xref: /linux/net/dccp/ccids/lib/packet_history.c (revision cdd38c5f1ce4398ec58fec95904b75824daab7b5)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  Copyright (c) 2007   The University of Aberdeen, Scotland, UK
4  *  Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
5  *
6  *  An implementation of the DCCP protocol
7  *
8  *  This code has been developed by the University of Waikato WAND
9  *  research group. For further information please see https://www.wand.net.nz/
10  *  or e-mail Ian McDonald - ian.mcdonald@jandi.co.nz
11  *
12  *  This code also uses code from Lulea University, rereleased as GPL by its
13  *  authors:
14  *  Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
15  *
16  *  Changes to meet Linux coding standards, to make it meet latest ccid3 draft
17  *  and to make it work as a loadable module in the DCCP stack written by
18  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
19  *
20  *  Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
21  */
22 
23 #include <linux/string.h>
24 #include <linux/slab.h>
25 #include "packet_history.h"
26 #include "../../dccp.h"
27 
28 /*
29  * Transmitter History Routines
30  */
31 static struct kmem_cache *tfrc_tx_hist_slab;
32 
tfrc_tx_packet_history_init(void)33 int __init tfrc_tx_packet_history_init(void)
34 {
35 	tfrc_tx_hist_slab = kmem_cache_create("tfrc_tx_hist",
36 					      sizeof(struct tfrc_tx_hist_entry),
37 					      0, SLAB_HWCACHE_ALIGN, NULL);
38 	return tfrc_tx_hist_slab == NULL ? -ENOBUFS : 0;
39 }
40 
tfrc_tx_packet_history_exit(void)41 void tfrc_tx_packet_history_exit(void)
42 {
43 	if (tfrc_tx_hist_slab != NULL) {
44 		kmem_cache_destroy(tfrc_tx_hist_slab);
45 		tfrc_tx_hist_slab = NULL;
46 	}
47 }
48 
tfrc_tx_hist_add(struct tfrc_tx_hist_entry ** headp,u64 seqno)49 int tfrc_tx_hist_add(struct tfrc_tx_hist_entry **headp, u64 seqno)
50 {
51 	struct tfrc_tx_hist_entry *entry = kmem_cache_alloc(tfrc_tx_hist_slab, gfp_any());
52 
53 	if (entry == NULL)
54 		return -ENOBUFS;
55 	entry->seqno = seqno;
56 	entry->stamp = ktime_get_real();
57 	entry->next  = *headp;
58 	*headp	     = entry;
59 	return 0;
60 }
61 
tfrc_tx_hist_purge(struct tfrc_tx_hist_entry ** headp)62 void tfrc_tx_hist_purge(struct tfrc_tx_hist_entry **headp)
63 {
64 	struct tfrc_tx_hist_entry *head = *headp;
65 
66 	while (head != NULL) {
67 		struct tfrc_tx_hist_entry *next = head->next;
68 
69 		kmem_cache_free(tfrc_tx_hist_slab, head);
70 		head = next;
71 	}
72 
73 	*headp = NULL;
74 }
75 
76 /*
77  *	Receiver History Routines
78  */
79 static struct kmem_cache *tfrc_rx_hist_slab;
80 
tfrc_rx_packet_history_init(void)81 int __init tfrc_rx_packet_history_init(void)
82 {
83 	tfrc_rx_hist_slab = kmem_cache_create("tfrc_rxh_cache",
84 					      sizeof(struct tfrc_rx_hist_entry),
85 					      0, SLAB_HWCACHE_ALIGN, NULL);
86 	return tfrc_rx_hist_slab == NULL ? -ENOBUFS : 0;
87 }
88 
tfrc_rx_packet_history_exit(void)89 void tfrc_rx_packet_history_exit(void)
90 {
91 	if (tfrc_rx_hist_slab != NULL) {
92 		kmem_cache_destroy(tfrc_rx_hist_slab);
93 		tfrc_rx_hist_slab = NULL;
94 	}
95 }
96 
tfrc_rx_hist_entry_from_skb(struct tfrc_rx_hist_entry * entry,const struct sk_buff * skb,const u64 ndp)97 static inline void tfrc_rx_hist_entry_from_skb(struct tfrc_rx_hist_entry *entry,
98 					       const struct sk_buff *skb,
99 					       const u64 ndp)
100 {
101 	const struct dccp_hdr *dh = dccp_hdr(skb);
102 
103 	entry->tfrchrx_seqno = DCCP_SKB_CB(skb)->dccpd_seq;
104 	entry->tfrchrx_ccval = dh->dccph_ccval;
105 	entry->tfrchrx_type  = dh->dccph_type;
106 	entry->tfrchrx_ndp   = ndp;
107 	entry->tfrchrx_tstamp = ktime_get_real();
108 }
109 
tfrc_rx_hist_add_packet(struct tfrc_rx_hist * h,const struct sk_buff * skb,const u64 ndp)110 void tfrc_rx_hist_add_packet(struct tfrc_rx_hist *h,
111 			     const struct sk_buff *skb,
112 			     const u64 ndp)
113 {
114 	struct tfrc_rx_hist_entry *entry = tfrc_rx_hist_last_rcv(h);
115 
116 	tfrc_rx_hist_entry_from_skb(entry, skb, ndp);
117 }
118 
119 /* has the packet contained in skb been seen before? */
tfrc_rx_hist_duplicate(struct tfrc_rx_hist * h,struct sk_buff * skb)120 int tfrc_rx_hist_duplicate(struct tfrc_rx_hist *h, struct sk_buff *skb)
121 {
122 	const u64 seq = DCCP_SKB_CB(skb)->dccpd_seq;
123 	int i;
124 
125 	if (dccp_delta_seqno(tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno, seq) <= 0)
126 		return 1;
127 
128 	for (i = 1; i <= h->loss_count; i++)
129 		if (tfrc_rx_hist_entry(h, i)->tfrchrx_seqno == seq)
130 			return 1;
131 
132 	return 0;
133 }
134 
tfrc_rx_hist_swap(struct tfrc_rx_hist * h,const u8 a,const u8 b)135 static void tfrc_rx_hist_swap(struct tfrc_rx_hist *h, const u8 a, const u8 b)
136 {
137 	const u8 idx_a = tfrc_rx_hist_index(h, a),
138 		 idx_b = tfrc_rx_hist_index(h, b);
139 
140 	swap(h->ring[idx_a], h->ring[idx_b]);
141 }
142 
143 /*
144  * Private helper functions for loss detection.
145  *
146  * In the descriptions, `Si' refers to the sequence number of entry number i,
147  * whose NDP count is `Ni' (lower case is used for variables).
148  * Note: All __xxx_loss functions expect that a test against duplicates has been
149  *       performed already: the seqno of the skb must not be less than the seqno
150  *       of loss_prev; and it must not equal that of any valid history entry.
151  */
__do_track_loss(struct tfrc_rx_hist * h,struct sk_buff * skb,u64 n1)152 static void __do_track_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u64 n1)
153 {
154 	u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
155 	    s1 = DCCP_SKB_CB(skb)->dccpd_seq;
156 
157 	if (!dccp_loss_free(s0, s1, n1)) {	/* gap between S0 and S1 */
158 		h->loss_count = 1;
159 		tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n1);
160 	}
161 }
162 
__one_after_loss(struct tfrc_rx_hist * h,struct sk_buff * skb,u32 n2)163 static void __one_after_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u32 n2)
164 {
165 	u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
166 	    s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
167 	    s2 = DCCP_SKB_CB(skb)->dccpd_seq;
168 
169 	if (likely(dccp_delta_seqno(s1, s2) > 0)) {	/* S1  <  S2 */
170 		h->loss_count = 2;
171 		tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 2), skb, n2);
172 		return;
173 	}
174 
175 	/* S0  <  S2  <  S1 */
176 
177 	if (dccp_loss_free(s0, s2, n2)) {
178 		u64 n1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_ndp;
179 
180 		if (dccp_loss_free(s2, s1, n1)) {
181 			/* hole is filled: S0, S2, and S1 are consecutive */
182 			h->loss_count = 0;
183 			h->loss_start = tfrc_rx_hist_index(h, 1);
184 		} else
185 			/* gap between S2 and S1: just update loss_prev */
186 			tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_loss_prev(h), skb, n2);
187 
188 	} else {	/* gap between S0 and S2 */
189 		/*
190 		 * Reorder history to insert S2 between S0 and S1
191 		 */
192 		tfrc_rx_hist_swap(h, 0, 3);
193 		h->loss_start = tfrc_rx_hist_index(h, 3);
194 		tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n2);
195 		h->loss_count = 2;
196 	}
197 }
198 
199 /* return 1 if a new loss event has been identified */
__two_after_loss(struct tfrc_rx_hist * h,struct sk_buff * skb,u32 n3)200 static int __two_after_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u32 n3)
201 {
202 	u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
203 	    s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
204 	    s2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_seqno,
205 	    s3 = DCCP_SKB_CB(skb)->dccpd_seq;
206 
207 	if (likely(dccp_delta_seqno(s2, s3) > 0)) {	/* S2  <  S3 */
208 		h->loss_count = 3;
209 		tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 3), skb, n3);
210 		return 1;
211 	}
212 
213 	/* S3  <  S2 */
214 
215 	if (dccp_delta_seqno(s1, s3) > 0) {		/* S1  <  S3  <  S2 */
216 		/*
217 		 * Reorder history to insert S3 between S1 and S2
218 		 */
219 		tfrc_rx_hist_swap(h, 2, 3);
220 		tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 2), skb, n3);
221 		h->loss_count = 3;
222 		return 1;
223 	}
224 
225 	/* S0  <  S3  <  S1 */
226 
227 	if (dccp_loss_free(s0, s3, n3)) {
228 		u64 n1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_ndp;
229 
230 		if (dccp_loss_free(s3, s1, n1)) {
231 			/* hole between S0 and S1 filled by S3 */
232 			u64 n2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_ndp;
233 
234 			if (dccp_loss_free(s1, s2, n2)) {
235 				/* entire hole filled by S0, S3, S1, S2 */
236 				h->loss_start = tfrc_rx_hist_index(h, 2);
237 				h->loss_count = 0;
238 			} else {
239 				/* gap remains between S1 and S2 */
240 				h->loss_start = tfrc_rx_hist_index(h, 1);
241 				h->loss_count = 1;
242 			}
243 
244 		} else /* gap exists between S3 and S1, loss_count stays at 2 */
245 			tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_loss_prev(h), skb, n3);
246 
247 		return 0;
248 	}
249 
250 	/*
251 	 * The remaining case:  S0  <  S3  <  S1  <  S2;  gap between S0 and S3
252 	 * Reorder history to insert S3 between S0 and S1.
253 	 */
254 	tfrc_rx_hist_swap(h, 0, 3);
255 	h->loss_start = tfrc_rx_hist_index(h, 3);
256 	tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n3);
257 	h->loss_count = 3;
258 
259 	return 1;
260 }
261 
262 /* recycle RX history records to continue loss detection if necessary */
__three_after_loss(struct tfrc_rx_hist * h)263 static void __three_after_loss(struct tfrc_rx_hist *h)
264 {
265 	/*
266 	 * At this stage we know already that there is a gap between S0 and S1
267 	 * (since S0 was the highest sequence number received before detecting
268 	 * the loss). To recycle the loss record, it is	thus only necessary to
269 	 * check for other possible gaps between S1/S2 and between S2/S3.
270 	 */
271 	u64 s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
272 	    s2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_seqno,
273 	    s3 = tfrc_rx_hist_entry(h, 3)->tfrchrx_seqno;
274 	u64 n2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_ndp,
275 	    n3 = tfrc_rx_hist_entry(h, 3)->tfrchrx_ndp;
276 
277 	if (dccp_loss_free(s1, s2, n2)) {
278 
279 		if (dccp_loss_free(s2, s3, n3)) {
280 			/* no gap between S2 and S3: entire hole is filled */
281 			h->loss_start = tfrc_rx_hist_index(h, 3);
282 			h->loss_count = 0;
283 		} else {
284 			/* gap between S2 and S3 */
285 			h->loss_start = tfrc_rx_hist_index(h, 2);
286 			h->loss_count = 1;
287 		}
288 
289 	} else {	/* gap between S1 and S2 */
290 		h->loss_start = tfrc_rx_hist_index(h, 1);
291 		h->loss_count = 2;
292 	}
293 }
294 
295 /**
296  *  tfrc_rx_handle_loss  -  Loss detection and further processing
297  *  @h:		    The non-empty RX history object
298  *  @lh:	    Loss Intervals database to update
299  *  @skb:	    Currently received packet
300  *  @ndp:	    The NDP count belonging to @skb
301  *  @calc_first_li: Caller-dependent computation of first loss interval in @lh
302  *  @sk:	    Used by @calc_first_li (see tfrc_lh_interval_add)
303  *
304  *  Chooses action according to pending loss, updates LI database when a new
305  *  loss was detected, and does required post-processing. Returns 1 when caller
306  *  should send feedback, 0 otherwise.
307  *  Since it also takes care of reordering during loss detection and updates the
308  *  records accordingly, the caller should not perform any more RX history
309  *  operations when loss_count is greater than 0 after calling this function.
310  */
tfrc_rx_handle_loss(struct tfrc_rx_hist * h,struct tfrc_loss_hist * lh,struct sk_buff * skb,const u64 ndp,u32 (* calc_first_li)(struct sock *),struct sock * sk)311 int tfrc_rx_handle_loss(struct tfrc_rx_hist *h,
312 			struct tfrc_loss_hist *lh,
313 			struct sk_buff *skb, const u64 ndp,
314 			u32 (*calc_first_li)(struct sock *), struct sock *sk)
315 {
316 	int is_new_loss = 0;
317 
318 	if (h->loss_count == 0) {
319 		__do_track_loss(h, skb, ndp);
320 	} else if (h->loss_count == 1) {
321 		__one_after_loss(h, skb, ndp);
322 	} else if (h->loss_count != 2) {
323 		DCCP_BUG("invalid loss_count %d", h->loss_count);
324 	} else if (__two_after_loss(h, skb, ndp)) {
325 		/*
326 		 * Update Loss Interval database and recycle RX records
327 		 */
328 		is_new_loss = tfrc_lh_interval_add(lh, h, calc_first_li, sk);
329 		__three_after_loss(h);
330 	}
331 	return is_new_loss;
332 }
333 
tfrc_rx_hist_alloc(struct tfrc_rx_hist * h)334 int tfrc_rx_hist_alloc(struct tfrc_rx_hist *h)
335 {
336 	int i;
337 
338 	for (i = 0; i <= TFRC_NDUPACK; i++) {
339 		h->ring[i] = kmem_cache_alloc(tfrc_rx_hist_slab, GFP_ATOMIC);
340 		if (h->ring[i] == NULL)
341 			goto out_free;
342 	}
343 
344 	h->loss_count = h->loss_start = 0;
345 	return 0;
346 
347 out_free:
348 	while (i-- != 0) {
349 		kmem_cache_free(tfrc_rx_hist_slab, h->ring[i]);
350 		h->ring[i] = NULL;
351 	}
352 	return -ENOBUFS;
353 }
354 
tfrc_rx_hist_purge(struct tfrc_rx_hist * h)355 void tfrc_rx_hist_purge(struct tfrc_rx_hist *h)
356 {
357 	int i;
358 
359 	for (i = 0; i <= TFRC_NDUPACK; ++i)
360 		if (h->ring[i] != NULL) {
361 			kmem_cache_free(tfrc_rx_hist_slab, h->ring[i]);
362 			h->ring[i] = NULL;
363 		}
364 }
365 
366 /**
367  * tfrc_rx_hist_rtt_last_s - reference entry to compute RTT samples against
368  * @h:	The non-empty RX history object
369  */
370 static inline struct tfrc_rx_hist_entry *
tfrc_rx_hist_rtt_last_s(const struct tfrc_rx_hist * h)371 			tfrc_rx_hist_rtt_last_s(const struct tfrc_rx_hist *h)
372 {
373 	return h->ring[0];
374 }
375 
376 /**
377  * tfrc_rx_hist_rtt_prev_s - previously suitable (wrt rtt_last_s) RTT-sampling entry
378  * @h:	The non-empty RX history object
379  */
380 static inline struct tfrc_rx_hist_entry *
tfrc_rx_hist_rtt_prev_s(const struct tfrc_rx_hist * h)381 			tfrc_rx_hist_rtt_prev_s(const struct tfrc_rx_hist *h)
382 {
383 	return h->ring[h->rtt_sample_prev];
384 }
385 
386 /**
387  * tfrc_rx_hist_sample_rtt  -  Sample RTT from timestamp / CCVal
388  * @h: receive histogram
389  * @skb: packet containing timestamp.
390  *
391  * Based on ideas presented in RFC 4342, 8.1. Returns 0 if it was not able
392  * to compute a sample with given data - calling function should check this.
393  */
tfrc_rx_hist_sample_rtt(struct tfrc_rx_hist * h,const struct sk_buff * skb)394 u32 tfrc_rx_hist_sample_rtt(struct tfrc_rx_hist *h, const struct sk_buff *skb)
395 {
396 	u32 sample = 0,
397 	    delta_v = SUB16(dccp_hdr(skb)->dccph_ccval,
398 			    tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
399 
400 	if (delta_v < 1 || delta_v > 4) {	/* unsuitable CCVal delta */
401 		if (h->rtt_sample_prev == 2) {	/* previous candidate stored */
402 			sample = SUB16(tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_ccval,
403 				       tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
404 			if (sample)
405 				sample = 4 / sample *
406 				         ktime_us_delta(tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_tstamp,
407 							tfrc_rx_hist_rtt_last_s(h)->tfrchrx_tstamp);
408 			else    /*
409 				 * FIXME: This condition is in principle not
410 				 * possible but occurs when CCID is used for
411 				 * two-way data traffic. I have tried to trace
412 				 * it, but the cause does not seem to be here.
413 				 */
414 				DCCP_BUG("please report to dccp@vger.kernel.org"
415 					 " => prev = %u, last = %u",
416 					 tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_ccval,
417 					 tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
418 		} else if (delta_v < 1) {
419 			h->rtt_sample_prev = 1;
420 			goto keep_ref_for_next_time;
421 		}
422 
423 	} else if (delta_v == 4) /* optimal match */
424 		sample = ktime_to_us(net_timedelta(tfrc_rx_hist_rtt_last_s(h)->tfrchrx_tstamp));
425 	else {			 /* suboptimal match */
426 		h->rtt_sample_prev = 2;
427 		goto keep_ref_for_next_time;
428 	}
429 
430 	if (unlikely(sample > DCCP_SANE_RTT_MAX)) {
431 		DCCP_WARN("RTT sample %u too large, using max\n", sample);
432 		sample = DCCP_SANE_RTT_MAX;
433 	}
434 
435 	h->rtt_sample_prev = 0;	       /* use current entry as next reference */
436 keep_ref_for_next_time:
437 
438 	return sample;
439 }
440