats.link_congs++;
return true;
return -ELINKCONG;
err:
__skb_queue_purge(list);
return -ENOBUFS;
}

/**
Expand All @@ -398,19 +419,22 @@ static bool link_schedule_user(struct tipc_link *link, u32 oport,
* 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
*/
void link_prepare_wakeup(struct tipc_link *link)
void link_prepare_wakeup(struct tipc_link *l)
{
uint pend_qsz = skb_queue_len(&link->backlogq);
int pnd[TIPC_SYSTEM_IMPORTANCE + 1] = {0,};
int imp, lim;
struct sk_buff *skb, *tmp;

skb_queue_walk_safe(&link->wakeupq, skb, tmp) {
if (pend_qsz >= link->queue_limit[TIPC_SKB_CB(skb)->chain_imp])
skb_queue_walk_safe(&l->wakeupq, skb, tmp) {
imp = TIPC_SKB_CB(skb)->chain_imp;
lim = l->window + l->backlog[imp].limit;
pnd[imp] += TIPC_SKB_CB(skb)->chain_sz;
if ((pnd[imp] + l->backlog[imp].len) >= lim)
break;
pend_qsz += TIPC_SKB_CB(skb)->chain_sz;
skb_unlink(skb, &link->wakeupq);
skb_queue_tail(&link->inputq, skb);
link->owner->inputq = &link->inputq;
link->owner->action_flags |= TIPC_MSG_EVT;
skb_unlink(skb, &l->wakeupq);
skb_queue_tail(&l->inputq, skb);
l->owner->inputq = &l->inputq;
l->owner->action_flags |= TIPC_MSG_EVT;
}
}

Expand All @@ -424,6 +448,16 @@ void tipc_link_reset_fragments(struct tipc_link *l_ptr)
l_ptr->reasm_buf = NULL;
}

static void tipc_link_purge_backlog(struct tipc_link *l)
{
__skb_queue_purge(&l->backlogq);
l->backlog[TIPC_LOW_IMPORTANCE].len = 0;
l->backlog[TIPC_MEDIUM_IMPORTANCE].len = 0;
l->backlog[TIPC_HIGH_IMPORTANCE].len = 0;
l->backlog[TIPC_CRITICAL_IMPORTANCE].len = 0;
l->backlog[TIPC_SYSTEM_IMPORTANCE].len = 0;
}

/**
* tipc_link_purge_queues - purge all pkt queues associated with link
* @l_ptr: pointer to link
Expand All @@ -432,7 +466,7 @@ void tipc_link_purge_queues(struct tipc_link *l_ptr)
{
__skb_queue_purge(&l_ptr->deferdq);
__skb_queue_purge(&l_ptr->transmq);
__skb_queue_purge(&l_ptr->backlogq);
tipc_link_purge_backlog(l_ptr);
tipc_link_reset_fragments(l_ptr);
}

Expand Down Expand Up @@ -466,13 +500,13 @@ void tipc_link_reset(struct tipc_link *l_ptr)

/* Clean up all queues, except inputq: */
__skb_queue_purge(&l_ptr->transmq);
__skb_queue_purge(&l_ptr->backlogq);
__skb_queue_purge(&l_ptr->deferdq);
if (!owner->inputq)
owner->inputq = &l_ptr->inputq;
skb_queue_splice_init(&l_ptr->wakeupq, owner->inputq);
if (!skb_queue_empty(owner->inputq))
owner->action_flags |= TIPC_MSG_EVT;
tipc_link_purge_backlog(l_ptr);
l_ptr->rcv_unacked = 0;
l_ptr->checkpoint = 1;
l_ptr->next_out_no = 1;
Expand Down Expand Up @@ -696,48 +730,15 @@ static void link_state_event(struct tipc_link *l_ptr, unsigned int event)
}
}

/* tipc_link_cong: determine return value and how to treat the
* sent buffer during link congestion.
* - For plain, errorless user data messages we keep the buffer and
* return -ELINKONG.
* - For all other messages we discard the buffer and return -EHOSTUNREACH
* - For TIPC internal messages we also reset the link
*/
static int tipc_link_cong(struct tipc_link *link, struct sk_buff_head *list)
{
struct sk_buff *skb = skb_peek(list);
struct tipc_msg *msg = buf_msg(skb);
int imp = msg_importance(msg);
u32 oport = msg_tot_origport(msg);

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(skb)->wakeup_pending)
return -ELINKCONG;
if (link_schedule_user(link, oport, skb_queue_len(list), imp))
return -ELINKCONG;
drop:
__skb_queue_purge(list);
return -EHOSTUNREACH;
}

/**
* __tipc_link_xmit(): same as tipc_link_xmit, but destlink is known & locked
* @link: link to use
* @list: chain of buffers containing message
*
* Consumes the buffer chain, except when returning -ELINKCONG
* Returns 0 if success, otherwise errno: -ELINKCONG, -EMSGSIZE (plain socket
* user data messages) or -EHOSTUNREACH (all other messages/senders)
* Only the socket functions tipc_send_stream() and tipc_send_packet() need
* to act on the return value, since they may need to do more send attempts.
* Consumes the buffer chain, except when returning -ELINKCONG,
* since the caller then may want to make more send attempts.
* Returns 0 if success, or errno: -ELINKCONG, -EMSGSIZE or -ENOBUFS
* Messages at TIPC_SYSTEM_IMPORTANCE are always accepted
*/
int __tipc_link_xmit(struct net *net, struct tipc_link *link,
struct sk_buff_head *list)
Expand All @@ -754,16 +755,14 @@ int __tipc_link_xmit(struct net *net, struct tipc_link *link,
struct sk_buff_head *backlogq = &link->backlogq;
struct sk_buff *skb, *tmp;

/* Match queue limit against msg importance: */
if (unlikely(skb_queue_len(backlogq) >= link->queue_limit[imp]))
return tipc_link_cong(link, list);
/* Match backlog limit against msg importance: */
if (unlikely(link->backlog[imp].len >= link->backlog[imp].limit))
return link_schedule_user(link, list);

/* Has valid packet limit been used ? */
if (unlikely(msg_size(msg) > mtu)) {
__skb_queue_purge(list);
return -EMSGSIZE;
}

/* Prepare each packet for sending, and add to relevant queue: */
skb_queue_walk_safe(list, skb, tmp) {
__skb_unlink(skb, list);
Expand All @@ -786,8 +785,10 @@ int __tipc_link_xmit(struct net *net, struct tipc_link *link,
if (tipc_msg_make_bundle(&skb, mtu, link->addr)) {
link->stats.sent_bundled++;
link->stats.sent_bundles++;
imp = msg_importance(buf_msg(skb));
}
__skb_queue_tail(backlogq, skb);
link->backlog[imp].len++;
seqno++;
}
link->next_out_no = seqno;
Expand All @@ -808,13 +809,25 @@ static int __tipc_link_xmit_skb(struct tipc_link *link, struct sk_buff *skb)
return __tipc_link_xmit(link->owner->net, link, &head);
}

/* tipc_link_xmit_skb(): send single buffer to destination
* Buffers sent via this functon are generally TIPC_SYSTEM_IMPORTANCE
* messages, which will not be rejected
* The only exception is datagram messages rerouted after secondary
* lookup, which are rare and safe to dispose of anyway.
* TODO: Return real return value, and let callers use
* tipc_wait_for_sendpkt() where applicable
*/
int tipc_link_xmit_skb(struct net *net, struct sk_buff *skb, u32 dnode,
u32 selector)
{
struct sk_buff_head head;
int rc;

skb2list(skb, &head);
return tipc_link_xmit(net, &head, dnode, selector);
rc = tipc_link_xmit(net, &head, dnode, selector);
if (rc == -ELINKCONG)
kfree_skb(skb);
return 0;
}

/**
Expand Down Expand Up @@ -914,6 +927,7 @@ void tipc_link_push_packets(struct tipc_link *link)
if (!skb)
break;
msg = buf_msg(skb);
link->backlog[msg_importance(msg)].len--;
msg_set_ack(msg, ack);
msg_set_bcast_ack(msg, link->owner->bclink.last_in);
link->rcv_unacked = 0;
Expand Down Expand Up @@ -1019,6 +1033,32 @@ void tipc_link_retransmit(struct tipc_link *l_ptr, struct sk_buff *skb,
}
}

/* link_synch(): check if all packets arrived before the synch
* point have been consumed
* Returns true if the parallel links are synched, otherwise false
*/
static bool link_synch(struct tipc_link *l)
{
unsigned int post_synch;
struct tipc_link *pl;

pl = tipc_parallel_link(l);
if (pl == l)
goto synched;

/* Was last pre-synch packet added to input queue ? */
if (less_eq(pl->next_in_no, l->synch_point))
return false;

/* Is it still in the input queue ? */
post_synch = mod(pl->next_in_no - l->synch_point) - 1;
if (skb_queue_len(&pl->inputq) > post_synch)
return false;
synched:
l->flags &= ~LINK_SYNCHING;
return true;
}

static void link_retrieve_defq(struct tipc_link *link,
struct sk_buff_head *list)
{
Expand Down Expand Up @@ -1149,6 +1189,14 @@ void tipc_rcv(struct net *net, struct sk_buff *skb, struct tipc_bearer *b_ptr)
skb = NULL;
goto unlock;
}
/* Synchronize with parallel link if applicable */
if (unlikely((l_ptr->flags & LINK_SYNCHING) && !msg_dup(msg))) {
link_handle_out_of_seq_msg(l_ptr, skb);
if (link_synch(l_ptr))
link_retrieve_defq(l_ptr, &head);
skb = NULL;
goto unlock;
}
l_ptr->next_in_no++;
if (unlikely(!skb_queue_empty(&l_ptr->deferdq)))
link_retrieve_defq(l_ptr, &head);
Expand Down Expand Up @@ -1224,6 +1272,10 @@ static void tipc_link_input(struct tipc_link *link, struct sk_buff *skb)

switch (msg_user(msg)) {
case CHANGEOVER_PROTOCOL:
if (msg_dup(msg)) {
link->flags |= LINK_SYNCHING;
link->synch_point = msg_seqno(msg_get_wrapped(msg));
}
if (!tipc_link_tunnel_rcv(node, &skb))
break;
if (msg_user(buf_msg(skb)) != MSG_BUNDLER) {
Expand Down Expand Up @@ -1610,6 +1662,7 @@ void tipc_link_failover_send_queue(struct tipc_link *l_ptr)
tipc_msg_init(link_own_addr(l_ptr), &tunnel_hdr, CHANGEOVER_PROTOCOL,
ORIGINAL_MSG, INT_H_SIZE, l_ptr->addr);
skb_queue_splice_tail_init(&l_ptr->backlogq, &l_ptr->transmq);
tipc_link_purge_backlog(l_ptr);
msgcount = skb_queue_len(&l_ptr->transmq);
msg_set_bearer_id(&tunnel_hdr, l_ptr->peer_bearer_id);
msg_set_msgcnt(&tunnel_hdr, msgcount);
Expand Down Expand Up @@ -1817,11 +1870,11 @@ void tipc_link_set_queue_limits(struct tipc_link *l, u32 win)
int max_bulk = TIPC_MAX_PUBLICATIONS / (l->max_pkt / ITEM_SIZE);

l->window = win;
l->queue_limit[TIPC_LOW_IMPORTANCE] = win / 2;
l->queue_limit[TIPC_MEDIUM_IMPORTANCE] = win;
l->queue_limit[TIPC_HIGH_IMPORTANCE] = win / 2 * 3;
l->queue_limit[TIPC_CRITICAL_IMPORTANCE] = win * 2;
l->queue_limit[TIPC_SYSTEM_IMPORTANCE] = max_bulk;
l->backlog[TIPC_LOW_IMPORTANCE].limit = win / 2;
l->backlog[TIPC_MEDIUM_IMPORTANCE].limit = win;
l->backlog[TIPC_HIGH_IMPORTANCE].limit = win / 2 * 3;
l->backlog[TIPC_CRITICAL_IMPORTANCE].limit = win * 2;
l->backlog[TIPC_SYSTEM_IMPORTANCE].limit = max_bulk;
}

/* tipc_link_find_owner - locate owner node of link by link's name
Expand Down Expand Up @@ -2120,7 +2173,7 @@ static int __tipc_nl_add_link(struct net *net, struct tipc_nl_msg *msg,
if (nla_put_u32(msg->skb, TIPC_NLA_PROP_TOL, link->tolerance))
goto prop_msg_full;
if (nla_put_u32(msg->skb, TIPC_NLA_PROP_WIN,
link->queue_limit[TIPC_LOW_IMPORTANCE]))
link->window))
goto prop_msg_full;
if (nla_put_u32(msg->skb, TIPC_NLA_PROP_PRIO, link->priority))
goto prop_msg_full;
Expand Down
Loading

0 comments on commit 8fa38a3

Please sign in to comment.