tcp_minisocks.c

Linux Kernel 2.6.9 for OMAP1710

		tw->tw_ts_recent_stamp	= tp->ts_recent_stamp;
		tw_dead_node_init(tw);

#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
		if (tw->tw_family == PF_INET6) {
			struct ipv6_pinfo *np = inet6_sk(sk);

			ipv6_addr_copy(&tw->tw_v6_daddr, &np->daddr);
			ipv6_addr_copy(&tw->tw_v6_rcv_saddr, &np->rcv_saddr);
			tw->tw_v6_ipv6only = np->ipv6only;
		} else {
			memset(&tw->tw_v6_daddr, 0, sizeof(tw->tw_v6_daddr));
			memset(&tw->tw_v6_rcv_saddr, 0, sizeof(tw->tw_v6_rcv_saddr));
			tw->tw_v6_ipv6only = 0;
		}
#endif
		/* Linkage updates. */
		__tcp_tw_hashdance(sk, tw);

		/* Get the TIME_WAIT timeout firing. */
		if (timeo < rto)
			timeo = rto;

		if (recycle_ok) {
			tw->tw_timeout = rto;
		} else {
			tw->tw_timeout = TCP_TIMEWAIT_LEN;
			if (state == TCP_TIME_WAIT)
				timeo = TCP_TIMEWAIT_LEN;
		}

		tcp_tw_schedule(tw, timeo);
		tcp_tw_put(tw);
	} else {
		/* Sorry, if we're out of memory, just CLOSE this
		 * socket up.  We've got bigger problems than
		 * non-graceful socket closings.
		 */
		if (net_ratelimit())
			printk(KERN_INFO "TCP: time wait bucket table overflow\n");
	}

	tcp_update_metrics(sk);
	tcp_done(sk);
}

/* Kill off TIME_WAIT sockets once their lifetime has expired. */
static int tcp_tw_death_row_slot;

static void tcp_twkill(unsigned long);

/* TIME_WAIT reaping mechanism. */
#define TCP_TWKILL_SLOTS	8	/* Please keep this a power of 2. */
#define TCP_TWKILL_PERIOD	(TCP_TIMEWAIT_LEN/TCP_TWKILL_SLOTS)

#define TCP_TWKILL_QUOTA	100

static struct hlist_head tcp_tw_death_row[TCP_TWKILL_SLOTS];
static spinlock_t tw_death_lock = SPIN_LOCK_UNLOCKED;
static struct timer_list tcp_tw_timer = TIMER_INITIALIZER(tcp_twkill, 0, 0);
static void twkill_work(void *);
static DECLARE_WORK(tcp_twkill_work, twkill_work, NULL);
static u32 twkill_thread_slots;

/* Returns non-zero if quota exceeded.  */
static int tcp_do_twkill_work(int slot, unsigned int quota)
{
	struct tcp_tw_bucket *tw;
	struct hlist_node *node;
	unsigned int killed;
	int ret;

	/* NOTE: compare this to previous version where lock
	 * was released after detaching chain. It was racy,
	 * because tw buckets are scheduled in not serialized context
	 * in 2.3 (with netfilter), and with softnet it is common, because
	 * soft irqs are not sequenced.
	 */
	killed = 0;
	ret = 0;
rescan:
	tw_for_each_inmate(tw, node, &tcp_tw_death_row[slot]) {
		__tw_del_dead_node(tw);
		spin_unlock(&tw_death_lock);
		tcp_timewait_kill(tw);
		tcp_tw_put(tw);
		killed++;
		spin_lock(&tw_death_lock);
		if (killed > quota) {
			ret = 1;
			break;
		}

		/* While we dropped tw_death_lock, another cpu may have
		 * killed off the next TW bucket in the list, therefore
		 * do a fresh re-read of the hlist head node with the
		 * lock reacquired.  We still use the hlist traversal
		 * macro in order to get the prefetches.
		 */
		goto rescan;
	}

	tcp_tw_count -= killed;
	NET_ADD_STATS_BH(LINUX_MIB_TIMEWAITED, killed);

	return ret;
}

static void tcp_twkill(unsigned long dummy)
{
	int need_timer, ret;

	spin_lock(&tw_death_lock);

	if (tcp_tw_count == 0)
		goto out;

	need_timer = 0;
	ret = tcp_do_twkill_work(tcp_tw_death_row_slot, TCP_TWKILL_QUOTA);
	if (ret) {
		twkill_thread_slots |= (1 << tcp_tw_death_row_slot);
		mb();
		schedule_work(&tcp_twkill_work);
		need_timer = 1;
	} else {
		/* We purged the entire slot, anything left?  */
		if (tcp_tw_count)
			need_timer = 1;
	}
	tcp_tw_death_row_slot =
		((tcp_tw_death_row_slot + 1) & (TCP_TWKILL_SLOTS - 1));
	if (need_timer)
		mod_timer(&tcp_tw_timer, jiffies + TCP_TWKILL_PERIOD);
out:
	spin_unlock(&tw_death_lock);
}

extern void twkill_slots_invalid(void);

static void twkill_work(void *dummy)
{
	int i;

	if ((TCP_TWKILL_SLOTS - 1) > (sizeof(twkill_thread_slots) * 8))
		twkill_slots_invalid();

	while (twkill_thread_slots) {
		spin_lock_bh(&tw_death_lock);
		for (i = 0; i < TCP_TWKILL_SLOTS; i++) {
			if (!(twkill_thread_slots & (1 << i)))
				continue;

			while (tcp_do_twkill_work(i, TCP_TWKILL_QUOTA) != 0) {
				if (need_resched()) {
					spin_unlock_bh(&tw_death_lock);
					schedule();
					spin_lock_bh(&tw_death_lock);
				}
			}

			twkill_thread_slots &= ~(1 << i);
		}
		spin_unlock_bh(&tw_death_lock);
	}
}

/* These are always called from BH context.  See callers in
 * tcp_input.c to verify this.
 */

/* This is for handling early-kills of TIME_WAIT sockets. */
void tcp_tw_deschedule(struct tcp_tw_bucket *tw)
{
	spin_lock(&tw_death_lock);
	if (tw_del_dead_node(tw)) {
		tcp_tw_put(tw);
		if (--tcp_tw_count == 0)
			del_timer(&tcp_tw_timer);
	}
	spin_unlock(&tw_death_lock);
	tcp_timewait_kill(tw);
}

/* Short-time timewait calendar */

static int tcp_twcal_hand = -1;
static int tcp_twcal_jiffie;
static void tcp_twcal_tick(unsigned long);
static struct timer_list tcp_twcal_timer =
		TIMER_INITIALIZER(tcp_twcal_tick, 0, 0);
static struct hlist_head tcp_twcal_row[TCP_TW_RECYCLE_SLOTS];

void tcp_tw_schedule(struct tcp_tw_bucket *tw, int timeo)
{
	struct hlist_head *list;
	int slot;

	/* timeout := RTO * 3.5
	 *
	 * 3.5 = 1+2+0.5 to wait for two retransmits.
	 *
	 * RATIONALE: if FIN arrived and we entered TIME-WAIT state,
	 * our ACK acking that FIN can be lost. If N subsequent retransmitted
	 * FINs (or previous seqments) are lost (probability of such event
	 * is p^(N+1), where p is probability to lose single packet and
	 * time to detect the loss is about RTO*(2^N - 1) with exponential
	 * backoff). Normal timewait length is calculated so, that we
	 * waited at least for one retransmitted FIN (maximal RTO is 120sec).
	 * [ BTW Linux. following BSD, violates this requirement waiting
	 *   only for 60sec, we should wait at least for 240 secs.
	 *   Well, 240 consumes too much of resources 8)
	 * ]
	 * This interval is not reduced to catch old duplicate and
	 * responces to our wandering segments living for two MSLs.
	 * However, if we use PAWS to detect
	 * old duplicates, we can reduce the interval to bounds required
	 * by RTO, rather than MSL. So, if peer understands PAWS, we
	 * kill tw bucket after 3.5*RTO (it is important that this number
	 * is greater than TS tick!) and detect old duplicates with help
	 * of PAWS.
	 */
	slot = (timeo + (1<<TCP_TW_RECYCLE_TICK) - 1) >> TCP_TW_RECYCLE_TICK;

	spin_lock(&tw_death_lock);

	/* Unlink it, if it was scheduled */
	if (tw_del_dead_node(tw))
		tcp_tw_count--;
	else
		atomic_inc(&tw->tw_refcnt);

	if (slot >= TCP_TW_RECYCLE_SLOTS) {
		/* Schedule to slow timer */
		if (timeo >= TCP_TIMEWAIT_LEN) {
			slot = TCP_TWKILL_SLOTS-1;
		} else {
			slot = (timeo + TCP_TWKILL_PERIOD-1) / TCP_TWKILL_PERIOD;
			if (slot >= TCP_TWKILL_SLOTS)
				slot = TCP_TWKILL_SLOTS-1;
		}
		tw->tw_ttd = jiffies + timeo;
		slot = (tcp_tw_death_row_slot + slot) & (TCP_TWKILL_SLOTS - 1);
		list = &tcp_tw_death_row[slot];
	} else {
		tw->tw_ttd = jiffies + (slot << TCP_TW_RECYCLE_TICK);

		if (tcp_twcal_hand < 0) {
			tcp_twcal_hand = 0;
			tcp_twcal_jiffie = jiffies;
			tcp_twcal_timer.expires = tcp_twcal_jiffie + (slot<<TCP_TW_RECYCLE_TICK);
			add_timer(&tcp_twcal_timer);
		} else {
			if (time_after(tcp_twcal_timer.expires, jiffies + (slot<<TCP_TW_RECYCLE_TICK)))
				mod_timer(&tcp_twcal_timer, jiffies + (slot<<TCP_TW_RECYCLE_TICK));
			slot = (tcp_twcal_hand + slot)&(TCP_TW_RECYCLE_SLOTS-1);
		}
		list = &tcp_twcal_row[slot];
	}

	hlist_add_head(&tw->tw_death_node, list);

	if (tcp_tw_count++ == 0)
		mod_timer(&tcp_tw_timer, jiffies+TCP_TWKILL_PERIOD);
	spin_unlock(&tw_death_lock);
}

void tcp_twcal_tick(unsigned long dummy)
{
	int n, slot;
	unsigned long j;
	unsigned long now = jiffies;
	int killed = 0;
	int adv = 0;

	spin_lock(&tw_death_lock);
	if (tcp_twcal_hand < 0)
		goto out;

	slot = tcp_twcal_hand;
	j = tcp_twcal_jiffie;

	for (n=0; n<TCP_TW_RECYCLE_SLOTS; n++) {
		if (time_before_eq(j, now)) {
			struct hlist_node *node, *safe;
			struct tcp_tw_bucket *tw;

			tw_for_each_inmate_safe(tw, node, safe,
					   &tcp_twcal_row[slot]) {
				__tw_del_dead_node(tw);
				tcp_timewait_kill(tw);
				tcp_tw_put(tw);
				killed++;
			}
		} else {
			if (!adv) {
				adv = 1;
				tcp_twcal_jiffie = j;
				tcp_twcal_hand = slot;
			}

			if (!hlist_empty(&tcp_twcal_row[slot])) {
				mod_timer(&tcp_twcal_timer, j);
				goto out;
			}
		}
		j += (1<<TCP_TW_RECYCLE_TICK);
		slot = (slot+1)&(TCP_TW_RECYCLE_SLOTS-1);
	}
	tcp_twcal_hand = -1;

out:
	if ((tcp_tw_count -= killed) == 0)
		del_timer(&tcp_tw_timer);
	NET_ADD_STATS_BH(LINUX_MIB_TIMEWAITKILLED, killed);
	spin_unlock(&tw_death_lock);
}

/* This is not only more efficient than what we used to do, it eliminates
 * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
 *
 * Actually, we could lots of memory writes here. tp of listening
 * socket contains all necessary default parameters.
 */
struct sock *tcp_create_openreq_child(struct sock *sk, struct open_request *req, struct sk_buff *skb)
{
	/* allocate the newsk from the same slab of the master sock,
	 * if not, at sk_free time we'll try to free it from the wrong
	 * slabcache (i.e. is it TCPv4 or v6?) -acme */
	struct sock *newsk = sk_alloc(PF_INET, GFP_ATOMIC, 0, sk->sk_prot->slab);

	if(newsk != NULL) {
		struct tcp_opt *newtp;
		struct sk_filter *filter;

		memcpy(newsk, sk, sizeof(struct tcp_sock));
		newsk->sk_state = TCP_SYN_RECV;

		/* SANITY */
		sk_node_init(&newsk->sk_node);
		tcp_sk(newsk)->bind_hash = NULL;

		/* Clone the TCP header template */
		inet_sk(newsk)->dport = req->rmt_port;

		sock_lock_init(newsk);
		bh_lock_sock(newsk);

		newsk->sk_dst_lock = RW_LOCK_UNLOCKED;
		atomic_set(&newsk->sk_rmem_alloc, 0);
		skb_queue_head_init(&newsk->sk_receive_queue);
		atomic_set(&newsk->sk_wmem_alloc, 0);
		skb_queue_head_init(&newsk->sk_write_queue);
		atomic_set(&newsk->sk_omem_alloc, 0);
		newsk->sk_wmem_queued = 0;
		newsk->sk_forward_alloc = 0;

		sock_reset_flag(newsk, SOCK_DONE);
		newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
		newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
		newsk->sk_send_head = NULL;
		newsk->sk_callback_lock = RW_LOCK_UNLOCKED;