]>
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 | |
15d99e02 RJ |
45 | /* People can turn this on to work with those rare, broken TCPs that |
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 | |
66f5fe62 | 62 | static void tcp_event_new_data_sent(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 63 | { |
9e412ba7 | 64 | struct tcp_sock *tp = tcp_sk(sk); |
66f5fe62 | 65 | unsigned int prior_packets = tp->packets_out; |
9e412ba7 | 66 | |
fe067e8a | 67 | tcp_advance_send_head(sk, skb); |
1da177e4 | 68 | tp->snd_nxt = TCP_SKB_CB(skb)->end_seq; |
8512430e IJ |
69 | |
70 | /* Don't override Nagle indefinately with F-RTO */ | |
71 | if (tp->frto_counter == 2) | |
72 | tp->frto_counter = 3; | |
66f5fe62 IJ |
73 | |
74 | tp->packets_out += tcp_skb_pcount(skb); | |
75 | if (!prior_packets) | |
76 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, | |
77 | inet_csk(sk)->icsk_rto, TCP_RTO_MAX); | |
1da177e4 LT |
78 | } |
79 | ||
80 | /* SND.NXT, if window was not shrunk. | |
81 | * If window has been shrunk, what should we make? It is not clear at all. | |
82 | * Using SND.UNA we will fail to open window, SND.NXT is out of window. :-( | |
83 | * Anything in between SND.UNA...SND.UNA+SND.WND also can be already | |
84 | * invalid. OK, let's make this for now: | |
85 | */ | |
9e412ba7 | 86 | static inline __u32 tcp_acceptable_seq(struct sock *sk) |
1da177e4 | 87 | { |
9e412ba7 IJ |
88 | struct tcp_sock *tp = tcp_sk(sk); |
89 | ||
90840def | 90 | if (!before(tcp_wnd_end(tp), tp->snd_nxt)) |
1da177e4 LT |
91 | return tp->snd_nxt; |
92 | else | |
90840def | 93 | return tcp_wnd_end(tp); |
1da177e4 LT |
94 | } |
95 | ||
96 | /* Calculate mss to advertise in SYN segment. | |
97 | * RFC1122, RFC1063, draft-ietf-tcpimpl-pmtud-01 state that: | |
98 | * | |
99 | * 1. It is independent of path mtu. | |
100 | * 2. Ideally, it is maximal possible segment size i.e. 65535-40. | |
101 | * 3. For IPv4 it is reasonable to calculate it from maximal MTU of | |
102 | * attached devices, because some buggy hosts are confused by | |
103 | * large MSS. | |
104 | * 4. We do not make 3, we advertise MSS, calculated from first | |
105 | * hop device mtu, but allow to raise it to ip_rt_min_advmss. | |
106 | * This may be overridden via information stored in routing table. | |
107 | * 5. Value 65535 for MSS is valid in IPv6 and means "as large as possible, | |
108 | * probably even Jumbo". | |
109 | */ | |
110 | static __u16 tcp_advertise_mss(struct sock *sk) | |
111 | { | |
112 | struct tcp_sock *tp = tcp_sk(sk); | |
113 | struct dst_entry *dst = __sk_dst_get(sk); | |
114 | int mss = tp->advmss; | |
115 | ||
116 | if (dst && dst_metric(dst, RTAX_ADVMSS) < mss) { | |
117 | mss = dst_metric(dst, RTAX_ADVMSS); | |
118 | tp->advmss = mss; | |
119 | } | |
120 | ||
121 | return (__u16)mss; | |
122 | } | |
123 | ||
124 | /* RFC2861. Reset CWND after idle period longer RTO to "restart window". | |
125 | * This is the first part of cwnd validation mechanism. */ | |
463c84b9 | 126 | static void tcp_cwnd_restart(struct sock *sk, struct dst_entry *dst) |
1da177e4 | 127 | { |
463c84b9 | 128 | struct tcp_sock *tp = tcp_sk(sk); |
1da177e4 LT |
129 | s32 delta = tcp_time_stamp - tp->lsndtime; |
130 | u32 restart_cwnd = tcp_init_cwnd(tp, dst); | |
131 | u32 cwnd = tp->snd_cwnd; | |
132 | ||
6687e988 | 133 | tcp_ca_event(sk, CA_EVENT_CWND_RESTART); |
1da177e4 | 134 | |
6687e988 | 135 | tp->snd_ssthresh = tcp_current_ssthresh(sk); |
1da177e4 LT |
136 | restart_cwnd = min(restart_cwnd, cwnd); |
137 | ||
463c84b9 | 138 | while ((delta -= inet_csk(sk)->icsk_rto) > 0 && cwnd > restart_cwnd) |
1da177e4 LT |
139 | cwnd >>= 1; |
140 | tp->snd_cwnd = max(cwnd, restart_cwnd); | |
141 | tp->snd_cwnd_stamp = tcp_time_stamp; | |
142 | tp->snd_cwnd_used = 0; | |
143 | } | |
144 | ||
40efc6fa SH |
145 | static void tcp_event_data_sent(struct tcp_sock *tp, |
146 | struct sk_buff *skb, struct sock *sk) | |
1da177e4 | 147 | { |
463c84b9 ACM |
148 | struct inet_connection_sock *icsk = inet_csk(sk); |
149 | const u32 now = tcp_time_stamp; | |
1da177e4 | 150 | |
35089bb2 DM |
151 | if (sysctl_tcp_slow_start_after_idle && |
152 | (!tp->packets_out && (s32)(now - tp->lsndtime) > icsk->icsk_rto)) | |
463c84b9 | 153 | tcp_cwnd_restart(sk, __sk_dst_get(sk)); |
1da177e4 LT |
154 | |
155 | tp->lsndtime = now; | |
156 | ||
157 | /* If it is a reply for ato after last received | |
158 | * packet, enter pingpong mode. | |
159 | */ | |
463c84b9 ACM |
160 | if ((u32)(now - icsk->icsk_ack.lrcvtime) < icsk->icsk_ack.ato) |
161 | icsk->icsk_ack.pingpong = 1; | |
1da177e4 LT |
162 | } |
163 | ||
40efc6fa | 164 | static inline void tcp_event_ack_sent(struct sock *sk, unsigned int pkts) |
1da177e4 | 165 | { |
463c84b9 ACM |
166 | tcp_dec_quickack_mode(sk, pkts); |
167 | inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK); | |
1da177e4 LT |
168 | } |
169 | ||
170 | /* Determine a window scaling and initial window to offer. | |
171 | * Based on the assumption that the given amount of space | |
172 | * will be offered. Store the results in the tp structure. | |
173 | * NOTE: for smooth operation initial space offering should | |
174 | * be a multiple of mss if possible. We assume here that mss >= 1. | |
175 | * This MUST be enforced by all callers. | |
176 | */ | |
177 | void tcp_select_initial_window(int __space, __u32 mss, | |
178 | __u32 *rcv_wnd, __u32 *window_clamp, | |
179 | int wscale_ok, __u8 *rcv_wscale) | |
180 | { | |
181 | unsigned int space = (__space < 0 ? 0 : __space); | |
182 | ||
183 | /* If no clamp set the clamp to the max possible scaled window */ | |
184 | if (*window_clamp == 0) | |
185 | (*window_clamp) = (65535 << 14); | |
186 | space = min(*window_clamp, space); | |
187 | ||
188 | /* Quantize space offering to a multiple of mss if possible. */ | |
189 | if (space > mss) | |
190 | space = (space / mss) * mss; | |
191 | ||
192 | /* NOTE: offering an initial window larger than 32767 | |
15d99e02 RJ |
193 | * will break some buggy TCP stacks. If the admin tells us |
194 | * it is likely we could be speaking with such a buggy stack | |
195 | * we will truncate our initial window offering to 32K-1 | |
196 | * unless the remote has sent us a window scaling option, | |
197 | * which we interpret as a sign the remote TCP is not | |
198 | * misinterpreting the window field as a signed quantity. | |
1da177e4 | 199 | */ |
15d99e02 RJ |
200 | if (sysctl_tcp_workaround_signed_windows) |
201 | (*rcv_wnd) = min(space, MAX_TCP_WINDOW); | |
202 | else | |
203 | (*rcv_wnd) = space; | |
204 | ||
1da177e4 LT |
205 | (*rcv_wscale) = 0; |
206 | if (wscale_ok) { | |
207 | /* Set window scaling on max possible window | |
e905a9ed | 208 | * See RFC1323 for an explanation of the limit to 14 |
1da177e4 LT |
209 | */ |
210 | space = max_t(u32, sysctl_tcp_rmem[2], sysctl_rmem_max); | |
316c1592 | 211 | space = min_t(u32, space, *window_clamp); |
1da177e4 LT |
212 | while (space > 65535 && (*rcv_wscale) < 14) { |
213 | space >>= 1; | |
214 | (*rcv_wscale)++; | |
215 | } | |
216 | } | |
217 | ||
218 | /* Set initial window to value enough for senders, | |
6b251858 | 219 | * following RFC2414. Senders, not following this RFC, |
1da177e4 LT |
220 | * will be satisfied with 2. |
221 | */ | |
056834d9 | 222 | if (mss > (1 << *rcv_wscale)) { |
01ff367e | 223 | int init_cwnd = 4; |
056834d9 | 224 | if (mss > 1460 * 3) |
1da177e4 | 225 | init_cwnd = 2; |
01ff367e DM |
226 | else if (mss > 1460) |
227 | init_cwnd = 3; | |
056834d9 IJ |
228 | if (*rcv_wnd > init_cwnd * mss) |
229 | *rcv_wnd = init_cwnd * mss; | |
1da177e4 LT |
230 | } |
231 | ||
232 | /* Set the clamp no higher than max representable value */ | |
233 | (*window_clamp) = min(65535U << (*rcv_wscale), *window_clamp); | |
234 | } | |
235 | ||
236 | /* Chose a new window to advertise, update state in tcp_sock for the | |
237 | * socket, and return result with RFC1323 scaling applied. The return | |
238 | * value can be stuffed directly into th->window for an outgoing | |
239 | * frame. | |
240 | */ | |
40efc6fa | 241 | static u16 tcp_select_window(struct sock *sk) |
1da177e4 LT |
242 | { |
243 | struct tcp_sock *tp = tcp_sk(sk); | |
244 | u32 cur_win = tcp_receive_window(tp); | |
245 | u32 new_win = __tcp_select_window(sk); | |
246 | ||
247 | /* Never shrink the offered window */ | |
2de979bd | 248 | if (new_win < cur_win) { |
1da177e4 LT |
249 | /* Danger Will Robinson! |
250 | * Don't update rcv_wup/rcv_wnd here or else | |
251 | * we will not be able to advertise a zero | |
252 | * window in time. --DaveM | |
253 | * | |
254 | * Relax Will Robinson. | |
255 | */ | |
607bfbf2 | 256 | new_win = ALIGN(cur_win, 1 << tp->rx_opt.rcv_wscale); |
1da177e4 LT |
257 | } |
258 | tp->rcv_wnd = new_win; | |
259 | tp->rcv_wup = tp->rcv_nxt; | |
260 | ||
261 | /* Make sure we do not exceed the maximum possible | |
262 | * scaled window. | |
263 | */ | |
15d99e02 | 264 | if (!tp->rx_opt.rcv_wscale && sysctl_tcp_workaround_signed_windows) |
1da177e4 LT |
265 | new_win = min(new_win, MAX_TCP_WINDOW); |
266 | else | |
267 | new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale)); | |
268 | ||
269 | /* RFC1323 scaling applied */ | |
270 | new_win >>= tp->rx_opt.rcv_wscale; | |
271 | ||
272 | /* If we advertise zero window, disable fast path. */ | |
273 | if (new_win == 0) | |
274 | tp->pred_flags = 0; | |
275 | ||
276 | return new_win; | |
277 | } | |
278 | ||
056834d9 | 279 | static inline void TCP_ECN_send_synack(struct tcp_sock *tp, struct sk_buff *skb) |
bdf1ee5d IJ |
280 | { |
281 | TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_CWR; | |
056834d9 | 282 | if (!(tp->ecn_flags & TCP_ECN_OK)) |
bdf1ee5d IJ |
283 | TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_ECE; |
284 | } | |
285 | ||
286 | static inline void TCP_ECN_send_syn(struct sock *sk, struct sk_buff *skb) | |
287 | { | |
288 | struct tcp_sock *tp = tcp_sk(sk); | |
289 | ||
290 | tp->ecn_flags = 0; | |
291 | if (sysctl_tcp_ecn) { | |
056834d9 | 292 | TCP_SKB_CB(skb)->flags |= TCPCB_FLAG_ECE | TCPCB_FLAG_CWR; |
bdf1ee5d IJ |
293 | tp->ecn_flags = TCP_ECN_OK; |
294 | } | |
295 | } | |
296 | ||
297 | static __inline__ void | |
298 | TCP_ECN_make_synack(struct request_sock *req, struct tcphdr *th) | |
299 | { | |
300 | if (inet_rsk(req)->ecn_ok) | |
301 | th->ece = 1; | |
302 | } | |
303 | ||
304 | static inline void TCP_ECN_send(struct sock *sk, struct sk_buff *skb, | |
305 | int tcp_header_len) | |
306 | { | |
307 | struct tcp_sock *tp = tcp_sk(sk); | |
308 | ||
309 | if (tp->ecn_flags & TCP_ECN_OK) { | |
310 | /* Not-retransmitted data segment: set ECT and inject CWR. */ | |
311 | if (skb->len != tcp_header_len && | |
312 | !before(TCP_SKB_CB(skb)->seq, tp->snd_nxt)) { | |
313 | INET_ECN_xmit(sk); | |
056834d9 | 314 | if (tp->ecn_flags & TCP_ECN_QUEUE_CWR) { |
bdf1ee5d IJ |
315 | tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR; |
316 | tcp_hdr(skb)->cwr = 1; | |
317 | skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; | |
318 | } | |
319 | } else { | |
320 | /* ACK or retransmitted segment: clear ECT|CE */ | |
321 | INET_ECN_dontxmit(sk); | |
322 | } | |
323 | if (tp->ecn_flags & TCP_ECN_DEMAND_CWR) | |
324 | tcp_hdr(skb)->ece = 1; | |
325 | } | |
326 | } | |
327 | ||
e870a8ef IJ |
328 | /* Constructs common control bits of non-data skb. If SYN/FIN is present, |
329 | * auto increment end seqno. | |
330 | */ | |
331 | static void tcp_init_nondata_skb(struct sk_buff *skb, u32 seq, u8 flags) | |
332 | { | |
333 | skb->csum = 0; | |
334 | ||
335 | TCP_SKB_CB(skb)->flags = flags; | |
336 | TCP_SKB_CB(skb)->sacked = 0; | |
337 | ||
338 | skb_shinfo(skb)->gso_segs = 1; | |
339 | skb_shinfo(skb)->gso_size = 0; | |
340 | skb_shinfo(skb)->gso_type = 0; | |
341 | ||
342 | TCP_SKB_CB(skb)->seq = seq; | |
343 | if (flags & (TCPCB_FLAG_SYN | TCPCB_FLAG_FIN)) | |
344 | seq++; | |
345 | TCP_SKB_CB(skb)->end_seq = seq; | |
346 | } | |
347 | ||
33f5f57e IJ |
348 | static inline int tcp_urg_mode(const struct tcp_sock *tp) |
349 | { | |
350 | return tp->snd_una != tp->snd_up; | |
351 | } | |
352 | ||
33ad798c AL |
353 | #define OPTION_SACK_ADVERTISE (1 << 0) |
354 | #define OPTION_TS (1 << 1) | |
355 | #define OPTION_MD5 (1 << 2) | |
356 | ||
357 | struct tcp_out_options { | |
358 | u8 options; /* bit field of OPTION_* */ | |
359 | u8 ws; /* window scale, 0 to disable */ | |
360 | u8 num_sack_blocks; /* number of SACK blocks to include */ | |
361 | u16 mss; /* 0 to disable */ | |
362 | __u32 tsval, tsecr; /* need to include OPTION_TS */ | |
363 | }; | |
364 | ||
365 | static void tcp_options_write(__be32 *ptr, struct tcp_sock *tp, | |
366 | const struct tcp_out_options *opts, | |
367 | __u8 **md5_hash) { | |
368 | if (unlikely(OPTION_MD5 & opts->options)) { | |
496c98df YH |
369 | *ptr++ = htonl((TCPOPT_NOP << 24) | |
370 | (TCPOPT_NOP << 16) | | |
33ad798c AL |
371 | (TCPOPT_MD5SIG << 8) | |
372 | TCPOLEN_MD5SIG); | |
373 | *md5_hash = (__u8 *)ptr; | |
374 | ptr += 4; | |
375 | } else { | |
376 | *md5_hash = NULL; | |
40efc6fa | 377 | } |
33ad798c AL |
378 | |
379 | if (likely(OPTION_TS & opts->options)) { | |
380 | if (unlikely(OPTION_SACK_ADVERTISE & opts->options)) { | |
381 | *ptr++ = htonl((TCPOPT_SACK_PERM << 24) | | |
382 | (TCPOLEN_SACK_PERM << 16) | | |
383 | (TCPOPT_TIMESTAMP << 8) | | |
384 | TCPOLEN_TIMESTAMP); | |
385 | } else { | |
386 | *ptr++ = htonl((TCPOPT_NOP << 24) | | |
387 | (TCPOPT_NOP << 16) | | |
388 | (TCPOPT_TIMESTAMP << 8) | | |
389 | TCPOLEN_TIMESTAMP); | |
390 | } | |
391 | *ptr++ = htonl(opts->tsval); | |
392 | *ptr++ = htonl(opts->tsecr); | |
393 | } | |
394 | ||
395 | if (unlikely(opts->mss)) { | |
396 | *ptr++ = htonl((TCPOPT_MSS << 24) | | |
397 | (TCPOLEN_MSS << 16) | | |
398 | opts->mss); | |
399 | } | |
400 | ||
401 | if (unlikely(OPTION_SACK_ADVERTISE & opts->options && | |
402 | !(OPTION_TS & opts->options))) { | |
403 | *ptr++ = htonl((TCPOPT_NOP << 24) | | |
404 | (TCPOPT_NOP << 16) | | |
405 | (TCPOPT_SACK_PERM << 8) | | |
406 | TCPOLEN_SACK_PERM); | |
407 | } | |
408 | ||
409 | if (unlikely(opts->ws)) { | |
410 | *ptr++ = htonl((TCPOPT_NOP << 24) | | |
411 | (TCPOPT_WINDOW << 16) | | |
412 | (TCPOLEN_WINDOW << 8) | | |
413 | opts->ws); | |
414 | } | |
415 | ||
416 | if (unlikely(opts->num_sack_blocks)) { | |
417 | struct tcp_sack_block *sp = tp->rx_opt.dsack ? | |
418 | tp->duplicate_sack : tp->selective_acks; | |
40efc6fa SH |
419 | int this_sack; |
420 | ||
421 | *ptr++ = htonl((TCPOPT_NOP << 24) | | |
422 | (TCPOPT_NOP << 16) | | |
423 | (TCPOPT_SACK << 8) | | |
33ad798c | 424 | (TCPOLEN_SACK_BASE + (opts->num_sack_blocks * |
40efc6fa | 425 | TCPOLEN_SACK_PERBLOCK))); |
2de979bd | 426 | |
33ad798c AL |
427 | for (this_sack = 0; this_sack < opts->num_sack_blocks; |
428 | ++this_sack) { | |
40efc6fa SH |
429 | *ptr++ = htonl(sp[this_sack].start_seq); |
430 | *ptr++ = htonl(sp[this_sack].end_seq); | |
431 | } | |
2de979bd | 432 | |
40efc6fa SH |
433 | if (tp->rx_opt.dsack) { |
434 | tp->rx_opt.dsack = 0; | |
435 | tp->rx_opt.eff_sacks--; | |
436 | } | |
437 | } | |
33ad798c AL |
438 | } |
439 | ||
440 | static unsigned tcp_syn_options(struct sock *sk, struct sk_buff *skb, | |
441 | struct tcp_out_options *opts, | |
442 | struct tcp_md5sig_key **md5) { | |
443 | struct tcp_sock *tp = tcp_sk(sk); | |
444 | unsigned size = 0; | |
445 | ||
cfb6eeb4 | 446 | #ifdef CONFIG_TCP_MD5SIG |
33ad798c AL |
447 | *md5 = tp->af_specific->md5_lookup(sk, sk); |
448 | if (*md5) { | |
449 | opts->options |= OPTION_MD5; | |
450 | size += TCPOLEN_MD5SIG_ALIGNED; | |
cfb6eeb4 | 451 | } |
33ad798c AL |
452 | #else |
453 | *md5 = NULL; | |
cfb6eeb4 | 454 | #endif |
33ad798c AL |
455 | |
456 | /* We always get an MSS option. The option bytes which will be seen in | |
457 | * normal data packets should timestamps be used, must be in the MSS | |
458 | * advertised. But we subtract them from tp->mss_cache so that | |
459 | * calculations in tcp_sendmsg are simpler etc. So account for this | |
460 | * fact here if necessary. If we don't do this correctly, as a | |
461 | * receiver we won't recognize data packets as being full sized when we | |
462 | * should, and thus we won't abide by the delayed ACK rules correctly. | |
463 | * SACKs don't matter, we never delay an ACK when we have any of those | |
464 | * going out. */ | |
465 | opts->mss = tcp_advertise_mss(sk); | |
466 | size += TCPOLEN_MSS_ALIGNED; | |
467 | ||
468 | if (likely(sysctl_tcp_timestamps && *md5 == NULL)) { | |
469 | opts->options |= OPTION_TS; | |
470 | opts->tsval = TCP_SKB_CB(skb)->when; | |
471 | opts->tsecr = tp->rx_opt.ts_recent; | |
472 | size += TCPOLEN_TSTAMP_ALIGNED; | |
473 | } | |
474 | if (likely(sysctl_tcp_window_scaling)) { | |
475 | opts->ws = tp->rx_opt.rcv_wscale; | |
7982d5e1 PL |
476 | if(likely(opts->ws)) |
477 | size += TCPOLEN_WSCALE_ALIGNED; | |
33ad798c AL |
478 | } |
479 | if (likely(sysctl_tcp_sack)) { | |
480 | opts->options |= OPTION_SACK_ADVERTISE; | |
b32d1310 | 481 | if (unlikely(!(OPTION_TS & opts->options))) |
33ad798c AL |
482 | size += TCPOLEN_SACKPERM_ALIGNED; |
483 | } | |
484 | ||
485 | return size; | |
40efc6fa SH |
486 | } |
487 | ||
33ad798c AL |
488 | static unsigned tcp_synack_options(struct sock *sk, |
489 | struct request_sock *req, | |
490 | unsigned mss, struct sk_buff *skb, | |
491 | struct tcp_out_options *opts, | |
492 | struct tcp_md5sig_key **md5) { | |
493 | unsigned size = 0; | |
494 | struct inet_request_sock *ireq = inet_rsk(req); | |
495 | char doing_ts; | |
496 | ||
cfb6eeb4 | 497 | #ifdef CONFIG_TCP_MD5SIG |
33ad798c AL |
498 | *md5 = tcp_rsk(req)->af_specific->md5_lookup(sk, req); |
499 | if (*md5) { | |
500 | opts->options |= OPTION_MD5; | |
501 | size += TCPOLEN_MD5SIG_ALIGNED; | |
cfb6eeb4 | 502 | } |
33ad798c AL |
503 | #else |
504 | *md5 = NULL; | |
cfb6eeb4 | 505 | #endif |
33ad798c AL |
506 | |
507 | /* we can't fit any SACK blocks in a packet with MD5 + TS | |
508 | options. There was discussion about disabling SACK rather than TS in | |
509 | order to fit in better with old, buggy kernels, but that was deemed | |
510 | to be unnecessary. */ | |
511 | doing_ts = ireq->tstamp_ok && !(*md5 && ireq->sack_ok); | |
512 | ||
513 | opts->mss = mss; | |
514 | size += TCPOLEN_MSS_ALIGNED; | |
515 | ||
516 | if (likely(ireq->wscale_ok)) { | |
517 | opts->ws = ireq->rcv_wscale; | |
7982d5e1 PL |
518 | if(likely(opts->ws)) |
519 | size += TCPOLEN_WSCALE_ALIGNED; | |
33ad798c AL |
520 | } |
521 | if (likely(doing_ts)) { | |
522 | opts->options |= OPTION_TS; | |
523 | opts->tsval = TCP_SKB_CB(skb)->when; | |
524 | opts->tsecr = req->ts_recent; | |
525 | size += TCPOLEN_TSTAMP_ALIGNED; | |
526 | } | |
527 | if (likely(ireq->sack_ok)) { | |
528 | opts->options |= OPTION_SACK_ADVERTISE; | |
529 | if (unlikely(!doing_ts)) | |
530 | size += TCPOLEN_SACKPERM_ALIGNED; | |
531 | } | |
532 | ||
533 | return size; | |
534 | } | |
535 | ||
536 | static unsigned tcp_established_options(struct sock *sk, struct sk_buff *skb, | |
537 | struct tcp_out_options *opts, | |
538 | struct tcp_md5sig_key **md5) { | |
539 | struct tcp_skb_cb *tcb = skb ? TCP_SKB_CB(skb) : NULL; | |
540 | struct tcp_sock *tp = tcp_sk(sk); | |
541 | unsigned size = 0; | |
542 | ||
543 | #ifdef CONFIG_TCP_MD5SIG | |
544 | *md5 = tp->af_specific->md5_lookup(sk, sk); | |
545 | if (unlikely(*md5)) { | |
546 | opts->options |= OPTION_MD5; | |
547 | size += TCPOLEN_MD5SIG_ALIGNED; | |
548 | } | |
549 | #else | |
550 | *md5 = NULL; | |
551 | #endif | |
552 | ||
553 | if (likely(tp->rx_opt.tstamp_ok)) { | |
554 | opts->options |= OPTION_TS; | |
555 | opts->tsval = tcb ? tcb->when : 0; | |
556 | opts->tsecr = tp->rx_opt.ts_recent; | |
557 | size += TCPOLEN_TSTAMP_ALIGNED; | |
558 | } | |
559 | ||
560 | if (unlikely(tp->rx_opt.eff_sacks)) { | |
561 | const unsigned remaining = MAX_TCP_OPTION_SPACE - size; | |
562 | opts->num_sack_blocks = | |
563 | min_t(unsigned, tp->rx_opt.eff_sacks, | |
564 | (remaining - TCPOLEN_SACK_BASE_ALIGNED) / | |
565 | TCPOLEN_SACK_PERBLOCK); | |
566 | size += TCPOLEN_SACK_BASE_ALIGNED + | |
567 | opts->num_sack_blocks * TCPOLEN_SACK_PERBLOCK; | |
568 | } | |
569 | ||
570 | return size; | |
40efc6fa | 571 | } |
1da177e4 LT |
572 | |
573 | /* This routine actually transmits TCP packets queued in by | |
574 | * tcp_do_sendmsg(). This is used by both the initial | |
575 | * transmission and possible later retransmissions. | |
576 | * All SKB's seen here are completely headerless. It is our | |
577 | * job to build the TCP header, and pass the packet down to | |
578 | * IP so it can do the same plus pass the packet off to the | |
579 | * device. | |
580 | * | |
581 | * We are working here with either a clone of the original | |
582 | * SKB, or a fresh unique copy made by the retransmit engine. | |
583 | */ | |
056834d9 IJ |
584 | static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it, |
585 | gfp_t gfp_mask) | |
1da177e4 | 586 | { |
dfb4b9dc DM |
587 | const struct inet_connection_sock *icsk = inet_csk(sk); |
588 | struct inet_sock *inet; | |
589 | struct tcp_sock *tp; | |
590 | struct tcp_skb_cb *tcb; | |
33ad798c AL |
591 | struct tcp_out_options opts; |
592 | unsigned tcp_options_size, tcp_header_size; | |
cfb6eeb4 YH |
593 | struct tcp_md5sig_key *md5; |
594 | __u8 *md5_hash_location; | |
dfb4b9dc | 595 | struct tcphdr *th; |
dfb4b9dc DM |
596 | int err; |
597 | ||
598 | BUG_ON(!skb || !tcp_skb_pcount(skb)); | |
599 | ||
600 | /* If congestion control is doing timestamping, we must | |
601 | * take such a timestamp before we potentially clone/copy. | |
602 | */ | |
164891aa | 603 | if (icsk->icsk_ca_ops->flags & TCP_CONG_RTT_STAMP) |
dfb4b9dc DM |
604 | __net_timestamp(skb); |
605 | ||
606 | if (likely(clone_it)) { | |
607 | if (unlikely(skb_cloned(skb))) | |
608 | skb = pskb_copy(skb, gfp_mask); | |
609 | else | |
610 | skb = skb_clone(skb, gfp_mask); | |
611 | if (unlikely(!skb)) | |
612 | return -ENOBUFS; | |
613 | } | |
1da177e4 | 614 | |
dfb4b9dc DM |
615 | inet = inet_sk(sk); |
616 | tp = tcp_sk(sk); | |
617 | tcb = TCP_SKB_CB(skb); | |
33ad798c | 618 | memset(&opts, 0, sizeof(opts)); |
1da177e4 | 619 | |
33ad798c AL |
620 | if (unlikely(tcb->flags & TCPCB_FLAG_SYN)) |
621 | tcp_options_size = tcp_syn_options(sk, skb, &opts, &md5); | |
622 | else | |
623 | tcp_options_size = tcp_established_options(sk, skb, &opts, | |
624 | &md5); | |
625 | tcp_header_size = tcp_options_size + sizeof(struct tcphdr); | |
e905a9ed | 626 | |
dfb4b9dc DM |
627 | if (tcp_packets_in_flight(tp) == 0) |
628 | tcp_ca_event(sk, CA_EVENT_TX_START); | |
629 | ||
aa8223c7 ACM |
630 | skb_push(skb, tcp_header_size); |
631 | skb_reset_transport_header(skb); | |
e89862f4 | 632 | skb_set_owner_w(skb, sk); |
dfb4b9dc DM |
633 | |
634 | /* Build TCP header and checksum it. */ | |
aa8223c7 | 635 | th = tcp_hdr(skb); |
dfb4b9dc DM |
636 | th->source = inet->sport; |
637 | th->dest = inet->dport; | |
638 | th->seq = htonl(tcb->seq); | |
639 | th->ack_seq = htonl(tp->rcv_nxt); | |
df7a3b07 | 640 | *(((__be16 *)th) + 6) = htons(((tcp_header_size >> 2) << 12) | |
dfb4b9dc DM |
641 | tcb->flags); |
642 | ||
643 | if (unlikely(tcb->flags & TCPCB_FLAG_SYN)) { | |
644 | /* RFC1323: The window in SYN & SYN/ACK segments | |
645 | * is never scaled. | |
646 | */ | |
600ff0c2 | 647 | th->window = htons(min(tp->rcv_wnd, 65535U)); |
dfb4b9dc DM |
648 | } else { |
649 | th->window = htons(tcp_select_window(sk)); | |
650 | } | |
651 | th->check = 0; | |
652 | th->urg_ptr = 0; | |
1da177e4 | 653 | |
33f5f57e IJ |
654 | /* The urg_mode check is necessary during a below snd_una win probe */ |
655 | if (unlikely(tcp_urg_mode(tp) && | |
056834d9 IJ |
656 | between(tp->snd_up, tcb->seq + 1, tcb->seq + 0xFFFF))) { |
657 | th->urg_ptr = htons(tp->snd_up - tcb->seq); | |
dfb4b9dc DM |
658 | th->urg = 1; |
659 | } | |
1da177e4 | 660 | |
33ad798c AL |
661 | tcp_options_write((__be32 *)(th + 1), tp, &opts, &md5_hash_location); |
662 | if (likely((tcb->flags & TCPCB_FLAG_SYN) == 0)) | |
9e412ba7 | 663 | TCP_ECN_send(sk, skb, tcp_header_size); |
1da177e4 | 664 | |
cfb6eeb4 YH |
665 | #ifdef CONFIG_TCP_MD5SIG |
666 | /* Calculate the MD5 hash, as we have all we need now */ | |
667 | if (md5) { | |
33ad798c | 668 | sk->sk_route_caps &= ~NETIF_F_GSO_MASK; |
cfb6eeb4 | 669 | tp->af_specific->calc_md5_hash(md5_hash_location, |
49a72dfb | 670 | md5, sk, NULL, skb); |
cfb6eeb4 YH |
671 | } |
672 | #endif | |
673 | ||
8292a17a | 674 | icsk->icsk_af_ops->send_check(sk, skb->len, skb); |
1da177e4 | 675 | |
dfb4b9dc DM |
676 | if (likely(tcb->flags & TCPCB_FLAG_ACK)) |
677 | tcp_event_ack_sent(sk, tcp_skb_pcount(skb)); | |
1da177e4 | 678 | |
dfb4b9dc DM |
679 | if (skb->len != tcp_header_size) |
680 | tcp_event_data_sent(tp, skb, sk); | |
1da177e4 | 681 | |
bd37a088 | 682 | if (after(tcb->end_seq, tp->snd_nxt) || tcb->seq == tcb->end_seq) |
81cc8a75 | 683 | TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTSEGS); |
1da177e4 | 684 | |
e89862f4 | 685 | err = icsk->icsk_af_ops->queue_xmit(skb, 0); |
83de47cd | 686 | if (likely(err <= 0)) |
dfb4b9dc DM |
687 | return err; |
688 | ||
3cfe3baa | 689 | tcp_enter_cwr(sk, 1); |
dfb4b9dc | 690 | |
b9df3cb8 | 691 | return net_xmit_eval(err); |
1da177e4 LT |
692 | } |
693 | ||
e905a9ed | 694 | /* This routine just queue's the buffer |
1da177e4 LT |
695 | * |
696 | * NOTE: probe0 timer is not checked, do not forget tcp_push_pending_frames, | |
697 | * otherwise socket can stall. | |
698 | */ | |
699 | static void tcp_queue_skb(struct sock *sk, struct sk_buff *skb) | |
700 | { | |
701 | struct tcp_sock *tp = tcp_sk(sk); | |
702 | ||
703 | /* Advance write_seq and place onto the write_queue. */ | |
704 | tp->write_seq = TCP_SKB_CB(skb)->end_seq; | |
705 | skb_header_release(skb); | |
fe067e8a | 706 | tcp_add_write_queue_tail(sk, skb); |
3ab224be HA |
707 | sk->sk_wmem_queued += skb->truesize; |
708 | sk_mem_charge(sk, skb->truesize); | |
1da177e4 LT |
709 | } |
710 | ||
056834d9 IJ |
711 | static void tcp_set_skb_tso_segs(struct sock *sk, struct sk_buff *skb, |
712 | unsigned int mss_now) | |
f6302d1d | 713 | { |
bcd76111 | 714 | if (skb->len <= mss_now || !sk_can_gso(sk)) { |
f6302d1d DM |
715 | /* Avoid the costly divide in the normal |
716 | * non-TSO case. | |
717 | */ | |
7967168c HX |
718 | skb_shinfo(skb)->gso_segs = 1; |
719 | skb_shinfo(skb)->gso_size = 0; | |
720 | skb_shinfo(skb)->gso_type = 0; | |
f6302d1d | 721 | } else { |
356f89e1 | 722 | skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss_now); |
7967168c | 723 | skb_shinfo(skb)->gso_size = mss_now; |
bcd76111 | 724 | skb_shinfo(skb)->gso_type = sk->sk_gso_type; |
1da177e4 LT |
725 | } |
726 | } | |
727 | ||
91fed7a1 | 728 | /* When a modification to fackets out becomes necessary, we need to check |
68f8353b | 729 | * skb is counted to fackets_out or not. |
91fed7a1 | 730 | */ |
a47e5a98 | 731 | static void tcp_adjust_fackets_out(struct sock *sk, struct sk_buff *skb, |
91fed7a1 IJ |
732 | int decr) |
733 | { | |
a47e5a98 IJ |
734 | struct tcp_sock *tp = tcp_sk(sk); |
735 | ||
dc86967b | 736 | if (!tp->sacked_out || tcp_is_reno(tp)) |
91fed7a1 IJ |
737 | return; |
738 | ||
6859d494 | 739 | if (after(tcp_highest_sack_seq(tp), TCP_SKB_CB(skb)->seq)) |
91fed7a1 | 740 | tp->fackets_out -= decr; |
91fed7a1 IJ |
741 | } |
742 | ||
1da177e4 LT |
743 | /* Function to create two new TCP segments. Shrinks the given segment |
744 | * to the specified size and appends a new segment with the rest of the | |
e905a9ed | 745 | * packet to the list. This won't be called frequently, I hope. |
1da177e4 LT |
746 | * Remember, these are still headerless SKBs at this point. |
747 | */ | |
056834d9 IJ |
748 | int tcp_fragment(struct sock *sk, struct sk_buff *skb, u32 len, |
749 | unsigned int mss_now) | |
1da177e4 LT |
750 | { |
751 | struct tcp_sock *tp = tcp_sk(sk); | |
752 | struct sk_buff *buff; | |
6475be16 | 753 | int nsize, old_factor; |
b60b49ea | 754 | int nlen; |
1da177e4 LT |
755 | u16 flags; |
756 | ||
b2cc99f0 | 757 | BUG_ON(len > skb->len); |
6a438bbe | 758 | |
ef9da47c | 759 | tcp_clear_retrans_hints_partial(tp); |
1da177e4 LT |
760 | nsize = skb_headlen(skb) - len; |
761 | if (nsize < 0) | |
762 | nsize = 0; | |
763 | ||
764 | if (skb_cloned(skb) && | |
765 | skb_is_nonlinear(skb) && | |
766 | pskb_expand_head(skb, 0, 0, GFP_ATOMIC)) | |
767 | return -ENOMEM; | |
768 | ||
769 | /* Get a new skb... force flag on. */ | |
770 | buff = sk_stream_alloc_skb(sk, nsize, GFP_ATOMIC); | |
771 | if (buff == NULL) | |
772 | return -ENOMEM; /* We'll just try again later. */ | |
ef5cb973 | 773 | |
3ab224be HA |
774 | sk->sk_wmem_queued += buff->truesize; |
775 | sk_mem_charge(sk, buff->truesize); | |
b60b49ea HX |
776 | nlen = skb->len - len - nsize; |
777 | buff->truesize += nlen; | |
778 | skb->truesize -= nlen; | |
1da177e4 LT |
779 | |
780 | /* Correct the sequence numbers. */ | |
781 | TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len; | |
782 | TCP_SKB_CB(buff)->end_seq = TCP_SKB_CB(skb)->end_seq; | |
783 | TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(buff)->seq; | |
784 | ||
785 | /* PSH and FIN should only be set in the second packet. */ | |
786 | flags = TCP_SKB_CB(skb)->flags; | |
056834d9 | 787 | TCP_SKB_CB(skb)->flags = flags & ~(TCPCB_FLAG_FIN | TCPCB_FLAG_PSH); |
1da177e4 | 788 | TCP_SKB_CB(buff)->flags = flags; |
e14c3caf | 789 | TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked; |
1da177e4 | 790 | |
84fa7933 | 791 | if (!skb_shinfo(skb)->nr_frags && skb->ip_summed != CHECKSUM_PARTIAL) { |
1da177e4 | 792 | /* Copy and checksum data tail into the new buffer. */ |
056834d9 IJ |
793 | buff->csum = csum_partial_copy_nocheck(skb->data + len, |
794 | skb_put(buff, nsize), | |
1da177e4 LT |
795 | nsize, 0); |
796 | ||
797 | skb_trim(skb, len); | |
798 | ||
799 | skb->csum = csum_block_sub(skb->csum, buff->csum, len); | |
800 | } else { | |
84fa7933 | 801 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 LT |
802 | skb_split(skb, buff, len); |
803 | } | |
804 | ||
805 | buff->ip_summed = skb->ip_summed; | |
806 | ||
807 | /* Looks stupid, but our code really uses when of | |
808 | * skbs, which it never sent before. --ANK | |
809 | */ | |
810 | TCP_SKB_CB(buff)->when = TCP_SKB_CB(skb)->when; | |
a61bbcf2 | 811 | buff->tstamp = skb->tstamp; |
1da177e4 | 812 | |
6475be16 DM |
813 | old_factor = tcp_skb_pcount(skb); |
814 | ||
1da177e4 | 815 | /* Fix up tso_factor for both original and new SKB. */ |
846998ae DM |
816 | tcp_set_skb_tso_segs(sk, skb, mss_now); |
817 | tcp_set_skb_tso_segs(sk, buff, mss_now); | |
1da177e4 | 818 | |
6475be16 DM |
819 | /* If this packet has been sent out already, we must |
820 | * adjust the various packet counters. | |
821 | */ | |
cf0b450c | 822 | if (!before(tp->snd_nxt, TCP_SKB_CB(buff)->end_seq)) { |
6475be16 DM |
823 | int diff = old_factor - tcp_skb_pcount(skb) - |
824 | tcp_skb_pcount(buff); | |
1da177e4 | 825 | |
6475be16 | 826 | tp->packets_out -= diff; |
e14c3caf HX |
827 | |
828 | if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) | |
829 | tp->sacked_out -= diff; | |
830 | if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) | |
831 | tp->retrans_out -= diff; | |
832 | ||
b5860bba | 833 | if (TCP_SKB_CB(skb)->sacked & TCPCB_LOST) |
6475be16 | 834 | tp->lost_out -= diff; |
83ca28be | 835 | |
91fed7a1 IJ |
836 | /* Adjust Reno SACK estimate. */ |
837 | if (tcp_is_reno(tp) && diff > 0) { | |
838 | tcp_dec_pcount_approx_int(&tp->sacked_out, diff); | |
839 | tcp_verify_left_out(tp); | |
6475be16 | 840 | } |
a47e5a98 | 841 | tcp_adjust_fackets_out(sk, skb, diff); |
1da177e4 LT |
842 | } |
843 | ||
844 | /* Link BUFF into the send queue. */ | |
f44b5271 | 845 | skb_header_release(buff); |
fe067e8a | 846 | tcp_insert_write_queue_after(skb, buff, sk); |
1da177e4 LT |
847 | |
848 | return 0; | |
849 | } | |
850 | ||
851 | /* This is similar to __pskb_pull_head() (it will go to core/skbuff.c | |
852 | * eventually). The difference is that pulled data not copied, but | |
853 | * immediately discarded. | |
854 | */ | |
f2911969 | 855 | static void __pskb_trim_head(struct sk_buff *skb, int len) |
1da177e4 LT |
856 | { |
857 | int i, k, eat; | |
858 | ||
859 | eat = len; | |
860 | k = 0; | |
056834d9 | 861 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
1da177e4 LT |
862 | if (skb_shinfo(skb)->frags[i].size <= eat) { |
863 | put_page(skb_shinfo(skb)->frags[i].page); | |
864 | eat -= skb_shinfo(skb)->frags[i].size; | |
865 | } else { | |
866 | skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; | |
867 | if (eat) { | |
868 | skb_shinfo(skb)->frags[k].page_offset += eat; | |
869 | skb_shinfo(skb)->frags[k].size -= eat; | |
870 | eat = 0; | |
871 | } | |
872 | k++; | |
873 | } | |
874 | } | |
875 | skb_shinfo(skb)->nr_frags = k; | |
876 | ||
27a884dc | 877 | skb_reset_tail_pointer(skb); |
1da177e4 LT |
878 | skb->data_len -= len; |
879 | skb->len = skb->data_len; | |
1da177e4 LT |
880 | } |
881 | ||
882 | int tcp_trim_head(struct sock *sk, struct sk_buff *skb, u32 len) | |
883 | { | |
056834d9 | 884 | if (skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC)) |
1da177e4 LT |
885 | return -ENOMEM; |
886 | ||
f2911969 HX |
887 | /* If len == headlen, we avoid __skb_pull to preserve alignment. */ |
888 | if (unlikely(len < skb_headlen(skb))) | |
1da177e4 | 889 | __skb_pull(skb, len); |
f2911969 HX |
890 | else |
891 | __pskb_trim_head(skb, len - skb_headlen(skb)); | |
1da177e4 LT |
892 | |
893 | TCP_SKB_CB(skb)->seq += len; | |
84fa7933 | 894 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 LT |
895 | |
896 | skb->truesize -= len; | |
897 | sk->sk_wmem_queued -= len; | |
3ab224be | 898 | sk_mem_uncharge(sk, len); |
1da177e4 LT |
899 | sock_set_flag(sk, SOCK_QUEUE_SHRUNK); |
900 | ||
901 | /* Any change of skb->len requires recalculation of tso | |
902 | * factor and mss. | |
903 | */ | |
904 | if (tcp_skb_pcount(skb) > 1) | |
846998ae | 905 | tcp_set_skb_tso_segs(sk, skb, tcp_current_mss(sk, 1)); |
1da177e4 LT |
906 | |
907 | return 0; | |
908 | } | |
909 | ||
5d424d5a JH |
910 | /* Not accounting for SACKs here. */ |
911 | int tcp_mtu_to_mss(struct sock *sk, int pmtu) | |
912 | { | |
913 | struct tcp_sock *tp = tcp_sk(sk); | |
914 | struct inet_connection_sock *icsk = inet_csk(sk); | |
915 | int mss_now; | |
916 | ||
917 | /* Calculate base mss without TCP options: | |
918 | It is MMS_S - sizeof(tcphdr) of rfc1122 | |
919 | */ | |
920 | mss_now = pmtu - icsk->icsk_af_ops->net_header_len - sizeof(struct tcphdr); | |
921 | ||
922 | /* Clamp it (mss_clamp does not include tcp options) */ | |
923 | if (mss_now > tp->rx_opt.mss_clamp) | |
924 | mss_now = tp->rx_opt.mss_clamp; | |
925 | ||
926 | /* Now subtract optional transport overhead */ | |
927 | mss_now -= icsk->icsk_ext_hdr_len; | |
928 | ||
929 | /* Then reserve room for full set of TCP options and 8 bytes of data */ | |
930 | if (mss_now < 48) | |
931 | mss_now = 48; | |
932 | ||
933 | /* Now subtract TCP options size, not including SACKs */ | |
934 | mss_now -= tp->tcp_header_len - sizeof(struct tcphdr); | |
935 | ||
936 | return mss_now; | |
937 | } | |
938 | ||
939 | /* Inverse of above */ | |
940 | int tcp_mss_to_mtu(struct sock *sk, int mss) | |
941 | { | |
942 | struct tcp_sock *tp = tcp_sk(sk); | |
943 | struct inet_connection_sock *icsk = inet_csk(sk); | |
944 | int mtu; | |
945 | ||
946 | mtu = mss + | |
947 | tp->tcp_header_len + | |
948 | icsk->icsk_ext_hdr_len + | |
949 | icsk->icsk_af_ops->net_header_len; | |
950 | ||
951 | return mtu; | |
952 | } | |
953 | ||
954 | void tcp_mtup_init(struct sock *sk) | |
955 | { | |
956 | struct tcp_sock *tp = tcp_sk(sk); | |
957 | struct inet_connection_sock *icsk = inet_csk(sk); | |
958 | ||
959 | icsk->icsk_mtup.enabled = sysctl_tcp_mtu_probing > 1; | |
960 | icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + sizeof(struct tcphdr) + | |
e905a9ed | 961 | icsk->icsk_af_ops->net_header_len; |
5d424d5a JH |
962 | icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, sysctl_tcp_base_mss); |
963 | icsk->icsk_mtup.probe_size = 0; | |
964 | } | |
965 | ||
409d22b4 IJ |
966 | /* Bound MSS / TSO packet size with the half of the window */ |
967 | static int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) | |
968 | { | |
969 | if (tp->max_window && pktsize > (tp->max_window >> 1)) | |
970 | return max(tp->max_window >> 1, 68U - tp->tcp_header_len); | |
971 | else | |
972 | return pktsize; | |
973 | } | |
974 | ||
1da177e4 LT |
975 | /* This function synchronize snd mss to current pmtu/exthdr set. |
976 | ||
977 | tp->rx_opt.user_mss is mss set by user by TCP_MAXSEG. It does NOT counts | |
978 | for TCP options, but includes only bare TCP header. | |
979 | ||
980 | tp->rx_opt.mss_clamp is mss negotiated at connection setup. | |
caa20d9a | 981 | It is minimum of user_mss and mss received with SYN. |
1da177e4 LT |
982 | It also does not include TCP options. |
983 | ||
d83d8461 | 984 | inet_csk(sk)->icsk_pmtu_cookie is last pmtu, seen by this function. |
1da177e4 LT |
985 | |
986 | tp->mss_cache is current effective sending mss, including | |
987 | all tcp options except for SACKs. It is evaluated, | |
988 | taking into account current pmtu, but never exceeds | |
989 | tp->rx_opt.mss_clamp. | |
990 | ||
991 | NOTE1. rfc1122 clearly states that advertised MSS | |
992 | DOES NOT include either tcp or ip options. | |
993 | ||
d83d8461 ACM |
994 | NOTE2. inet_csk(sk)->icsk_pmtu_cookie and tp->mss_cache |
995 | are READ ONLY outside this function. --ANK (980731) | |
1da177e4 | 996 | */ |
1da177e4 LT |
997 | unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu) |
998 | { | |
999 | struct tcp_sock *tp = tcp_sk(sk); | |
d83d8461 | 1000 | struct inet_connection_sock *icsk = inet_csk(sk); |
5d424d5a | 1001 | int mss_now; |
1da177e4 | 1002 | |
5d424d5a JH |
1003 | if (icsk->icsk_mtup.search_high > pmtu) |
1004 | icsk->icsk_mtup.search_high = pmtu; | |
1da177e4 | 1005 | |
5d424d5a | 1006 | mss_now = tcp_mtu_to_mss(sk, pmtu); |
409d22b4 | 1007 | mss_now = tcp_bound_to_half_wnd(tp, mss_now); |
1da177e4 LT |
1008 | |
1009 | /* And store cached results */ | |
d83d8461 | 1010 | icsk->icsk_pmtu_cookie = pmtu; |
5d424d5a JH |
1011 | if (icsk->icsk_mtup.enabled) |
1012 | mss_now = min(mss_now, tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low)); | |
c1b4a7e6 | 1013 | tp->mss_cache = mss_now; |
1da177e4 LT |
1014 | |
1015 | return mss_now; | |
1016 | } | |
1017 | ||
1018 | /* Compute the current effective MSS, taking SACKs and IP options, | |
1019 | * and even PMTU discovery events into account. | |
1020 | * | |
33f5f57e | 1021 | * LARGESEND note: !tcp_urg_mode is overkill, only frames up to snd_up |
1da177e4 LT |
1022 | * cannot be large. However, taking into account rare use of URG, this |
1023 | * is not a big flaw. | |
1024 | */ | |
c1b4a7e6 | 1025 | unsigned int tcp_current_mss(struct sock *sk, int large_allowed) |
1da177e4 LT |
1026 | { |
1027 | struct tcp_sock *tp = tcp_sk(sk); | |
1028 | struct dst_entry *dst = __sk_dst_get(sk); | |
c1b4a7e6 DM |
1029 | u32 mss_now; |
1030 | u16 xmit_size_goal; | |
1031 | int doing_tso = 0; | |
33ad798c AL |
1032 | unsigned header_len; |
1033 | struct tcp_out_options opts; | |
1034 | struct tcp_md5sig_key *md5; | |
c1b4a7e6 DM |
1035 | |
1036 | mss_now = tp->mss_cache; | |
1037 | ||
33f5f57e | 1038 | if (large_allowed && sk_can_gso(sk) && !tcp_urg_mode(tp)) |
c1b4a7e6 | 1039 | doing_tso = 1; |
1da177e4 | 1040 | |
1da177e4 LT |
1041 | if (dst) { |
1042 | u32 mtu = dst_mtu(dst); | |
d83d8461 | 1043 | if (mtu != inet_csk(sk)->icsk_pmtu_cookie) |
1da177e4 LT |
1044 | mss_now = tcp_sync_mss(sk, mtu); |
1045 | } | |
1046 | ||
33ad798c AL |
1047 | header_len = tcp_established_options(sk, NULL, &opts, &md5) + |
1048 | sizeof(struct tcphdr); | |
1049 | /* The mss_cache is sized based on tp->tcp_header_len, which assumes | |
1050 | * some common options. If this is an odd packet (because we have SACK | |
1051 | * blocks etc) then our calculated header_len will be different, and | |
1052 | * we have to adjust mss_now correspondingly */ | |
1053 | if (header_len != tp->tcp_header_len) { | |
1054 | int delta = (int) header_len - tp->tcp_header_len; | |
1055 | mss_now -= delta; | |
1056 | } | |
cfb6eeb4 | 1057 | |
c1b4a7e6 | 1058 | xmit_size_goal = mss_now; |
1da177e4 | 1059 | |
c1b4a7e6 | 1060 | if (doing_tso) { |
82cc1a7a | 1061 | xmit_size_goal = ((sk->sk_gso_max_size - 1) - |
8292a17a | 1062 | inet_csk(sk)->icsk_af_ops->net_header_len - |
d83d8461 ACM |
1063 | inet_csk(sk)->icsk_ext_hdr_len - |
1064 | tp->tcp_header_len); | |
1da177e4 | 1065 | |
409d22b4 | 1066 | xmit_size_goal = tcp_bound_to_half_wnd(tp, xmit_size_goal); |
c1b4a7e6 | 1067 | xmit_size_goal -= (xmit_size_goal % mss_now); |
1da177e4 | 1068 | } |
c1b4a7e6 | 1069 | tp->xmit_size_goal = xmit_size_goal; |
1da177e4 | 1070 | |
1da177e4 LT |
1071 | return mss_now; |
1072 | } | |
1073 | ||
a762a980 | 1074 | /* Congestion window validation. (RFC2861) */ |
9e412ba7 | 1075 | static void tcp_cwnd_validate(struct sock *sk) |
a762a980 | 1076 | { |
9e412ba7 | 1077 | struct tcp_sock *tp = tcp_sk(sk); |
a762a980 | 1078 | |
d436d686 | 1079 | if (tp->packets_out >= tp->snd_cwnd) { |
a762a980 DM |
1080 | /* Network is feed fully. */ |
1081 | tp->snd_cwnd_used = 0; | |
1082 | tp->snd_cwnd_stamp = tcp_time_stamp; | |
1083 | } else { | |
1084 | /* Network starves. */ | |
1085 | if (tp->packets_out > tp->snd_cwnd_used) | |
1086 | tp->snd_cwnd_used = tp->packets_out; | |
1087 | ||
15d33c07 DM |
1088 | if (sysctl_tcp_slow_start_after_idle && |
1089 | (s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= inet_csk(sk)->icsk_rto) | |
a762a980 DM |
1090 | tcp_cwnd_application_limited(sk); |
1091 | } | |
1092 | } | |
1093 | ||
0e3a4803 IJ |
1094 | /* Returns the portion of skb which can be sent right away without |
1095 | * introducing MSS oddities to segment boundaries. In rare cases where | |
1096 | * mss_now != mss_cache, we will request caller to create a small skb | |
1097 | * per input skb which could be mostly avoided here (if desired). | |
5ea3a748 IJ |
1098 | * |
1099 | * We explicitly want to create a request for splitting write queue tail | |
1100 | * to a small skb for Nagle purposes while avoiding unnecessary modulos, | |
1101 | * thus all the complexity (cwnd_len is always MSS multiple which we | |
1102 | * return whenever allowed by the other factors). Basically we need the | |
1103 | * modulo only when the receiver window alone is the limiting factor or | |
1104 | * when we would be allowed to send the split-due-to-Nagle skb fully. | |
0e3a4803 IJ |
1105 | */ |
1106 | static unsigned int tcp_mss_split_point(struct sock *sk, struct sk_buff *skb, | |
056834d9 | 1107 | unsigned int mss_now, unsigned int cwnd) |
c1b4a7e6 | 1108 | { |
0e3a4803 IJ |
1109 | struct tcp_sock *tp = tcp_sk(sk); |
1110 | u32 needed, window, cwnd_len; | |
c1b4a7e6 | 1111 | |
90840def | 1112 | window = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; |
c1b4a7e6 | 1113 | cwnd_len = mss_now * cwnd; |
0e3a4803 IJ |
1114 | |
1115 | if (likely(cwnd_len <= window && skb != tcp_write_queue_tail(sk))) | |
1116 | return cwnd_len; | |
1117 | ||
5ea3a748 IJ |
1118 | needed = min(skb->len, window); |
1119 | ||
17515408 | 1120 | if (cwnd_len <= needed) |
0e3a4803 IJ |
1121 | return cwnd_len; |
1122 | ||
0e3a4803 | 1123 | return needed - needed % mss_now; |
c1b4a7e6 DM |
1124 | } |
1125 | ||
1126 | /* Can at least one segment of SKB be sent right now, according to the | |
1127 | * congestion window rules? If so, return how many segments are allowed. | |
1128 | */ | |
056834d9 IJ |
1129 | static inline unsigned int tcp_cwnd_test(struct tcp_sock *tp, |
1130 | struct sk_buff *skb) | |
c1b4a7e6 DM |
1131 | { |
1132 | u32 in_flight, cwnd; | |
1133 | ||
1134 | /* Don't be strict about the congestion window for the final FIN. */ | |
104439a8 JH |
1135 | if ((TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN) && |
1136 | tcp_skb_pcount(skb) == 1) | |
c1b4a7e6 DM |
1137 | return 1; |
1138 | ||
1139 | in_flight = tcp_packets_in_flight(tp); | |
1140 | cwnd = tp->snd_cwnd; | |
1141 | if (in_flight < cwnd) | |
1142 | return (cwnd - in_flight); | |
1143 | ||
1144 | return 0; | |
1145 | } | |
1146 | ||
1147 | /* This must be invoked the first time we consider transmitting | |
1148 | * SKB onto the wire. | |
1149 | */ | |
056834d9 IJ |
1150 | static int tcp_init_tso_segs(struct sock *sk, struct sk_buff *skb, |
1151 | unsigned int mss_now) | |
c1b4a7e6 DM |
1152 | { |
1153 | int tso_segs = tcp_skb_pcount(skb); | |
1154 | ||
056834d9 | 1155 | if (!tso_segs || (tso_segs > 1 && tcp_skb_mss(skb) != mss_now)) { |
846998ae | 1156 | tcp_set_skb_tso_segs(sk, skb, mss_now); |
c1b4a7e6 DM |
1157 | tso_segs = tcp_skb_pcount(skb); |
1158 | } | |
1159 | return tso_segs; | |
1160 | } | |
1161 | ||
1162 | static inline int tcp_minshall_check(const struct tcp_sock *tp) | |
1163 | { | |
1164 | return after(tp->snd_sml,tp->snd_una) && | |
1165 | !after(tp->snd_sml, tp->snd_nxt); | |
1166 | } | |
1167 | ||
1168 | /* Return 0, if packet can be sent now without violation Nagle's rules: | |
1169 | * 1. It is full sized. | |
1170 | * 2. Or it contains FIN. (already checked by caller) | |
1171 | * 3. Or TCP_NODELAY was set. | |
1172 | * 4. Or TCP_CORK is not set, and all sent packets are ACKed. | |
1173 | * With Minshall's modification: all sent small packets are ACKed. | |
1174 | */ | |
c1b4a7e6 | 1175 | static inline int tcp_nagle_check(const struct tcp_sock *tp, |
e905a9ed | 1176 | const struct sk_buff *skb, |
c1b4a7e6 DM |
1177 | unsigned mss_now, int nonagle) |
1178 | { | |
1179 | return (skb->len < mss_now && | |
056834d9 IJ |
1180 | ((nonagle & TCP_NAGLE_CORK) || |
1181 | (!nonagle && tp->packets_out && tcp_minshall_check(tp)))); | |
c1b4a7e6 DM |
1182 | } |
1183 | ||
1184 | /* Return non-zero if the Nagle test allows this packet to be | |
1185 | * sent now. | |
1186 | */ | |
1187 | static inline int tcp_nagle_test(struct tcp_sock *tp, struct sk_buff *skb, | |
1188 | unsigned int cur_mss, int nonagle) | |
1189 | { | |
1190 | /* Nagle rule does not apply to frames, which sit in the middle of the | |
1191 | * write_queue (they have no chances to get new data). | |
1192 | * | |
1193 | * This is implemented in the callers, where they modify the 'nonagle' | |
1194 | * argument based upon the location of SKB in the send queue. | |
1195 | */ | |
1196 | if (nonagle & TCP_NAGLE_PUSH) | |
1197 | return 1; | |
1198 | ||
d551e454 IJ |
1199 | /* Don't use the nagle rule for urgent data (or for the final FIN). |
1200 | * Nagle can be ignored during F-RTO too (see RFC4138). | |
1201 | */ | |
33f5f57e | 1202 | if (tcp_urg_mode(tp) || (tp->frto_counter == 2) || |
c1b4a7e6 DM |
1203 | (TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN)) |
1204 | return 1; | |
1205 | ||
1206 | if (!tcp_nagle_check(tp, skb, cur_mss, nonagle)) | |
1207 | return 1; | |
1208 | ||
1209 | return 0; | |
1210 | } | |
1211 | ||
1212 | /* Does at least the first segment of SKB fit into the send window? */ | |
056834d9 IJ |
1213 | static inline int tcp_snd_wnd_test(struct tcp_sock *tp, struct sk_buff *skb, |
1214 | unsigned int cur_mss) | |
c1b4a7e6 DM |
1215 | { |
1216 | u32 end_seq = TCP_SKB_CB(skb)->end_seq; | |
1217 | ||
1218 | if (skb->len > cur_mss) | |
1219 | end_seq = TCP_SKB_CB(skb)->seq + cur_mss; | |
1220 | ||
90840def | 1221 | return !after(end_seq, tcp_wnd_end(tp)); |
c1b4a7e6 DM |
1222 | } |
1223 | ||
fe067e8a | 1224 | /* This checks if the data bearing packet SKB (usually tcp_send_head(sk)) |
c1b4a7e6 DM |
1225 | * should be put on the wire right now. If so, it returns the number of |
1226 | * packets allowed by the congestion window. | |
1227 | */ | |
1228 | static unsigned int tcp_snd_test(struct sock *sk, struct sk_buff *skb, | |
1229 | unsigned int cur_mss, int nonagle) | |
1230 | { | |
1231 | struct tcp_sock *tp = tcp_sk(sk); | |
1232 | unsigned int cwnd_quota; | |
1233 | ||
846998ae | 1234 | tcp_init_tso_segs(sk, skb, cur_mss); |
c1b4a7e6 DM |
1235 | |
1236 | if (!tcp_nagle_test(tp, skb, cur_mss, nonagle)) | |
1237 | return 0; | |
1238 | ||
1239 | cwnd_quota = tcp_cwnd_test(tp, skb); | |
056834d9 | 1240 | if (cwnd_quota && !tcp_snd_wnd_test(tp, skb, cur_mss)) |
c1b4a7e6 DM |
1241 | cwnd_quota = 0; |
1242 | ||
1243 | return cwnd_quota; | |
1244 | } | |
1245 | ||
9e412ba7 | 1246 | int tcp_may_send_now(struct sock *sk) |
c1b4a7e6 | 1247 | { |
9e412ba7 | 1248 | struct tcp_sock *tp = tcp_sk(sk); |
fe067e8a | 1249 | struct sk_buff *skb = tcp_send_head(sk); |
c1b4a7e6 DM |
1250 | |
1251 | return (skb && | |
1252 | tcp_snd_test(sk, skb, tcp_current_mss(sk, 1), | |
1253 | (tcp_skb_is_last(sk, skb) ? | |