From: Yaogong Wang wygivan@google.com

[ Upstream commit 9f5afeae51526b3ad7b7cb21ee8b145ce6ea7a7a ]

Over the years, TCP BDP has increased by several orders of magnitude, and some people are considering to reach the 2 Gbytes limit.

Even with current window scale limit of 14, ~1 Gbytes maps to ~740,000 MSS.

In presence of packet losses (or reorders), TCP stores incoming packets into an out of order queue, and number of skbs sitting there waiting for the missing packets to be received can be in the 10^5 range.

Most packets are appended to the tail of this queue, and when packets can finally be transferred to receive queue, we scan the queue from its head.

However, in presence of heavy losses, we might have to find an arbitrary point in this queue, involving a linear scan for every incoming packet, throwing away cpu caches.

This patch converts it to a RB tree, to get bounded latencies.

Yaogong wrote a preliminary patch about 2 years ago. Eric did the rebase, added ofo_last_skb cache, polishing and tests.

Tested with network dropping between 1 and 10 % packets, with good success (about 30 % increase of throughput in stress tests)

Next step would be to also use an RB tree for the write queue at sender side ;)

Signed-off-by: Yaogong Wang wygivan@google.com Signed-off-by: Eric Dumazet edumazet@google.com Cc: Yuchung Cheng ycheng@google.com Cc: Neal Cardwell ncardwell@google.com Cc: Ilpo Järvinen ilpo.jarvinen@helsinki.fi Acked-By: Ilpo Järvinen ilpo.jarvinen@helsinki.fi Signed-off-by: David S. Miller davem@davemloft.net Signed-off-by: Mao Wenan maowenan@huawei.com --- include/linux/skbuff.h | 8 ++ include/linux/tcp.h | 7 +- include/net/tcp.h | 2 +- net/core/skbuff.c | 19 +++ net/ipv4/tcp.c | 4 +- net/ipv4/tcp_input.c | 358 +++++++++++++++++++++++++++-------------------- net/ipv4/tcp_ipv4.c | 2 +- net/ipv4/tcp_minisocks.c | 1 - 8 files changed, 242 insertions(+), 159 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h index c28bd8b..a490dd7 100644 --- a/include/linux/skbuff.h +++ b/include/linux/skbuff.h @@ -2273,6 +2273,8 @@ static inline void __skb_queue_purge(struct sk_buff_head *list) kfree_skb(skb); }

+void skb_rbtree_purge(struct rb_root *root); + void *netdev_alloc_frag(unsigned int fragsz);

struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, @@ -2807,6 +2809,12 @@ static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) return __pskb_trim(skb, len); }

+#define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) +#define skb_rb_first(root) rb_to_skb(rb_first(root)) +#define skb_rb_last(root) rb_to_skb(rb_last(root)) +#define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) +#define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) + #define skb_queue_walk(queue, skb) \ for (skb = (queue)->next; \ skb != (struct sk_buff *)(queue); \ diff --git a/include/linux/tcp.h b/include/linux/tcp.h index 5b6df1a..747404d 100644 --- a/include/linux/tcp.h +++ b/include/linux/tcp.h @@ -279,10 +279,9 @@ struct tcp_sock { struct sk_buff* lost_skb_hint; struct sk_buff *retransmit_skb_hint;

- /* OOO segments go in this list. Note that socket lock must be held, - * as we do not use sk_buff_head lock. - */ - struct sk_buff_head out_of_order_queue; + /* OOO segments go in this rbtree. Socket lock must be held. */ + struct rb_root out_of_order_queue; + struct sk_buff *ooo_last_skb; /* cache rb_last(out_of_order_queue) */

/* SACKs data, these 2 need to be together (see tcp_options_write) */ struct tcp_sack_block duplicate_sack[1]; /* D-SACK block */ diff --git a/include/net/tcp.h b/include/net/tcp.h index 6c89238..a99f75e 100644 --- a/include/net/tcp.h +++ b/include/net/tcp.h @@ -649,7 +649,7 @@ static inline void tcp_fast_path_check(struct sock *sk) { struct tcp_sock *tp = tcp_sk(sk);

- if (skb_queue_empty(&tp->out_of_order_queue) && + if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && tp->rcv_wnd && atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && !tp->urg_data) diff --git a/net/core/skbuff.c b/net/core/skbuff.c index 55be076..9703924 100644 --- a/net/core/skbuff.c +++ b/net/core/skbuff.c @@ -2378,6 +2378,25 @@ void skb_queue_purge(struct sk_buff_head *list) EXPORT_SYMBOL(skb_queue_purge);

/** + * skb_rbtree_purge - empty a skb rbtree + * @root: root of the rbtree to empty + * + * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from + * the list and one reference dropped. This function does not take + * any lock. Synchronization should be handled by the caller (e.g., TCP + * out-of-order queue is protected by the socket lock). + */ +void skb_rbtree_purge(struct rb_root *root) +{ + struct sk_buff *skb, *next; + + rbtree_postorder_for_each_entry_safe(skb, next, root, rbnode) + kfree_skb(skb); + + *root = RB_ROOT; +} + +/** * skb_queue_head - queue a buffer at the list head * @list: list to use * @newsk: buffer to queue diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c index 5e162b8..b7492aa 100644 --- a/net/ipv4/tcp.c +++ b/net/ipv4/tcp.c @@ -382,7 +382,7 @@ void tcp_init_sock(struct sock *sk) struct inet_connection_sock *icsk = inet_csk(sk); struct tcp_sock *tp = tcp_sk(sk);

- __skb_queue_head_init(&tp->out_of_order_queue); + tp->out_of_order_queue = RB_ROOT; tcp_init_xmit_timers(sk); tcp_prequeue_init(tp); INIT_LIST_HEAD(&tp->tsq_node); @@ -2240,7 +2240,7 @@ int tcp_disconnect(struct sock *sk, int flags) tcp_clear_xmit_timers(sk); __skb_queue_purge(&sk->sk_receive_queue); tcp_write_queue_purge(sk); - __skb_queue_purge(&tp->out_of_order_queue); + skb_rbtree_purge(&tp->out_of_order_queue);

inet->inet_dport = 0;

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index 9df4cbb..7832d0d 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -4073,7 +4073,7 @@ static void tcp_fin(struct sock *sk) /* It _is_ possible, that we have something out-of-order _after_ FIN. * Probably, we should reset in this case. For now drop them. */ - __skb_queue_purge(&tp->out_of_order_queue); + skb_rbtree_purge(&tp->out_of_order_queue); if (tcp_is_sack(tp)) tcp_sack_reset(&tp->rx_opt); sk_mem_reclaim(sk); @@ -4233,7 +4233,7 @@ static void tcp_sack_remove(struct tcp_sock *tp) int this_sack;

/* Empty ofo queue, hence, all the SACKs are eaten. Clear. */ - if (skb_queue_empty(&tp->out_of_order_queue)) { + if (RB_EMPTY_ROOT(&tp->out_of_order_queue)) { tp->rx_opt.num_sacks = 0; return; } @@ -4309,10 +4309,13 @@ static void tcp_ofo_queue(struct sock *sk) { struct tcp_sock *tp = tcp_sk(sk); __u32 dsack_high = tp->rcv_nxt; + bool fin, fragstolen, eaten; struct sk_buff *skb, *tail; - bool fragstolen, eaten; + struct rb_node *p;

- while ((skb = skb_peek(&tp->out_of_order_queue)) != NULL) { + p = rb_first(&tp->out_of_order_queue); + while (p) { + skb = rb_entry(p, struct sk_buff, rbnode); if (after(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) break;

@@ -4322,9 +4325,10 @@ static void tcp_ofo_queue(struct sock *sk) dsack_high = TCP_SKB_CB(skb)->end_seq; tcp_dsack_extend(sk, TCP_SKB_CB(skb)->seq, dsack); } + p = rb_next(p); + rb_erase(&skb->rbnode, &tp->out_of_order_queue);

- __skb_unlink(skb, &tp->out_of_order_queue); - if (!after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt)) { + if (unlikely(!after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt))) { SOCK_DEBUG(sk, "ofo packet was already received\n"); tcp_drop(sk, skb); continue; @@ -4336,12 +4340,19 @@ static void tcp_ofo_queue(struct sock *sk) tail = skb_peek_tail(&sk->sk_receive_queue); eaten = tail && tcp_try_coalesce(sk, tail, skb, &fragstolen); tcp_rcv_nxt_update(tp, TCP_SKB_CB(skb)->end_seq); + fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN; if (!eaten) __skb_queue_tail(&sk->sk_receive_queue, skb); - if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) - tcp_fin(sk); - if (eaten) + else kfree_skb_partial(skb, fragstolen); + + if (unlikely(fin)) { + tcp_fin(sk); + /* tcp_fin() purges tp->out_of_order_queue, + * so we must end this loop right now. + */ + break; + } } }

@@ -4371,8 +4382,10 @@ static int tcp_try_rmem_schedule(struct sock *sk, struct sk_buff *skb, static void tcp_data_queue_ofo(struct sock *sk, struct sk_buff *skb) { struct tcp_sock *tp = tcp_sk(sk); + struct rb_node **p, *q, *parent; struct sk_buff *skb1; u32 seq, end_seq; + bool fragstolen;

tcp_ecn_check_ce(sk, skb);

@@ -4387,89 +4400,86 @@ static void tcp_data_queue_ofo(struct sock *sk, struct sk_buff *skb) inet_csk_schedule_ack(sk);

NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPOFOQUEUE); + seq = TCP_SKB_CB(skb)->seq; + end_seq = TCP_SKB_CB(skb)->end_seq; SOCK_DEBUG(sk, "out of order segment: rcv_next %X seq %X - %X\n", - tp->rcv_nxt, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq); + tp->rcv_nxt, seq, end_seq);

- skb1 = skb_peek_tail(&tp->out_of_order_queue); - if (!skb1) { + p = &tp->out_of_order_queue.rb_node; + if (RB_EMPTY_ROOT(&tp->out_of_order_queue)) { /* Initial out of order segment, build 1 SACK. */ if (tcp_is_sack(tp)) { tp->rx_opt.num_sacks = 1; - tp->selective_acks[0].start_seq = TCP_SKB_CB(skb)->seq; - tp->selective_acks[0].end_seq = - TCP_SKB_CB(skb)->end_seq; + tp->selective_acks[0].start_seq = seq; + tp->selective_acks[0].end_seq = end_seq; } - __skb_queue_head(&tp->out_of_order_queue, skb); - goto end; - } - - seq = TCP_SKB_CB(skb)->seq; - end_seq = TCP_SKB_CB(skb)->end_seq; - - if (seq == TCP_SKB_CB(skb1)->end_seq) { - bool fragstolen; - - if (!tcp_try_coalesce(sk, skb1, skb, &fragstolen)) { - __skb_queue_after(&tp->out_of_order_queue, skb1, skb); - } else { - tcp_grow_window(sk, skb); - kfree_skb_partial(skb, fragstolen); - skb = NULL; - } - - if (!tp->rx_opt.num_sacks || - tp->selective_acks[0].end_seq != seq) - goto add_sack; - - /* Common case: data arrive in order after hole. */ - tp->selective_acks[0].end_seq = end_seq; + rb_link_node(&skb->rbnode, NULL, p); + rb_insert_color(&skb->rbnode, &tp->out_of_order_queue); + tp->ooo_last_skb = skb; goto end; }

- /* Find place to insert this segment. */ - while (1) { - if (!after(TCP_SKB_CB(skb1)->seq, seq)) - break; - if (skb_queue_is_first(&tp->out_of_order_queue, skb1)) { - skb1 = NULL; - break; + /* In the typical case, we are adding an skb to the end of the list. + * Use of ooo_last_skb avoids the O(Log(N)) rbtree lookup. + */ + if (tcp_try_coalesce(sk, tp->ooo_last_skb, skb, &fragstolen)) { +coalesce_done: + tcp_grow_window(sk, skb); + kfree_skb_partial(skb, fragstolen); + skb = NULL; + goto add_sack; + } + + /* Find place to insert this segment. Handle overlaps on the way. */ + parent = NULL; + while (*p) { + parent = *p; + skb1 = rb_entry(parent, struct sk_buff, rbnode); + if (before(seq, TCP_SKB_CB(skb1)->seq)) { + p = &parent->rb_left; + continue; } - skb1 = skb_queue_prev(&tp->out_of_order_queue, skb1); - }

- /* Do skb overlap to previous one? */ - if (skb1 && before(seq, TCP_SKB_CB(skb1)->end_seq)) { - if (!after(end_seq, TCP_SKB_CB(skb1)->end_seq)) { - /* All the bits are present. Drop. */ - NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPOFOMERGE); - tcp_drop(sk, skb); - skb = NULL; - tcp_dsack_set(sk, seq, end_seq); - goto add_sack; - } - if (after(seq, TCP_SKB_CB(skb1)->seq)) { - /* Partial overlap. */ - tcp_dsack_set(sk, seq, - TCP_SKB_CB(skb1)->end_seq); - } else { - if (skb_queue_is_first(&tp->out_of_order_queue, - skb1)) - skb1 = NULL; - else - skb1 = skb_queue_prev( - &tp->out_of_order_queue, - skb1); + if (before(seq, TCP_SKB_CB(skb1)->end_seq)) { + if (!after(end_seq, TCP_SKB_CB(skb1)->end_seq)) { + /* All the bits are present. Drop. */ + NET_INC_STATS(sock_net(sk), + LINUX_MIB_TCPOFOMERGE); + __kfree_skb(skb); + skb = NULL; + tcp_dsack_set(sk, seq, end_seq); + goto add_sack; + } + if (after(seq, TCP_SKB_CB(skb1)->seq)) { + /* Partial overlap. */ + tcp_dsack_set(sk, seq, TCP_SKB_CB(skb1)->end_seq); + } else { + /* skb's seq == skb1's seq and skb covers skb1. + * Replace skb1 with skb. + */ + rb_replace_node(&skb1->rbnode, &skb->rbnode, + &tp->out_of_order_queue); + tcp_dsack_extend(sk, + TCP_SKB_CB(skb1)->seq, + TCP_SKB_CB(skb1)->end_seq); + NET_INC_STATS(sock_net(sk), + LINUX_MIB_TCPOFOMERGE); + __kfree_skb(skb1); + goto add_sack; + } + } else if (tcp_try_coalesce(sk, skb1, skb, &fragstolen)) { + goto coalesce_done; } + p = &parent->rb_right; } - if (!skb1) - __skb_queue_head(&tp->out_of_order_queue, skb); - else - __skb_queue_after(&tp->out_of_order_queue, skb1, skb);

- /* And clean segments covered by new one as whole. */ - while (!skb_queue_is_last(&tp->out_of_order_queue, skb)) { - skb1 = skb_queue_next(&tp->out_of_order_queue, skb); + /* Insert segment into RB tree. */ + rb_link_node(&skb->rbnode, parent, p); + rb_insert_color(&skb->rbnode, &tp->out_of_order_queue);

+ /* Remove other segments covered by skb. */ + while ((q = rb_next(&skb->rbnode)) != NULL) { + skb1 = rb_entry(q, struct sk_buff, rbnode); if (!after(end_seq, TCP_SKB_CB(skb1)->seq)) break; if (before(end_seq, TCP_SKB_CB(skb1)->end_seq)) { @@ -4477,12 +4487,15 @@ static void tcp_data_queue_ofo(struct sock *sk, struct sk_buff *skb) end_seq); break; } - __skb_unlink(skb1, &tp->out_of_order_queue); + rb_erase(&skb1->rbnode, &tp->out_of_order_queue); tcp_dsack_extend(sk, TCP_SKB_CB(skb1)->seq, TCP_SKB_CB(skb1)->end_seq); NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPOFOMERGE); tcp_drop(sk, skb1); } + /* If there is no skb after us, we are the last_skb ! */ + if (!q) + tp->ooo_last_skb = skb;

add_sack: if (tcp_is_sack(tp)) @@ -4621,13 +4634,13 @@ queue_and_out: if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) tcp_fin(sk);

- if (!skb_queue_empty(&tp->out_of_order_queue)) { + if (!RB_EMPTY_ROOT(&tp->out_of_order_queue)) { tcp_ofo_queue(sk);

/* RFC2581. 4.2. SHOULD send immediate ACK, when * gap in queue is filled. */ - if (skb_queue_empty(&tp->out_of_order_queue)) + if (RB_EMPTY_ROOT(&tp->out_of_order_queue)) inet_csk(sk)->icsk_ack.pingpong = 0; }

@@ -4679,48 +4692,76 @@ drop: tcp_data_queue_ofo(sk, skb); }

+static struct sk_buff *tcp_skb_next(struct sk_buff *skb, struct sk_buff_head *list) +{ + if (list) + return !skb_queue_is_last(list, skb) ? skb->next : NULL; + + return rb_entry_safe(rb_next(&skb->rbnode), struct sk_buff, rbnode); +} + static struct sk_buff *tcp_collapse_one(struct sock *sk, struct sk_buff *skb, - struct sk_buff_head *list) + struct sk_buff_head *list, + struct rb_root *root) { - struct sk_buff *next = NULL; + struct sk_buff *next = tcp_skb_next(skb, list);

- if (!skb_queue_is_last(list, skb)) - next = skb_queue_next(list, skb); + if (list) + __skb_unlink(skb, list); + else + rb_erase(&skb->rbnode, root);

- __skb_unlink(skb, list); __kfree_skb(skb); NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPRCVCOLLAPSED);

return next; }

+/* Insert skb into rb tree, ordered by TCP_SKB_CB(skb)->seq */ +static void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb) +{ + struct rb_node **p = &root->rb_node; + struct rb_node *parent = NULL; + struct sk_buff *skb1; + + while (*p) { + parent = *p; + skb1 = rb_entry(parent, struct sk_buff, rbnode); + if (before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb1)->seq)) + p = &parent->rb_left; + else + p = &parent->rb_right; + } + rb_link_node(&skb->rbnode, parent, p); + rb_insert_color(&skb->rbnode, root); +} + /* Collapse contiguous sequence of skbs head..tail with * sequence numbers start..end. * - * If tail is NULL, this means until the end of the list. + * If tail is NULL, this means until the end of the queue. * * Segments with FIN/SYN are not collapsed (only because this * simplifies code) */ static void -tcp_collapse(struct sock *sk, struct sk_buff_head *list, - struct sk_buff *head, struct sk_buff *tail, - u32 start, u32 end) +tcp_collapse(struct sock *sk, struct sk_buff_head *list, struct rb_root *root, + struct sk_buff *head, struct sk_buff *tail, u32 start, u32 end) { - struct sk_buff *skb, *n; + struct sk_buff *skb = head, *n; + struct sk_buff_head tmp; bool end_of_skbs;

/* First, check that queue is collapsible and find - * the point where collapsing can be useful. */ - skb = head; + * the point where collapsing can be useful. + */ restart: - end_of_skbs = true; - skb_queue_walk_from_safe(list, skb, n) { - if (skb == tail) - break; + for (end_of_skbs = true; skb != NULL && skb != tail; skb = n) { + n = tcp_skb_next(skb, list); + /* No new bits? It is possible on ofo queue. */ if (!before(start, TCP_SKB_CB(skb)->end_seq)) { - skb = tcp_collapse_one(sk, skb, list); + skb = tcp_collapse_one(sk, skb, list, root); if (!skb) break; goto restart; @@ -4738,13 +4779,10 @@ restart: break; }

- if (!skb_queue_is_last(list, skb)) { - struct sk_buff *next = skb_queue_next(list, skb); - if (next != tail && - TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(next)->seq) { - end_of_skbs = false; - break; - } + if (n && n != tail && + TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(n)->seq) { + end_of_skbs = false; + break; }

/* Decided to skip this, advance start seq. */ @@ -4754,17 +4792,22 @@ restart: (TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN))) return;

+ __skb_queue_head_init(&tmp); + while (before(start, end)) { int copy = min_t(int, SKB_MAX_ORDER(0, 0), end - start); struct sk_buff *nskb;

nskb = alloc_skb(copy, GFP_ATOMIC); if (!nskb) - return; + break;

memcpy(nskb->cb, skb->cb, sizeof(skb->cb)); TCP_SKB_CB(nskb)->seq = TCP_SKB_CB(nskb)->end_seq = start; - __skb_queue_before(list, skb, nskb); + if (list) + __skb_queue_before(list, skb, nskb); + else + __skb_queue_tail(&tmp, nskb); /* defer rbtree insertion */ skb_set_owner_r(nskb, sk);

/* Copy data, releasing collapsed skbs. */ @@ -4782,14 +4825,17 @@ restart: start += size; } if (!before(start, TCP_SKB_CB(skb)->end_seq)) { - skb = tcp_collapse_one(sk, skb, list); + skb = tcp_collapse_one(sk, skb, list, root); if (!skb || skb == tail || (TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN))) - return; + goto end; } } } +end: + skb_queue_walk_safe(&tmp, skb, n) + tcp_rbtree_insert(root, skb); }

/* Collapse ofo queue. Algorithm: select contiguous sequence of skbs @@ -4799,34 +4845,39 @@ static void tcp_collapse_ofo_queue(struct sock *sk) { struct tcp_sock *tp = tcp_sk(sk); u32 range_truesize, sum_tiny = 0; - struct sk_buff *skb = skb_peek(&tp->out_of_order_queue); - struct sk_buff *head; + struct sk_buff *skb, *head; + struct rb_node *p; u32 start, end;

- if (!skb) + p = rb_first(&tp->out_of_order_queue); + skb = rb_entry_safe(p, struct sk_buff, rbnode); +new_range: + if (!skb) { + p = rb_last(&tp->out_of_order_queue); + /* Note: This is possible p is NULL here. We do not + * use rb_entry_safe(), as ooo_last_skb is valid only + * if rbtree is not empty. + */ + tp->ooo_last_skb = rb_entry(p, struct sk_buff, rbnode); return; - + } start = TCP_SKB_CB(skb)->seq; end = TCP_SKB_CB(skb)->end_seq; range_truesize = skb->truesize; - head = skb; - - for (;;) { - struct sk_buff *next = NULL;

- if (!skb_queue_is_last(&tp->out_of_order_queue, skb)) - next = skb_queue_next(&tp->out_of_order_queue, skb); - skb = next; + for (head = skb;;) { + skb = tcp_skb_next(skb, NULL);

- /* Segment is terminated when we see gap or when - * we are at the end of all the queue. */ + /* Range is terminated when we see a gap or when + * we are at the queue end. + */ if (!skb || after(TCP_SKB_CB(skb)->seq, end) || before(TCP_SKB_CB(skb)->end_seq, start)) { /* Do not attempt collapsing tiny skbs */ if (range_truesize != head->truesize || end - start >= SKB_WITH_OVERHEAD(SK_MEM_QUANTUM)) { - tcp_collapse(sk, &tp->out_of_order_queue, + tcp_collapse(sk, NULL, &tp->out_of_order_queue, head, skb, start, end); } else { sum_tiny += range_truesize; @@ -4834,20 +4885,14 @@ static void tcp_collapse_ofo_queue(struct sock *sk) return; }

- head = skb; - if (!skb) - break; - /* Start new segment */ + goto new_range; + } + + range_truesize += skb->truesize; + if (unlikely(before(TCP_SKB_CB(skb)->seq, start))) start = TCP_SKB_CB(skb)->seq; + if (after(TCP_SKB_CB(skb)->end_seq, end)) end = TCP_SKB_CB(skb)->end_seq; - range_truesize = skb->truesize; - } else { - range_truesize += skb->truesize; - if (before(TCP_SKB_CB(skb)->seq, start)) - start = TCP_SKB_CB(skb)->seq; - if (after(TCP_SKB_CB(skb)->end_seq, end)) - end = TCP_SKB_CB(skb)->end_seq; - } } }

@@ -4858,23 +4903,36 @@ static void tcp_collapse_ofo_queue(struct sock *sk) static bool tcp_prune_ofo_queue(struct sock *sk) { struct tcp_sock *tp = tcp_sk(sk); - bool res = false; + struct rb_node *node, *prev;

- if (!skb_queue_empty(&tp->out_of_order_queue)) { - NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_OFOPRUNED); - __skb_queue_purge(&tp->out_of_order_queue); + if (RB_EMPTY_ROOT(&tp->out_of_order_queue)) + return false;

- /* Reset SACK state. A conforming SACK implementation will - * do the same at a timeout based retransmit. When a connection - * is in a sad state like this, we care only about integrity - * of the connection not performance. - */ - if (tp->rx_opt.sack_ok) - tcp_sack_reset(&tp->rx_opt); + NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_OFOPRUNED); + + node = &tp->ooo_last_skb->rbnode; + do { + prev = rb_prev(node); + rb_erase(node, &tp->out_of_order_queue); + __kfree_skb(rb_to_skb(node)); sk_mem_reclaim(sk); - res = true; - } - return res; + if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf && + !tcp_under_memory_pressure(sk)) + break; + + node = prev; + } while (node); + tp->ooo_last_skb = rb_entry(prev, struct sk_buff, rbnode); + + /* Reset SACK state. A conforming SACK implementation will + * do the same at a timeout based retransmit. When a connection + * is in a sad state like this, we care only about integrity + * of the connection not performance. + */ + if (tp->rx_opt.sack_ok) + tcp_sack_reset(&tp->rx_opt); + + return true; }

/* Reduce allocated memory if we can, trying to get @@ -4902,7 +4960,7 @@ static int tcp_prune_queue(struct sock *sk)

tcp_collapse_ofo_queue(sk); if (!skb_queue_empty(&sk->sk_receive_queue)) - tcp_collapse(sk, &sk->sk_receive_queue, + tcp_collapse(sk, &sk->sk_receive_queue, NULL, skb_peek(&sk->sk_receive_queue), NULL, tp->copied_seq, tp->rcv_nxt); @@ -5007,7 +5065,7 @@ static void __tcp_ack_snd_check(struct sock *sk, int ofo_possible) /* We ACK each frame or... */ tcp_in_quickack_mode(sk) || /* We have out of order data. */ - (ofo_possible && skb_peek(&tp->out_of_order_queue))) { + (ofo_possible && !RB_EMPTY_ROOT(&tp->out_of_order_queue))) { /* Then ack it now */ tcp_send_ack(sk); } else { diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c index 01715fc..ee8399f 100644 --- a/net/ipv4/tcp_ipv4.c +++ b/net/ipv4/tcp_ipv4.c @@ -1830,7 +1830,7 @@ void tcp_v4_destroy_sock(struct sock *sk) tcp_write_queue_purge(sk);

/* Cleans up our, hopefully empty, out_of_order_queue. */ - __skb_queue_purge(&tp->out_of_order_queue); + skb_rbtree_purge(&tp->out_of_order_queue);

#ifdef CONFIG_TCP_MD5SIG /* Clean up the MD5 key list, if any */ diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c index 4c1c94f..81c633d 100644 --- a/net/ipv4/tcp_minisocks.c +++ b/net/ipv4/tcp_minisocks.c @@ -495,7 +495,6 @@ struct sock *tcp_create_openreq_child(const struct sock *sk, newtp->snd_cwnd_cnt = 0;

tcp_init_xmit_timers(newsk); - __skb_queue_head_init(&newtp->out_of_order_queue); newtp->write_seq = newtp->pushed_seq = treq->snt_isn + 1;

newtp->rx_opt.saw_tstamp = 0;

From: Eric Dumazet edumazet@google.com

[ Upstream commit 72cd43ba64fc172a443410ce01645895850844c8 ]

Juha-Matti Tilli reported that malicious peers could inject tiny packets in out_of_order_queue, forcing very expensive calls to tcp_collapse_ofo_queue() and tcp_prune_ofo_queue() for every incoming packet. out_of_order_queue rb-tree can contain thousands of nodes, iterating over all of them is not nice.

Before linux-4.9, we would have pruned all packets in ofo_queue in one go, every XXXX packets. XXXX depends on sk_rcvbuf and skbs truesize, but is about 7000 packets with tcp_rmem[2] default of 6 MB.

Since we plan to increase tcp_rmem[2] in the future to cope with modern BDP, can not revert to the old behavior, without great pain.

Strategy taken in this patch is to purge ~12.5 % of the queue capacity.

Fixes: 36a6503fedda ("tcp: refine tcp_prune_ofo_queue() to not drop all packets") Signed-off-by: Eric Dumazet edumazet@google.com Reported-by: Juha-Matti Tilli juha-matti.tilli@iki.fi Acked-by: Yuchung Cheng ycheng@google.com Acked-by: Soheil Hassas Yeganeh soheil@google.com Signed-off-by: David S. Miller davem@davemloft.net Signed-off-by: Mao Wenan maowenan@huawei.com --- net/ipv4/tcp_input.c | 16 +++++++++++----- 1 file changed, 11 insertions(+), 5 deletions(-)

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index 4cc0a53..4739a93 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -4899,27 +4899,33 @@ new_range:

/* * Purge the out-of-order queue. + * Drop at least 12.5 % of sk_rcvbuf to avoid malicious attacks. * Return true if queue was pruned. */ static bool tcp_prune_ofo_queue(struct sock *sk) { struct tcp_sock *tp = tcp_sk(sk); struct rb_node *node, *prev; + int goal;

if (RB_EMPTY_ROOT(&tp->out_of_order_queue)) return false;

NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_OFOPRUNED); - + goal = sk->sk_rcvbuf >> 3; node = &tp->ooo_last_skb->rbnode; do { prev = rb_prev(node); rb_erase(node, &tp->out_of_order_queue); + goal -= rb_to_skb(node)->truesize; __kfree_skb(rb_to_skb(node)); - sk_mem_reclaim(sk); - if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf && - !tcp_under_memory_pressure(sk)) - break; + if (!prev || goal <= 0) { + sk_mem_reclaim(sk); + if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf && + !tcp_under_memory_pressure(sk)) + break; + goal = sk->sk_rcvbuf >> 3; + }

node = prev; } while (node);

From: Eric Dumazet edumazet@google.com

[ Upstream commit 58152ecbbcc6a0ce7fddd5bf5f6ee535834ece0c ]

In case skb in out_or_order_queue is the result of multiple skbs coalescing, we would like to get a proper gso_segs counter tracking, so that future tcp_drop() can report an accurate number.

I chose to not implement this tracking for skbs in receive queue, since they are not dropped, unless socket is disconnected.

Signed-off-by: Eric Dumazet edumazet@google.com Acked-by: Soheil Hassas Yeganeh soheil@google.com Acked-by: Yuchung Cheng ycheng@google.com Signed-off-by: David S. Miller davem@davemloft.net Signed-off-by: Mao Wenan maowenan@huawei.com --- net/ipv4/tcp_input.c | 23 +++++++++++++++++++++-- 1 file changed, 21 insertions(+), 2 deletions(-)

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index cbe0ca0..1aff93d 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -4296,6 +4296,23 @@ static bool tcp_try_coalesce(struct sock *sk, return true; }

+static bool tcp_ooo_try_coalesce(struct sock *sk, + struct sk_buff *to, + struct sk_buff *from, + bool *fragstolen) +{ + bool res = tcp_try_coalesce(sk, to, from, fragstolen); + + /* In case tcp_drop() is called later, update to->gso_segs */ + if (res) { + u32 gso_segs = max_t(u16, 1, skb_shinfo(to)->gso_segs) + + max_t(u16, 1, skb_shinfo(from)->gso_segs); + + skb_shinfo(to)->gso_segs = min_t(u32, gso_segs, 0xFFFF); + } + return res; +} + static void tcp_drop(struct sock *sk, struct sk_buff *skb) { sk_drops_add(sk, skb); @@ -4422,7 +4439,8 @@ static void tcp_data_queue_ofo(struct sock *sk, struct sk_buff *skb) /* In the typical case, we are adding an skb to the end of the list. * Use of ooo_last_skb avoids the O(Log(N)) rbtree lookup. */ - if (tcp_try_coalesce(sk, tp->ooo_last_skb, skb, &fragstolen)) { + if (tcp_ooo_try_coalesce(sk, tp->ooo_last_skb, + skb, &fragstolen)) { coalesce_done: tcp_grow_window(sk, skb); kfree_skb_partial(skb, fragstolen); @@ -4467,7 +4485,8 @@ coalesce_done: tcp_drop(sk, skb1); goto merge_right; } - } else if (tcp_try_coalesce(sk, skb1, skb, &fragstolen)) { + } else if (tcp_ooo_try_coalesce(sk, skb1, + skb, &fragstolen)) { goto coalesce_done; } p = &parent->rb_right;