[TIMEWAIT]: Introduce inet_timewait_death_row
authorArnaldo Carvalho de Melo <acme@mandriva.com>
Wed, 10 Aug 2005 03:44:40 +0000 (20:44 -0700)
committerDavid S. Miller <davem@sunset.davemloft.net>
Mon, 29 Aug 2005 22:55:48 +0000 (15:55 -0700)
That groups all of the tables and variables associated to the TCP timewait
schedulling/recycling/killing code, that now can be isolated from the TCP
specific code and used by other transport protocols, such as DCCP.

Next changeset will move this code to net/ipv4/inet_timewait_sock.c

Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/net/inet_timewait_sock.h
include/net/tcp.h
net/ipv4/proc.c
net/ipv4/sysctl_net_ipv4.c
net/ipv4/tcp.c
net/ipv4/tcp_ipv4.c
net/ipv4/tcp_minisocks.c
net/ipv6/tcp_ipv6.c

index e00861b..a7e8052 100644 (file)
 
 #include <linux/ip.h>
 #include <linux/list.h>
+#include <linux/timer.h>
 #include <linux/types.h>
+#include <linux/workqueue.h>
 
 #include <net/sock.h>
 #include <net/tcp_states.h>
 
 #include <asm/atomic.h>
 
+struct inet_hashinfo;
+
+#define INET_TWDR_RECYCLE_SLOTS_LOG    5
+#define INET_TWDR_RECYCLE_SLOTS                (1 << INET_TWDR_RECYCLE_SLOTS_LOG)
+
+/*
+ * If time > 4sec, it is "slow" path, no recycling is required,
+ * so that we select tick to get range about 4 seconds.
+ */
+#if HZ <= 16 || HZ > 4096
+# error Unsupported: HZ <= 16 or HZ > 4096
+#elif HZ <= 32
+# define INET_TWDR_RECYCLE_TICK (5 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
+#elif HZ <= 64
+# define INET_TWDR_RECYCLE_TICK (6 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
+#elif HZ <= 128
+# define INET_TWDR_RECYCLE_TICK (7 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
+#elif HZ <= 256
+# define INET_TWDR_RECYCLE_TICK (8 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
+#elif HZ <= 512
+# define INET_TWDR_RECYCLE_TICK (9 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
+#elif HZ <= 1024
+# define INET_TWDR_RECYCLE_TICK (10 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
+#elif HZ <= 2048
+# define INET_TWDR_RECYCLE_TICK (11 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
+#else
+# define INET_TWDR_RECYCLE_TICK (12 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
+#endif
+
+/* TIME_WAIT reaping mechanism. */
+#define INET_TWDR_TWKILL_SLOTS 8 /* Please keep this a power of 2. */
+
+#define INET_TWDR_TWKILL_QUOTA 100
+
+struct inet_timewait_death_row {
+       /* Short-time timewait calendar */
+       int                     twcal_hand;
+       int                     twcal_jiffie;
+       struct timer_list       twcal_timer;
+       struct hlist_head       twcal_row[INET_TWDR_RECYCLE_SLOTS];
+
+       spinlock_t              death_lock;
+       int                     tw_count;
+       int                     period;
+       u32                     thread_slots;
+       struct work_struct      twkill_work;
+       struct timer_list       tw_timer;
+       int                     slot;
+       struct hlist_head       cells[INET_TWDR_TWKILL_SLOTS];
+       struct inet_hashinfo    *hashinfo;
+       int                     sysctl_tw_recycle;
+       int                     sysctl_max_tw_buckets;
+};
+
 #if (BITS_PER_LONG == 64)
 #define INET_TIMEWAIT_ADDRCMP_ALIGN_BYTES 8
 #else
@@ -33,7 +89,6 @@
 #endif
 
 struct inet_bind_bucket;
-struct inet_hashinfo;
 
 /*
  * This is a TIME_WAIT sock. It works around the memory consumption
index 077db85..4c4cd4f 100644 (file)
@@ -29,6 +29,7 @@
 #include <linux/percpu.h>
 
 #include <net/inet_connection_sock.h>
+#include <net/inet_timewait_sock.h>
 #include <net/inet_hashtables.h>
 #include <net/checksum.h>
 #include <net/request_sock.h>
@@ -42,9 +43,9 @@
 extern struct inet_hashinfo tcp_hashinfo;
 
 extern atomic_t tcp_orphan_count;
-extern int tcp_tw_count;
 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
-extern void tcp_tw_deschedule(struct inet_timewait_sock *tw);
+extern void inet_twsk_deschedule(struct inet_timewait_sock *tw,
+                                struct inet_timewait_death_row *twdr);
 
 #define MAX_TCP_HEADER (128 + MAX_HEADER)
 
@@ -148,33 +149,6 @@ extern void tcp_tw_deschedule(struct inet_timewait_sock *tw);
                                         * timestamps. It must be less than
                                         * minimal timewait lifetime.
                                         */
-
-#define TCP_TW_RECYCLE_SLOTS_LOG       5
-#define TCP_TW_RECYCLE_SLOTS           (1<<TCP_TW_RECYCLE_SLOTS_LOG)
-
-/* If time > 4sec, it is "slow" path, no recycling is required,
-   so that we select tick to get range about 4 seconds.
- */
-
-#if HZ <= 16 || HZ > 4096
-# error Unsupported: HZ <= 16 or HZ > 4096
-#elif HZ <= 32
-# define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG)
-#elif HZ <= 64
-# define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG)
-#elif HZ <= 128
-# define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG)
-#elif HZ <= 256
-# define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG)
-#elif HZ <= 512
-# define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG)
-#elif HZ <= 1024
-# define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG)
-#elif HZ <= 2048
-# define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG)
-#else
-# define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG)
-#endif
 /*
  *     TCP option
  */
@@ -209,12 +183,13 @@ extern void tcp_tw_deschedule(struct inet_timewait_sock *tw);
 #define TCP_NAGLE_CORK         2       /* Socket is corked         */
 #define TCP_NAGLE_PUSH         4       /* Cork is overriden for already queued data */
 
+extern struct inet_timewait_death_row tcp_death_row;
+
 /* sysctl variables for tcp */
 extern int sysctl_tcp_timestamps;
 extern int sysctl_tcp_window_scaling;
 extern int sysctl_tcp_sack;
 extern int sysctl_tcp_fin_timeout;
-extern int sysctl_tcp_tw_recycle;
 extern int sysctl_tcp_keepalive_time;
 extern int sysctl_tcp_keepalive_probes;
 extern int sysctl_tcp_keepalive_intvl;
@@ -229,7 +204,6 @@ extern int sysctl_tcp_stdurg;
 extern int sysctl_tcp_rfc1337;
 extern int sysctl_tcp_abort_on_overflow;
 extern int sysctl_tcp_max_orphans;
-extern int sysctl_tcp_max_tw_buckets;
 extern int sysctl_tcp_fack;
 extern int sysctl_tcp_reordering;
 extern int sysctl_tcp_ecn;
index 912bbcc..3eadbb2 100644 (file)
@@ -65,7 +65,7 @@ static int sockstat_seq_show(struct seq_file *seq, void *v)
        socket_seq_show(seq);
        seq_printf(seq, "TCP: inuse %d orphan %d tw %d alloc %d mem %d\n",
                   fold_prot_inuse(&tcp_prot), atomic_read(&tcp_orphan_count),
-                  tcp_tw_count, atomic_read(&tcp_sockets_allocated),
+                  tcp_death_row.tw_count, atomic_read(&tcp_sockets_allocated),
                   atomic_read(&tcp_memory_allocated));
        seq_printf(seq, "UDP: inuse %d\n", fold_prot_inuse(&udp_prot));
        seq_printf(seq, "RAW: inuse %d\n", fold_prot_inuse(&raw_prot));
index e328945..ce47a34 100644 (file)
@@ -259,7 +259,7 @@ ctl_table ipv4_table[] = {
        {
                .ctl_name       = NET_TCP_MAX_TW_BUCKETS,
                .procname       = "tcp_max_tw_buckets",
-               .data           = &sysctl_tcp_max_tw_buckets,
+               .data           = &tcp_death_row.sysctl_max_tw_buckets,
                .maxlen         = sizeof(int),
                .mode           = 0644,
                .proc_handler   = &proc_dointvec
@@ -363,7 +363,7 @@ ctl_table ipv4_table[] = {
        {
                .ctl_name       = NET_TCP_TW_RECYCLE,
                .procname       = "tcp_tw_recycle",
-               .data           = &sysctl_tcp_tw_recycle,
+               .data           = &tcp_death_row.sysctl_tw_recycle,
                .maxlen         = sizeof(int),
                .mode           = 0644,
                .proc_handler   = &proc_dointvec
index 4bda522..0eed64a 100644 (file)
@@ -2109,12 +2109,12 @@ void __init tcp_init(void)
        if (order >= 4) {
                sysctl_local_port_range[0] = 32768;
                sysctl_local_port_range[1] = 61000;
-               sysctl_tcp_max_tw_buckets = 180000;
+               tcp_death_row.sysctl_max_tw_buckets = 180000;
                sysctl_tcp_max_orphans = 4096 << (order - 4);
                sysctl_max_syn_backlog = 1024;
        } else if (order < 3) {
                sysctl_local_port_range[0] = 1024 * (3 - order);
-               sysctl_tcp_max_tw_buckets >>= (3 - order);
+               tcp_death_row.sysctl_max_tw_buckets >>= (3 - order);
                sysctl_tcp_max_orphans >>= (3 - order);
                sysctl_max_syn_backlog = 128;
        }
index b966102..83f7234 100644 (file)
@@ -199,7 +199,7 @@ unique:
                NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
        } else if (tw) {
                /* Silly. Should hash-dance instead... */
-               tcp_tw_deschedule(tw);
+               inet_twsk_deschedule(tw, &tcp_death_row);
                NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 
                inet_twsk_put(tw);
@@ -291,7 +291,7 @@ ok:
                spin_unlock(&head->lock);
 
                if (tw) {
-                       tcp_tw_deschedule(tw);
+                       inet_twsk_deschedule(tw, &tcp_death_row);;
                        inet_twsk_put(tw);
                }
 
@@ -366,7 +366,7 @@ int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
                tp->write_seq              = 0;
        }
 
-       if (sysctl_tcp_tw_recycle &&
+       if (tcp_death_row.sysctl_tw_recycle &&
            !tp->rx_opt.ts_recent_stamp && rt->rt_dst == daddr) {
                struct inet_peer *peer = rt_get_peer(rt);
 
@@ -965,7 +965,7 @@ int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
                 * are made in the function processing timewait state.
                 */
                if (tmp_opt.saw_tstamp &&
-                   sysctl_tcp_tw_recycle &&
+                   tcp_death_row.sysctl_tw_recycle &&
                    (dst = inet_csk_route_req(sk, req)) != NULL &&
                    (peer = rt_get_peer((struct rtable *)dst)) != NULL &&
                    peer->v4daddr == saddr) {
@@ -1305,7 +1305,8 @@ do_time_wait:
                                                        ntohs(th->dest),
                                                        inet_iif(skb));
                if (sk2) {
-                       tcp_tw_deschedule((struct inet_timewait_sock *)sk);
+                       inet_twsk_deschedule((struct inet_timewait_sock *)sk,
+                                            &tcp_death_row);
                        inet_twsk_put((struct inet_timewait_sock *)sk);
                        sk = sk2;
                        goto process;
index 2d95afe..81b9a52 100644 (file)
 #define SYNC_INIT 1
 #endif
 
-int sysctl_tcp_tw_recycle;
-int sysctl_tcp_max_tw_buckets = NR_FILE*2;
+/* New-style handling of TIME_WAIT sockets. */
+
+static void inet_twdr_hangman(unsigned long data);
+static void inet_twdr_twkill_work(void *data);
+static void inet_twdr_twcal_tick(unsigned long data);
 
 int sysctl_tcp_syncookies = SYNC_INIT; 
 int sysctl_tcp_abort_on_overflow;
 
-static void tcp_tw_schedule(struct inet_timewait_sock *tw, int timeo);
+struct inet_timewait_death_row tcp_death_row = {
+       .sysctl_max_tw_buckets = NR_FILE * 2,
+       .period         = TCP_TIMEWAIT_LEN / INET_TWDR_TWKILL_SLOTS,
+       .death_lock     = SPIN_LOCK_UNLOCKED,
+       .hashinfo       = &tcp_hashinfo,
+       .tw_timer       = TIMER_INITIALIZER(inet_twdr_hangman, 0,
+                                           (unsigned long)&tcp_death_row),
+       .twkill_work    = __WORK_INITIALIZER(tcp_death_row.twkill_work,
+                                            inet_twdr_twkill_work,
+                                            &tcp_death_row),
+/* Short-time timewait calendar */
+
+       .twcal_hand     = -1,
+       .twcal_timer    = TIMER_INITIALIZER(inet_twdr_twcal_tick, 0,
+                                           (unsigned long)&tcp_death_row),
+};
+
+EXPORT_SYMBOL_GPL(tcp_death_row);
+
+static void inet_twsk_schedule(struct inet_timewait_sock *tw,
+                              struct inet_timewait_death_row *twdr,
+                              const int timeo);
 
 static __inline__ int tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win)
 {
@@ -52,10 +76,6 @@ static __inline__ int tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win)
        return (seq == e_win && seq == end_seq);
 }
 
-/* New-style handling of TIME_WAIT sockets. */
-
-int tcp_tw_count;
-
 /* 
  * * Main purpose of TIME-WAIT state is to close connection gracefully,
  *   when one of ends sits in LAST-ACK or CLOSING retransmitting FIN
@@ -132,7 +152,7 @@ tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
                if (!th->fin ||
                    TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1) {
 kill_with_rst:
-                       tcp_tw_deschedule(tw);
+                       inet_twsk_deschedule(tw, &tcp_death_row);
                        inet_twsk_put(tw);
                        return TCP_TW_RST;
                }
@@ -151,11 +171,11 @@ kill_with_rst:
                 * do not undertsnad recycling in any case, it not
                 * a big problem in practice. --ANK */
                if (tw->tw_family == AF_INET &&
-                   sysctl_tcp_tw_recycle && tcptw->tw_ts_recent_stamp &&
+                   tcp_death_row.sysctl_tw_recycle && tcptw->tw_ts_recent_stamp &&
                    tcp_v4_tw_remember_stamp(tw))
-                       tcp_tw_schedule(tw, tw->tw_timeout);
+                       inet_twsk_schedule(tw, &tcp_death_row, tw->tw_timeout);
                else
-                       tcp_tw_schedule(tw, TCP_TIMEWAIT_LEN);
+                       inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN);
                return TCP_TW_ACK;
        }
 
@@ -188,12 +208,12 @@ kill_with_rst:
                         */
                        if (sysctl_tcp_rfc1337 == 0) {
 kill:
-                               tcp_tw_deschedule(tw);
+                               inet_twsk_deschedule(tw, &tcp_death_row);
                                inet_twsk_put(tw);
                                return TCP_TW_SUCCESS;
                        }
                }
-               tcp_tw_schedule(tw, TCP_TIMEWAIT_LEN);
+               inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN);
 
                if (tmp_opt.saw_tstamp) {
                        tcptw->tw_ts_recent       = tmp_opt.rcv_tsval;
@@ -243,7 +263,7 @@ kill:
                 * Do not reschedule in the last case.
                 */
                if (paws_reject || th->ack)
-                       tcp_tw_schedule(tw, TCP_TIMEWAIT_LEN);
+                       inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN);
 
                /* Send ACK. Note, we do not put the bucket,
                 * it will be released by caller.
@@ -263,10 +283,10 @@ void tcp_time_wait(struct sock *sk, int state, int timeo)
        const struct tcp_sock *tp = tcp_sk(sk);
        int recycle_ok = 0;
 
-       if (sysctl_tcp_tw_recycle && tp->rx_opt.ts_recent_stamp)
+       if (tcp_death_row.sysctl_tw_recycle && tp->rx_opt.ts_recent_stamp)
                recycle_ok = tp->af_specific->remember_stamp(sk);
 
-       if (tcp_tw_count < sysctl_tcp_max_tw_buckets)
+       if (tcp_death_row.tw_count < tcp_death_row.sysctl_max_tw_buckets)
                tw = inet_twsk_alloc(sk, state);
 
        if (tw != NULL) {
@@ -306,7 +326,7 @@ void tcp_time_wait(struct sock *sk, int state, int timeo)
                                timeo = TCP_TIMEWAIT_LEN;
                }
 
-               tcp_tw_schedule(tw, timeo);
+               inet_twsk_schedule(tw, &tcp_death_row, timeo);
                inet_twsk_put(tw);
        } else {
                /* Sorry, if we're out of memory, just CLOSE this
@@ -321,26 +341,9 @@ void tcp_time_wait(struct sock *sk, int state, int timeo)
        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 DEFINE_SPINLOCK(tw_death_lock);
-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)
+static int inet_twdr_do_twkill_work(struct inet_timewait_death_row *twdr,
+                                   const int slot)
 {
        struct inet_timewait_sock *tw;
        struct hlist_node *node;
@@ -356,19 +359,19 @@ static int tcp_do_twkill_work(int slot, unsigned int quota)
        killed = 0;
        ret = 0;
 rescan:
-       inet_twsk_for_each_inmate(tw, node, &tcp_tw_death_row[slot]) {
+       inet_twsk_for_each_inmate(tw, node, &twdr->cells[slot]) {
                __inet_twsk_del_dead_node(tw);
-               spin_unlock(&tw_death_lock);
-               __inet_twsk_kill(tw, &tcp_hashinfo);
+               spin_unlock(&twdr->death_lock);
+               __inet_twsk_kill(tw, twdr->hashinfo);
                inet_twsk_put(tw);
                killed++;
-               spin_lock(&tw_death_lock);
-               if (killed > quota) {
+               spin_lock(&twdr->death_lock);
+               if (killed > INET_TWDR_TWKILL_QUOTA) {
                        ret = 1;
                        break;
                }
 
-               /* While we dropped tw_death_lock, another cpu may have
+               /* While we dropped twdr->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
@@ -377,67 +380,68 @@ rescan:
                goto rescan;
        }
 
-       tcp_tw_count -= killed;
+       twdr->tw_count -= killed;
        NET_ADD_STATS_BH(LINUX_MIB_TIMEWAITED, killed);
 
        return ret;
 }
 
-static void tcp_twkill(unsigned long dummy)
+static void inet_twdr_hangman(unsigned long data)
 {
-       int need_timer, ret;
+       struct inet_timewait_death_row *twdr;
+       int unsigned need_timer;
 
-       spin_lock(&tw_death_lock);
+       twdr = (struct inet_timewait_death_row *)data;
+       spin_lock(&twdr->death_lock);
 
-       if (tcp_tw_count == 0)
+       if (twdr->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);
+       if (inet_twdr_do_twkill_work(twdr, twdr->slot)) {
+               twdr->thread_slots |= (1 << twdr->slot);
                mb();
-               schedule_work(&tcp_twkill_work);
+               schedule_work(&twdr->twkill_work);
                need_timer = 1;
        } else {
                /* We purged the entire slot, anything left?  */
-               if (tcp_tw_count)
+               if (twdr->tw_count)
                        need_timer = 1;
        }
-       tcp_tw_death_row_slot =
-               ((tcp_tw_death_row_slot + 1) & (TCP_TWKILL_SLOTS - 1));
+       twdr->slot = ((twdr->slot + 1) & (INET_TWDR_TWKILL_SLOTS - 1));
        if (need_timer)
-               mod_timer(&tcp_tw_timer, jiffies + TCP_TWKILL_PERIOD);
+               mod_timer(&twdr->tw_timer, jiffies + twdr->period);
 out:
-       spin_unlock(&tw_death_lock);
+       spin_unlock(&twdr->death_lock);
 }
 
 extern void twkill_slots_invalid(void);
 
-static void twkill_work(void *dummy)
+static void inet_twdr_twkill_work(void *data)
 {
+       struct inet_timewait_death_row *twdr = data;
        int i;
 
-       if ((TCP_TWKILL_SLOTS - 1) > (sizeof(twkill_thread_slots) * 8))
+       if ((INET_TWDR_TWKILL_SLOTS - 1) > (sizeof(twdr->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)))
+       while (twdr->thread_slots) {
+               spin_lock_bh(&twdr->death_lock);
+               for (i = 0; i < INET_TWDR_TWKILL_SLOTS; i++) {
+                       if (!(twdr->thread_slots & (1 << i)))
                                continue;
 
-                       while (tcp_do_twkill_work(i, TCP_TWKILL_QUOTA) != 0) {
+                       while (inet_twdr_do_twkill_work(twdr, i) != 0) {
                                if (need_resched()) {
-                                       spin_unlock_bh(&tw_death_lock);
+                                       spin_unlock_bh(&twdr->death_lock);
                                        schedule();
-                                       spin_lock_bh(&tw_death_lock);
+                                       spin_lock_bh(&twdr->death_lock);
                                }
                        }
 
-                       twkill_thread_slots &= ~(1 << i);
+                       twdr->thread_slots &= ~(1 << i);
                }
-               spin_unlock_bh(&tw_death_lock);
+               spin_unlock_bh(&twdr->death_lock);
        }
 }
 
@@ -446,28 +450,22 @@ static void twkill_work(void *dummy)
  */
 
 /* This is for handling early-kills of TIME_WAIT sockets. */
-void tcp_tw_deschedule(struct inet_timewait_sock *tw)
+void inet_twsk_deschedule(struct inet_timewait_sock *tw,
+                         struct inet_timewait_death_row *twdr)
 {
-       spin_lock(&tw_death_lock);
+       spin_lock(&twdr->death_lock);
        if (inet_twsk_del_dead_node(tw)) {
                inet_twsk_put(tw);
-               if (--tcp_tw_count == 0)
-                       del_timer(&tcp_tw_timer);
+               if (--twdr->tw_count == 0)
+                       del_timer(&twdr->tw_timer);
        }
-       spin_unlock(&tw_death_lock);
-       __inet_twsk_kill(tw, &tcp_hashinfo);
+       spin_unlock(&twdr->death_lock);
+       __inet_twsk_kill(tw, twdr->hashinfo);
 }
 
-/* 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];
-
-static void tcp_tw_schedule(struct inet_timewait_sock *tw, const int timeo)
+static void inet_twsk_schedule(struct inet_timewait_sock *tw,
+                              struct inet_timewait_death_row *twdr,
+                              const int timeo)
 {
        struct hlist_head *list;
        int slot;
@@ -496,100 +494,106 @@ static void tcp_tw_schedule(struct inet_timewait_sock *tw, const int timeo)
         * 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;
+       slot = (timeo + (1 << INET_TWDR_RECYCLE_TICK) - 1) >> INET_TWDR_RECYCLE_TICK;
 
-       spin_lock(&tw_death_lock);
+       spin_lock(&twdr->death_lock);
 
        /* Unlink it, if it was scheduled */
        if (inet_twsk_del_dead_node(tw))
-               tcp_tw_count--;
+               twdr->tw_count--;
        else
                atomic_inc(&tw->tw_refcnt);
 
-       if (slot >= TCP_TW_RECYCLE_SLOTS) {
+       if (slot >= INET_TWDR_RECYCLE_SLOTS) {
                /* Schedule to slow timer */
                if (timeo >= TCP_TIMEWAIT_LEN) {
-                       slot = TCP_TWKILL_SLOTS-1;
+                       slot = INET_TWDR_TWKILL_SLOTS - 1;
                } else {
-                       slot = (timeo + TCP_TWKILL_PERIOD-1) / TCP_TWKILL_PERIOD;
-                       if (slot >= TCP_TWKILL_SLOTS)
-                               slot = TCP_TWKILL_SLOTS-1;
+                       slot = (timeo + twdr->period - 1) / twdr->period;
+                       if (slot >= INET_TWDR_TWKILL_SLOTS)
+                               slot = INET_TWDR_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];
+               slot = (twdr->slot + slot) & (INET_TWDR_TWKILL_SLOTS - 1);
+               list = &twdr->cells[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);
+               tw->tw_ttd = jiffies + (slot << INET_TWDR_RECYCLE_TICK);
+
+               if (twdr->twcal_hand < 0) {
+                       twdr->twcal_hand = 0;
+                       twdr->twcal_jiffie = jiffies;
+                       twdr->twcal_timer.expires = twdr->twcal_jiffie +
+                                             (slot << INET_TWDR_RECYCLE_TICK);
+                       add_timer(&twdr->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);
+                       if (time_after(twdr->twcal_timer.expires,
+                                      jiffies + (slot << INET_TWDR_RECYCLE_TICK)))
+                               mod_timer(&twdr->twcal_timer,
+                                         jiffies + (slot << INET_TWDR_RECYCLE_TICK));
+                       slot = (twdr->twcal_hand + slot) & (INET_TWDR_RECYCLE_SLOTS - 1);
                }
-               list = &tcp_twcal_row[slot];
+               list = &twdr->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);
+       if (twdr->tw_count++ == 0)
+               mod_timer(&twdr->tw_timer, jiffies + twdr->period);
+       spin_unlock(&twdr->death_lock);
 }
 
-void tcp_twcal_tick(unsigned long dummy)
+void inet_twdr_twcal_tick(unsigned long data)
 {
+       struct inet_timewait_death_row *twdr;
        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)
+       twdr = (struct inet_timewait_death_row *)data;
+
+       spin_lock(&twdr->death_lock);
+       if (twdr->twcal_hand < 0)
                goto out;
 
-       slot = tcp_twcal_hand;
-       j = tcp_twcal_jiffie;
+       slot = twdr->twcal_hand;
+       j = twdr->twcal_jiffie;
 
-       for (n=0; n<TCP_TW_RECYCLE_SLOTS; n++) {
+       for (n = 0; n < INET_TWDR_RECYCLE_SLOTS; n++) {
                if (time_before_eq(j, now)) {
                        struct hlist_node *node, *safe;
                        struct inet_timewait_sock *tw;
 
                        inet_twsk_for_each_inmate_safe(tw, node, safe,
-                                                      &tcp_twcal_row[slot]) {
+                                                      &twdr->twcal_row[slot]) {
                                __inet_twsk_del_dead_node(tw);
-                               __inet_twsk_kill(tw, &tcp_hashinfo);
+                               __inet_twsk_kill(tw, twdr->hashinfo);
                                inet_twsk_put(tw);
                                killed++;
                        }
                } else {
                        if (!adv) {
                                adv = 1;
-                               tcp_twcal_jiffie = j;
-                               tcp_twcal_hand = slot;
+                               twdr->twcal_jiffie = j;
+                               twdr->twcal_hand = slot;
                        }
 
-                       if (!hlist_empty(&tcp_twcal_row[slot])) {
-                               mod_timer(&tcp_twcal_timer, j);
+                       if (!hlist_empty(&twdr->twcal_row[slot])) {
+                               mod_timer(&twdr->twcal_timer, j);
                                goto out;
                        }
                }
-               j += (1<<TCP_TW_RECYCLE_TICK);
-               slot = (slot+1)&(TCP_TW_RECYCLE_SLOTS-1);
+               j += 1 << INET_TWDR_RECYCLE_TICK;
+               slot = (slot + 1) & (INET_TWDR_RECYCLE_SLOTS - 1);
        }
-       tcp_twcal_hand = -1;
+       twdr->twcal_hand = -1;
 
 out:
-       if ((tcp_tw_count -= killed) == 0)
-               del_timer(&tcp_tw_timer);
+       if ((twdr->tw_count -= killed) == 0)
+               del_timer(&twdr->tw_timer);
        NET_ADD_STATS_BH(LINUX_MIB_TIMEWAITKILLED, killed);
-       spin_unlock(&tw_death_lock);
+       spin_unlock(&twdr->death_lock);
 }
 
 /* This is not only more efficient than what we used to do, it eliminates
@@ -929,4 +933,4 @@ EXPORT_SYMBOL(tcp_check_req);
 EXPORT_SYMBOL(tcp_child_process);
 EXPORT_SYMBOL(tcp_create_openreq_child);
 EXPORT_SYMBOL(tcp_timewait_state_process);
-EXPORT_SYMBOL(tcp_tw_deschedule);
+EXPORT_SYMBOL(inet_twsk_deschedule);
index 0b51ec3..1c21ad6 100644 (file)
@@ -521,7 +521,7 @@ unique:
                NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
        } else if (tw) {
                /* Silly. Should hash-dance instead... */
-               tcp_tw_deschedule(tw);
+               inet_twsk_deschedule(tw, &tcp_death_row);
                NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 
                inet_twsk_put(tw);
@@ -611,7 +611,7 @@ ok:
                spin_unlock(&head->lock);
 
                if (tw) {
-                       tcp_tw_deschedule(tw);
+                       inet_twsk_deschedule(tw, &tcp_death_row);
                        inet_twsk_put(tw);
                }
 
@@ -1820,8 +1820,9 @@ do_time_wait:
 
                sk2 = tcp_v6_lookup_listener(&skb->nh.ipv6h->daddr, ntohs(th->dest), tcp_v6_iif(skb));
                if (sk2 != NULL) {
-                       tcp_tw_deschedule((struct inet_timewait_sock *)sk);
-                       inet_twsk_put((struct inet_timewait_sock *)sk);
+                       struct inet_timewait_sock *tw = inet_twsk(sk);
+                       inet_twsk_deschedule(tw, &tcp_death_row);
+                       inet_twsk_put(tw);
                        sk = sk2;
                        goto process;
                }