1 /*
2  *  Copyright (c) 2007   The University of Aberdeen, Scotland, UK
3  *  Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
4  *
5  *  An implementation of the DCCP protocol
6  *
7  *  This code has been developed by the University of Waikato WAND
8  *  research group. For further information please see http://www.wand.net.nz/
9  *  or e-mail Ian McDonald - ian.mcdonald@jandi.co.nz
10  *
11  *  This code also uses code from Lulea University, rereleased as GPL by its
12  *  authors:
13  *  Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
14  *
15  *  Changes to meet Linux coding standards, to make it meet latest ccid3 draft
16  *  and to make it work as a loadable module in the DCCP stack written by
17  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
18  *
19  *  Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
20  *
21  *  This program is free software; you can redistribute it and/or modify
22  *  it under the terms of the GNU General Public License as published by
23  *  the Free Software Foundation; either version 2 of the License, or
24  *  (at your option) any later version.
25  *
26  *  This program is distributed in the hope that it will be useful,
27  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
28  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
29  *  GNU General Public License for more details.
30  *
31  *  You should have received a copy of the GNU General Public License
32  *  along with this program; if not, write to the Free Software
33  *  Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
34  */
35 
36 #include <linux/string.h>
37 #include <linux/slab.h>
38 #include "packet_history.h"
39 #include "../../dccp.h"
40 
41 /**
42  *  tfrc_tx_hist_entry  -  Simple singly-linked TX history list
43  *  @next:  next oldest entry (LIFO order)
44  *  @seqno: sequence number of this entry
45  *  @stamp: send time of packet with sequence number @seqno
46  */
47 struct tfrc_tx_hist_entry {
48 	struct tfrc_tx_hist_entry *next;
49 	u64			  seqno;
50 	ktime_t			  stamp;
51 };
52 
53 /*
54  * Transmitter History Routines
55  */
56 static struct kmem_cache *tfrc_tx_hist_slab;
57 
58 int __init tfrc_tx_packet_history_init(void)
59 {
60 	tfrc_tx_hist_slab = kmem_cache_create("tfrc_tx_hist",
61 					      sizeof(struct tfrc_tx_hist_entry),
62 					      0, SLAB_HWCACHE_ALIGN, NULL);
63 	return tfrc_tx_hist_slab == NULL ? -ENOBUFS : 0;
64 }
65 
66 void tfrc_tx_packet_history_exit(void)
67 {
68 	if (tfrc_tx_hist_slab != NULL) {
69 		kmem_cache_destroy(tfrc_tx_hist_slab);
70 		tfrc_tx_hist_slab = NULL;
71 	}
72 }
73 
74 static struct tfrc_tx_hist_entry *
75 	tfrc_tx_hist_find_entry(struct tfrc_tx_hist_entry *head, u64 seqno)
76 {
77 	while (head != NULL && head->seqno != seqno)
78 		head = head->next;
79 
80 	return head;
81 }
82 
83 int tfrc_tx_hist_add(struct tfrc_tx_hist_entry **headp, u64 seqno)
84 {
85 	struct tfrc_tx_hist_entry *entry = kmem_cache_alloc(tfrc_tx_hist_slab, gfp_any());
86 
87 	if (entry == NULL)
88 		return -ENOBUFS;
89 	entry->seqno = seqno;
90 	entry->stamp = ktime_get_real();
91 	entry->next  = *headp;
92 	*headp	     = entry;
93 	return 0;
94 }
95 
96 void tfrc_tx_hist_purge(struct tfrc_tx_hist_entry **headp)
97 {
98 	struct tfrc_tx_hist_entry *head = *headp;
99 
100 	while (head != NULL) {
101 		struct tfrc_tx_hist_entry *next = head->next;
102 
103 		kmem_cache_free(tfrc_tx_hist_slab, head);
104 		head = next;
105 	}
106 
107 	*headp = NULL;
108 }
109 
110 u32 tfrc_tx_hist_rtt(struct tfrc_tx_hist_entry *head, const u64 seqno,
111 		     const ktime_t now)
112 {
113 	u32 rtt = 0;
114 	struct tfrc_tx_hist_entry *packet = tfrc_tx_hist_find_entry(head, seqno);
115 
116 	if (packet != NULL) {
117 		rtt = ktime_us_delta(now, packet->stamp);
118 		/*
119 		 * Garbage-collect older (irrelevant) entries:
120 		 */
121 		tfrc_tx_hist_purge(&packet->next);
122 	}
123 
124 	return rtt;
125 }
126 
127 
128 /*
129  *	Receiver History Routines
130  */
131 static struct kmem_cache *tfrc_rx_hist_slab;
132 
133 int __init tfrc_rx_packet_history_init(void)
134 {
135 	tfrc_rx_hist_slab = kmem_cache_create("tfrc_rxh_cache",
136 					      sizeof(struct tfrc_rx_hist_entry),
137 					      0, SLAB_HWCACHE_ALIGN, NULL);
138 	return tfrc_rx_hist_slab == NULL ? -ENOBUFS : 0;
139 }
140 
141 void tfrc_rx_packet_history_exit(void)
142 {
143 	if (tfrc_rx_hist_slab != NULL) {
144 		kmem_cache_destroy(tfrc_rx_hist_slab);
145 		tfrc_rx_hist_slab = NULL;
146 	}
147 }
148 
149 static inline void tfrc_rx_hist_entry_from_skb(struct tfrc_rx_hist_entry *entry,
150 					       const struct sk_buff *skb,
151 					       const u64 ndp)
152 {
153 	const struct dccp_hdr *dh = dccp_hdr(skb);
154 
155 	entry->tfrchrx_seqno = DCCP_SKB_CB(skb)->dccpd_seq;
156 	entry->tfrchrx_ccval = dh->dccph_ccval;
157 	entry->tfrchrx_type  = dh->dccph_type;
158 	entry->tfrchrx_ndp   = ndp;
159 	entry->tfrchrx_tstamp = ktime_get_real();
160 }
161 
162 void tfrc_rx_hist_add_packet(struct tfrc_rx_hist *h,
163 			     const struct sk_buff *skb,
164 			     const u64 ndp)
165 {
166 	struct tfrc_rx_hist_entry *entry = tfrc_rx_hist_last_rcv(h);
167 
168 	tfrc_rx_hist_entry_from_skb(entry, skb, ndp);
169 }
170 
171 /* has the packet contained in skb been seen before? */
172 int tfrc_rx_hist_duplicate(struct tfrc_rx_hist *h, struct sk_buff *skb)
173 {
174 	const u64 seq = DCCP_SKB_CB(skb)->dccpd_seq;
175 	int i;
176 
177 	if (dccp_delta_seqno(tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno, seq) <= 0)
178 		return 1;
179 
180 	for (i = 1; i <= h->loss_count; i++)
181 		if (tfrc_rx_hist_entry(h, i)->tfrchrx_seqno == seq)
182 			return 1;
183 
184 	return 0;
185 }
186 
187 static void tfrc_rx_hist_swap(struct tfrc_rx_hist *h, const u8 a, const u8 b)
188 {
189 	const u8 idx_a = tfrc_rx_hist_index(h, a),
190 		 idx_b = tfrc_rx_hist_index(h, b);
191 	struct tfrc_rx_hist_entry *tmp = h->ring[idx_a];
192 
193 	h->ring[idx_a] = h->ring[idx_b];
194 	h->ring[idx_b] = tmp;
195 }
196 
197 /*
198  * Private helper functions for loss detection.
199  *
200  * In the descriptions, `Si' refers to the sequence number of entry number i,
201  * whose NDP count is `Ni' (lower case is used for variables).
202  * Note: All __xxx_loss functions expect that a test against duplicates has been
203  *       performed already: the seqno of the skb must not be less than the seqno
204  *       of loss_prev; and it must not equal that of any valid history entry.
205  */
206 static void __do_track_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u64 n1)
207 {
208 	u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
209 	    s1 = DCCP_SKB_CB(skb)->dccpd_seq;
210 
211 	if (!dccp_loss_free(s0, s1, n1)) {	/* gap between S0 and S1 */
212 		h->loss_count = 1;
213 		tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n1);
214 	}
215 }
216 
217 static void __one_after_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u32 n2)
218 {
219 	u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
220 	    s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
221 	    s2 = DCCP_SKB_CB(skb)->dccpd_seq;
222 
223 	if (likely(dccp_delta_seqno(s1, s2) > 0)) {	/* S1  <  S2 */
224 		h->loss_count = 2;
225 		tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 2), skb, n2);
226 		return;
227 	}
228 
229 	/* S0  <  S2  <  S1 */
230 
231 	if (dccp_loss_free(s0, s2, n2)) {
232 		u64 n1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_ndp;
233 
234 		if (dccp_loss_free(s2, s1, n1)) {
235 			/* hole is filled: S0, S2, and S1 are consecutive */
236 			h->loss_count = 0;
237 			h->loss_start = tfrc_rx_hist_index(h, 1);
238 		} else
239 			/* gap between S2 and S1: just update loss_prev */
240 			tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_loss_prev(h), skb, n2);
241 
242 	} else {	/* gap between S0 and S2 */
243 		/*
244 		 * Reorder history to insert S2 between S0 and S1
245 		 */
246 		tfrc_rx_hist_swap(h, 0, 3);
247 		h->loss_start = tfrc_rx_hist_index(h, 3);
248 		tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n2);
249 		h->loss_count = 2;
250 	}
251 }
252 
253 /* return 1 if a new loss event has been identified */
254 static int __two_after_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u32 n3)
255 {
256 	u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
257 	    s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
258 	    s2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_seqno,
259 	    s3 = DCCP_SKB_CB(skb)->dccpd_seq;
260 
261 	if (likely(dccp_delta_seqno(s2, s3) > 0)) {	/* S2  <  S3 */
262 		h->loss_count = 3;
263 		tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 3), skb, n3);
264 		return 1;
265 	}
266 
267 	/* S3  <  S2 */
268 
269 	if (dccp_delta_seqno(s1, s3) > 0) {		/* S1  <  S3  <  S2 */
270 		/*
271 		 * Reorder history to insert S3 between S1 and S2
272 		 */
273 		tfrc_rx_hist_swap(h, 2, 3);
274 		tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 2), skb, n3);
275 		h->loss_count = 3;
276 		return 1;
277 	}
278 
279 	/* S0  <  S3  <  S1 */
280 
281 	if (dccp_loss_free(s0, s3, n3)) {
282 		u64 n1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_ndp;
283 
284 		if (dccp_loss_free(s3, s1, n1)) {
285 			/* hole between S0 and S1 filled by S3 */
286 			u64 n2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_ndp;
287 
288 			if (dccp_loss_free(s1, s2, n2)) {
289 				/* entire hole filled by S0, S3, S1, S2 */
290 				h->loss_start = tfrc_rx_hist_index(h, 2);
291 				h->loss_count = 0;
292 			} else {
293 				/* gap remains between S1 and S2 */
294 				h->loss_start = tfrc_rx_hist_index(h, 1);
295 				h->loss_count = 1;
296 			}
297 
298 		} else /* gap exists between S3 and S1, loss_count stays at 2 */
299 			tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_loss_prev(h), skb, n3);
300 
301 		return 0;
302 	}
303 
304 	/*
305 	 * The remaining case:  S0  <  S3  <  S1  <  S2;  gap between S0 and S3
306 	 * Reorder history to insert S3 between S0 and S1.
307 	 */
308 	tfrc_rx_hist_swap(h, 0, 3);
309 	h->loss_start = tfrc_rx_hist_index(h, 3);
310 	tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n3);
311 	h->loss_count = 3;
312 
313 	return 1;
314 }
315 
316 /* recycle RX history records to continue loss detection if necessary */
317 static void __three_after_loss(struct tfrc_rx_hist *h)
318 {
319 	/*
320 	 * At this stage we know already that there is a gap between S0 and S1
321 	 * (since S0 was the highest sequence number received before detecting
322 	 * the loss). To recycle the loss record, it is	thus only necessary to
323 	 * check for other possible gaps between S1/S2 and between S2/S3.
324 	 */
325 	u64 s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
326 	    s2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_seqno,
327 	    s3 = tfrc_rx_hist_entry(h, 3)->tfrchrx_seqno;
328 	u64 n2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_ndp,
329 	    n3 = tfrc_rx_hist_entry(h, 3)->tfrchrx_ndp;
330 
331 	if (dccp_loss_free(s1, s2, n2)) {
332 
333 		if (dccp_loss_free(s2, s3, n3)) {
334 			/* no gap between S2 and S3: entire hole is filled */
335 			h->loss_start = tfrc_rx_hist_index(h, 3);
336 			h->loss_count = 0;
337 		} else {
338 			/* gap between S2 and S3 */
339 			h->loss_start = tfrc_rx_hist_index(h, 2);
340 			h->loss_count = 1;
341 		}
342 
343 	} else {	/* gap between S1 and S2 */
344 		h->loss_start = tfrc_rx_hist_index(h, 1);
345 		h->loss_count = 2;
346 	}
347 }
348 
349 /**
350  *  tfrc_rx_handle_loss  -  Loss detection and further processing
351  *  @h:		    The non-empty RX history object
352  *  @lh:	    Loss Intervals database to update
353  *  @skb:	    Currently received packet
354  *  @ndp:	    The NDP count belonging to @skb
355  *  @calc_first_li: Caller-dependent computation of first loss interval in @lh
356  *  @sk:	    Used by @calc_first_li (see tfrc_lh_interval_add)
357  *  Chooses action according to pending loss, updates LI database when a new
358  *  loss was detected, and does required post-processing. Returns 1 when caller
359  *  should send feedback, 0 otherwise.
360  *  Since it also takes care of reordering during loss detection and updates the
361  *  records accordingly, the caller should not perform any more RX history
362  *  operations when loss_count is greater than 0 after calling this function.
363  */
364 int tfrc_rx_handle_loss(struct tfrc_rx_hist *h,
365 			struct tfrc_loss_hist *lh,
366 			struct sk_buff *skb, const u64 ndp,
367 			u32 (*calc_first_li)(struct sock *), struct sock *sk)
368 {
369 	int is_new_loss = 0;
370 
371 	if (h->loss_count == 0) {
372 		__do_track_loss(h, skb, ndp);
373 	} else if (h->loss_count == 1) {
374 		__one_after_loss(h, skb, ndp);
375 	} else if (h->loss_count != 2) {
376 		DCCP_BUG("invalid loss_count %d", h->loss_count);
377 	} else if (__two_after_loss(h, skb, ndp)) {
378 		/*
379 		 * Update Loss Interval database and recycle RX records
380 		 */
381 		is_new_loss = tfrc_lh_interval_add(lh, h, calc_first_li, sk);
382 		__three_after_loss(h);
383 	}
384 	return is_new_loss;
385 }
386 
387 int tfrc_rx_hist_alloc(struct tfrc_rx_hist *h)
388 {
389 	int i;
390 
391 	for (i = 0; i <= TFRC_NDUPACK; i++) {
392 		h->ring[i] = kmem_cache_alloc(tfrc_rx_hist_slab, GFP_ATOMIC);
393 		if (h->ring[i] == NULL)
394 			goto out_free;
395 	}
396 
397 	h->loss_count = h->loss_start = 0;
398 	return 0;
399 
400 out_free:
401 	while (i-- != 0) {
402 		kmem_cache_free(tfrc_rx_hist_slab, h->ring[i]);
403 		h->ring[i] = NULL;
404 	}
405 	return -ENOBUFS;
406 }
407 
408 void tfrc_rx_hist_purge(struct tfrc_rx_hist *h)
409 {
410 	int i;
411 
412 	for (i = 0; i <= TFRC_NDUPACK; ++i)
413 		if (h->ring[i] != NULL) {
414 			kmem_cache_free(tfrc_rx_hist_slab, h->ring[i]);
415 			h->ring[i] = NULL;
416 		}
417 }
418 
419 /**
420  * tfrc_rx_hist_rtt_last_s - reference entry to compute RTT samples against
421  */
422 static inline struct tfrc_rx_hist_entry *
423 			tfrc_rx_hist_rtt_last_s(const struct tfrc_rx_hist *h)
424 {
425 	return h->ring[0];
426 }
427 
428 /**
429  * tfrc_rx_hist_rtt_prev_s: previously suitable (wrt rtt_last_s) RTT-sampling entry
430  */
431 static inline struct tfrc_rx_hist_entry *
432 			tfrc_rx_hist_rtt_prev_s(const struct tfrc_rx_hist *h)
433 {
434 	return h->ring[h->rtt_sample_prev];
435 }
436 
437 /**
438  * tfrc_rx_hist_sample_rtt  -  Sample RTT from timestamp / CCVal
439  * Based on ideas presented in RFC 4342, 8.1. Returns 0 if it was not able
440  * to compute a sample with given data - calling function should check this.
441  */
442 u32 tfrc_rx_hist_sample_rtt(struct tfrc_rx_hist *h, const struct sk_buff *skb)
443 {
444 	u32 sample = 0,
445 	    delta_v = SUB16(dccp_hdr(skb)->dccph_ccval,
446 			    tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
447 
448 	if (delta_v < 1 || delta_v > 4) {	/* unsuitable CCVal delta */
449 		if (h->rtt_sample_prev == 2) {	/* previous candidate stored */
450 			sample = SUB16(tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_ccval,
451 				       tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
452 			if (sample)
453 				sample = 4 / sample *
454 				         ktime_us_delta(tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_tstamp,
455 							tfrc_rx_hist_rtt_last_s(h)->tfrchrx_tstamp);
456 			else    /*
457 				 * FIXME: This condition is in principle not
458 				 * possible but occurs when CCID is used for
459 				 * two-way data traffic. I have tried to trace
460 				 * it, but the cause does not seem to be here.
461 				 */
462 				DCCP_BUG("please report to dccp@vger.kernel.org"
463 					 " => prev = %u, last = %u",
464 					 tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_ccval,
465 					 tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
466 		} else if (delta_v < 1) {
467 			h->rtt_sample_prev = 1;
468 			goto keep_ref_for_next_time;
469 		}
470 
471 	} else if (delta_v == 4) /* optimal match */
472 		sample = ktime_to_us(net_timedelta(tfrc_rx_hist_rtt_last_s(h)->tfrchrx_tstamp));
473 	else {			 /* suboptimal match */
474 		h->rtt_sample_prev = 2;
475 		goto keep_ref_for_next_time;
476 	}
477 
478 	if (unlikely(sample > DCCP_SANE_RTT_MAX)) {
479 		DCCP_WARN("RTT sample %u too large, using max\n", sample);
480 		sample = DCCP_SANE_RTT_MAX;
481 	}
482 
483 	h->rtt_sample_prev = 0;	       /* use current entry as next reference */
484 keep_ref_for_next_time:
485 
486 	return sample;
487 }
488