TCPCT part 1d: define TCP cookie option, extend existing struct's
[safe/jmp/linux-2.6] / net / ipv4 / tcp_minisocks.c
index 36a8fbd..96852af 100644 (file)
@@ -5,8 +5,6 @@
  *
  *             Implementation of the Transmission Control Protocol(TCP).
  *
- * Version:    $Id: tcp_minisocks.c,v 1.15 2002/02/01 22:01:04 davem Exp $
- *
  * Authors:    Ross Biro
  *             Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  *             Mark Evans, <evansmp@uhura.aston.ac.uk>
 #include <net/inet_common.h>
 #include <net/xfrm.h>
 
-#ifdef CONFIG_SYSCTL
-#define SYNC_INIT 0 /* let the user enable it */
-#else
-#define SYNC_INIT 1
-#endif
+int sysctl_tcp_syncookies __read_mostly = 1;
+EXPORT_SYMBOL(sysctl_tcp_syncookies);
 
-int sysctl_tcp_syncookies __read_mostly = SYNC_INIT;
 int sysctl_tcp_abort_on_overflow __read_mostly;
 
 struct inet_timewait_death_row tcp_death_row = {
@@ -100,14 +94,14 @@ tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
        struct tcp_options_received tmp_opt;
        int paws_reject = 0;
 
-       tmp_opt.saw_tstamp = 0;
        if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) {
-               tcp_parse_options(skb, &tmp_opt, 0);
+               tmp_opt.tstamp_ok = 1;
+               tcp_parse_options(skb, &tmp_opt, 1, NULL);
 
                if (tmp_opt.saw_tstamp) {
                        tmp_opt.ts_recent       = tcptw->tw_ts_recent;
                        tmp_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
-                       paws_reject = tcp_paws_check(&tmp_opt, th->rst);
+                       paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
                }
        }
 
@@ -128,7 +122,8 @@ tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
                        goto kill_with_rst;
 
                /* Dup ACK? */
-               if (!after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) ||
+               if (!th->ack ||
+                   !after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) ||
                    TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) {
                        inet_twsk_put(tw);
                        return TCP_TW_SUCCESS;
@@ -244,7 +239,7 @@ kill:
        }
 
        if (paws_reject)
-               NET_INC_STATS_BH(LINUX_MIB_PAWSESTABREJECTED);
+               NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED);
 
        if (!th->rst) {
                /* In this case we must reset the TIMEWAIT timer.
@@ -321,7 +316,7 @@ void tcp_time_wait(struct sock *sk, int state, int timeo)
                        if (key != NULL) {
                                memcpy(&tcptw->tw_md5_key, key->key, key->keylen);
                                tcptw->tw_md5_keylen = key->keylen;
-                               if (tcp_alloc_md5sig_pool() == NULL)
+                               if (tcp_alloc_md5sig_pool(sk) == NULL)
                                        BUG();
                        }
                } while (0);
@@ -362,7 +357,7 @@ void tcp_twsk_destructor(struct sock *sk)
 #ifdef CONFIG_TCP_MD5SIG
        struct tcp_timewait_sock *twsk = tcp_twsk(sk);
        if (twsk->tw_md5_keylen)
-               tcp_put_md5sig_pool();
+               tcp_free_md5sig_pool();
 #endif
 }
 
@@ -388,28 +383,57 @@ struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req,
                const struct inet_request_sock *ireq = inet_rsk(req);
                struct tcp_request_sock *treq = tcp_rsk(req);
                struct inet_connection_sock *newicsk = inet_csk(newsk);
-               struct tcp_sock *newtp;
+               struct tcp_sock *newtp = tcp_sk(newsk);
+               struct tcp_sock *oldtp = tcp_sk(sk);
+               struct tcp_cookie_values *oldcvp = oldtp->cookie_values;
+
+               /* TCP Cookie Transactions require space for the cookie pair,
+                * as it differs for each connection.  There is no need to
+                * copy any s_data_payload stored at the original socket.
+                * Failure will prevent resuming the connection.
+                *
+                * Presumed copied, in order of appearance:
+                *      cookie_in_always, cookie_out_never
+                */
+               if (oldcvp != NULL) {
+                       struct tcp_cookie_values *newcvp =
+                               kzalloc(sizeof(*newtp->cookie_values),
+                                       GFP_ATOMIC);
+
+                       if (newcvp != NULL) {
+                               kref_init(&newcvp->kref);
+                               newcvp->cookie_desired =
+                                               oldcvp->cookie_desired;
+                               newtp->cookie_values = newcvp;
+                       } else {
+                               /* Not Yet Implemented */
+                               newtp->cookie_values = NULL;
+                       }
+               }
 
                /* Now setup tcp_sock */
-               newtp = tcp_sk(newsk);
                newtp->pred_flags = 0;
-               newtp->rcv_wup = newtp->copied_seq = newtp->rcv_nxt = treq->rcv_isn + 1;
-               newtp->snd_sml = newtp->snd_una = newtp->snd_nxt = treq->snt_isn + 1;
+
+               newtp->rcv_wup = newtp->copied_seq =
+               newtp->rcv_nxt = treq->rcv_isn + 1;
+
+               newtp->snd_sml = newtp->snd_una =
+               newtp->snd_nxt = newtp->snd_up =
+                       treq->snt_isn + 1 + tcp_s_data_size(oldtp);
 
                tcp_prequeue_init(newtp);
 
-               tcp_init_wl(newtp, treq->snt_isn, treq->rcv_isn);
+               tcp_init_wl(newtp, treq->rcv_isn);
 
                newtp->srtt = 0;
                newtp->mdev = TCP_TIMEOUT_INIT;
                newicsk->icsk_rto = TCP_TIMEOUT_INIT;
 
                newtp->packets_out = 0;
-               newtp->left_out = 0;
                newtp->retrans_out = 0;
                newtp->sacked_out = 0;
                newtp->fackets_out = 0;
-               newtp->snd_ssthresh = 0x7fffffff;
+               newtp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
 
                /* So many TCP implementations out there (incorrectly) count the
                 * initial SYN frame in their delayed-ACK and congestion control
@@ -428,15 +452,14 @@ struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req,
                tcp_set_ca_state(newsk, TCP_CA_Open);
                tcp_init_xmit_timers(newsk);
                skb_queue_head_init(&newtp->out_of_order_queue);
-               newtp->write_seq = treq->snt_isn + 1;
-               newtp->pushed_seq = newtp->write_seq;
+               newtp->write_seq = newtp->pushed_seq =
+                       treq->snt_isn + 1 + tcp_s_data_size(oldtp);
 
                newtp->rx_opt.saw_tstamp = 0;
 
                newtp->rx_opt.dsack = 0;
-               newtp->rx_opt.eff_sacks = 0;
-
                newtp->rx_opt.num_sacks = 0;
+
                newtp->urg_data = 0;
 
                if (sock_flag(newsk, SOCK_KEEPOPEN))
@@ -446,7 +469,7 @@ struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req,
                newtp->rx_opt.tstamp_ok = ireq->tstamp_ok;
                if ((newtp->rx_opt.sack_ok = ireq->sack_ok) != 0) {
                        if (sysctl_tcp_fack)
-                               newtp->rx_opt.sack_ok |= 2;
+                               tcp_enable_fack(newtp);
                }
                newtp->window_clamp = req->window_clamp;
                newtp->rcv_ssthresh = req->rcv_wnd;
@@ -476,12 +499,12 @@ struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req,
                if (newtp->af_specific->md5_lookup(sk, newsk))
                        newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
 #endif
-               if (skb->len >= TCP_MIN_RCVMSS+newtp->tcp_header_len)
+               if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len)
                        newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
                newtp->rx_opt.mss_clamp = req->mss;
                TCP_ECN_openreq_child(newtp, req);
 
-               TCP_INC_STATS_BH(TCP_MIB_PASSIVEOPENS);
+               TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_PASSIVEOPENS);
        }
        return newsk;
 }
@@ -491,7 +514,7 @@ struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req,
  *     as a request_sock.
  */
 
-struct sock *tcp_check_req(struct sock *sk,struct sk_buff *skb,
+struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
                           struct request_sock *req,
                           struct request_sock **prev)
 {
@@ -501,9 +524,9 @@ struct sock *tcp_check_req(struct sock *sk,struct sk_buff *skb,
        struct tcp_options_received tmp_opt;
        struct sock *child;
 
-       tmp_opt.saw_tstamp = 0;
-       if (th->doff > (sizeof(struct tcphdr)>>2)) {
-               tcp_parse_options(skb, &tmp_opt, 0);
+       if ((th->doff > (sizeof(struct tcphdr)>>2)) && (req->ts_recent)) {
+               tmp_opt.tstamp_ok = 1;
+               tcp_parse_options(skb, &tmp_opt, 1, NULL);
 
                if (tmp_opt.saw_tstamp) {
                        tmp_opt.ts_recent = req->ts_recent;
@@ -512,7 +535,7 @@ struct sock *tcp_check_req(struct sock *sk,struct sk_buff *skb,
                         * from another data.
                         */
                        tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->retrans);
-                       paws_reject = tcp_paws_check(&tmp_opt, th->rst);
+                       paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
                }
        }
 
@@ -596,7 +619,8 @@ struct sock *tcp_check_req(struct sock *sk,struct sk_buff *skb,
         * Invalid ACK: reset will be sent by listening socket
         */
        if ((flg & TCP_FLAG_ACK) &&
-           (TCP_SKB_CB(skb)->ack_seq != tcp_rsk(req)->snt_isn + 1))
+           (TCP_SKB_CB(skb)->ack_seq !=
+            tcp_rsk(req)->snt_isn + 1 + tcp_s_data_size(tcp_sk(sk))))
                return sk;
 
        /* Also, it would be not so bad idea to check rcv_tsecr, which
@@ -610,98 +634,73 @@ struct sock *tcp_check_req(struct sock *sk,struct sk_buff *skb,
                                          tcp_rsk(req)->rcv_isn + 1, tcp_rsk(req)->rcv_isn + 1 + req->rcv_wnd)) {
                /* Out of window: send ACK and drop. */
                if (!(flg & TCP_FLAG_RST))
-                       req->rsk_ops->send_ack(skb, req);
+                       req->rsk_ops->send_ack(sk, skb, req);
                if (paws_reject)
-                       NET_INC_STATS_BH(LINUX_MIB_PAWSESTABREJECTED);
+                       NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED);
                return NULL;
        }
 
        /* In sequence, PAWS is OK. */
 
        if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_isn + 1))
-                       req->ts_recent = tmp_opt.rcv_tsval;
-
-               if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
-                       /* Truncate SYN, it is out of window starting
-                          at tcp_rsk(req)->rcv_isn + 1. */
-                       flg &= ~TCP_FLAG_SYN;
-               }
+               req->ts_recent = tmp_opt.rcv_tsval;
 
-               /* RFC793: "second check the RST bit" and
-                *         "fourth, check the SYN bit"
-                */
-               if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) {
-                       TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
-                       goto embryonic_reset;
-               }
+       if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
+               /* Truncate SYN, it is out of window starting
+                  at tcp_rsk(req)->rcv_isn + 1. */
+               flg &= ~TCP_FLAG_SYN;
+       }
 
-               /* ACK sequence verified above, just make sure ACK is
-                * set.  If ACK not set, just silently drop the packet.
-                */
-               if (!(flg & TCP_FLAG_ACK))
-                       return NULL;
-
-               /* If TCP_DEFER_ACCEPT is set, drop bare ACK. */
-               if (inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
-                   TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
-                       inet_rsk(req)->acked = 1;
-                       return NULL;
-               }
+       /* RFC793: "second check the RST bit" and
+        *         "fourth, check the SYN bit"
+        */
+       if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) {
+               TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
+               goto embryonic_reset;
+       }
 
-               /* OK, ACK is valid, create big socket and
-                * feed this segment to it. It will repeat all
-                * the tests. THIS SEGMENT MUST MOVE SOCKET TO
-                * ESTABLISHED STATE. If it will be dropped after
-                * socket is created, wait for troubles.
-                */
-               child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb,
-                                                                req, NULL);
-               if (child == NULL)
-                       goto listen_overflow;
-#ifdef CONFIG_TCP_MD5SIG
-               else {
-                       /* Copy over the MD5 key from the original socket */
-                       struct tcp_md5sig_key *key;
-                       struct tcp_sock *tp = tcp_sk(sk);
-                       key = tp->af_specific->md5_lookup(sk, child);
-                       if (key != NULL) {
-                               /*
-                                * We're using one, so create a matching key on the
-                                * newsk structure. If we fail to get memory then we
-                                * end up not copying the key across. Shucks.
-                                */
-                               char *newkey = kmemdup(key->key, key->keylen,
-                                                      GFP_ATOMIC);
-                               if (newkey) {
-                                       if (!tcp_alloc_md5sig_pool())
-                                               BUG();
-                                       tp->af_specific->md5_add(child, child,
-                                                                newkey,
-                                                                key->keylen);
-                               }
-                       }
-               }
-#endif
+       /* ACK sequence verified above, just make sure ACK is
+        * set.  If ACK not set, just silently drop the packet.
+        */
+       if (!(flg & TCP_FLAG_ACK))
+               return NULL;
 
-               inet_csk_reqsk_queue_unlink(sk, req, prev);
-               inet_csk_reqsk_queue_removed(sk, req);
+       /* While TCP_DEFER_ACCEPT is active, drop bare ACK. */
+       if (req->retrans < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
+           TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
+               inet_rsk(req)->acked = 1;
+               return NULL;
+       }
 
-               inet_csk_reqsk_queue_add(sk, req, child);
-               return child;
+       /* OK, ACK is valid, create big socket and
+        * feed this segment to it. It will repeat all
+        * the tests. THIS SEGMENT MUST MOVE SOCKET TO
+        * ESTABLISHED STATE. If it will be dropped after
+        * socket is created, wait for troubles.
+        */
+       child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
+       if (child == NULL)
+               goto listen_overflow;
 
-       listen_overflow:
-               if (!sysctl_tcp_abort_on_overflow) {
-                       inet_rsk(req)->acked = 1;
-                       return NULL;
-               }
+       inet_csk_reqsk_queue_unlink(sk, req, prev);
+       inet_csk_reqsk_queue_removed(sk, req);
 
-       embryonic_reset:
-               NET_INC_STATS_BH(LINUX_MIB_EMBRYONICRSTS);
-               if (!(flg & TCP_FLAG_RST))
-                       req->rsk_ops->send_reset(sk, skb);
+       inet_csk_reqsk_queue_add(sk, req, child);
+       return child;
 
-               inet_csk_reqsk_queue_drop(sk, req, prev);
+listen_overflow:
+       if (!sysctl_tcp_abort_on_overflow) {
+               inet_rsk(req)->acked = 1;
                return NULL;
+       }
+
+embryonic_reset:
+       NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
+       if (!(flg & TCP_FLAG_RST))
+               req->rsk_ops->send_reset(sk, skb);
+
+       inet_csk_reqsk_queue_drop(sk, req, prev);
+       return NULL;
 }
 
 /*