xref: /linux/net/dccp/ccids/lib/packet_history.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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  * Transmitter History Routines
43  */
44 static struct kmem_cache *tfrc_tx_hist_slab;
45 
46 int __init tfrc_tx_packet_history_init(void)
47 {
48 	tfrc_tx_hist_slab = kmem_cache_create("tfrc_tx_hist",
49 					      sizeof(struct tfrc_tx_hist_entry),
50 					      0, SLAB_HWCACHE_ALIGN, NULL);
51 	return tfrc_tx_hist_slab == NULL ? -ENOBUFS : 0;
52 }
53 
54 void tfrc_tx_packet_history_exit(void)
55 {
56 	if (tfrc_tx_hist_slab != NULL) {
57 		kmem_cache_destroy(tfrc_tx_hist_slab);
58 		tfrc_tx_hist_slab = NULL;
59 	}
60 }
61 
62 int tfrc_tx_hist_add(struct tfrc_tx_hist_entry **headp, u64 seqno)
63 {
64 	struct tfrc_tx_hist_entry *entry = kmem_cache_alloc(tfrc_tx_hist_slab, gfp_any());
65 
66 	if (entry == NULL)
67 		return -ENOBUFS;
68 	entry->seqno = seqno;
69 	entry->stamp = ktime_get_real();
70 	entry->next  = *headp;
71 	*headp	     = entry;
72 	return 0;
73 }
74 
75 void tfrc_tx_hist_purge(struct tfrc_tx_hist_entry **headp)
76 {
77 	struct tfrc_tx_hist_entry *head = *headp;
78 
79 	while (head != NULL) {
80 		struct tfrc_tx_hist_entry *next = head->next;
81 
82 		kmem_cache_free(tfrc_tx_hist_slab, head);
83 		head = next;
84 	}
85 
86 	*headp = NULL;
87 }
88 
89 /*
90  *	Receiver History Routines
91  */
92 static struct kmem_cache *tfrc_rx_hist_slab;
93 
94 int __init tfrc_rx_packet_history_init(void)
95 {
96 	tfrc_rx_hist_slab = kmem_cache_create("tfrc_rxh_cache",
97 					      sizeof(struct tfrc_rx_hist_entry),
98 					      0, SLAB_HWCACHE_ALIGN, NULL);
99 	return tfrc_rx_hist_slab == NULL ? -ENOBUFS : 0;
100 }
101 
102 void tfrc_rx_packet_history_exit(void)
103 {
104 	if (tfrc_rx_hist_slab != NULL) {
105 		kmem_cache_destroy(tfrc_rx_hist_slab);
106 		tfrc_rx_hist_slab = NULL;
107 	}
108 }
109 
110 static inline void tfrc_rx_hist_entry_from_skb(struct tfrc_rx_hist_entry *entry,
111 					       const struct sk_buff *skb,
112 					       const u64 ndp)
113 {
114 	const struct dccp_hdr *dh = dccp_hdr(skb);
115 
116 	entry->tfrchrx_seqno = DCCP_SKB_CB(skb)->dccpd_seq;
117 	entry->tfrchrx_ccval = dh->dccph_ccval;
118 	entry->tfrchrx_type  = dh->dccph_type;
119 	entry->tfrchrx_ndp   = ndp;
120 	entry->tfrchrx_tstamp = ktime_get_real();
121 }
122 
123 void tfrc_rx_hist_add_packet(struct tfrc_rx_hist *h,
124 			     const struct sk_buff *skb,
125 			     const u64 ndp)
126 {
127 	struct tfrc_rx_hist_entry *entry = tfrc_rx_hist_last_rcv(h);
128 
129 	tfrc_rx_hist_entry_from_skb(entry, skb, ndp);
130 }
131 
132 /* has the packet contained in skb been seen before? */
133 int tfrc_rx_hist_duplicate(struct tfrc_rx_hist *h, struct sk_buff *skb)
134 {
135 	const u64 seq = DCCP_SKB_CB(skb)->dccpd_seq;
136 	int i;
137 
138 	if (dccp_delta_seqno(tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno, seq) <= 0)
139 		return 1;
140 
141 	for (i = 1; i <= h->loss_count; i++)
142 		if (tfrc_rx_hist_entry(h, i)->tfrchrx_seqno == seq)
143 			return 1;
144 
145 	return 0;
146 }
147 
148 static void tfrc_rx_hist_swap(struct tfrc_rx_hist *h, const u8 a, const u8 b)
149 {
150 	const u8 idx_a = tfrc_rx_hist_index(h, a),
151 		 idx_b = tfrc_rx_hist_index(h, b);
152 	struct tfrc_rx_hist_entry *tmp = h->ring[idx_a];
153 
154 	h->ring[idx_a] = h->ring[idx_b];
155 	h->ring[idx_b] = tmp;
156 }
157 
158 /*
159  * Private helper functions for loss detection.
160  *
161  * In the descriptions, `Si' refers to the sequence number of entry number i,
162  * whose NDP count is `Ni' (lower case is used for variables).
163  * Note: All __xxx_loss functions expect that a test against duplicates has been
164  *       performed already: the seqno of the skb must not be less than the seqno
165  *       of loss_prev; and it must not equal that of any valid history entry.
166  */
167 static void __do_track_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u64 n1)
168 {
169 	u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
170 	    s1 = DCCP_SKB_CB(skb)->dccpd_seq;
171 
172 	if (!dccp_loss_free(s0, s1, n1)) {	/* gap between S0 and S1 */
173 		h->loss_count = 1;
174 		tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n1);
175 	}
176 }
177 
178 static void __one_after_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u32 n2)
179 {
180 	u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
181 	    s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
182 	    s2 = DCCP_SKB_CB(skb)->dccpd_seq;
183 
184 	if (likely(dccp_delta_seqno(s1, s2) > 0)) {	/* S1  <  S2 */
185 		h->loss_count = 2;
186 		tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 2), skb, n2);
187 		return;
188 	}
189 
190 	/* S0  <  S2  <  S1 */
191 
192 	if (dccp_loss_free(s0, s2, n2)) {
193 		u64 n1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_ndp;
194 
195 		if (dccp_loss_free(s2, s1, n1)) {
196 			/* hole is filled: S0, S2, and S1 are consecutive */
197 			h->loss_count = 0;
198 			h->loss_start = tfrc_rx_hist_index(h, 1);
199 		} else
200 			/* gap between S2 and S1: just update loss_prev */
201 			tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_loss_prev(h), skb, n2);
202 
203 	} else {	/* gap between S0 and S2 */
204 		/*
205 		 * Reorder history to insert S2 between S0 and S1
206 		 */
207 		tfrc_rx_hist_swap(h, 0, 3);
208 		h->loss_start = tfrc_rx_hist_index(h, 3);
209 		tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n2);
210 		h->loss_count = 2;
211 	}
212 }
213 
214 /* return 1 if a new loss event has been identified */
215 static int __two_after_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u32 n3)
216 {
217 	u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
218 	    s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
219 	    s2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_seqno,
220 	    s3 = DCCP_SKB_CB(skb)->dccpd_seq;
221 
222 	if (likely(dccp_delta_seqno(s2, s3) > 0)) {	/* S2  <  S3 */
223 		h->loss_count = 3;
224 		tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 3), skb, n3);
225 		return 1;
226 	}
227 
228 	/* S3  <  S2 */
229 
230 	if (dccp_delta_seqno(s1, s3) > 0) {		/* S1  <  S3  <  S2 */
231 		/*
232 		 * Reorder history to insert S3 between S1 and S2
233 		 */
234 		tfrc_rx_hist_swap(h, 2, 3);
235 		tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 2), skb, n3);
236 		h->loss_count = 3;
237 		return 1;
238 	}
239 
240 	/* S0  <  S3  <  S1 */
241 
242 	if (dccp_loss_free(s0, s3, n3)) {
243 		u64 n1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_ndp;
244 
245 		if (dccp_loss_free(s3, s1, n1)) {
246 			/* hole between S0 and S1 filled by S3 */
247 			u64 n2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_ndp;
248 
249 			if (dccp_loss_free(s1, s2, n2)) {
250 				/* entire hole filled by S0, S3, S1, S2 */
251 				h->loss_start = tfrc_rx_hist_index(h, 2);
252 				h->loss_count = 0;
253 			} else {
254 				/* gap remains between S1 and S2 */
255 				h->loss_start = tfrc_rx_hist_index(h, 1);
256 				h->loss_count = 1;
257 			}
258 
259 		} else /* gap exists between S3 and S1, loss_count stays at 2 */
260 			tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_loss_prev(h), skb, n3);
261 
262 		return 0;
263 	}
264 
265 	/*
266 	 * The remaining case:  S0  <  S3  <  S1  <  S2;  gap between S0 and S3
267 	 * Reorder history to insert S3 between S0 and S1.
268 	 */
269 	tfrc_rx_hist_swap(h, 0, 3);
270 	h->loss_start = tfrc_rx_hist_index(h, 3);
271 	tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n3);
272 	h->loss_count = 3;
273 
274 	return 1;
275 }
276 
277 /* recycle RX history records to continue loss detection if necessary */
278 static void __three_after_loss(struct tfrc_rx_hist *h)
279 {
280 	/*
281 	 * At this stage we know already that there is a gap between S0 and S1
282 	 * (since S0 was the highest sequence number received before detecting
283 	 * the loss). To recycle the loss record, it is	thus only necessary to
284 	 * check for other possible gaps between S1/S2 and between S2/S3.
285 	 */
286 	u64 s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
287 	    s2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_seqno,
288 	    s3 = tfrc_rx_hist_entry(h, 3)->tfrchrx_seqno;
289 	u64 n2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_ndp,
290 	    n3 = tfrc_rx_hist_entry(h, 3)->tfrchrx_ndp;
291 
292 	if (dccp_loss_free(s1, s2, n2)) {
293 
294 		if (dccp_loss_free(s2, s3, n3)) {
295 			/* no gap between S2 and S3: entire hole is filled */
296 			h->loss_start = tfrc_rx_hist_index(h, 3);
297 			h->loss_count = 0;
298 		} else {
299 			/* gap between S2 and S3 */
300 			h->loss_start = tfrc_rx_hist_index(h, 2);
301 			h->loss_count = 1;
302 		}
303 
304 	} else {	/* gap between S1 and S2 */
305 		h->loss_start = tfrc_rx_hist_index(h, 1);
306 		h->loss_count = 2;
307 	}
308 }
309 
310 /**
311  *  tfrc_rx_handle_loss  -  Loss detection and further processing
312  *  @h:		    The non-empty RX history object
313  *  @lh:	    Loss Intervals database to update
314  *  @skb:	    Currently received packet
315  *  @ndp:	    The NDP count belonging to @skb
316  *  @calc_first_li: Caller-dependent computation of first loss interval in @lh
317  *  @sk:	    Used by @calc_first_li (see tfrc_lh_interval_add)
318  *
319  *  Chooses action according to pending loss, updates LI database when a new
320  *  loss was detected, and does required post-processing. Returns 1 when caller
321  *  should send feedback, 0 otherwise.
322  *  Since it also takes care of reordering during loss detection and updates the
323  *  records accordingly, the caller should not perform any more RX history
324  *  operations when loss_count is greater than 0 after calling this function.
325  */
326 int tfrc_rx_handle_loss(struct tfrc_rx_hist *h,
327 			struct tfrc_loss_hist *lh,
328 			struct sk_buff *skb, const u64 ndp,
329 			u32 (*calc_first_li)(struct sock *), struct sock *sk)
330 {
331 	int is_new_loss = 0;
332 
333 	if (h->loss_count == 0) {
334 		__do_track_loss(h, skb, ndp);
335 	} else if (h->loss_count == 1) {
336 		__one_after_loss(h, skb, ndp);
337 	} else if (h->loss_count != 2) {
338 		DCCP_BUG("invalid loss_count %d", h->loss_count);
339 	} else if (__two_after_loss(h, skb, ndp)) {
340 		/*
341 		 * Update Loss Interval database and recycle RX records
342 		 */
343 		is_new_loss = tfrc_lh_interval_add(lh, h, calc_first_li, sk);
344 		__three_after_loss(h);
345 	}
346 	return is_new_loss;
347 }
348 
349 int tfrc_rx_hist_alloc(struct tfrc_rx_hist *h)
350 {
351 	int i;
352 
353 	for (i = 0; i <= TFRC_NDUPACK; i++) {
354 		h->ring[i] = kmem_cache_alloc(tfrc_rx_hist_slab, GFP_ATOMIC);
355 		if (h->ring[i] == NULL)
356 			goto out_free;
357 	}
358 
359 	h->loss_count = h->loss_start = 0;
360 	return 0;
361 
362 out_free:
363 	while (i-- != 0) {
364 		kmem_cache_free(tfrc_rx_hist_slab, h->ring[i]);
365 		h->ring[i] = NULL;
366 	}
367 	return -ENOBUFS;
368 }
369 
370 void tfrc_rx_hist_purge(struct tfrc_rx_hist *h)
371 {
372 	int i;
373 
374 	for (i = 0; i <= TFRC_NDUPACK; ++i)
375 		if (h->ring[i] != NULL) {
376 			kmem_cache_free(tfrc_rx_hist_slab, h->ring[i]);
377 			h->ring[i] = NULL;
378 		}
379 }
380 
381 /**
382  * tfrc_rx_hist_rtt_last_s - reference entry to compute RTT samples against
383  */
384 static inline struct tfrc_rx_hist_entry *
385 			tfrc_rx_hist_rtt_last_s(const struct tfrc_rx_hist *h)
386 {
387 	return h->ring[0];
388 }
389 
390 /**
391  * tfrc_rx_hist_rtt_prev_s - previously suitable (wrt rtt_last_s) RTT-sampling entry
392  */
393 static inline struct tfrc_rx_hist_entry *
394 			tfrc_rx_hist_rtt_prev_s(const struct tfrc_rx_hist *h)
395 {
396 	return h->ring[h->rtt_sample_prev];
397 }
398 
399 /**
400  * tfrc_rx_hist_sample_rtt  -  Sample RTT from timestamp / CCVal
401  * Based on ideas presented in RFC 4342, 8.1. Returns 0 if it was not able
402  * to compute a sample with given data - calling function should check this.
403  */
404 u32 tfrc_rx_hist_sample_rtt(struct tfrc_rx_hist *h, const struct sk_buff *skb)
405 {
406 	u32 sample = 0,
407 	    delta_v = SUB16(dccp_hdr(skb)->dccph_ccval,
408 			    tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
409 
410 	if (delta_v < 1 || delta_v > 4) {	/* unsuitable CCVal delta */
411 		if (h->rtt_sample_prev == 2) {	/* previous candidate stored */
412 			sample = SUB16(tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_ccval,
413 				       tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
414 			if (sample)
415 				sample = 4 / sample *
416 				         ktime_us_delta(tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_tstamp,
417 							tfrc_rx_hist_rtt_last_s(h)->tfrchrx_tstamp);
418 			else    /*
419 				 * FIXME: This condition is in principle not
420 				 * possible but occurs when CCID is used for
421 				 * two-way data traffic. I have tried to trace
422 				 * it, but the cause does not seem to be here.
423 				 */
424 				DCCP_BUG("please report to dccp@vger.kernel.org"
425 					 " => prev = %u, last = %u",
426 					 tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_ccval,
427 					 tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
428 		} else if (delta_v < 1) {
429 			h->rtt_sample_prev = 1;
430 			goto keep_ref_for_next_time;
431 		}
432 
433 	} else if (delta_v == 4) /* optimal match */
434 		sample = ktime_to_us(net_timedelta(tfrc_rx_hist_rtt_last_s(h)->tfrchrx_tstamp));
435 	else {			 /* suboptimal match */
436 		h->rtt_sample_prev = 2;
437 		goto keep_ref_for_next_time;
438 	}
439 
440 	if (unlikely(sample > DCCP_SANE_RTT_MAX)) {
441 		DCCP_WARN("RTT sample %u too large, using max\n", sample);
442 		sample = DCCP_SANE_RTT_MAX;
443 	}
444 
445 	h->rtt_sample_prev = 0;	       /* use current entry as next reference */
446 keep_ref_for_next_time:
447 
448 	return sample;
449 }
450