Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
3 | * operating system. INET is implemented using the BSD Socket | |
4 | * interface as the means of communication with the user level. | |
5 | * | |
6 | * Implementation of the Transmission Control Protocol(TCP). | |
7 | * | |
02c30a84 | 8 | * Authors: Ross Biro |
1da177e4 LT |
9 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
10 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
11 | * Corey Minyard <wf-rch!minyard@relay.EU.net> | |
12 | * Florian La Roche, <flla@stud.uni-sb.de> | |
13 | * Charles Hedrick, <hedrick@klinzhai.rutgers.edu> | |
14 | * Linus Torvalds, <torvalds@cs.helsinki.fi> | |
15 | * Alan Cox, <gw4pts@gw4pts.ampr.org> | |
16 | * Matthew Dillon, <dillon@apollo.west.oic.com> | |
17 | * Arnt Gulbrandsen, <agulbra@nvg.unit.no> | |
18 | * Jorge Cwik, <jorge@laser.satlink.net> | |
19 | */ | |
20 | ||
21 | /* | |
22 | * Changes: Pedro Roque : Retransmit queue handled by TCP. | |
23 | * : Fragmentation on mtu decrease | |
24 | * : Segment collapse on retransmit | |
25 | * : AF independence | |
26 | * | |
27 | * Linus Torvalds : send_delayed_ack | |
28 | * David S. Miller : Charge memory using the right skb | |
29 | * during syn/ack processing. | |
30 | * David S. Miller : Output engine completely rewritten. | |
31 | * Andrea Arcangeli: SYNACK carry ts_recent in tsecr. | |
32 | * Cacophonix Gaul : draft-minshall-nagle-01 | |
33 | * J Hadi Salim : ECN support | |
34 | * | |
35 | */ | |
36 | ||
37 | #include <net/tcp.h> | |
38 | ||
39 | #include <linux/compiler.h> | |
40 | #include <linux/module.h> | |
1da177e4 LT |
41 | |
42 | /* People can turn this off for buggy TCP's found in printers etc. */ | |
ab32ea5d | 43 | int sysctl_tcp_retrans_collapse __read_mostly = 1; |
1da177e4 | 44 | |
09cb105e | 45 | /* People can turn this on to work with those rare, broken TCPs that |
15d99e02 RJ |
46 | * interpret the window field as a signed quantity. |
47 | */ | |
ab32ea5d | 48 | int sysctl_tcp_workaround_signed_windows __read_mostly = 0; |
15d99e02 | 49 | |
1da177e4 LT |
50 | /* This limits the percentage of the congestion window which we |
51 | * will allow a single TSO frame to consume. Building TSO frames | |
52 | * which are too large can cause TCP streams to be bursty. | |
53 | */ | |
ab32ea5d | 54 | int sysctl_tcp_tso_win_divisor __read_mostly = 3; |
1da177e4 | 55 | |
ab32ea5d BH |
56 | int sysctl_tcp_mtu_probing __read_mostly = 0; |
57 | int sysctl_tcp_base_mss __read_mostly = 512; | |
5d424d5a | 58 | |
35089bb2 | 59 | /* By default, RFC2861 behavior. */ |
ab32ea5d | 60 | int sysctl_tcp_slow_start_after_idle __read_mostly = 1; |
35089bb2 | 61 | |
67edfef7 | 62 | /* Account for new data that has been sent to the network. */ |
66f5fe62 | 63 | static void tcp_event_new_data_sent(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 64 | { |
9e412ba7 | 65 | struct tcp_sock *tp = tcp_sk(sk); |
66f5fe62 | 66 | unsigned int prior_packets = tp->packets_out; |
9e412ba7 | 67 | |
fe067e8a | 68 | tcp_advance_send_head(sk, skb); |
1da177e4 | 69 | tp->snd_nxt = TCP_SKB_CB(skb)->end_seq; |
8512430e IJ |
70 | |
71 | /* Don't override Nagle indefinately with F-RTO */ | |
72 | if (tp->frto_counter == 2) | |
73 | tp->frto_counter = 3; | |
66f5fe62 IJ |
74 | |
75 | tp->packets_out += tcp_skb_pcount(skb); | |
76 | if (!prior_packets) | |
77 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, | |
78 | inet_csk(sk)->icsk_rto, TCP_RTO_MAX); | |
1da177e4 LT |
79 | } |
80 | ||
81 | /* SND.NXT, if window was not shrunk. | |
82 | * If window has been shrunk, what should we make? It is not clear at all. | |
83 | * Using SND.UNA we will fail to open window, SND.NXT is out of window. :-( | |
84 | * Anything in between SND.UNA...SND.UNA+SND.WND also can be already | |
85 | * invalid. OK, let's make this for now: | |
86 | */ | |
9e412ba7 | 87 | static inline __u32 tcp_acceptable_seq(struct sock *sk) |
1da177e4 | 88 | { |
9e412ba7 IJ |
89 | struct tcp_sock *tp = tcp_sk(sk); |
90 | ||
90840def | 91 | if (!before(tcp_wnd_end(tp), tp->snd_nxt)) |
1da177e4 LT |
92 | return tp->snd_nxt; |
93 | else | |
90840def | 94 | return tcp_wnd_end(tp); |
1da177e4 LT |
95 | } |
96 | ||
97 | /* Calculate mss to advertise in SYN segment. | |
98 | * RFC1122, RFC1063, draft-ietf-tcpimpl-pmtud-01 state that: | |
99 | * | |
100 | * 1. It is independent of path mtu. | |
101 | * 2. Ideally, it is maximal possible segment size i.e. 65535-40. | |
102 | * 3. For IPv4 it is reasonable to calculate it from maximal MTU of | |
103 | * attached devices, because some buggy hosts are confused by | |
104 | * large MSS. | |
105 | * 4. We do not make 3, we advertise MSS, calculated from first | |
106 | * hop device mtu, but allow to raise it to ip_rt_min_advmss. | |
107 | * This may be overridden via information stored in routing table. | |
108 | * 5. Value 65535 for MSS is valid in IPv6 and means "as large as possible, | |
109 | * probably even Jumbo". | |
110 | */ | |
111 | static __u16 tcp_advertise_mss(struct sock *sk) | |
112 | { | |
113 | struct tcp_sock *tp = tcp_sk(sk); | |
114 | struct dst_entry *dst = __sk_dst_get(sk); | |
115 | int mss = tp->advmss; | |
116 | ||
117 | if (dst && dst_metric(dst, RTAX_ADVMSS) < mss) { | |
118 | mss = dst_metric(dst, RTAX_ADVMSS); | |
119 | tp->advmss = mss; | |
120 | } | |
121 | ||
122 | return (__u16)mss; | |
123 | } | |
124 | ||
125 | /* RFC2861. Reset CWND after idle period longer RTO to "restart window". | |
126 | * This is the first part of cwnd validation mechanism. */ | |
463c84b9 | 127 | static void tcp_cwnd_restart(struct sock *sk, struct dst_entry *dst) |
1da177e4 | 128 | { |
463c84b9 | 129 | struct tcp_sock *tp = tcp_sk(sk); |
1da177e4 LT |
130 | s32 delta = tcp_time_stamp - tp->lsndtime; |
131 | u32 restart_cwnd = tcp_init_cwnd(tp, dst); | |
132 | u32 cwnd = tp->snd_cwnd; | |
133 | ||
6687e988 | 134 | tcp_ca_event(sk, CA_EVENT_CWND_RESTART); |
1da177e4 | 135 | |
6687e988 | 136 | tp->snd_ssthresh = tcp_current_ssthresh(sk); |
1da177e4 LT |
137 | restart_cwnd = min(restart_cwnd, cwnd); |
138 | ||
463c84b9 | 139 | while ((delta -= inet_csk(sk)->icsk_rto) > 0 && cwnd > restart_cwnd) |
1da177e4 LT |
140 | cwnd >>= 1; |
141 | tp->snd_cwnd = max(cwnd, restart_cwnd); | |
142 | tp->snd_cwnd_stamp = tcp_time_stamp; | |
143 | tp->snd_cwnd_used = 0; | |
144 | } | |
145 | ||
67edfef7 | 146 | /* Congestion state accounting after a packet has been sent. */ |
40efc6fa SH |
147 | static void tcp_event_data_sent(struct tcp_sock *tp, |
148 | struct sk_buff *skb, struct sock *sk) | |
1da177e4 | 149 | { |
463c84b9 ACM |
150 | struct inet_connection_sock *icsk = inet_csk(sk); |
151 | const u32 now = tcp_time_stamp; | |
1da177e4 | 152 | |
35089bb2 DM |
153 | if (sysctl_tcp_slow_start_after_idle && |
154 | (!tp->packets_out && (s32)(now - tp->lsndtime) > icsk->icsk_rto)) | |
463c84b9 | 155 | tcp_cwnd_restart(sk, __sk_dst_get(sk)); |
1da177e4 LT |
156 | |
157 | tp->lsndtime = now; | |
158 | ||
159 | /* If it is a reply for ato after last received | |
160 | * packet, enter pingpong mode. | |
161 | */ | |
463c84b9 ACM |
162 | if ((u32)(now - icsk->icsk_ack.lrcvtime) < icsk->icsk_ack.ato) |
163 | icsk->icsk_ack.pingpong = 1; | |
1da177e4 LT |
164 | } |
165 | ||
67edfef7 | 166 | /* Account for an ACK we sent. */ |
40efc6fa | 167 | static inline void tcp_event_ack_sent(struct sock *sk, unsigned int pkts) |
1da177e4 | 168 | { |
463c84b9 ACM |
169 | tcp_dec_quickack_mode(sk, pkts); |
170 | inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK); | |
1da177e4 LT |
171 | } |
172 | ||
173 | /* Determine a window scaling and initial window to offer. | |
174 | * Based on the assumption that the given amount of space | |
175 | * will be offered. Store the results in the tp structure. | |
176 | * NOTE: for smooth operation initial space offering should | |
177 | * be a multiple of mss if possible. We assume here that mss >= 1. | |
178 | * This MUST be enforced by all callers. | |
179 | */ | |
180 | void tcp_select_initial_window(int __space, __u32 mss, | |
181 | __u32 *rcv_wnd, __u32 *window_clamp, | |
182 | int wscale_ok, __u8 *rcv_wscale) | |
183 | { | |
184 | unsigned int space = (__space < 0 ? 0 : __space); | |
185 | ||
186 | /* If no clamp set the clamp to the max possible scaled window */ | |
187 | if (*window_clamp == 0) | |
188 | (*window_clamp) = (65535 << 14); | |
189 | space = min(*window_clamp, space); | |
190 | ||
191 | /* Quantize space offering to a multiple of mss if possible. */ | |
192 | if (space > mss) | |
193 | space = (space / mss) * mss; | |
194 | ||
195 | /* NOTE: offering an initial window larger than 32767 | |
15d99e02 RJ |
196 | * will break some buggy TCP stacks. If the admin tells us |
197 | * it is likely we could be speaking with such a buggy stack | |
198 | * we will truncate our initial window offering to 32K-1 | |
199 | * unless the remote has sent us a window scaling option, | |
200 | * which we interpret as a sign the remote TCP is not | |
201 | * misinterpreting the window field as a signed quantity. | |
1da177e4 | 202 | */ |
15d99e02 RJ |
203 | if (sysctl_tcp_workaround_signed_windows) |
204 | (*rcv_wnd) = min(space, MAX_TCP_WINDOW); | |
205 | else | |
206 | (*rcv_wnd) = space; | |
207 | ||
1da177e4 LT |
208 | (*rcv_wscale) = 0; |
209 | if (wscale_ok) { | |
210 | /* Set window scaling on max possible window | |
e905a9ed | 211 | * See RFC1323 for an explanation of the limit to 14 |
1da177e4 LT |
212 | */ |
213 | space = max_t(u32, sysctl_tcp_rmem[2], sysctl_rmem_max); | |
316c1592 | 214 | space = min_t(u32, space, *window_clamp); |
1da177e4 LT |
215 | while (space > 65535 && (*rcv_wscale) < 14) { |
216 | space >>= 1; | |
217 | (*rcv_wscale)++; | |
218 | } | |
219 | } | |
220 | ||
221 | /* Set initial window to value enough for senders, | |
6b251858 | 222 | * following RFC2414. Senders, not following this RFC, |
1da177e4 LT |
223 | * will be satisfied with 2. |
224 | */ | |
056834d9 | 225 | if (mss > (1 << *rcv_wscale)) { |
01ff367e | 226 | int init_cwnd = 4; |
056834d9 | 227 | if (mss > 1460 * 3) |
1da177e4 | 228 | init_cwnd = 2; |
01ff367e DM |
229 | else if (mss > 1460) |
230 | init_cwnd = 3; | |
056834d9 IJ |
231 | if (*rcv_wnd > init_cwnd * mss) |
232 | *rcv_wnd = init_cwnd * mss; | |
1da177e4 LT |
233 | } |
234 | ||
235 | /* Set the clamp no higher than max representable value */ | |
236 | (*window_clamp) = min(65535U << (*rcv_wscale), *window_clamp); | |
237 | } | |
238 | ||
239 | /* Chose a new window to advertise, update state in tcp_sock for the | |
240 | * socket, and return result with RFC1323 scaling applied. The return | |
241 | * value can be stuffed directly into th->window for an outgoing | |
242 | * frame. | |
243 | */ | |
40efc6fa | 244 | static u16 tcp_select_window(struct sock *sk) |
1da177e4 LT |
245 | { |
246 | struct tcp_sock *tp = tcp_sk(sk); | |
247 | u32 cur_win = tcp_receive_window(tp); | |
248 | u32 new_win = __tcp_select_window(sk); | |
249 | ||
250 | /* Never shrink the offered window */ | |
2de979bd | 251 | if (new_win < cur_win) { |
1da177e4 LT |
252 | /* Danger Will Robinson! |
253 | * Don't update rcv_wup/rcv_wnd here or else | |
254 | * we will not be able to advertise a zero | |
255 | * window in time. --DaveM | |
256 | * | |
257 | * Relax Will Robinson. | |
258 | */ | |
607bfbf2 | 259 | new_win = ALIGN(cur_win, 1 << tp->rx_opt.rcv_wscale); |
1da177e4 LT |
260 | } |
261 | tp->rcv_wnd = new_win; | |
262 | tp->rcv_wup = tp->rcv_nxt; | |
263 | ||
264 | /* Make sure we do not exceed the maximum possible | |
265 | * scaled window. | |
266 | */ | |
15d99e02 | 267 | if (!tp->rx_opt.rcv_wscale && sysctl_tcp_workaround_signed_windows) |
1da177e4 LT |
268 | new_win = min(new_win, MAX_TCP_WINDOW); |
269 | else | |
270 | new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale)); | |
271 | ||
272 | /* RFC1323 scaling applied */ | |
273 | new_win >>= tp->rx_opt.rcv_wscale; | |
274 | ||
275 | /* If we advertise zero window, disable fast path. */ | |
276 | if (new_win == 0) | |
277 | tp->pred_flags = 0; | |
278 | ||
279 | return new_win; | |
280 | } | |
281 | ||
67edfef7 | 282 | /* Packet ECN state for a SYN-ACK */ |
056834d9 | 283 | static inline void TCP_ECN_send_synack(struct tcp_sock *tp, struct sk_buff *skb) |
bdf1ee5d IJ |
284 | { |
285 | TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_CWR; | |
056834d9 | 286 | if (!(tp->ecn_flags & TCP_ECN_OK)) |
bdf1ee5d IJ |
287 | TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_ECE; |
288 | } | |
289 | ||
67edfef7 | 290 | /* Packet ECN state for a SYN. */ |
bdf1ee5d IJ |
291 | static inline void TCP_ECN_send_syn(struct sock *sk, struct sk_buff *skb) |
292 | { | |
293 | struct tcp_sock *tp = tcp_sk(sk); | |
294 | ||
295 | tp->ecn_flags = 0; | |
255cac91 | 296 | if (sysctl_tcp_ecn == 1) { |
056834d9 | 297 | TCP_SKB_CB(skb)->flags |= TCPCB_FLAG_ECE | TCPCB_FLAG_CWR; |
bdf1ee5d IJ |
298 | tp->ecn_flags = TCP_ECN_OK; |
299 | } | |
300 | } | |
301 | ||
302 | static __inline__ void | |
303 | TCP_ECN_make_synack(struct request_sock *req, struct tcphdr *th) | |
304 | { | |
305 | if (inet_rsk(req)->ecn_ok) | |
306 | th->ece = 1; | |
307 | } | |
308 | ||
67edfef7 AK |
309 | /* Set up ECN state for a packet on a ESTABLISHED socket that is about to |
310 | * be sent. | |
311 | */ | |
bdf1ee5d IJ |
312 | static inline void TCP_ECN_send(struct sock *sk, struct sk_buff *skb, |
313 | int tcp_header_len) | |
314 | { | |
315 | struct tcp_sock *tp = tcp_sk(sk); | |
316 | ||
317 | if (tp->ecn_flags & TCP_ECN_OK) { | |
318 | /* Not-retransmitted data segment: set ECT and inject CWR. */ | |
319 | if (skb->len != tcp_header_len && | |
320 | !before(TCP_SKB_CB(skb)->seq, tp->snd_nxt)) { | |
321 | INET_ECN_xmit(sk); | |
056834d9 | 322 | if (tp->ecn_flags & TCP_ECN_QUEUE_CWR) { |
bdf1ee5d IJ |
323 | tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR; |
324 | tcp_hdr(skb)->cwr = 1; | |
325 | skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; | |
326 | } | |
327 | } else { | |
328 | /* ACK or retransmitted segment: clear ECT|CE */ | |
329 | INET_ECN_dontxmit(sk); | |
330 | } | |
331 | if (tp->ecn_flags & TCP_ECN_DEMAND_CWR) | |
332 | tcp_hdr(skb)->ece = 1; | |
333 | } | |
334 | } | |
335 | ||
e870a8ef IJ |
336 | /* Constructs common control bits of non-data skb. If SYN/FIN is present, |
337 | * auto increment end seqno. | |
338 | */ | |
339 | static void tcp_init_nondata_skb(struct sk_buff *skb, u32 seq, u8 flags) | |
340 | { | |
341 | skb->csum = 0; | |
342 | ||
343 | TCP_SKB_CB(skb)->flags = flags; | |
344 | TCP_SKB_CB(skb)->sacked = 0; | |
345 | ||
346 | skb_shinfo(skb)->gso_segs = 1; | |
347 | skb_shinfo(skb)->gso_size = 0; | |
348 | skb_shinfo(skb)->gso_type = 0; | |
349 | ||
350 | TCP_SKB_CB(skb)->seq = seq; | |
351 | if (flags & (TCPCB_FLAG_SYN | TCPCB_FLAG_FIN)) | |
352 | seq++; | |
353 | TCP_SKB_CB(skb)->end_seq = seq; | |
354 | } | |
355 | ||
33f5f57e IJ |
356 | static inline int tcp_urg_mode(const struct tcp_sock *tp) |
357 | { | |
358 | return tp->snd_una != tp->snd_up; | |
359 | } | |
360 | ||
33ad798c AL |
361 | #define OPTION_SACK_ADVERTISE (1 << 0) |
362 | #define OPTION_TS (1 << 1) | |
363 | #define OPTION_MD5 (1 << 2) | |
364 | ||
365 | struct tcp_out_options { | |
366 | u8 options; /* bit field of OPTION_* */ | |
367 | u8 ws; /* window scale, 0 to disable */ | |
368 | u8 num_sack_blocks; /* number of SACK blocks to include */ | |
369 | u16 mss; /* 0 to disable */ | |
370 | __u32 tsval, tsecr; /* need to include OPTION_TS */ | |
371 | }; | |
372 | ||
67edfef7 AK |
373 | /* Write previously computed TCP options to the packet. |
374 | * | |
375 | * Beware: Something in the Internet is very sensitive to the ordering of | |
fd6149d3 IJ |
376 | * TCP options, we learned this through the hard way, so be careful here. |
377 | * Luckily we can at least blame others for their non-compliance but from | |
378 | * inter-operatibility perspective it seems that we're somewhat stuck with | |
379 | * the ordering which we have been using if we want to keep working with | |
380 | * those broken things (not that it currently hurts anybody as there isn't | |
381 | * particular reason why the ordering would need to be changed). | |
382 | * | |
383 | * At least SACK_PERM as the first option is known to lead to a disaster | |
384 | * (but it may well be that other scenarios fail similarly). | |
385 | */ | |
33ad798c AL |
386 | static void tcp_options_write(__be32 *ptr, struct tcp_sock *tp, |
387 | const struct tcp_out_options *opts, | |
388 | __u8 **md5_hash) { | |
389 | if (unlikely(OPTION_MD5 & opts->options)) { | |
496c98df YH |
390 | *ptr++ = htonl((TCPOPT_NOP << 24) | |
391 | (TCPOPT_NOP << 16) | | |
33ad798c AL |
392 | (TCPOPT_MD5SIG << 8) | |
393 | TCPOLEN_MD5SIG); | |
394 | *md5_hash = (__u8 *)ptr; | |
395 | ptr += 4; | |
396 | } else { | |
397 | *md5_hash = NULL; | |
40efc6fa | 398 | } |
33ad798c | 399 | |
fd6149d3 IJ |
400 | if (unlikely(opts->mss)) { |
401 | *ptr++ = htonl((TCPOPT_MSS << 24) | | |
402 | (TCPOLEN_MSS << 16) | | |
403 | opts->mss); | |
404 | } | |
405 | ||
33ad798c AL |
406 | if (likely(OPTION_TS & opts->options)) { |
407 | if (unlikely(OPTION_SACK_ADVERTISE & opts->options)) { | |
408 | *ptr++ = htonl((TCPOPT_SACK_PERM << 24) | | |
409 | (TCPOLEN_SACK_PERM << 16) | | |
410 | (TCPOPT_TIMESTAMP << 8) | | |
411 | TCPOLEN_TIMESTAMP); | |
412 | } else { | |
413 | *ptr++ = htonl((TCPOPT_NOP << 24) | | |
414 | (TCPOPT_NOP << 16) | | |
415 | (TCPOPT_TIMESTAMP << 8) | | |
416 | TCPOLEN_TIMESTAMP); | |
417 | } | |
418 | *ptr++ = htonl(opts->tsval); | |
419 | *ptr++ = htonl(opts->tsecr); | |
420 | } | |
421 | ||
33ad798c AL |
422 | if (unlikely(OPTION_SACK_ADVERTISE & opts->options && |
423 | !(OPTION_TS & opts->options))) { | |
424 | *ptr++ = htonl((TCPOPT_NOP << 24) | | |
425 | (TCPOPT_NOP << 16) | | |
426 | (TCPOPT_SACK_PERM << 8) | | |
427 | TCPOLEN_SACK_PERM); | |
428 | } | |
429 | ||
430 | if (unlikely(opts->ws)) { | |
431 | *ptr++ = htonl((TCPOPT_NOP << 24) | | |
432 | (TCPOPT_WINDOW << 16) | | |
433 | (TCPOLEN_WINDOW << 8) | | |
434 | opts->ws); | |
435 | } | |
436 | ||
437 | if (unlikely(opts->num_sack_blocks)) { | |
438 | struct tcp_sack_block *sp = tp->rx_opt.dsack ? | |
439 | tp->duplicate_sack : tp->selective_acks; | |
40efc6fa SH |
440 | int this_sack; |
441 | ||
442 | *ptr++ = htonl((TCPOPT_NOP << 24) | | |
443 | (TCPOPT_NOP << 16) | | |
444 | (TCPOPT_SACK << 8) | | |
33ad798c | 445 | (TCPOLEN_SACK_BASE + (opts->num_sack_blocks * |
40efc6fa | 446 | TCPOLEN_SACK_PERBLOCK))); |
2de979bd | 447 | |
33ad798c AL |
448 | for (this_sack = 0; this_sack < opts->num_sack_blocks; |
449 | ++this_sack) { | |
40efc6fa SH |
450 | *ptr++ = htonl(sp[this_sack].start_seq); |
451 | *ptr++ = htonl(sp[this_sack].end_seq); | |
452 | } | |
2de979bd | 453 | |
5861f8e5 | 454 | tp->rx_opt.dsack = 0; |
40efc6fa | 455 | } |
33ad798c AL |
456 | } |
457 | ||
67edfef7 AK |
458 | /* Compute TCP options for SYN packets. This is not the final |
459 | * network wire format yet. | |
460 | */ | |
33ad798c AL |
461 | static unsigned tcp_syn_options(struct sock *sk, struct sk_buff *skb, |
462 | struct tcp_out_options *opts, | |
463 | struct tcp_md5sig_key **md5) { | |
464 | struct tcp_sock *tp = tcp_sk(sk); | |
465 | unsigned size = 0; | |
466 | ||
cfb6eeb4 | 467 | #ifdef CONFIG_TCP_MD5SIG |
33ad798c AL |
468 | *md5 = tp->af_specific->md5_lookup(sk, sk); |
469 | if (*md5) { | |
470 | opts->options |= OPTION_MD5; | |
471 | size += TCPOLEN_MD5SIG_ALIGNED; | |
cfb6eeb4 | 472 | } |
33ad798c AL |
473 | #else |
474 | *md5 = NULL; | |
cfb6eeb4 | 475 | #endif |
33ad798c AL |
476 | |
477 | /* We always get an MSS option. The option bytes which will be seen in | |
478 | * normal data packets should timestamps be used, must be in the MSS | |
479 | * advertised. But we subtract them from tp->mss_cache so that | |
480 | * calculations in tcp_sendmsg are simpler etc. So account for this | |
481 | * fact here if necessary. If we don't do this correctly, as a | |
482 | * receiver we won't recognize data packets as being full sized when we | |
483 | * should, and thus we won't abide by the delayed ACK rules correctly. | |
484 | * SACKs don't matter, we never delay an ACK when we have any of those | |
485 | * going out. */ | |
486 | opts->mss = tcp_advertise_mss(sk); | |
487 | size += TCPOLEN_MSS_ALIGNED; | |
488 | ||
489 | if (likely(sysctl_tcp_timestamps && *md5 == NULL)) { | |
490 | opts->options |= OPTION_TS; | |
491 | opts->tsval = TCP_SKB_CB(skb)->when; | |
492 | opts->tsecr = tp->rx_opt.ts_recent; | |
493 | size += TCPOLEN_TSTAMP_ALIGNED; | |
494 | } | |
495 | if (likely(sysctl_tcp_window_scaling)) { | |
496 | opts->ws = tp->rx_opt.rcv_wscale; | |
09cb105e | 497 | if (likely(opts->ws)) |
7982d5e1 | 498 | size += TCPOLEN_WSCALE_ALIGNED; |
33ad798c AL |
499 | } |
500 | if (likely(sysctl_tcp_sack)) { | |
501 | opts->options |= OPTION_SACK_ADVERTISE; | |
b32d1310 | 502 | if (unlikely(!(OPTION_TS & opts->options))) |
33ad798c AL |
503 | size += TCPOLEN_SACKPERM_ALIGNED; |
504 | } | |
505 | ||
506 | return size; | |
40efc6fa SH |
507 | } |
508 | ||
67edfef7 | 509 | /* Set up TCP options for SYN-ACKs. */ |
33ad798c AL |
510 | static unsigned tcp_synack_options(struct sock *sk, |
511 | struct request_sock *req, | |
512 | unsigned mss, struct sk_buff *skb, | |
513 | struct tcp_out_options *opts, | |
514 | struct tcp_md5sig_key **md5) { | |
515 | unsigned size = 0; | |
516 | struct inet_request_sock *ireq = inet_rsk(req); | |
517 | char doing_ts; | |
518 | ||
cfb6eeb4 | 519 | #ifdef CONFIG_TCP_MD5SIG |
33ad798c AL |
520 | *md5 = tcp_rsk(req)->af_specific->md5_lookup(sk, req); |
521 | if (*md5) { | |
522 | opts->options |= OPTION_MD5; | |
523 | size += TCPOLEN_MD5SIG_ALIGNED; | |
cfb6eeb4 | 524 | } |
33ad798c AL |
525 | #else |
526 | *md5 = NULL; | |
cfb6eeb4 | 527 | #endif |
33ad798c AL |
528 | |
529 | /* we can't fit any SACK blocks in a packet with MD5 + TS | |
530 | options. There was discussion about disabling SACK rather than TS in | |
531 | order to fit in better with old, buggy kernels, but that was deemed | |
532 | to be unnecessary. */ | |
533 | doing_ts = ireq->tstamp_ok && !(*md5 && ireq->sack_ok); | |
534 | ||
535 | opts->mss = mss; | |
536 | size += TCPOLEN_MSS_ALIGNED; | |
537 | ||
538 | if (likely(ireq->wscale_ok)) { | |
539 | opts->ws = ireq->rcv_wscale; | |
09cb105e | 540 | if (likely(opts->ws)) |
7982d5e1 | 541 | size += TCPOLEN_WSCALE_ALIGNED; |
33ad798c AL |
542 | } |
543 | if (likely(doing_ts)) { | |
544 | opts->options |= OPTION_TS; | |
545 | opts->tsval = TCP_SKB_CB(skb)->when; | |
546 | opts->tsecr = req->ts_recent; | |
547 | size += TCPOLEN_TSTAMP_ALIGNED; | |
548 | } | |
549 | if (likely(ireq->sack_ok)) { | |
550 | opts->options |= OPTION_SACK_ADVERTISE; | |
551 | if (unlikely(!doing_ts)) | |
552 | size += TCPOLEN_SACKPERM_ALIGNED; | |
553 | } | |
554 | ||
555 | return size; | |
556 | } | |
557 | ||
67edfef7 AK |
558 | /* Compute TCP options for ESTABLISHED sockets. This is not the |
559 | * final wire format yet. | |
560 | */ | |
33ad798c AL |
561 | static unsigned tcp_established_options(struct sock *sk, struct sk_buff *skb, |
562 | struct tcp_out_options *opts, | |
563 | struct tcp_md5sig_key **md5) { | |
564 | struct tcp_skb_cb *tcb = skb ? TCP_SKB_CB(skb) : NULL; | |
565 | struct tcp_sock *tp = tcp_sk(sk); | |
566 | unsigned size = 0; | |
cabeccbd | 567 | unsigned int eff_sacks; |
33ad798c AL |
568 | |
569 | #ifdef CONFIG_TCP_MD5SIG | |
570 | *md5 = tp->af_specific->md5_lookup(sk, sk); | |
571 | if (unlikely(*md5)) { | |
572 | opts->options |= OPTION_MD5; | |
573 | size += TCPOLEN_MD5SIG_ALIGNED; | |
574 | } | |
575 | #else | |
576 | *md5 = NULL; | |
577 | #endif | |
578 | ||
579 | if (likely(tp->rx_opt.tstamp_ok)) { | |
580 | opts->options |= OPTION_TS; | |
581 | opts->tsval = tcb ? tcb->when : 0; | |
582 | opts->tsecr = tp->rx_opt.ts_recent; | |
583 | size += TCPOLEN_TSTAMP_ALIGNED; | |
584 | } | |
585 | ||
cabeccbd IJ |
586 | eff_sacks = tp->rx_opt.num_sacks + tp->rx_opt.dsack; |
587 | if (unlikely(eff_sacks)) { | |
33ad798c AL |
588 | const unsigned remaining = MAX_TCP_OPTION_SPACE - size; |
589 | opts->num_sack_blocks = | |
cabeccbd | 590 | min_t(unsigned, eff_sacks, |
33ad798c AL |
591 | (remaining - TCPOLEN_SACK_BASE_ALIGNED) / |
592 | TCPOLEN_SACK_PERBLOCK); | |
593 | size += TCPOLEN_SACK_BASE_ALIGNED + | |
594 | opts->num_sack_blocks * TCPOLEN_SACK_PERBLOCK; | |
595 | } | |
596 | ||
597 | return size; | |
40efc6fa | 598 | } |
1da177e4 LT |
599 | |
600 | /* This routine actually transmits TCP packets queued in by | |
601 | * tcp_do_sendmsg(). This is used by both the initial | |
602 | * transmission and possible later retransmissions. | |
603 | * All SKB's seen here are completely headerless. It is our | |
604 | * job to build the TCP header, and pass the packet down to | |
605 | * IP so it can do the same plus pass the packet off to the | |
606 | * device. | |
607 | * | |
608 | * We are working here with either a clone of the original | |
609 | * SKB, or a fresh unique copy made by the retransmit engine. | |
610 | */ | |
056834d9 IJ |
611 | static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it, |
612 | gfp_t gfp_mask) | |
1da177e4 | 613 | { |
dfb4b9dc DM |
614 | const struct inet_connection_sock *icsk = inet_csk(sk); |
615 | struct inet_sock *inet; | |
616 | struct tcp_sock *tp; | |
617 | struct tcp_skb_cb *tcb; | |
33ad798c AL |
618 | struct tcp_out_options opts; |
619 | unsigned tcp_options_size, tcp_header_size; | |
cfb6eeb4 YH |
620 | struct tcp_md5sig_key *md5; |
621 | __u8 *md5_hash_location; | |
dfb4b9dc | 622 | struct tcphdr *th; |
dfb4b9dc DM |
623 | int err; |
624 | ||
625 | BUG_ON(!skb || !tcp_skb_pcount(skb)); | |
626 | ||
627 | /* If congestion control is doing timestamping, we must | |
628 | * take such a timestamp before we potentially clone/copy. | |
629 | */ | |
164891aa | 630 | if (icsk->icsk_ca_ops->flags & TCP_CONG_RTT_STAMP) |
dfb4b9dc DM |
631 | __net_timestamp(skb); |
632 | ||
633 | if (likely(clone_it)) { | |
634 | if (unlikely(skb_cloned(skb))) | |
635 | skb = pskb_copy(skb, gfp_mask); | |
636 | else | |
637 | skb = skb_clone(skb, gfp_mask); | |
638 | if (unlikely(!skb)) | |
639 | return -ENOBUFS; | |
640 | } | |
1da177e4 | 641 | |
dfb4b9dc DM |
642 | inet = inet_sk(sk); |
643 | tp = tcp_sk(sk); | |
644 | tcb = TCP_SKB_CB(skb); | |
33ad798c | 645 | memset(&opts, 0, sizeof(opts)); |
1da177e4 | 646 | |
33ad798c AL |
647 | if (unlikely(tcb->flags & TCPCB_FLAG_SYN)) |
648 | tcp_options_size = tcp_syn_options(sk, skb, &opts, &md5); | |
649 | else | |
650 | tcp_options_size = tcp_established_options(sk, skb, &opts, | |
651 | &md5); | |
652 | tcp_header_size = tcp_options_size + sizeof(struct tcphdr); | |
e905a9ed | 653 | |
dfb4b9dc DM |
654 | if (tcp_packets_in_flight(tp) == 0) |
655 | tcp_ca_event(sk, CA_EVENT_TX_START); | |
656 | ||
aa8223c7 ACM |
657 | skb_push(skb, tcp_header_size); |
658 | skb_reset_transport_header(skb); | |
e89862f4 | 659 | skb_set_owner_w(skb, sk); |
dfb4b9dc DM |
660 | |
661 | /* Build TCP header and checksum it. */ | |
aa8223c7 | 662 | th = tcp_hdr(skb); |
dfb4b9dc DM |
663 | th->source = inet->sport; |
664 | th->dest = inet->dport; | |
665 | th->seq = htonl(tcb->seq); | |
666 | th->ack_seq = htonl(tp->rcv_nxt); | |
df7a3b07 | 667 | *(((__be16 *)th) + 6) = htons(((tcp_header_size >> 2) << 12) | |
dfb4b9dc DM |
668 | tcb->flags); |
669 | ||
670 | if (unlikely(tcb->flags & TCPCB_FLAG_SYN)) { | |
671 | /* RFC1323: The window in SYN & SYN/ACK segments | |
672 | * is never scaled. | |
673 | */ | |
600ff0c2 | 674 | th->window = htons(min(tp->rcv_wnd, 65535U)); |
dfb4b9dc DM |
675 | } else { |
676 | th->window = htons(tcp_select_window(sk)); | |
677 | } | |
678 | th->check = 0; | |
679 | th->urg_ptr = 0; | |
1da177e4 | 680 | |
33f5f57e | 681 | /* The urg_mode check is necessary during a below snd_una win probe */ |
7691367d HX |
682 | if (unlikely(tcp_urg_mode(tp) && before(tcb->seq, tp->snd_up))) { |
683 | if (before(tp->snd_up, tcb->seq + 0x10000)) { | |
684 | th->urg_ptr = htons(tp->snd_up - tcb->seq); | |
685 | th->urg = 1; | |
686 | } else if (after(tcb->seq + 0xFFFF, tp->snd_nxt)) { | |
687 | th->urg_ptr = 0xFFFF; | |
688 | th->urg = 1; | |
689 | } | |
dfb4b9dc | 690 | } |
1da177e4 | 691 | |
33ad798c AL |
692 | tcp_options_write((__be32 *)(th + 1), tp, &opts, &md5_hash_location); |
693 | if (likely((tcb->flags & TCPCB_FLAG_SYN) == 0)) | |
9e412ba7 | 694 | TCP_ECN_send(sk, skb, tcp_header_size); |
1da177e4 | 695 | |
cfb6eeb4 YH |
696 | #ifdef CONFIG_TCP_MD5SIG |
697 | /* Calculate the MD5 hash, as we have all we need now */ | |
698 | if (md5) { | |
33ad798c | 699 | sk->sk_route_caps &= ~NETIF_F_GSO_MASK; |
cfb6eeb4 | 700 | tp->af_specific->calc_md5_hash(md5_hash_location, |
49a72dfb | 701 | md5, sk, NULL, skb); |
cfb6eeb4 YH |
702 | } |
703 | #endif | |
704 | ||
8292a17a | 705 | icsk->icsk_af_ops->send_check(sk, skb->len, skb); |
1da177e4 | 706 | |
dfb4b9dc DM |
707 | if (likely(tcb->flags & TCPCB_FLAG_ACK)) |
708 | tcp_event_ack_sent(sk, tcp_skb_pcount(skb)); | |
1da177e4 | 709 | |
dfb4b9dc DM |
710 | if (skb->len != tcp_header_size) |
711 | tcp_event_data_sent(tp, skb, sk); | |
1da177e4 | 712 | |
bd37a088 | 713 | if (after(tcb->end_seq, tp->snd_nxt) || tcb->seq == tcb->end_seq) |
81cc8a75 | 714 | TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTSEGS); |
1da177e4 | 715 | |
e89862f4 | 716 | err = icsk->icsk_af_ops->queue_xmit(skb, 0); |
83de47cd | 717 | if (likely(err <= 0)) |
dfb4b9dc DM |
718 | return err; |
719 | ||
3cfe3baa | 720 | tcp_enter_cwr(sk, 1); |
dfb4b9dc | 721 | |
b9df3cb8 | 722 | return net_xmit_eval(err); |
1da177e4 LT |
723 | } |
724 | ||
67edfef7 | 725 | /* This routine just queues the buffer for sending. |
1da177e4 LT |
726 | * |
727 | * NOTE: probe0 timer is not checked, do not forget tcp_push_pending_frames, | |
728 | * otherwise socket can stall. | |
729 | */ | |
730 | static void tcp_queue_skb(struct sock *sk, struct sk_buff *skb) | |
731 | { | |
732 | struct tcp_sock *tp = tcp_sk(sk); | |
733 | ||
734 | /* Advance write_seq and place onto the write_queue. */ | |
735 | tp->write_seq = TCP_SKB_CB(skb)->end_seq; | |
736 | skb_header_release(skb); | |
fe067e8a | 737 | tcp_add_write_queue_tail(sk, skb); |
3ab224be HA |
738 | sk->sk_wmem_queued += skb->truesize; |
739 | sk_mem_charge(sk, skb->truesize); | |
1da177e4 LT |
740 | } |
741 | ||
67edfef7 | 742 | /* Initialize TSO segments for a packet. */ |
056834d9 IJ |
743 | static void tcp_set_skb_tso_segs(struct sock *sk, struct sk_buff *skb, |
744 | unsigned int mss_now) | |
f6302d1d | 745 | { |
8e5b9dda HX |
746 | if (skb->len <= mss_now || !sk_can_gso(sk) || |
747 | skb->ip_summed == CHECKSUM_NONE) { | |
f6302d1d DM |
748 | /* Avoid the costly divide in the normal |
749 | * non-TSO case. | |
750 | */ | |
7967168c HX |
751 | skb_shinfo(skb)->gso_segs = 1; |
752 | skb_shinfo(skb)->gso_size = 0; | |
753 | skb_shinfo(skb)->gso_type = 0; | |
f6302d1d | 754 | } else { |
356f89e1 | 755 | skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss_now); |
7967168c | 756 | skb_shinfo(skb)->gso_size = mss_now; |
bcd76111 | 757 | skb_shinfo(skb)->gso_type = sk->sk_gso_type; |
1da177e4 LT |
758 | } |
759 | } | |
760 | ||
91fed7a1 | 761 | /* When a modification to fackets out becomes necessary, we need to check |
68f8353b | 762 | * skb is counted to fackets_out or not. |
91fed7a1 | 763 | */ |
a47e5a98 | 764 | static void tcp_adjust_fackets_out(struct sock *sk, struct sk_buff *skb, |
91fed7a1 IJ |
765 | int decr) |
766 | { | |
a47e5a98 IJ |
767 | struct tcp_sock *tp = tcp_sk(sk); |
768 | ||
dc86967b | 769 | if (!tp->sacked_out || tcp_is_reno(tp)) |
91fed7a1 IJ |
770 | return; |
771 | ||
6859d494 | 772 | if (after(tcp_highest_sack_seq(tp), TCP_SKB_CB(skb)->seq)) |
91fed7a1 | 773 | tp->fackets_out -= decr; |
91fed7a1 IJ |
774 | } |
775 | ||
797108d1 IJ |
776 | /* Pcount in the middle of the write queue got changed, we need to do various |
777 | * tweaks to fix counters | |
778 | */ | |
779 | static void tcp_adjust_pcount(struct sock *sk, struct sk_buff *skb, int decr) | |
780 | { | |
781 | struct tcp_sock *tp = tcp_sk(sk); | |
782 | ||
783 | tp->packets_out -= decr; | |
784 | ||
785 | if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) | |
786 | tp->sacked_out -= decr; | |
787 | if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) | |
788 | tp->retrans_out -= decr; | |
789 | if (TCP_SKB_CB(skb)->sacked & TCPCB_LOST) | |
790 | tp->lost_out -= decr; | |
791 | ||
792 | /* Reno case is special. Sigh... */ | |
793 | if (tcp_is_reno(tp) && decr > 0) | |
794 | tp->sacked_out -= min_t(u32, tp->sacked_out, decr); | |
795 | ||
796 | tcp_adjust_fackets_out(sk, skb, decr); | |
797 | ||
798 | if (tp->lost_skb_hint && | |
799 | before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(tp->lost_skb_hint)->seq) && | |
52cf3cc8 | 800 | (tcp_is_fack(tp) || (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED))) |
797108d1 IJ |
801 | tp->lost_cnt_hint -= decr; |
802 | ||
803 | tcp_verify_left_out(tp); | |
804 | } | |
805 | ||
1da177e4 LT |
806 | /* Function to create two new TCP segments. Shrinks the given segment |
807 | * to the specified size and appends a new segment with the rest of the | |
e905a9ed | 808 | * packet to the list. This won't be called frequently, I hope. |
1da177e4 LT |
809 | * Remember, these are still headerless SKBs at this point. |
810 | */ | |
056834d9 IJ |
811 | int tcp_fragment(struct sock *sk, struct sk_buff *skb, u32 len, |
812 | unsigned int mss_now) | |
1da177e4 LT |
813 | { |
814 | struct tcp_sock *tp = tcp_sk(sk); | |
815 | struct sk_buff *buff; | |
6475be16 | 816 | int nsize, old_factor; |
b60b49ea | 817 | int nlen; |
9ce01461 | 818 | u8 flags; |
1da177e4 | 819 | |
b2cc99f0 | 820 | BUG_ON(len > skb->len); |
6a438bbe | 821 | |
1da177e4 LT |
822 | nsize = skb_headlen(skb) - len; |
823 | if (nsize < 0) | |
824 | nsize = 0; | |
825 | ||
826 | if (skb_cloned(skb) && | |
827 | skb_is_nonlinear(skb) && | |
828 | pskb_expand_head(skb, 0, 0, GFP_ATOMIC)) | |
829 | return -ENOMEM; | |
830 | ||
831 | /* Get a new skb... force flag on. */ | |
832 | buff = sk_stream_alloc_skb(sk, nsize, GFP_ATOMIC); | |
833 | if (buff == NULL) | |
834 | return -ENOMEM; /* We'll just try again later. */ | |
ef5cb973 | 835 | |
3ab224be HA |
836 | sk->sk_wmem_queued += buff->truesize; |
837 | sk_mem_charge(sk, buff->truesize); | |
b60b49ea HX |
838 | nlen = skb->len - len - nsize; |
839 | buff->truesize += nlen; | |
840 | skb->truesize -= nlen; | |
1da177e4 LT |
841 | |
842 | /* Correct the sequence numbers. */ | |
843 | TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len; | |
844 | TCP_SKB_CB(buff)->end_seq = TCP_SKB_CB(skb)->end_seq; | |
845 | TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(buff)->seq; | |
846 | ||
847 | /* PSH and FIN should only be set in the second packet. */ | |
848 | flags = TCP_SKB_CB(skb)->flags; | |
056834d9 | 849 | TCP_SKB_CB(skb)->flags = flags & ~(TCPCB_FLAG_FIN | TCPCB_FLAG_PSH); |
1da177e4 | 850 | TCP_SKB_CB(buff)->flags = flags; |
e14c3caf | 851 | TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked; |
1da177e4 | 852 | |
84fa7933 | 853 | if (!skb_shinfo(skb)->nr_frags && skb->ip_summed != CHECKSUM_PARTIAL) { |
1da177e4 | 854 | /* Copy and checksum data tail into the new buffer. */ |
056834d9 IJ |
855 | buff->csum = csum_partial_copy_nocheck(skb->data + len, |
856 | skb_put(buff, nsize), | |
1da177e4 LT |
857 | nsize, 0); |
858 | ||
859 | skb_trim(skb, len); | |
860 | ||
861 | skb->csum = csum_block_sub(skb->csum, buff->csum, len); | |
862 | } else { | |
84fa7933 | 863 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 LT |
864 | skb_split(skb, buff, len); |
865 | } | |
866 | ||
867 | buff->ip_summed = skb->ip_summed; | |
868 | ||
869 | /* Looks stupid, but our code really uses when of | |
870 | * skbs, which it never sent before. --ANK | |
871 | */ | |
872 | TCP_SKB_CB(buff)->when = TCP_SKB_CB(skb)->when; | |
a61bbcf2 | 873 | buff->tstamp = skb->tstamp; |
1da177e4 | 874 | |
6475be16 DM |
875 | old_factor = tcp_skb_pcount(skb); |
876 | ||
1da177e4 | 877 | /* Fix up tso_factor for both original and new SKB. */ |
846998ae DM |
878 | tcp_set_skb_tso_segs(sk, skb, mss_now); |
879 | tcp_set_skb_tso_segs(sk, buff, mss_now); | |
1da177e4 | 880 | |
6475be16 DM |
881 | /* If this packet has been sent out already, we must |
882 | * adjust the various packet counters. | |
883 | */ | |
cf0b450c | 884 | if (!before(tp->snd_nxt, TCP_SKB_CB(buff)->end_seq)) { |
6475be16 DM |
885 | int diff = old_factor - tcp_skb_pcount(skb) - |
886 | tcp_skb_pcount(buff); | |
1da177e4 | 887 | |
797108d1 IJ |
888 | if (diff) |
889 | tcp_adjust_pcount(sk, skb, diff); | |
1da177e4 LT |
890 | } |
891 | ||
892 | /* Link BUFF into the send queue. */ | |
f44b5271 | 893 | skb_header_release(buff); |
fe067e8a | 894 | tcp_insert_write_queue_after(skb, buff, sk); |
1da177e4 LT |
895 | |
896 | return 0; | |
897 | } | |
898 | ||
899 | /* This is similar to __pskb_pull_head() (it will go to core/skbuff.c | |
900 | * eventually). The difference is that pulled data not copied, but | |
901 | * immediately discarded. | |
902 | */ | |
f2911969 | 903 | static void __pskb_trim_head(struct sk_buff *skb, int len) |
1da177e4 LT |
904 | { |
905 | int i, k, eat; | |
906 | ||
907 | eat = len; | |
908 | k = 0; | |
056834d9 | 909 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
1da177e4 LT |
910 | if (skb_shinfo(skb)->frags[i].size <= eat) { |
911 | put_page(skb_shinfo(skb)->frags[i].page); | |
912 | eat -= skb_shinfo(skb)->frags[i].size; | |
913 | } else { | |
914 | skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; | |
915 | if (eat) { | |
916 | skb_shinfo(skb)->frags[k].page_offset += eat; | |
917 | skb_shinfo(skb)->frags[k].size -= eat; | |
918 | eat = 0; | |
919 | } | |
920 | k++; | |
921 | } | |
922 | } | |
923 | skb_shinfo(skb)->nr_frags = k; | |
924 | ||
27a884dc | 925 | skb_reset_tail_pointer(skb); |
1da177e4 LT |
926 | skb->data_len -= len; |
927 | skb->len = skb->data_len; | |
1da177e4 LT |
928 | } |
929 | ||
67edfef7 | 930 | /* Remove acked data from a packet in the transmit queue. */ |
1da177e4 LT |
931 | int tcp_trim_head(struct sock *sk, struct sk_buff *skb, u32 len) |
932 | { | |
056834d9 | 933 | if (skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC)) |
1da177e4 LT |
934 | return -ENOMEM; |
935 | ||
f2911969 HX |
936 | /* If len == headlen, we avoid __skb_pull to preserve alignment. */ |
937 | if (unlikely(len < skb_headlen(skb))) | |
1da177e4 | 938 | __skb_pull(skb, len); |
f2911969 HX |
939 | else |
940 | __pskb_trim_head(skb, len - skb_headlen(skb)); | |
1da177e4 LT |
941 | |
942 | TCP_SKB_CB(skb)->seq += len; | |
84fa7933 | 943 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 LT |
944 | |
945 | skb->truesize -= len; | |
946 | sk->sk_wmem_queued -= len; | |
3ab224be | 947 | sk_mem_uncharge(sk, len); |
1da177e4 LT |
948 | sock_set_flag(sk, SOCK_QUEUE_SHRUNK); |
949 | ||
950 | /* Any change of skb->len requires recalculation of tso | |
951 | * factor and mss. | |
952 | */ | |
953 | if (tcp_skb_pcount(skb) > 1) | |
0c54b85f | 954 | tcp_set_skb_tso_segs(sk, skb, tcp_current_mss(sk)); |
1da177e4 LT |
955 | |
956 | return 0; | |
957 | } | |
958 | ||
67edfef7 | 959 | /* Calculate MSS. Not accounting for SACKs here. */ |
5d424d5a JH |
960 | int tcp_mtu_to_mss(struct sock *sk, int pmtu) |
961 | { | |
962 | struct tcp_sock *tp = tcp_sk(sk); | |
963 | struct inet_connection_sock *icsk = inet_csk(sk); | |
964 | int mss_now; | |
965 | ||
966 | /* Calculate base mss without TCP options: | |
967 | It is MMS_S - sizeof(tcphdr) of rfc1122 | |
968 | */ | |
969 | mss_now = pmtu - icsk->icsk_af_ops->net_header_len - sizeof(struct tcphdr); | |
970 | ||
971 | /* Clamp it (mss_clamp does not include tcp options) */ | |
972 | if (mss_now > tp->rx_opt.mss_clamp) | |
973 | mss_now = tp->rx_opt.mss_clamp; | |
974 | ||
975 | /* Now subtract optional transport overhead */ | |
976 | mss_now -= icsk->icsk_ext_hdr_len; | |
977 | ||
978 | /* Then reserve room for full set of TCP options and 8 bytes of data */ | |
979 | if (mss_now < 48) | |
980 | mss_now = 48; | |
981 | ||
982 | /* Now subtract TCP options size, not including SACKs */ | |
983 | mss_now -= tp->tcp_header_len - sizeof(struct tcphdr); | |
984 | ||
985 | return mss_now; | |
986 | } | |
987 | ||
988 | /* Inverse of above */ | |
989 | int tcp_mss_to_mtu(struct sock *sk, int mss) | |
990 | { | |
991 | struct tcp_sock *tp = tcp_sk(sk); | |
992 | struct inet_connection_sock *icsk = inet_csk(sk); | |
993 | int mtu; | |
994 | ||
995 | mtu = mss + | |
996 | tp->tcp_header_len + | |
997 | icsk->icsk_ext_hdr_len + | |
998 | icsk->icsk_af_ops->net_header_len; | |
999 | ||
1000 | return mtu; | |
1001 | } | |
1002 | ||
67edfef7 | 1003 | /* MTU probing init per socket */ |
5d424d5a JH |
1004 | void tcp_mtup_init(struct sock *sk) |
1005 | { | |
1006 | struct tcp_sock *tp = tcp_sk(sk); | |
1007 | struct inet_connection_sock *icsk = inet_csk(sk); | |
1008 | ||
1009 | icsk->icsk_mtup.enabled = sysctl_tcp_mtu_probing > 1; | |
1010 | icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + sizeof(struct tcphdr) + | |
e905a9ed | 1011 | icsk->icsk_af_ops->net_header_len; |
5d424d5a JH |
1012 | icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, sysctl_tcp_base_mss); |
1013 | icsk->icsk_mtup.probe_size = 0; | |
1014 | } | |
1015 | ||
1da177e4 LT |
1016 | /* This function synchronize snd mss to current pmtu/exthdr set. |
1017 | ||
1018 | tp->rx_opt.user_mss is mss set by user by TCP_MAXSEG. It does NOT counts | |
1019 | for TCP options, but includes only bare TCP header. | |
1020 | ||
1021 | tp->rx_opt.mss_clamp is mss negotiated at connection setup. | |
caa20d9a | 1022 | It is minimum of user_mss and mss received with SYN. |
1da177e4 LT |
1023 | It also does not include TCP options. |
1024 | ||
d83d8461 | 1025 | inet_csk(sk)->icsk_pmtu_cookie is last pmtu, seen by this function. |
1da177e4 LT |
1026 | |
1027 | tp->mss_cache is current effective sending mss, including | |
1028 | all tcp options except for SACKs. It is evaluated, | |
1029 | taking into account current pmtu, but never exceeds | |
1030 | tp->rx_opt.mss_clamp. | |
1031 | ||
1032 | NOTE1. rfc1122 clearly states that advertised MSS | |
1033 | DOES NOT include either tcp or ip options. | |
1034 | ||
d83d8461 ACM |
1035 | NOTE2. inet_csk(sk)->icsk_pmtu_cookie and tp->mss_cache |
1036 | are READ ONLY outside this function. --ANK (980731) | |
1da177e4 | 1037 | */ |
1da177e4 LT |
1038 | unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu) |
1039 | { | |
1040 | struct tcp_sock *tp = tcp_sk(sk); | |
d83d8461 | 1041 | struct inet_connection_sock *icsk = inet_csk(sk); |
5d424d5a | 1042 | int mss_now; |
1da177e4 | 1043 | |
5d424d5a JH |
1044 | if (icsk->icsk_mtup.search_high > pmtu) |
1045 | icsk->icsk_mtup.search_high = pmtu; | |
1da177e4 | 1046 | |
5d424d5a | 1047 | mss_now = tcp_mtu_to_mss(sk, pmtu); |
409d22b4 | 1048 | mss_now = tcp_bound_to_half_wnd(tp, mss_now); |
1da177e4 LT |
1049 | |
1050 | /* And store cached results */ | |
d83d8461 | 1051 | icsk->icsk_pmtu_cookie = pmtu; |
5d424d5a JH |
1052 | if (icsk->icsk_mtup.enabled) |
1053 | mss_now = min(mss_now, tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low)); | |
c1b4a7e6 | 1054 | tp->mss_cache = mss_now; |
1da177e4 LT |
1055 | |
1056 | return mss_now; | |
1057 | } | |
1058 | ||
1059 | /* Compute the current effective MSS, taking SACKs and IP options, | |
1060 | * and even PMTU discovery events into account. | |
1da177e4 | 1061 | */ |
0c54b85f | 1062 | unsigned int tcp_current_mss(struct sock *sk) |
1da177e4 LT |
1063 | { |
1064 | struct tcp_sock *tp = tcp_sk(sk); | |
1065 | struct dst_entry *dst = __sk_dst_get(sk); | |
c1b4a7e6 | 1066 | u32 mss_now; |
33ad798c AL |
1067 | unsigned header_len; |
1068 | struct tcp_out_options opts; | |
1069 | struct tcp_md5sig_key *md5; | |
c1b4a7e6 DM |
1070 | |
1071 | mss_now = tp->mss_cache; | |
1072 | ||
1da177e4 LT |
1073 | if (dst) { |
1074 | u32 mtu = dst_mtu(dst); | |
d83d8461 | 1075 | if (mtu != inet_csk(sk)->icsk_pmtu_cookie) |
1da177e4 LT |
1076 | mss_now = tcp_sync_mss(sk, mtu); |
1077 | } | |
1078 | ||
33ad798c AL |
1079 | header_len = tcp_established_options(sk, NULL, &opts, &md5) + |
1080 | sizeof(struct tcphdr); | |
1081 | /* The mss_cache is sized based on tp->tcp_header_len, which assumes | |
1082 | * some common options. If this is an odd packet (because we have SACK | |
1083 | * blocks etc) then our calculated header_len will be different, and | |
1084 | * we have to adjust mss_now correspondingly */ | |
1085 | if (header_len != tp->tcp_header_len) { | |
1086 | int delta = (int) header_len - tp->tcp_header_len; | |
1087 | mss_now -= delta; | |
1088 | } | |
cfb6eeb4 | 1089 | |
1da177e4 LT |
1090 | return mss_now; |
1091 | } | |
1092 | ||
a762a980 | 1093 | /* Congestion window validation. (RFC2861) */ |
9e412ba7 | 1094 | static void tcp_cwnd_validate(struct sock *sk) |
a762a980 | 1095 | { |
9e412ba7 | 1096 | struct tcp_sock *tp = tcp_sk(sk); |
a762a980 | 1097 | |
d436d686 | 1098 | if (tp->packets_out >= tp->snd_cwnd) { |
a762a980 DM |
1099 | /* Network is feed fully. */ |
1100 | tp->snd_cwnd_used = 0; | |
1101 | tp->snd_cwnd_stamp = tcp_time_stamp; | |
1102 | } else { | |
1103 | /* Network starves. */ | |
1104 | if (tp->packets_out > tp->snd_cwnd_used) | |
1105 | tp->snd_cwnd_used = tp->packets_out; | |
1106 | ||
15d33c07 DM |
1107 | if (sysctl_tcp_slow_start_after_idle && |
1108 | (s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= inet_csk(sk)->icsk_rto) | |
a762a980 DM |
1109 | tcp_cwnd_application_limited(sk); |
1110 | } | |
1111 | } | |
1112 | ||
0e3a4803 IJ |
1113 | /* Returns the portion of skb which can be sent right away without |
1114 | * introducing MSS oddities to segment boundaries. In rare cases where | |
1115 | * mss_now != mss_cache, we will request caller to create a small skb | |
1116 | * per input skb which could be mostly avoided here (if desired). | |
5ea3a748 IJ |
1117 | * |
1118 | * We explicitly want to create a request for splitting write queue tail | |
1119 | * to a small skb for Nagle purposes while avoiding unnecessary modulos, | |
1120 | * thus all the complexity (cwnd_len is always MSS multiple which we | |
1121 | * return whenever allowed by the other factors). Basically we need the | |
1122 | * modulo only when the receiver window alone is the limiting factor or | |
1123 | * when we would be allowed to send the split-due-to-Nagle skb fully. | |
0e3a4803 IJ |
1124 | */ |
1125 | static unsigned int tcp_mss_split_point(struct sock *sk, struct sk_buff *skb, | |
056834d9 | 1126 | unsigned int mss_now, unsigned int cwnd) |
c1b4a7e6 | 1127 | { |
0e3a4803 IJ |
1128 | struct tcp_sock *tp = tcp_sk(sk); |
1129 | u32 needed, window, cwnd_len; | |
c1b4a7e6 | 1130 | |
90840def | 1131 | window = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; |
c1b4a7e6 | 1132 | cwnd_len = mss_now * cwnd; |
0e3a4803 IJ |
1133 | |
1134 | if (likely(cwnd_len <= window && skb != tcp_write_queue_tail(sk))) | |
1135 | return cwnd_len; | |
1136 | ||
5ea3a748 IJ |
1137 | needed = min(skb->len, window); |
1138 | ||
17515408 | 1139 | if (cwnd_len <= needed) |
0e3a4803 IJ |
1140 | return cwnd_len; |
1141 | ||
0e3a4803 | 1142 | return needed - needed % mss_now; |
c1b4a7e6 DM |
1143 | } |
1144 | ||
1145 | /* Can at least one segment of SKB be sent right now, according to the | |
1146 | * congestion window rules? If so, return how many segments are allowed. | |
1147 | */ | |
056834d9 IJ |
1148 | static inline unsigned int tcp_cwnd_test(struct tcp_sock *tp, |
1149 | struct sk_buff *skb) | |
c1b4a7e6 DM |
1150 | { |
1151 | u32 in_flight, cwnd; | |
1152 | ||
1153 | /* Don't be strict about the congestion window for the final FIN. */ | |
104439a8 JH |
1154 | if ((TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN) && |
1155 | tcp_skb_pcount(skb) == 1) | |
c1b4a7e6 DM |
1156 | return 1; |
1157 | ||
1158 | in_flight = tcp_packets_in_flight(tp); | |
1159 | cwnd = tp->snd_cwnd; | |
1160 | if (in_flight < cwnd) | |
1161 | return (cwnd - in_flight); | |
1162 | ||
1163 | return 0; | |
1164 | } | |
1165 | ||
67edfef7 AK |
1166 | /* Intialize TSO state of a skb. |
1167 | * This must be invoked the first time we consider transmitting | |
c1b4a7e6 DM |
1168 | * SKB onto the wire. |
1169 | */ | |
056834d9 IJ |
1170 | static int tcp_init_tso_segs(struct sock *sk, struct sk_buff *skb, |
1171 | unsigned int mss_now) | |
c1b4a7e6 DM |
1172 | { |
1173 | int tso_segs = tcp_skb_pcount(skb); | |
1174 | ||
f8269a49 | 1175 | if (!tso_segs || (tso_segs > 1 && tcp_skb_mss(skb) != mss_now)) { |
846998ae | 1176 | tcp_set_skb_tso_segs(sk, skb, mss_now); |
c1b4a7e6 DM |
1177 | tso_segs = tcp_skb_pcount(skb); |
1178 | } | |
1179 | return tso_segs; | |
1180 | } | |
1181 | ||
67edfef7 | 1182 | /* Minshall's variant of the Nagle send check. */ |
c1b4a7e6 DM |
1183 | static inline int tcp_minshall_check(const struct tcp_sock *tp) |
1184 | { | |
09cb105e | 1185 | return after(tp->snd_sml, tp->snd_una) && |
c1b4a7e6 DM |
1186 | !after(tp->snd_sml, tp->snd_nxt); |
1187 | } | |
1188 | ||
1189 | /* Return 0, if packet can be sent now without violation Nagle's rules: | |
1190 | * 1. It is full sized. | |
1191 | * 2. Or it contains FIN. (already checked by caller) | |
1192 | * 3. Or TCP_NODELAY was set. | |
1193 | * 4. Or TCP_CORK is not set, and all sent packets are ACKed. | |
1194 | * With Minshall's modification: all sent small packets are ACKed. | |
1195 | */ | |
c1b4a7e6 | 1196 | static inline int tcp_nagle_check(const struct tcp_sock *tp, |
e905a9ed | 1197 | const struct sk_buff *skb, |
c1b4a7e6 DM |
1198 | unsigned mss_now, int nonagle) |
1199 | { | |
1200 | return (skb->len < mss_now && | |
056834d9 IJ |
1201 | ((nonagle & TCP_NAGLE_CORK) || |
1202 | (!nonagle && tp->packets_out && tcp_minshall_check(tp)))); | |
c1b4a7e6 DM |
1203 | } |
1204 | ||
1205 | /* Return non-zero if the Nagle test allows this packet to be | |
1206 | * sent now. | |
1207 | */ | |
1208 | static inline int tcp_nagle_test(struct tcp_sock *tp, struct sk_buff *skb, | |
1209 | unsigned int cur_mss, int nonagle) | |
1210 | { | |
1211 | /* Nagle rule does not apply to frames, which sit in the middle of the | |
1212 | * write_queue (they have no chances to get new data). | |
1213 | * | |
1214 | * This is implemented in the callers, where they modify the 'nonagle' | |
1215 | * argument based upon the location of SKB in the send queue. | |
1216 | */ | |
1217 | if (nonagle & TCP_NAGLE_PUSH) | |
1218 | return 1; | |
1219 | ||
d551e454 IJ |
1220 | /* Don't use the nagle rule for urgent data (or for the final FIN). |
1221 | * Nagle can be ignored during F-RTO too (see RFC4138). | |
1222 | */ | |
33f5f57e | 1223 | if (tcp_urg_mode(tp) || (tp->frto_counter == 2) || |
c1b4a7e6 DM |
1224 | (TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN)) |
1225 | return 1; | |
1226 | ||
1227 | if (!tcp_nagle_check(tp, skb, cur_mss, nonagle)) | |
1228 | return 1; | |
1229 | ||
1230 | return 0; | |
1231 | } | |
1232 | ||
1233 | /* Does at least the first segment of SKB fit into the send window? */ | |
056834d9 IJ |
1234 | static inline int tcp_snd_wnd_test(struct tcp_sock *tp, struct sk_buff *skb, |
1235 | unsigned int cur_mss) | |
c1b4a7e6 DM |
1236 | { |
1237 | u32 end_seq = TCP_SKB_CB(skb)->end_seq; | |
1238 | ||
1239 | if (skb->len > cur_mss) | |
1240 | end_seq = TCP_SKB_CB(skb)->seq + cur_mss; | |
1241 | ||
90840def | 1242 | return !after(end_seq, tcp_wnd_end(tp)); |
c1b4a7e6 DM |
1243 | } |
1244 | ||
fe067e8a | 1245 | /* This checks if the data bearing packet SKB (usually tcp_send_head(sk)) |
c1b4a7e6 DM |
1246 | * should be put on the wire right now. If so, it returns the number of |
1247 | * packets allowed by the congestion window. | |
1248 | */ | |
1249 | static unsigned int tcp_snd_test(struct sock *sk, struct sk_buff *skb, | |
1250 | unsigned int cur_mss, int nonagle) | |
1251 | { | |
1252 | struct tcp_sock *tp = tcp_sk(sk); | |
1253 | unsigned int cwnd_quota; | |
1254 | ||
846998ae | 1255 | tcp_init_tso_segs(sk, skb, cur_mss); |
c1b4a7e6 DM |
1256 | |
1257 | if (!tcp_nagle_test(tp, skb, cur_mss, nonagle)) | |
1258 | return 0; | |
1259 | ||
1260 | cwnd_quota = tcp_cwnd_test(tp, skb); | |
056834d9 | 1261 | if (cwnd_quota && !tcp_snd_wnd_test(tp, skb, cur_mss)) |
c1b4a7e6 DM |
1262 | cwnd_quota = 0; |
1263 | ||
1264 | return cwnd_quota; | |
1265 | } | |
1266 | ||
67edfef7 | 1267 | /* Test if sending is allowed right now. */ |
9e412ba7 | 1268 | int tcp_may_send_now(struct sock *sk) |
c1b4a7e6 | 1269 | { |
9e412ba7 | 1270 | struct tcp_sock *tp = tcp_sk(sk); |
fe067e8a | 1271 | struct sk_buff *skb = tcp_send_head(sk); |
c1b4a7e6 DM |
1272 | |
1273 | return (skb && | |
0c54b85f | 1274 | tcp_snd_test(sk, skb, tcp_current_mss(sk), |
c1b4a7e6 | 1275 | (tcp_skb_is_last(sk, skb) ? |