ccid: ccid-2/3 code cosmetics
[linux-2.6.git] / net / dccp / ackvec.c
index 4181984..92a6fcb 100644 (file)
 /*
  *  net/dccp/ackvec.c
  *
- *  An implementation of Ack Vectors for the DCCP protocol
- *  Copyright (c) 2007 University of Aberdeen, Scotland, UK
+ *  An implementation of the DCCP protocol
  *  Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
  *
  *      This program is free software; you can redistribute it and/or modify it
  *      under the terms of the GNU General Public License as published by the
  *      Free Software Foundation; version 2 of the License;
  */
+
+#include "ackvec.h"
 #include "dccp.h"
+
+#include <linux/init.h>
+#include <linux/errno.h>
 #include <linux/kernel.h>
+#include <linux/skbuff.h>
 #include <linux/slab.h>
 
+#include <net/sock.h>
+
 static struct kmem_cache *dccp_ackvec_slab;
 static struct kmem_cache *dccp_ackvec_record_slab;
 
-struct dccp_ackvec *dccp_ackvec_alloc(const gfp_t priority)
+static struct dccp_ackvec_record *dccp_ackvec_record_new(void)
 {
-       struct dccp_ackvec *av = kmem_cache_zalloc(dccp_ackvec_slab, priority);
+       struct dccp_ackvec_record *avr =
+                       kmem_cache_alloc(dccp_ackvec_record_slab, GFP_ATOMIC);
 
-       if (av != NULL) {
-               av->av_buf_head = av->av_buf_tail = DCCPAV_MAX_ACKVEC_LEN - 1;
-               INIT_LIST_HEAD(&av->av_records);
-       }
-       return av;
+       if (avr != NULL)
+               INIT_LIST_HEAD(&avr->avr_node);
+
+       return avr;
 }
 
-static void dccp_ackvec_purge_records(struct dccp_ackvec *av)
+static void dccp_ackvec_record_delete(struct dccp_ackvec_record *avr)
 {
-       struct dccp_ackvec_record *cur, *next;
-
-       list_for_each_entry_safe(cur, next, &av->av_records, avr_node)
-               kmem_cache_free(dccp_ackvec_record_slab, cur);
-       INIT_LIST_HEAD(&av->av_records);
+       if (unlikely(avr == NULL))
+               return;
+       /* Check if deleting a linked record */
+       WARN_ON(!list_empty(&avr->avr_node));
+       kmem_cache_free(dccp_ackvec_record_slab, avr);
 }
 
-void dccp_ackvec_free(struct dccp_ackvec *av)
+static void dccp_ackvec_insert_avr(struct dccp_ackvec *av,
+                                  struct dccp_ackvec_record *avr)
 {
-       if (likely(av != NULL)) {
-               dccp_ackvec_purge_records(av);
-               kmem_cache_free(dccp_ackvec_slab, av);
+       /*
+        * AVRs are sorted by seqno. Since we are sending them in order, we
+        * just add the AVR at the head of the list.
+        * -sorbo.
+        */
+       if (!list_empty(&av->av_records)) {
+               const struct dccp_ackvec_record *head =
+                                       list_entry(av->av_records.next,
+                                                  struct dccp_ackvec_record,
+                                                  avr_node);
+               BUG_ON(before48(avr->avr_ack_seqno, head->avr_ack_seqno));
        }
+
+       list_add(&avr->avr_node, &av->av_records);
 }
 
-/**
- * dccp_ackvec_update_records  -  Record information about sent Ack Vectors
- * @av:                Ack Vector records to update
- * @seqno:     Sequence number of the packet carrying the Ack Vector just sent
- * @nonce_sum: The sum of all buffer nonces contained in the Ack Vector
- */
-int dccp_ackvec_update_records(struct dccp_ackvec *av, u64 seqno, u8 nonce_sum)
+int dccp_insert_option_ackvec(struct sock *sk, struct sk_buff *skb)
 {
+       struct dccp_sock *dp = dccp_sk(sk);
+       struct dccp_ackvec *av = dp->dccps_hc_rx_ackvec;
+       /* Figure out how many options do we need to represent the ackvec */
+       const u8 nr_opts = DIV_ROUND_UP(av->av_vec_len, DCCP_SINGLE_OPT_MAXLEN);
+       u16 len = av->av_vec_len + 2 * nr_opts, i;
+       u32 elapsed_time;
+       const unsigned char *tail, *from;
+       unsigned char *to;
        struct dccp_ackvec_record *avr;
+       suseconds_t delta;
+
+       if (DCCP_SKB_CB(skb)->dccpd_opt_len + len > DCCP_MAX_OPT_LEN)
+               return -1;
+
+       delta = ktime_us_delta(ktime_get_real(), av->av_time);
+       elapsed_time = delta / 10;
 
-       avr = kmem_cache_alloc(dccp_ackvec_record_slab, GFP_ATOMIC);
+       if (elapsed_time != 0 &&
+           dccp_insert_option_elapsed_time(skb, elapsed_time))
+               return -1;
+
+       avr = dccp_ackvec_record_new();
        if (avr == NULL)
-               return -ENOBUFS;
+               return -1;
+
+       DCCP_SKB_CB(skb)->dccpd_opt_len += len;
+
+       to   = skb_push(skb, len);
+       len  = av->av_vec_len;
+       from = av->av_buf + av->av_buf_head;
+       tail = av->av_buf + DCCP_MAX_ACKVEC_LEN;
+
+       for (i = 0; i < nr_opts; ++i) {
+               int copylen = len;
+
+               if (len > DCCP_SINGLE_OPT_MAXLEN)
+                       copylen = DCCP_SINGLE_OPT_MAXLEN;
+
+               *to++ = DCCPO_ACK_VECTOR_0;
+               *to++ = copylen + 2;
+
+               /* Check if buf_head wraps */
+               if (from + copylen > tail) {
+                       const u16 tailsize = tail - from;
+
+                       memcpy(to, from, tailsize);
+                       to      += tailsize;
+                       len     -= tailsize;
+                       copylen -= tailsize;
+                       from    = av->av_buf;
+               }
+
+               memcpy(to, from, copylen);
+               from += copylen;
+               to   += copylen;
+               len  -= copylen;
+       }
 
-       avr->avr_ack_seqno  = seqno;
-       avr->avr_ack_ptr    = av->av_buf_head;
-       avr->avr_ack_ackno  = av->av_buf_ackno;
-       avr->avr_ack_nonce  = nonce_sum;
-       avr->avr_ack_runlen = dccp_ackvec_runlen(av->av_buf + av->av_buf_head);
-       /*
-        * When the buffer overflows, we keep no more than one record. This is
-        * the simplest way of disambiguating sender-Acks dating from before the
-        * overflow from sender-Acks which refer to after the overflow; a simple
-        * solution is preferable here since we are handling an exception.
-        */
-       if (av->av_overflow)
-               dccp_ackvec_purge_records(av);
        /*
-        * Since GSS is incremented for each packet, the list is automatically
-        * arranged in descending order of @ack_seqno.
+        *      From RFC 4340, A.2:
+        *
+        *      For each acknowledgement it sends, the HC-Receiver will add an
+        *      acknowledgement record.  ack_seqno will equal the HC-Receiver
+        *      sequence number it used for the ack packet; ack_ptr will equal
+        *      buf_head; ack_ackno will equal buf_ackno; and ack_nonce will
+        *      equal buf_nonce.
         */
-       list_add(&avr->avr_node, &av->av_records);
+       avr->avr_ack_seqno = DCCP_SKB_CB(skb)->dccpd_seq;
+       avr->avr_ack_ptr   = av->av_buf_head;
+       avr->avr_ack_ackno = av->av_buf_ackno;
+       avr->avr_ack_nonce = av->av_buf_nonce;
+       avr->avr_sent_len  = av->av_vec_len;
 
-       dccp_pr_debug("Added Vector, ack_seqno=%llu, ack_ackno=%llu (rl=%u)\n",
+       dccp_ackvec_insert_avr(av, avr);
+
+       dccp_pr_debug("%s ACK Vector 0, len=%d, ack_seqno=%llu, "
+                     "ack_ackno=%llu\n",
+                     dccp_role(sk), avr->avr_sent_len,
                      (unsigned long long)avr->avr_ack_seqno,
-                     (unsigned long long)avr->avr_ack_ackno,
-                     avr->avr_ack_runlen);
+                     (unsigned long long)avr->avr_ack_ackno);
        return 0;
 }
 
-static struct dccp_ackvec_record *dccp_ackvec_lookup(struct list_head *av_list,
-                                                    const u64 ackno)
+struct dccp_ackvec *dccp_ackvec_alloc(const gfp_t priority)
 {
-       struct dccp_ackvec_record *avr;
-       /*
-        * Exploit that records are inserted in descending order of sequence
-        * number, start with the oldest record first. If @ackno is `before'
-        * the earliest ack_ackno, the packet is too old to be considered.
-        */
-       list_for_each_entry_reverse(avr, av_list, avr_node) {
-               if (avr->avr_ack_seqno == ackno)
-                       return avr;
-               if (before48(ackno, avr->avr_ack_seqno))
-                       break;
+       struct dccp_ackvec *av = kmem_cache_alloc(dccp_ackvec_slab, priority);
+
+       if (av != NULL) {
+               av->av_buf_head  = DCCP_MAX_ACKVEC_LEN - 1;
+               av->av_buf_ackno = UINT48_MAX + 1;
+               av->av_buf_nonce = 0;
+               av->av_time      = ktime_set(0, 0);
+               av->av_vec_len   = 0;
+               INIT_LIST_HEAD(&av->av_records);
        }
-       return NULL;
+
+       return av;
 }
 
-/*
- * Buffer index and length computation using modulo-buffersize arithmetic.
- * Note that, as pointers move from right to left, head is `before' tail.
- */
-static inline u16 __ackvec_idx_add(const u16 a, const u16 b)
+void dccp_ackvec_free(struct dccp_ackvec *av)
 {
-       return (a + b) % DCCPAV_MAX_ACKVEC_LEN;
+       if (unlikely(av == NULL))
+               return;
+
+       if (!list_empty(&av->av_records)) {
+               struct dccp_ackvec_record *avr, *next;
+
+               list_for_each_entry_safe(avr, next, &av->av_records, avr_node) {
+                       list_del_init(&avr->avr_node);
+                       dccp_ackvec_record_delete(avr);
+               }
+       }
+
+       kmem_cache_free(dccp_ackvec_slab, av);
 }
 
-static inline u16 __ackvec_idx_sub(const u16 a, const u16 b)
+static inline u8 dccp_ackvec_state(const struct dccp_ackvec *av,
+                                  const u32 index)
 {
-       return __ackvec_idx_add(a, DCCPAV_MAX_ACKVEC_LEN - b);
+       return av->av_buf[index] & DCCP_ACKVEC_STATE_MASK;
 }
 
-u16 dccp_ackvec_buflen(const struct dccp_ackvec *av)
+static inline u8 dccp_ackvec_len(const struct dccp_ackvec *av,
+                                const u32 index)
 {
-       if (unlikely(av->av_overflow))
-               return DCCPAV_MAX_ACKVEC_LEN;
-       return __ackvec_idx_sub(av->av_buf_tail, av->av_buf_head);
+       return av->av_buf[index] & DCCP_ACKVEC_LEN_MASK;
 }
 
-/**
- * dccp_ackvec_update_old  -  Update previous state as per RFC 4340, 11.4.1
- * @av:                non-empty buffer to update
- * @distance:   negative or zero distance of @seqno from buf_ackno downward
- * @seqno:     the (old) sequence number whose record is to be updated
- * @state:     state in which packet carrying @seqno was received
+/*
+ * If several packets are missing, the HC-Receiver may prefer to enter multiple
+ * bytes with run length 0, rather than a single byte with a larger run length;
+ * this simplifies table updates if one of the missing packets arrives.
  */
-static void dccp_ackvec_update_old(struct dccp_ackvec *av, s64 distance,
-                                  u64 seqno, enum dccp_ackvec_states state)
+static inline int dccp_ackvec_set_buf_head_state(struct dccp_ackvec *av,
+                                                const unsigned int packets,
+                                                const unsigned char state)
 {
-       u16 ptr = av->av_buf_head;
+       long gap;
+       long new_head;
 
-       BUG_ON(distance > 0);
-       if (unlikely(dccp_ackvec_is_empty(av)))
-               return;
+       if (av->av_vec_len + packets > DCCP_MAX_ACKVEC_LEN)
+               return -ENOBUFS;
 
-       do {
-               u8 runlen = dccp_ackvec_runlen(av->av_buf + ptr);
+       gap      = packets - 1;
+       new_head = av->av_buf_head - packets;
 
-               if (distance + runlen >= 0) {
-                       /*
-                        * Only update the state if packet has not been received
-                        * yet. This is OK as per the second table in RFC 4340,
-                        * 11.4.1; i.e. here we are using the following table:
-                        *                     RECEIVED
-                        *                      0   1   3
-                        *              S     +---+---+---+
-                        *              T   0 | 0 | 0 | 0 |
-                        *              O     +---+---+---+
-                        *              R   1 | 1 | 1 | 1 |
-                        *              E     +---+---+---+
-                        *              D   3 | 0 | 1 | 3 |
-                        *                    +---+---+---+
-                        * The "Not Received" state was set by reserve_seats().
-                        */
-                       if (av->av_buf[ptr] == DCCPAV_NOT_RECEIVED)
-                               av->av_buf[ptr] = state;
-                       else
-                               dccp_pr_debug("Not changing %llu state to %u\n",
-                                             (unsigned long long)seqno, state);
-                       break;
+       if (new_head < 0) {
+               if (gap > 0) {
+                       memset(av->av_buf, DCCP_ACKVEC_STATE_NOT_RECEIVED,
+                              gap + new_head + 1);
+                       gap = -new_head;
                }
+               new_head += DCCP_MAX_ACKVEC_LEN;
+       }
 
-               distance += runlen + 1;
-               ptr       = __ackvec_idx_add(ptr, 1);
+       av->av_buf_head = new_head;
 
-       } while (ptr != av->av_buf_tail);
-}
+       if (gap > 0)
+               memset(av->av_buf + av->av_buf_head + 1,
+                      DCCP_ACKVEC_STATE_NOT_RECEIVED, gap);
 
-/* Mark @num entries after buf_head as "Not yet received". */
-static void dccp_ackvec_reserve_seats(struct dccp_ackvec *av, u16 num)
-{
-       u16 start = __ackvec_idx_add(av->av_buf_head, 1),
-           len   = DCCPAV_MAX_ACKVEC_LEN - start;
-
-       /* check for buffer wrap-around */
-       if (num > len) {
-               memset(av->av_buf + start, DCCPAV_NOT_RECEIVED, len);
-               start = 0;
-               num  -= len;
-       }
-       if (num)
-               memset(av->av_buf + start, DCCPAV_NOT_RECEIVED, num);
+       av->av_buf[av->av_buf_head] = state;
+       av->av_vec_len += packets;
+       return 0;
 }
 
-/**
- * dccp_ackvec_add_new  -  Record one or more new entries in Ack Vector buffer
- * @av:                 container of buffer to update (can be empty or non-empty)
- * @num_packets: number of packets to register (must be >= 1)
- * @seqno:      sequence number of the first packet in @num_packets
- * @state:      state in which packet carrying @seqno was received
+/*
+ * Implements the RFC 4340, Appendix A
  */
-static void dccp_ackvec_add_new(struct dccp_ackvec *av, u32 num_packets,
-                               u64 seqno, enum dccp_ackvec_states state)
+int dccp_ackvec_add(struct dccp_ackvec *av, const struct sock *sk,
+                   const u64 ackno, const u8 state)
 {
-       u32 num_cells = num_packets;
+       /*
+        * Check at the right places if the buffer is full, if it is, tell the
+        * caller to start dropping packets till the HC-Sender acks our ACK
+        * vectors, when we will free up space in av_buf.
+        *
+        * We may well decide to do buffer compression, etc, but for now lets
+        * just drop.
+        *
+        * From Appendix A.1.1 (`New Packets'):
+        *
+        *      Of course, the circular buffer may overflow, either when the
+        *      HC-Sender is sending data at a very high rate, when the
+        *      HC-Receiver's acknowledgements are not reaching the HC-Sender,
+        *      or when the HC-Sender is forgetting to acknowledge those acks
+        *      (so the HC-Receiver is unable to clean up old state). In this
+        *      case, the HC-Receiver should either compress the buffer (by
+        *      increasing run lengths when possible), transfer its state to
+        *      a larger buffer, or, as a last resort, drop all received
+        *      packets, without processing them whatsoever, until its buffer
+        *      shrinks again.
+        */
 
-       if (num_packets > DCCPAV_BURST_THRESH) {
-               u32 lost_packets = num_packets - 1;
+       /* See if this is the first ackno being inserted */
+       if (av->av_vec_len == 0) {
+               av->av_buf[av->av_buf_head] = state;
+               av->av_vec_len = 1;
+       } else if (after48(ackno, av->av_buf_ackno)) {
+               const u64 delta = dccp_delta_seqno(av->av_buf_ackno, ackno);
 
-               DCCP_WARN("Warning: large burst loss (%u)\n", lost_packets);
                /*
-                * We received 1 packet and have a loss of size "num_packets-1"
-                * which we squeeze into num_cells-1 rather than reserving an
-                * entire byte for each lost packet.
-                * The reason is that the vector grows in O(burst_length); when
-                * it grows too large there will no room left for the payload.
-                * This is a trade-off: if a few packets out of the burst show
-                * up later, their state will not be changed; it is simply too
-                * costly to reshuffle/reallocate/copy the buffer each time.
-                * Should such problems persist, we will need to switch to a
-                * different underlying data structure.
+                * Look if the state of this packet is the same as the
+                * previous ackno and if so if we can bump the head len.
                 */
-               for (num_packets = num_cells = 1; lost_packets; ++num_cells) {
-                       u8 len = min(lost_packets, (u32)DCCPAV_MAX_RUNLEN);
-
-                       av->av_buf_head = __ackvec_idx_sub(av->av_buf_head, 1);
-                       av->av_buf[av->av_buf_head] = DCCPAV_NOT_RECEIVED | len;
+               if (delta == 1 &&
+                   dccp_ackvec_state(av, av->av_buf_head) == state &&
+                   dccp_ackvec_len(av, av->av_buf_head) < DCCP_ACKVEC_LEN_MASK)
+                       av->av_buf[av->av_buf_head]++;
+               else if (dccp_ackvec_set_buf_head_state(av, delta, state))
+                       return -ENOBUFS;
+       } else {
+               /*
+                * A.1.2.  Old Packets
+                *
+                *      When a packet with Sequence Number S <= buf_ackno
+                *      arrives, the HC-Receiver will scan the table for
+                *      the byte corresponding to S. (Indexing structures
+                *      could reduce the complexity of this scan.)
+                */
+               u64 delta = dccp_delta_seqno(ackno, av->av_buf_ackno);
+               u32 index = av->av_buf_head;
 
-                       lost_packets -= len;
+               while (1) {
+                       const u8 len = dccp_ackvec_len(av, index);
+                       const u8 av_state = dccp_ackvec_state(av, index);
+                       /*
+                        * valid packets not yet in av_buf have a reserved
+                        * entry, with a len equal to 0.
+                        */
+                       if (av_state == DCCP_ACKVEC_STATE_NOT_RECEIVED &&
+                           len == 0 && delta == 0) { /* Found our
+                                                        reserved seat! */
+                               dccp_pr_debug("Found %llu reserved seat!\n",
+                                             (unsigned long long)ackno);
+                               av->av_buf[index] = state;
+                               goto out;
+                       }
+                       /* len == 0 means one packet */
+                       if (delta < len + 1)
+                               goto out_duplicate;
+
+                       delta -= len + 1;
+                       if (++index == DCCP_MAX_ACKVEC_LEN)
+                               index = 0;
                }
        }
 
-       if (num_cells + dccp_ackvec_buflen(av) >= DCCPAV_MAX_ACKVEC_LEN) {
-               DCCP_CRIT("Ack Vector buffer overflow: dropping old entries\n");
-               av->av_overflow = true;
-       }
-
-       av->av_buf_head = __ackvec_idx_sub(av->av_buf_head, num_packets);
-       if (av->av_overflow)
-               av->av_buf_tail = av->av_buf_head;
-
-       av->av_buf[av->av_buf_head] = state;
-       av->av_buf_ackno            = seqno;
+       av->av_buf_ackno = ackno;
+       av->av_time = ktime_get_real();
+out:
+       return 0;
 
-       if (num_packets > 1)
-               dccp_ackvec_reserve_seats(av, num_packets - 1);
+out_duplicate:
+       /* Duplicate packet */
+       dccp_pr_debug("Received a dup or already considered lost "
+                     "packet: %llu\n", (unsigned long long)ackno);
+       return -EILSEQ;
 }
 
-/**
- * dccp_ackvec_input  -  Register incoming packet in the buffer
- */
-void dccp_ackvec_input(struct dccp_ackvec *av, struct sk_buff *skb)
+static void dccp_ackvec_throw_record(struct dccp_ackvec *av,
+                                    struct dccp_ackvec_record *avr)
 {
-       u64 seqno = DCCP_SKB_CB(skb)->dccpd_seq;
-       enum dccp_ackvec_states state = DCCPAV_RECEIVED;
+       struct dccp_ackvec_record *next;
 
-       if (dccp_ackvec_is_empty(av)) {
-               dccp_ackvec_add_new(av, 1, seqno, state);
-               av->av_tail_ackno = seqno;
+       /* sort out vector length */
+       if (av->av_buf_head <= avr->avr_ack_ptr)
+               av->av_vec_len = avr->avr_ack_ptr - av->av_buf_head;
+       else
+               av->av_vec_len = DCCP_MAX_ACKVEC_LEN - 1 -
+                                av->av_buf_head + avr->avr_ack_ptr;
 
-       } else {
-               s64 num_packets = dccp_delta_seqno(av->av_buf_ackno, seqno);
-               u8 *current_head = av->av_buf + av->av_buf_head;
-
-               if (num_packets == 1 &&
-                   dccp_ackvec_state(current_head) == state &&
-                   dccp_ackvec_runlen(current_head) < DCCPAV_MAX_RUNLEN) {
+       /* free records */
+       list_for_each_entry_safe_from(avr, next, &av->av_records, avr_node) {
+               list_del_init(&avr->avr_node);
+               dccp_ackvec_record_delete(avr);
+       }
+}
 
-                       *current_head   += 1;
-                       av->av_buf_ackno = seqno;
+void dccp_ackvec_check_rcv_ackno(struct dccp_ackvec *av, struct sock *sk,
+                                const u64 ackno)
+{
+       struct dccp_ackvec_record *avr;
 
-               } else if (num_packets > 0) {
-                       dccp_ackvec_add_new(av, num_packets, seqno, state);
-               } else {
-                       dccp_ackvec_update_old(av, num_packets, seqno, state);
-               }
+       /*
+        * If we traverse backwards, it should be faster when we have large
+        * windows. We will be receiving ACKs for stuff we sent a while back
+        * -sorbo.
+        */
+       list_for_each_entry_reverse(avr, &av->av_records, avr_node) {
+               if (ackno == avr->avr_ack_seqno) {
+                       dccp_pr_debug("%s ACK packet 0, len=%d, ack_seqno=%llu, "
+                                     "ack_ackno=%llu, ACKED!\n",
+                                     dccp_role(sk), 1,
+                                     (unsigned long long)avr->avr_ack_seqno,
+                                     (unsigned long long)avr->avr_ack_ackno);
+                       dccp_ackvec_throw_record(av, avr);
+                       break;
+               } else if (avr->avr_ack_seqno > ackno)
+                       break; /* old news */
        }
 }
 
-/**
- * dccp_ackvec_clear_state  -  Perform house-keeping / garbage-collection
- * This routine is called when the peer acknowledges the receipt of Ack Vectors
- * up to and including @ackno. While based on on section A.3 of RFC 4340, here
- * are additional precautions to prevent corrupted buffer state. In particular,
- * we use tail_ackno to identify outdated records; it always marks the earliest
- * packet of group (2) in 11.4.2.
- */
-void dccp_ackvec_clear_state(struct dccp_ackvec *av, const u64 ackno)
- {
-       struct dccp_ackvec_record *avr, *next;
-       u8 runlen_now, eff_runlen;
-       s64 delta;
+static void dccp_ackvec_check_rcv_ackvector(struct dccp_ackvec *av,
+                                           struct sock *sk, u64 *ackno,
+                                           const unsigned char len,
+                                           const unsigned char *vector)
+{
+       unsigned char i;
+       struct dccp_ackvec_record *avr;
 
-       avr = dccp_ackvec_lookup(&av->av_records, ackno);
-       if (avr == NULL)
+       /* Check if we actually sent an ACK vector */
+       if (list_empty(&av->av_records))
                return;
-       /*
-        * Deal with outdated acknowledgments: this arises when e.g. there are
-        * several old records and the acks from the peer come in slowly. In
-        * that case we may still have records that pre-date tail_ackno.
-        */
-       delta = dccp_delta_seqno(av->av_tail_ackno, avr->avr_ack_ackno);
-       if (delta < 0)
-               goto free_records;
-       /*
-        * Deal with overlapping Ack Vectors: don't subtract more than the
-        * number of packets between tail_ackno and ack_ackno.
-        */
-       eff_runlen = delta < avr->avr_ack_runlen ? delta : avr->avr_ack_runlen;
 
-       runlen_now = dccp_ackvec_runlen(av->av_buf + avr->avr_ack_ptr);
+       i = len;
        /*
-        * The run length of Ack Vector cells does not decrease over time. If
-        * the run length is the same as at the time the Ack Vector was sent, we
-        * free the ack_ptr cell. That cell can however not be freed if the run
-        * length has increased: in this case we need to move the tail pointer
-        * backwards (towards higher indices), to its next-oldest neighbour.
+        * XXX
+        * I think it might be more efficient to work backwards. See comment on
+        * rcv_ackno. -sorbo.
         */
-       if (runlen_now > eff_runlen) {
+       avr = list_entry(av->av_records.next, struct dccp_ackvec_record, avr_node);
+       while (i--) {
+               const u8 rl = *vector & DCCP_ACKVEC_LEN_MASK;
+               u64 ackno_end_rl;
 
-               av->av_buf[avr->avr_ack_ptr] -= eff_runlen + 1;
-               av->av_buf_tail = __ackvec_idx_add(avr->avr_ack_ptr, 1);
+               dccp_set_seqno(&ackno_end_rl, *ackno - rl);
 
-               /* This move may not have cleared the overflow flag. */
-               if (av->av_overflow)
-                       av->av_overflow = (av->av_buf_head == av->av_buf_tail);
-       } else {
-               av->av_buf_tail = avr->avr_ack_ptr;
                /*
-                * We have made sure that avr points to a valid cell within the
-                * buffer. This cell is either older than head, or equals head
-                * (empty buffer): in both cases we no longer have any overflow.
+                * If our AVR sequence number is greater than the ack, go
+                * forward in the AVR list until it is not so.
                 */
-               av->av_overflow = 0;
-       }
-
-       /*
-        * The peer has acknowledged up to and including ack_ackno. Hence the
-        * first packet in group (2) of 11.4.2 is the successor of ack_ackno.
-        */
-       av->av_tail_ackno = ADD48(avr->avr_ack_ackno, 1);
+               list_for_each_entry_from(avr, &av->av_records, avr_node) {
+                       if (!after48(avr->avr_ack_seqno, *ackno))
+                               goto found;
+               }
+               /* End of the av_records list, not found, exit */
+               break;
+found:
+               if (between48(avr->avr_ack_seqno, ackno_end_rl, *ackno)) {
+                       const u8 state = *vector & DCCP_ACKVEC_STATE_MASK;
+                       if (state != DCCP_ACKVEC_STATE_NOT_RECEIVED) {
+                               dccp_pr_debug("%s ACK vector 0, len=%d, "
+                                             "ack_seqno=%llu, ack_ackno=%llu, "
+                                             "ACKED!\n",
+                                             dccp_role(sk), len,
+                                             (unsigned long long)
+                                             avr->avr_ack_seqno,
+                                             (unsigned long long)
+                                             avr->avr_ack_ackno);
+                               dccp_ackvec_throw_record(av, avr);
+                               break;
+                       }
+                       /*
+                        * If it wasn't received, continue scanning... we might
+                        * find another one.
+                        */
+               }
 
-free_records:
-       list_for_each_entry_safe_from(avr, next, &av->av_records, avr_node) {
-               list_del(&avr->avr_node);
-               kmem_cache_free(dccp_ackvec_record_slab, avr);
+               dccp_set_seqno(ackno, ackno_end_rl - 1);
+               ++vector;
        }
 }
 
-/*
- *     Routines to keep track of Ack Vectors received in an skb
- */
-int dccp_ackvec_parsed_add(struct list_head *head, u8 *vec, u8 len, u8 nonce)
+int dccp_ackvec_parse(struct sock *sk, const struct sk_buff *skb,
+                     u64 *ackno, const u8 opt, const u8 *value, const u8 len)
 {
-       struct dccp_ackvec_parsed *new = kmalloc(sizeof(*new), GFP_ATOMIC);
-
-       if (new == NULL)
-               return -ENOBUFS;
-       new->vec   = vec;
-       new->len   = len;
-       new->nonce = nonce;
+       if (len > DCCP_SINGLE_OPT_MAXLEN)
+               return -1;
 
-       list_add_tail(&new->node, head);
+       /* dccp_ackvector_print(DCCP_SKB_CB(skb)->dccpd_ack_seq, value, len); */
+       dccp_ackvec_check_rcv_ackvector(dccp_sk(sk)->dccps_hc_rx_ackvec, sk,
+                                       ackno, len, value);
        return 0;
 }
-EXPORT_SYMBOL_GPL(dccp_ackvec_parsed_add);
-
-void dccp_ackvec_parsed_cleanup(struct list_head *parsed_chunks)
-{
-       struct dccp_ackvec_parsed *cur, *next;
-
-       list_for_each_entry_safe(cur, next, parsed_chunks, node)
-               kfree(cur);
-       INIT_LIST_HEAD(parsed_chunks);
-}
-EXPORT_SYMBOL_GPL(dccp_ackvec_parsed_cleanup);
 
 int __init dccp_ackvec_init(void)
 {
@@ -379,9 +448,10 @@ int __init dccp_ackvec_init(void)
        if (dccp_ackvec_slab == NULL)
                goto out_err;
 
-       dccp_ackvec_record_slab = kmem_cache_create("dccp_ackvec_record",
-                                            sizeof(struct dccp_ackvec_record),
-                                            0, SLAB_HWCACHE_ALIGN, NULL);
+       dccp_ackvec_record_slab =
+                       kmem_cache_create("dccp_ackvec_record",
+                                         sizeof(struct dccp_ackvec_record),
+                                         0, SLAB_HWCACHE_ALIGN, NULL);
        if (dccp_ackvec_record_slab == NULL)
                goto out_destroy_slab;