[DCCP] ackvec: Introduce ack vector records
Based on a patch by Andrea Bittau.
Signed-off-by: Andrea Bittau <a.bittau@cs.ucl.ac.uk>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/net/dccp/ackvec.c b/net/dccp/ackvec.c
index 6440825..b4ff14f 100644
--- a/net/dccp/ackvec.c
+++ b/net/dccp/ackvec.c
@@ -22,6 +22,47 @@
#include <net/sock.h>
static kmem_cache_t *dccp_ackvec_slab;
+static kmem_cache_t *dccp_ackvec_record_slab;
+
+static struct dccp_ackvec_record *dccp_ackvec_record_new(void)
+{
+ struct dccp_ackvec_record *avr =
+ kmem_cache_alloc(dccp_ackvec_record_slab, GFP_ATOMIC);
+
+ if (avr != NULL)
+ INIT_LIST_HEAD(&avr->dccpavr_node);
+
+ return avr;
+}
+
+static void dccp_ackvec_record_delete(struct dccp_ackvec_record *avr)
+{
+ if (unlikely(avr == NULL))
+ return;
+ /* Check if deleting a linked record */
+ WARN_ON(!list_empty(&avr->dccpavr_node));
+ kmem_cache_free(dccp_ackvec_record_slab, avr);
+}
+
+static void dccp_ackvec_insert_avr(struct dccp_ackvec *av,
+ struct dccp_ackvec_record *avr)
+{
+ /*
+ * 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->dccpav_records)) {
+ const struct dccp_ackvec_record *head =
+ list_entry(av->dccpav_records.next,
+ struct dccp_ackvec_record,
+ dccpavr_node);
+ BUG_ON(before48(avr->dccpavr_ack_seqno,
+ head->dccpavr_ack_seqno));
+ }
+
+ list_add(&avr->dccpavr_node, &av->dccpav_records);
+}
int dccp_insert_option_ackvec(struct sock *sk, struct sk_buff *skb)
{
@@ -35,6 +76,14 @@
struct timeval now;
u32 elapsed_time;
unsigned char *to, *from;
+ struct dccp_ackvec_record *avr;
+
+ if (DCCP_SKB_CB(skb)->dccpd_opt_len + len > DCCP_MAX_OPT_LEN)
+ return -1;
+
+ avr = dccp_ackvec_record_new();
+ if (avr == NULL)
+ return -1;
dccp_timestamp(sk, &now);
elapsed_time = timeval_delta(&now, &av->dccpav_time) / 10;
@@ -42,19 +91,6 @@
if (elapsed_time != 0)
dccp_insert_option_elapsed_time(sk, skb, elapsed_time);
- if (DCCP_SKB_CB(skb)->dccpd_opt_len + len > DCCP_MAX_OPT_LEN)
- return -1;
-
- /*
- * XXX: now we have just one ack vector sent record, so
- * we have to wait for it to be cleared.
- *
- * Of course this is not acceptable, but this is just for
- * basic testing now.
- */
- if (av->dccpav_ack_seqno != DCCP_MAX_SEQNO + 1)
- return -1;
-
DCCP_SKB_CB(skb)->dccpd_opt_len += len;
to = skb_push(skb, len);
@@ -65,8 +101,8 @@
from = av->dccpav_buf + av->dccpav_buf_head;
/* Check if buf_head wraps */
- if ((int)av->dccpav_buf_head + len > av->dccpav_vec_len) {
- const u32 tailsize = av->dccpav_vec_len - av->dccpav_buf_head;
+ if ((int)av->dccpav_buf_head + len > DCCP_MAX_ACKVEC_LEN) {
+ const u32 tailsize = DCCP_MAX_ACKVEC_LEN - av->dccpav_buf_head;
memcpy(to, from, tailsize);
to += tailsize;
@@ -83,21 +119,21 @@
* 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.
- *
- * This implemention uses just one ack record for now.
*/
- av->dccpav_ack_seqno = DCCP_SKB_CB(skb)->dccpd_seq;
- av->dccpav_ack_ptr = av->dccpav_buf_head;
- av->dccpav_ack_ackno = av->dccpav_buf_ackno;
- av->dccpav_ack_nonce = av->dccpav_buf_nonce;
- av->dccpav_sent_len = av->dccpav_vec_len;
+ avr->dccpavr_ack_seqno = DCCP_SKB_CB(skb)->dccpd_seq;
+ avr->dccpavr_ack_ptr = av->dccpav_buf_head;
+ avr->dccpavr_ack_ackno = av->dccpav_buf_ackno;
+ avr->dccpavr_ack_nonce = av->dccpav_buf_nonce;
+ avr->dccpavr_sent_len = av->dccpav_vec_len;
+
+ dccp_ackvec_insert_avr(av, avr);
dccp_pr_debug("%sACK Vector 0, len=%d, ack_seqno=%llu, "
"ack_ackno=%llu\n",
- debug_prefix, av->dccpav_sent_len,
- (unsigned long long)av->dccpav_ack_seqno,
- (unsigned long long)av->dccpav_ack_ackno);
- return -1;
+ debug_prefix, avr->dccpavr_sent_len,
+ (unsigned long long)avr->dccpavr_ack_seqno,
+ (unsigned long long)avr->dccpavr_ack_ackno);
+ return 0;
}
struct dccp_ackvec *dccp_ackvec_alloc(const gfp_t priority)
@@ -107,13 +143,13 @@
if (av != NULL) {
av->dccpav_buf_head =
av->dccpav_buf_tail = DCCP_MAX_ACKVEC_LEN - 1;
- av->dccpav_buf_ackno =
- av->dccpav_ack_ackno = av->dccpav_ack_seqno = ~0LLU;
+ av->dccpav_buf_ackno = DCCP_MAX_SEQNO + 1;
av->dccpav_buf_nonce = av->dccpav_buf_nonce = 0;
av->dccpav_ack_ptr = 0;
av->dccpav_time.tv_sec = 0;
av->dccpav_time.tv_usec = 0;
av->dccpav_sent_len = av->dccpav_vec_len = 0;
+ INIT_LIST_HEAD(&av->dccpav_records);
}
return av;
@@ -121,6 +157,9 @@
void dccp_ackvec_free(struct dccp_ackvec *av)
{
+ if (unlikely(av == NULL))
+ return;
+ WARN_ON(!list_empty(&av->dccpav_records));
kmem_cache_free(dccp_ackvec_slab, av);
}
@@ -299,44 +338,50 @@
}
#endif
-static void dccp_ackvec_throw_away_ack_record(struct dccp_ackvec *av)
+static void dccp_ackvec_throw_record(struct dccp_ackvec *av,
+ struct dccp_ackvec_record *avr)
{
- /*
- * As we're keeping track of the ack vector size (dccpav_vec_len) and
- * the sent ack vector size (dccpav_sent_len) we don't need
- * dccpav_buf_tail at all, but keep this code here as in the future
- * we'll implement a vector of ack records, as suggested in
- * draft-ietf-dccp-spec-11.txt Appendix A. -acme
- */
-#if 0
- u32 new_buf_tail = av->dccpav_ack_ptr + 1;
- if (new_buf_tail >= av->dccpav_vec_len)
- new_buf_tail -= av->dccpav_vec_len;
- av->dccpav_buf_tail = new_buf_tail;
-#endif
- av->dccpav_vec_len -= av->dccpav_sent_len;
+ struct dccp_ackvec_record *next;
+
+ av->dccpav_buf_tail = avr->dccpavr_ack_ptr - 1;
+ if (av->dccpav_buf_tail == 0)
+ av->dccpav_buf_tail = DCCP_MAX_ACKVEC_LEN - 1;
+
+ av->dccpav_vec_len -= avr->dccpavr_sent_len;
+
+ /* free records */
+ list_for_each_entry_safe_from(avr, next, &av->dccpav_records,
+ dccpavr_node) {
+ list_del_init(&avr->dccpavr_node);
+ dccp_ackvec_record_delete(avr);
+ }
}
void dccp_ackvec_check_rcv_ackno(struct dccp_ackvec *av, struct sock *sk,
const u64 ackno)
{
- /* Check if we actually sent an ACK vector */
- if (av->dccpav_ack_seqno == DCCP_MAX_SEQNO + 1)
- return;
+ struct dccp_ackvec_record *avr;
- if (ackno == av->dccpav_ack_seqno) {
+ /*
+ * 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->dccpav_records, dccpavr_node) {
+ if (ackno == avr->dccpavr_ack_seqno) {
#ifdef CONFIG_IP_DCCP_DEBUG
- struct dccp_sock *dp = dccp_sk(sk);
- const char *debug_prefix = dp->dccps_role == DCCP_ROLE_CLIENT ?
- "CLIENT rx ack: " : "server rx ack: ";
+ struct dccp_sock *dp = dccp_sk(sk);
+ const char *debug_prefix = dp->dccps_role == DCCP_ROLE_CLIENT ?
+ "CLIENT rx ack: " : "server rx ack: ";
#endif
- dccp_pr_debug("%sACK packet 0, len=%d, ack_seqno=%llu, "
- "ack_ackno=%llu, ACKED!\n",
- debug_prefix, 1,
- (unsigned long long)av->dccpav_ack_seqno,
- (unsigned long long)av->dccpav_ack_ackno);
- dccp_ackvec_throw_away_ack_record(av);
- av->dccpav_ack_seqno = DCCP_MAX_SEQNO + 1;
+ dccp_pr_debug("%sACK packet 0, len=%d, ack_seqno=%llu, "
+ "ack_ackno=%llu, ACKED!\n",
+ debug_prefix, 1,
+ (unsigned long long)avr->dccpavr_ack_seqno,
+ (unsigned long long)avr->dccpavr_ack_ackno);
+ dccp_ackvec_throw_record(av, avr);
+ break;
+ }
}
}
@@ -346,28 +391,20 @@
const unsigned char *vector)
{
unsigned char i;
+ struct dccp_ackvec_record *avr;
/* Check if we actually sent an ACK vector */
- if (av->dccpav_ack_seqno == DCCP_MAX_SEQNO + 1)
+ if (list_empty(&av->dccpav_records))
return;
- /*
- * We're in the receiver half connection, so if the received an ACK
- * vector ackno (e.g. 50) before dccpav_ack_seqno (e.g. 52), we're
- * not interested.
- *
- * Extra explanation with example:
- *
- * if we received an ACK vector with ackno 50, it can only be acking
- * 50, 49, 48, etc, not 52 (the seqno for the ACK vector we sent).
- */
- /* dccp_pr_debug("is %llu < %llu? ", ackno, av->dccpav_ack_seqno); */
- if (before48(ackno, av->dccpav_ack_seqno)) {
- /* dccp_pr_debug_cat("yes\n"); */
- return;
- }
- /* dccp_pr_debug_cat("no\n"); */
i = len;
+ /*
+ * XXX
+ * I think it might be more efficient to work backwards. See comment on
+ * rcv_ackno. -sorbo.
+ */
+ avr = list_entry(av->dccpav_records.next, struct dccp_ackvec_record,
+ dccpavr_node);
while (i--) {
const u8 rl = *vector & DCCP_ACKVEC_LEN_MASK;
u64 ackno_end_rl;
@@ -375,14 +412,20 @@
dccp_set_seqno(&ackno_end_rl, ackno - rl);
/*
- * dccp_pr_debug("is %llu <= %llu <= %llu? ", ackno_end_rl,
- * av->dccpav_ack_seqno, ackno);
+ * If our AVR sequence number is greater than the ack, go
+ * forward in the AVR list until it is not so.
*/
- if (between48(av->dccpav_ack_seqno, ackno_end_rl, ackno)) {
+ list_for_each_entry_from(avr, &av->dccpav_records,
+ dccpavr_node) {
+ if (!after48(avr->dccpavr_ack_seqno, ackno))
+ goto found;
+ }
+ /* End of the dccpav_records list, not found, exit */
+ break;
+found:
+ if (between48(avr->dccpavr_ack_seqno, ackno_end_rl, ackno)) {
const u8 state = (*vector &
DCCP_ACKVEC_STATE_MASK) >> 6;
- /* dccp_pr_debug_cat("yes\n"); */
-
if (state != DCCP_ACKVEC_STATE_NOT_RECEIVED) {
#ifdef CONFIG_IP_DCCP_DEBUG
struct dccp_sock *dp = dccp_sk(sk);
@@ -395,19 +438,16 @@
"ACKED!\n",
debug_prefix, len,
(unsigned long long)
- av->dccpav_ack_seqno,
+ avr->dccpavr_ack_seqno,
(unsigned long long)
- av->dccpav_ack_ackno);
- dccp_ackvec_throw_away_ack_record(av);
+ avr->dccpavr_ack_ackno);
+ dccp_ackvec_throw_record(av, avr);
}
/*
- * If dccpav_ack_seqno was not received, no problem
- * we'll send another ACK vector.
+ * If it wasn't received, continue scanning... we might
+ * find another one.
*/
- av->dccpav_ack_seqno = DCCP_MAX_SEQNO + 1;
- break;
}
- /* dccp_pr_debug_cat("no\n"); */
dccp_set_seqno(&ackno, ackno_end_rl - 1);
++vector;
@@ -428,19 +468,31 @@
}
static char dccp_ackvec_slab_msg[] __initdata =
- KERN_CRIT "DCCP: Unable to create ack vectors slab cache\n";
+ KERN_CRIT "DCCP: Unable to create ack vectors slab caches\n";
int __init dccp_ackvec_init(void)
{
dccp_ackvec_slab = kmem_cache_create("dccp_ackvec",
sizeof(struct dccp_ackvec), 0,
SLAB_HWCACHE_ALIGN, NULL, NULL);
- if (dccp_ackvec_slab == NULL) {
- printk(dccp_ackvec_slab_msg);
- return -ENOBUFS;
- }
+ 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, NULL);
+ if (dccp_ackvec_record_slab == NULL)
+ goto out_destroy_slab;
return 0;
+
+out_destroy_slab:
+ kmem_cache_destroy(dccp_ackvec_slab);
+ dccp_ackvec_slab = NULL;
+out_err:
+ printk(dccp_ackvec_slab_msg);
+ return -ENOBUFS;
}
void dccp_ackvec_exit(void)
@@ -449,4 +501,8 @@
kmem_cache_destroy(dccp_ackvec_slab);
dccp_ackvec_slab = NULL;
}
+ if (dccp_ackvec_record_slab != NULL) {
+ kmem_cache_destroy(dccp_ackvec_record_slab);
+ dccp_ackvec_record_slab = NULL;
+ }
}