aboutsummaryrefslogtreecommitdiffstats
path: root/net/ipv4
diff options
context:
space:
mode:
Diffstat (limited to 'net/ipv4')
-rw-r--r--net/ipv4/Kconfig10
-rw-r--r--net/ipv4/Makefile1
-rw-r--r--net/ipv4/af_inet.c36
-rw-r--r--net/ipv4/fib_rules.c4
-rw-r--r--net/ipv4/fib_trie.c2
-rw-r--r--net/ipv4/inetpeer.c2
-rw-r--r--net/ipv4/ip_input.c3
-rw-r--r--net/ipv4/ip_output.c4
-rw-r--r--net/ipv4/ipcomp.c3
-rw-r--r--net/ipv4/tcp_compound.c448
-rw-r--r--net/ipv4/tcp_highspeed.c13
-rw-r--r--net/ipv4/tcp_ipv4.c18
-rw-r--r--net/ipv4/xfrm4_output.c2
13 files changed, 75 insertions, 471 deletions
diff --git a/net/ipv4/Kconfig b/net/ipv4/Kconfig
index da33393be45f..8514106761b0 100644
--- a/net/ipv4/Kconfig
+++ b/net/ipv4/Kconfig
@@ -572,16 +572,6 @@ config TCP_CONG_VENO
572 loss packets. 572 loss packets.
573 See http://www.ntu.edu.sg/home5/ZHOU0022/papers/CPFu03a.pdf 573 See http://www.ntu.edu.sg/home5/ZHOU0022/papers/CPFu03a.pdf
574 574
575config TCP_CONG_COMPOUND
576 tristate "TCP Compound"
577 depends on EXPERIMENTAL
578 default n
579 ---help---
580 TCP Compound is a sender-side only change to TCP that uses
581 a mixed Reno/Vegas approach to calculate the cwnd.
582 For further details look here:
583 ftp://ftp.research.microsoft.com/pub/tr/TR-2005-86.pdf
584
585endmenu 575endmenu
586 576
587config TCP_CONG_BIC 577config TCP_CONG_BIC
diff --git a/net/ipv4/Makefile b/net/ipv4/Makefile
index 38b8039bdd55..4878fc5be85f 100644
--- a/net/ipv4/Makefile
+++ b/net/ipv4/Makefile
@@ -47,7 +47,6 @@ obj-$(CONFIG_TCP_CONG_VEGAS) += tcp_vegas.o
47obj-$(CONFIG_TCP_CONG_VENO) += tcp_veno.o 47obj-$(CONFIG_TCP_CONG_VENO) += tcp_veno.o
48obj-$(CONFIG_TCP_CONG_SCALABLE) += tcp_scalable.o 48obj-$(CONFIG_TCP_CONG_SCALABLE) += tcp_scalable.o
49obj-$(CONFIG_TCP_CONG_LP) += tcp_lp.o 49obj-$(CONFIG_TCP_CONG_LP) += tcp_lp.o
50obj-$(CONFIG_TCP_CONG_COMPOUND) += tcp_compound.o
51 50
52obj-$(CONFIG_XFRM) += xfrm4_policy.o xfrm4_state.o xfrm4_input.o \ 51obj-$(CONFIG_XFRM) += xfrm4_policy.o xfrm4_state.o xfrm4_input.o \
53 xfrm4_output.o 52 xfrm4_output.o
diff --git a/net/ipv4/af_inet.c b/net/ipv4/af_inet.c
index 318d4674faa1..c84a32070f8d 100644
--- a/net/ipv4/af_inet.c
+++ b/net/ipv4/af_inet.c
@@ -1097,6 +1097,40 @@ int inet_sk_rebuild_header(struct sock *sk)
1097 1097
1098EXPORT_SYMBOL(inet_sk_rebuild_header); 1098EXPORT_SYMBOL(inet_sk_rebuild_header);
1099 1099
1100static int inet_gso_send_check(struct sk_buff *skb)
1101{
1102 struct iphdr *iph;
1103 struct net_protocol *ops;
1104 int proto;
1105 int ihl;
1106 int err = -EINVAL;
1107
1108 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1109 goto out;
1110
1111 iph = skb->nh.iph;
1112 ihl = iph->ihl * 4;
1113 if (ihl < sizeof(*iph))
1114 goto out;
1115
1116 if (unlikely(!pskb_may_pull(skb, ihl)))
1117 goto out;
1118
1119 skb->h.raw = __skb_pull(skb, ihl);
1120 iph = skb->nh.iph;
1121 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1122 err = -EPROTONOSUPPORT;
1123
1124 rcu_read_lock();
1125 ops = rcu_dereference(inet_protos[proto]);
1126 if (likely(ops && ops->gso_send_check))
1127 err = ops->gso_send_check(skb);
1128 rcu_read_unlock();
1129
1130out:
1131 return err;
1132}
1133
1100static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features) 1134static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1101{ 1135{
1102 struct sk_buff *segs = ERR_PTR(-EINVAL); 1136 struct sk_buff *segs = ERR_PTR(-EINVAL);
@@ -1162,6 +1196,7 @@ static struct net_protocol igmp_protocol = {
1162static struct net_protocol tcp_protocol = { 1196static struct net_protocol tcp_protocol = {
1163 .handler = tcp_v4_rcv, 1197 .handler = tcp_v4_rcv,
1164 .err_handler = tcp_v4_err, 1198 .err_handler = tcp_v4_err,
1199 .gso_send_check = tcp_v4_gso_send_check,
1165 .gso_segment = tcp_tso_segment, 1200 .gso_segment = tcp_tso_segment,
1166 .no_policy = 1, 1201 .no_policy = 1,
1167}; 1202};
@@ -1208,6 +1243,7 @@ static int ipv4_proc_init(void);
1208static struct packet_type ip_packet_type = { 1243static struct packet_type ip_packet_type = {
1209 .type = __constant_htons(ETH_P_IP), 1244 .type = __constant_htons(ETH_P_IP),
1210 .func = ip_rcv, 1245 .func = ip_rcv,
1246 .gso_send_check = inet_gso_send_check,
1211 .gso_segment = inet_gso_segment, 1247 .gso_segment = inet_gso_segment,
1212}; 1248};
1213 1249
diff --git a/net/ipv4/fib_rules.c b/net/ipv4/fib_rules.c
index 6c642d11d4ca..773b12ba4e3c 100644
--- a/net/ipv4/fib_rules.c
+++ b/net/ipv4/fib_rules.c
@@ -457,13 +457,13 @@ int inet_dump_rules(struct sk_buff *skb, struct netlink_callback *cb)
457 457
458 rcu_read_lock(); 458 rcu_read_lock();
459 hlist_for_each_entry(r, node, &fib_rules, hlist) { 459 hlist_for_each_entry(r, node, &fib_rules, hlist) {
460
461 if (idx < s_idx) 460 if (idx < s_idx)
462 continue; 461 goto next;
463 if (inet_fill_rule(skb, r, NETLINK_CB(cb->skb).pid, 462 if (inet_fill_rule(skb, r, NETLINK_CB(cb->skb).pid,
464 cb->nlh->nlmsg_seq, 463 cb->nlh->nlmsg_seq,
465 RTM_NEWRULE, NLM_F_MULTI) < 0) 464 RTM_NEWRULE, NLM_F_MULTI) < 0)
466 break; 465 break;
466next:
467 idx++; 467 idx++;
468 } 468 }
469 rcu_read_unlock(); 469 rcu_read_unlock();
diff --git a/net/ipv4/fib_trie.c b/net/ipv4/fib_trie.c
index 1cb65305e102..23fb9d9768e3 100644
--- a/net/ipv4/fib_trie.c
+++ b/net/ipv4/fib_trie.c
@@ -1252,8 +1252,8 @@ fn_trie_insert(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
1252 */ 1252 */
1253 1253
1254 if (!fa_head) { 1254 if (!fa_head) {
1255 fa_head = fib_insert_node(t, &err, key, plen);
1256 err = 0; 1255 err = 0;
1256 fa_head = fib_insert_node(t, &err, key, plen);
1257 if (err) 1257 if (err)
1258 goto out_free_new_fa; 1258 goto out_free_new_fa;
1259 } 1259 }
diff --git a/net/ipv4/inetpeer.c b/net/ipv4/inetpeer.c
index 2160874ce7aa..03ff62ebcfeb 100644
--- a/net/ipv4/inetpeer.c
+++ b/net/ipv4/inetpeer.c
@@ -86,7 +86,7 @@ static struct inet_peer *peer_root = peer_avl_empty;
86static DEFINE_RWLOCK(peer_pool_lock); 86static DEFINE_RWLOCK(peer_pool_lock);
87#define PEER_MAXDEPTH 40 /* sufficient for about 2^27 nodes */ 87#define PEER_MAXDEPTH 40 /* sufficient for about 2^27 nodes */
88 88
89static volatile int peer_total; 89static int peer_total;
90/* Exported for sysctl_net_ipv4. */ 90/* Exported for sysctl_net_ipv4. */
91int inet_peer_threshold = 65536 + 128; /* start to throw entries more 91int inet_peer_threshold = 65536 + 128; /* start to throw entries more
92 * aggressively at this stage */ 92 * aggressively at this stage */
diff --git a/net/ipv4/ip_input.c b/net/ipv4/ip_input.c
index e1a7dba2fa8a..184c78ca79e6 100644
--- a/net/ipv4/ip_input.c
+++ b/net/ipv4/ip_input.c
@@ -428,6 +428,9 @@ int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt,
428 goto drop; 428 goto drop;
429 } 429 }
430 430
431 /* Remove any debris in the socket control block */
432 memset(&(IPCB(skb)->opt), 0, sizeof(struct ip_options));
433
431 return NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, dev, NULL, 434 return NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, dev, NULL,
432 ip_rcv_finish); 435 ip_rcv_finish);
433 436
diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
index ca0e714613fb..7c9f9a6421b8 100644
--- a/net/ipv4/ip_output.c
+++ b/net/ipv4/ip_output.c
@@ -209,7 +209,7 @@ static inline int ip_finish_output(struct sk_buff *skb)
209 return dst_output(skb); 209 return dst_output(skb);
210 } 210 }
211#endif 211#endif
212 if (skb->len > dst_mtu(skb->dst) && !skb_shinfo(skb)->gso_size) 212 if (skb->len > dst_mtu(skb->dst) && !skb_is_gso(skb))
213 return ip_fragment(skb, ip_finish_output2); 213 return ip_fragment(skb, ip_finish_output2);
214 else 214 else
215 return ip_finish_output2(skb); 215 return ip_finish_output2(skb);
@@ -1095,7 +1095,7 @@ ssize_t ip_append_page(struct sock *sk, struct page *page,
1095 while (size > 0) { 1095 while (size > 0) {
1096 int i; 1096 int i;
1097 1097
1098 if (skb_shinfo(skb)->gso_size) 1098 if (skb_is_gso(skb))
1099 len = size; 1099 len = size;
1100 else { 1100 else {
1101 1101
diff --git a/net/ipv4/ipcomp.c b/net/ipv4/ipcomp.c
index 8e0374847532..8a8b5cf2f7fe 100644
--- a/net/ipv4/ipcomp.c
+++ b/net/ipv4/ipcomp.c
@@ -70,7 +70,8 @@ static int ipcomp_decompress(struct xfrm_state *x, struct sk_buff *skb)
70 if (err) 70 if (err)
71 goto out; 71 goto out;
72 72
73 skb_put(skb, dlen - plen); 73 skb->truesize += dlen - plen;
74 __skb_put(skb, dlen - plen);
74 memcpy(skb->data, scratch, dlen); 75 memcpy(skb->data, scratch, dlen);
75out: 76out:
76 put_cpu(); 77 put_cpu();
diff --git a/net/ipv4/tcp_compound.c b/net/ipv4/tcp_compound.c
deleted file mode 100644
index bc54f7e9aea9..000000000000
--- a/net/ipv4/tcp_compound.c
+++ /dev/null
@@ -1,448 +0,0 @@
1/*
2 * TCP Vegas congestion control
3 *
4 * This is based on the congestion detection/avoidance scheme described in
5 * Lawrence S. Brakmo and Larry L. Peterson.
6 * "TCP Vegas: End to end congestion avoidance on a global internet."
7 * IEEE Journal on Selected Areas in Communication, 13(8):1465--1480,
8 * October 1995. Available from:
9 * ftp://ftp.cs.arizona.edu/xkernel/Papers/jsac.ps
10 *
11 * See http://www.cs.arizona.edu/xkernel/ for their implementation.
12 * The main aspects that distinguish this implementation from the
13 * Arizona Vegas implementation are:
14 * o We do not change the loss detection or recovery mechanisms of
15 * Linux in any way. Linux already recovers from losses quite well,
16 * using fine-grained timers, NewReno, and FACK.
17 * o To avoid the performance penalty imposed by increasing cwnd
18 * only every-other RTT during slow start, we increase during
19 * every RTT during slow start, just like Reno.
20 * o Largely to allow continuous cwnd growth during slow start,
21 * we use the rate at which ACKs come back as the "actual"
22 * rate, rather than the rate at which data is sent.
23 * o To speed convergence to the right rate, we set the cwnd
24 * to achieve the right ("actual") rate when we exit slow start.
25 * o To filter out the noise caused by delayed ACKs, we use the
26 * minimum RTT sample observed during the last RTT to calculate
27 * the actual rate.
28 * o When the sender re-starts from idle, it waits until it has
29 * received ACKs for an entire flight of new data before making
30 * a cwnd adjustment decision. The original Vegas implementation
31 * assumed senders never went idle.
32 *
33 *
34 * TCP Compound based on TCP Vegas
35 *
36 * further details can be found here:
37 * ftp://ftp.research.microsoft.com/pub/tr/TR-2005-86.pdf
38 */
39
40#include <linux/config.h>
41#include <linux/mm.h>
42#include <linux/module.h>
43#include <linux/skbuff.h>
44#include <linux/inet_diag.h>
45
46#include <net/tcp.h>
47
48/* Default values of the Vegas variables, in fixed-point representation
49 * with V_PARAM_SHIFT bits to the right of the binary point.
50 */
51#define V_PARAM_SHIFT 1
52
53#define TCP_COMPOUND_ALPHA 3U
54#define TCP_COMPOUND_BETA 1U
55#define TCP_COMPOUND_GAMMA 30
56#define TCP_COMPOUND_ZETA 1
57
58/* TCP compound variables */
59struct compound {
60 u32 beg_snd_nxt; /* right edge during last RTT */
61 u32 beg_snd_una; /* left edge during last RTT */
62 u32 beg_snd_cwnd; /* saves the size of the cwnd */
63 u8 doing_vegas_now; /* if true, do vegas for this RTT */
64 u16 cntRTT; /* # of RTTs measured within last RTT */
65 u32 minRTT; /* min of RTTs measured within last RTT (in usec) */
66 u32 baseRTT; /* the min of all Vegas RTT measurements seen (in usec) */
67
68 u32 cwnd;
69 u32 dwnd;
70};
71
72/* There are several situations when we must "re-start" Vegas:
73 *
74 * o when a connection is established
75 * o after an RTO
76 * o after fast recovery
77 * o when we send a packet and there is no outstanding
78 * unacknowledged data (restarting an idle connection)
79 *
80 * In these circumstances we cannot do a Vegas calculation at the
81 * end of the first RTT, because any calculation we do is using
82 * stale info -- both the saved cwnd and congestion feedback are
83 * stale.
84 *
85 * Instead we must wait until the completion of an RTT during
86 * which we actually receive ACKs.
87 */
88static inline void vegas_enable(struct sock *sk)
89{
90 const struct tcp_sock *tp = tcp_sk(sk);
91 struct compound *vegas = inet_csk_ca(sk);
92
93 /* Begin taking Vegas samples next time we send something. */
94 vegas->doing_vegas_now = 1;
95
96 /* Set the beginning of the next send window. */
97 vegas->beg_snd_nxt = tp->snd_nxt;
98
99 vegas->cntRTT = 0;
100 vegas->minRTT = 0x7fffffff;
101}
102
103/* Stop taking Vegas samples for now. */
104static inline void vegas_disable(struct sock *sk)
105{
106 struct compound *vegas = inet_csk_ca(sk);
107
108 vegas->doing_vegas_now = 0;
109}
110
111static void tcp_compound_init(struct sock *sk)
112{
113 struct compound *vegas = inet_csk_ca(sk);
114 const struct tcp_sock *tp = tcp_sk(sk);
115
116 vegas->baseRTT = 0x7fffffff;
117 vegas_enable(sk);
118
119 vegas->dwnd = 0;
120 vegas->cwnd = tp->snd_cwnd;
121}
122
123/* Do RTT sampling needed for Vegas.
124 * Basically we:
125 * o min-filter RTT samples from within an RTT to get the current
126 * propagation delay + queuing delay (we are min-filtering to try to
127 * avoid the effects of delayed ACKs)
128 * o min-filter RTT samples from a much longer window (forever for now)
129 * to find the propagation delay (baseRTT)
130 */
131static void tcp_compound_rtt_calc(struct sock *sk, u32 usrtt)
132{
133 struct compound *vegas = inet_csk_ca(sk);
134 u32 vrtt = usrtt + 1; /* Never allow zero rtt or baseRTT */
135
136 /* Filter to find propagation delay: */
137 if (vrtt < vegas->baseRTT)
138 vegas->baseRTT = vrtt;
139
140 /* Find the min RTT during the last RTT to find
141 * the current prop. delay + queuing delay:
142 */
143
144 vegas->minRTT = min(vegas->minRTT, vrtt);
145 vegas->cntRTT++;
146}
147
148static void tcp_compound_state(struct sock *sk, u8 ca_state)
149{
150
151 if (ca_state == TCP_CA_Open)
152 vegas_enable(sk);
153 else
154 vegas_disable(sk);
155}
156
157
158/* 64bit divisor, dividend and result. dynamic precision */
159static inline u64 div64_64(u64 dividend, u64 divisor)
160{
161 u32 d = divisor;
162
163 if (divisor > 0xffffffffULL) {
164 unsigned int shift = fls(divisor >> 32);
165
166 d = divisor >> shift;
167 dividend >>= shift;
168 }
169
170 /* avoid 64 bit division if possible */
171 if (dividend >> 32)
172 do_div(dividend, d);
173 else
174 dividend = (u32) dividend / d;
175
176 return dividend;
177}
178
179/* calculate the quartic root of "a" using Newton-Raphson */
180static u32 qroot(u64 a)
181{
182 u32 x, x1;
183
184 /* Initial estimate is based on:
185 * qrt(x) = exp(log(x) / 4)
186 */
187 x = 1u << (fls64(a) >> 2);
188
189 /*
190 * Iteration based on:
191 * 3
192 * x = ( 3 * x + a / x ) / 4
193 * k+1 k k
194 */
195 do {
196 u64 x3 = x;
197
198 x1 = x;
199 x3 *= x;
200 x3 *= x;
201
202 x = (3 * x + (u32) div64_64(a, x3)) / 4;
203 } while (abs(x1 - x) > 1);
204
205 return x;
206}
207
208
209/*
210 * If the connection is idle and we are restarting,
211 * then we don't want to do any Vegas calculations
212 * until we get fresh RTT samples. So when we
213 * restart, we reset our Vegas state to a clean
214 * slate. After we get acks for this flight of
215 * packets, _then_ we can make Vegas calculations
216 * again.
217 */
218static void tcp_compound_cwnd_event(struct sock *sk, enum tcp_ca_event event)
219{
220 if (event == CA_EVENT_CWND_RESTART || event == CA_EVENT_TX_START)
221 tcp_compound_init(sk);
222}
223
224static void tcp_compound_cong_avoid(struct sock *sk, u32 ack,
225 u32 seq_rtt, u32 in_flight, int flag)
226{
227 struct tcp_sock *tp = tcp_sk(sk);
228 struct compound *vegas = inet_csk_ca(sk);
229 u8 inc = 0;
230
231 if (vegas->cwnd + vegas->dwnd > tp->snd_cwnd) {
232 if (vegas->cwnd > tp->snd_cwnd || vegas->dwnd > tp->snd_cwnd) {
233 vegas->cwnd = tp->snd_cwnd;
234 vegas->dwnd = 0;
235 } else
236 vegas->cwnd = tp->snd_cwnd - vegas->dwnd;
237
238 }
239
240 if (!tcp_is_cwnd_limited(sk, in_flight))
241 return;
242
243 if (vegas->cwnd <= tp->snd_ssthresh)
244 inc = 1;
245 else if (tp->snd_cwnd_cnt < tp->snd_cwnd)
246 tp->snd_cwnd_cnt++;
247
248 if (tp->snd_cwnd_cnt >= tp->snd_cwnd) {
249 inc = 1;
250 tp->snd_cwnd_cnt = 0;
251 }
252
253 if (inc && tp->snd_cwnd < tp->snd_cwnd_clamp)
254 vegas->cwnd++;
255
256 /* The key players are v_beg_snd_una and v_beg_snd_nxt.
257 *
258 * These are so named because they represent the approximate values
259 * of snd_una and snd_nxt at the beginning of the current RTT. More
260 * precisely, they represent the amount of data sent during the RTT.
261 * At the end of the RTT, when we receive an ACK for v_beg_snd_nxt,
262 * we will calculate that (v_beg_snd_nxt - v_beg_snd_una) outstanding
263 * bytes of data have been ACKed during the course of the RTT, giving
264 * an "actual" rate of:
265 *
266 * (v_beg_snd_nxt - v_beg_snd_una) / (rtt duration)
267 *
268 * Unfortunately, v_beg_snd_una is not exactly equal to snd_una,
269 * because delayed ACKs can cover more than one segment, so they
270 * don't line up nicely with the boundaries of RTTs.
271 *
272 * Another unfortunate fact of life is that delayed ACKs delay the
273 * advance of the left edge of our send window, so that the number
274 * of bytes we send in an RTT is often less than our cwnd will allow.
275 * So we keep track of our cwnd separately, in v_beg_snd_cwnd.
276 */
277
278 if (after(ack, vegas->beg_snd_nxt)) {
279 /* Do the Vegas once-per-RTT cwnd adjustment. */
280 u32 old_wnd, old_snd_cwnd;
281
282 /* Here old_wnd is essentially the window of data that was
283 * sent during the previous RTT, and has all
284 * been acknowledged in the course of the RTT that ended
285 * with the ACK we just received. Likewise, old_snd_cwnd
286 * is the cwnd during the previous RTT.
287 */
288 if (!tp->mss_cache)
289 return;
290
291 old_wnd = (vegas->beg_snd_nxt - vegas->beg_snd_una) /
292 tp->mss_cache;
293 old_snd_cwnd = vegas->beg_snd_cwnd;
294
295 /* Save the extent of the current window so we can use this
296 * at the end of the next RTT.
297 */
298 vegas->beg_snd_una = vegas->beg_snd_nxt;
299 vegas->beg_snd_nxt = tp->snd_nxt;
300 vegas->beg_snd_cwnd = tp->snd_cwnd;
301
302 /* We do the Vegas calculations only if we got enough RTT
303 * samples that we can be reasonably sure that we got
304 * at least one RTT sample that wasn't from a delayed ACK.
305 * If we only had 2 samples total,
306 * then that means we're getting only 1 ACK per RTT, which
307 * means they're almost certainly delayed ACKs.
308 * If we have 3 samples, we should be OK.
309 */
310
311 if (vegas->cntRTT > 2) {
312 u32 rtt, target_cwnd, diff;
313 u32 brtt, dwnd;
314
315 /* We have enough RTT samples, so, using the Vegas
316 * algorithm, we determine if we should increase or
317 * decrease cwnd, and by how much.
318 */
319
320 /* Pluck out the RTT we are using for the Vegas
321 * calculations. This is the min RTT seen during the
322 * last RTT. Taking the min filters out the effects
323 * of delayed ACKs, at the cost of noticing congestion
324 * a bit later.
325 */
326 rtt = vegas->minRTT;
327
328 /* Calculate the cwnd we should have, if we weren't
329 * going too fast.
330 *
331 * This is:
332 * (actual rate in segments) * baseRTT
333 * We keep it as a fixed point number with
334 * V_PARAM_SHIFT bits to the right of the binary point.
335 */
336 if (!rtt)
337 return;
338
339 brtt = vegas->baseRTT;
340 target_cwnd = ((old_wnd * brtt)
341 << V_PARAM_SHIFT) / rtt;
342
343 /* Calculate the difference between the window we had,
344 * and the window we would like to have. This quantity
345 * is the "Diff" from the Arizona Vegas papers.
346 *
347 * Again, this is a fixed point number with
348 * V_PARAM_SHIFT bits to the right of the binary
349 * point.
350 */
351
352 diff = (old_wnd << V_PARAM_SHIFT) - target_cwnd;
353
354 dwnd = vegas->dwnd;
355
356 if (diff < (TCP_COMPOUND_GAMMA << V_PARAM_SHIFT)) {
357 u64 v;
358 u32 x;
359
360 /*
361 * The TCP Compound paper describes the choice
362 * of "k" determines the agressiveness,
363 * ie. slope of the response function.
364 *
365 * For same value as HSTCP would be 0.8
366 * but for computaional reasons, both the
367 * original authors and this implementation
368 * use 0.75.
369 */
370 v = old_wnd;
371 x = qroot(v * v * v) >> TCP_COMPOUND_ALPHA;
372 if (x > 1)
373 dwnd = x - 1;
374 else
375 dwnd = 0;
376
377 dwnd += vegas->dwnd;
378
379 } else if ((dwnd << V_PARAM_SHIFT) <
380 (diff * TCP_COMPOUND_BETA))
381 dwnd = 0;
382 else
383 dwnd =
384 ((dwnd << V_PARAM_SHIFT) -
385 (diff *
386 TCP_COMPOUND_BETA)) >> V_PARAM_SHIFT;
387
388 vegas->dwnd = dwnd;
389
390 }
391
392 /* Wipe the slate clean for the next RTT. */
393 vegas->cntRTT = 0;
394 vegas->minRTT = 0x7fffffff;
395 }
396
397 tp->snd_cwnd = vegas->cwnd + vegas->dwnd;
398}
399
400/* Extract info for Tcp socket info provided via netlink. */
401static void tcp_compound_get_info(struct sock *sk, u32 ext, struct sk_buff *skb)
402{
403 const struct compound *ca = inet_csk_ca(sk);
404 if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
405 struct tcpvegas_info *info;
406
407 info = RTA_DATA(__RTA_PUT(skb, INET_DIAG_VEGASINFO,
408 sizeof(*info)));
409
410 info->tcpv_enabled = ca->doing_vegas_now;
411 info->tcpv_rttcnt = ca->cntRTT;
412 info->tcpv_rtt = ca->baseRTT;
413 info->tcpv_minrtt = ca->minRTT;
414 rtattr_failure:;
415 }
416}
417
418static struct tcp_congestion_ops tcp_compound = {
419 .init = tcp_compound_init,
420 .ssthresh = tcp_reno_ssthresh,
421 .cong_avoid = tcp_compound_cong_avoid,
422 .rtt_sample = tcp_compound_rtt_calc,
423 .set_state = tcp_compound_state,
424 .cwnd_event = tcp_compound_cwnd_event,
425 .get_info = tcp_compound_get_info,
426
427 .owner = THIS_MODULE,
428 .name = "compound",
429};
430
431static int __init tcp_compound_register(void)
432{
433 BUG_ON(sizeof(struct compound) > ICSK_CA_PRIV_SIZE);
434 tcp_register_congestion_control(&tcp_compound);
435 return 0;
436}
437
438static void __exit tcp_compound_unregister(void)
439{
440 tcp_unregister_congestion_control(&tcp_compound);
441}
442
443module_init(tcp_compound_register);
444module_exit(tcp_compound_unregister);
445
446MODULE_AUTHOR("Angelo P. Castellani, Stephen Hemminger");
447MODULE_LICENSE("GPL");
448MODULE_DESCRIPTION("TCP Compound");
diff --git a/net/ipv4/tcp_highspeed.c b/net/ipv4/tcp_highspeed.c
index aaa1538c0692..fa3e1aad660c 100644
--- a/net/ipv4/tcp_highspeed.c
+++ b/net/ipv4/tcp_highspeed.c
@@ -139,14 +139,19 @@ static void hstcp_cong_avoid(struct sock *sk, u32 adk, u32 rtt,
139 tp->snd_cwnd++; 139 tp->snd_cwnd++;
140 } 140 }
141 } else { 141 } else {
142 /* Update AIMD parameters */ 142 /* Update AIMD parameters.
143 *
144 * We want to guarantee that:
145 * hstcp_aimd_vals[ca->ai-1].cwnd <
146 * snd_cwnd <=
147 * hstcp_aimd_vals[ca->ai].cwnd
148 */
143 if (tp->snd_cwnd > hstcp_aimd_vals[ca->ai].cwnd) { 149 if (tp->snd_cwnd > hstcp_aimd_vals[ca->ai].cwnd) {
144 while (tp->snd_cwnd > hstcp_aimd_vals[ca->ai].cwnd && 150 while (tp->snd_cwnd > hstcp_aimd_vals[ca->ai].cwnd &&
145 ca->ai < HSTCP_AIMD_MAX - 1) 151 ca->ai < HSTCP_AIMD_MAX - 1)
146 ca->ai++; 152 ca->ai++;
147 } else if (tp->snd_cwnd < hstcp_aimd_vals[ca->ai].cwnd) { 153 } else if (ca->ai && tp->snd_cwnd <= hstcp_aimd_vals[ca->ai-1].cwnd) {
148 while (tp->snd_cwnd > hstcp_aimd_vals[ca->ai].cwnd && 154 while (ca->ai && tp->snd_cwnd <= hstcp_aimd_vals[ca->ai-1].cwnd)
149 ca->ai > 0)
150 ca->ai--; 155 ca->ai--;
151 } 156 }
152 157
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index 5a886e6efbbe..a891133f00e4 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -496,6 +496,24 @@ void tcp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
496 } 496 }
497} 497}
498 498
499int tcp_v4_gso_send_check(struct sk_buff *skb)
500{
501 struct iphdr *iph;
502 struct tcphdr *th;
503
504 if (!pskb_may_pull(skb, sizeof(*th)))
505 return -EINVAL;
506
507 iph = skb->nh.iph;
508 th = skb->h.th;
509
510 th->check = 0;
511 th->check = ~tcp_v4_check(th, skb->len, iph->saddr, iph->daddr, 0);
512 skb->csum = offsetof(struct tcphdr, check);
513 skb->ip_summed = CHECKSUM_HW;
514 return 0;
515}
516
499/* 517/*
500 * This routine will send an RST to the other tcp. 518 * This routine will send an RST to the other tcp.
501 * 519 *
diff --git a/net/ipv4/xfrm4_output.c b/net/ipv4/xfrm4_output.c
index 193363e22932..d16f863cf687 100644
--- a/net/ipv4/xfrm4_output.c
+++ b/net/ipv4/xfrm4_output.c
@@ -134,7 +134,7 @@ static int xfrm4_output_finish(struct sk_buff *skb)
134 } 134 }
135#endif 135#endif
136 136
137 if (!skb_shinfo(skb)->gso_size) 137 if (!skb_is_gso(skb))
138 return xfrm4_output_finish2(skb); 138 return xfrm4_output_finish2(skb);
139 139
140 skb->protocol = htons(ETH_P_IP); 140 skb->protocol = htons(ETH_P_IP);
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-28 00:16:02 -0500