aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@g5.osdl.org>2005-11-07 11:05:11 -0500
committerLinus Torvalds <torvalds@g5.osdl.org>2005-11-07 11:05:11 -0500
commit8e33ba49765484bc6de3a2f8143733713fa93bc1 (patch)
tree2ea080e478e4ee86a893b75db2d5c81ce14cbf10
parent8cde0776ec1e86c270f65bf482f96288e6bf0023 (diff)
parent2d43f1128a4282fbe8442f40b4cbbac05d8f10aa (diff)
Merge master.kernel.org:/pub/scm/linux/kernel/git/acme/net-2.6
-rw-r--r--include/linux/pkt_sched.h50
-rw-r--r--include/linux/skbuff.h38
-rw-r--r--include/net/inet_ecn.h28
-rw-r--r--include/net/inet_hashtables.h2
-rw-r--r--include/net/red.h325
-rw-r--r--net/core/stream.c12
-rw-r--r--net/dccp/ipv4.c32
-rw-r--r--net/ipv4/inet_connection_sock.c14
-rw-r--r--net/ipv4/netfilter/ip_conntrack_helper_pptp.c4
-rw-r--r--net/ipv4/netfilter/ip_conntrack_netlink.c19
-rw-r--r--net/ipv4/netfilter/ip_nat_core.c6
-rw-r--r--net/ipv4/netfilter/ip_nat_helper_pptp.c2
-rw-r--r--net/ipv4/netfilter/ip_nat_proto_gre.c4
-rw-r--r--net/ipv4/netfilter/ip_nat_proto_unknown.c2
-rw-r--r--net/ipv4/netfilter/ipt_CONNMARK.c1
-rw-r--r--net/ipv4/tcp.c1
-rw-r--r--net/ipv4/tcp_ipv4.c2
-rw-r--r--net/ipv6/tcp_ipv6.c15
-rw-r--r--net/netfilter/nf_queue.c2
-rw-r--r--net/netfilter/nfnetlink_log.c6
-rw-r--r--net/netfilter/nfnetlink_queue.c6
-rw-r--r--net/sched/sch_gred.c841
-rw-r--r--net/sched/sch_netem.c122
-rw-r--r--net/sched/sch_red.c418
24 files changed, 1066 insertions, 886 deletions
diff --git a/include/linux/pkt_sched.h b/include/linux/pkt_sched.h
index 60ffcb9c5791..e87b233615b3 100644
--- a/include/linux/pkt_sched.h
+++ b/include/linux/pkt_sched.h
@@ -93,6 +93,7 @@ struct tc_fifo_qopt
93/* PRIO section */ 93/* PRIO section */
94 94
95#define TCQ_PRIO_BANDS 16 95#define TCQ_PRIO_BANDS 16
96#define TCQ_MIN_PRIO_BANDS 2
96 97
97struct tc_prio_qopt 98struct tc_prio_qopt
98{ 99{
@@ -169,6 +170,7 @@ struct tc_red_qopt
169 unsigned char Scell_log; /* cell size for idle damping */ 170 unsigned char Scell_log; /* cell size for idle damping */
170 unsigned char flags; 171 unsigned char flags;
171#define TC_RED_ECN 1 172#define TC_RED_ECN 1
173#define TC_RED_HARDDROP 2
172}; 174};
173 175
174struct tc_red_xstats 176struct tc_red_xstats
@@ -194,38 +196,34 @@ enum
194 196
195#define TCA_GRED_MAX (__TCA_GRED_MAX - 1) 197#define TCA_GRED_MAX (__TCA_GRED_MAX - 1)
196 198
197#define TCA_SET_OFF TCA_GRED_PARMS
198struct tc_gred_qopt 199struct tc_gred_qopt
199{ 200{
200 __u32 limit; /* HARD maximal queue length (bytes) 201 __u32 limit; /* HARD maximal queue length (bytes) */
201*/ 202 __u32 qth_min; /* Min average length threshold (bytes) */
202 __u32 qth_min; /* Min average length threshold (bytes) 203 __u32 qth_max; /* Max average length threshold (bytes) */
203*/ 204 __u32 DP; /* upto 2^32 DPs */
204 __u32 qth_max; /* Max average length threshold (bytes) 205 __u32 backlog;
205*/ 206 __u32 qave;
206 __u32 DP; /* upto 2^32 DPs */ 207 __u32 forced;
207 __u32 backlog; 208 __u32 early;
208 __u32 qave; 209 __u32 other;
209 __u32 forced; 210 __u32 pdrop;
210 __u32 early; 211 __u8 Wlog; /* log(W) */
211 __u32 other; 212 __u8 Plog; /* log(P_max/(qth_max-qth_min)) */
212 __u32 pdrop; 213 __u8 Scell_log; /* cell size for idle damping */
213 214 __u8 prio; /* prio of this VQ */
214 unsigned char Wlog; /* log(W) */ 215 __u32 packets;
215 unsigned char Plog; /* log(P_max/(qth_max-qth_min)) */ 216 __u32 bytesin;
216 unsigned char Scell_log; /* cell size for idle damping */
217 __u8 prio; /* prio of this VQ */
218 __u32 packets;
219 __u32 bytesin;
220}; 217};
218
221/* gred setup */ 219/* gred setup */
222struct tc_gred_sopt 220struct tc_gred_sopt
223{ 221{
224 __u32 DPs; 222 __u32 DPs;
225 __u32 def_DP; 223 __u32 def_DP;
226 __u8 grio; 224 __u8 grio;
227 __u8 pad1; 225 __u8 flags;
228 __u16 pad2; 226 __u16 pad1;
229}; 227};
230 228
231/* HTB section */ 229/* HTB section */
diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 4286d832166f..fdfb8fe8c38c 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -603,23 +603,23 @@ static inline void skb_queue_head_init(struct sk_buff_head *list)
603 */ 603 */
604 604
605/** 605/**
606 * __skb_queue_head - queue a buffer at the list head 606 * __skb_queue_after - queue a buffer at the list head
607 * @list: list to use 607 * @list: list to use
608 * @prev: place after this buffer
608 * @newsk: buffer to queue 609 * @newsk: buffer to queue
609 * 610 *
610 * Queue a buffer at the start of a list. This function takes no locks 611 * Queue a buffer int the middle of a list. This function takes no locks
611 * and you must therefore hold required locks before calling it. 612 * and you must therefore hold required locks before calling it.
612 * 613 *
613 * A buffer cannot be placed on two lists at the same time. 614 * A buffer cannot be placed on two lists at the same time.
614 */ 615 */
615extern void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 616static inline void __skb_queue_after(struct sk_buff_head *list,
616static inline void __skb_queue_head(struct sk_buff_head *list, 617 struct sk_buff *prev,
617 struct sk_buff *newsk) 618 struct sk_buff *newsk)
618{ 619{
619 struct sk_buff *prev, *next; 620 struct sk_buff *next;
620
621 list->qlen++; 621 list->qlen++;
622 prev = (struct sk_buff *)list; 622
623 next = prev->next; 623 next = prev->next;
624 newsk->next = next; 624 newsk->next = next;
625 newsk->prev = prev; 625 newsk->prev = prev;
@@ -627,6 +627,23 @@ static inline void __skb_queue_head(struct sk_buff_head *list,
627} 627}
628 628
629/** 629/**
630 * __skb_queue_head - queue a buffer at the list head
631 * @list: list to use
632 * @newsk: buffer to queue
633 *
634 * Queue a buffer at the start of a list. This function takes no locks
635 * and you must therefore hold required locks before calling it.
636 *
637 * A buffer cannot be placed on two lists at the same time.
638 */
639extern void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
640static inline void __skb_queue_head(struct sk_buff_head *list,
641 struct sk_buff *newsk)
642{
643 __skb_queue_after(list, (struct sk_buff *)list, newsk);
644}
645
646/**
630 * __skb_queue_tail - queue a buffer at the list tail 647 * __skb_queue_tail - queue a buffer at the list tail
631 * @list: list to use 648 * @list: list to use
632 * @newsk: buffer to queue 649 * @newsk: buffer to queue
@@ -1203,6 +1220,11 @@ static inline void kunmap_skb_frag(void *vaddr)
1203 prefetch(skb->next), (skb != (struct sk_buff *)(queue)); \ 1220 prefetch(skb->next), (skb != (struct sk_buff *)(queue)); \
1204 skb = skb->next) 1221 skb = skb->next)
1205 1222
1223#define skb_queue_reverse_walk(queue, skb) \
1224 for (skb = (queue)->prev; \
1225 prefetch(skb->prev), (skb != (struct sk_buff *)(queue)); \
1226 skb = skb->prev)
1227
1206 1228
1207extern struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, 1229extern struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
1208 int noblock, int *err); 1230 int noblock, int *err);
diff --git a/include/net/inet_ecn.h b/include/net/inet_ecn.h
index f87845e2e965..b0c47e2eccf1 100644
--- a/include/net/inet_ecn.h
+++ b/include/net/inet_ecn.h
@@ -2,6 +2,7 @@
2#define _INET_ECN_H_ 2#define _INET_ECN_H_
3 3
4#include <linux/ip.h> 4#include <linux/ip.h>
5#include <linux/skbuff.h>
5#include <net/dsfield.h> 6#include <net/dsfield.h>
6 7
7enum { 8enum {
@@ -48,7 +49,7 @@ static inline __u8 INET_ECN_encapsulate(__u8 outer, __u8 inner)
48 (label) |= __constant_htons(INET_ECN_ECT_0 << 4); \ 49 (label) |= __constant_htons(INET_ECN_ECT_0 << 4); \
49 } while (0) 50 } while (0)
50 51
51static inline void IP_ECN_set_ce(struct iphdr *iph) 52static inline int IP_ECN_set_ce(struct iphdr *iph)
52{ 53{
53 u32 check = iph->check; 54 u32 check = iph->check;
54 u32 ecn = (iph->tos + 1) & INET_ECN_MASK; 55 u32 ecn = (iph->tos + 1) & INET_ECN_MASK;
@@ -61,7 +62,7 @@ static inline void IP_ECN_set_ce(struct iphdr *iph)
61 * INET_ECN_CE => 00 62 * INET_ECN_CE => 00
62 */ 63 */
63 if (!(ecn & 2)) 64 if (!(ecn & 2))
64 return; 65 return !ecn;
65 66
66 /* 67 /*
67 * The following gives us: 68 * The following gives us:
@@ -72,6 +73,7 @@ static inline void IP_ECN_set_ce(struct iphdr *iph)
72 73
73 iph->check = check + (check>=0xFFFF); 74 iph->check = check + (check>=0xFFFF);
74 iph->tos |= INET_ECN_CE; 75 iph->tos |= INET_ECN_CE;
76 return 1;
75} 77}
76 78
77static inline void IP_ECN_clear(struct iphdr *iph) 79static inline void IP_ECN_clear(struct iphdr *iph)
@@ -87,11 +89,12 @@ static inline void ipv4_copy_dscp(struct iphdr *outer, struct iphdr *inner)
87 89
88struct ipv6hdr; 90struct ipv6hdr;
89 91
90static inline void IP6_ECN_set_ce(struct ipv6hdr *iph) 92static inline int IP6_ECN_set_ce(struct ipv6hdr *iph)
91{ 93{
92 if (INET_ECN_is_not_ect(ipv6_get_dsfield(iph))) 94 if (INET_ECN_is_not_ect(ipv6_get_dsfield(iph)))
93 return; 95 return 0;
94 *(u32*)iph |= htonl(INET_ECN_CE << 20); 96 *(u32*)iph |= htonl(INET_ECN_CE << 20);
97 return 1;
95} 98}
96 99
97static inline void IP6_ECN_clear(struct ipv6hdr *iph) 100static inline void IP6_ECN_clear(struct ipv6hdr *iph)
@@ -105,4 +108,21 @@ static inline void ipv6_copy_dscp(struct ipv6hdr *outer, struct ipv6hdr *inner)
105 ipv6_change_dsfield(inner, INET_ECN_MASK, dscp); 108 ipv6_change_dsfield(inner, INET_ECN_MASK, dscp);
106} 109}
107 110
111static inline int INET_ECN_set_ce(struct sk_buff *skb)
112{
113 switch (skb->protocol) {
114 case __constant_htons(ETH_P_IP):
115 if (skb->nh.raw + sizeof(struct iphdr) <= skb->tail)
116 return IP_ECN_set_ce(skb->nh.iph);
117 break;
118
119 case __constant_htons(ETH_P_IPV6):
120 if (skb->nh.raw + sizeof(struct ipv6hdr) <= skb->tail)
121 return IP6_ECN_set_ce(skb->nh.ipv6h);
122 break;
123 }
124
125 return 0;
126}
127
108#endif 128#endif
diff --git a/include/net/inet_hashtables.h b/include/net/inet_hashtables.h
index f50f95968340..07840baa9341 100644
--- a/include/net/inet_hashtables.h
+++ b/include/net/inet_hashtables.h
@@ -125,9 +125,7 @@ struct inet_hashinfo {
125 rwlock_t lhash_lock ____cacheline_aligned; 125 rwlock_t lhash_lock ____cacheline_aligned;
126 atomic_t lhash_users; 126 atomic_t lhash_users;
127 wait_queue_head_t lhash_wait; 127 wait_queue_head_t lhash_wait;
128 spinlock_t portalloc_lock;
129 kmem_cache_t *bind_bucket_cachep; 128 kmem_cache_t *bind_bucket_cachep;
130 int port_rover;
131}; 129};
132 130
133static inline unsigned int inet_ehashfn(const __u32 laddr, const __u16 lport, 131static inline unsigned int inet_ehashfn(const __u32 laddr, const __u16 lport,
diff --git a/include/net/red.h b/include/net/red.h
new file mode 100644
index 000000000000..2ed4358e3295
--- /dev/null
+++ b/include/net/red.h
@@ -0,0 +1,325 @@
1#ifndef __NET_SCHED_RED_H
2#define __NET_SCHED_RED_H
3
4#include <linux/config.h>
5#include <linux/types.h>
6#include <net/pkt_sched.h>
7#include <net/inet_ecn.h>
8#include <net/dsfield.h>
9
10/* Random Early Detection (RED) algorithm.
11 =======================================
12
13 Source: Sally Floyd and Van Jacobson, "Random Early Detection Gateways
14 for Congestion Avoidance", 1993, IEEE/ACM Transactions on Networking.
15
16 This file codes a "divisionless" version of RED algorithm
17 as written down in Fig.17 of the paper.
18
19 Short description.
20 ------------------
21
22 When a new packet arrives we calculate the average queue length:
23
24 avg = (1-W)*avg + W*current_queue_len,
25
26 W is the filter time constant (chosen as 2^(-Wlog)), it controls
27 the inertia of the algorithm. To allow larger bursts, W should be
28 decreased.
29
30 if (avg > th_max) -> packet marked (dropped).
31 if (avg < th_min) -> packet passes.
32 if (th_min < avg < th_max) we calculate probability:
33
34 Pb = max_P * (avg - th_min)/(th_max-th_min)
35
36 and mark (drop) packet with this probability.
37 Pb changes from 0 (at avg==th_min) to max_P (avg==th_max).
38 max_P should be small (not 1), usually 0.01..0.02 is good value.
39
40 max_P is chosen as a number, so that max_P/(th_max-th_min)
41 is a negative power of two in order arithmetics to contain
42 only shifts.
43
44
45 Parameters, settable by user:
46 -----------------------------
47
48 qth_min - bytes (should be < qth_max/2)
49 qth_max - bytes (should be at least 2*qth_min and less limit)
50 Wlog - bits (<32) log(1/W).
51 Plog - bits (<32)
52
53 Plog is related to max_P by formula:
54
55 max_P = (qth_max-qth_min)/2^Plog;
56
57 F.e. if qth_max=128K and qth_min=32K, then Plog=22
58 corresponds to max_P=0.02
59
60 Scell_log
61 Stab
62
63 Lookup table for log((1-W)^(t/t_ave).
64
65
66 NOTES:
67
68 Upper bound on W.
69 -----------------
70
71 If you want to allow bursts of L packets of size S,
72 you should choose W:
73
74 L + 1 - th_min/S < (1-(1-W)^L)/W
75
76 th_min/S = 32 th_min/S = 4
77
78 log(W) L
79 -1 33
80 -2 35
81 -3 39
82 -4 46
83 -5 57
84 -6 75
85 -7 101
86 -8 135
87 -9 190
88 etc.
89 */
90
91#define RED_STAB_SIZE 256
92#define RED_STAB_MASK (RED_STAB_SIZE - 1)
93
94struct red_stats
95{
96 u32 prob_drop; /* Early probability drops */
97 u32 prob_mark; /* Early probability marks */
98 u32 forced_drop; /* Forced drops, qavg > max_thresh */
99 u32 forced_mark; /* Forced marks, qavg > max_thresh */
100 u32 pdrop; /* Drops due to queue limits */
101 u32 other; /* Drops due to drop() calls */
102 u32 backlog;
103};
104
105struct red_parms
106{
107 /* Parameters */
108 u32 qth_min; /* Min avg length threshold: A scaled */
109 u32 qth_max; /* Max avg length threshold: A scaled */
110 u32 Scell_max;
111 u32 Rmask; /* Cached random mask, see red_rmask */
112 u8 Scell_log;
113 u8 Wlog; /* log(W) */
114 u8 Plog; /* random number bits */
115 u8 Stab[RED_STAB_SIZE];
116
117 /* Variables */
118 int qcount; /* Number of packets since last random
119 number generation */
120 u32 qR; /* Cached random number */
121
122 unsigned long qavg; /* Average queue length: A scaled */
123 psched_time_t qidlestart; /* Start of current idle period */
124};
125
126static inline u32 red_rmask(u8 Plog)
127{
128 return Plog < 32 ? ((1 << Plog) - 1) : ~0UL;
129}
130
131static inline void red_set_parms(struct red_parms *p,
132 u32 qth_min, u32 qth_max, u8 Wlog, u8 Plog,
133 u8 Scell_log, u8 *stab)
134{
135 /* Reset average queue length, the value is strictly bound
136 * to the parameters below, reseting hurts a bit but leaving
137 * it might result in an unreasonable qavg for a while. --TGR
138 */
139 p->qavg = 0;
140
141 p->qcount = -1;
142 p->qth_min = qth_min << Wlog;
143 p->qth_max = qth_max << Wlog;
144 p->Wlog = Wlog;
145 p->Plog = Plog;
146 p->Rmask = red_rmask(Plog);
147 p->Scell_log = Scell_log;
148 p->Scell_max = (255 << Scell_log);
149
150 memcpy(p->Stab, stab, sizeof(p->Stab));
151}
152
153static inline int red_is_idling(struct red_parms *p)
154{
155 return !PSCHED_IS_PASTPERFECT(p->qidlestart);
156}
157
158static inline void red_start_of_idle_period(struct red_parms *p)
159{
160 PSCHED_GET_TIME(p->qidlestart);
161}
162
163static inline void red_end_of_idle_period(struct red_parms *p)
164{
165 PSCHED_SET_PASTPERFECT(p->qidlestart);
166}
167
168static inline void red_restart(struct red_parms *p)
169{
170 red_end_of_idle_period(p);
171 p->qavg = 0;
172 p->qcount = -1;
173}
174
175static inline unsigned long red_calc_qavg_from_idle_time(struct red_parms *p)
176{
177 psched_time_t now;
178 long us_idle;
179 int shift;
180
181 PSCHED_GET_TIME(now);
182 us_idle = PSCHED_TDIFF_SAFE(now, p->qidlestart, p->Scell_max);
183
184 /*
185 * The problem: ideally, average length queue recalcultion should
186 * be done over constant clock intervals. This is too expensive, so
187 * that the calculation is driven by outgoing packets.
188 * When the queue is idle we have to model this clock by hand.
189 *
190 * SF+VJ proposed to "generate":
191 *
192 * m = idletime / (average_pkt_size / bandwidth)
193 *
194 * dummy packets as a burst after idle time, i.e.
195 *
196 * p->qavg *= (1-W)^m
197 *
198 * This is an apparently overcomplicated solution (f.e. we have to
199 * precompute a table to make this calculation in reasonable time)
200 * I believe that a simpler model may be used here,
201 * but it is field for experiments.
202 */
203
204 shift = p->Stab[(us_idle >> p->Scell_log) & RED_STAB_MASK];
205
206 if (shift)
207 return p->qavg >> shift;
208 else {
209 /* Approximate initial part of exponent with linear function:
210 *
211 * (1-W)^m ~= 1-mW + ...
212 *
213 * Seems, it is the best solution to
214 * problem of too coarse exponent tabulation.
215 */
216 us_idle = (p->qavg * us_idle) >> p->Scell_log;
217
218 if (us_idle < (p->qavg >> 1))
219 return p->qavg - us_idle;
220 else
221 return p->qavg >> 1;
222 }
223}
224
225static inline unsigned long red_calc_qavg_no_idle_time(struct red_parms *p,
226 unsigned int backlog)
227{
228 /*
229 * NOTE: p->qavg is fixed point number with point at Wlog.
230 * The formula below is equvalent to floating point
231 * version:
232 *
233 * qavg = qavg*(1-W) + backlog*W;
234 *
235 * --ANK (980924)
236 */
237 return p->qavg + (backlog - (p->qavg >> p->Wlog));
238}
239
240static inline unsigned long red_calc_qavg(struct red_parms *p,
241 unsigned int backlog)
242{
243 if (!red_is_idling(p))
244 return red_calc_qavg_no_idle_time(p, backlog);
245 else
246 return red_calc_qavg_from_idle_time(p);
247}
248
249static inline u32 red_random(struct red_parms *p)
250{
251 return net_random() & p->Rmask;
252}
253
254static inline int red_mark_probability(struct red_parms *p, unsigned long qavg)
255{
256 /* The formula used below causes questions.
257
258 OK. qR is random number in the interval 0..Rmask
259 i.e. 0..(2^Plog). If we used floating point
260 arithmetics, it would be: (2^Plog)*rnd_num,
261 where rnd_num is less 1.
262
263 Taking into account, that qavg have fixed
264 point at Wlog, and Plog is related to max_P by
265 max_P = (qth_max-qth_min)/2^Plog; two lines
266 below have the following floating point equivalent:
267
268 max_P*(qavg - qth_min)/(qth_max-qth_min) < rnd/qcount
269
270 Any questions? --ANK (980924)
271 */
272 return !(((qavg - p->qth_min) >> p->Wlog) * p->qcount < p->qR);
273}
274
275enum {
276 RED_BELOW_MIN_THRESH,
277 RED_BETWEEN_TRESH,
278 RED_ABOVE_MAX_TRESH,
279};
280
281static inline int red_cmp_thresh(struct red_parms *p, unsigned long qavg)
282{
283 if (qavg < p->qth_min)
284 return RED_BELOW_MIN_THRESH;
285 else if (qavg >= p->qth_max)
286 return RED_ABOVE_MAX_TRESH;
287 else
288 return RED_BETWEEN_TRESH;
289}
290
291enum {
292 RED_DONT_MARK,
293 RED_PROB_MARK,
294 RED_HARD_MARK,
295};
296
297static inline int red_action(struct red_parms *p, unsigned long qavg)
298{
299 switch (red_cmp_thresh(p, qavg)) {
300 case RED_BELOW_MIN_THRESH:
301 p->qcount = -1;
302 return RED_DONT_MARK;
303
304 case RED_BETWEEN_TRESH:
305 if (++p->qcount) {
306 if (red_mark_probability(p, qavg)) {
307 p->qcount = 0;
308 p->qR = red_random(p);
309 return RED_PROB_MARK;
310 }
311 } else
312 p->qR = red_random(p);
313
314 return RED_DONT_MARK;
315
316 case RED_ABOVE_MAX_TRESH:
317 p->qcount = -1;
318 return RED_HARD_MARK;
319 }
320
321 BUG();
322 return RED_DONT_MARK;
323}
324
325#endif
diff --git a/net/core/stream.c b/net/core/stream.c
index ac9edfdf8742..15bfd03e8024 100644
--- a/net/core/stream.c
+++ b/net/core/stream.c
@@ -52,8 +52,9 @@ int sk_stream_wait_connect(struct sock *sk, long *timeo_p)
52{ 52{
53 struct task_struct *tsk = current; 53 struct task_struct *tsk = current;
54 DEFINE_WAIT(wait); 54 DEFINE_WAIT(wait);
55 int done;
55 56
56 while (1) { 57 do {
57 if (sk->sk_err) 58 if (sk->sk_err)
58 return sock_error(sk); 59 return sock_error(sk);
59 if ((1 << sk->sk_state) & ~(TCPF_SYN_SENT | TCPF_SYN_RECV)) 60 if ((1 << sk->sk_state) & ~(TCPF_SYN_SENT | TCPF_SYN_RECV))
@@ -65,13 +66,12 @@ int sk_stream_wait_connect(struct sock *sk, long *timeo_p)
65 66
66 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE); 67 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
67 sk->sk_write_pending++; 68 sk->sk_write_pending++;
68 if (sk_wait_event(sk, timeo_p, 69 done = sk_wait_event(sk, timeo_p,
69 !((1 << sk->sk_state) & 70 !((1 << sk->sk_state) &
70 ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)))) 71 ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)));
71 break;
72 finish_wait(sk->sk_sleep, &wait); 72 finish_wait(sk->sk_sleep, &wait);
73 sk->sk_write_pending--; 73 sk->sk_write_pending--;
74 } 74 } while (!done);
75 return 0; 75 return 0;
76} 76}
77 77
diff --git a/net/dccp/ipv4.c b/net/dccp/ipv4.c
index 6298cf58ff9e..4b9bc81ae1a3 100644
--- a/net/dccp/ipv4.c
+++ b/net/dccp/ipv4.c
@@ -31,8 +31,6 @@ struct inet_hashinfo __cacheline_aligned dccp_hashinfo = {
31 .lhash_lock = RW_LOCK_UNLOCKED, 31 .lhash_lock = RW_LOCK_UNLOCKED,
32 .lhash_users = ATOMIC_INIT(0), 32 .lhash_users = ATOMIC_INIT(0),
33 .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(dccp_hashinfo.lhash_wait), 33 .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(dccp_hashinfo.lhash_wait),
34 .portalloc_lock = SPIN_LOCK_UNLOCKED,
35 .port_rover = 1024 - 1,
36}; 34};
37 35
38EXPORT_SYMBOL_GPL(dccp_hashinfo); 36EXPORT_SYMBOL_GPL(dccp_hashinfo);
@@ -125,36 +123,15 @@ static int dccp_v4_hash_connect(struct sock *sk)
125 int ret; 123 int ret;
126 124
127 if (snum == 0) { 125 if (snum == 0) {
128 int rover;
129 int low = sysctl_local_port_range[0]; 126 int low = sysctl_local_port_range[0];
130 int high = sysctl_local_port_range[1]; 127 int high = sysctl_local_port_range[1];
131 int remaining = (high - low) + 1; 128 int remaining = (high - low) + 1;
129 int rover = net_random() % (high - low) + low;
132 struct hlist_node *node; 130 struct hlist_node *node;
133 struct inet_timewait_sock *tw = NULL; 131 struct inet_timewait_sock *tw = NULL;
134 132
135 local_bh_disable(); 133 local_bh_disable();
136
137 /* TODO. Actually it is not so bad idea to remove
138 * dccp_hashinfo.portalloc_lock before next submission to
139 * Linus.
140 * As soon as we touch this place at all it is time to think.
141 *
142 * Now it protects single _advisory_ variable
143 * dccp_hashinfo.port_rover, hence it is mostly useless.
144 * Code will work nicely if we just delete it, but
145 * I am afraid in contented case it will work not better or
146 * even worse: another cpu just will hit the same bucket
147 * and spin there.
148 * So some cpu salt could remove both contention and
149 * memory pingpong. Any ideas how to do this in a nice way?
150 */
151 spin_lock(&dccp_hashinfo.portalloc_lock);
152 rover = dccp_hashinfo.port_rover;
153
154 do { 134 do {
155 rover++;
156 if ((rover < low) || (rover > high))
157 rover = low;
158 head = &dccp_hashinfo.bhash[inet_bhashfn(rover, 135 head = &dccp_hashinfo.bhash[inet_bhashfn(rover,
159 dccp_hashinfo.bhash_size)]; 136 dccp_hashinfo.bhash_size)];
160 spin_lock(&head->lock); 137 spin_lock(&head->lock);
@@ -187,9 +164,9 @@ static int dccp_v4_hash_connect(struct sock *sk)
187 164
188 next_port: 165 next_port:
189 spin_unlock(&head->lock); 166 spin_unlock(&head->lock);
167 if (++rover > high)
168 rover = low;
190 } while (--remaining > 0); 169 } while (--remaining > 0);
191 dccp_hashinfo.port_rover = rover;
192 spin_unlock(&dccp_hashinfo.portalloc_lock);
193 170
194 local_bh_enable(); 171 local_bh_enable();
195 172
@@ -197,9 +174,6 @@ static int dccp_v4_hash_connect(struct sock *sk)
197 174
198ok: 175ok:
199 /* All locks still held and bhs disabled */ 176 /* All locks still held and bhs disabled */
200 dccp_hashinfo.port_rover = rover;
201 spin_unlock(&dccp_hashinfo.portalloc_lock);
202
203 inet_bind_hash(sk, tb, rover); 177 inet_bind_hash(sk, tb, rover);
204 if (sk_unhashed(sk)) { 178 if (sk_unhashed(sk)) {
205 inet_sk(sk)->sport = htons(rover); 179 inet_sk(sk)->sport = htons(rover);
diff --git a/net/ipv4/inet_connection_sock.c b/net/ipv4/inet_connection_sock.c
index 94468a76c5b4..3fe021f1a566 100644
--- a/net/ipv4/inet_connection_sock.c
+++ b/net/ipv4/inet_connection_sock.c
@@ -78,17 +78,9 @@ int inet_csk_get_port(struct inet_hashinfo *hashinfo,
78 int low = sysctl_local_port_range[0]; 78 int low = sysctl_local_port_range[0];
79 int high = sysctl_local_port_range[1]; 79 int high = sysctl_local_port_range[1];
80 int remaining = (high - low) + 1; 80 int remaining = (high - low) + 1;
81 int rover; 81 int rover = net_random() % (high - low) + low;
82 82
83 spin_lock(&hashinfo->portalloc_lock);
84 if (hashinfo->port_rover < low)
85 rover = low;
86 else
87 rover = hashinfo->port_rover;
88 do { 83 do {
89 rover++;
90 if (rover > high)
91 rover = low;
92 head = &hashinfo->bhash[inet_bhashfn(rover, hashinfo->bhash_size)]; 84 head = &hashinfo->bhash[inet_bhashfn(rover, hashinfo->bhash_size)];
93 spin_lock(&head->lock); 85 spin_lock(&head->lock);
94 inet_bind_bucket_for_each(tb, node, &head->chain) 86 inet_bind_bucket_for_each(tb, node, &head->chain)
@@ -97,9 +89,9 @@ int inet_csk_get_port(struct inet_hashinfo *hashinfo,
97 break; 89 break;
98 next: 90 next:
99 spin_unlock(&head->lock); 91 spin_unlock(&head->lock);
92 if (++rover > high)
93 rover = low;
100 } while (--remaining > 0); 94 } while (--remaining > 0);
101 hashinfo->port_rover = rover;
102 spin_unlock(&hashinfo->portalloc_lock);
103 95
104 /* Exhausted local port range during search? It is not 96 /* Exhausted local port range during search? It is not
105 * possible for us to be holding one of the bind hash 97 * possible for us to be holding one of the bind hash
diff --git a/net/ipv4/netfilter/ip_conntrack_helper_pptp.c b/net/ipv4/netfilter/ip_conntrack_helper_pptp.c
index 926a6684643d..4108a5e12b3c 100644
--- a/net/ipv4/netfilter/ip_conntrack_helper_pptp.c
+++ b/net/ipv4/netfilter/ip_conntrack_helper_pptp.c
@@ -270,14 +270,10 @@ exp_gre(struct ip_conntrack *master,
270 exp_orig->expectfn = pptp_expectfn; 270 exp_orig->expectfn = pptp_expectfn;
271 exp_orig->flags = 0; 271 exp_orig->flags = 0;
272 272
273 exp_orig->dir = IP_CT_DIR_ORIGINAL;
274
275 /* both expectations are identical apart from tuple */ 273 /* both expectations are identical apart from tuple */
276 memcpy(exp_reply, exp_orig, sizeof(*exp_reply)); 274 memcpy(exp_reply, exp_orig, sizeof(*exp_reply));
277 memcpy(&exp_reply->tuple, &exp_tuples[1], sizeof(exp_reply->tuple)); 275 memcpy(&exp_reply->tuple, &exp_tuples[1], sizeof(exp_reply->tuple));
278 276
279 exp_reply->dir = !exp_orig->dir;
280
281 if (ip_nat_pptp_hook_exp_gre) 277 if (ip_nat_pptp_hook_exp_gre)
282 ret = ip_nat_pptp_hook_exp_gre(exp_orig, exp_reply); 278 ret = ip_nat_pptp_hook_exp_gre(exp_orig, exp_reply);
283 else { 279 else {
diff --git a/net/ipv4/netfilter/ip_conntrack_netlink.c b/net/ipv4/netfilter/ip_conntrack_netlink.c
index 166e6069f121..82a65043a8ef 100644
--- a/net/ipv4/netfilter/ip_conntrack_netlink.c
+++ b/net/ipv4/netfilter/ip_conntrack_netlink.c
@@ -815,7 +815,7 @@ ctnetlink_get_conntrack(struct sock *ctnl, struct sk_buff *skb,
815 IPCTNL_MSG_CT_NEW, 1, ct); 815 IPCTNL_MSG_CT_NEW, 1, ct);
816 ip_conntrack_put(ct); 816 ip_conntrack_put(ct);
817 if (err <= 0) 817 if (err <= 0)
818 goto out; 818 goto free;
819 819
820 err = netlink_unicast(ctnl, skb2, NETLINK_CB(skb).pid, MSG_DONTWAIT); 820 err = netlink_unicast(ctnl, skb2, NETLINK_CB(skb).pid, MSG_DONTWAIT);
821 if (err < 0) 821 if (err < 0)
@@ -824,9 +824,9 @@ ctnetlink_get_conntrack(struct sock *ctnl, struct sk_buff *skb,
824 DEBUGP("leaving\n"); 824 DEBUGP("leaving\n");
825 return 0; 825 return 0;
826 826
827free:
828 kfree_skb(skb2);
827out: 829out:
828 if (skb2)
829 kfree_skb(skb2);
830 return -1; 830 return -1;
831} 831}
832 832
@@ -1322,21 +1322,16 @@ ctnetlink_get_expect(struct sock *ctnl, struct sk_buff *skb,
1322 nlh->nlmsg_seq, IPCTNL_MSG_EXP_NEW, 1322 nlh->nlmsg_seq, IPCTNL_MSG_EXP_NEW,
1323 1, exp); 1323 1, exp);
1324 if (err <= 0) 1324 if (err <= 0)
1325 goto out; 1325 goto free;
1326 1326
1327 ip_conntrack_expect_put(exp); 1327 ip_conntrack_expect_put(exp);
1328 1328
1329 err = netlink_unicast(ctnl, skb2, NETLINK_CB(skb).pid, MSG_DONTWAIT); 1329 return netlink_unicast(ctnl, skb2, NETLINK_CB(skb).pid, MSG_DONTWAIT);
1330 if (err < 0)
1331 goto free;
1332
1333 return err;
1334 1330
1331free:
1332 kfree_skb(skb2);
1335out: 1333out:
1336 ip_conntrack_expect_put(exp); 1334 ip_conntrack_expect_put(exp);
1337free:
1338 if (skb2)
1339 kfree_skb(skb2);
1340 return err; 1335 return err;
1341} 1336}
1342 1337
diff --git a/net/ipv4/netfilter/ip_nat_core.c b/net/ipv4/netfilter/ip_nat_core.c
index c5e3abd24672..762f4d93936b 100644
--- a/net/ipv4/netfilter/ip_nat_core.c
+++ b/net/ipv4/netfilter/ip_nat_core.c
@@ -66,10 +66,8 @@ ip_nat_proto_find_get(u_int8_t protonum)
66 * removed until we've grabbed the reference */ 66 * removed until we've grabbed the reference */
67 preempt_disable(); 67 preempt_disable();
68 p = __ip_nat_proto_find(protonum); 68 p = __ip_nat_proto_find(protonum);
69 if (p) { 69 if (!try_module_get(p->me))
70 if (!try_module_get(p->me)) 70 p = &ip_nat_unknown_protocol;
71 p = &ip_nat_unknown_protocol;
72 }
73 preempt_enable(); 71 preempt_enable();
74 72
75 return p; 73 return p;
diff --git a/net/ipv4/netfilter/ip_nat_helper_pptp.c b/net/ipv4/netfilter/ip_nat_helper_pptp.c
index 3cdd0684d30d..ee6ab74ad3a9 100644
--- a/net/ipv4/netfilter/ip_nat_helper_pptp.c
+++ b/net/ipv4/netfilter/ip_nat_helper_pptp.c
@@ -216,6 +216,7 @@ pptp_exp_gre(struct ip_conntrack_expect *expect_orig,
216 expect_orig->saved_proto.gre.key = htons(nat_pptp_info->pac_call_id); 216 expect_orig->saved_proto.gre.key = htons(nat_pptp_info->pac_call_id);
217 expect_orig->tuple.src.u.gre.key = htons(nat_pptp_info->pns_call_id); 217 expect_orig->tuple.src.u.gre.key = htons(nat_pptp_info->pns_call_id);
218 expect_orig->tuple.dst.u.gre.key = htons(ct_pptp_info->pac_call_id); 218 expect_orig->tuple.dst.u.gre.key = htons(ct_pptp_info->pac_call_id);
219 expect_orig->dir = IP_CT_DIR_ORIGINAL;
219 inv_t.src.ip = reply_t->src.ip; 220 inv_t.src.ip = reply_t->src.ip;
220 inv_t.dst.ip = reply_t->dst.ip; 221 inv_t.dst.ip = reply_t->dst.ip;
221 inv_t.src.u.gre.key = htons(nat_pptp_info->pac_call_id); 222 inv_t.src.u.gre.key = htons(nat_pptp_info->pac_call_id);
@@ -233,6 +234,7 @@ pptp_exp_gre(struct ip_conntrack_expect *expect_orig,
233 expect_reply->saved_proto.gre.key = htons(nat_pptp_info->pns_call_id); 234 expect_reply->saved_proto.gre.key = htons(nat_pptp_info->pns_call_id);
234 expect_reply->tuple.src.u.gre.key = htons(nat_pptp_info->pac_call_id); 235 expect_reply->tuple.src.u.gre.key = htons(nat_pptp_info->pac_call_id);
235 expect_reply->tuple.dst.u.gre.key = htons(ct_pptp_info->pns_call_id); 236 expect_reply->tuple.dst.u.gre.key = htons(ct_pptp_info->pns_call_id);
237 expect_reply->dir = IP_CT_DIR_REPLY;
236 inv_t.src.ip = orig_t->src.ip; 238 inv_t.src.ip = orig_t->src.ip;
237 inv_t.dst.ip = orig_t->dst.ip; 239 inv_t.dst.ip = orig_t->dst.ip;
238 inv_t.src.u.gre.key = htons(nat_pptp_info->pns_call_id); 240 inv_t.src.u.gre.key = htons(nat_pptp_info->pns_call_id);
diff --git a/net/ipv4/netfilter/ip_nat_proto_gre.c b/net/ipv4/netfilter/ip_nat_proto_gre.c
index 7c1285401672..f7cad7cf1aec 100644
--- a/net/ipv4/netfilter/ip_nat_proto_gre.c
+++ b/net/ipv4/netfilter/ip_nat_proto_gre.c
@@ -139,8 +139,8 @@ gre_manip_pkt(struct sk_buff **pskb,
139 break; 139 break;
140 case GRE_VERSION_PPTP: 140 case GRE_VERSION_PPTP:
141 DEBUGP("call_id -> 0x%04x\n", 141 DEBUGP("call_id -> 0x%04x\n",
142 ntohl(tuple->dst.u.gre.key)); 142 ntohs(tuple->dst.u.gre.key));
143 pgreh->call_id = htons(ntohl(tuple->dst.u.gre.key)); 143 pgreh->call_id = tuple->dst.u.gre.key;
144 break; 144 break;
145 default: 145 default:
146 DEBUGP("can't nat unknown GRE version\n"); 146 DEBUGP("can't nat unknown GRE version\n");
diff --git a/net/ipv4/netfilter/ip_nat_proto_unknown.c b/net/ipv4/netfilter/ip_nat_proto_unknown.c
index 99bbef56f84e..f0099a646a0b 100644
--- a/net/ipv4/netfilter/ip_nat_proto_unknown.c
+++ b/net/ipv4/netfilter/ip_nat_proto_unknown.c
@@ -62,7 +62,7 @@ unknown_print_range(char *buffer, const struct ip_nat_range *range)
62 62
63struct ip_nat_protocol ip_nat_unknown_protocol = { 63struct ip_nat_protocol ip_nat_unknown_protocol = {
64 .name = "unknown", 64 .name = "unknown",
65 .me = THIS_MODULE, 65 /* .me isn't set: getting a ref to this cannot fail. */
66 .manip_pkt = unknown_manip_pkt, 66 .manip_pkt = unknown_manip_pkt,
67 .in_range = unknown_in_range, 67 .in_range = unknown_in_range,
68 .unique_tuple = unknown_unique_tuple, 68 .unique_tuple = unknown_unique_tuple,
diff --git a/net/ipv4/netfilter/ipt_CONNMARK.c b/net/ipv4/netfilter/ipt_CONNMARK.c
index 134638021339..05d66ab59424 100644
--- a/net/ipv4/netfilter/ipt_CONNMARK.c
+++ b/net/ipv4/netfilter/ipt_CONNMARK.c
@@ -109,6 +109,7 @@ static struct ipt_target ipt_connmark_reg = {
109 109
110static int __init init(void) 110static int __init init(void)
111{ 111{
112 need_ip_conntrack();
112 return ipt_register_target(&ipt_connmark_reg); 113 return ipt_register_target(&ipt_connmark_reg);
113} 114}
114 115
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index f3f0013a9580..72b7c22e1ea5 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -2112,7 +2112,6 @@ void __init tcp_init(void)
2112 sysctl_tcp_max_orphans >>= (3 - order); 2112 sysctl_tcp_max_orphans >>= (3 - order);
2113 sysctl_max_syn_backlog = 128; 2113 sysctl_max_syn_backlog = 128;
2114 } 2114 }
2115 tcp_hashinfo.port_rover = sysctl_local_port_range[0] - 1;
2116 2115
2117 sysctl_tcp_mem[0] = 768 << order; 2116 sysctl_tcp_mem[0] = 768 << order;
2118 sysctl_tcp_mem[1] = 1024 << order; 2117 sysctl_tcp_mem[1] = 1024 << order;
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index c85819d8474b..49d67cd75edd 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -93,8 +93,6 @@ struct inet_hashinfo __cacheline_aligned tcp_hashinfo = {
93 .lhash_lock = RW_LOCK_UNLOCKED, 93 .lhash_lock = RW_LOCK_UNLOCKED,
94 .lhash_users = ATOMIC_INIT(0), 94 .lhash_users = ATOMIC_INIT(0),
95 .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(tcp_hashinfo.lhash_wait), 95 .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(tcp_hashinfo.lhash_wait),
96 .portalloc_lock = SPIN_LOCK_UNLOCKED,
97 .port_rover = 1024 - 1,
98}; 96};
99 97
100static int tcp_v4_get_port(struct sock *sk, unsigned short snum) 98static int tcp_v4_get_port(struct sock *sk, unsigned short snum)
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index d693cb988b78..d746d3b27efb 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -114,16 +114,9 @@ static int tcp_v6_get_port(struct sock *sk, unsigned short snum)
114 int low = sysctl_local_port_range[0]; 114 int low = sysctl_local_port_range[0];
115 int high = sysctl_local_port_range[1]; 115 int high = sysctl_local_port_range[1];
116 int remaining = (high - low) + 1; 116 int remaining = (high - low) + 1;
117 int rover; 117 int rover = net_random() % (high - low) + low;
118 118
119 spin_lock(&tcp_hashinfo.portalloc_lock); 119 do {
120 if (tcp_hashinfo.port_rover < low)
121 rover = low;
122 else
123 rover = tcp_hashinfo.port_rover;
124 do { rover++;
125 if (rover > high)
126 rover = low;
127 head = &tcp_hashinfo.bhash[inet_bhashfn(rover, tcp_hashinfo.bhash_size)]; 120 head = &tcp_hashinfo.bhash[inet_bhashfn(rover, tcp_hashinfo.bhash_size)];
128 spin_lock(&head->lock); 121 spin_lock(&head->lock);
129 inet_bind_bucket_for_each(tb, node, &head->chain) 122 inet_bind_bucket_for_each(tb, node, &head->chain)
@@ -132,9 +125,9 @@ static int tcp_v6_get_port(struct sock *sk, unsigned short snum)
132 break; 125 break;
133 next: 126 next:
134 spin_unlock(&head->lock); 127 spin_unlock(&head->lock);
128 if (++rover > high)
129 rover = low;
135 } while (--remaining > 0); 130 } while (--remaining > 0);
136 tcp_hashinfo.port_rover = rover;
137 spin_unlock(&tcp_hashinfo.portalloc_lock);
138 131
139 /* Exhausted local port range during search? It is not 132 /* Exhausted local port range during search? It is not
140 * possible for us to be holding one of the bind hash 133 * possible for us to be holding one of the bind hash
diff --git a/net/netfilter/nf_queue.c b/net/netfilter/nf_queue.c
index d10d552d9c40..d3a4f30a7f22 100644
--- a/net/netfilter/nf_queue.c
+++ b/net/netfilter/nf_queue.c
@@ -117,7 +117,7 @@ int nf_queue(struct sk_buff **skb,
117 117
118 /* QUEUE == DROP if noone is waiting, to be safe. */ 118 /* QUEUE == DROP if noone is waiting, to be safe. */
119 read_lock(&queue_handler_lock); 119 read_lock(&queue_handler_lock);
120 if (!queue_handler[pf]->outfn) { 120 if (!queue_handler[pf] || !queue_handler[pf]->outfn) {
121 read_unlock(&queue_handler_lock); 121 read_unlock(&queue_handler_lock);
122 kfree_skb(*skb); 122 kfree_skb(*skb);
123 return 1; 123 return 1;
diff --git a/net/netfilter/nfnetlink_log.c b/net/netfilter/nfnetlink_log.c
index efcd10f996ba..d194676f3655 100644
--- a/net/netfilter/nfnetlink_log.c
+++ b/net/netfilter/nfnetlink_log.c
@@ -146,11 +146,10 @@ instance_create(u_int16_t group_num, int pid)
146 goto out_unlock; 146 goto out_unlock;
147 } 147 }
148 148
149 inst = kmalloc(sizeof(*inst), GFP_ATOMIC); 149 inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
150 if (!inst) 150 if (!inst)
151 goto out_unlock; 151 goto out_unlock;
152 152
153 memset(inst, 0, sizeof(*inst));
154 INIT_HLIST_NODE(&inst->hlist); 153 INIT_HLIST_NODE(&inst->hlist);
155 inst->lock = SPIN_LOCK_UNLOCKED; 154 inst->lock = SPIN_LOCK_UNLOCKED;
156 /* needs to be two, since we _put() after creation */ 155 /* needs to be two, since we _put() after creation */
@@ -962,10 +961,9 @@ static int nful_open(struct inode *inode, struct file *file)
962 struct iter_state *is; 961 struct iter_state *is;
963 int ret; 962 int ret;
964 963
965 is = kmalloc(sizeof(*is), GFP_KERNEL); 964 is = kzalloc(sizeof(*is), GFP_KERNEL);
966 if (!is) 965 if (!is)
967 return -ENOMEM; 966 return -ENOMEM;
968 memset(is, 0, sizeof(*is));
969 ret = seq_open(file, &nful_seq_ops); 967 ret = seq_open(file, &nful_seq_ops);
970 if (ret < 0) 968 if (ret < 0)
971 goto out_free; 969 goto out_free;
diff --git a/net/netfilter/nfnetlink_queue.c b/net/netfilter/nfnetlink_queue.c
index eaa44c49567b..f065a6c94953 100644
--- a/net/netfilter/nfnetlink_queue.c
+++ b/net/netfilter/nfnetlink_queue.c
@@ -136,11 +136,10 @@ instance_create(u_int16_t queue_num, int pid)
136 goto out_unlock; 136 goto out_unlock;
137 } 137 }
138 138
139 inst = kmalloc(sizeof(*inst), GFP_ATOMIC); 139 inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
140 if (!inst) 140 if (!inst)
141 goto out_unlock; 141 goto out_unlock;
142 142
143 memset(inst, 0, sizeof(*inst));
144 inst->queue_num = queue_num; 143 inst->queue_num = queue_num;
145 inst->peer_pid = pid; 144 inst->peer_pid = pid;
146 inst->queue_maxlen = NFQNL_QMAX_DEFAULT; 145 inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
@@ -1036,10 +1035,9 @@ static int nfqnl_open(struct inode *inode, struct file *file)
1036 struct iter_state *is; 1035 struct iter_state *is;
1037 int ret; 1036 int ret;
1038 1037
1039 is = kmalloc(sizeof(*is), GFP_KERNEL); 1038 is = kzalloc(sizeof(*is), GFP_KERNEL);
1040 if (!is) 1039 if (!is)
1041 return -ENOMEM; 1040 return -ENOMEM;
1042 memset(is, 0, sizeof(*is));
1043 ret = seq_open(file, &nfqnl_seq_ops); 1041 ret = seq_open(file, &nfqnl_seq_ops);
1044 if (ret < 0) 1042 if (ret < 0)
1045 goto out_free; 1043 goto out_free;
diff --git a/net/sched/sch_gred.c b/net/sched/sch_gred.c
index 25c171c32715..29a2dd9f3029 100644
--- a/net/sched/sch_gred.c
+++ b/net/sched/sch_gred.c
@@ -15,247 +15,281 @@
15 * from Ren Liu 15 * from Ren Liu
16 * - More error checks 16 * - More error checks
17 * 17 *
18 * 18 * For all the glorious comments look at include/net/red.h
19 *
20 * For all the glorious comments look at Alexey's sch_red.c
21 */ 19 */
22 20
23#include <linux/config.h> 21#include <linux/config.h>
24#include <linux/module.h> 22#include <linux/module.h>
25#include <asm/uaccess.h>
26#include <asm/system.h>
27#include <linux/bitops.h>
28#include <linux/types.h> 23#include <linux/types.h>
29#include <linux/kernel.h> 24#include <linux/kernel.h>
30#include <linux/sched.h>
31#include <linux/string.h>
32#include <linux/mm.h>
33#include <linux/socket.h>
34#include <linux/sockios.h>
35#include <linux/in.h>
36#include <linux/errno.h>
37#include <linux/interrupt.h>
38#include <linux/if_ether.h>
39#include <linux/inet.h>
40#include <linux/netdevice.h> 25#include <linux/netdevice.h>
41#include <linux/etherdevice.h>
42#include <linux/notifier.h>
43#include <net/ip.h>
44#include <net/route.h>
45#include <linux/skbuff.h> 26#include <linux/skbuff.h>
46#include <net/sock.h>
47#include <net/pkt_sched.h> 27#include <net/pkt_sched.h>
28#include <net/red.h>
48 29
49#if 1 /* control */ 30#define GRED_DEF_PRIO (MAX_DPs / 2)
50#define DPRINTK(format,args...) printk(KERN_DEBUG format,##args) 31#define GRED_VQ_MASK (MAX_DPs - 1)
51#else
52#define DPRINTK(format,args...)
53#endif
54
55#if 0 /* data */
56#define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args)
57#else
58#define D2PRINTK(format,args...)
59#endif
60 32
61struct gred_sched_data; 33struct gred_sched_data;
62struct gred_sched; 34struct gred_sched;
63 35
64struct gred_sched_data 36struct gred_sched_data
65{ 37{
66/* Parameters */
67 u32 limit; /* HARD maximal queue length */ 38 u32 limit; /* HARD maximal queue length */
68 u32 qth_min; /* Min average length threshold: A scaled */
69 u32 qth_max; /* Max average length threshold: A scaled */
70 u32 DP; /* the drop pramaters */ 39 u32 DP; /* the drop pramaters */
71 char Wlog; /* log(W) */
72 char Plog; /* random number bits */
73 u32 Scell_max;
74 u32 Rmask;
75 u32 bytesin; /* bytes seen on virtualQ so far*/ 40 u32 bytesin; /* bytes seen on virtualQ so far*/
76 u32 packetsin; /* packets seen on virtualQ so far*/ 41 u32 packetsin; /* packets seen on virtualQ so far*/
77 u32 backlog; /* bytes on the virtualQ */ 42 u32 backlog; /* bytes on the virtualQ */
78 u32 forced; /* packets dropped for exceeding limits */ 43 u8 prio; /* the prio of this vq */
79 u32 early; /* packets dropped as a warning */ 44
80 u32 other; /* packets dropped by invoking drop() */ 45 struct red_parms parms;
81 u32 pdrop; /* packets dropped because we exceeded physical queue limits */ 46 struct red_stats stats;
82 char Scell_log; 47};
83 u8 Stab[256]; 48
84 u8 prio; /* the prio of this vq */ 49enum {
85 50 GRED_WRED_MODE = 1,
86/* Variables */ 51 GRED_RIO_MODE,
87 unsigned long qave; /* Average queue length: A scaled */
88 int qcount; /* Packets since last random number generation */
89 u32 qR; /* Cached random number */
90
91 psched_time_t qidlestart; /* Start of idle period */
92}; 52};
93 53
94struct gred_sched 54struct gred_sched
95{ 55{
96 struct gred_sched_data *tab[MAX_DPs]; 56 struct gred_sched_data *tab[MAX_DPs];
97 u32 DPs; 57 unsigned long flags;
98 u32 def; 58 u32 red_flags;
99 u8 initd; 59 u32 DPs;
100 u8 grio; 60 u32 def;
101 u8 eqp; 61 struct red_parms wred_set;
102}; 62};
103 63
104static int 64static inline int gred_wred_mode(struct gred_sched *table)
105gred_enqueue(struct sk_buff *skb, struct Qdisc* sch)
106{ 65{
107 psched_time_t now; 66 return test_bit(GRED_WRED_MODE, &table->flags);
108 struct gred_sched_data *q=NULL; 67}
109 struct gred_sched *t= qdisc_priv(sch); 68
110 unsigned long qave=0; 69static inline void gred_enable_wred_mode(struct gred_sched *table)
111 int i=0; 70{
71 __set_bit(GRED_WRED_MODE, &table->flags);
72}
73
74static inline void gred_disable_wred_mode(struct gred_sched *table)
75{
76 __clear_bit(GRED_WRED_MODE, &table->flags);
77}
78
79static inline int gred_rio_mode(struct gred_sched *table)
80{
81 return test_bit(GRED_RIO_MODE, &table->flags);
82}
83
84static inline void gred_enable_rio_mode(struct gred_sched *table)
85{
86 __set_bit(GRED_RIO_MODE, &table->flags);
87}
88
89static inline void gred_disable_rio_mode(struct gred_sched *table)
90{
91 __clear_bit(GRED_RIO_MODE, &table->flags);
92}
93
94static inline int gred_wred_mode_check(struct Qdisc *sch)
95{
96 struct gred_sched *table = qdisc_priv(sch);
97 int i;
112 98
113 if (!t->initd && skb_queue_len(&sch->q) < (sch->dev->tx_queue_len ? : 1)) { 99 /* Really ugly O(n^2) but shouldn't be necessary too frequent. */
114 D2PRINTK("NO GRED Queues setup yet! Enqueued anyway\n"); 100 for (i = 0; i < table->DPs; i++) {
115 goto do_enqueue; 101 struct gred_sched_data *q = table->tab[i];
102 int n;
103
104 if (q == NULL)
105 continue;
106
107 for (n = 0; n < table->DPs; n++)
108 if (table->tab[n] && table->tab[n] != q &&
109 table->tab[n]->prio == q->prio)
110 return 1;
116 } 111 }
117 112
113 return 0;
114}
115
116static inline unsigned int gred_backlog(struct gred_sched *table,
117 struct gred_sched_data *q,
118 struct Qdisc *sch)
119{
120 if (gred_wred_mode(table))
121 return sch->qstats.backlog;
122 else
123 return q->backlog;
124}
125
126static inline u16 tc_index_to_dp(struct sk_buff *skb)
127{
128 return skb->tc_index & GRED_VQ_MASK;
129}
130
131static inline void gred_load_wred_set(struct gred_sched *table,
132 struct gred_sched_data *q)
133{
134 q->parms.qavg = table->wred_set.qavg;
135 q->parms.qidlestart = table->wred_set.qidlestart;
136}
137
138static inline void gred_store_wred_set(struct gred_sched *table,
139 struct gred_sched_data *q)
140{
141 table->wred_set.qavg = q->parms.qavg;
142}
143
144static inline int gred_use_ecn(struct gred_sched *t)
145{
146 return t->red_flags & TC_RED_ECN;
147}
118 148
119 if ( ((skb->tc_index&0xf) > (t->DPs -1)) || !(q=t->tab[skb->tc_index&0xf])) { 149static inline int gred_use_harddrop(struct gred_sched *t)
120 printk("GRED: setting to default (%d)\n ",t->def); 150{
121 if (!(q=t->tab[t->def])) { 151 return t->red_flags & TC_RED_HARDDROP;
122 DPRINTK("GRED: setting to default FAILED! dropping!! " 152}
123 "(%d)\n ", t->def); 153
124 goto drop; 154static int gred_enqueue(struct sk_buff *skb, struct Qdisc* sch)
155{
156 struct gred_sched_data *q=NULL;
157 struct gred_sched *t= qdisc_priv(sch);
158 unsigned long qavg = 0;
159 u16 dp = tc_index_to_dp(skb);
160
161 if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
162 dp = t->def;
163
164 if ((q = t->tab[dp]) == NULL) {
165 /* Pass through packets not assigned to a DP
166 * if no default DP has been configured. This
167 * allows for DP flows to be left untouched.
168 */
169 if (skb_queue_len(&sch->q) < sch->dev->tx_queue_len)
170 return qdisc_enqueue_tail(skb, sch);
171 else
172 goto drop;
125 } 173 }
174
126 /* fix tc_index? --could be controvesial but needed for 175 /* fix tc_index? --could be controvesial but needed for
127 requeueing */ 176 requeueing */
128 skb->tc_index=(skb->tc_index&0xfffffff0) | t->def; 177 skb->tc_index = (skb->tc_index & ~GRED_VQ_MASK) | dp;
129 } 178 }
130 179
131 D2PRINTK("gred_enqueue virtualQ 0x%x classid %x backlog %d " 180 /* sum up all the qaves of prios <= to ours to get the new qave */
132 "general backlog %d\n",skb->tc_index&0xf,sch->handle,q->backlog, 181 if (!gred_wred_mode(t) && gred_rio_mode(t)) {
133 sch->qstats.backlog); 182 int i;
134 /* sum up all the qaves of prios <= to ours to get the new qave*/ 183
135 if (!t->eqp && t->grio) { 184 for (i = 0; i < t->DPs; i++) {
136 for (i=0;i<t->DPs;i++) { 185 if (t->tab[i] && t->tab[i]->prio < q->prio &&
137 if ((!t->tab[i]) || (i==q->DP)) 186 !red_is_idling(&t->tab[i]->parms))
138 continue; 187 qavg +=t->tab[i]->parms.qavg;
139
140 if ((t->tab[i]->prio < q->prio) && (PSCHED_IS_PASTPERFECT(t->tab[i]->qidlestart)))
141 qave +=t->tab[i]->qave;
142 } 188 }
143 189
144 } 190 }
145 191
146 q->packetsin++; 192 q->packetsin++;
147 q->bytesin+=skb->len; 193 q->bytesin += skb->len;
148 194
149 if (t->eqp && t->grio) { 195 if (gred_wred_mode(t))
150 qave=0; 196 gred_load_wred_set(t, q);
151 q->qave=t->tab[t->def]->qave;
152 q->qidlestart=t->tab[t->def]->qidlestart;
153 }
154 197
155 if (!PSCHED_IS_PASTPERFECT(q->qidlestart)) { 198 q->parms.qavg = red_calc_qavg(&q->parms, gred_backlog(t, q, sch));
156 long us_idle;
157 PSCHED_GET_TIME(now);
158 us_idle = PSCHED_TDIFF_SAFE(now, q->qidlestart, q->Scell_max);
159 PSCHED_SET_PASTPERFECT(q->qidlestart);
160 199
161 q->qave >>= q->Stab[(us_idle>>q->Scell_log)&0xFF]; 200 if (red_is_idling(&q->parms))
162 } else { 201 red_end_of_idle_period(&q->parms);
163 if (t->eqp) {
164 q->qave += sch->qstats.backlog - (q->qave >> q->Wlog);
165 } else {
166 q->qave += q->backlog - (q->qave >> q->Wlog);
167 }
168 202
169 } 203 if (gred_wred_mode(t))
170 204 gred_store_wred_set(t, q);
171
172 if (t->eqp && t->grio)
173 t->tab[t->def]->qave=q->qave;
174
175 if ((q->qave+qave) < q->qth_min) {
176 q->qcount = -1;
177enqueue:
178 if (q->backlog + skb->len <= q->limit) {
179 q->backlog += skb->len;
180do_enqueue:
181 __skb_queue_tail(&sch->q, skb);
182 sch->qstats.backlog += skb->len;
183 sch->bstats.bytes += skb->len;
184 sch->bstats.packets++;
185 return 0;
186 } else {
187 q->pdrop++;
188 }
189 205
190drop: 206 switch (red_action(&q->parms, q->parms.qavg + qavg)) {
191 kfree_skb(skb); 207 case RED_DONT_MARK:
192 sch->qstats.drops++; 208 break;
193 return NET_XMIT_DROP; 209
194 } 210 case RED_PROB_MARK:
195 if ((q->qave+qave) >= q->qth_max) { 211 sch->qstats.overlimits++;
196 q->qcount = -1; 212 if (!gred_use_ecn(t) || !INET_ECN_set_ce(skb)) {
197 sch->qstats.overlimits++; 213 q->stats.prob_drop++;
198 q->forced++; 214 goto congestion_drop;
199 goto drop; 215 }
216
217 q->stats.prob_mark++;
218 break;
219
220 case RED_HARD_MARK:
221 sch->qstats.overlimits++;
222 if (gred_use_harddrop(t) || !gred_use_ecn(t) ||
223 !INET_ECN_set_ce(skb)) {
224 q->stats.forced_drop++;
225 goto congestion_drop;
226 }
227 q->stats.forced_mark++;
228 break;
200 } 229 }
201 if (++q->qcount) { 230
202 if ((((qave+q->qave) - q->qth_min)>>q->Wlog)*q->qcount < q->qR) 231 if (q->backlog + skb->len <= q->limit) {
203 goto enqueue; 232 q->backlog += skb->len;
204 q->qcount = 0; 233 return qdisc_enqueue_tail(skb, sch);
205 q->qR = net_random()&q->Rmask;
206 sch->qstats.overlimits++;
207 q->early++;
208 goto drop;
209 } 234 }
210 q->qR = net_random()&q->Rmask; 235
211 goto enqueue; 236 q->stats.pdrop++;
237drop:
238 return qdisc_drop(skb, sch);
239
240congestion_drop:
241 qdisc_drop(skb, sch);
242 return NET_XMIT_CN;
212} 243}
213 244
214static int 245static int gred_requeue(struct sk_buff *skb, struct Qdisc* sch)
215gred_requeue(struct sk_buff *skb, struct Qdisc* sch)
216{ 246{
247 struct gred_sched *t = qdisc_priv(sch);
217 struct gred_sched_data *q; 248 struct gred_sched_data *q;
218 struct gred_sched *t= qdisc_priv(sch); 249 u16 dp = tc_index_to_dp(skb);
219 q= t->tab[(skb->tc_index&0xf)]; 250
220/* error checking here -- probably unnecessary */ 251 if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
221 PSCHED_SET_PASTPERFECT(q->qidlestart); 252 if (net_ratelimit())
222 253 printk(KERN_WARNING "GRED: Unable to relocate VQ 0x%x "
223 __skb_queue_head(&sch->q, skb); 254 "for requeue, screwing up backlog.\n",
224 sch->qstats.backlog += skb->len; 255 tc_index_to_dp(skb));
225 sch->qstats.requeues++; 256 } else {
226 q->backlog += skb->len; 257 if (red_is_idling(&q->parms))
227 return 0; 258 red_end_of_idle_period(&q->parms);
259 q->backlog += skb->len;
260 }
261
262 return qdisc_requeue(skb, sch);
228} 263}
229 264
230static struct sk_buff * 265static struct sk_buff *gred_dequeue(struct Qdisc* sch)
231gred_dequeue(struct Qdisc* sch)
232{ 266{
233 struct sk_buff *skb; 267 struct sk_buff *skb;
234 struct gred_sched_data *q; 268 struct gred_sched *t = qdisc_priv(sch);
235 struct gred_sched *t= qdisc_priv(sch); 269
270 skb = qdisc_dequeue_head(sch);
236 271
237 skb = __skb_dequeue(&sch->q);
238 if (skb) { 272 if (skb) {
239 sch->qstats.backlog -= skb->len; 273 struct gred_sched_data *q;
240 q= t->tab[(skb->tc_index&0xf)]; 274 u16 dp = tc_index_to_dp(skb);
241 if (q) { 275
242 q->backlog -= skb->len; 276 if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
243 if (!q->backlog && !t->eqp) 277 if (net_ratelimit())
244 PSCHED_GET_TIME(q->qidlestart); 278 printk(KERN_WARNING "GRED: Unable to relocate "
279 "VQ 0x%x after dequeue, screwing up "
280 "backlog.\n", tc_index_to_dp(skb));
245 } else { 281 } else {
246 D2PRINTK("gred_dequeue: skb has bad tcindex %x\n",skb->tc_index&0xf); 282 q->backlog -= skb->len;
283
284 if (!q->backlog && !gred_wred_mode(t))
285 red_start_of_idle_period(&q->parms);
247 } 286 }
287
248 return skb; 288 return skb;
249 } 289 }
250 290
251 if (t->eqp) { 291 if (gred_wred_mode(t) && !red_is_idling(&t->wred_set))
252 q= t->tab[t->def]; 292 red_start_of_idle_period(&t->wred_set);
253 if (!q)
254 D2PRINTK("no default VQ set: Results will be "
255 "screwed up\n");
256 else
257 PSCHED_GET_TIME(q->qidlestart);
258 }
259 293
260 return NULL; 294 return NULL;
261} 295}
@@ -263,36 +297,34 @@ gred_dequeue(struct Qdisc* sch)
263static unsigned int gred_drop(struct Qdisc* sch) 297static unsigned int gred_drop(struct Qdisc* sch)
264{ 298{
265 struct sk_buff *skb; 299 struct sk_buff *skb;
300 struct gred_sched *t = qdisc_priv(sch);
266 301
267 struct gred_sched_data *q; 302 skb = qdisc_dequeue_tail(sch);
268 struct gred_sched *t= qdisc_priv(sch);
269
270 skb = __skb_dequeue_tail(&sch->q);
271 if (skb) { 303 if (skb) {
272 unsigned int len = skb->len; 304 unsigned int len = skb->len;
273 sch->qstats.backlog -= len; 305 struct gred_sched_data *q;
274 sch->qstats.drops++; 306 u16 dp = tc_index_to_dp(skb);
275 q= t->tab[(skb->tc_index&0xf)]; 307
276 if (q) { 308 if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
277 q->backlog -= len; 309 if (net_ratelimit())
278 q->other++; 310 printk(KERN_WARNING "GRED: Unable to relocate "
279 if (!q->backlog && !t->eqp) 311 "VQ 0x%x while dropping, screwing up "
280 PSCHED_GET_TIME(q->qidlestart); 312 "backlog.\n", tc_index_to_dp(skb));
281 } else { 313 } else {
282 D2PRINTK("gred_dequeue: skb has bad tcindex %x\n",skb->tc_index&0xf); 314 q->backlog -= len;
315 q->stats.other++;
316
317 if (!q->backlog && !gred_wred_mode(t))
318 red_start_of_idle_period(&q->parms);
283 } 319 }
284 320
285 kfree_skb(skb); 321 qdisc_drop(skb, sch);
286 return len; 322 return len;
287 } 323 }
288 324
289 q=t->tab[t->def]; 325 if (gred_wred_mode(t) && !red_is_idling(&t->wred_set))
290 if (!q) { 326 red_start_of_idle_period(&t->wred_set);
291 D2PRINTK("no default VQ set: Results might be screwed up\n");
292 return 0;
293 }
294 327
295 PSCHED_GET_TIME(q->qidlestart);
296 return 0; 328 return 0;
297 329
298} 330}
@@ -300,293 +332,241 @@ static unsigned int gred_drop(struct Qdisc* sch)
300static void gred_reset(struct Qdisc* sch) 332static void gred_reset(struct Qdisc* sch)
301{ 333{
302 int i; 334 int i;
303 struct gred_sched_data *q; 335 struct gred_sched *t = qdisc_priv(sch);
304 struct gred_sched *t= qdisc_priv(sch); 336
337 qdisc_reset_queue(sch);
305 338
306 __skb_queue_purge(&sch->q); 339 for (i = 0; i < t->DPs; i++) {
340 struct gred_sched_data *q = t->tab[i];
307 341
308 sch->qstats.backlog = 0; 342 if (!q)
343 continue;
309 344
310 for (i=0;i<t->DPs;i++) { 345 red_restart(&q->parms);
311 q= t->tab[i];
312 if (!q)
313 continue;
314 PSCHED_SET_PASTPERFECT(q->qidlestart);
315 q->qave = 0;
316 q->qcount = -1;
317 q->backlog = 0; 346 q->backlog = 0;
318 q->other=0;
319 q->forced=0;
320 q->pdrop=0;
321 q->early=0;
322 } 347 }
323} 348}
324 349
325static int gred_change(struct Qdisc *sch, struct rtattr *opt) 350static inline void gred_destroy_vq(struct gred_sched_data *q)
351{
352 kfree(q);
353}
354
355static inline int gred_change_table_def(struct Qdisc *sch, struct rtattr *dps)
326{ 356{
327 struct gred_sched *table = qdisc_priv(sch); 357 struct gred_sched *table = qdisc_priv(sch);
328 struct gred_sched_data *q;
329 struct tc_gred_qopt *ctl;
330 struct tc_gred_sopt *sopt; 358 struct tc_gred_sopt *sopt;
331 struct rtattr *tb[TCA_GRED_STAB];
332 struct rtattr *tb2[TCA_GRED_DPS];
333 int i; 359 int i;
334 360
335 if (opt == NULL || rtattr_parse_nested(tb, TCA_GRED_STAB, opt)) 361 if (dps == NULL || RTA_PAYLOAD(dps) < sizeof(*sopt))
336 return -EINVAL; 362 return -EINVAL;
337 363
338 if (tb[TCA_GRED_PARMS-1] == 0 && tb[TCA_GRED_STAB-1] == 0) { 364 sopt = RTA_DATA(dps);
339 rtattr_parse_nested(tb2, TCA_GRED_DPS, opt); 365
366 if (sopt->DPs > MAX_DPs || sopt->DPs == 0 || sopt->def_DP >= sopt->DPs)
367 return -EINVAL;
340 368
341 if (tb2[TCA_GRED_DPS-1] == 0) 369 sch_tree_lock(sch);
342 return -EINVAL; 370 table->DPs = sopt->DPs;
371 table->def = sopt->def_DP;
372 table->red_flags = sopt->flags;
373
374 /*
375 * Every entry point to GRED is synchronized with the above code
376 * and the DP is checked against DPs, i.e. shadowed VQs can no
377 * longer be found so we can unlock right here.
378 */
379 sch_tree_unlock(sch);
380
381 if (sopt->grio) {
382 gred_enable_rio_mode(table);
383 gred_disable_wred_mode(table);
384 if (gred_wred_mode_check(sch))
385 gred_enable_wred_mode(table);
386 } else {
387 gred_disable_rio_mode(table);
388 gred_disable_wred_mode(table);
389 }
343 390
344 sopt = RTA_DATA(tb2[TCA_GRED_DPS-1]); 391 for (i = table->DPs; i < MAX_DPs; i++) {
345 table->DPs=sopt->DPs; 392 if (table->tab[i]) {
346 table->def=sopt->def_DP; 393 printk(KERN_WARNING "GRED: Warning: Destroying "
347 table->grio=sopt->grio; 394 "shadowed VQ 0x%x\n", i);
348 table->initd=0; 395 gred_destroy_vq(table->tab[i]);
349 /* probably need to clear all the table DP entries as well */ 396 table->tab[i] = NULL;
350 return 0; 397 }
351 } 398 }
352 399
400 return 0;
401}
353 402
354 if (!table->DPs || tb[TCA_GRED_PARMS-1] == 0 || tb[TCA_GRED_STAB-1] == 0 || 403static inline int gred_change_vq(struct Qdisc *sch, int dp,
355 RTA_PAYLOAD(tb[TCA_GRED_PARMS-1]) < sizeof(*ctl) || 404 struct tc_gred_qopt *ctl, int prio, u8 *stab)
356 RTA_PAYLOAD(tb[TCA_GRED_STAB-1]) < 256) 405{
357 return -EINVAL; 406 struct gred_sched *table = qdisc_priv(sch);
407 struct gred_sched_data *q;
358 408
359 ctl = RTA_DATA(tb[TCA_GRED_PARMS-1]); 409 if (table->tab[dp] == NULL) {
360 if (ctl->DP > MAX_DPs-1 ) { 410 table->tab[dp] = kmalloc(sizeof(*q), GFP_KERNEL);
361 /* misbehaving is punished! Put in the default drop probability */ 411 if (table->tab[dp] == NULL)
362 DPRINTK("\nGRED: DP %u not in the proper range fixed. New DP "
363 "set to default at %d\n",ctl->DP,table->def);
364 ctl->DP=table->def;
365 }
366
367 if (table->tab[ctl->DP] == NULL) {
368 table->tab[ctl->DP]=kmalloc(sizeof(struct gred_sched_data),
369 GFP_KERNEL);
370 if (NULL == table->tab[ctl->DP])
371 return -ENOMEM; 412 return -ENOMEM;
372 memset(table->tab[ctl->DP], 0, (sizeof(struct gred_sched_data))); 413 memset(table->tab[dp], 0, sizeof(*q));
373 }
374 q= table->tab[ctl->DP];
375
376 if (table->grio) {
377 if (ctl->prio <=0) {
378 if (table->def && table->tab[table->def]) {
379 DPRINTK("\nGRED: DP %u does not have a prio"
380 "setting default to %d\n",ctl->DP,
381 table->tab[table->def]->prio);
382 q->prio=table->tab[table->def]->prio;
383 } else {
384 DPRINTK("\nGRED: DP %u does not have a prio"
385 " setting default to 8\n",ctl->DP);
386 q->prio=8;
387 }
388 } else {
389 q->prio=ctl->prio;
390 }
391 } else {
392 q->prio=8;
393 } 414 }
394 415
395 416 q = table->tab[dp];
396 q->DP=ctl->DP; 417 q->DP = dp;
397 q->Wlog = ctl->Wlog; 418 q->prio = prio;
398 q->Plog = ctl->Plog;
399 q->limit = ctl->limit; 419 q->limit = ctl->limit;
400 q->Scell_log = ctl->Scell_log;
401 q->Rmask = ctl->Plog < 32 ? ((1<<ctl->Plog) - 1) : ~0UL;
402 q->Scell_max = (255<<q->Scell_log);
403 q->qth_min = ctl->qth_min<<ctl->Wlog;
404 q->qth_max = ctl->qth_max<<ctl->Wlog;
405 q->qave=0;
406 q->backlog=0;
407 q->qcount = -1;
408 q->other=0;
409 q->forced=0;
410 q->pdrop=0;
411 q->early=0;
412
413 PSCHED_SET_PASTPERFECT(q->qidlestart);
414 memcpy(q->Stab, RTA_DATA(tb[TCA_GRED_STAB-1]), 256);
415
416 if ( table->initd && table->grio) {
417 /* this looks ugly but it's not in the fast path */
418 for (i=0;i<table->DPs;i++) {
419 if ((!table->tab[i]) || (i==q->DP) )
420 continue;
421 if (table->tab[i]->prio == q->prio ){
422 /* WRED mode detected */
423 table->eqp=1;
424 break;
425 }
426 }
427 }
428 420
429 if (!table->initd) { 421 if (q->backlog == 0)
430 table->initd=1; 422 red_end_of_idle_period(&q->parms);
431 /*
432 the first entry also goes into the default until
433 over-written
434 */
435
436 if (table->tab[table->def] == NULL) {
437 table->tab[table->def]=
438 kmalloc(sizeof(struct gred_sched_data), GFP_KERNEL);
439 if (NULL == table->tab[table->def])
440 return -ENOMEM;
441
442 memset(table->tab[table->def], 0,
443 (sizeof(struct gred_sched_data)));
444 }
445 q= table->tab[table->def];
446 q->DP=table->def;
447 q->Wlog = ctl->Wlog;
448 q->Plog = ctl->Plog;
449 q->limit = ctl->limit;
450 q->Scell_log = ctl->Scell_log;
451 q->Rmask = ctl->Plog < 32 ? ((1<<ctl->Plog) - 1) : ~0UL;
452 q->Scell_max = (255<<q->Scell_log);
453 q->qth_min = ctl->qth_min<<ctl->Wlog;
454 q->qth_max = ctl->qth_max<<ctl->Wlog;
455
456 if (table->grio)
457 q->prio=table->tab[ctl->DP]->prio;
458 else
459 q->prio=8;
460
461 q->qcount = -1;
462 PSCHED_SET_PASTPERFECT(q->qidlestart);
463 memcpy(q->Stab, RTA_DATA(tb[TCA_GRED_STAB-1]), 256);
464 }
465 return 0;
466 423
424 red_set_parms(&q->parms,
425 ctl->qth_min, ctl->qth_max, ctl->Wlog, ctl->Plog,
426 ctl->Scell_log, stab);
427
428 return 0;
467} 429}
468 430
469static int gred_init(struct Qdisc *sch, struct rtattr *opt) 431static int gred_change(struct Qdisc *sch, struct rtattr *opt)
470{ 432{
471 struct gred_sched *table = qdisc_priv(sch); 433 struct gred_sched *table = qdisc_priv(sch);
472 struct tc_gred_sopt *sopt; 434 struct tc_gred_qopt *ctl;
473 struct rtattr *tb[TCA_GRED_STAB]; 435 struct rtattr *tb[TCA_GRED_MAX];
474 struct rtattr *tb2[TCA_GRED_DPS]; 436 int err = -EINVAL, prio = GRED_DEF_PRIO;
437 u8 *stab;
475 438
476 if (opt == NULL || rtattr_parse_nested(tb, TCA_GRED_STAB, opt)) 439 if (opt == NULL || rtattr_parse_nested(tb, TCA_GRED_MAX, opt))
477 return -EINVAL; 440 return -EINVAL;
478 441
479 if (tb[TCA_GRED_PARMS-1] == 0 && tb[TCA_GRED_STAB-1] == 0) { 442 if (tb[TCA_GRED_PARMS-1] == NULL && tb[TCA_GRED_STAB-1] == NULL)
480 rtattr_parse_nested(tb2, TCA_GRED_DPS, opt); 443 return gred_change_table_def(sch, opt);
444
445 if (tb[TCA_GRED_PARMS-1] == NULL ||
446 RTA_PAYLOAD(tb[TCA_GRED_PARMS-1]) < sizeof(*ctl) ||
447 tb[TCA_GRED_STAB-1] == NULL ||
448 RTA_PAYLOAD(tb[TCA_GRED_STAB-1]) < 256)
449 return -EINVAL;
450
451 ctl = RTA_DATA(tb[TCA_GRED_PARMS-1]);
452 stab = RTA_DATA(tb[TCA_GRED_STAB-1]);
453
454 if (ctl->DP >= table->DPs)
455 goto errout;
481 456
482 if (tb2[TCA_GRED_DPS-1] == 0) 457 if (gred_rio_mode(table)) {
483 return -EINVAL; 458 if (ctl->prio == 0) {
459 int def_prio = GRED_DEF_PRIO;
484 460
485 sopt = RTA_DATA(tb2[TCA_GRED_DPS-1]); 461 if (table->tab[table->def])
486 table->DPs=sopt->DPs; 462 def_prio = table->tab[table->def]->prio;
487 table->def=sopt->def_DP; 463
488 table->grio=sopt->grio; 464 printk(KERN_DEBUG "GRED: DP %u does not have a prio "
489 table->initd=0; 465 "setting default to %d\n", ctl->DP, def_prio);
490 return 0; 466
467 prio = def_prio;
468 } else
469 prio = ctl->prio;
470 }
471
472 sch_tree_lock(sch);
473
474 err = gred_change_vq(sch, ctl->DP, ctl, prio, stab);
475 if (err < 0)
476 goto errout_locked;
477
478 if (gred_rio_mode(table)) {
479 gred_disable_wred_mode(table);
480 if (gred_wred_mode_check(sch))
481 gred_enable_wred_mode(table);
491 } 482 }
492 483
493 DPRINTK("\n GRED_INIT error!\n"); 484 err = 0;
494 return -EINVAL; 485
486errout_locked:
487 sch_tree_unlock(sch);
488errout:
489 return err;
495} 490}
496 491
497static int gred_dump(struct Qdisc *sch, struct sk_buff *skb) 492static int gred_init(struct Qdisc *sch, struct rtattr *opt)
498{ 493{
499 unsigned long qave; 494 struct rtattr *tb[TCA_GRED_MAX];
500 struct rtattr *rta;
501 struct tc_gred_qopt *opt = NULL ;
502 struct tc_gred_qopt *dst;
503 struct gred_sched *table = qdisc_priv(sch);
504 struct gred_sched_data *q;
505 int i;
506 unsigned char *b = skb->tail;
507 495
508 rta = (struct rtattr*)b; 496 if (opt == NULL || rtattr_parse_nested(tb, TCA_GRED_MAX, opt))
509 RTA_PUT(skb, TCA_OPTIONS, 0, NULL); 497 return -EINVAL;
510 498
511 opt=kmalloc(sizeof(struct tc_gred_qopt)*MAX_DPs, GFP_KERNEL); 499 if (tb[TCA_GRED_PARMS-1] || tb[TCA_GRED_STAB-1])
500 return -EINVAL;
512 501
513 if (opt == NULL) { 502 return gred_change_table_def(sch, tb[TCA_GRED_DPS-1]);
514 DPRINTK("gred_dump:failed to malloc for %Zd\n", 503}
515 sizeof(struct tc_gred_qopt)*MAX_DPs);
516 goto rtattr_failure;
517 }
518 504
519 memset(opt, 0, (sizeof(struct tc_gred_qopt))*table->DPs); 505static int gred_dump(struct Qdisc *sch, struct sk_buff *skb)
506{
507 struct gred_sched *table = qdisc_priv(sch);
508 struct rtattr *parms, *opts = NULL;
509 int i;
510 struct tc_gred_sopt sopt = {
511 .DPs = table->DPs,
512 .def_DP = table->def,
513 .grio = gred_rio_mode(table),
514 .flags = table->red_flags,
515 };
520 516
521 if (!table->initd) { 517 opts = RTA_NEST(skb, TCA_OPTIONS);
522 DPRINTK("NO GRED Queues setup!\n"); 518 RTA_PUT(skb, TCA_GRED_DPS, sizeof(sopt), &sopt);
523 } 519 parms = RTA_NEST(skb, TCA_GRED_PARMS);
520
521 for (i = 0; i < MAX_DPs; i++) {
522 struct gred_sched_data *q = table->tab[i];
523 struct tc_gred_qopt opt;
524 524
525 for (i=0;i<MAX_DPs;i++) { 525 memset(&opt, 0, sizeof(opt));
526 dst= &opt[i];
527 q= table->tab[i];
528 526
529 if (!q) { 527 if (!q) {
530 /* hack -- fix at some point with proper message 528 /* hack -- fix at some point with proper message
531 This is how we indicate to tc that there is no VQ 529 This is how we indicate to tc that there is no VQ
532 at this DP */ 530 at this DP */
533 531
534 dst->DP=MAX_DPs+i; 532 opt.DP = MAX_DPs + i;
535 continue; 533 goto append_opt;
536 } 534 }
537 535
538 dst->limit=q->limit; 536 opt.limit = q->limit;
539 dst->qth_min=q->qth_min>>q->Wlog; 537 opt.DP = q->DP;
540 dst->qth_max=q->qth_max>>q->Wlog; 538 opt.backlog = q->backlog;
541 dst->DP=q->DP; 539 opt.prio = q->prio;
542 dst->backlog=q->backlog; 540 opt.qth_min = q->parms.qth_min >> q->parms.Wlog;
543 if (q->qave) { 541 opt.qth_max = q->parms.qth_max >> q->parms.Wlog;
544 if (table->eqp && table->grio) { 542 opt.Wlog = q->parms.Wlog;
545 q->qidlestart=table->tab[table->def]->qidlestart; 543 opt.Plog = q->parms.Plog;
546 q->qave=table->tab[table->def]->qave; 544 opt.Scell_log = q->parms.Scell_log;
547 } 545 opt.other = q->stats.other;
548 if (!PSCHED_IS_PASTPERFECT(q->qidlestart)) { 546 opt.early = q->stats.prob_drop;
549 long idle; 547 opt.forced = q->stats.forced_drop;
550 psched_time_t now; 548 opt.pdrop = q->stats.pdrop;
551 PSCHED_GET_TIME(now); 549 opt.packets = q->packetsin;
552 idle = PSCHED_TDIFF_SAFE(now, q->qidlestart, q->Scell_max); 550 opt.bytesin = q->bytesin;
553 qave = q->qave >> q->Stab[(idle>>q->Scell_log)&0xFF]; 551
554 dst->qave = qave >> q->Wlog; 552 if (gred_wred_mode(table)) {
555 553 q->parms.qidlestart =
556 } else { 554 table->tab[table->def]->parms.qidlestart;
557 dst->qave = q->qave >> q->Wlog; 555 q->parms.qavg = table->tab[table->def]->parms.qavg;
558 }
559 } else {
560 dst->qave = 0;
561 } 556 }
562 557
563 558 opt.qave = red_calc_qavg(&q->parms, q->parms.qavg);
564 dst->Wlog = q->Wlog; 559
565 dst->Plog = q->Plog; 560append_opt:
566 dst->Scell_log = q->Scell_log; 561 RTA_APPEND(skb, sizeof(opt), &opt);
567 dst->other = q->other;
568 dst->forced = q->forced;
569 dst->early = q->early;
570 dst->pdrop = q->pdrop;
571 dst->prio = q->prio;
572 dst->packets=q->packetsin;
573 dst->bytesin=q->bytesin;
574 } 562 }
575 563
576 RTA_PUT(skb, TCA_GRED_PARMS, sizeof(struct tc_gred_qopt)*MAX_DPs, opt); 564 RTA_NEST_END(skb, parms);
577 rta->rta_len = skb->tail - b;
578 565
579 kfree(opt); 566 return RTA_NEST_END(skb, opts);
580 return skb->len;
581 567
582rtattr_failure: 568rtattr_failure:
583 if (opt) 569 return RTA_NEST_CANCEL(skb, opts);
584 kfree(opt);
585 DPRINTK("gred_dump: FAILURE!!!!\n");
586
587/* also free the opt struct here */
588 skb_trim(skb, b - skb->data);
589 return -1;
590} 570}
591 571
592static void gred_destroy(struct Qdisc *sch) 572static void gred_destroy(struct Qdisc *sch)
@@ -594,15 +574,13 @@ static void gred_destroy(struct Qdisc *sch)
594 struct gred_sched *table = qdisc_priv(sch); 574 struct gred_sched *table = qdisc_priv(sch);
595 int i; 575 int i;
596 576
597 for (i = 0;i < table->DPs; i++) { 577 for (i = 0; i < table->DPs; i++) {
598 if (table->tab[i]) 578 if (table->tab[i])
599 kfree(table->tab[i]); 579 gred_destroy_vq(table->tab[i]);
600 } 580 }
601} 581}
602 582
603static struct Qdisc_ops gred_qdisc_ops = { 583static struct Qdisc_ops gred_qdisc_ops = {
604 .next = NULL,
605 .cl_ops = NULL,
606 .id = "gred", 584 .id = "gred",
607 .priv_size = sizeof(struct gred_sched), 585 .priv_size = sizeof(struct gred_sched),
608 .enqueue = gred_enqueue, 586 .enqueue = gred_enqueue,
@@ -621,10 +599,13 @@ static int __init gred_module_init(void)
621{ 599{
622 return register_qdisc(&gred_qdisc_ops); 600 return register_qdisc(&gred_qdisc_ops);
623} 601}
624static void __exit gred_module_exit(void) 602
603static void __exit gred_module_exit(void)
625{ 604{
626 unregister_qdisc(&gred_qdisc_ops); 605 unregister_qdisc(&gred_qdisc_ops);
627} 606}
607
628module_init(gred_module_init) 608module_init(gred_module_init)
629module_exit(gred_module_exit) 609module_exit(gred_module_exit)
610
630MODULE_LICENSE("GPL"); 611MODULE_LICENSE("GPL");
diff --git a/net/sched/sch_netem.c b/net/sched/sch_netem.c
index bb9bf8d5003c..cdc8d283791c 100644
--- a/net/sched/sch_netem.c
+++ b/net/sched/sch_netem.c
@@ -25,6 +25,8 @@
25 25
26#include <net/pkt_sched.h> 26#include <net/pkt_sched.h>
27 27
28#define VERSION "1.1"
29
28/* Network Emulation Queuing algorithm. 30/* Network Emulation Queuing algorithm.
29 ==================================== 31 ====================================
30 32
@@ -185,10 +187,13 @@ static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch)
185 || q->counter < q->gap /* inside last reordering gap */ 187 || q->counter < q->gap /* inside last reordering gap */
186 || q->reorder < get_crandom(&q->reorder_cor)) { 188 || q->reorder < get_crandom(&q->reorder_cor)) {
187 psched_time_t now; 189 psched_time_t now;
190 psched_tdiff_t delay;
191
192 delay = tabledist(q->latency, q->jitter,
193 &q->delay_cor, q->delay_dist);
194
188 PSCHED_GET_TIME(now); 195 PSCHED_GET_TIME(now);
189 PSCHED_TADD2(now, tabledist(q->latency, q->jitter, 196 PSCHED_TADD2(now, delay, cb->time_to_send);
190 &q->delay_cor, q->delay_dist),
191 cb->time_to_send);
192 ++q->counter; 197 ++q->counter;
193 ret = q->qdisc->enqueue(skb, q->qdisc); 198 ret = q->qdisc->enqueue(skb, q->qdisc);
194 } else { 199 } else {
@@ -248,24 +253,31 @@ static struct sk_buff *netem_dequeue(struct Qdisc *sch)
248 const struct netem_skb_cb *cb 253 const struct netem_skb_cb *cb
249 = (const struct netem_skb_cb *)skb->cb; 254 = (const struct netem_skb_cb *)skb->cb;
250 psched_time_t now; 255 psched_time_t now;
251 long delay;
252 256
253 /* if more time remaining? */ 257 /* if more time remaining? */
254 PSCHED_GET_TIME(now); 258 PSCHED_GET_TIME(now);
255 delay = PSCHED_US2JIFFIE(PSCHED_TDIFF(cb->time_to_send, now)); 259
256 pr_debug("netem_run: skb=%p delay=%ld\n", skb, delay); 260 if (PSCHED_TLESS(cb->time_to_send, now)) {
257 if (delay <= 0) {
258 pr_debug("netem_dequeue: return skb=%p\n", skb); 261 pr_debug("netem_dequeue: return skb=%p\n", skb);
259 sch->q.qlen--; 262 sch->q.qlen--;
260 sch->flags &= ~TCQ_F_THROTTLED; 263 sch->flags &= ~TCQ_F_THROTTLED;
261 return skb; 264 return skb;
262 } 265 } else {
266 psched_tdiff_t delay = PSCHED_TDIFF(cb->time_to_send, now);
267
268 if (q->qdisc->ops->requeue(skb, q->qdisc) != NET_XMIT_SUCCESS) {
269 sch->qstats.drops++;
263 270
264 mod_timer(&q->timer, jiffies + delay); 271 /* After this qlen is confused */
265 sch->flags |= TCQ_F_THROTTLED; 272 printk(KERN_ERR "netem: queue discpline %s could not requeue\n",
273 q->qdisc->ops->id);
266 274
267 if (q->qdisc->ops->requeue(skb, q->qdisc) != 0) 275 sch->q.qlen--;
268 sch->qstats.drops++; 276 }
277
278 mod_timer(&q->timer, jiffies + PSCHED_US2JIFFIE(delay));
279 sch->flags |= TCQ_F_THROTTLED;
280 }
269 } 281 }
270 282
271 return NULL; 283 return NULL;
@@ -290,11 +302,16 @@ static void netem_reset(struct Qdisc *sch)
290 del_timer_sync(&q->timer); 302 del_timer_sync(&q->timer);
291} 303}
292 304
305/* Pass size change message down to embedded FIFO */
293static int set_fifo_limit(struct Qdisc *q, int limit) 306static int set_fifo_limit(struct Qdisc *q, int limit)
294{ 307{
295 struct rtattr *rta; 308 struct rtattr *rta;
296 int ret = -ENOMEM; 309 int ret = -ENOMEM;
297 310
311 /* Hack to avoid sending change message to non-FIFO */
312 if (strncmp(q->ops->id + 1, "fifo", 4) != 0)
313 return 0;
314
298 rta = kmalloc(RTA_LENGTH(sizeof(struct tc_fifo_qopt)), GFP_KERNEL); 315 rta = kmalloc(RTA_LENGTH(sizeof(struct tc_fifo_qopt)), GFP_KERNEL);
299 if (rta) { 316 if (rta) {
300 rta->rta_type = RTM_NEWQDISC; 317 rta->rta_type = RTM_NEWQDISC;
@@ -426,6 +443,84 @@ static int netem_change(struct Qdisc *sch, struct rtattr *opt)
426 return 0; 443 return 0;
427} 444}
428 445
446/*
447 * Special case version of FIFO queue for use by netem.
448 * It queues in order based on timestamps in skb's
449 */
450struct fifo_sched_data {
451 u32 limit;
452};
453
454static int tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch)
455{
456 struct fifo_sched_data *q = qdisc_priv(sch);
457 struct sk_buff_head *list = &sch->q;
458 const struct netem_skb_cb *ncb
459 = (const struct netem_skb_cb *)nskb->cb;
460 struct sk_buff *skb;
461
462 if (likely(skb_queue_len(list) < q->limit)) {
463 skb_queue_reverse_walk(list, skb) {
464 const struct netem_skb_cb *cb
465 = (const struct netem_skb_cb *)skb->cb;
466
467 if (PSCHED_TLESS(cb->time_to_send, ncb->time_to_send))
468 break;
469 }
470
471 __skb_queue_after(list, skb, nskb);
472
473 sch->qstats.backlog += nskb->len;
474 sch->bstats.bytes += nskb->len;
475 sch->bstats.packets++;
476
477 return NET_XMIT_SUCCESS;
478 }
479
480 return qdisc_drop(nskb, sch);
481}
482
483static int tfifo_init(struct Qdisc *sch, struct rtattr *opt)
484{
485 struct fifo_sched_data *q = qdisc_priv(sch);
486
487 if (opt) {
488 struct tc_fifo_qopt *ctl = RTA_DATA(opt);
489 if (RTA_PAYLOAD(opt) < sizeof(*ctl))
490 return -EINVAL;
491
492 q->limit = ctl->limit;
493 } else
494 q->limit = max_t(u32, sch->dev->tx_queue_len, 1);
495
496 return 0;
497}
498
499static int tfifo_dump(struct Qdisc *sch, struct sk_buff *skb)
500{
501 struct fifo_sched_data *q = qdisc_priv(sch);
502 struct tc_fifo_qopt opt = { .limit = q->limit };
503
504 RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
505 return skb->len;
506
507rtattr_failure:
508 return -1;
509}
510
511static struct Qdisc_ops tfifo_qdisc_ops = {
512 .id = "tfifo",
513 .priv_size = sizeof(struct fifo_sched_data),
514 .enqueue = tfifo_enqueue,
515 .dequeue = qdisc_dequeue_head,
516 .requeue = qdisc_requeue,
517 .drop = qdisc_queue_drop,
518 .init = tfifo_init,
519 .reset = qdisc_reset_queue,
520 .change = tfifo_init,
521 .dump = tfifo_dump,
522};
523
429static int netem_init(struct Qdisc *sch, struct rtattr *opt) 524static int netem_init(struct Qdisc *sch, struct rtattr *opt)
430{ 525{
431 struct netem_sched_data *q = qdisc_priv(sch); 526 struct netem_sched_data *q = qdisc_priv(sch);
@@ -438,7 +533,7 @@ static int netem_init(struct Qdisc *sch, struct rtattr *opt)
438 q->timer.function = netem_watchdog; 533 q->timer.function = netem_watchdog;
439 q->timer.data = (unsigned long) sch; 534 q->timer.data = (unsigned long) sch;
440 535
441 q->qdisc = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops); 536 q->qdisc = qdisc_create_dflt(sch->dev, &tfifo_qdisc_ops);
442 if (!q->qdisc) { 537 if (!q->qdisc) {
443 pr_debug("netem: qdisc create failed\n"); 538 pr_debug("netem: qdisc create failed\n");
444 return -ENOMEM; 539 return -ENOMEM;
@@ -601,6 +696,7 @@ static struct Qdisc_ops netem_qdisc_ops = {
601 696
602static int __init netem_module_init(void) 697static int __init netem_module_init(void)
603{ 698{
699 pr_info("netem: version " VERSION "\n");
604 return register_qdisc(&netem_qdisc_ops); 700 return register_qdisc(&netem_qdisc_ops);
605} 701}
606static void __exit netem_module_exit(void) 702static void __exit netem_module_exit(void)
diff --git a/net/sched/sch_red.c b/net/sched/sch_red.c
index 7845d045eec4..dccfa44c2d71 100644
--- a/net/sched/sch_red.c
+++ b/net/sched/sch_red.c
@@ -9,76 +9,23 @@
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 * 10 *
11 * Changes: 11 * Changes:
12 * J Hadi Salim <hadi@nortel.com> 980914: computation fixes 12 * J Hadi Salim 980914: computation fixes
13 * Alexey Makarenko <makar@phoenix.kharkov.ua> 990814: qave on idle link was calculated incorrectly. 13 * Alexey Makarenko <makar@phoenix.kharkov.ua> 990814: qave on idle link was calculated incorrectly.
14 * J Hadi Salim <hadi@nortelnetworks.com> 980816: ECN support 14 * J Hadi Salim 980816: ECN support
15 */ 15 */
16 16
17#include <linux/config.h> 17#include <linux/config.h>
18#include <linux/module.h> 18#include <linux/module.h>
19#include <asm/uaccess.h>
20#include <asm/system.h>
21#include <linux/bitops.h>
22#include <linux/types.h> 19#include <linux/types.h>
23#include <linux/kernel.h> 20#include <linux/kernel.h>
24#include <linux/sched.h>
25#include <linux/string.h>
26#include <linux/mm.h>
27#include <linux/socket.h>
28#include <linux/sockios.h>
29#include <linux/in.h>
30#include <linux/errno.h>
31#include <linux/interrupt.h>
32#include <linux/if_ether.h>
33#include <linux/inet.h>
34#include <linux/netdevice.h> 21#include <linux/netdevice.h>
35#include <linux/etherdevice.h>
36#include <linux/notifier.h>
37#include <net/ip.h>
38#include <net/route.h>
39#include <linux/skbuff.h> 22#include <linux/skbuff.h>
40#include <net/sock.h>
41#include <net/pkt_sched.h> 23#include <net/pkt_sched.h>
42#include <net/inet_ecn.h> 24#include <net/inet_ecn.h>
43#include <net/dsfield.h> 25#include <net/red.h>
44 26
45 27
46/* Random Early Detection (RED) algorithm. 28/* Parameters, settable by user:
47 =======================================
48
49 Source: Sally Floyd and Van Jacobson, "Random Early Detection Gateways
50 for Congestion Avoidance", 1993, IEEE/ACM Transactions on Networking.
51
52 This file codes a "divisionless" version of RED algorithm
53 as written down in Fig.17 of the paper.
54
55Short description.
56------------------
57
58 When a new packet arrives we calculate the average queue length:
59
60 avg = (1-W)*avg + W*current_queue_len,
61
62 W is the filter time constant (chosen as 2^(-Wlog)), it controls
63 the inertia of the algorithm. To allow larger bursts, W should be
64 decreased.
65
66 if (avg > th_max) -> packet marked (dropped).
67 if (avg < th_min) -> packet passes.
68 if (th_min < avg < th_max) we calculate probability:
69
70 Pb = max_P * (avg - th_min)/(th_max-th_min)
71
72 and mark (drop) packet with this probability.
73 Pb changes from 0 (at avg==th_min) to max_P (avg==th_max).
74 max_P should be small (not 1), usually 0.01..0.02 is good value.
75
76 max_P is chosen as a number, so that max_P/(th_max-th_min)
77 is a negative power of two in order arithmetics to contain
78 only shifts.
79
80
81 Parameters, settable by user:
82 ----------------------------- 29 -----------------------------
83 30
84 limit - bytes (must be > qth_max + burst) 31 limit - bytes (must be > qth_max + burst)
@@ -89,243 +36,93 @@ Short description.
89 arbitrarily high (well, less than ram size) 36 arbitrarily high (well, less than ram size)
90 Really, this limit will never be reached 37 Really, this limit will never be reached
91 if RED works correctly. 38 if RED works correctly.
92
93 qth_min - bytes (should be < qth_max/2)
94 qth_max - bytes (should be at least 2*qth_min and less limit)
95 Wlog - bits (<32) log(1/W).
96 Plog - bits (<32)
97
98 Plog is related to max_P by formula:
99
100 max_P = (qth_max-qth_min)/2^Plog;
101
102 F.e. if qth_max=128K and qth_min=32K, then Plog=22
103 corresponds to max_P=0.02
104
105 Scell_log
106 Stab
107
108 Lookup table for log((1-W)^(t/t_ave).
109
110
111NOTES:
112
113Upper bound on W.
114-----------------
115
116 If you want to allow bursts of L packets of size S,
117 you should choose W:
118
119 L + 1 - th_min/S < (1-(1-W)^L)/W
120
121 th_min/S = 32 th_min/S = 4
122
123 log(W) L
124 -1 33
125 -2 35
126 -3 39
127 -4 46
128 -5 57
129 -6 75
130 -7 101
131 -8 135
132 -9 190
133 etc.
134 */ 39 */
135 40
136struct red_sched_data 41struct red_sched_data
137{ 42{
138/* Parameters */ 43 u32 limit; /* HARD maximal queue length */
139 u32 limit; /* HARD maximal queue length */ 44 unsigned char flags;
140 u32 qth_min; /* Min average length threshold: A scaled */ 45 struct red_parms parms;
141 u32 qth_max; /* Max average length threshold: A scaled */ 46 struct red_stats stats;
142 u32 Rmask;
143 u32 Scell_max;
144 unsigned char flags;
145 char Wlog; /* log(W) */
146 char Plog; /* random number bits */
147 char Scell_log;
148 u8 Stab[256];
149
150/* Variables */
151 unsigned long qave; /* Average queue length: A scaled */
152 int qcount; /* Packets since last random number generation */
153 u32 qR; /* Cached random number */
154
155 psched_time_t qidlestart; /* Start of idle period */
156 struct tc_red_xstats st;
157}; 47};
158 48
159static int red_ecn_mark(struct sk_buff *skb) 49static inline int red_use_ecn(struct red_sched_data *q)
160{ 50{
161 if (skb->nh.raw + 20 > skb->tail) 51 return q->flags & TC_RED_ECN;
162 return 0;
163
164 switch (skb->protocol) {
165 case __constant_htons(ETH_P_IP):
166 if (INET_ECN_is_not_ect(skb->nh.iph->tos))
167 return 0;
168 IP_ECN_set_ce(skb->nh.iph);
169 return 1;
170 case __constant_htons(ETH_P_IPV6):
171 if (INET_ECN_is_not_ect(ipv6_get_dsfield(skb->nh.ipv6h)))
172 return 0;
173 IP6_ECN_set_ce(skb->nh.ipv6h);
174 return 1;
175 default:
176 return 0;
177 }
178} 52}
179 53
180static int 54static inline int red_use_harddrop(struct red_sched_data *q)
181red_enqueue(struct sk_buff *skb, struct Qdisc* sch) 55{
56 return q->flags & TC_RED_HARDDROP;
57}
58
59static int red_enqueue(struct sk_buff *skb, struct Qdisc* sch)
182{ 60{
183 struct red_sched_data *q = qdisc_priv(sch); 61 struct red_sched_data *q = qdisc_priv(sch);
184 62
185 psched_time_t now; 63 q->parms.qavg = red_calc_qavg(&q->parms, sch->qstats.backlog);
186 64
187 if (!PSCHED_IS_PASTPERFECT(q->qidlestart)) { 65 if (red_is_idling(&q->parms))
188 long us_idle; 66 red_end_of_idle_period(&q->parms);
189 int shift;
190 67
191 PSCHED_GET_TIME(now); 68 switch (red_action(&q->parms, q->parms.qavg)) {
192 us_idle = PSCHED_TDIFF_SAFE(now, q->qidlestart, q->Scell_max); 69 case RED_DONT_MARK:
193 PSCHED_SET_PASTPERFECT(q->qidlestart); 70 break;
194 71
195/* 72 case RED_PROB_MARK:
196 The problem: ideally, average length queue recalcultion should 73 sch->qstats.overlimits++;
197 be done over constant clock intervals. This is too expensive, so that 74 if (!red_use_ecn(q) || !INET_ECN_set_ce(skb)) {
198 the calculation is driven by outgoing packets. 75 q->stats.prob_drop++;
199 When the queue is idle we have to model this clock by hand. 76 goto congestion_drop;
200 77 }
201 SF+VJ proposed to "generate" m = idletime/(average_pkt_size/bandwidth)
202 dummy packets as a burst after idle time, i.e.
203
204 q->qave *= (1-W)^m
205
206 This is an apparently overcomplicated solution (f.e. we have to precompute
207 a table to make this calculation in reasonable time)
208 I believe that a simpler model may be used here,
209 but it is field for experiments.
210*/
211 shift = q->Stab[us_idle>>q->Scell_log];
212
213 if (shift) {
214 q->qave >>= shift;
215 } else {
216 /* Approximate initial part of exponent
217 with linear function:
218 (1-W)^m ~= 1-mW + ...
219
220 Seems, it is the best solution to
221 problem of too coarce exponent tabulation.
222 */
223
224 us_idle = (q->qave * us_idle)>>q->Scell_log;
225 if (us_idle < q->qave/2)
226 q->qave -= us_idle;
227 else
228 q->qave >>= 1;
229 }
230 } else {
231 q->qave += sch->qstats.backlog - (q->qave >> q->Wlog);
232 /* NOTE:
233 q->qave is fixed point number with point at Wlog.
234 The formulae above is equvalent to floating point
235 version:
236
237 qave = qave*(1-W) + sch->qstats.backlog*W;
238 --ANK (980924)
239 */
240 }
241 78
242 if (q->qave < q->qth_min) { 79 q->stats.prob_mark++;
243 q->qcount = -1; 80 break;
244enqueue: 81
245 if (sch->qstats.backlog + skb->len <= q->limit) { 82 case RED_HARD_MARK:
246 __skb_queue_tail(&sch->q, skb); 83 sch->qstats.overlimits++;
247 sch->qstats.backlog += skb->len; 84 if (red_use_harddrop(q) || !red_use_ecn(q) ||
248 sch->bstats.bytes += skb->len; 85 !INET_ECN_set_ce(skb)) {
249 sch->bstats.packets++; 86 q->stats.forced_drop++;
250 return NET_XMIT_SUCCESS; 87 goto congestion_drop;
251 } else { 88 }
252 q->st.pdrop++;
253 }
254 kfree_skb(skb);
255 sch->qstats.drops++;
256 return NET_XMIT_DROP;
257 }
258 if (q->qave >= q->qth_max) {
259 q->qcount = -1;
260 sch->qstats.overlimits++;
261mark:
262 if (!(q->flags&TC_RED_ECN) || !red_ecn_mark(skb)) {
263 q->st.early++;
264 goto drop;
265 }
266 q->st.marked++;
267 goto enqueue;
268 }
269 89
270 if (++q->qcount) { 90 q->stats.forced_mark++;
271 /* The formula used below causes questions. 91 break;
272
273 OK. qR is random number in the interval 0..Rmask
274 i.e. 0..(2^Plog). If we used floating point
275 arithmetics, it would be: (2^Plog)*rnd_num,
276 where rnd_num is less 1.
277
278 Taking into account, that qave have fixed
279 point at Wlog, and Plog is related to max_P by
280 max_P = (qth_max-qth_min)/2^Plog; two lines
281 below have the following floating point equivalent:
282
283 max_P*(qave - qth_min)/(qth_max-qth_min) < rnd/qcount
284
285 Any questions? --ANK (980924)
286 */
287 if (((q->qave - q->qth_min)>>q->Wlog)*q->qcount < q->qR)
288 goto enqueue;
289 q->qcount = 0;
290 q->qR = net_random()&q->Rmask;
291 sch->qstats.overlimits++;
292 goto mark;
293 } 92 }
294 q->qR = net_random()&q->Rmask;
295 goto enqueue;
296 93
297drop: 94 if (sch->qstats.backlog + skb->len <= q->limit)
298 kfree_skb(skb); 95 return qdisc_enqueue_tail(skb, sch);
299 sch->qstats.drops++; 96
97 q->stats.pdrop++;
98 return qdisc_drop(skb, sch);
99
100congestion_drop:
101 qdisc_drop(skb, sch);
300 return NET_XMIT_CN; 102 return NET_XMIT_CN;
301} 103}
302 104
303static int 105static int red_requeue(struct sk_buff *skb, struct Qdisc* sch)
304red_requeue(struct sk_buff *skb, struct Qdisc* sch)
305{ 106{
306 struct red_sched_data *q = qdisc_priv(sch); 107 struct red_sched_data *q = qdisc_priv(sch);
307 108
308 PSCHED_SET_PASTPERFECT(q->qidlestart); 109 if (red_is_idling(&q->parms))
110 red_end_of_idle_period(&q->parms);
309 111
310 __skb_queue_head(&sch->q, skb); 112 return qdisc_requeue(skb, sch);
311 sch->qstats.backlog += skb->len;
312 sch->qstats.requeues++;
313 return 0;
314} 113}
315 114
316static struct sk_buff * 115static struct sk_buff * red_dequeue(struct Qdisc* sch)
317red_dequeue(struct Qdisc* sch)
318{ 116{
319 struct sk_buff *skb; 117 struct sk_buff *skb;
320 struct red_sched_data *q = qdisc_priv(sch); 118 struct red_sched_data *q = qdisc_priv(sch);
321 119
322 skb = __skb_dequeue(&sch->q); 120 skb = qdisc_dequeue_head(sch);
323 if (skb) { 121
324 sch->qstats.backlog -= skb->len; 122 if (skb == NULL && !red_is_idling(&q->parms))
325 return skb; 123 red_start_of_idle_period(&q->parms);
326 } 124
327 PSCHED_GET_TIME(q->qidlestart); 125 return skb;
328 return NULL;
329} 126}
330 127
331static unsigned int red_drop(struct Qdisc* sch) 128static unsigned int red_drop(struct Qdisc* sch)
@@ -333,16 +130,17 @@ static unsigned int red_drop(struct Qdisc* sch)
333 struct sk_buff *skb; 130 struct sk_buff *skb;
334 struct red_sched_data *q = qdisc_priv(sch); 131 struct red_sched_data *q = qdisc_priv(sch);
335 132
336 skb = __skb_dequeue_tail(&sch->q); 133 skb = qdisc_dequeue_tail(sch);
337 if (skb) { 134 if (skb) {
338 unsigned int len = skb->len; 135 unsigned int len = skb->len;
339 sch->qstats.backlog -= len; 136 q->stats.other++;
340 sch->qstats.drops++; 137 qdisc_drop(skb, sch);
341 q->st.other++;
342 kfree_skb(skb);
343 return len; 138 return len;
344 } 139 }
345 PSCHED_GET_TIME(q->qidlestart); 140
141 if (!red_is_idling(&q->parms))
142 red_start_of_idle_period(&q->parms);
143
346 return 0; 144 return 0;
347} 145}
348 146
@@ -350,43 +148,38 @@ static void red_reset(struct Qdisc* sch)
350{ 148{
351 struct red_sched_data *q = qdisc_priv(sch); 149 struct red_sched_data *q = qdisc_priv(sch);
352 150
353 __skb_queue_purge(&sch->q); 151 qdisc_reset_queue(sch);
354 sch->qstats.backlog = 0; 152 red_restart(&q->parms);
355 PSCHED_SET_PASTPERFECT(q->qidlestart);
356 q->qave = 0;
357 q->qcount = -1;
358} 153}
359 154
360static int red_change(struct Qdisc *sch, struct rtattr *opt) 155static int red_change(struct Qdisc *sch, struct rtattr *opt)
361{ 156{
362 struct red_sched_data *q = qdisc_priv(sch); 157 struct red_sched_data *q = qdisc_priv(sch);
363 struct rtattr *tb[TCA_RED_STAB]; 158 struct rtattr *tb[TCA_RED_MAX];
364 struct tc_red_qopt *ctl; 159 struct tc_red_qopt *ctl;
365 160
366 if (opt == NULL || 161 if (opt == NULL || rtattr_parse_nested(tb, TCA_RED_MAX, opt))
367 rtattr_parse_nested(tb, TCA_RED_STAB, opt) || 162 return -EINVAL;
368 tb[TCA_RED_PARMS-1] == 0 || tb[TCA_RED_STAB-1] == 0 || 163
164 if (tb[TCA_RED_PARMS-1] == NULL ||
369 RTA_PAYLOAD(tb[TCA_RED_PARMS-1]) < sizeof(*ctl) || 165 RTA_PAYLOAD(tb[TCA_RED_PARMS-1]) < sizeof(*ctl) ||
370 RTA_PAYLOAD(tb[TCA_RED_STAB-1]) < 256) 166 tb[TCA_RED_STAB-1] == NULL ||
167 RTA_PAYLOAD(tb[TCA_RED_STAB-1]) < RED_STAB_SIZE)
371 return -EINVAL; 168 return -EINVAL;
372 169
373 ctl = RTA_DATA(tb[TCA_RED_PARMS-1]); 170 ctl = RTA_DATA(tb[TCA_RED_PARMS-1]);
374 171
375 sch_tree_lock(sch); 172 sch_tree_lock(sch);
376 q->flags = ctl->flags; 173 q->flags = ctl->flags;
377 q->Wlog = ctl->Wlog;
378 q->Plog = ctl->Plog;
379 q->Rmask = ctl->Plog < 32 ? ((1<<ctl->Plog) - 1) : ~0UL;
380 q->Scell_log = ctl->Scell_log;
381 q->Scell_max = (255<<q->Scell_log);
382 q->qth_min = ctl->qth_min<<ctl->Wlog;
383 q->qth_max = ctl->qth_max<<ctl->Wlog;
384 q->limit = ctl->limit; 174 q->limit = ctl->limit;
385 memcpy(q->Stab, RTA_DATA(tb[TCA_RED_STAB-1]), 256);
386 175
387 q->qcount = -1; 176 red_set_parms(&q->parms, ctl->qth_min, ctl->qth_max, ctl->Wlog,
177 ctl->Plog, ctl->Scell_log,
178 RTA_DATA(tb[TCA_RED_STAB-1]));
179
388 if (skb_queue_empty(&sch->q)) 180 if (skb_queue_empty(&sch->q))
389 PSCHED_SET_PASTPERFECT(q->qidlestart); 181 red_end_of_idle_period(&q->parms);
182
390 sch_tree_unlock(sch); 183 sch_tree_unlock(sch);
391 return 0; 184 return 0;
392} 185}
@@ -399,39 +192,39 @@ static int red_init(struct Qdisc* sch, struct rtattr *opt)
399static int red_dump(struct Qdisc *sch, struct sk_buff *skb) 192static int red_dump(struct Qdisc *sch, struct sk_buff *skb)
400{ 193{
401 struct red_sched_data *q = qdisc_priv(sch); 194 struct red_sched_data *q = qdisc_priv(sch);
402 unsigned char *b = skb->tail; 195 struct rtattr *opts = NULL;
403 struct rtattr *rta; 196 struct tc_red_qopt opt = {
404 struct tc_red_qopt opt; 197 .limit = q->limit,
405 198 .flags = q->flags,
406 rta = (struct rtattr*)b; 199 .qth_min = q->parms.qth_min >> q->parms.Wlog,
407 RTA_PUT(skb, TCA_OPTIONS, 0, NULL); 200 .qth_max = q->parms.qth_max >> q->parms.Wlog,
408 opt.limit = q->limit; 201 .Wlog = q->parms.Wlog,
409 opt.qth_min = q->qth_min>>q->Wlog; 202 .Plog = q->parms.Plog,
410 opt.qth_max = q->qth_max>>q->Wlog; 203 .Scell_log = q->parms.Scell_log,
411 opt.Wlog = q->Wlog; 204 };
412 opt.Plog = q->Plog; 205
413 opt.Scell_log = q->Scell_log; 206 opts = RTA_NEST(skb, TCA_OPTIONS);
414 opt.flags = q->flags;
415 RTA_PUT(skb, TCA_RED_PARMS, sizeof(opt), &opt); 207 RTA_PUT(skb, TCA_RED_PARMS, sizeof(opt), &opt);
416 rta->rta_len = skb->tail - b; 208 return RTA_NEST_END(skb, opts);
417
418 return skb->len;
419 209
420rtattr_failure: 210rtattr_failure:
421 skb_trim(skb, b - skb->data); 211 return RTA_NEST_CANCEL(skb, opts);
422 return -1;
423} 212}
424 213
425static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d) 214static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
426{ 215{
427 struct red_sched_data *q = qdisc_priv(sch); 216 struct red_sched_data *q = qdisc_priv(sch);
428 217 struct tc_red_xstats st = {
429 return gnet_stats_copy_app(d, &q->st, sizeof(q->st)); 218 .early = q->stats.prob_drop + q->stats.forced_drop,
219 .pdrop = q->stats.pdrop,
220 .other = q->stats.other,
221 .marked = q->stats.prob_mark + q->stats.forced_mark,
222 };
223
224 return gnet_stats_copy_app(d, &st, sizeof(st));
430} 225}
431 226
432static struct Qdisc_ops red_qdisc_ops = { 227static struct Qdisc_ops red_qdisc_ops = {
433 .next = NULL,
434 .cl_ops = NULL,
435 .id = "red", 228 .id = "red",
436 .priv_size = sizeof(struct red_sched_data), 229 .priv_size = sizeof(struct red_sched_data),
437 .enqueue = red_enqueue, 230 .enqueue = red_enqueue,
@@ -450,10 +243,13 @@ static int __init red_module_init(void)
450{ 243{
451 return register_qdisc(&red_qdisc_ops); 244 return register_qdisc(&red_qdisc_ops);
452} 245}
453static void __exit red_module_exit(void) 246
247static void __exit red_module_exit(void)
454{ 248{
455 unregister_qdisc(&red_qdisc_ops); 249 unregister_qdisc(&red_qdisc_ops);
456} 250}
251
457module_init(red_module_init) 252module_init(red_module_init)
458module_exit(red_module_exit) 253module_exit(red_module_exit)
254
459MODULE_LICENSE("GPL"); 255MODULE_LICENSE("GPL");