4 * An implementation of the DCCP protocol
5 * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
7 * This program is free software; you can redistribute it and/or modify it
8 * under the terms of the GNU General Public License as published by the
9 * Free Software Foundation; version 2 of the License;
15 #include <linux/dccp.h>
16 #include <linux/init.h>
17 #include <linux/errno.h>
18 #include <linux/kernel.h>
19 #include <linux/skbuff.h>
20 #include <linux/slab.h>
24 static struct kmem_cache
*dccp_ackvec_slab
;
25 static struct kmem_cache
*dccp_ackvec_record_slab
;
27 static struct dccp_ackvec_record
*dccp_ackvec_record_new(void)
29 struct dccp_ackvec_record
*avr
=
30 kmem_cache_alloc(dccp_ackvec_record_slab
, GFP_ATOMIC
);
33 INIT_LIST_HEAD(&avr
->avr_node
);
38 static void dccp_ackvec_record_delete(struct dccp_ackvec_record
*avr
)
40 if (unlikely(avr
== NULL
))
42 /* Check if deleting a linked record */
43 WARN_ON(!list_empty(&avr
->avr_node
));
44 kmem_cache_free(dccp_ackvec_record_slab
, avr
);
47 static void dccp_ackvec_insert_avr(struct dccp_ackvec
*av
,
48 struct dccp_ackvec_record
*avr
)
51 * AVRs are sorted by seqno. Since we are sending them in order, we
52 * just add the AVR at the head of the list.
55 if (!list_empty(&av
->av_records
)) {
56 const struct dccp_ackvec_record
*head
=
57 list_entry(av
->av_records
.next
,
58 struct dccp_ackvec_record
,
60 BUG_ON(before48(avr
->avr_ack_seqno
, head
->avr_ack_seqno
));
63 list_add(&avr
->avr_node
, &av
->av_records
);
66 int dccp_insert_option_ackvec(struct sock
*sk
, struct sk_buff
*skb
)
68 struct dccp_sock
*dp
= dccp_sk(sk
);
69 struct dccp_ackvec
*av
= dp
->dccps_hc_rx_ackvec
;
70 /* Figure out how many options do we need to represent the ackvec */
71 const u16 nr_opts
= DIV_ROUND_UP(av
->av_vec_len
, DCCP_MAX_ACKVEC_OPT_LEN
);
72 u16 len
= av
->av_vec_len
+ 2 * nr_opts
, i
;
74 const unsigned char *tail
, *from
;
76 struct dccp_ackvec_record
*avr
;
79 if (DCCP_SKB_CB(skb
)->dccpd_opt_len
+ len
> DCCP_MAX_OPT_LEN
)
82 delta
= ktime_us_delta(ktime_get_real(), av
->av_time
);
83 elapsed_time
= delta
/ 10;
85 if (elapsed_time
!= 0 &&
86 dccp_insert_option_elapsed_time(sk
, skb
, elapsed_time
))
89 avr
= dccp_ackvec_record_new();
93 DCCP_SKB_CB(skb
)->dccpd_opt_len
+= len
;
95 to
= skb_push(skb
, len
);
97 from
= av
->av_buf
+ av
->av_buf_head
;
98 tail
= av
->av_buf
+ DCCP_MAX_ACKVEC_LEN
;
100 for (i
= 0; i
< nr_opts
; ++i
) {
103 if (len
> DCCP_MAX_ACKVEC_OPT_LEN
)
104 copylen
= DCCP_MAX_ACKVEC_OPT_LEN
;
106 *to
++ = DCCPO_ACK_VECTOR_0
;
109 /* Check if buf_head wraps */
110 if (from
+ copylen
> tail
) {
111 const u16 tailsize
= tail
- from
;
113 memcpy(to
, from
, tailsize
);
120 memcpy(to
, from
, copylen
);
127 * From RFC 4340, A.2:
129 * For each acknowledgement it sends, the HC-Receiver will add an
130 * acknowledgement record. ack_seqno will equal the HC-Receiver
131 * sequence number it used for the ack packet; ack_ptr will equal
132 * buf_head; ack_ackno will equal buf_ackno; and ack_nonce will
135 avr
->avr_ack_seqno
= DCCP_SKB_CB(skb
)->dccpd_seq
;
136 avr
->avr_ack_ptr
= av
->av_buf_head
;
137 avr
->avr_ack_ackno
= av
->av_buf_ackno
;
138 avr
->avr_ack_nonce
= av
->av_buf_nonce
;
139 avr
->avr_sent_len
= av
->av_vec_len
;
141 dccp_ackvec_insert_avr(av
, avr
);
143 dccp_pr_debug("%s ACK Vector 0, len=%d, ack_seqno=%llu, "
145 dccp_role(sk
), avr
->avr_sent_len
,
146 (unsigned long long)avr
->avr_ack_seqno
,
147 (unsigned long long)avr
->avr_ack_ackno
);
151 struct dccp_ackvec
*dccp_ackvec_alloc(const gfp_t priority
)
153 struct dccp_ackvec
*av
= kmem_cache_alloc(dccp_ackvec_slab
, priority
);
156 av
->av_buf_head
= DCCP_MAX_ACKVEC_LEN
- 1;
157 av
->av_buf_ackno
= UINT48_MAX
+ 1;
158 av
->av_buf_nonce
= 0;
159 av
->av_time
= ktime_set(0, 0);
161 INIT_LIST_HEAD(&av
->av_records
);
167 void dccp_ackvec_free(struct dccp_ackvec
*av
)
169 if (unlikely(av
== NULL
))
172 if (!list_empty(&av
->av_records
)) {
173 struct dccp_ackvec_record
*avr
, *next
;
175 list_for_each_entry_safe(avr
, next
, &av
->av_records
, avr_node
) {
176 list_del_init(&avr
->avr_node
);
177 dccp_ackvec_record_delete(avr
);
181 kmem_cache_free(dccp_ackvec_slab
, av
);
184 static inline u8
dccp_ackvec_state(const struct dccp_ackvec
*av
,
187 return av
->av_buf
[index
] & DCCP_ACKVEC_STATE_MASK
;
190 static inline u8
dccp_ackvec_len(const struct dccp_ackvec
*av
,
193 return av
->av_buf
[index
] & DCCP_ACKVEC_LEN_MASK
;
197 * If several packets are missing, the HC-Receiver may prefer to enter multiple
198 * bytes with run length 0, rather than a single byte with a larger run length;
199 * this simplifies table updates if one of the missing packets arrives.
201 static inline int dccp_ackvec_set_buf_head_state(struct dccp_ackvec
*av
,
202 const unsigned int packets
,
203 const unsigned char state
)
208 if (av
->av_vec_len
+ packets
> DCCP_MAX_ACKVEC_LEN
)
212 new_head
= av
->av_buf_head
- packets
;
216 memset(av
->av_buf
, DCCP_ACKVEC_STATE_NOT_RECEIVED
,
220 new_head
+= DCCP_MAX_ACKVEC_LEN
;
223 av
->av_buf_head
= new_head
;
226 memset(av
->av_buf
+ av
->av_buf_head
+ 1,
227 DCCP_ACKVEC_STATE_NOT_RECEIVED
, gap
);
229 av
->av_buf
[av
->av_buf_head
] = state
;
230 av
->av_vec_len
+= packets
;
235 * Implements the RFC 4340, Appendix A
237 int dccp_ackvec_add(struct dccp_ackvec
*av
, const struct sock
*sk
,
238 const u64 ackno
, const u8 state
)
241 * Check at the right places if the buffer is full, if it is, tell the
242 * caller to start dropping packets till the HC-Sender acks our ACK
243 * vectors, when we will free up space in av_buf.
245 * We may well decide to do buffer compression, etc, but for now lets
248 * From Appendix A.1.1 (`New Packets'):
250 * Of course, the circular buffer may overflow, either when the
251 * HC-Sender is sending data at a very high rate, when the
252 * HC-Receiver's acknowledgements are not reaching the HC-Sender,
253 * or when the HC-Sender is forgetting to acknowledge those acks
254 * (so the HC-Receiver is unable to clean up old state). In this
255 * case, the HC-Receiver should either compress the buffer (by
256 * increasing run lengths when possible), transfer its state to
257 * a larger buffer, or, as a last resort, drop all received
258 * packets, without processing them whatsoever, until its buffer
262 /* See if this is the first ackno being inserted */
263 if (av
->av_vec_len
== 0) {
264 av
->av_buf
[av
->av_buf_head
] = state
;
266 } else if (after48(ackno
, av
->av_buf_ackno
)) {
267 const u64 delta
= dccp_delta_seqno(av
->av_buf_ackno
, ackno
);
270 * Look if the state of this packet is the same as the
271 * previous ackno and if so if we can bump the head len.
274 dccp_ackvec_state(av
, av
->av_buf_head
) == state
&&
275 dccp_ackvec_len(av
, av
->av_buf_head
) < DCCP_ACKVEC_LEN_MASK
)
276 av
->av_buf
[av
->av_buf_head
]++;
277 else if (dccp_ackvec_set_buf_head_state(av
, delta
, state
))
283 * When a packet with Sequence Number S <= buf_ackno
284 * arrives, the HC-Receiver will scan the table for
285 * the byte corresponding to S. (Indexing structures
286 * could reduce the complexity of this scan.)
288 u64 delta
= dccp_delta_seqno(ackno
, av
->av_buf_ackno
);
289 u32 index
= av
->av_buf_head
;
292 const u8 len
= dccp_ackvec_len(av
, index
);
293 const u8 state
= dccp_ackvec_state(av
, index
);
295 * valid packets not yet in av_buf have a reserved
296 * entry, with a len equal to 0.
298 if (state
== DCCP_ACKVEC_STATE_NOT_RECEIVED
&&
299 len
== 0 && delta
== 0) { /* Found our
301 dccp_pr_debug("Found %llu reserved seat!\n",
302 (unsigned long long)ackno
);
303 av
->av_buf
[index
] = state
;
306 /* len == 0 means one packet */
311 if (++index
== DCCP_MAX_ACKVEC_LEN
)
316 av
->av_buf_ackno
= ackno
;
317 av
->av_time
= ktime_get_real();
322 /* Duplicate packet */
323 dccp_pr_debug("Received a dup or already considered lost "
324 "packet: %llu\n", (unsigned long long)ackno
);
328 #ifdef CONFIG_IP_DCCP_DEBUG
329 void dccp_ackvector_print(const u64 ackno
, const unsigned char *vector
, int len
)
331 dccp_pr_debug_cat("ACK vector len=%d, ackno=%llu |", len
,
332 (unsigned long long)ackno
);
335 const u8 state
= (*vector
& DCCP_ACKVEC_STATE_MASK
) >> 6;
336 const u8 rl
= *vector
& DCCP_ACKVEC_LEN_MASK
;
338 dccp_pr_debug_cat("%d,%d|", state
, rl
);
342 dccp_pr_debug_cat("\n");
345 void dccp_ackvec_print(const struct dccp_ackvec
*av
)
347 dccp_ackvector_print(av
->av_buf_ackno
,
348 av
->av_buf
+ av
->av_buf_head
,
353 static void dccp_ackvec_throw_record(struct dccp_ackvec
*av
,
354 struct dccp_ackvec_record
*avr
)
356 struct dccp_ackvec_record
*next
;
358 /* sort out vector length */
359 if (av
->av_buf_head
<= avr
->avr_ack_ptr
)
360 av
->av_vec_len
= avr
->avr_ack_ptr
- av
->av_buf_head
;
362 av
->av_vec_len
= DCCP_MAX_ACKVEC_LEN
- 1 -
363 av
->av_buf_head
+ avr
->avr_ack_ptr
;
366 list_for_each_entry_safe_from(avr
, next
, &av
->av_records
, avr_node
) {
367 list_del_init(&avr
->avr_node
);
368 dccp_ackvec_record_delete(avr
);
372 void dccp_ackvec_check_rcv_ackno(struct dccp_ackvec
*av
, struct sock
*sk
,
375 struct dccp_ackvec_record
*avr
;
378 * If we traverse backwards, it should be faster when we have large
379 * windows. We will be receiving ACKs for stuff we sent a while back
382 list_for_each_entry_reverse(avr
, &av
->av_records
, avr_node
) {
383 if (ackno
== avr
->avr_ack_seqno
) {
384 dccp_pr_debug("%s ACK packet 0, len=%d, ack_seqno=%llu, "
385 "ack_ackno=%llu, ACKED!\n",
387 (unsigned long long)avr
->avr_ack_seqno
,
388 (unsigned long long)avr
->avr_ack_ackno
);
389 dccp_ackvec_throw_record(av
, avr
);
391 } else if (avr
->avr_ack_seqno
> ackno
)
392 break; /* old news */
396 static void dccp_ackvec_check_rcv_ackvector(struct dccp_ackvec
*av
,
397 struct sock
*sk
, u64
*ackno
,
398 const unsigned char len
,
399 const unsigned char *vector
)
402 struct dccp_ackvec_record
*avr
;
404 /* Check if we actually sent an ACK vector */
405 if (list_empty(&av
->av_records
))
411 * I think it might be more efficient to work backwards. See comment on
414 avr
= list_entry(av
->av_records
.next
, struct dccp_ackvec_record
, avr_node
);
416 const u8 rl
= *vector
& DCCP_ACKVEC_LEN_MASK
;
419 dccp_set_seqno(&ackno_end_rl
, *ackno
- rl
);
422 * If our AVR sequence number is greater than the ack, go
423 * forward in the AVR list until it is not so.
425 list_for_each_entry_from(avr
, &av
->av_records
, avr_node
) {
426 if (!after48(avr
->avr_ack_seqno
, *ackno
))
429 /* End of the av_records list, not found, exit */
432 if (between48(avr
->avr_ack_seqno
, ackno_end_rl
, *ackno
)) {
433 const u8 state
= *vector
& DCCP_ACKVEC_STATE_MASK
;
434 if (state
!= DCCP_ACKVEC_STATE_NOT_RECEIVED
) {
435 dccp_pr_debug("%s ACK vector 0, len=%d, "
436 "ack_seqno=%llu, ack_ackno=%llu, "
443 dccp_ackvec_throw_record(av
, avr
);
447 * If it wasn't received, continue scanning... we might
452 dccp_set_seqno(ackno
, ackno_end_rl
- 1);
457 int dccp_ackvec_parse(struct sock
*sk
, const struct sk_buff
*skb
,
458 u64
*ackno
, const u8 opt
, const u8
*value
, const u8 len
)
460 if (len
> DCCP_MAX_ACKVEC_OPT_LEN
)
463 /* dccp_ackvector_print(DCCP_SKB_CB(skb)->dccpd_ack_seq, value, len); */
464 dccp_ackvec_check_rcv_ackvector(dccp_sk(sk
)->dccps_hc_rx_ackvec
, sk
,
469 int __init
dccp_ackvec_init(void)
471 dccp_ackvec_slab
= kmem_cache_create("dccp_ackvec",
472 sizeof(struct dccp_ackvec
), 0,
473 SLAB_HWCACHE_ALIGN
, NULL
);
474 if (dccp_ackvec_slab
== NULL
)
477 dccp_ackvec_record_slab
=
478 kmem_cache_create("dccp_ackvec_record",
479 sizeof(struct dccp_ackvec_record
),
480 0, SLAB_HWCACHE_ALIGN
, NULL
);
481 if (dccp_ackvec_record_slab
== NULL
)
482 goto out_destroy_slab
;
487 kmem_cache_destroy(dccp_ackvec_slab
);
488 dccp_ackvec_slab
= NULL
;
490 DCCP_CRIT("Unable to create Ack Vector slab cache");
494 void dccp_ackvec_exit(void)
496 if (dccp_ackvec_slab
!= NULL
) {
497 kmem_cache_destroy(dccp_ackvec_slab
);
498 dccp_ackvec_slab
= NULL
;
500 if (dccp_ackvec_record_slab
!= NULL
) {
501 kmem_cache_destroy(dccp_ackvec_record_slab
);
502 dccp_ackvec_record_slab
= NULL
;