Merge master.kernel.org:/pub/scm/linux/kernel/git/acme/net-2.6

include/linux/pkt_sched.h

@@ -93,6 +93,7 @@ struct tc_fifo_qopt
 /* PRIO section */
 
 #define TCQ_PRIO_BANDS	16
+#define TCQ_MIN_PRIO_BANDS 2
 
 struct tc_prio_qopt
 {
@@ -169,6 +170,7 @@ struct tc_red_qopt
 	unsigned char   Scell_log;	/* cell size for idle damping */
 	unsigned char	flags;
 #define TC_RED_ECN	1
+#define TC_RED_HARDDROP	2
 };
 
 struct tc_red_xstats
@@ -194,15 +196,11 @@ enum
 
 #define TCA_GRED_MAX (__TCA_GRED_MAX - 1)
 
-#define TCA_SET_OFF TCA_GRED_PARMS
 struct tc_gred_qopt
 {
-       __u32           limit;          /* HARD maximal queue length (bytes)    
-*/
-       __u32           qth_min;        /* Min average length threshold (bytes) 
-*/
-       __u32           qth_max;        /* Max average length threshold (bytes) 
-*/
+	__u32		limit;        /* HARD maximal queue length (bytes)    */
+	__u32		qth_min;      /* Min average length threshold (bytes) */
+	__u32		qth_max;      /* Max average length threshold (bytes) */
 	__u32		DP;           /* upto 2^32 DPs */
 	__u32		backlog;
 	__u32		qave;
@@ -210,22 +208,22 @@ struct tc_gred_qopt
 	__u32		early;
 	__u32		other;
 	__u32		pdrop;
-
-       unsigned char   Wlog;           /* log(W)               */
-       unsigned char   Plog;           /* log(P_max/(qth_max-qth_min)) */
-       unsigned char   Scell_log;      /* cell size for idle damping */
+	__u8		Wlog;         /* log(W)               */
+	__u8		Plog;         /* log(P_max/(qth_max-qth_min)) */
+	__u8		Scell_log;    /* cell size for idle damping */
 	__u8		prio;         /* prio of this VQ */
 	__u32		packets;
 	__u32		bytesin;
 };
+
 /* gred setup */
 struct tc_gred_sopt
 {
 	__u32		DPs;
 	__u32		def_DP;
 	__u8		grio;
-       __u8		pad1;
-       __u16		pad2;
+	__u8		flags;
+	__u16		pad1;
 };
 
 /* HTB section */

include/linux/skbuff.h

@@ -603,29 +603,46 @@ static inline void skb_queue_head_init(struct sk_buff_head *list)
  */
 
 /**
- *	__skb_queue_head - queue a buffer at the list head
+ *	__skb_queue_after - queue a buffer at the list head
  *	@list: list to use
+ *	@prev: place after this buffer
  *	@newsk: buffer to queue
  *
- *	Queue a buffer at the start of a list. This function takes no locks
+ *	Queue a buffer int the middle of a list. This function takes no locks
  *	and you must therefore hold required locks before calling it.
  *
  *	A buffer cannot be placed on two lists at the same time.
  */
-extern void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
-static inline void __skb_queue_head(struct sk_buff_head *list,
+static inline void __skb_queue_after(struct sk_buff_head *list,
+				     struct sk_buff *prev,
 				     struct sk_buff *newsk)
 {
-	struct sk_buff *prev, *next;
-
+	struct sk_buff *next;
 	list->qlen++;
-	prev = (struct sk_buff *)list;
+
 	next = prev->next;
 	newsk->next = next;
 	newsk->prev = prev;
 	next->prev  = prev->next = newsk;
 }
 
+/**
+ *	__skb_queue_head - queue a buffer at the list head
+ *	@list: list to use
+ *	@newsk: buffer to queue
+ *
+ *	Queue a buffer at the start of a list. This function takes no locks
+ *	and you must therefore hold required locks before calling it.
+ *
+ *	A buffer cannot be placed on two lists at the same time.
+ */
+extern void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
+static inline void __skb_queue_head(struct sk_buff_head *list,
+				    struct sk_buff *newsk)
+{
+	__skb_queue_after(list, (struct sk_buff *)list, newsk);
+}
+
 /**
  *	__skb_queue_tail - queue a buffer at the list tail
  *	@list: list to use
@@ -1203,6 +1220,11 @@ static inline void kunmap_skb_frag(void *vaddr)
 		     prefetch(skb->next), (skb != (struct sk_buff *)(queue));	\
 		     skb = skb->next)
 
+#define skb_queue_reverse_walk(queue, skb) \
+		for (skb = (queue)->prev;					\
+		     prefetch(skb->prev), (skb != (struct sk_buff *)(queue));	\
+		     skb = skb->prev)
+
 
 extern struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
 					 int noblock, int *err);

include/net/inet_ecn.h

@@ -2,6 +2,7 @@
 #define _INET_ECN_H_
 
 #include <linux/ip.h>
+#include <linux/skbuff.h>
 #include <net/dsfield.h>
 
 enum {
@@ -48,7 +49,7 @@ static inline __u8 INET_ECN_encapsulate(__u8 outer, __u8 inner)
 		(label) |= __constant_htons(INET_ECN_ECT_0 << 4);	\
     } while (0)
 
-static inline void IP_ECN_set_ce(struct iphdr *iph)
+static inline int IP_ECN_set_ce(struct iphdr *iph)
 {
 	u32 check = iph->check;
 	u32 ecn = (iph->tos + 1) & INET_ECN_MASK;
@@ -61,7 +62,7 @@ static inline void IP_ECN_set_ce(struct iphdr *iph)
 	 * INET_ECN_CE      => 00
 	 */
 	if (!(ecn & 2))
-		return;
+		return !ecn;
 
 	/*
 	 * The following gives us:
@@ -72,6 +73,7 @@ static inline void IP_ECN_set_ce(struct iphdr *iph)
 
 	iph->check = check + (check>=0xFFFF);
 	iph->tos |= INET_ECN_CE;
+	return 1;
 }
 
 static inline void IP_ECN_clear(struct iphdr *iph)
@@ -87,11 +89,12 @@ static inline void ipv4_copy_dscp(struct iphdr *outer, struct iphdr *inner)
 
 struct ipv6hdr;
 
-static inline void IP6_ECN_set_ce(struct ipv6hdr *iph)
+static inline int IP6_ECN_set_ce(struct ipv6hdr *iph)
 {
 	if (INET_ECN_is_not_ect(ipv6_get_dsfield(iph)))
-		return;
+		return 0;
 	*(u32*)iph |= htonl(INET_ECN_CE << 20);
+	return 1;
 }
 
 static inline void IP6_ECN_clear(struct ipv6hdr *iph)
@@ -105,4 +108,21 @@ static inline void ipv6_copy_dscp(struct ipv6hdr *outer, struct ipv6hdr *inner)
 	ipv6_change_dsfield(inner, INET_ECN_MASK, dscp);
 }
 
+static inline int INET_ECN_set_ce(struct sk_buff *skb)
+{
+	switch (skb->protocol) {
+	case __constant_htons(ETH_P_IP):
+		if (skb->nh.raw + sizeof(struct iphdr) <= skb->tail)
+			return IP_ECN_set_ce(skb->nh.iph);
+		break;
+
+	case __constant_htons(ETH_P_IPV6):
+		if (skb->nh.raw + sizeof(struct ipv6hdr) <= skb->tail)
+			return IP6_ECN_set_ce(skb->nh.ipv6h);
+		break;
+	}
+
+	return 0;
+}
+
 #endif

include/net/inet_hashtables.h

@@ -125,9 +125,7 @@ struct inet_hashinfo {
 	rwlock_t			lhash_lock ____cacheline_aligned;
 	atomic_t			lhash_users;
 	wait_queue_head_t		lhash_wait;
-	spinlock_t			portalloc_lock;
 	kmem_cache_t			*bind_bucket_cachep;
-	int				port_rover;
 };
 
 static inline unsigned int inet_ehashfn(const __u32 laddr, const __u16 lport,

include/net/red.h

+#ifndef __NET_SCHED_RED_H
+#define __NET_SCHED_RED_H
+
+#include <linux/config.h>
+#include <linux/types.h>
+#include <net/pkt_sched.h>
+#include <net/inet_ecn.h>
+#include <net/dsfield.h>
+
+/*	Random Early Detection (RED) algorithm.
+	=======================================
+
+	Source: Sally Floyd and Van Jacobson, "Random Early Detection Gateways
+	for Congestion Avoidance", 1993, IEEE/ACM Transactions on Networking.
+
+	This file codes a "divisionless" version of RED algorithm
+	as written down in Fig.17 of the paper.
+
+	Short description.
+	------------------
+
+	When a new packet arrives we calculate the average queue length:
+
+	avg = (1-W)*avg + W*current_queue_len,
+
+	W is the filter time constant (chosen as 2^(-Wlog)), it controls
+	the inertia of the algorithm. To allow larger bursts, W should be
+	decreased.
+
+	if (avg > th_max) -> packet marked (dropped).
+	if (avg < th_min) -> packet passes.
+	if (th_min < avg < th_max) we calculate probability:
+
+	Pb = max_P * (avg - th_min)/(th_max-th_min)
+
+	and mark (drop) packet with this probability.
+	Pb changes from 0 (at avg==th_min) to max_P (avg==th_max).
+	max_P should be small (not 1), usually 0.01..0.02 is good value.
+
+	max_P is chosen as a number, so that max_P/(th_max-th_min)
+	is a negative power of two in order arithmetics to contain
+	only shifts.
+
+
+	Parameters, settable by user:
+	-----------------------------
+
+	qth_min		- bytes (should be < qth_max/2)
+	qth_max		- bytes (should be at least 2*qth_min and less limit)
+	Wlog	       	- bits (<32) log(1/W).
+	Plog	       	- bits (<32)
+
+	Plog is related to max_P by formula:
+
+	max_P = (qth_max-qth_min)/2^Plog;
+
+	F.e. if qth_max=128K and qth_min=32K, then Plog=22
+	corresponds to max_P=0.02
+
+	Scell_log
+	Stab
+
+	Lookup table for log((1-W)^(t/t_ave).
+
+
+	NOTES:
+
+	Upper bound on W.
+	-----------------
+
+	If you want to allow bursts of L packets of size S,
+	you should choose W:
+
+	L + 1 - th_min/S < (1-(1-W)^L)/W
+
+	th_min/S = 32         th_min/S = 4
+
+	log(W)	L
+	-1	33
+	-2	35
+	-3	39
+	-4	46
+	-5	57
+	-6	75
+	-7	101
+	-8	135
+	-9	190
+	etc.
+ */
+
+#define RED_STAB_SIZE	256
+#define RED_STAB_MASK	(RED_STAB_SIZE - 1)
+
+struct red_stats
+{
+	u32		prob_drop;	/* Early probability drops */
+	u32		prob_mark;	/* Early probability marks */
+	u32		forced_drop;	/* Forced drops, qavg > max_thresh */
+	u32		forced_mark;	/* Forced marks, qavg > max_thresh */
+	u32		pdrop;          /* Drops due to queue limits */
+	u32		other;          /* Drops due to drop() calls */
+	u32		backlog;
+};
+
+struct red_parms
+{
+	/* Parameters */
+	u32		qth_min;	/* Min avg length threshold: A scaled */
+	u32		qth_max;	/* Max avg length threshold: A scaled */
+	u32		Scell_max;
+	u32		Rmask;		/* Cached random mask, see red_rmask */
+	u8		Scell_log;
+	u8		Wlog;		/* log(W)		*/
+	u8		Plog;		/* random number bits	*/
+	u8		Stab[RED_STAB_SIZE];
+
+	/* Variables */
+	int		qcount;		/* Number of packets since last random
+					   number generation */
+	u32		qR;		/* Cached random number */
+
+	unsigned long	qavg;		/* Average queue length: A scaled */
+	psched_time_t	qidlestart;	/* Start of current idle period */
+};
+
+static inline u32 red_rmask(u8 Plog)
+{
+	return Plog < 32 ? ((1 << Plog) - 1) : ~0UL;
+}
+
+static inline void red_set_parms(struct red_parms *p,
+				 u32 qth_min, u32 qth_max, u8 Wlog, u8 Plog,
+				 u8 Scell_log, u8 *stab)
+{
+	/* Reset average queue length, the value is strictly bound
+	 * to the parameters below, reseting hurts a bit but leaving
+	 * it might result in an unreasonable qavg for a while. --TGR
+	 */
+	p->qavg		= 0;
+
+	p->qcount	= -1;
+	p->qth_min	= qth_min << Wlog;
+	p->qth_max	= qth_max << Wlog;
+	p->Wlog		= Wlog;
+	p->Plog		= Plog;
+	p->Rmask	= red_rmask(Plog);
+	p->Scell_log	= Scell_log;
+	p->Scell_max	= (255 << Scell_log);
+
+	memcpy(p->Stab, stab, sizeof(p->Stab));
+}
+
+static inline int red_is_idling(struct red_parms *p)
+{
+	return !PSCHED_IS_PASTPERFECT(p->qidlestart);
+}
+
+static inline void red_start_of_idle_period(struct red_parms *p)
+{
+	PSCHED_GET_TIME(p->qidlestart);
+}
+
+static inline void red_end_of_idle_period(struct red_parms *p)
+{
+	PSCHED_SET_PASTPERFECT(p->qidlestart);
+}
+
+static inline void red_restart(struct red_parms *p)
+{
+	red_end_of_idle_period(p);
+	p->qavg = 0;
+	p->qcount = -1;
+}
+
+static inline unsigned long red_calc_qavg_from_idle_time(struct red_parms *p)
+{
+	psched_time_t now;
+	long us_idle;
+	int  shift;
+
+	PSCHED_GET_TIME(now);
+	us_idle = PSCHED_TDIFF_SAFE(now, p->qidlestart, p->Scell_max);
+
+	/*
+	 * The problem: ideally, average length queue recalcultion should
+	 * be done over constant clock intervals. This is too expensive, so
+	 * that the calculation is driven by outgoing packets.
+	 * When the queue is idle we have to model this clock by hand.
+	 *
+	 * SF+VJ proposed to "generate":
+	 *
+	 *	m = idletime / (average_pkt_size / bandwidth)
+	 *
+	 * dummy packets as a burst after idle time, i.e.
+	 *
+	 * 	p->qavg *= (1-W)^m
+	 *
+	 * This is an apparently overcomplicated solution (f.e. we have to
+	 * precompute a table to make this calculation in reasonable time)
+	 * I believe that a simpler model may be used here,
+	 * but it is field for experiments.
+	 */
+
+	shift = p->Stab[(us_idle >> p->Scell_log) & RED_STAB_MASK];
+
+	if (shift)
+		return p->qavg >> shift;
+	else {
+		/* Approximate initial part of exponent with linear function:
+		 *
+		 * 	(1-W)^m ~= 1-mW + ...
+		 *
+		 * Seems, it is the best solution to
+		 * problem of too coarse exponent tabulation.
+		 */
+		us_idle = (p->qavg * us_idle) >> p->Scell_log;
+
+		if (us_idle < (p->qavg >> 1))
+			return p->qavg - us_idle;
+		else
+			return p->qavg >> 1;
+	}
+}
+
+static inline unsigned long red_calc_qavg_no_idle_time(struct red_parms *p,
+						       unsigned int backlog)
+{
+	/*
+	 * NOTE: p->qavg is fixed point number with point at Wlog.
+	 * The formula below is equvalent to floating point
+	 * version:
+	 *
+	 * 	qavg = qavg*(1-W) + backlog*W;
+	 *
+	 * --ANK (980924)
+	 */
+	return p->qavg + (backlog - (p->qavg >> p->Wlog));
+}
+
+static inline unsigned long red_calc_qavg(struct red_parms *p,
+					  unsigned int backlog)
+{
+	if (!red_is_idling(p))
+		return red_calc_qavg_no_idle_time(p, backlog);
+	else
+		return red_calc_qavg_from_idle_time(p);
+}
+
+static inline u32 red_random(struct red_parms *p)
+{
+	return net_random() & p->Rmask;
+}
+
+static inline int red_mark_probability(struct red_parms *p, unsigned long qavg)
+{
+	/* The formula used below causes questions.
+
+	   OK. qR is random number in the interval 0..Rmask
+	   i.e. 0..(2^Plog). If we used floating point
+	   arithmetics, it would be: (2^Plog)*rnd_num,
+	   where rnd_num is less 1.
+
+	   Taking into account, that qavg have fixed
+	   point at Wlog, and Plog is related to max_P by
+	   max_P = (qth_max-qth_min)/2^Plog; two lines
+	   below have the following floating point equivalent:
+
+	   max_P*(qavg - qth_min)/(qth_max-qth_min) < rnd/qcount
+
+	   Any questions? --ANK (980924)
+	 */
+	return !(((qavg - p->qth_min) >> p->Wlog) * p->qcount < p->qR);
+}
+
+enum {
+	RED_BELOW_MIN_THRESH,
+	RED_BETWEEN_TRESH,
+	RED_ABOVE_MAX_TRESH,
+};
+
+static inline int red_cmp_thresh(struct red_parms *p, unsigned long qavg)
+{
+	if (qavg < p->qth_min)
+		return RED_BELOW_MIN_THRESH;
+	else if (qavg >= p->qth_max)
+		return RED_ABOVE_MAX_TRESH;
+	else
+		return RED_BETWEEN_TRESH;
+}
+
+enum {
+	RED_DONT_MARK,
+	RED_PROB_MARK,
+	RED_HARD_MARK,
+};
+
+static inline int red_action(struct red_parms *p, unsigned long qavg)
+{
+	switch (red_cmp_thresh(p, qavg)) {
+		case RED_BELOW_MIN_THRESH:
+			p->qcount = -1;
+			return RED_DONT_MARK;
+
+		case RED_BETWEEN_TRESH:
+			if (++p->qcount) {
+				if (red_mark_probability(p, qavg)) {
+					p->qcount = 0;
+					p->qR = red_random(p);
+					return RED_PROB_MARK;
+				}
+			} else
+				p->qR = red_random(p);
+
+			return RED_DONT_MARK;
+
+		case RED_ABOVE_MAX_TRESH:
+			p->qcount = -1;
+			return RED_HARD_MARK;
+	}
+
+	BUG();
+	return RED_DONT_MARK;
+}
+
+#endif

net/core/stream.c

@@ -52,8 +52,9 @@ int sk_stream_wait_connect(struct sock *sk, long *timeo_p)
 {
 	struct task_struct *tsk = current;
 	DEFINE_WAIT(wait);
+	int done;
 
-	while (1) {
+	do {
 		if (sk->sk_err)
 			return sock_error(sk);
 		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)
 
 		prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 		sk->sk_write_pending++;
-		if (sk_wait_event(sk, timeo_p,
+		done = sk_wait_event(sk, timeo_p,
 				     !((1 << sk->sk_state) & 
-				    ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))))
-			break;
+				       ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)));
 		finish_wait(sk->sk_sleep, &wait);
 		sk->sk_write_pending--;
-	}
+	} while (!done);
 	return 0;
 }
 

net/dccp/ipv4.c

@@ -31,8 +31,6 @@ struct inet_hashinfo __cacheline_aligned dccp_hashinfo = {
 	.lhash_lock	= RW_LOCK_UNLOCKED,
 	.lhash_users	= ATOMIC_INIT(0),
 	.lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(dccp_hashinfo.lhash_wait),
-	.portalloc_lock	= SPIN_LOCK_UNLOCKED,
-	.port_rover	= 1024 - 1,
 };
 
 EXPORT_SYMBOL_GPL(dccp_hashinfo);
@@ -125,36 +123,15 @@ static int dccp_v4_hash_connect(struct sock *sk)
 	int ret;
 
  	if (snum == 0) {
- 		int rover;
  		int low = sysctl_local_port_range[0];
  		int high = sysctl_local_port_range[1];
  		int remaining = (high - low) + 1;
+ 		int rover = net_random() % (high - low) + low;
 		struct hlist_node *node;
  		struct inet_timewait_sock *tw = NULL;
 
  		local_bh_disable();
-
- 		/* TODO. Actually it is not so bad idea to remove
- 		 * dccp_hashinfo.portalloc_lock before next submission to
-		 * Linus.
- 		 * As soon as we touch this place at all it is time to think.
- 		 *
- 		 * Now it protects single _advisory_ variable
-		 * dccp_hashinfo.port_rover, hence it is mostly useless.
- 		 * Code will work nicely if we just delete it, but
- 		 * I am afraid in contented case it will work not better or
- 		 * even worse: another cpu just will hit the same bucket
- 		 * and spin there.
- 		 * So some cpu salt could remove both contention and
- 		 * memory pingpong. Any ideas how to do this in a nice way?
- 		 */
- 		spin_lock(&dccp_hashinfo.portalloc_lock);
- 		rover = dccp_hashinfo.port_rover;
-
  		do {
- 			rover++;
- 			if ((rover < low) || (rover > high))
- 				rover = low;
  			head = &dccp_hashinfo.bhash[inet_bhashfn(rover,
 						    dccp_hashinfo.bhash_size)];
  			spin_lock(&head->lock);
@@ -187,9 +164,9 @@ static int dccp_v4_hash_connect(struct sock *sk)
 
  		next_port:
  			spin_unlock(&head->lock);
+ 			if (++rover > high)
+ 				rover = low;
  		} while (--remaining > 0);
- 		dccp_hashinfo.port_rover = rover;
- 		spin_unlock(&dccp_hashinfo.portalloc_lock);
 
  		local_bh_enable();
 
@@ -197,9 +174,6 @@ static int dccp_v4_hash_connect(struct sock *sk)
 
 ok:
  		/* All locks still held and bhs disabled */
- 		dccp_hashinfo.port_rover = rover;
- 		spin_unlock(&dccp_hashinfo.portalloc_lock);
-
  		inet_bind_hash(sk, tb, rover);
 		if (sk_unhashed(sk)) {
  			inet_sk(sk)->sport = htons(rover);

net/ipv4/inet_connection_sock.c

@@ -78,17 +78,9 @@ int inet_csk_get_port(struct inet_hashinfo *hashinfo,
 		int low = sysctl_local_port_range[0];
 		int high = sysctl_local_port_range[1];
 		int remaining = (high - low) + 1;
-		int rover;
+		int rover = net_random() % (high - low) + low;
 
-		spin_lock(&hashinfo->portalloc_lock);
-		if (hashinfo->port_rover < low)
-			rover = low;
-		else
-			rover = hashinfo->port_rover;
 		do {
-			rover++;
-			if (rover > high)
-				rover = low;
 			head = &hashinfo->bhash[inet_bhashfn(rover, hashinfo->bhash_size)];
 			spin_lock(&head->lock);
 			inet_bind_bucket_for_each(tb, node, &head->chain)
@@ -97,9 +89,9 @@ int inet_csk_get_port(struct inet_hashinfo *hashinfo,
 			break;
 		next:
 			spin_unlock(&head->lock);
+			if (++rover > high)
+				rover = low;
 		} while (--remaining > 0);
-		hashinfo->port_rover = rover;
-		spin_unlock(&hashinfo->portalloc_lock);
 
 		/* Exhausted local port range during search?  It is not
 		 * possible for us to be holding one of the bind hash

net/ipv4/netfilter/ip_conntrack_helper_pptp.c

@@ -270,14 +270,10 @@ exp_gre(struct ip_conntrack *master,
 	exp_orig->expectfn = pptp_expectfn;
 	exp_orig->flags = 0;
 
-	exp_orig->dir = IP_CT_DIR_ORIGINAL;
-
 	/* both expectations are identical apart from tuple */
 	memcpy(exp_reply, exp_orig, sizeof(*exp_reply));
 	memcpy(&exp_reply->tuple, &exp_tuples[1], sizeof(exp_reply->tuple));
 
-	exp_reply->dir = !exp_orig->dir;
-
 	if (ip_nat_pptp_hook_exp_gre)
 		ret = ip_nat_pptp_hook_exp_gre(exp_orig, exp_reply);
 	else {

net/ipv4/netfilter/ip_conntrack_netlink.c

@@ -815,7 +815,7 @@ ctnetlink_get_conntrack(struct sock *ctnl, struct sk_buff *skb,
 				  IPCTNL_MSG_CT_NEW, 1, ct);
 	ip_conntrack_put(ct);
 	if (err <= 0)
-		goto out;
+		goto free;
 
 	err = netlink_unicast(ctnl, skb2, NETLINK_CB(skb).pid, MSG_DONTWAIT);
 	if (err < 0)
@@ -824,9 +824,9 @@ ctnetlink_get_conntrack(struct sock *ctnl, struct sk_buff *skb,
 	DEBUGP("leaving\n");
 	return 0;
 
-out:
-	if (skb2)
+free:
 	kfree_skb(skb2);
+out:
 	return -1;
 }
 
@@ -1322,21 +1322,16 @@ ctnetlink_get_expect(struct sock *ctnl, struct sk_buff *skb,
 				      nlh->nlmsg_seq, IPCTNL_MSG_EXP_NEW,
 				      1, exp);
 	if (err <= 0)
-		goto out;
+		goto free;
 
 	ip_conntrack_expect_put(exp);
 
-	err = netlink_unicast(ctnl, skb2, NETLINK_CB(skb).pid, MSG_DONTWAIT);
-	if (err < 0)
-		goto free;
-
-	return err;
+	return netlink_unicast(ctnl, skb2, NETLINK_CB(skb).pid, MSG_DONTWAIT);
 
-out:
-	ip_conntrack_expect_put(exp);
 free:
-	if (skb2)
 	kfree_skb(skb2);
+out:
+	ip_conntrack_expect_put(exp);
 	return err;
 }
 

net/ipv4/netfilter/ip_nat_core.c

@@ -66,10 +66,8 @@ ip_nat_proto_find_get(u_int8_t protonum)
 	 * removed until we've grabbed the reference */
 	preempt_disable();
 	p = __ip_nat_proto_find(protonum);
-	if (p) {
 	if (!try_module_get(p->me))
 		p = &ip_nat_unknown_protocol;
-	}
 	preempt_enable();
 
 	return p;

net/ipv4/netfilter/ip_nat_helper_pptp.c

@@ -216,6 +216,7 @@ pptp_exp_gre(struct ip_conntrack_expect *expect_orig,
 	expect_orig->saved_proto.gre.key = htons(nat_pptp_info->pac_call_id);
 	expect_orig->tuple.src.u.gre.key = htons(nat_pptp_info->pns_call_id);
 	expect_orig->tuple.dst.u.gre.key = htons(ct_pptp_info->pac_call_id);
+	expect_orig->dir = IP_CT_DIR_ORIGINAL;
 	inv_t.src.ip = reply_t->src.ip;
 	inv_t.dst.ip = reply_t->dst.ip;
 	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,
 	expect_reply->saved_proto.gre.key = htons(nat_pptp_info->pns_call_id);
 	expect_reply->tuple.src.u.gre.key = htons(nat_pptp_info->pac_call_id);
 	expect_reply->tuple.dst.u.gre.key = htons(ct_pptp_info->pns_call_id);
+	expect_reply->dir = IP_CT_DIR_REPLY;
 	inv_t.src.ip = orig_t->src.ip;
 	inv_t.dst.ip = orig_t->dst.ip;
 	inv_t.src.u.gre.key = htons(nat_pptp_info->pns_call_id);

net/ipv4/netfilter/ip_nat_proto_gre.c

@@ -139,8 +139,8 @@ gre_manip_pkt(struct sk_buff **pskb,
 			break;
 		case GRE_VERSION_PPTP:
 			DEBUGP("call_id -> 0x%04x\n", 
-				ntohl(tuple->dst.u.gre.key));
-			pgreh->call_id = htons(ntohl(tuple->dst.u.gre.key));
+				ntohs(tuple->dst.u.gre.key));
+			pgreh->call_id = tuple->dst.u.gre.key;
 			break;
 		default:
 			DEBUGP("can't nat unknown GRE version\n");

net/ipv4/netfilter/ip_nat_proto_unknown.c

@@ -62,7 +62,7 @@ unknown_print_range(char *buffer, const struct ip_nat_range *range)
 
 struct ip_nat_protocol ip_nat_unknown_protocol = {
 	.name			= "unknown",
-	.me			= THIS_MODULE,
+	/* .me isn't set: getting a ref to this cannot fail. */
 	.manip_pkt		= unknown_manip_pkt,
 	.in_range		= unknown_in_range,
 	.unique_tuple		= unknown_unique_tuple,

net/ipv4/netfilter/ipt_CONNMARK.c

@@ -109,6 +109,7 @@ static struct ipt_target ipt_connmark_reg = {
 
 static int __init init(void)
 {
+	need_ip_conntrack();
 	return ipt_register_target(&ipt_connmark_reg);
 }
 

net/ipv4/tcp.c

@@ -2112,7 +2112,6 @@ void __init tcp_init(void)
 		sysctl_tcp_max_orphans >>= (3 - order);
 		sysctl_max_syn_backlog = 128;
 	}
-	tcp_hashinfo.port_rover = sysctl_local_port_range[0] - 1;
 
 	sysctl_tcp_mem[0] =  768 << order;
 	sysctl_tcp_mem[1] = 1024 << order;

net/ipv4/tcp_ipv4.c

@@ -93,8 +93,6 @@ struct inet_hashinfo __cacheline_aligned tcp_hashinfo = {
 	.lhash_lock	= RW_LOCK_UNLOCKED,
 	.lhash_users	= ATOMIC_INIT(0),
 	.lhash_wait	= __WAIT_QUEUE_HEAD_INITIALIZER(tcp_hashinfo.lhash_wait),
-	.portalloc_lock	= SPIN_LOCK_UNLOCKED,
-	.port_rover	= 1024 - 1,
 };
 
 static int tcp_v4_get_port(struct sock *sk, unsigned short snum)

net/ipv6/tcp_ipv6.c

@@ -114,16 +114,9 @@ static int tcp_v6_get_port(struct sock *sk, unsigned short snum)
 		int low = sysctl_local_port_range[0];
 		int high = sysctl_local_port_range[1];
 		int remaining = (high - low) + 1;
-		int rover;
+		int rover = net_random() % (high - low) + low;
 
-		spin_lock(&tcp_hashinfo.portalloc_lock);
-		if (tcp_hashinfo.port_rover < low)
-			rover = low;
-		else
-			rover = tcp_hashinfo.port_rover;
-		do {	rover++;
-			if (rover > high)
-				rover = low;
+		do {
 			head = &tcp_hashinfo.bhash[inet_bhashfn(rover, tcp_hashinfo.bhash_size)];
 			spin_lock(&head->lock);
 			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)
 			break;
 		next:
 			spin_unlock(&head->lock);
+			if (++rover > high)
+				rover = low;
 		} while (--remaining > 0);
-		tcp_hashinfo.port_rover = rover;
-		spin_unlock(&tcp_hashinfo.portalloc_lock);
 
 		/* Exhausted local port range during search?  It is not
 		 * possible for us to be holding one of the bind hash

net/netfilter/nf_queue.c

@@ -117,7 +117,7 @@ int nf_queue(struct sk_buff **skb,
 
 	/* QUEUE == DROP if noone is waiting, to be safe. */
 	read_lock(&queue_handler_lock);
-	if (!queue_handler[pf]->outfn) {
+	if (!queue_handler[pf] || !queue_handler[pf]->outfn) {
 		read_unlock(&queue_handler_lock);
 		kfree_skb(*skb);
 		return 1;

net/netfilter/nfnetlink_log.c

@@ -146,11 +146,10 @@ instance_create(u_int16_t group_num, int pid)
 		goto out_unlock;
 	}
 
-	inst = kmalloc(sizeof(*inst), GFP_ATOMIC);
+	inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 	if (!inst)
 		goto out_unlock;
 
-	memset(inst, 0, sizeof(*inst));
 	INIT_HLIST_NODE(&inst->hlist);
 	inst->lock = SPIN_LOCK_UNLOCKED;
 	/* needs to be two, since we _put() after creation */
@@ -962,10 +961,9 @@ static int nful_open(struct inode *inode, struct file *file)
 	struct iter_state *is;
 	int ret;
 
-	is = kmalloc(sizeof(*is), GFP_KERNEL);
+	is = kzalloc(sizeof(*is), GFP_KERNEL);
 	if (!is)
 		return -ENOMEM;
-	memset(is, 0, sizeof(*is));
 	ret = seq_open(file, &nful_seq_ops);
 	if (ret < 0)
 		goto out_free;

net/netfilter/nfnetlink_queue.c

@@ -136,11 +136,10 @@ instance_create(u_int16_t queue_num, int pid)
 		goto out_unlock;
 	}
 
-	inst = kmalloc(sizeof(*inst), GFP_ATOMIC);
+	inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 	if (!inst)
 		goto out_unlock;
 
-	memset(inst, 0, sizeof(*inst));
 	inst->queue_num = queue_num;
 	inst->peer_pid = pid;
 	inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
@@ -1036,10 +1035,9 @@ static int nfqnl_open(struct inode *inode, struct file *file)
 	struct iter_state *is;
 	int ret;
 
-	is = kmalloc(sizeof(*is), GFP_KERNEL);
+	is = kzalloc(sizeof(*is), GFP_KERNEL);
 	if (!is)
 		return -ENOMEM;
-	memset(is, 0, sizeof(*is));
 	ret = seq_open(file, &nfqnl_seq_ops);
 	if (ret < 0)
 		goto out_free;

net/sched/sch_gred.c

@@ -15,247 +15,281 @@
  *		         from Ren Liu
  *		       - More error checks
  *
- *
- *
- *  For all the glorious comments look at Alexey's sch_red.c
+ *  For all the glorious comments look at include/net/red.h
  */
 
 #include <linux/config.h>
 #include <linux/module.h>
-#include <asm/uaccess.h>
-#include <asm/system.h>
-#include <linux/bitops.h>
 #include <linux/types.h>
 #include <linux/kernel.h>
-#include <linux/sched.h>
-#include <linux/string.h>
-#include <linux/mm.h>
-#include <linux/socket.h>
-#include <linux/sockios.h>
-#include <linux/in.h>
-#include <linux/errno.h>
-#include <linux/interrupt.h>
-#include <linux/if_ether.h>
-#include <linux/inet.h>
 #include <linux/netdevice.h>
-#include <linux/etherdevice.h>
-#include <linux/notifier.h>
-#include <net/ip.h>
-#include <net/route.h>
 #include <linux/skbuff.h>
-#include <net/sock.h>
 #include <net/pkt_sched.h>
+#include <net/red.h>
 
-#if 1 /* control */
-#define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
-#else
-#define DPRINTK(format,args...)
-#endif
-
-#if 0 /* data */
-#define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args)
-#else
-#define D2PRINTK(format,args...)
-#endif
+#define GRED_DEF_PRIO (MAX_DPs / 2)
+#define GRED_VQ_MASK (MAX_DPs - 1)
 
 struct gred_sched_data;
 struct gred_sched;
 
 struct gred_sched_data
 {
-/* Parameters */
 	u32		limit;		/* HARD maximal queue length	*/
-	u32		qth_min;	/* Min average length threshold: A scaled */
-	u32		qth_max;	/* Max average length threshold: A scaled */
 	u32      	DP;		/* the drop pramaters */
-	char		Wlog;		/* log(W)		*/
-	char		Plog;		/* random number bits	*/
-	u32		Scell_max;
-	u32		Rmask;
 	u32		bytesin;	/* bytes seen on virtualQ so far*/
 	u32		packetsin;	/* packets seen on virtualQ so far*/
 	u32		backlog;	/* bytes on the virtualQ */
-	u32		forced;	/* packets dropped for exceeding limits */
-	u32		early;	/* packets dropped as a warning */
-	u32		other;	/* packets dropped by invoking drop() */
-	u32		pdrop;	/* packets dropped because we exceeded physical queue limits */
-	char		Scell_log;
-	u8		Stab[256];
 	u8		prio;		/* the prio of this vq */
 
-/* Variables */
-	unsigned long	qave;		/* Average queue length: A scaled */
-	int		qcount;		/* Packets since last random number generation */
-	u32		qR;		/* Cached random number */
+	struct red_parms parms;
+	struct red_stats stats;
+};
 
-	psched_time_t	qidlestart;	/* Start of idle period	*/
+enum {
+	GRED_WRED_MODE = 1,
+	GRED_RIO_MODE,
 };
 
 struct gred_sched
 {
 	struct gred_sched_data *tab[MAX_DPs];
+	unsigned long	flags;
+	u32		red_flags;
 	u32 		DPs;
 	u32 		def;
-	u8 		initd; 
-	u8 		grio; 
-	u8 		eqp; 
+	struct red_parms wred_set;
 };
 
-static int
-gred_enqueue(struct sk_buff *skb, struct Qdisc* sch)
+static inline int gred_wred_mode(struct gred_sched *table)
 {
-	psched_time_t now;
-	struct gred_sched_data *q=NULL;
-	struct gred_sched *t= qdisc_priv(sch);
-	unsigned long	qave=0;	
-	int i=0;
+	return test_bit(GRED_WRED_MODE, &table->flags);
+}
+
+static inline void gred_enable_wred_mode(struct gred_sched *table)
+{
+	__set_bit(GRED_WRED_MODE, &table->flags);
+}
+
+static inline void gred_disable_wred_mode(struct gred_sched *table)
+{
+	__clear_bit(GRED_WRED_MODE, &table->flags);
+}
 
-	if (!t->initd && skb_queue_len(&sch->q) < (sch->dev->tx_queue_len ? : 1)) {
-		D2PRINTK("NO GRED Queues setup yet! Enqueued anyway\n");
-		goto do_enqueue;
+static inline int gred_rio_mode(struct gred_sched *table)
+{
+	return test_bit(GRED_RIO_MODE, &table->flags);
+}
+
+static inline void gred_enable_rio_mode(struct gred_sched *table)
+{
+	__set_bit(GRED_RIO_MODE, &table->flags);
+}
+
+static inline void gred_disable_rio_mode(struct gred_sched *table)
+{
+	__clear_bit(GRED_RIO_MODE, &table->flags);
+}
+
+static inline int gred_wred_mode_check(struct Qdisc *sch)
+{
+	struct gred_sched *table = qdisc_priv(sch);
+	int i;
+
+	/* Really ugly O(n^2) but shouldn't be necessary too frequent. */
+	for (i = 0; i < table->DPs; i++) {
+		struct gred_sched_data *q = table->tab[i];
+		int n;
+
+		if (q == NULL)
+			continue;
+
+		for (n = 0; n < table->DPs; n++)
+			if (table->tab[n] && table->tab[n] != q &&
+			    table->tab[n]->prio == q->prio)
+				return 1;
 	}
 
+	return 0;
+}
+
+static inline unsigned int gred_backlog(struct gred_sched *table,
+					struct gred_sched_data *q,
+					struct Qdisc *sch)
+{
+	if (gred_wred_mode(table))
+		return sch->qstats.backlog;
+	else
+		return q->backlog;
+}
+
+static inline u16 tc_index_to_dp(struct sk_buff *skb)
+{
+	return skb->tc_index & GRED_VQ_MASK;
+}
+
+static inline void gred_load_wred_set(struct gred_sched *table,
+				      struct gred_sched_data *q)
+{
+	q->parms.qavg = table->wred_set.qavg;
+	q->parms.qidlestart = table->wred_set.qidlestart;
+}
+
+static inline void gred_store_wred_set(struct gred_sched *table,
+				       struct gred_sched_data *q)
+{
+	table->wred_set.qavg = q->parms.qavg;
+}
+
+static inline int gred_use_ecn(struct gred_sched *t)
+{
+	return t->red_flags & TC_RED_ECN;
+}
+
+static inline int gred_use_harddrop(struct gred_sched *t)
+{
+	return t->red_flags & TC_RED_HARDDROP;
+}
+
+static int gred_enqueue(struct sk_buff *skb, struct Qdisc* sch)
+{
+	struct gred_sched_data *q=NULL;
+	struct gred_sched *t= qdisc_priv(sch);
+	unsigned long qavg = 0;
+	u16 dp = tc_index_to_dp(skb);
+
+	if (dp >= t->DPs  || (q = t->tab[dp]) == NULL) {
+		dp = t->def;
 
-	if ( ((skb->tc_index&0xf) > (t->DPs -1)) || !(q=t->tab[skb->tc_index&0xf])) {
-		printk("GRED: setting to default (%d)\n ",t->def);
-		if (!(q=t->tab[t->def])) {
-			DPRINTK("GRED: setting to default FAILED! dropping!! "
-			    "(%d)\n ", t->def);
+		if ((q = t->tab[dp]) == NULL) {
+			/* Pass through packets not assigned to a DP
+			 * if no default DP has been configured. This
+			 * allows for DP flows to be left untouched.
+			 */
+			if (skb_queue_len(&sch->q) < sch->dev->tx_queue_len)
+				return qdisc_enqueue_tail(skb, sch);
+			else
 				goto drop;
 		}
+
 		/* fix tc_index? --could be controvesial but needed for
 		   requeueing */
-		skb->tc_index=(skb->tc_index&0xfffffff0) | t->def;
+		skb->tc_index = (skb->tc_index & ~GRED_VQ_MASK) | dp;
 	}
 
-	D2PRINTK("gred_enqueue virtualQ 0x%x classid %x backlog %d "
-	    "general backlog %d\n",skb->tc_index&0xf,sch->handle,q->backlog,
-	    sch->qstats.backlog);
-	/* sum up all the qaves of prios <= to ours to get the new qave*/
-	if (!t->eqp && t->grio) {
-		for (i=0;i<t->DPs;i++) {
-			if ((!t->tab[i]) || (i==q->DP))	
-				continue; 
+	/* sum up all the qaves of prios <= to ours to get the new qave */
+	if (!gred_wred_mode(t) && gred_rio_mode(t)) {
+		int i;
 
-			if ((t->tab[i]->prio < q->prio) && (PSCHED_IS_PASTPERFECT(t->tab[i]->qidlestart)))
-				qave +=t->tab[i]->qave;
+		for (i = 0; i < t->DPs; i++) {
+			if (t->tab[i] && t->tab[i]->prio < q->prio &&
+			    !red_is_idling(&t->tab[i]->parms))
+				qavg +=t->tab[i]->parms.qavg;
 		}
 
 	}
 
 	q->packetsin++;
-	q->bytesin+=skb->len;
+	q->bytesin += skb->len;
 
-	if (t->eqp && t->grio) {
-		qave=0;
-		q->qave=t->tab[t->def]->qave;
-		q->qidlestart=t->tab[t->def]->qidlestart;
-	}
+	if (gred_wred_mode(t))
+		gred_load_wred_set(t, q);
 
-	if (!PSCHED_IS_PASTPERFECT(q->qidlestart)) {
-		long us_idle;
-		PSCHED_GET_TIME(now);
-		us_idle = PSCHED_TDIFF_SAFE(now, q->qidlestart, q->Scell_max);
-		PSCHED_SET_PASTPERFECT(q->qidlestart);
+	q->parms.qavg = red_calc_qavg(&q->parms, gred_backlog(t, q, sch));
 
-		q->qave >>= q->Stab[(us_idle>>q->Scell_log)&0xFF];
-	} else {
-		if (t->eqp) {
-			q->qave += sch->qstats.backlog - (q->qave >> q->Wlog);
-		} else {
-			q->qave += q->backlog - (q->qave >> q->Wlog);
-		}
+	if (red_is_idling(&q->parms))
+		red_end_of_idle_period(&q->parms);
+
+	if (gred_wred_mode(t))
+		gred_store_wred_set(t, q);
 
+	switch (red_action(&q->parms, q->parms.qavg + qavg)) {
+		case RED_DONT_MARK:
+			break;
+
+		case RED_PROB_MARK:
+			sch->qstats.overlimits++;
+			if (!gred_use_ecn(t) || !INET_ECN_set_ce(skb)) {
+				q->stats.prob_drop++;
+				goto congestion_drop;
 			}
 
+			q->stats.prob_mark++;
+			break;
 
-	if (t->eqp && t->grio) 
-		t->tab[t->def]->qave=q->qave;
+		case RED_HARD_MARK:
+			sch->qstats.overlimits++;
+			if (gred_use_harddrop(t) || !gred_use_ecn(t) ||
+			    !INET_ECN_set_ce(skb)) {
+				q->stats.forced_drop++;
+				goto congestion_drop;
+			}
+			q->stats.forced_mark++;
+			break;
+	}
 
-	if ((q->qave+qave) < q->qth_min) {
-		q->qcount = -1;
-enqueue:
 	if (q->backlog + skb->len <= q->limit) {
 		q->backlog += skb->len;
-do_enqueue:
-			__skb_queue_tail(&sch->q, skb);
-			sch->qstats.backlog += skb->len;
-			sch->bstats.bytes += skb->len;
-			sch->bstats.packets++;
-			return 0;
-		} else {
-			q->pdrop++;
+		return qdisc_enqueue_tail(skb, sch);
 	}
 
+	q->stats.pdrop++;
 drop:
-		kfree_skb(skb);
-		sch->qstats.drops++;
-		return NET_XMIT_DROP;
-	}
-	if ((q->qave+qave) >= q->qth_max) {
-		q->qcount = -1;
-		sch->qstats.overlimits++;
-		q->forced++;
-		goto drop;
-	}
-	if (++q->qcount) {
-		if ((((qave+q->qave) - q->qth_min)>>q->Wlog)*q->qcount < q->qR)
-			goto enqueue;
-		q->qcount = 0;
-		q->qR = net_random()&q->Rmask;
-		sch->qstats.overlimits++;
-		q->early++;
-		goto drop;
-	}
-	q->qR = net_random()&q->Rmask;
-	goto enqueue;
+	return qdisc_drop(skb, sch);
+
+congestion_drop:
+	qdisc_drop(skb, sch);
+	return NET_XMIT_CN;
 }
 
-static int
-gred_requeue(struct sk_buff *skb, struct Qdisc* sch)
+static int gred_requeue(struct sk_buff *skb, struct Qdisc* sch)
 {
+	struct gred_sched *t = qdisc_priv(sch);
 	struct gred_sched_data *q;
-	struct gred_sched *t= qdisc_priv(sch);
-	q= t->tab[(skb->tc_index&0xf)];
-/* error checking here -- probably unnecessary */
-	PSCHED_SET_PASTPERFECT(q->qidlestart);
+	u16 dp = tc_index_to_dp(skb);
 
-	__skb_queue_head(&sch->q, skb);
-	sch->qstats.backlog += skb->len;
-	sch->qstats.requeues++;
+	if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
+		if (net_ratelimit())
+			printk(KERN_WARNING "GRED: Unable to relocate VQ 0x%x "
+			       "for requeue, screwing up backlog.\n",
+			       tc_index_to_dp(skb));
+	} else {
+		if (red_is_idling(&q->parms))
+			red_end_of_idle_period(&q->parms);
 		q->backlog += skb->len;
-	return 0;
+	}
+
+	return qdisc_requeue(skb, sch);
 }
 
-static struct sk_buff *
-gred_dequeue(struct Qdisc* sch)
+static struct sk_buff *gred_dequeue(struct Qdisc* sch)
 {
 	struct sk_buff *skb;
-	struct gred_sched_data *q;
-	struct gred_sched *t= qdisc_priv(sch);
+	struct gred_sched *t = qdisc_priv(sch);
+
+	skb = qdisc_dequeue_head(sch);
 
-	skb = __skb_dequeue(&sch->q);
 	if (skb) {
-		sch->qstats.backlog -= skb->len;
-		q= t->tab[(skb->tc_index&0xf)];
-		if (q) {
-			q->backlog -= skb->len;
-			if (!q->backlog && !t->eqp)
-				PSCHED_GET_TIME(q->qidlestart);
+		struct gred_sched_data *q;
+		u16 dp = tc_index_to_dp(skb);
+
+		if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
+			if (net_ratelimit())
+				printk(KERN_WARNING "GRED: Unable to relocate "
+				       "VQ 0x%x after dequeue, screwing up "
+				       "backlog.\n", tc_index_to_dp(skb));
 		} else {
-			D2PRINTK("gred_dequeue: skb has bad tcindex %x\n",skb->tc_index&0xf); 
+			q->backlog -= skb->len;
+
+			if (!q->backlog && !gred_wred_mode(t))
+				red_start_of_idle_period(&q->parms);
 		}
+
 		return skb;
 	}
 
-	if (t->eqp) {
-			q= t->tab[t->def];
-			if (!q)	
-				D2PRINTK("no default VQ set: Results will be "
-				       "screwed up\n");
-			else
-				PSCHED_GET_TIME(q->qidlestart);
-	}
+	if (gred_wred_mode(t) && !red_is_idling(&t->wred_set))
+		red_start_of_idle_period(&t->wred_set);
 
 	return NULL;
 }
@@ -263,36 +297,34 @@ gred_dequeue(struct Qdisc* sch)
 static unsigned int gred_drop(struct Qdisc* sch)
 {
 	struct sk_buff *skb;
+	struct gred_sched *t = qdisc_priv(sch);
 
-	struct gred_sched_data *q;
-	struct gred_sched *t= qdisc_priv(sch);
-
-	skb = __skb_dequeue_tail(&sch->q);
+	skb = qdisc_dequeue_tail(sch);
 	if (skb) {
 		unsigned int len = skb->len;
-		sch->qstats.backlog -= len;
-		sch->qstats.drops++;
-		q= t->tab[(skb->tc_index&0xf)];
-		if (q) {
-			q->backlog -= len;
-			q->other++;
-			if (!q->backlog && !t->eqp)
-				PSCHED_GET_TIME(q->qidlestart);
+		struct gred_sched_data *q;
+		u16 dp = tc_index_to_dp(skb);
+
+		if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
+			if (net_ratelimit())
+				printk(KERN_WARNING "GRED: Unable to relocate "
+				       "VQ 0x%x while dropping, screwing up "
+				       "backlog.\n", tc_index_to_dp(skb));
 		} else {
-			D2PRINTK("gred_dequeue: skb has bad tcindex %x\n",skb->tc_index&0xf); 
+			q->backlog -= len;
+			q->stats.other++;
+
+			if (!q->backlog && !gred_wred_mode(t))
+				red_start_of_idle_period(&q->parms);
 		}
 
-		kfree_skb(skb);
+		qdisc_drop(skb, sch);
 		return len;
 	}
 
-	q=t->tab[t->def];
-	if (!q) {
-		D2PRINTK("no default VQ set: Results might be screwed up\n");
-		return 0;
-	}
+	if (gred_wred_mode(t) && !red_is_idling(&t->wred_set))
+		red_start_of_idle_period(&t->wred_set);
 
-	PSCHED_GET_TIME(q->qidlestart);
 	return 0;
 
 }
@@ -300,293 +332,241 @@ static unsigned int gred_drop(struct Qdisc* sch)
 static void gred_reset(struct Qdisc* sch)
 {
 	int i;
-	struct gred_sched_data *q;
-	struct gred_sched *t= qdisc_priv(sch);
+	struct gred_sched *t = qdisc_priv(sch);
 
-	__skb_queue_purge(&sch->q);
+	qdisc_reset_queue(sch);
 
-	sch->qstats.backlog = 0;
+        for (i = 0; i < t->DPs; i++) {
+		struct gred_sched_data *q = t->tab[i];
 
-        for (i=0;i<t->DPs;i++) {
-	        q= t->tab[i];
 		if (!q)
 			continue;
-		PSCHED_SET_PASTPERFECT(q->qidlestart);
-		q->qave = 0;
-		q->qcount = -1;
+
+		red_restart(&q->parms);
 		q->backlog = 0;
-		q->other=0;
-		q->forced=0;
-		q->pdrop=0;
-		q->early=0;
 	}
 }
 
-static int gred_change(struct Qdisc *sch, struct rtattr *opt)
+static inline void gred_destroy_vq(struct gred_sched_data *q)
+{
+	kfree(q);
+}
+
+static inline int gred_change_table_def(struct Qdisc *sch, struct rtattr *dps)
 {
 	struct gred_sched *table = qdisc_priv(sch);
-	struct gred_sched_data *q;
-	struct tc_gred_qopt *ctl;
 	struct tc_gred_sopt *sopt;
-	struct rtattr *tb[TCA_GRED_STAB];
-	struct rtattr *tb2[TCA_GRED_DPS];
 	int i;
 
-	if (opt == NULL || rtattr_parse_nested(tb, TCA_GRED_STAB, opt))
+	if (dps == NULL || RTA_PAYLOAD(dps) < sizeof(*sopt))
 		return -EINVAL;
 
-	if (tb[TCA_GRED_PARMS-1] == 0 && tb[TCA_GRED_STAB-1] == 0) {
-		rtattr_parse_nested(tb2, TCA_GRED_DPS, opt);
+	sopt = RTA_DATA(dps);
 
-	    if (tb2[TCA_GRED_DPS-1] == 0) 
+	if (sopt->DPs > MAX_DPs || sopt->DPs == 0 || sopt->def_DP >= sopt->DPs)
 		return -EINVAL;
 
-		sopt = RTA_DATA(tb2[TCA_GRED_DPS-1]);
-		table->DPs=sopt->DPs;   
-		table->def=sopt->def_DP; 
-		table->grio=sopt->grio; 
-		table->initd=0;
-		/* probably need to clear all the table DP entries as well */
+	sch_tree_lock(sch);
+	table->DPs = sopt->DPs;
+	table->def = sopt->def_DP;
+	table->red_flags = sopt->flags;
+
+	/*
+	 * Every entry point to GRED is synchronized with the above code
+	 * and the DP is checked against DPs, i.e. shadowed VQs can no
+	 * longer be found so we can unlock right here.
+	 */
+	sch_tree_unlock(sch);
+
+	if (sopt->grio) {
+		gred_enable_rio_mode(table);
+		gred_disable_wred_mode(table);
+		if (gred_wred_mode_check(sch))
+			gred_enable_wred_mode(table);
+	} else {
+		gred_disable_rio_mode(table);
+		gred_disable_wred_mode(table);
+	}
+
+	for (i = table->DPs; i < MAX_DPs; i++) {
+		if (table->tab[i]) {
+			printk(KERN_WARNING "GRED: Warning: Destroying "
+			       "shadowed VQ 0x%x\n", i);
+			gred_destroy_vq(table->tab[i]);
+			table->tab[i] = NULL;
+  		}
+	}
+
 	return 0;
+}
+
+static inline int gred_change_vq(struct Qdisc *sch, int dp,
+				 struct tc_gred_qopt *ctl, int prio, u8 *stab)
+{
+	struct gred_sched *table = qdisc_priv(sch);
+	struct gred_sched_data *q;
+
+	if (table->tab[dp] == NULL) {
+		table->tab[dp] = kmalloc(sizeof(*q), GFP_KERNEL);
+		if (table->tab[dp] == NULL)
+			return -ENOMEM;
+		memset(table->tab[dp], 0, sizeof(*q));
 	}
 
+	q = table->tab[dp];
+	q->DP = dp;
+	q->prio = prio;
+	q->limit = ctl->limit;
+
+	if (q->backlog == 0)
+		red_end_of_idle_period(&q->parms);
+
+	red_set_parms(&q->parms,
+		      ctl->qth_min, ctl->qth_max, ctl->Wlog, ctl->Plog,
+		      ctl->Scell_log, stab);
+
+	return 0;
+}
+
+static int gred_change(struct Qdisc *sch, struct rtattr *opt)
+{
+	struct gred_sched *table = qdisc_priv(sch);
+	struct tc_gred_qopt *ctl;
+	struct rtattr *tb[TCA_GRED_MAX];
+	int err = -EINVAL, prio = GRED_DEF_PRIO;
+	u8 *stab;
+
+	if (opt == NULL || rtattr_parse_nested(tb, TCA_GRED_MAX, opt))
+		return -EINVAL;
+
+	if (tb[TCA_GRED_PARMS-1] == NULL && tb[TCA_GRED_STAB-1] == NULL)
+		return gred_change_table_def(sch, opt);
 
-	if (!table->DPs || tb[TCA_GRED_PARMS-1] == 0 || tb[TCA_GRED_STAB-1] == 0 ||
+	if (tb[TCA_GRED_PARMS-1] == NULL ||
 	    RTA_PAYLOAD(tb[TCA_GRED_PARMS-1]) < sizeof(*ctl) ||
+	    tb[TCA_GRED_STAB-1] == NULL ||
 	    RTA_PAYLOAD(tb[TCA_GRED_STAB-1]) < 256)
 		return -EINVAL;
 
 	ctl = RTA_DATA(tb[TCA_GRED_PARMS-1]);
-	if (ctl->DP > MAX_DPs-1 ) {
-		/* misbehaving is punished! Put in the default drop probability */
-		DPRINTK("\nGRED: DP %u not in  the proper range fixed. New DP "
-			"set to default at %d\n",ctl->DP,table->def);
-		ctl->DP=table->def;
-	}
+	stab = RTA_DATA(tb[TCA_GRED_STAB-1]);
 
-	if (table->tab[ctl->DP] == NULL) {
-		table->tab[ctl->DP]=kmalloc(sizeof(struct gred_sched_data),
-					    GFP_KERNEL);
-		if (NULL == table->tab[ctl->DP])
-			return -ENOMEM;
-		memset(table->tab[ctl->DP], 0, (sizeof(struct gred_sched_data)));
-	}
-	q= table->tab[ctl->DP]; 
-
-	if (table->grio) {
-		if (ctl->prio <=0) {
-			if (table->def && table->tab[table->def]) {
-				DPRINTK("\nGRED: DP %u does not have a prio"
-					"setting default to %d\n",ctl->DP,
-					table->tab[table->def]->prio);
-				q->prio=table->tab[table->def]->prio;
-			} else { 
-				DPRINTK("\nGRED: DP %u does not have a prio"
-					" setting default to 8\n",ctl->DP);
-				q->prio=8;
-			}
-		} else {
-			q->prio=ctl->prio;
-		}
-	} else {
-		q->prio=8;
-	}
+	if (ctl->DP >= table->DPs)
+		goto errout;
 
+	if (gred_rio_mode(table)) {
+		if (ctl->prio == 0) {
+			int def_prio = GRED_DEF_PRIO;
 
-	q->DP=ctl->DP;
-	q->Wlog = ctl->Wlog;
-	q->Plog = ctl->Plog;
-	q->limit = ctl->limit;
-	q->Scell_log = ctl->Scell_log;
-	q->Rmask = ctl->Plog < 32 ? ((1<<ctl->Plog) - 1) : ~0UL;
-	q->Scell_max = (255<<q->Scell_log);
-	q->qth_min = ctl->qth_min<<ctl->Wlog;
-	q->qth_max = ctl->qth_max<<ctl->Wlog;
-	q->qave=0;
-	q->backlog=0;
-	q->qcount = -1;
-	q->other=0;
-	q->forced=0;
-	q->pdrop=0;
-	q->early=0;
-
-	PSCHED_SET_PASTPERFECT(q->qidlestart);
-	memcpy(q->Stab, RTA_DATA(tb[TCA_GRED_STAB-1]), 256);
-
-	if ( table->initd && table->grio) {
-	/* this looks ugly but it's not in the fast path */
-		for (i=0;i<table->DPs;i++) {
-			if ((!table->tab[i]) || (i==q->DP) )    
-				continue; 
-			if (table->tab[i]->prio == q->prio ){
-				/* WRED mode detected */
-				table->eqp=1;
-				break;
-			}
-		}
+			if (table->tab[table->def])
+				def_prio = table->tab[table->def]->prio;
+
+			printk(KERN_DEBUG "GRED: DP %u does not have a prio "
+			       "setting default to %d\n", ctl->DP, def_prio);
+
+			prio = def_prio;
+		} else
+			prio = ctl->prio;
 	}
 
-	if (!table->initd) {
-		table->initd=1;
-		/* 
-        	the first entry also goes into the default until
-        	over-written 
-		*/
+	sch_tree_lock(sch);
 
-		if (table->tab[table->def] == NULL) {
-			table->tab[table->def]=
-				kmalloc(sizeof(struct gred_sched_data), GFP_KERNEL);
-			if (NULL == table->tab[table->def])
-				return -ENOMEM;
+	err = gred_change_vq(sch, ctl->DP, ctl, prio, stab);
+	if (err < 0)
+		goto errout_locked;
 
-			memset(table->tab[table->def], 0,
-			       (sizeof(struct gred_sched_data)));
+	if (gred_rio_mode(table)) {
+		gred_disable_wred_mode(table);
+		if (gred_wred_mode_check(sch))
+			gred_enable_wred_mode(table);
 	}
-		q= table->tab[table->def]; 
-		q->DP=table->def;
-		q->Wlog = ctl->Wlog;
-		q->Plog = ctl->Plog;
-		q->limit = ctl->limit;
-		q->Scell_log = ctl->Scell_log;
-		q->Rmask = ctl->Plog < 32 ? ((1<<ctl->Plog) - 1) : ~0UL;
-		q->Scell_max = (255<<q->Scell_log);
-		q->qth_min = ctl->qth_min<<ctl->Wlog;
-		q->qth_max = ctl->qth_max<<ctl->Wlog;
-
-		if (table->grio)
-			q->prio=table->tab[ctl->DP]->prio;
-		else
-			q->prio=8;
 
-		q->qcount = -1;
-		PSCHED_SET_PASTPERFECT(q->qidlestart);
-		memcpy(q->Stab, RTA_DATA(tb[TCA_GRED_STAB-1]), 256);
-	}
-	return 0;
+	err = 0;
 
+errout_locked:
+	sch_tree_unlock(sch);
+errout:
+	return err;
 }
 
 static int gred_init(struct Qdisc *sch, struct rtattr *opt)
 {
-	struct gred_sched *table = qdisc_priv(sch);
-	struct tc_gred_sopt *sopt;
-	struct rtattr *tb[TCA_GRED_STAB];
-	struct rtattr *tb2[TCA_GRED_DPS];
+	struct rtattr *tb[TCA_GRED_MAX];
 
-	if (opt == NULL || rtattr_parse_nested(tb, TCA_GRED_STAB, opt))
+	if (opt == NULL || rtattr_parse_nested(tb, TCA_GRED_MAX, opt))
 		return -EINVAL;
 
-	if (tb[TCA_GRED_PARMS-1] == 0 && tb[TCA_GRED_STAB-1] == 0) {
-		rtattr_parse_nested(tb2, TCA_GRED_DPS, opt);
-
-	    if (tb2[TCA_GRED_DPS-1] == 0) 
+	if (tb[TCA_GRED_PARMS-1] || tb[TCA_GRED_STAB-1])
 		return -EINVAL;
 
-		sopt = RTA_DATA(tb2[TCA_GRED_DPS-1]);
-		table->DPs=sopt->DPs;   
-		table->def=sopt->def_DP; 
-		table->grio=sopt->grio; 
-		table->initd=0;
-		return 0;
-	}
-
-	DPRINTK("\n GRED_INIT error!\n");
-	return -EINVAL;
+	return gred_change_table_def(sch, tb[TCA_GRED_DPS-1]);
 }
 
 static int gred_dump(struct Qdisc *sch, struct sk_buff *skb)
 {
-	unsigned long qave;
-	struct rtattr *rta;
-	struct tc_gred_qopt *opt = NULL ;
-	struct tc_gred_qopt *dst;
 	struct gred_sched *table = qdisc_priv(sch);
-	struct gred_sched_data *q;
+	struct rtattr *parms, *opts = NULL;
 	int i;
-	unsigned char	 *b = skb->tail;
-
-	rta = (struct rtattr*)b;
-	RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
-
-	opt=kmalloc(sizeof(struct tc_gred_qopt)*MAX_DPs, GFP_KERNEL);
+	struct tc_gred_sopt sopt = {
+		.DPs	= table->DPs,
+		.def_DP	= table->def,
+		.grio	= gred_rio_mode(table),
+		.flags	= table->red_flags,
+	};
 
-	if (opt  == NULL) {
-		DPRINTK("gred_dump:failed to malloc for %Zd\n",
-		    sizeof(struct tc_gred_qopt)*MAX_DPs);
-		goto rtattr_failure;
-	}
-
-	memset(opt, 0, (sizeof(struct tc_gred_qopt))*table->DPs);
+	opts = RTA_NEST(skb, TCA_OPTIONS);
+	RTA_PUT(skb, TCA_GRED_DPS, sizeof(sopt), &sopt);
+	parms = RTA_NEST(skb, TCA_GRED_PARMS);
 
-	if (!table->initd) {
-		DPRINTK("NO GRED Queues setup!\n");
-	}
+	for (i = 0; i < MAX_DPs; i++) {
+		struct gred_sched_data *q = table->tab[i];
+		struct tc_gred_qopt opt;
 
-	for (i=0;i<MAX_DPs;i++) {
-		dst= &opt[i]; 
-		q= table->tab[i]; 
+		memset(&opt, 0, sizeof(opt));
 
 		if (!q) {
 			/* hack -- fix at some point with proper message
 			   This is how we indicate to tc that there is no VQ
 			   at this DP */
 
-			dst->DP=MAX_DPs+i;
-			continue;
+			opt.DP = MAX_DPs + i;
+			goto append_opt;
 		}
 
-		dst->limit=q->limit;
-		dst->qth_min=q->qth_min>>q->Wlog;
-		dst->qth_max=q->qth_max>>q->Wlog;
-		dst->DP=q->DP;
-		dst->backlog=q->backlog;
-		if (q->qave) {
-			if (table->eqp && table->grio) {
-				q->qidlestart=table->tab[table->def]->qidlestart;
-				q->qave=table->tab[table->def]->qave;
-			}
-			if (!PSCHED_IS_PASTPERFECT(q->qidlestart)) {
-				long idle;
-				psched_time_t now;
-				PSCHED_GET_TIME(now);
-				idle = PSCHED_TDIFF_SAFE(now, q->qidlestart, q->Scell_max);
-				qave  = q->qave >> q->Stab[(idle>>q->Scell_log)&0xFF];
-				dst->qave = qave >> q->Wlog;
+		opt.limit	= q->limit;
+		opt.DP		= q->DP;
+		opt.backlog	= q->backlog;
+		opt.prio	= q->prio;
+		opt.qth_min	= q->parms.qth_min >> q->parms.Wlog;
+		opt.qth_max	= q->parms.qth_max >> q->parms.Wlog;
+		opt.Wlog	= q->parms.Wlog;
+		opt.Plog	= q->parms.Plog;
+		opt.Scell_log	= q->parms.Scell_log;
+		opt.other	= q->stats.other;
+		opt.early	= q->stats.prob_drop;
+		opt.forced	= q->stats.forced_drop;
+		opt.pdrop	= q->stats.pdrop;
+		opt.packets	= q->packetsin;
+		opt.bytesin	= q->bytesin;
 
-			} else {
-				dst->qave = q->qave >> q->Wlog;
-			}
-		} else {
-			dst->qave = 0;
+		if (gred_wred_mode(table)) {
+			q->parms.qidlestart =
+				table->tab[table->def]->parms.qidlestart;
+			q->parms.qavg = table->tab[table->def]->parms.qavg;
 		}
 
+		opt.qave = red_calc_qavg(&q->parms, q->parms.qavg);
 
-		dst->Wlog = q->Wlog;
-		dst->Plog = q->Plog;
-		dst->Scell_log = q->Scell_log;
-		dst->other = q->other;
-		dst->forced = q->forced;
-		dst->early = q->early;
-		dst->pdrop = q->pdrop;
-		dst->prio = q->prio;
-		dst->packets=q->packetsin;
-		dst->bytesin=q->bytesin;
+append_opt:
+		RTA_APPEND(skb, sizeof(opt), &opt);
 	}
 
-	RTA_PUT(skb, TCA_GRED_PARMS, sizeof(struct tc_gred_qopt)*MAX_DPs, opt);
-	rta->rta_len = skb->tail - b;
+	RTA_NEST_END(skb, parms);
 
-	kfree(opt);
-	return skb->len;
+	return RTA_NEST_END(skb, opts);
 
 rtattr_failure:
-	if (opt)
-		kfree(opt);
-	DPRINTK("gred_dump: FAILURE!!!!\n");
-
-/* also free the opt struct here */
-	skb_trim(skb, b - skb->data);
-	return -1;
+	return RTA_NEST_CANCEL(skb, opts);
 }
 
 static void gred_destroy(struct Qdisc *sch)
@@ -594,15 +574,13 @@ static void gred_destroy(struct Qdisc *sch)
 	struct gred_sched *table = qdisc_priv(sch);
 	int i;
 
-	for (i = 0;i < table->DPs; i++) {
+	for (i = 0; i < table->DPs; i++) {
 		if (table->tab[i])
-			kfree(table->tab[i]);
+			gred_destroy_vq(table->tab[i]);
 	}
 }
 
 static struct Qdisc_ops gred_qdisc_ops = {
-	.next		=	NULL,
-	.cl_ops		=	NULL,
 	.id		=	"gred",
 	.priv_size	=	sizeof(struct gred_sched),
 	.enqueue	=	gred_enqueue,
@@ -621,10 +599,13 @@ static int __init gred_module_init(void)
 {
 	return register_qdisc(&gred_qdisc_ops);
 }
+
 static void __exit gred_module_exit(void)
 {
 	unregister_qdisc(&gred_qdisc_ops);
 }
+
 module_init(gred_module_init)
 module_exit(gred_module_exit)
+
 MODULE_LICENSE("GPL");

net/sched/sch_netem.c

@@ -25,6 +25,8 @@
 
 #include <net/pkt_sched.h>
 
+#define VERSION "1.1"
+
 /*	Network Emulation Queuing algorithm.
 	====================================
 
@@ -185,10 +187,13 @@ static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 	    || q->counter < q->gap 	/* inside last reordering gap */
 	    || q->reorder < get_crandom(&q->reorder_cor)) {
 		psched_time_t now;
+		psched_tdiff_t delay;
+
+		delay = tabledist(q->latency, q->jitter,
+				  &q->delay_cor, q->delay_dist);
+
 		PSCHED_GET_TIME(now);
-		PSCHED_TADD2(now, tabledist(q->latency, q->jitter, 
-					    &q->delay_cor, q->delay_dist),
-			     cb->time_to_send);
+		PSCHED_TADD2(now, delay, cb->time_to_send);
 		++q->counter;
 		ret = q->qdisc->enqueue(skb, q->qdisc);
 	} else {
@@ -248,24 +253,31 @@ static struct sk_buff *netem_dequeue(struct Qdisc *sch)
 		const struct netem_skb_cb *cb
 			= (const struct netem_skb_cb *)skb->cb;
 		psched_time_t now;
-		long delay;
 
 		/* if more time remaining? */
 		PSCHED_GET_TIME(now);
-		delay = PSCHED_US2JIFFIE(PSCHED_TDIFF(cb->time_to_send, now));
-		pr_debug("netem_run: skb=%p delay=%ld\n", skb, delay);
-		if (delay <= 0) {
+
+		if (PSCHED_TLESS(cb->time_to_send, now)) {
 			pr_debug("netem_dequeue: return skb=%p\n", skb);
 			sch->q.qlen--;
 			sch->flags &= ~TCQ_F_THROTTLED;
 			return skb;
+		} else {
+			psched_tdiff_t delay = PSCHED_TDIFF(cb->time_to_send, now);
+
+			if (q->qdisc->ops->requeue(skb, q->qdisc) != NET_XMIT_SUCCESS) {
+				sch->qstats.drops++;
+
+				/* After this qlen is confused */
+				printk(KERN_ERR "netem: queue discpline %s could not requeue\n",
+				       q->qdisc->ops->id);
+
+				sch->q.qlen--;
 			}
 
-		mod_timer(&q->timer, jiffies + delay);
+			mod_timer(&q->timer, jiffies + PSCHED_US2JIFFIE(delay));
 			sch->flags |= TCQ_F_THROTTLED;
-
-		if (q->qdisc->ops->requeue(skb, q->qdisc) != 0)
-			sch->qstats.drops++;
+		}
 	}
 
 	return NULL;
@@ -290,11 +302,16 @@ static void netem_reset(struct Qdisc *sch)
 	del_timer_sync(&q->timer);
 }
 
+/* Pass size change message down to embedded FIFO */
 static int set_fifo_limit(struct Qdisc *q, int limit)
 {
         struct rtattr *rta;
 	int ret = -ENOMEM;
 
+	/* Hack to avoid sending change message to non-FIFO */
+	if (strncmp(q->ops->id + 1, "fifo", 4) != 0)
+		return 0;
+
 	rta = kmalloc(RTA_LENGTH(sizeof(struct tc_fifo_qopt)), GFP_KERNEL);
 	if (rta) {
 		rta->rta_type = RTM_NEWQDISC;
@@ -426,6 +443,84 @@ static int netem_change(struct Qdisc *sch, struct rtattr *opt)
 	return 0;
 }
 
+/*
+ * Special case version of FIFO queue for use by netem.
+ * It queues in order based on timestamps in skb's
+ */
+struct fifo_sched_data {
+	u32 limit;
+};
+
+static int tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch)
+{
+	struct fifo_sched_data *q = qdisc_priv(sch);
+	struct sk_buff_head *list = &sch->q;
+	const struct netem_skb_cb *ncb
+		= (const struct netem_skb_cb *)nskb->cb;
+	struct sk_buff *skb;
+
+	if (likely(skb_queue_len(list) < q->limit)) {
+		skb_queue_reverse_walk(list, skb) {
+			const struct netem_skb_cb *cb
+				= (const struct netem_skb_cb *)skb->cb;
+
+			if (PSCHED_TLESS(cb->time_to_send, ncb->time_to_send))
+				break;
+		}
+
+		__skb_queue_after(list, skb, nskb);
+
+		sch->qstats.backlog += nskb->len;
+		sch->bstats.bytes += nskb->len;
+		sch->bstats.packets++;
+
+		return NET_XMIT_SUCCESS;
+	}
+
+	return qdisc_drop(nskb, sch);
+}
+
+static int tfifo_init(struct Qdisc *sch, struct rtattr *opt)
+{
+	struct fifo_sched_data *q = qdisc_priv(sch);
+
+	if (opt) {
+		struct tc_fifo_qopt *ctl = RTA_DATA(opt);
+		if (RTA_PAYLOAD(opt) < sizeof(*ctl))
+			return -EINVAL;
+
+		q->limit = ctl->limit;
+	} else
+		q->limit = max_t(u32, sch->dev->tx_queue_len, 1);
+
+	return 0;
+}
+
+static int tfifo_dump(struct Qdisc *sch, struct sk_buff *skb)
+{
+	struct fifo_sched_data *q = qdisc_priv(sch);
+	struct tc_fifo_qopt opt = { .limit = q->limit };
+
+	RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
+	return skb->len;
+
+rtattr_failure:
+	return -1;
+}
+
+static struct Qdisc_ops tfifo_qdisc_ops = {
+	.id		=	"tfifo",
+	.priv_size	=	sizeof(struct fifo_sched_data),
+	.enqueue	=	tfifo_enqueue,
+	.dequeue	=	qdisc_dequeue_head,
+	.requeue	=	qdisc_requeue,
+	.drop		=	qdisc_queue_drop,
+	.init		=	tfifo_init,
+	.reset		=	qdisc_reset_queue,
+	.change		=	tfifo_init,
+	.dump		=	tfifo_dump,
+};
+
 static int netem_init(struct Qdisc *sch, struct rtattr *opt)
 {
 	struct netem_sched_data *q = qdisc_priv(sch);
@@ -438,7 +533,7 @@ static int netem_init(struct Qdisc *sch, struct rtattr *opt)
 	q->timer.function = netem_watchdog;
 	q->timer.data = (unsigned long) sch;
 
-	q->qdisc = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops);
+	q->qdisc = qdisc_create_dflt(sch->dev, &tfifo_qdisc_ops);
 	if (!q->qdisc) {
 		pr_debug("netem: qdisc create failed\n");
 		return -ENOMEM;
@@ -601,6 +696,7 @@ static struct Qdisc_ops netem_qdisc_ops = {
 
 static int __init netem_module_init(void)
 {
+	pr_info("netem: version " VERSION "\n");
 	return register_qdisc(&netem_qdisc_ops);
 }
 static void __exit netem_module_exit(void)

net/sched/sch_red.c

@@ -9,76 +9,23 @@
  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  *
  * Changes:
- * J Hadi Salim <hadi@nortel.com> 980914:	computation fixes
+ * J Hadi Salim 980914:	computation fixes
  * Alexey Makarenko <makar@phoenix.kharkov.ua> 990814: qave on idle link was calculated incorrectly.
- * J Hadi Salim <hadi@nortelnetworks.com> 980816:  ECN support	
+ * J Hadi Salim 980816:  ECN support
  */
 
 #include <linux/config.h>
 #include <linux/module.h>
-#include <asm/uaccess.h>
-#include <asm/system.h>
-#include <linux/bitops.h>
 #include <linux/types.h>
 #include <linux/kernel.h>
-#include <linux/sched.h>
-#include <linux/string.h>
-#include <linux/mm.h>
-#include <linux/socket.h>
-#include <linux/sockios.h>
-#include <linux/in.h>
-#include <linux/errno.h>
-#include <linux/interrupt.h>
-#include <linux/if_ether.h>
-#include <linux/inet.h>
 #include <linux/netdevice.h>
-#include <linux/etherdevice.h>
-#include <linux/notifier.h>
-#include <net/ip.h>
-#include <net/route.h>
 #include <linux/skbuff.h>
-#include <net/sock.h>
 #include <net/pkt_sched.h>
 #include <net/inet_ecn.h>
-#include <net/dsfield.h>
+#include <net/red.h>
 
 
-/*	Random Early Detection (RED) algorithm.
-	=======================================
-
-	Source: Sally Floyd and Van Jacobson, "Random Early Detection Gateways
-	for Congestion Avoidance", 1993, IEEE/ACM Transactions on Networking.
-
-	This file codes a "divisionless" version of RED algorithm
-	as written down in Fig.17 of the paper.
-
-Short description.
-------------------
-
-	When a new packet arrives we calculate the average queue length:
-
-	avg = (1-W)*avg + W*current_queue_len,
-
-	W is the filter time constant (chosen as 2^(-Wlog)), it controls
-	the inertia of the algorithm. To allow larger bursts, W should be
-	decreased.
-
-	if (avg > th_max) -> packet marked (dropped).
-	if (avg < th_min) -> packet passes.
-	if (th_min < avg < th_max) we calculate probability:
-
-	Pb = max_P * (avg - th_min)/(th_max-th_min)
-
-	and mark (drop) packet with this probability.
-	Pb changes from 0 (at avg==th_min) to max_P (avg==th_max).
-	max_P should be small (not 1), usually 0.01..0.02 is good value.
-
-	max_P is chosen as a number, so that max_P/(th_max-th_min)
-	is a negative power of two in order arithmetics to contain
-	only shifts.
-
-
-	Parameters, settable by user:
+/*	Parameters, settable by user:
 	-----------------------------
 
 	limit		- bytes (must be > qth_max + burst)
@@ -89,243 +36,93 @@ Short description.
 	arbitrarily high (well, less than ram size)
 	Really, this limit will never be reached
 	if RED works correctly.
-
-	qth_min		- bytes (should be < qth_max/2)
-	qth_max		- bytes (should be at least 2*qth_min and less limit)
-	Wlog	       	- bits (<32) log(1/W).
-	Plog	       	- bits (<32)
-
-	Plog is related to max_P by formula:
-
-	max_P = (qth_max-qth_min)/2^Plog;
-
-	F.e. if qth_max=128K and qth_min=32K, then Plog=22
-	corresponds to max_P=0.02
-
-	Scell_log
-	Stab
-
-	Lookup table for log((1-W)^(t/t_ave).
-
-
-NOTES:
-
-Upper bound on W.
------------------
-
-	If you want to allow bursts of L packets of size S,
-	you should choose W:
-
-	L + 1 - th_min/S < (1-(1-W)^L)/W
-
-	th_min/S = 32         th_min/S = 4
-			                       
-	log(W)	L
-	-1	33
-	-2	35
-	-3	39
-	-4	46
-	-5	57
-	-6	75
-	-7	101
-	-8	135
-	-9	190
-	etc.
  */
 
 struct red_sched_data
 {
-/* Parameters */
 	u32			limit;		/* HARD maximal queue length */
-	u32		qth_min;	/* Min average length threshold: A scaled */
-	u32		qth_max;	/* Max average length threshold: A scaled */
-	u32		Rmask;
-	u32		Scell_max;
 	unsigned char		flags;
-	char		Wlog;		/* log(W)		*/
-	char		Plog;		/* random number bits	*/
-	char		Scell_log;
-	u8		Stab[256];
-
-/* Variables */
-	unsigned long	qave;		/* Average queue length: A scaled */
-	int		qcount;		/* Packets since last random number generation */
-	u32		qR;		/* Cached random number */
-
-	psched_time_t	qidlestart;	/* Start of idle period		*/
-	struct tc_red_xstats st;
+	struct red_parms	parms;
+	struct red_stats	stats;
 };
 
-static int red_ecn_mark(struct sk_buff *skb)
+static inline int red_use_ecn(struct red_sched_data *q)
 {
-	if (skb->nh.raw + 20 > skb->tail)
-		return 0;
+	return q->flags & TC_RED_ECN;
+}
 
-	switch (skb->protocol) {
-	case __constant_htons(ETH_P_IP):
-		if (INET_ECN_is_not_ect(skb->nh.iph->tos))
-			return 0;
-		IP_ECN_set_ce(skb->nh.iph);
-		return 1;
-	case __constant_htons(ETH_P_IPV6):
-		if (INET_ECN_is_not_ect(ipv6_get_dsfield(skb->nh.ipv6h)))
-			return 0;
-		IP6_ECN_set_ce(skb->nh.ipv6h);
-		return 1;
-	default:
-		return 0;
-	}
+static inline int red_use_harddrop(struct red_sched_data *q)
+{
+	return q->flags & TC_RED_HARDDROP;
 }
 
-static int
-red_enqueue(struct sk_buff *skb, struct Qdisc* sch)
+static int red_enqueue(struct sk_buff *skb, struct Qdisc* sch)
 {
 	struct red_sched_data *q = qdisc_priv(sch);
 
-	psched_time_t now;
+	q->parms.qavg = red_calc_qavg(&q->parms, sch->qstats.backlog);
 
-	if (!PSCHED_IS_PASTPERFECT(q->qidlestart)) {
-		long us_idle;
-		int  shift;
+	if (red_is_idling(&q->parms))
+		red_end_of_idle_period(&q->parms);
 
-		PSCHED_GET_TIME(now);
-		us_idle = PSCHED_TDIFF_SAFE(now, q->qidlestart, q->Scell_max);
-		PSCHED_SET_PASTPERFECT(q->qidlestart);
+	switch (red_action(&q->parms, q->parms.qavg)) {
+		case RED_DONT_MARK:
+			break;
 
-/*
-   The problem: ideally, average length queue recalcultion should
-   be done over constant clock intervals. This is too expensive, so that
-   the calculation is driven by outgoing packets.
-   When the queue is idle we have to model this clock by hand.
-
-   SF+VJ proposed to "generate" m = idletime/(average_pkt_size/bandwidth)
-   dummy packets as a burst after idle time, i.e.
-
-          q->qave *= (1-W)^m
-
-   This is an apparently overcomplicated solution (f.e. we have to precompute
-   a table to make this calculation in reasonable time)
-   I believe that a simpler model may be used here,
-   but it is field for experiments.
-*/
-		shift = q->Stab[us_idle>>q->Scell_log];
-
-		if (shift) {
-			q->qave >>= shift;
-		} else {
-			/* Approximate initial part of exponent
-			   with linear function:
-			   (1-W)^m ~= 1-mW + ...
-
-			   Seems, it is the best solution to
-			   problem of too coarce exponent tabulation.
-			 */
-
-			us_idle = (q->qave * us_idle)>>q->Scell_log;
-			if (us_idle < q->qave/2)
-				q->qave -= us_idle;
-			else
-				q->qave >>= 1;
-		}
-	} else {
-		q->qave += sch->qstats.backlog - (q->qave >> q->Wlog);
-		/* NOTE:
-		   q->qave is fixed point number with point at Wlog.
-		   The formulae above is equvalent to floating point
-		   version:
-
-		   qave = qave*(1-W) + sch->qstats.backlog*W;
-		                                           --ANK (980924)
-		 */
-	}
-
-	if (q->qave < q->qth_min) {
-		q->qcount = -1;
-enqueue:
-		if (sch->qstats.backlog + skb->len <= q->limit) {
-			__skb_queue_tail(&sch->q, skb);
-			sch->qstats.backlog += skb->len;
-			sch->bstats.bytes += skb->len;
-			sch->bstats.packets++;
-			return NET_XMIT_SUCCESS;
-		} else {
-			q->st.pdrop++;
-		}
-		kfree_skb(skb);
-		sch->qstats.drops++;
-		return NET_XMIT_DROP;
-	}
-	if (q->qave >= q->qth_max) {
-		q->qcount = -1;
+		case RED_PROB_MARK:
 			sch->qstats.overlimits++;
-mark:
-		if  (!(q->flags&TC_RED_ECN) || !red_ecn_mark(skb)) {
-			q->st.early++;
-			goto drop;
-		}
-		q->st.marked++;
-		goto enqueue;
+			if (!red_use_ecn(q) || !INET_ECN_set_ce(skb)) {
+				q->stats.prob_drop++;
+				goto congestion_drop;
 			}
 
-	if (++q->qcount) {
-		/* The formula used below causes questions.
+			q->stats.prob_mark++;
+			break;
 
-		   OK. qR is random number in the interval 0..Rmask
-		   i.e. 0..(2^Plog). If we used floating point
-		   arithmetics, it would be: (2^Plog)*rnd_num,
-		   where rnd_num is less 1.
+		case RED_HARD_MARK:
+			sch->qstats.overlimits++;
+			if (red_use_harddrop(q) || !red_use_ecn(q) ||
+			    !INET_ECN_set_ce(skb)) {
+				q->stats.forced_drop++;
+				goto congestion_drop;
+			}
 
-		   Taking into account, that qave have fixed
-		   point at Wlog, and Plog is related to max_P by
-		   max_P = (qth_max-qth_min)/2^Plog; two lines
-		   below have the following floating point equivalent:
+			q->stats.forced_mark++;
+			break;
+	}
 
-		   max_P*(qave - qth_min)/(qth_max-qth_min) < rnd/qcount
+	if (sch->qstats.backlog + skb->len <= q->limit)
+		return qdisc_enqueue_tail(skb, sch);
 
-		   Any questions? --ANK (980924)
-		 */
-		if (((q->qave - q->qth_min)>>q->Wlog)*q->qcount < q->qR)
-			goto enqueue;
-		q->qcount = 0;
-		q->qR = net_random()&q->Rmask;
-		sch->qstats.overlimits++;
-		goto mark;
-	}
-	q->qR = net_random()&q->Rmask;
-	goto enqueue;
+	q->stats.pdrop++;
+	return qdisc_drop(skb, sch);
 
-drop:
-	kfree_skb(skb);
-	sch->qstats.drops++;
+congestion_drop:
+	qdisc_drop(skb, sch);
 	return NET_XMIT_CN;
 }
 
-static int
-red_requeue(struct sk_buff *skb, struct Qdisc* sch)
+static int red_requeue(struct sk_buff *skb, struct Qdisc* sch)
 {
 	struct red_sched_data *q = qdisc_priv(sch);
 
-	PSCHED_SET_PASTPERFECT(q->qidlestart);
+	if (red_is_idling(&q->parms))
+		red_end_of_idle_period(&q->parms);
 
-	__skb_queue_head(&sch->q, skb);
-	sch->qstats.backlog += skb->len;
-	sch->qstats.requeues++;
-	return 0;
+	return qdisc_requeue(skb, sch);
 }
 
-static struct sk_buff *
-red_dequeue(struct Qdisc* sch)
+static struct sk_buff * red_dequeue(struct Qdisc* sch)
 {
 	struct sk_buff *skb;
 	struct red_sched_data *q = qdisc_priv(sch);
 
-	skb = __skb_dequeue(&sch->q);
-	if (skb) {
-		sch->qstats.backlog -= skb->len;
+	skb = qdisc_dequeue_head(sch);
+
+	if (skb == NULL && !red_is_idling(&q->parms))
+		red_start_of_idle_period(&q->parms);
+
 	return skb;
-	}
-	PSCHED_GET_TIME(q->qidlestart);
-	return NULL;
 }
 
 static unsigned int red_drop(struct Qdisc* sch)
@@ -333,16 +130,17 @@ static unsigned int red_drop(struct Qdisc* sch)
 	struct sk_buff *skb;
 	struct red_sched_data *q = qdisc_priv(sch);
 
-	skb = __skb_dequeue_tail(&sch->q);
+	skb = qdisc_dequeue_tail(sch);
 	if (skb) {
 		unsigned int len = skb->len;
-		sch->qstats.backlog -= len;
-		sch->qstats.drops++;
-		q->st.other++;
-		kfree_skb(skb);
+		q->stats.other++;
+		qdisc_drop(skb, sch);
 		return len;
 	}
-	PSCHED_GET_TIME(q->qidlestart);
+
+	if (!red_is_idling(&q->parms))
+		red_start_of_idle_period(&q->parms);
+
 	return 0;
 }
 
@@ -350,43 +148,38 @@ static void red_reset(struct Qdisc* sch)
 {
 	struct red_sched_data *q = qdisc_priv(sch);
 
-	__skb_queue_purge(&sch->q);
-	sch->qstats.backlog = 0;
-	PSCHED_SET_PASTPERFECT(q->qidlestart);
-	q->qave = 0;
-	q->qcount = -1;
+	qdisc_reset_queue(sch);
+	red_restart(&q->parms);
 }
 
 static int red_change(struct Qdisc *sch, struct rtattr *opt)
 {
 	struct red_sched_data *q = qdisc_priv(sch);
-	struct rtattr *tb[TCA_RED_STAB];
+	struct rtattr *tb[TCA_RED_MAX];
 	struct tc_red_qopt *ctl;
 
-	if (opt == NULL ||
-	    rtattr_parse_nested(tb, TCA_RED_STAB, opt) ||
-	    tb[TCA_RED_PARMS-1] == 0 || tb[TCA_RED_STAB-1] == 0 ||
+	if (opt == NULL || rtattr_parse_nested(tb, TCA_RED_MAX, opt))
+		return -EINVAL;
+
+	if (tb[TCA_RED_PARMS-1] == NULL ||
 	    RTA_PAYLOAD(tb[TCA_RED_PARMS-1]) < sizeof(*ctl) ||
-	    RTA_PAYLOAD(tb[TCA_RED_STAB-1]) < 256)
+	    tb[TCA_RED_STAB-1] == NULL ||
+	    RTA_PAYLOAD(tb[TCA_RED_STAB-1]) < RED_STAB_SIZE)
 		return -EINVAL;
 
 	ctl = RTA_DATA(tb[TCA_RED_PARMS-1]);
 
 	sch_tree_lock(sch);
 	q->flags = ctl->flags;
-	q->Wlog = ctl->Wlog;
-	q->Plog = ctl->Plog;
-	q->Rmask = ctl->Plog < 32 ? ((1<<ctl->Plog) - 1) : ~0UL;
-	q->Scell_log = ctl->Scell_log;
-	q->Scell_max = (255<<q->Scell_log);
-	q->qth_min = ctl->qth_min<<ctl->Wlog;
-	q->qth_max = ctl->qth_max<<ctl->Wlog;
 	q->limit = ctl->limit;
-	memcpy(q->Stab, RTA_DATA(tb[TCA_RED_STAB-1]), 256);
 
-	q->qcount = -1;
+	red_set_parms(&q->parms, ctl->qth_min, ctl->qth_max, ctl->Wlog,
+				 ctl->Plog, ctl->Scell_log,
+				 RTA_DATA(tb[TCA_RED_STAB-1]));
+
 	if (skb_queue_empty(&sch->q))
-		PSCHED_SET_PASTPERFECT(q->qidlestart);
+		red_end_of_idle_period(&q->parms);
+
 	sch_tree_unlock(sch);
 	return 0;
 }
@@ -399,39 +192,39 @@ static int red_init(struct Qdisc* sch, struct rtattr *opt)
 static int red_dump(struct Qdisc *sch, struct sk_buff *skb)
 {
 	struct red_sched_data *q = qdisc_priv(sch);
-	unsigned char	 *b = skb->tail;
-	struct rtattr *rta;
-	struct tc_red_qopt opt;
-
-	rta = (struct rtattr*)b;
-	RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
-	opt.limit = q->limit;
-	opt.qth_min = q->qth_min>>q->Wlog;
-	opt.qth_max = q->qth_max>>q->Wlog;
-	opt.Wlog = q->Wlog;
-	opt.Plog = q->Plog;
-	opt.Scell_log = q->Scell_log;
-	opt.flags = q->flags;
+	struct rtattr *opts = NULL;
+	struct tc_red_qopt opt = {
+		.limit		= q->limit,
+		.flags		= q->flags,
+		.qth_min	= q->parms.qth_min >> q->parms.Wlog,
+		.qth_max	= q->parms.qth_max >> q->parms.Wlog,
+		.Wlog		= q->parms.Wlog,
+		.Plog		= q->parms.Plog,
+		.Scell_log	= q->parms.Scell_log,
+	};
+
+	opts = RTA_NEST(skb, TCA_OPTIONS);
 	RTA_PUT(skb, TCA_RED_PARMS, sizeof(opt), &opt);
-	rta->rta_len = skb->tail - b;
-
-	return skb->len;
+	return RTA_NEST_END(skb, opts);
 
 rtattr_failure:
-	skb_trim(skb, b - skb->data);
-	return -1;
+	return RTA_NEST_CANCEL(skb, opts);
 }
 
 static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
 {
 	struct red_sched_data *q = qdisc_priv(sch);
-
-	return gnet_stats_copy_app(d, &q->st, sizeof(q->st));
+	struct tc_red_xstats st = {
+		.early	= q->stats.prob_drop + q->stats.forced_drop,
+		.pdrop	= q->stats.pdrop,
+		.other	= q->stats.other,
+		.marked	= q->stats.prob_mark + q->stats.forced_mark,
+	};
+
+	return gnet_stats_copy_app(d, &st, sizeof(st));
 }
 
 static struct Qdisc_ops red_qdisc_ops = {
-	.next		=	NULL,
-	.cl_ops		=	NULL,
 	.id		=	"red",
 	.priv_size	=	sizeof(struct red_sched_data),
 	.enqueue	=	red_enqueue,
@@ -450,10 +243,13 @@ static int __init red_module_init(void)
 {
 	return register_qdisc(&red_qdisc_ops);
 }
+
 static void __exit red_module_exit(void)
 {
 	unregister_qdisc(&red_qdisc_ops);
 }
+
 module_init(red_module_init)
 module_exit(red_module_exit)
+
 MODULE_LICENSE("GPL");