Doug / smarc-fsl-linux-kernel

Download as
Browse Code ยป

Commit 6d01a026b7d3009a418326bdcf313503a314f1ea

Authored by Willy Tarreau
Committed by David S. Miller
1 parent e3eef534c5
Exists in master and in 7 other branches imx_3.0.35_4.1.0, imx_3.10.17_1.0.1_ga, imx_3.10.53_1.1.0_ga, imx_3.14.28_1.0.0_ga, smarc-imx_3.10.53_1.1.0_ga, smarc-imx_3.14.28_1.0.0_ga, smarc-l5.0.0_1.0.0-ga

tcp: fix tcp_defer_accept to consider the timeout 

I was trying to use TCP_DEFER_ACCEPT and noticed that if the
client does not talk, the connection is never accepted and
remains in SYN_RECV state until the retransmits expire, where
it finally is deleted. This is bad when some firewall such as
netfilter sits between the client and the server because the
firewall sees the connection in ESTABLISHED state while the
server will finally silently drop it without sending an RST.

This behaviour contradicts the man page which says it should
wait only for some time :

       TCP_DEFER_ACCEPT (since Linux 2.4)
          Allows a listener to be awakened only when data arrives
          on the socket.  Takes an integer value  (seconds), this
          can  bound  the  maximum  number  of attempts TCP will
          make to complete the connection. This option should not
          be used in code intended to be portable.

Also, looking at ipv4/tcp.c, a retransmit counter is correctly
computed :

        case TCP_DEFER_ACCEPT:
                icsk->icsk_accept_queue.rskq_defer_accept = 0;
                if (val > 0) {
                        /* Translate value in seconds to number of
                         * retransmits */
                        while (icsk->icsk_accept_queue.rskq_defer_accept < 32 &&
                               val > ((TCP_TIMEOUT_INIT / HZ) <<
                                       icsk->icsk_accept_queue.rskq_defer_accept))
                                icsk->icsk_accept_queue.rskq_defer_accept++;
                        icsk->icsk_accept_queue.rskq_defer_accept++;
                }
                break;

==> rskq_defer_accept is used as a counter of retransmits.

But in tcp_minisocks.c, this counter is only checked. And in
fact, I have found no location which updates it. So I think
that what was intended was to decrease it in tcp_minisocks
whenever it is checked, which the trivial patch below does.

Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: David S. Miller <davem@davemloft.net>

Showing 1 changed file with 1 additions and 0 deletions Inline Diff

net/ipv4/tcp_minisocks.c
Diff comments   View file @ 6d01a02
1 /* 1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX 2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket 3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level. 4 * interface as the means of communication with the user level.
5 * 5 *
6 * Implementation of the Transmission Control Protocol(TCP). 6 * Implementation of the Transmission Control Protocol(TCP).
7 * 7 *
8 * Authors: Ross Biro 8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Mark Evans, <evansmp@uhura.aston.ac.uk> 10 * Mark Evans, <evansmp@uhura.aston.ac.uk>
11 * Corey Minyard <wf-rch!minyard@relay.EU.net> 11 * Corey Minyard <wf-rch!minyard@relay.EU.net>
12 * Florian La Roche, <flla@stud.uni-sb.de> 12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Charles Hedrick, <hedrick@klinzhai.rutgers.edu> 13 * Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
14 * Linus Torvalds, <torvalds@cs.helsinki.fi> 14 * Linus Torvalds, <torvalds@cs.helsinki.fi>
15 * Alan Cox, <gw4pts@gw4pts.ampr.org> 15 * Alan Cox, <gw4pts@gw4pts.ampr.org>
16 * Matthew Dillon, <dillon@apollo.west.oic.com> 16 * Matthew Dillon, <dillon@apollo.west.oic.com>
17 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 17 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
18 * Jorge Cwik, <jorge@laser.satlink.net> 18 * Jorge Cwik, <jorge@laser.satlink.net>
19 */ 19 */
20 20
21 #include <linux/mm.h> 21 #include <linux/mm.h>
22 #include <linux/module.h> 22 #include <linux/module.h>
23 #include <linux/sysctl.h> 23 #include <linux/sysctl.h>
24 #include <linux/workqueue.h> 24 #include <linux/workqueue.h>
25 #include <net/tcp.h> 25 #include <net/tcp.h>
26 #include <net/inet_common.h> 26 #include <net/inet_common.h>
27 #include <net/xfrm.h> 27 #include <net/xfrm.h>
28 28
29 #ifdef CONFIG_SYSCTL 29 #ifdef CONFIG_SYSCTL
30 #define SYNC_INIT 0 /* let the user enable it */ 30 #define SYNC_INIT 0 /* let the user enable it */
31 #else 31 #else
32 #define SYNC_INIT 1 32 #define SYNC_INIT 1
33 #endif 33 #endif
34 34
35 int sysctl_tcp_syncookies __read_mostly = SYNC_INIT; 35 int sysctl_tcp_syncookies __read_mostly = SYNC_INIT;
36 EXPORT_SYMBOL(sysctl_tcp_syncookies); 36 EXPORT_SYMBOL(sysctl_tcp_syncookies);
37 37
38 int sysctl_tcp_abort_on_overflow __read_mostly; 38 int sysctl_tcp_abort_on_overflow __read_mostly;
39 39
40 struct inet_timewait_death_row tcp_death_row = { 40 struct inet_timewait_death_row tcp_death_row = {
41 .sysctl_max_tw_buckets = NR_FILE * 2, 41 .sysctl_max_tw_buckets = NR_FILE * 2,
42 .period = TCP_TIMEWAIT_LEN / INET_TWDR_TWKILL_SLOTS, 42 .period = TCP_TIMEWAIT_LEN / INET_TWDR_TWKILL_SLOTS,
43 .death_lock = __SPIN_LOCK_UNLOCKED(tcp_death_row.death_lock), 43 .death_lock = __SPIN_LOCK_UNLOCKED(tcp_death_row.death_lock),
44 .hashinfo = &tcp_hashinfo, 44 .hashinfo = &tcp_hashinfo,
45 .tw_timer = TIMER_INITIALIZER(inet_twdr_hangman, 0, 45 .tw_timer = TIMER_INITIALIZER(inet_twdr_hangman, 0,
46 (unsigned long)&tcp_death_row), 46 (unsigned long)&tcp_death_row),
47 .twkill_work = __WORK_INITIALIZER(tcp_death_row.twkill_work, 47 .twkill_work = __WORK_INITIALIZER(tcp_death_row.twkill_work,
48 inet_twdr_twkill_work), 48 inet_twdr_twkill_work),
49 /* Short-time timewait calendar */ 49 /* Short-time timewait calendar */
50 50
51 .twcal_hand = -1, 51 .twcal_hand = -1,
52 .twcal_timer = TIMER_INITIALIZER(inet_twdr_twcal_tick, 0, 52 .twcal_timer = TIMER_INITIALIZER(inet_twdr_twcal_tick, 0,
53 (unsigned long)&tcp_death_row), 53 (unsigned long)&tcp_death_row),
54 }; 54 };
55 55
56 EXPORT_SYMBOL_GPL(tcp_death_row); 56 EXPORT_SYMBOL_GPL(tcp_death_row);
57 57
58 static __inline__ int tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win) 58 static __inline__ int tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win)
59 { 59 {
60 if (seq == s_win) 60 if (seq == s_win)
61 return 1; 61 return 1;
62 if (after(end_seq, s_win) && before(seq, e_win)) 62 if (after(end_seq, s_win) && before(seq, e_win))
63 return 1; 63 return 1;
64 return (seq == e_win && seq == end_seq); 64 return (seq == e_win && seq == end_seq);
65 } 65 }
66 66
67 /* 67 /*
68 * * Main purpose of TIME-WAIT state is to close connection gracefully, 68 * * Main purpose of TIME-WAIT state is to close connection gracefully,
69 * when one of ends sits in LAST-ACK or CLOSING retransmitting FIN 69 * when one of ends sits in LAST-ACK or CLOSING retransmitting FIN
70 * (and, probably, tail of data) and one or more our ACKs are lost. 70 * (and, probably, tail of data) and one or more our ACKs are lost.
71 * * What is TIME-WAIT timeout? It is associated with maximal packet 71 * * What is TIME-WAIT timeout? It is associated with maximal packet
72 * lifetime in the internet, which results in wrong conclusion, that 72 * lifetime in the internet, which results in wrong conclusion, that
73 * it is set to catch "old duplicate segments" wandering out of their path. 73 * it is set to catch "old duplicate segments" wandering out of their path.
74 * It is not quite correct. This timeout is calculated so that it exceeds 74 * It is not quite correct. This timeout is calculated so that it exceeds
75 * maximal retransmission timeout enough to allow to lose one (or more) 75 * maximal retransmission timeout enough to allow to lose one (or more)
76 * segments sent by peer and our ACKs. This time may be calculated from RTO. 76 * segments sent by peer and our ACKs. This time may be calculated from RTO.
77 * * When TIME-WAIT socket receives RST, it means that another end 77 * * When TIME-WAIT socket receives RST, it means that another end
78 * finally closed and we are allowed to kill TIME-WAIT too. 78 * finally closed and we are allowed to kill TIME-WAIT too.
79 * * Second purpose of TIME-WAIT is catching old duplicate segments. 79 * * Second purpose of TIME-WAIT is catching old duplicate segments.
80 * Well, certainly it is pure paranoia, but if we load TIME-WAIT 80 * Well, certainly it is pure paranoia, but if we load TIME-WAIT
81 * with this semantics, we MUST NOT kill TIME-WAIT state with RSTs. 81 * with this semantics, we MUST NOT kill TIME-WAIT state with RSTs.
82 * * If we invented some more clever way to catch duplicates 82 * * If we invented some more clever way to catch duplicates
83 * (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs. 83 * (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs.
84 * 84 *
85 * The algorithm below is based on FORMAL INTERPRETATION of RFCs. 85 * The algorithm below is based on FORMAL INTERPRETATION of RFCs.
86 * When you compare it to RFCs, please, read section SEGMENT ARRIVES 86 * When you compare it to RFCs, please, read section SEGMENT ARRIVES
87 * from the very beginning. 87 * from the very beginning.
88 * 88 *
89 * NOTE. With recycling (and later with fin-wait-2) TW bucket 89 * NOTE. With recycling (and later with fin-wait-2) TW bucket
90 * is _not_ stateless. It means, that strictly speaking we must 90 * is _not_ stateless. It means, that strictly speaking we must
91 * spinlock it. I do not want! Well, probability of misbehaviour 91 * spinlock it. I do not want! Well, probability of misbehaviour
92 * is ridiculously low and, seems, we could use some mb() tricks 92 * is ridiculously low and, seems, we could use some mb() tricks
93 * to avoid misread sequence numbers, states etc. --ANK 93 * to avoid misread sequence numbers, states etc. --ANK
94 */ 94 */
95 enum tcp_tw_status 95 enum tcp_tw_status
96 tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb, 96 tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
97 const struct tcphdr *th) 97 const struct tcphdr *th)
98 { 98 {
99 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); 99 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
100 struct tcp_options_received tmp_opt; 100 struct tcp_options_received tmp_opt;
101 int paws_reject = 0; 101 int paws_reject = 0;
102 102
103 tmp_opt.saw_tstamp = 0; 103 tmp_opt.saw_tstamp = 0;
104 if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) { 104 if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) {
105 tcp_parse_options(skb, &tmp_opt, 0); 105 tcp_parse_options(skb, &tmp_opt, 0);
106 106
107 if (tmp_opt.saw_tstamp) { 107 if (tmp_opt.saw_tstamp) {
108 tmp_opt.ts_recent = tcptw->tw_ts_recent; 108 tmp_opt.ts_recent = tcptw->tw_ts_recent;
109 tmp_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp; 109 tmp_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
110 paws_reject = tcp_paws_reject(&tmp_opt, th->rst); 110 paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
111 } 111 }
112 } 112 }
113 113
114 if (tw->tw_substate == TCP_FIN_WAIT2) { 114 if (tw->tw_substate == TCP_FIN_WAIT2) {
115 /* Just repeat all the checks of tcp_rcv_state_process() */ 115 /* Just repeat all the checks of tcp_rcv_state_process() */
116 116
117 /* Out of window, send ACK */ 117 /* Out of window, send ACK */
118 if (paws_reject || 118 if (paws_reject ||
119 !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq, 119 !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
120 tcptw->tw_rcv_nxt, 120 tcptw->tw_rcv_nxt,
121 tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd)) 121 tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd))
122 return TCP_TW_ACK; 122 return TCP_TW_ACK;
123 123
124 if (th->rst) 124 if (th->rst)
125 goto kill; 125 goto kill;
126 126
127 if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt)) 127 if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt))
128 goto kill_with_rst; 128 goto kill_with_rst;
129 129
130 /* Dup ACK? */ 130 /* Dup ACK? */
131 if (!th->ack || 131 if (!th->ack ||
132 !after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) || 132 !after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) ||
133 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) { 133 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) {
134 inet_twsk_put(tw); 134 inet_twsk_put(tw);
135 return TCP_TW_SUCCESS; 135 return TCP_TW_SUCCESS;
136 } 136 }
137 137
138 /* New data or FIN. If new data arrive after half-duplex close, 138 /* New data or FIN. If new data arrive after half-duplex close,
139 * reset. 139 * reset.
140 */ 140 */
141 if (!th->fin || 141 if (!th->fin ||
142 TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1) { 142 TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1) {
143 kill_with_rst: 143 kill_with_rst:
144 inet_twsk_deschedule(tw, &tcp_death_row); 144 inet_twsk_deschedule(tw, &tcp_death_row);
145 inet_twsk_put(tw); 145 inet_twsk_put(tw);
146 return TCP_TW_RST; 146 return TCP_TW_RST;
147 } 147 }
148 148
149 /* FIN arrived, enter true time-wait state. */ 149 /* FIN arrived, enter true time-wait state. */
150 tw->tw_substate = TCP_TIME_WAIT; 150 tw->tw_substate = TCP_TIME_WAIT;
151 tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq; 151 tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq;
152 if (tmp_opt.saw_tstamp) { 152 if (tmp_opt.saw_tstamp) {
153 tcptw->tw_ts_recent_stamp = get_seconds(); 153 tcptw->tw_ts_recent_stamp = get_seconds();
154 tcptw->tw_ts_recent = tmp_opt.rcv_tsval; 154 tcptw->tw_ts_recent = tmp_opt.rcv_tsval;
155 } 155 }
156 156
157 /* I am shamed, but failed to make it more elegant. 157 /* I am shamed, but failed to make it more elegant.
158 * Yes, it is direct reference to IP, which is impossible 158 * Yes, it is direct reference to IP, which is impossible
159 * to generalize to IPv6. Taking into account that IPv6 159 * to generalize to IPv6. Taking into account that IPv6
160 * do not understand recycling in any case, it not 160 * do not understand recycling in any case, it not
161 * a big problem in practice. --ANK */ 161 * a big problem in practice. --ANK */
162 if (tw->tw_family == AF_INET && 162 if (tw->tw_family == AF_INET &&
163 tcp_death_row.sysctl_tw_recycle && tcptw->tw_ts_recent_stamp && 163 tcp_death_row.sysctl_tw_recycle && tcptw->tw_ts_recent_stamp &&
164 tcp_v4_tw_remember_stamp(tw)) 164 tcp_v4_tw_remember_stamp(tw))
165 inet_twsk_schedule(tw, &tcp_death_row, tw->tw_timeout, 165 inet_twsk_schedule(tw, &tcp_death_row, tw->tw_timeout,
166 TCP_TIMEWAIT_LEN); 166 TCP_TIMEWAIT_LEN);
167 else 167 else
168 inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN, 168 inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
169 TCP_TIMEWAIT_LEN); 169 TCP_TIMEWAIT_LEN);
170 return TCP_TW_ACK; 170 return TCP_TW_ACK;
171 } 171 }
172 172
173 /* 173 /*
174 * Now real TIME-WAIT state. 174 * Now real TIME-WAIT state.
175 * 175 *
176 * RFC 1122: 176 * RFC 1122:
177 * "When a connection is [...] on TIME-WAIT state [...] 177 * "When a connection is [...] on TIME-WAIT state [...]
178 * [a TCP] MAY accept a new SYN from the remote TCP to 178 * [a TCP] MAY accept a new SYN from the remote TCP to
179 * reopen the connection directly, if it: 179 * reopen the connection directly, if it:
180 * 180 *
181 * (1) assigns its initial sequence number for the new 181 * (1) assigns its initial sequence number for the new
182 * connection to be larger than the largest sequence 182 * connection to be larger than the largest sequence
183 * number it used on the previous connection incarnation, 183 * number it used on the previous connection incarnation,
184 * and 184 * and
185 * 185 *
186 * (2) returns to TIME-WAIT state if the SYN turns out 186 * (2) returns to TIME-WAIT state if the SYN turns out
187 * to be an old duplicate". 187 * to be an old duplicate".
188 */ 188 */
189 189
190 if (!paws_reject && 190 if (!paws_reject &&
191 (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt && 191 (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt &&
192 (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) { 192 (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) {
193 /* In window segment, it may be only reset or bare ack. */ 193 /* In window segment, it may be only reset or bare ack. */
194 194
195 if (th->rst) { 195 if (th->rst) {
196 /* This is TIME_WAIT assassination, in two flavors. 196 /* This is TIME_WAIT assassination, in two flavors.
197 * Oh well... nobody has a sufficient solution to this 197 * Oh well... nobody has a sufficient solution to this
198 * protocol bug yet. 198 * protocol bug yet.
199 */ 199 */
200 if (sysctl_tcp_rfc1337 == 0) { 200 if (sysctl_tcp_rfc1337 == 0) {
201 kill: 201 kill:
202 inet_twsk_deschedule(tw, &tcp_death_row); 202 inet_twsk_deschedule(tw, &tcp_death_row);
203 inet_twsk_put(tw); 203 inet_twsk_put(tw);
204 return TCP_TW_SUCCESS; 204 return TCP_TW_SUCCESS;
205 } 205 }
206 } 206 }
207 inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN, 207 inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
208 TCP_TIMEWAIT_LEN); 208 TCP_TIMEWAIT_LEN);
209 209
210 if (tmp_opt.saw_tstamp) { 210 if (tmp_opt.saw_tstamp) {
211 tcptw->tw_ts_recent = tmp_opt.rcv_tsval; 211 tcptw->tw_ts_recent = tmp_opt.rcv_tsval;
212 tcptw->tw_ts_recent_stamp = get_seconds(); 212 tcptw->tw_ts_recent_stamp = get_seconds();
213 } 213 }
214 214
215 inet_twsk_put(tw); 215 inet_twsk_put(tw);
216 return TCP_TW_SUCCESS; 216 return TCP_TW_SUCCESS;
217 } 217 }
218 218
219 /* Out of window segment. 219 /* Out of window segment.
220 220
221 All the segments are ACKed immediately. 221 All the segments are ACKed immediately.
222 222
223 The only exception is new SYN. We accept it, if it is 223 The only exception is new SYN. We accept it, if it is
224 not old duplicate and we are not in danger to be killed 224 not old duplicate and we are not in danger to be killed
225 by delayed old duplicates. RFC check is that it has 225 by delayed old duplicates. RFC check is that it has
226 newer sequence number works at rates <40Mbit/sec. 226 newer sequence number works at rates <40Mbit/sec.
227 However, if paws works, it is reliable AND even more, 227 However, if paws works, it is reliable AND even more,
228 we even may relax silly seq space cutoff. 228 we even may relax silly seq space cutoff.
229 229
230 RED-PEN: we violate main RFC requirement, if this SYN will appear 230 RED-PEN: we violate main RFC requirement, if this SYN will appear
231 old duplicate (i.e. we receive RST in reply to SYN-ACK), 231 old duplicate (i.e. we receive RST in reply to SYN-ACK),
232 we must return socket to time-wait state. It is not good, 232 we must return socket to time-wait state. It is not good,
233 but not fatal yet. 233 but not fatal yet.
234 */ 234 */
235 235
236 if (th->syn && !th->rst && !th->ack && !paws_reject && 236 if (th->syn && !th->rst && !th->ack && !paws_reject &&
237 (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) || 237 (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) ||
238 (tmp_opt.saw_tstamp && 238 (tmp_opt.saw_tstamp &&
239 (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) { 239 (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) {
240 u32 isn = tcptw->tw_snd_nxt + 65535 + 2; 240 u32 isn = tcptw->tw_snd_nxt + 65535 + 2;
241 if (isn == 0) 241 if (isn == 0)
242 isn++; 242 isn++;
243 TCP_SKB_CB(skb)->when = isn; 243 TCP_SKB_CB(skb)->when = isn;
244 return TCP_TW_SYN; 244 return TCP_TW_SYN;
245 } 245 }
246 246
247 if (paws_reject) 247 if (paws_reject)
248 NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED); 248 NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED);
249 249
250 if (!th->rst) { 250 if (!th->rst) {
251 /* In this case we must reset the TIMEWAIT timer. 251 /* In this case we must reset the TIMEWAIT timer.
252 * 252 *
253 * If it is ACKless SYN it may be both old duplicate 253 * If it is ACKless SYN it may be both old duplicate
254 * and new good SYN with random sequence number <rcv_nxt. 254 * and new good SYN with random sequence number <rcv_nxt.
255 * Do not reschedule in the last case. 255 * Do not reschedule in the last case.
256 */ 256 */
257 if (paws_reject || th->ack) 257 if (paws_reject || th->ack)
258 inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN, 258 inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
259 TCP_TIMEWAIT_LEN); 259 TCP_TIMEWAIT_LEN);
260 260
261 /* Send ACK. Note, we do not put the bucket, 261 /* Send ACK. Note, we do not put the bucket,
262 * it will be released by caller. 262 * it will be released by caller.
263 */ 263 */
264 return TCP_TW_ACK; 264 return TCP_TW_ACK;
265 } 265 }
266 inet_twsk_put(tw); 266 inet_twsk_put(tw);
267 return TCP_TW_SUCCESS; 267 return TCP_TW_SUCCESS;
268 } 268 }
269 269
270 /* 270 /*
271 * Move a socket to time-wait or dead fin-wait-2 state. 271 * Move a socket to time-wait or dead fin-wait-2 state.
272 */ 272 */
273 void tcp_time_wait(struct sock *sk, int state, int timeo) 273 void tcp_time_wait(struct sock *sk, int state, int timeo)
274 { 274 {
275 struct inet_timewait_sock *tw = NULL; 275 struct inet_timewait_sock *tw = NULL;
276 const struct inet_connection_sock *icsk = inet_csk(sk); 276 const struct inet_connection_sock *icsk = inet_csk(sk);
277 const struct tcp_sock *tp = tcp_sk(sk); 277 const struct tcp_sock *tp = tcp_sk(sk);
278 int recycle_ok = 0; 278 int recycle_ok = 0;
279 279
280 if (tcp_death_row.sysctl_tw_recycle && tp->rx_opt.ts_recent_stamp) 280 if (tcp_death_row.sysctl_tw_recycle && tp->rx_opt.ts_recent_stamp)
281 recycle_ok = icsk->icsk_af_ops->remember_stamp(sk); 281 recycle_ok = icsk->icsk_af_ops->remember_stamp(sk);
282 282
283 if (tcp_death_row.tw_count < tcp_death_row.sysctl_max_tw_buckets) 283 if (tcp_death_row.tw_count < tcp_death_row.sysctl_max_tw_buckets)
284 tw = inet_twsk_alloc(sk, state); 284 tw = inet_twsk_alloc(sk, state);
285 285
286 if (tw != NULL) { 286 if (tw != NULL) {
287 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); 287 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
288 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1); 288 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
289 289
290 tw->tw_rcv_wscale = tp->rx_opt.rcv_wscale; 290 tw->tw_rcv_wscale = tp->rx_opt.rcv_wscale;
291 tcptw->tw_rcv_nxt = tp->rcv_nxt; 291 tcptw->tw_rcv_nxt = tp->rcv_nxt;
292 tcptw->tw_snd_nxt = tp->snd_nxt; 292 tcptw->tw_snd_nxt = tp->snd_nxt;
293 tcptw->tw_rcv_wnd = tcp_receive_window(tp); 293 tcptw->tw_rcv_wnd = tcp_receive_window(tp);
294 tcptw->tw_ts_recent = tp->rx_opt.ts_recent; 294 tcptw->tw_ts_recent = tp->rx_opt.ts_recent;
295 tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp; 295 tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp;
296 296
297 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 297 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
298 if (tw->tw_family == PF_INET6) { 298 if (tw->tw_family == PF_INET6) {
299 struct ipv6_pinfo *np = inet6_sk(sk); 299 struct ipv6_pinfo *np = inet6_sk(sk);
300 struct inet6_timewait_sock *tw6; 300 struct inet6_timewait_sock *tw6;
301 301
302 tw->tw_ipv6_offset = inet6_tw_offset(sk->sk_prot); 302 tw->tw_ipv6_offset = inet6_tw_offset(sk->sk_prot);
303 tw6 = inet6_twsk((struct sock *)tw); 303 tw6 = inet6_twsk((struct sock *)tw);
304 ipv6_addr_copy(&tw6->tw_v6_daddr, &np->daddr); 304 ipv6_addr_copy(&tw6->tw_v6_daddr, &np->daddr);
305 ipv6_addr_copy(&tw6->tw_v6_rcv_saddr, &np->rcv_saddr); 305 ipv6_addr_copy(&tw6->tw_v6_rcv_saddr, &np->rcv_saddr);
306 tw->tw_ipv6only = np->ipv6only; 306 tw->tw_ipv6only = np->ipv6only;
307 } 307 }
308 #endif 308 #endif
309 309
310 #ifdef CONFIG_TCP_MD5SIG 310 #ifdef CONFIG_TCP_MD5SIG
311 /* 311 /*
312 * The timewait bucket does not have the key DB from the 312 * The timewait bucket does not have the key DB from the
313 * sock structure. We just make a quick copy of the 313 * sock structure. We just make a quick copy of the
314 * md5 key being used (if indeed we are using one) 314 * md5 key being used (if indeed we are using one)
315 * so the timewait ack generating code has the key. 315 * so the timewait ack generating code has the key.
316 */ 316 */
317 do { 317 do {
318 struct tcp_md5sig_key *key; 318 struct tcp_md5sig_key *key;
319 memset(tcptw->tw_md5_key, 0, sizeof(tcptw->tw_md5_key)); 319 memset(tcptw->tw_md5_key, 0, sizeof(tcptw->tw_md5_key));
320 tcptw->tw_md5_keylen = 0; 320 tcptw->tw_md5_keylen = 0;
321 key = tp->af_specific->md5_lookup(sk, sk); 321 key = tp->af_specific->md5_lookup(sk, sk);
322 if (key != NULL) { 322 if (key != NULL) {
323 memcpy(&tcptw->tw_md5_key, key->key, key->keylen); 323 memcpy(&tcptw->tw_md5_key, key->key, key->keylen);
324 tcptw->tw_md5_keylen = key->keylen; 324 tcptw->tw_md5_keylen = key->keylen;
325 if (tcp_alloc_md5sig_pool(sk) == NULL) 325 if (tcp_alloc_md5sig_pool(sk) == NULL)
326 BUG(); 326 BUG();
327 } 327 }
328 } while (0); 328 } while (0);
329 #endif 329 #endif
330 330
331 /* Linkage updates. */ 331 /* Linkage updates. */
332 __inet_twsk_hashdance(tw, sk, &tcp_hashinfo); 332 __inet_twsk_hashdance(tw, sk, &tcp_hashinfo);
333 333
334 /* Get the TIME_WAIT timeout firing. */ 334 /* Get the TIME_WAIT timeout firing. */
335 if (timeo < rto) 335 if (timeo < rto)
336 timeo = rto; 336 timeo = rto;
337 337
338 if (recycle_ok) { 338 if (recycle_ok) {
339 tw->tw_timeout = rto; 339 tw->tw_timeout = rto;
340 } else { 340 } else {
341 tw->tw_timeout = TCP_TIMEWAIT_LEN; 341 tw->tw_timeout = TCP_TIMEWAIT_LEN;
342 if (state == TCP_TIME_WAIT) 342 if (state == TCP_TIME_WAIT)
343 timeo = TCP_TIMEWAIT_LEN; 343 timeo = TCP_TIMEWAIT_LEN;
344 } 344 }
345 345
346 inet_twsk_schedule(tw, &tcp_death_row, timeo, 346 inet_twsk_schedule(tw, &tcp_death_row, timeo,
347 TCP_TIMEWAIT_LEN); 347 TCP_TIMEWAIT_LEN);
348 inet_twsk_put(tw); 348 inet_twsk_put(tw);
349 } else { 349 } else {
350 /* Sorry, if we're out of memory, just CLOSE this 350 /* Sorry, if we're out of memory, just CLOSE this
351 * socket up. We've got bigger problems than 351 * socket up. We've got bigger problems than
352 * non-graceful socket closings. 352 * non-graceful socket closings.
353 */ 353 */
354 LIMIT_NETDEBUG(KERN_INFO "TCP: time wait bucket table overflow\n"); 354 LIMIT_NETDEBUG(KERN_INFO "TCP: time wait bucket table overflow\n");
355 } 355 }
356 356
357 tcp_update_metrics(sk); 357 tcp_update_metrics(sk);
358 tcp_done(sk); 358 tcp_done(sk);
359 } 359 }
360 360
361 void tcp_twsk_destructor(struct sock *sk) 361 void tcp_twsk_destructor(struct sock *sk)
362 { 362 {
363 #ifdef CONFIG_TCP_MD5SIG 363 #ifdef CONFIG_TCP_MD5SIG
364 struct tcp_timewait_sock *twsk = tcp_twsk(sk); 364 struct tcp_timewait_sock *twsk = tcp_twsk(sk);
365 if (twsk->tw_md5_keylen) 365 if (twsk->tw_md5_keylen)
366 tcp_free_md5sig_pool(); 366 tcp_free_md5sig_pool();
367 #endif 367 #endif
368 } 368 }
369 369
370 EXPORT_SYMBOL_GPL(tcp_twsk_destructor); 370 EXPORT_SYMBOL_GPL(tcp_twsk_destructor);
371 371
372 static inline void TCP_ECN_openreq_child(struct tcp_sock *tp, 372 static inline void TCP_ECN_openreq_child(struct tcp_sock *tp,
373 struct request_sock *req) 373 struct request_sock *req)
374 { 374 {
375 tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0; 375 tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0;
376 } 376 }
377 377
378 /* This is not only more efficient than what we used to do, it eliminates 378 /* This is not only more efficient than what we used to do, it eliminates
379 * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM 379 * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
380 * 380 *
381 * Actually, we could lots of memory writes here. tp of listening 381 * Actually, we could lots of memory writes here. tp of listening
382 * socket contains all necessary default parameters. 382 * socket contains all necessary default parameters.
383 */ 383 */
384 struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req, struct sk_buff *skb) 384 struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req, struct sk_buff *skb)
385 { 385 {
386 struct sock *newsk = inet_csk_clone(sk, req, GFP_ATOMIC); 386 struct sock *newsk = inet_csk_clone(sk, req, GFP_ATOMIC);
387 387
388 if (newsk != NULL) { 388 if (newsk != NULL) {
389 const struct inet_request_sock *ireq = inet_rsk(req); 389 const struct inet_request_sock *ireq = inet_rsk(req);
390 struct tcp_request_sock *treq = tcp_rsk(req); 390 struct tcp_request_sock *treq = tcp_rsk(req);
391 struct inet_connection_sock *newicsk = inet_csk(newsk); 391 struct inet_connection_sock *newicsk = inet_csk(newsk);
392 struct tcp_sock *newtp; 392 struct tcp_sock *newtp;
393 393
394 /* Now setup tcp_sock */ 394 /* Now setup tcp_sock */
395 newtp = tcp_sk(newsk); 395 newtp = tcp_sk(newsk);
396 newtp->pred_flags = 0; 396 newtp->pred_flags = 0;
397 newtp->rcv_wup = newtp->copied_seq = newtp->rcv_nxt = treq->rcv_isn + 1; 397 newtp->rcv_wup = newtp->copied_seq = newtp->rcv_nxt = treq->rcv_isn + 1;
398 newtp->snd_sml = newtp->snd_una = newtp->snd_nxt = treq->snt_isn + 1; 398 newtp->snd_sml = newtp->snd_una = newtp->snd_nxt = treq->snt_isn + 1;
399 newtp->snd_up = treq->snt_isn + 1; 399 newtp->snd_up = treq->snt_isn + 1;
400 400
401 tcp_prequeue_init(newtp); 401 tcp_prequeue_init(newtp);
402 402
403 tcp_init_wl(newtp, treq->rcv_isn); 403 tcp_init_wl(newtp, treq->rcv_isn);
404 404
405 newtp->srtt = 0; 405 newtp->srtt = 0;
406 newtp->mdev = TCP_TIMEOUT_INIT; 406 newtp->mdev = TCP_TIMEOUT_INIT;
407 newicsk->icsk_rto = TCP_TIMEOUT_INIT; 407 newicsk->icsk_rto = TCP_TIMEOUT_INIT;
408 408
409 newtp->packets_out = 0; 409 newtp->packets_out = 0;
410 newtp->retrans_out = 0; 410 newtp->retrans_out = 0;
411 newtp->sacked_out = 0; 411 newtp->sacked_out = 0;
412 newtp->fackets_out = 0; 412 newtp->fackets_out = 0;
413 newtp->snd_ssthresh = TCP_INFINITE_SSTHRESH; 413 newtp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
414 414
415 /* So many TCP implementations out there (incorrectly) count the 415 /* So many TCP implementations out there (incorrectly) count the
416 * initial SYN frame in their delayed-ACK and congestion control 416 * initial SYN frame in their delayed-ACK and congestion control
417 * algorithms that we must have the following bandaid to talk 417 * algorithms that we must have the following bandaid to talk
418 * efficiently to them. -DaveM 418 * efficiently to them. -DaveM
419 */ 419 */
420 newtp->snd_cwnd = 2; 420 newtp->snd_cwnd = 2;
421 newtp->snd_cwnd_cnt = 0; 421 newtp->snd_cwnd_cnt = 0;
422 newtp->bytes_acked = 0; 422 newtp->bytes_acked = 0;
423 423
424 newtp->frto_counter = 0; 424 newtp->frto_counter = 0;
425 newtp->frto_highmark = 0; 425 newtp->frto_highmark = 0;
426 426
427 newicsk->icsk_ca_ops = &tcp_init_congestion_ops; 427 newicsk->icsk_ca_ops = &tcp_init_congestion_ops;
428 428
429 tcp_set_ca_state(newsk, TCP_CA_Open); 429 tcp_set_ca_state(newsk, TCP_CA_Open);
430 tcp_init_xmit_timers(newsk); 430 tcp_init_xmit_timers(newsk);
431 skb_queue_head_init(&newtp->out_of_order_queue); 431 skb_queue_head_init(&newtp->out_of_order_queue);
432 newtp->write_seq = treq->snt_isn + 1; 432 newtp->write_seq = treq->snt_isn + 1;
433 newtp->pushed_seq = newtp->write_seq; 433 newtp->pushed_seq = newtp->write_seq;
434 434
435 newtp->rx_opt.saw_tstamp = 0; 435 newtp->rx_opt.saw_tstamp = 0;
436 436
437 newtp->rx_opt.dsack = 0; 437 newtp->rx_opt.dsack = 0;
438 newtp->rx_opt.num_sacks = 0; 438 newtp->rx_opt.num_sacks = 0;
439 439
440 newtp->urg_data = 0; 440 newtp->urg_data = 0;
441 441
442 if (sock_flag(newsk, SOCK_KEEPOPEN)) 442 if (sock_flag(newsk, SOCK_KEEPOPEN))
443 inet_csk_reset_keepalive_timer(newsk, 443 inet_csk_reset_keepalive_timer(newsk,
444 keepalive_time_when(newtp)); 444 keepalive_time_when(newtp));
445 445
446 newtp->rx_opt.tstamp_ok = ireq->tstamp_ok; 446 newtp->rx_opt.tstamp_ok = ireq->tstamp_ok;
447 if ((newtp->rx_opt.sack_ok = ireq->sack_ok) != 0) { 447 if ((newtp->rx_opt.sack_ok = ireq->sack_ok) != 0) {
448 if (sysctl_tcp_fack) 448 if (sysctl_tcp_fack)
449 tcp_enable_fack(newtp); 449 tcp_enable_fack(newtp);
450 } 450 }
451 newtp->window_clamp = req->window_clamp; 451 newtp->window_clamp = req->window_clamp;
452 newtp->rcv_ssthresh = req->rcv_wnd; 452 newtp->rcv_ssthresh = req->rcv_wnd;
453 newtp->rcv_wnd = req->rcv_wnd; 453 newtp->rcv_wnd = req->rcv_wnd;
454 newtp->rx_opt.wscale_ok = ireq->wscale_ok; 454 newtp->rx_opt.wscale_ok = ireq->wscale_ok;
455 if (newtp->rx_opt.wscale_ok) { 455 if (newtp->rx_opt.wscale_ok) {
456 newtp->rx_opt.snd_wscale = ireq->snd_wscale; 456 newtp->rx_opt.snd_wscale = ireq->snd_wscale;
457 newtp->rx_opt.rcv_wscale = ireq->rcv_wscale; 457 newtp->rx_opt.rcv_wscale = ireq->rcv_wscale;
458 } else { 458 } else {
459 newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0; 459 newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0;
460 newtp->window_clamp = min(newtp->window_clamp, 65535U); 460 newtp->window_clamp = min(newtp->window_clamp, 65535U);
461 } 461 }
462 newtp->snd_wnd = (ntohs(tcp_hdr(skb)->window) << 462 newtp->snd_wnd = (ntohs(tcp_hdr(skb)->window) <<
463 newtp->rx_opt.snd_wscale); 463 newtp->rx_opt.snd_wscale);
464 newtp->max_window = newtp->snd_wnd; 464 newtp->max_window = newtp->snd_wnd;
465 465
466 if (newtp->rx_opt.tstamp_ok) { 466 if (newtp->rx_opt.tstamp_ok) {
467 newtp->rx_opt.ts_recent = req->ts_recent; 467 newtp->rx_opt.ts_recent = req->ts_recent;
468 newtp->rx_opt.ts_recent_stamp = get_seconds(); 468 newtp->rx_opt.ts_recent_stamp = get_seconds();
469 newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED; 469 newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
470 } else { 470 } else {
471 newtp->rx_opt.ts_recent_stamp = 0; 471 newtp->rx_opt.ts_recent_stamp = 0;
472 newtp->tcp_header_len = sizeof(struct tcphdr); 472 newtp->tcp_header_len = sizeof(struct tcphdr);
473 } 473 }
474 #ifdef CONFIG_TCP_MD5SIG 474 #ifdef CONFIG_TCP_MD5SIG
475 newtp->md5sig_info = NULL; /*XXX*/ 475 newtp->md5sig_info = NULL; /*XXX*/
476 if (newtp->af_specific->md5_lookup(sk, newsk)) 476 if (newtp->af_specific->md5_lookup(sk, newsk))
477 newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED; 477 newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
478 #endif 478 #endif
479 if (skb->len >= TCP_MIN_RCVMSS+newtp->tcp_header_len) 479 if (skb->len >= TCP_MIN_RCVMSS+newtp->tcp_header_len)
480 newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len; 480 newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
481 newtp->rx_opt.mss_clamp = req->mss; 481 newtp->rx_opt.mss_clamp = req->mss;
482 TCP_ECN_openreq_child(newtp, req); 482 TCP_ECN_openreq_child(newtp, req);
483 483
484 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_PASSIVEOPENS); 484 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_PASSIVEOPENS);
485 } 485 }
486 return newsk; 486 return newsk;
487 } 487 }
488 488
489 /* 489 /*
490 * Process an incoming packet for SYN_RECV sockets represented 490 * Process an incoming packet for SYN_RECV sockets represented
491 * as a request_sock. 491 * as a request_sock.
492 */ 492 */
493 493
494 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 494 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
495 struct request_sock *req, 495 struct request_sock *req,
496 struct request_sock **prev) 496 struct request_sock **prev)
497 { 497 {
498 const struct tcphdr *th = tcp_hdr(skb); 498 const struct tcphdr *th = tcp_hdr(skb);
499 __be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK); 499 __be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
500 int paws_reject = 0; 500 int paws_reject = 0;
501 struct tcp_options_received tmp_opt; 501 struct tcp_options_received tmp_opt;
502 struct sock *child; 502 struct sock *child;
503 503
504 tmp_opt.saw_tstamp = 0; 504 tmp_opt.saw_tstamp = 0;
505 if (th->doff > (sizeof(struct tcphdr)>>2)) { 505 if (th->doff > (sizeof(struct tcphdr)>>2)) {
506 tcp_parse_options(skb, &tmp_opt, 0); 506 tcp_parse_options(skb, &tmp_opt, 0);
507 507
508 if (tmp_opt.saw_tstamp) { 508 if (tmp_opt.saw_tstamp) {
509 tmp_opt.ts_recent = req->ts_recent; 509 tmp_opt.ts_recent = req->ts_recent;
510 /* We do not store true stamp, but it is not required, 510 /* We do not store true stamp, but it is not required,
511 * it can be estimated (approximately) 511 * it can be estimated (approximately)
512 * from another data. 512 * from another data.
513 */ 513 */
514 tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->retrans); 514 tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->retrans);
515 paws_reject = tcp_paws_reject(&tmp_opt, th->rst); 515 paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
516 } 516 }
517 } 517 }
518 518
519 /* Check for pure retransmitted SYN. */ 519 /* Check for pure retransmitted SYN. */
520 if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn && 520 if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn &&
521 flg == TCP_FLAG_SYN && 521 flg == TCP_FLAG_SYN &&
522 !paws_reject) { 522 !paws_reject) {
523 /* 523 /*
524 * RFC793 draws (Incorrectly! It was fixed in RFC1122) 524 * RFC793 draws (Incorrectly! It was fixed in RFC1122)
525 * this case on figure 6 and figure 8, but formal 525 * this case on figure 6 and figure 8, but formal
526 * protocol description says NOTHING. 526 * protocol description says NOTHING.
527 * To be more exact, it says that we should send ACK, 527 * To be more exact, it says that we should send ACK,
528 * because this segment (at least, if it has no data) 528 * because this segment (at least, if it has no data)
529 * is out of window. 529 * is out of window.
530 * 530 *
531 * CONCLUSION: RFC793 (even with RFC1122) DOES NOT 531 * CONCLUSION: RFC793 (even with RFC1122) DOES NOT
532 * describe SYN-RECV state. All the description 532 * describe SYN-RECV state. All the description
533 * is wrong, we cannot believe to it and should 533 * is wrong, we cannot believe to it and should
534 * rely only on common sense and implementation 534 * rely only on common sense and implementation
535 * experience. 535 * experience.
536 * 536 *
537 * Enforce "SYN-ACK" according to figure 8, figure 6 537 * Enforce "SYN-ACK" according to figure 8, figure 6
538 * of RFC793, fixed by RFC1122. 538 * of RFC793, fixed by RFC1122.
539 */ 539 */
540 req->rsk_ops->rtx_syn_ack(sk, req); 540 req->rsk_ops->rtx_syn_ack(sk, req);
541 return NULL; 541 return NULL;
542 } 542 }
543 543
544 /* Further reproduces section "SEGMENT ARRIVES" 544 /* Further reproduces section "SEGMENT ARRIVES"
545 for state SYN-RECEIVED of RFC793. 545 for state SYN-RECEIVED of RFC793.
546 It is broken, however, it does not work only 546 It is broken, however, it does not work only
547 when SYNs are crossed. 547 when SYNs are crossed.
548 548
549 You would think that SYN crossing is impossible here, since 549 You would think that SYN crossing is impossible here, since
550 we should have a SYN_SENT socket (from connect()) on our end, 550 we should have a SYN_SENT socket (from connect()) on our end,
551 but this is not true if the crossed SYNs were sent to both 551 but this is not true if the crossed SYNs were sent to both
552 ends by a malicious third party. We must defend against this, 552 ends by a malicious third party. We must defend against this,
553 and to do that we first verify the ACK (as per RFC793, page 553 and to do that we first verify the ACK (as per RFC793, page
554 36) and reset if it is invalid. Is this a true full defense? 554 36) and reset if it is invalid. Is this a true full defense?
555 To convince ourselves, let us consider a way in which the ACK 555 To convince ourselves, let us consider a way in which the ACK
556 test can still pass in this 'malicious crossed SYNs' case. 556 test can still pass in this 'malicious crossed SYNs' case.
557 Malicious sender sends identical SYNs (and thus identical sequence 557 Malicious sender sends identical SYNs (and thus identical sequence
558 numbers) to both A and B: 558 numbers) to both A and B:
559 559
560 A: gets SYN, seq=7 560 A: gets SYN, seq=7
561 B: gets SYN, seq=7 561 B: gets SYN, seq=7
562 562
563 By our good fortune, both A and B select the same initial 563 By our good fortune, both A and B select the same initial
564 send sequence number of seven :-) 564 send sequence number of seven :-)
565 565
566 A: sends SYN|ACK, seq=7, ack_seq=8 566 A: sends SYN|ACK, seq=7, ack_seq=8
567 B: sends SYN|ACK, seq=7, ack_seq=8 567 B: sends SYN|ACK, seq=7, ack_seq=8
568 568
569 So we are now A eating this SYN|ACK, ACK test passes. So 569 So we are now A eating this SYN|ACK, ACK test passes. So
570 does sequence test, SYN is truncated, and thus we consider 570 does sequence test, SYN is truncated, and thus we consider
571 it a bare ACK. 571 it a bare ACK.
572 572
573 If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this 573 If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this
574 bare ACK. Otherwise, we create an established connection. Both 574 bare ACK. Otherwise, we create an established connection. Both
575 ends (listening sockets) accept the new incoming connection and try 575 ends (listening sockets) accept the new incoming connection and try
576 to talk to each other. 8-) 576 to talk to each other. 8-)
577 577
578 Note: This case is both harmless, and rare. Possibility is about the 578 Note: This case is both harmless, and rare. Possibility is about the
579 same as us discovering intelligent life on another plant tomorrow. 579 same as us discovering intelligent life on another plant tomorrow.
580 580
581 But generally, we should (RFC lies!) to accept ACK 581 But generally, we should (RFC lies!) to accept ACK
582 from SYNACK both here and in tcp_rcv_state_process(). 582 from SYNACK both here and in tcp_rcv_state_process().
583 tcp_rcv_state_process() does not, hence, we do not too. 583 tcp_rcv_state_process() does not, hence, we do not too.
584 584
585 Note that the case is absolutely generic: 585 Note that the case is absolutely generic:
586 we cannot optimize anything here without 586 we cannot optimize anything here without
587 violating protocol. All the checks must be made 587 violating protocol. All the checks must be made
588 before attempt to create socket. 588 before attempt to create socket.
589 */ 589 */
590 590
591 /* RFC793 page 36: "If the connection is in any non-synchronized state ... 591 /* RFC793 page 36: "If the connection is in any non-synchronized state ...
592 * and the incoming segment acknowledges something not yet 592 * and the incoming segment acknowledges something not yet
593 * sent (the segment carries an unacceptable ACK) ... 593 * sent (the segment carries an unacceptable ACK) ...
594 * a reset is sent." 594 * a reset is sent."
595 * 595 *
596 * Invalid ACK: reset will be sent by listening socket 596 * Invalid ACK: reset will be sent by listening socket
597 */ 597 */
598 if ((flg & TCP_FLAG_ACK) && 598 if ((flg & TCP_FLAG_ACK) &&
599 (TCP_SKB_CB(skb)->ack_seq != tcp_rsk(req)->snt_isn + 1)) 599 (TCP_SKB_CB(skb)->ack_seq != tcp_rsk(req)->snt_isn + 1))
600 return sk; 600 return sk;
601 601
602 /* Also, it would be not so bad idea to check rcv_tsecr, which 602 /* Also, it would be not so bad idea to check rcv_tsecr, which
603 * is essentially ACK extension and too early or too late values 603 * is essentially ACK extension and too early or too late values
604 * should cause reset in unsynchronized states. 604 * should cause reset in unsynchronized states.
605 */ 605 */
606 606
607 /* RFC793: "first check sequence number". */ 607 /* RFC793: "first check sequence number". */
608 608
609 if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq, 609 if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
610 tcp_rsk(req)->rcv_isn + 1, tcp_rsk(req)->rcv_isn + 1 + req->rcv_wnd)) { 610 tcp_rsk(req)->rcv_isn + 1, tcp_rsk(req)->rcv_isn + 1 + req->rcv_wnd)) {
611 /* Out of window: send ACK and drop. */ 611 /* Out of window: send ACK and drop. */
612 if (!(flg & TCP_FLAG_RST)) 612 if (!(flg & TCP_FLAG_RST))
613 req->rsk_ops->send_ack(sk, skb, req); 613 req->rsk_ops->send_ack(sk, skb, req);
614 if (paws_reject) 614 if (paws_reject)
615 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED); 615 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED);
616 return NULL; 616 return NULL;
617 } 617 }
618 618
619 /* In sequence, PAWS is OK. */ 619 /* In sequence, PAWS is OK. */
620 620
621 if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_isn + 1)) 621 if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_isn + 1))
622 req->ts_recent = tmp_opt.rcv_tsval; 622 req->ts_recent = tmp_opt.rcv_tsval;
623 623
624 if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) { 624 if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
625 /* Truncate SYN, it is out of window starting 625 /* Truncate SYN, it is out of window starting
626 at tcp_rsk(req)->rcv_isn + 1. */ 626 at tcp_rsk(req)->rcv_isn + 1. */
627 flg &= ~TCP_FLAG_SYN; 627 flg &= ~TCP_FLAG_SYN;
628 } 628 }
629 629
630 /* RFC793: "second check the RST bit" and 630 /* RFC793: "second check the RST bit" and
631 * "fourth, check the SYN bit" 631 * "fourth, check the SYN bit"
632 */ 632 */
633 if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) { 633 if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) {
634 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS); 634 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
635 goto embryonic_reset; 635 goto embryonic_reset;
636 } 636 }
637 637
638 /* ACK sequence verified above, just make sure ACK is 638 /* ACK sequence verified above, just make sure ACK is
639 * set. If ACK not set, just silently drop the packet. 639 * set. If ACK not set, just silently drop the packet.
640 */ 640 */
641 if (!(flg & TCP_FLAG_ACK)) 641 if (!(flg & TCP_FLAG_ACK))
642 return NULL; 642 return NULL;
643 643
644 /* If TCP_DEFER_ACCEPT is set, drop bare ACK. */ 644 /* If TCP_DEFER_ACCEPT is set, drop bare ACK. */
645 if (inet_csk(sk)->icsk_accept_queue.rskq_defer_accept && 645 if (inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
646 TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) { 646 TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
647 inet_csk(sk)->icsk_accept_queue.rskq_defer_accept--;
647 inet_rsk(req)->acked = 1; 648 inet_rsk(req)->acked = 1;
648 return NULL; 649 return NULL;
649 } 650 }
650 651
651 /* OK, ACK is valid, create big socket and 652 /* OK, ACK is valid, create big socket and
652 * feed this segment to it. It will repeat all 653 * feed this segment to it. It will repeat all
653 * the tests. THIS SEGMENT MUST MOVE SOCKET TO 654 * the tests. THIS SEGMENT MUST MOVE SOCKET TO
654 * ESTABLISHED STATE. If it will be dropped after 655 * ESTABLISHED STATE. If it will be dropped after
655 * socket is created, wait for troubles. 656 * socket is created, wait for troubles.
656 */ 657 */
657 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL); 658 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
658 if (child == NULL) 659 if (child == NULL)
659 goto listen_overflow; 660 goto listen_overflow;
660 661
661 inet_csk_reqsk_queue_unlink(sk, req, prev); 662 inet_csk_reqsk_queue_unlink(sk, req, prev);
662 inet_csk_reqsk_queue_removed(sk, req); 663 inet_csk_reqsk_queue_removed(sk, req);
663 664
664 inet_csk_reqsk_queue_add(sk, req, child); 665 inet_csk_reqsk_queue_add(sk, req, child);
665 return child; 666 return child;
666 667
667 listen_overflow: 668 listen_overflow:
668 if (!sysctl_tcp_abort_on_overflow) { 669 if (!sysctl_tcp_abort_on_overflow) {
669 inet_rsk(req)->acked = 1; 670 inet_rsk(req)->acked = 1;
670 return NULL; 671 return NULL;
671 } 672 }
672 673
673 embryonic_reset: 674 embryonic_reset:
674 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_EMBRYONICRSTS); 675 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
675 if (!(flg & TCP_FLAG_RST)) 676 if (!(flg & TCP_FLAG_RST))
676 req->rsk_ops->send_reset(sk, skb); 677 req->rsk_ops->send_reset(sk, skb);
677 678
678 inet_csk_reqsk_queue_drop(sk, req, prev); 679 inet_csk_reqsk_queue_drop(sk, req, prev);
679 return NULL; 680 return NULL;
680 } 681 }
681 682
682 /* 683 /*
683 * Queue segment on the new socket if the new socket is active, 684 * Queue segment on the new socket if the new socket is active,
684 * otherwise we just shortcircuit this and continue with 685 * otherwise we just shortcircuit this and continue with
685 * the new socket. 686 * the new socket.
686 */ 687 */
687 688
688 int tcp_child_process(struct sock *parent, struct sock *child, 689 int tcp_child_process(struct sock *parent, struct sock *child,
689 struct sk_buff *skb) 690 struct sk_buff *skb)
690 { 691 {
691 int ret = 0; 692 int ret = 0;
692 int state = child->sk_state; 693 int state = child->sk_state;
693 694
694 if (!sock_owned_by_user(child)) { 695 if (!sock_owned_by_user(child)) {
695 ret = tcp_rcv_state_process(child, skb, tcp_hdr(skb), 696 ret = tcp_rcv_state_process(child, skb, tcp_hdr(skb),
696 skb->len); 697 skb->len);
697 /* Wakeup parent, send SIGIO */ 698 /* Wakeup parent, send SIGIO */
698 if (state == TCP_SYN_RECV && child->sk_state != state) 699 if (state == TCP_SYN_RECV && child->sk_state != state)
699 parent->sk_data_ready(parent, 0); 700 parent->sk_data_ready(parent, 0);
700 } else { 701 } else {
701 /* Alas, it is possible again, because we do lookup 702 /* Alas, it is possible again, because we do lookup
702 * in main socket hash table and lock on listening 703 * in main socket hash table and lock on listening
703 * socket does not protect us more. 704 * socket does not protect us more.
704 */ 705 */
705 sk_add_backlog(child, skb); 706 sk_add_backlog(child, skb);
706 } 707 }
707 708
708 bh_unlock_sock(child); 709 bh_unlock_sock(child);
709 sock_put(child); 710 sock_put(child);
710 return ret; 711 return ret;
711 } 712 }
712 713
713 EXPORT_SYMBOL(tcp_check_req); 714 EXPORT_SYMBOL(tcp_check_req);
714 EXPORT_SYMBOL(tcp_child_process); 715 EXPORT_SYMBOL(tcp_child_process);
715 EXPORT_SYMBOL(tcp_create_openreq_child); 716 EXPORT_SYMBOL(tcp_create_openreq_child);
716 EXPORT_SYMBOL(tcp_timewait_state_process); 717 EXPORT_SYMBOL(tcp_timewait_state_process);
717 718