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> | |
5a0e3ad6 | 40 | #include <linux/gfp.h> |
1da177e4 | 41 | #include <linux/module.h> |
1da177e4 LT |
42 | |
43 | /* People can turn this off for buggy TCP's found in printers etc. */ | |
ab32ea5d | 44 | int sysctl_tcp_retrans_collapse __read_mostly = 1; |
1da177e4 | 45 | |
09cb105e | 46 | /* People can turn this on to work with those rare, broken TCPs that |
15d99e02 RJ |
47 | * interpret the window field as a signed quantity. |
48 | */ | |
ab32ea5d | 49 | int sysctl_tcp_workaround_signed_windows __read_mostly = 0; |
15d99e02 | 50 | |
1da177e4 LT |
51 | /* This limits the percentage of the congestion window which we |
52 | * will allow a single TSO frame to consume. Building TSO frames | |
53 | * which are too large can cause TCP streams to be bursty. | |
54 | */ | |
ab32ea5d | 55 | int sysctl_tcp_tso_win_divisor __read_mostly = 3; |
1da177e4 | 56 | |
ab32ea5d BH |
57 | int sysctl_tcp_mtu_probing __read_mostly = 0; |
58 | int sysctl_tcp_base_mss __read_mostly = 512; | |
5d424d5a | 59 | |
35089bb2 | 60 | /* By default, RFC2861 behavior. */ |
ab32ea5d | 61 | int sysctl_tcp_slow_start_after_idle __read_mostly = 1; |
35089bb2 | 62 | |
519855c5 | 63 | int sysctl_tcp_cookie_size __read_mostly = 0; /* TCP_COOKIE_MAX */ |
e6b09cca | 64 | EXPORT_SYMBOL_GPL(sysctl_tcp_cookie_size); |
519855c5 WAS |
65 | |
66 | ||
67edfef7 | 67 | /* Account for new data that has been sent to the network. */ |
66f5fe62 | 68 | static void tcp_event_new_data_sent(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 69 | { |
9e412ba7 | 70 | struct tcp_sock *tp = tcp_sk(sk); |
66f5fe62 | 71 | unsigned int prior_packets = tp->packets_out; |
9e412ba7 | 72 | |
fe067e8a | 73 | tcp_advance_send_head(sk, skb); |
1da177e4 | 74 | tp->snd_nxt = TCP_SKB_CB(skb)->end_seq; |
8512430e IJ |
75 | |
76 | /* Don't override Nagle indefinately with F-RTO */ | |
77 | if (tp->frto_counter == 2) | |
78 | tp->frto_counter = 3; | |
66f5fe62 IJ |
79 | |
80 | tp->packets_out += tcp_skb_pcount(skb); | |
81 | if (!prior_packets) | |
82 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, | |
83 | inet_csk(sk)->icsk_rto, TCP_RTO_MAX); | |
1da177e4 LT |
84 | } |
85 | ||
86 | /* SND.NXT, if window was not shrunk. | |
87 | * If window has been shrunk, what should we make? It is not clear at all. | |
88 | * Using SND.UNA we will fail to open window, SND.NXT is out of window. :-( | |
89 | * Anything in between SND.UNA...SND.UNA+SND.WND also can be already | |
90 | * invalid. OK, let's make this for now: | |
91 | */ | |
9e412ba7 | 92 | static inline __u32 tcp_acceptable_seq(struct sock *sk) |
1da177e4 | 93 | { |
9e412ba7 IJ |
94 | struct tcp_sock *tp = tcp_sk(sk); |
95 | ||
90840def | 96 | if (!before(tcp_wnd_end(tp), tp->snd_nxt)) |
1da177e4 LT |
97 | return tp->snd_nxt; |
98 | else | |
90840def | 99 | return tcp_wnd_end(tp); |
1da177e4 LT |
100 | } |
101 | ||
102 | /* Calculate mss to advertise in SYN segment. | |
103 | * RFC1122, RFC1063, draft-ietf-tcpimpl-pmtud-01 state that: | |
104 | * | |
105 | * 1. It is independent of path mtu. | |
106 | * 2. Ideally, it is maximal possible segment size i.e. 65535-40. | |
107 | * 3. For IPv4 it is reasonable to calculate it from maximal MTU of | |
108 | * attached devices, because some buggy hosts are confused by | |
109 | * large MSS. | |
110 | * 4. We do not make 3, we advertise MSS, calculated from first | |
111 | * hop device mtu, but allow to raise it to ip_rt_min_advmss. | |
112 | * This may be overridden via information stored in routing table. | |
113 | * 5. Value 65535 for MSS is valid in IPv6 and means "as large as possible, | |
114 | * probably even Jumbo". | |
115 | */ | |
116 | static __u16 tcp_advertise_mss(struct sock *sk) | |
117 | { | |
118 | struct tcp_sock *tp = tcp_sk(sk); | |
119 | struct dst_entry *dst = __sk_dst_get(sk); | |
120 | int mss = tp->advmss; | |
121 | ||
122 | if (dst && dst_metric(dst, RTAX_ADVMSS) < mss) { | |
123 | mss = dst_metric(dst, RTAX_ADVMSS); | |
124 | tp->advmss = mss; | |
125 | } | |
126 | ||
127 | return (__u16)mss; | |
128 | } | |
129 | ||
130 | /* RFC2861. Reset CWND after idle period longer RTO to "restart window". | |
131 | * This is the first part of cwnd validation mechanism. */ | |
463c84b9 | 132 | static void tcp_cwnd_restart(struct sock *sk, struct dst_entry *dst) |
1da177e4 | 133 | { |
463c84b9 | 134 | struct tcp_sock *tp = tcp_sk(sk); |
1da177e4 LT |
135 | s32 delta = tcp_time_stamp - tp->lsndtime; |
136 | u32 restart_cwnd = tcp_init_cwnd(tp, dst); | |
137 | u32 cwnd = tp->snd_cwnd; | |
138 | ||
6687e988 | 139 | tcp_ca_event(sk, CA_EVENT_CWND_RESTART); |
1da177e4 | 140 | |
6687e988 | 141 | tp->snd_ssthresh = tcp_current_ssthresh(sk); |
1da177e4 LT |
142 | restart_cwnd = min(restart_cwnd, cwnd); |
143 | ||
463c84b9 | 144 | while ((delta -= inet_csk(sk)->icsk_rto) > 0 && cwnd > restart_cwnd) |
1da177e4 LT |
145 | cwnd >>= 1; |
146 | tp->snd_cwnd = max(cwnd, restart_cwnd); | |
147 | tp->snd_cwnd_stamp = tcp_time_stamp; | |
148 | tp->snd_cwnd_used = 0; | |
149 | } | |
150 | ||
67edfef7 | 151 | /* Congestion state accounting after a packet has been sent. */ |
40efc6fa SH |
152 | static void tcp_event_data_sent(struct tcp_sock *tp, |
153 | struct sk_buff *skb, struct sock *sk) | |
1da177e4 | 154 | { |
463c84b9 ACM |
155 | struct inet_connection_sock *icsk = inet_csk(sk); |
156 | const u32 now = tcp_time_stamp; | |
1da177e4 | 157 | |
35089bb2 DM |
158 | if (sysctl_tcp_slow_start_after_idle && |
159 | (!tp->packets_out && (s32)(now - tp->lsndtime) > icsk->icsk_rto)) | |
463c84b9 | 160 | tcp_cwnd_restart(sk, __sk_dst_get(sk)); |
1da177e4 LT |
161 | |
162 | tp->lsndtime = now; | |
163 | ||
164 | /* If it is a reply for ato after last received | |
165 | * packet, enter pingpong mode. | |
166 | */ | |
463c84b9 ACM |
167 | if ((u32)(now - icsk->icsk_ack.lrcvtime) < icsk->icsk_ack.ato) |
168 | icsk->icsk_ack.pingpong = 1; | |
1da177e4 LT |
169 | } |
170 | ||
67edfef7 | 171 | /* Account for an ACK we sent. */ |
40efc6fa | 172 | static inline void tcp_event_ack_sent(struct sock *sk, unsigned int pkts) |
1da177e4 | 173 | { |
463c84b9 ACM |
174 | tcp_dec_quickack_mode(sk, pkts); |
175 | inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK); | |
1da177e4 LT |
176 | } |
177 | ||
178 | /* Determine a window scaling and initial window to offer. | |
179 | * Based on the assumption that the given amount of space | |
180 | * will be offered. Store the results in the tp structure. | |
181 | * NOTE: for smooth operation initial space offering should | |
182 | * be a multiple of mss if possible. We assume here that mss >= 1. | |
183 | * This MUST be enforced by all callers. | |
184 | */ | |
185 | void tcp_select_initial_window(int __space, __u32 mss, | |
186 | __u32 *rcv_wnd, __u32 *window_clamp, | |
31d12926 | 187 | int wscale_ok, __u8 *rcv_wscale, |
188 | __u32 init_rcv_wnd) | |
1da177e4 LT |
189 | { |
190 | unsigned int space = (__space < 0 ? 0 : __space); | |
191 | ||
192 | /* If no clamp set the clamp to the max possible scaled window */ | |
193 | if (*window_clamp == 0) | |
194 | (*window_clamp) = (65535 << 14); | |
195 | space = min(*window_clamp, space); | |
196 | ||
197 | /* Quantize space offering to a multiple of mss if possible. */ | |
198 | if (space > mss) | |
199 | space = (space / mss) * mss; | |
200 | ||
201 | /* NOTE: offering an initial window larger than 32767 | |
15d99e02 RJ |
202 | * will break some buggy TCP stacks. If the admin tells us |
203 | * it is likely we could be speaking with such a buggy stack | |
204 | * we will truncate our initial window offering to 32K-1 | |
205 | * unless the remote has sent us a window scaling option, | |
206 | * which we interpret as a sign the remote TCP is not | |
207 | * misinterpreting the window field as a signed quantity. | |
1da177e4 | 208 | */ |
15d99e02 RJ |
209 | if (sysctl_tcp_workaround_signed_windows) |
210 | (*rcv_wnd) = min(space, MAX_TCP_WINDOW); | |
211 | else | |
212 | (*rcv_wnd) = space; | |
213 | ||
1da177e4 LT |
214 | (*rcv_wscale) = 0; |
215 | if (wscale_ok) { | |
216 | /* Set window scaling on max possible window | |
e905a9ed | 217 | * See RFC1323 for an explanation of the limit to 14 |
1da177e4 LT |
218 | */ |
219 | space = max_t(u32, sysctl_tcp_rmem[2], sysctl_rmem_max); | |
316c1592 | 220 | space = min_t(u32, space, *window_clamp); |
1da177e4 LT |
221 | while (space > 65535 && (*rcv_wscale) < 14) { |
222 | space >>= 1; | |
223 | (*rcv_wscale)++; | |
224 | } | |
225 | } | |
226 | ||
227 | /* Set initial window to value enough for senders, | |
6b251858 | 228 | * following RFC2414. Senders, not following this RFC, |
1da177e4 LT |
229 | * will be satisfied with 2. |
230 | */ | |
056834d9 | 231 | if (mss > (1 << *rcv_wscale)) { |
01ff367e | 232 | int init_cwnd = 4; |
056834d9 | 233 | if (mss > 1460 * 3) |
1da177e4 | 234 | init_cwnd = 2; |
01ff367e DM |
235 | else if (mss > 1460) |
236 | init_cwnd = 3; | |
31d12926 | 237 | /* when initializing use the value from init_rcv_wnd |
238 | * rather than the default from above | |
239 | */ | |
240 | if (init_rcv_wnd && | |
241 | (*rcv_wnd > init_rcv_wnd * mss)) | |
242 | *rcv_wnd = init_rcv_wnd * mss; | |
243 | else if (*rcv_wnd > init_cwnd * mss) | |
056834d9 | 244 | *rcv_wnd = init_cwnd * mss; |
1da177e4 LT |
245 | } |
246 | ||
247 | /* Set the clamp no higher than max representable value */ | |
248 | (*window_clamp) = min(65535U << (*rcv_wscale), *window_clamp); | |
249 | } | |
250 | ||
251 | /* Chose a new window to advertise, update state in tcp_sock for the | |
252 | * socket, and return result with RFC1323 scaling applied. The return | |
253 | * value can be stuffed directly into th->window for an outgoing | |
254 | * frame. | |
255 | */ | |
40efc6fa | 256 | static u16 tcp_select_window(struct sock *sk) |
1da177e4 LT |
257 | { |
258 | struct tcp_sock *tp = tcp_sk(sk); | |
259 | u32 cur_win = tcp_receive_window(tp); | |
260 | u32 new_win = __tcp_select_window(sk); | |
261 | ||
262 | /* Never shrink the offered window */ | |
2de979bd | 263 | if (new_win < cur_win) { |
1da177e4 LT |
264 | /* Danger Will Robinson! |
265 | * Don't update rcv_wup/rcv_wnd here or else | |
266 | * we will not be able to advertise a zero | |
267 | * window in time. --DaveM | |
268 | * | |
269 | * Relax Will Robinson. | |
270 | */ | |
607bfbf2 | 271 | new_win = ALIGN(cur_win, 1 << tp->rx_opt.rcv_wscale); |
1da177e4 LT |
272 | } |
273 | tp->rcv_wnd = new_win; | |
274 | tp->rcv_wup = tp->rcv_nxt; | |
275 | ||
276 | /* Make sure we do not exceed the maximum possible | |
277 | * scaled window. | |
278 | */ | |
15d99e02 | 279 | if (!tp->rx_opt.rcv_wscale && sysctl_tcp_workaround_signed_windows) |
1da177e4 LT |
280 | new_win = min(new_win, MAX_TCP_WINDOW); |
281 | else | |
282 | new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale)); | |
283 | ||
284 | /* RFC1323 scaling applied */ | |
285 | new_win >>= tp->rx_opt.rcv_wscale; | |
286 | ||
287 | /* If we advertise zero window, disable fast path. */ | |
288 | if (new_win == 0) | |
289 | tp->pred_flags = 0; | |
290 | ||
291 | return new_win; | |
292 | } | |
293 | ||
67edfef7 | 294 | /* Packet ECN state for a SYN-ACK */ |
056834d9 | 295 | static inline void TCP_ECN_send_synack(struct tcp_sock *tp, struct sk_buff *skb) |
bdf1ee5d IJ |
296 | { |
297 | TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_CWR; | |
056834d9 | 298 | if (!(tp->ecn_flags & TCP_ECN_OK)) |
bdf1ee5d IJ |
299 | TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_ECE; |
300 | } | |
301 | ||
67edfef7 | 302 | /* Packet ECN state for a SYN. */ |
bdf1ee5d IJ |
303 | static inline void TCP_ECN_send_syn(struct sock *sk, struct sk_buff *skb) |
304 | { | |
305 | struct tcp_sock *tp = tcp_sk(sk); | |
306 | ||
307 | tp->ecn_flags = 0; | |
255cac91 | 308 | if (sysctl_tcp_ecn == 1) { |
056834d9 | 309 | TCP_SKB_CB(skb)->flags |= TCPCB_FLAG_ECE | TCPCB_FLAG_CWR; |
bdf1ee5d IJ |
310 | tp->ecn_flags = TCP_ECN_OK; |
311 | } | |
312 | } | |
313 | ||
314 | static __inline__ void | |
315 | TCP_ECN_make_synack(struct request_sock *req, struct tcphdr *th) | |
316 | { | |
317 | if (inet_rsk(req)->ecn_ok) | |
318 | th->ece = 1; | |
319 | } | |
320 | ||
67edfef7 AK |
321 | /* Set up ECN state for a packet on a ESTABLISHED socket that is about to |
322 | * be sent. | |
323 | */ | |
bdf1ee5d IJ |
324 | static inline void TCP_ECN_send(struct sock *sk, struct sk_buff *skb, |
325 | int tcp_header_len) | |
326 | { | |
327 | struct tcp_sock *tp = tcp_sk(sk); | |
328 | ||
329 | if (tp->ecn_flags & TCP_ECN_OK) { | |
330 | /* Not-retransmitted data segment: set ECT and inject CWR. */ | |
331 | if (skb->len != tcp_header_len && | |
332 | !before(TCP_SKB_CB(skb)->seq, tp->snd_nxt)) { | |
333 | INET_ECN_xmit(sk); | |
056834d9 | 334 | if (tp->ecn_flags & TCP_ECN_QUEUE_CWR) { |
bdf1ee5d IJ |
335 | tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR; |
336 | tcp_hdr(skb)->cwr = 1; | |
337 | skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; | |
338 | } | |
339 | } else { | |
340 | /* ACK or retransmitted segment: clear ECT|CE */ | |
341 | INET_ECN_dontxmit(sk); | |
342 | } | |
343 | if (tp->ecn_flags & TCP_ECN_DEMAND_CWR) | |
344 | tcp_hdr(skb)->ece = 1; | |
345 | } | |
346 | } | |
347 | ||
e870a8ef IJ |
348 | /* Constructs common control bits of non-data skb. If SYN/FIN is present, |
349 | * auto increment end seqno. | |
350 | */ | |
351 | static void tcp_init_nondata_skb(struct sk_buff *skb, u32 seq, u8 flags) | |
352 | { | |
2e8e18ef | 353 | skb->ip_summed = CHECKSUM_PARTIAL; |
e870a8ef IJ |
354 | skb->csum = 0; |
355 | ||
356 | TCP_SKB_CB(skb)->flags = flags; | |
357 | TCP_SKB_CB(skb)->sacked = 0; | |
358 | ||
359 | skb_shinfo(skb)->gso_segs = 1; | |
360 | skb_shinfo(skb)->gso_size = 0; | |
361 | skb_shinfo(skb)->gso_type = 0; | |
362 | ||
363 | TCP_SKB_CB(skb)->seq = seq; | |
364 | if (flags & (TCPCB_FLAG_SYN | TCPCB_FLAG_FIN)) | |
365 | seq++; | |
366 | TCP_SKB_CB(skb)->end_seq = seq; | |
367 | } | |
368 | ||
33f5f57e IJ |
369 | static inline int tcp_urg_mode(const struct tcp_sock *tp) |
370 | { | |
371 | return tp->snd_una != tp->snd_up; | |
372 | } | |
373 | ||
33ad798c AL |
374 | #define OPTION_SACK_ADVERTISE (1 << 0) |
375 | #define OPTION_TS (1 << 1) | |
376 | #define OPTION_MD5 (1 << 2) | |
89e95a61 | 377 | #define OPTION_WSCALE (1 << 3) |
bd0388ae | 378 | #define OPTION_COOKIE_EXTENSION (1 << 4) |
33ad798c AL |
379 | |
380 | struct tcp_out_options { | |
381 | u8 options; /* bit field of OPTION_* */ | |
382 | u8 ws; /* window scale, 0 to disable */ | |
383 | u8 num_sack_blocks; /* number of SACK blocks to include */ | |
bd0388ae | 384 | u8 hash_size; /* bytes in hash_location */ |
33ad798c AL |
385 | u16 mss; /* 0 to disable */ |
386 | __u32 tsval, tsecr; /* need to include OPTION_TS */ | |
bd0388ae | 387 | __u8 *hash_location; /* temporary pointer, overloaded */ |
33ad798c AL |
388 | }; |
389 | ||
bd0388ae WAS |
390 | /* The sysctl int routines are generic, so check consistency here. |
391 | */ | |
392 | static u8 tcp_cookie_size_check(u8 desired) | |
393 | { | |
394 | if (desired > 0) { | |
395 | /* previously specified */ | |
396 | return desired; | |
397 | } | |
398 | if (sysctl_tcp_cookie_size <= 0) { | |
399 | /* no default specified */ | |
400 | return 0; | |
401 | } | |
402 | if (sysctl_tcp_cookie_size <= TCP_COOKIE_MIN) { | |
403 | /* value too small, specify minimum */ | |
404 | return TCP_COOKIE_MIN; | |
405 | } | |
406 | if (sysctl_tcp_cookie_size >= TCP_COOKIE_MAX) { | |
407 | /* value too large, specify maximum */ | |
408 | return TCP_COOKIE_MAX; | |
409 | } | |
410 | if (0x1 & sysctl_tcp_cookie_size) { | |
411 | /* 8-bit multiple, illegal, fix it */ | |
412 | return (u8)(sysctl_tcp_cookie_size + 0x1); | |
413 | } | |
414 | return (u8)sysctl_tcp_cookie_size; | |
415 | } | |
416 | ||
67edfef7 AK |
417 | /* Write previously computed TCP options to the packet. |
418 | * | |
419 | * Beware: Something in the Internet is very sensitive to the ordering of | |
fd6149d3 IJ |
420 | * TCP options, we learned this through the hard way, so be careful here. |
421 | * Luckily we can at least blame others for their non-compliance but from | |
422 | * inter-operatibility perspective it seems that we're somewhat stuck with | |
423 | * the ordering which we have been using if we want to keep working with | |
424 | * those broken things (not that it currently hurts anybody as there isn't | |
425 | * particular reason why the ordering would need to be changed). | |
426 | * | |
427 | * At least SACK_PERM as the first option is known to lead to a disaster | |
428 | * (but it may well be that other scenarios fail similarly). | |
429 | */ | |
33ad798c | 430 | static void tcp_options_write(__be32 *ptr, struct tcp_sock *tp, |
bd0388ae WAS |
431 | struct tcp_out_options *opts) |
432 | { | |
433 | u8 options = opts->options; /* mungable copy */ | |
434 | ||
435 | /* Having both authentication and cookies for security is redundant, | |
436 | * and there's certainly not enough room. Instead, the cookie-less | |
437 | * extension variant is proposed. | |
438 | * | |
439 | * Consider the pessimal case with authentication. The options | |
440 | * could look like: | |
441 | * COOKIE|MD5(20) + MSS(4) + SACK|TS(12) + WSCALE(4) == 40 | |
442 | */ | |
443 | if (unlikely(OPTION_MD5 & options)) { | |
444 | if (unlikely(OPTION_COOKIE_EXTENSION & options)) { | |
445 | *ptr++ = htonl((TCPOPT_COOKIE << 24) | | |
446 | (TCPOLEN_COOKIE_BASE << 16) | | |
447 | (TCPOPT_MD5SIG << 8) | | |
448 | TCPOLEN_MD5SIG); | |
449 | } else { | |
450 | *ptr++ = htonl((TCPOPT_NOP << 24) | | |
451 | (TCPOPT_NOP << 16) | | |
452 | (TCPOPT_MD5SIG << 8) | | |
453 | TCPOLEN_MD5SIG); | |
454 | } | |
455 | options &= ~OPTION_COOKIE_EXTENSION; | |
456 | /* overload cookie hash location */ | |
457 | opts->hash_location = (__u8 *)ptr; | |
33ad798c | 458 | ptr += 4; |
40efc6fa | 459 | } |
33ad798c | 460 | |
fd6149d3 IJ |
461 | if (unlikely(opts->mss)) { |
462 | *ptr++ = htonl((TCPOPT_MSS << 24) | | |
463 | (TCPOLEN_MSS << 16) | | |
464 | opts->mss); | |
465 | } | |
466 | ||
bd0388ae WAS |
467 | if (likely(OPTION_TS & options)) { |
468 | if (unlikely(OPTION_SACK_ADVERTISE & options)) { | |
33ad798c AL |
469 | *ptr++ = htonl((TCPOPT_SACK_PERM << 24) | |
470 | (TCPOLEN_SACK_PERM << 16) | | |
471 | (TCPOPT_TIMESTAMP << 8) | | |
472 | TCPOLEN_TIMESTAMP); | |
bd0388ae | 473 | options &= ~OPTION_SACK_ADVERTISE; |
33ad798c AL |
474 | } else { |
475 | *ptr++ = htonl((TCPOPT_NOP << 24) | | |
476 | (TCPOPT_NOP << 16) | | |
477 | (TCPOPT_TIMESTAMP << 8) | | |
478 | TCPOLEN_TIMESTAMP); | |
479 | } | |
480 | *ptr++ = htonl(opts->tsval); | |
481 | *ptr++ = htonl(opts->tsecr); | |
482 | } | |
483 | ||
bd0388ae WAS |
484 | /* Specification requires after timestamp, so do it now. |
485 | * | |
486 | * Consider the pessimal case without authentication. The options | |
487 | * could look like: | |
488 | * MSS(4) + SACK|TS(12) + COOKIE(20) + WSCALE(4) == 40 | |
489 | */ | |
490 | if (unlikely(OPTION_COOKIE_EXTENSION & options)) { | |
491 | __u8 *cookie_copy = opts->hash_location; | |
492 | u8 cookie_size = opts->hash_size; | |
493 | ||
494 | /* 8-bit multiple handled in tcp_cookie_size_check() above, | |
495 | * and elsewhere. | |
496 | */ | |
497 | if (0x2 & cookie_size) { | |
498 | __u8 *p = (__u8 *)ptr; | |
499 | ||
500 | /* 16-bit multiple */ | |
501 | *p++ = TCPOPT_COOKIE; | |
502 | *p++ = TCPOLEN_COOKIE_BASE + cookie_size; | |
503 | *p++ = *cookie_copy++; | |
504 | *p++ = *cookie_copy++; | |
505 | ptr++; | |
506 | cookie_size -= 2; | |
507 | } else { | |
508 | /* 32-bit multiple */ | |
509 | *ptr++ = htonl(((TCPOPT_NOP << 24) | | |
510 | (TCPOPT_NOP << 16) | | |
511 | (TCPOPT_COOKIE << 8) | | |
512 | TCPOLEN_COOKIE_BASE) + | |
513 | cookie_size); | |
514 | } | |
515 | ||
516 | if (cookie_size > 0) { | |
517 | memcpy(ptr, cookie_copy, cookie_size); | |
518 | ptr += (cookie_size / 4); | |
519 | } | |
520 | } | |
521 | ||
522 | if (unlikely(OPTION_SACK_ADVERTISE & options)) { | |
33ad798c AL |
523 | *ptr++ = htonl((TCPOPT_NOP << 24) | |
524 | (TCPOPT_NOP << 16) | | |
525 | (TCPOPT_SACK_PERM << 8) | | |
526 | TCPOLEN_SACK_PERM); | |
527 | } | |
528 | ||
bd0388ae | 529 | if (unlikely(OPTION_WSCALE & options)) { |
33ad798c AL |
530 | *ptr++ = htonl((TCPOPT_NOP << 24) | |
531 | (TCPOPT_WINDOW << 16) | | |
532 | (TCPOLEN_WINDOW << 8) | | |
533 | opts->ws); | |
534 | } | |
535 | ||
536 | if (unlikely(opts->num_sack_blocks)) { | |
537 | struct tcp_sack_block *sp = tp->rx_opt.dsack ? | |
538 | tp->duplicate_sack : tp->selective_acks; | |
40efc6fa SH |
539 | int this_sack; |
540 | ||
541 | *ptr++ = htonl((TCPOPT_NOP << 24) | | |
542 | (TCPOPT_NOP << 16) | | |
543 | (TCPOPT_SACK << 8) | | |
33ad798c | 544 | (TCPOLEN_SACK_BASE + (opts->num_sack_blocks * |
40efc6fa | 545 | TCPOLEN_SACK_PERBLOCK))); |
2de979bd | 546 | |
33ad798c AL |
547 | for (this_sack = 0; this_sack < opts->num_sack_blocks; |
548 | ++this_sack) { | |
40efc6fa SH |
549 | *ptr++ = htonl(sp[this_sack].start_seq); |
550 | *ptr++ = htonl(sp[this_sack].end_seq); | |
551 | } | |
2de979bd | 552 | |
5861f8e5 | 553 | tp->rx_opt.dsack = 0; |
40efc6fa | 554 | } |
33ad798c AL |
555 | } |
556 | ||
67edfef7 AK |
557 | /* Compute TCP options for SYN packets. This is not the final |
558 | * network wire format yet. | |
559 | */ | |
33ad798c AL |
560 | static unsigned tcp_syn_options(struct sock *sk, struct sk_buff *skb, |
561 | struct tcp_out_options *opts, | |
562 | struct tcp_md5sig_key **md5) { | |
563 | struct tcp_sock *tp = tcp_sk(sk); | |
bd0388ae | 564 | struct tcp_cookie_values *cvp = tp->cookie_values; |
bd0388ae WAS |
565 | unsigned remaining = MAX_TCP_OPTION_SPACE; |
566 | u8 cookie_size = (!tp->rx_opt.cookie_out_never && cvp != NULL) ? | |
567 | tcp_cookie_size_check(cvp->cookie_desired) : | |
568 | 0; | |
33ad798c | 569 | |
cfb6eeb4 | 570 | #ifdef CONFIG_TCP_MD5SIG |
33ad798c AL |
571 | *md5 = tp->af_specific->md5_lookup(sk, sk); |
572 | if (*md5) { | |
573 | opts->options |= OPTION_MD5; | |
bd0388ae | 574 | remaining -= TCPOLEN_MD5SIG_ALIGNED; |
cfb6eeb4 | 575 | } |
33ad798c AL |
576 | #else |
577 | *md5 = NULL; | |
cfb6eeb4 | 578 | #endif |
33ad798c AL |
579 | |
580 | /* We always get an MSS option. The option bytes which will be seen in | |
581 | * normal data packets should timestamps be used, must be in the MSS | |
582 | * advertised. But we subtract them from tp->mss_cache so that | |
583 | * calculations in tcp_sendmsg are simpler etc. So account for this | |
584 | * fact here if necessary. If we don't do this correctly, as a | |
585 | * receiver we won't recognize data packets as being full sized when we | |
586 | * should, and thus we won't abide by the delayed ACK rules correctly. | |
587 | * SACKs don't matter, we never delay an ACK when we have any of those | |
588 | * going out. */ | |
589 | opts->mss = tcp_advertise_mss(sk); | |
bd0388ae | 590 | remaining -= TCPOLEN_MSS_ALIGNED; |
33ad798c | 591 | |
bb5b7c11 | 592 | if (likely(sysctl_tcp_timestamps && *md5 == NULL)) { |
33ad798c AL |
593 | opts->options |= OPTION_TS; |
594 | opts->tsval = TCP_SKB_CB(skb)->when; | |
595 | opts->tsecr = tp->rx_opt.ts_recent; | |
bd0388ae | 596 | remaining -= TCPOLEN_TSTAMP_ALIGNED; |
33ad798c | 597 | } |
bb5b7c11 | 598 | if (likely(sysctl_tcp_window_scaling)) { |
33ad798c | 599 | opts->ws = tp->rx_opt.rcv_wscale; |
89e95a61 | 600 | opts->options |= OPTION_WSCALE; |
bd0388ae | 601 | remaining -= TCPOLEN_WSCALE_ALIGNED; |
33ad798c | 602 | } |
bb5b7c11 | 603 | if (likely(sysctl_tcp_sack)) { |
33ad798c | 604 | opts->options |= OPTION_SACK_ADVERTISE; |
b32d1310 | 605 | if (unlikely(!(OPTION_TS & opts->options))) |
bd0388ae | 606 | remaining -= TCPOLEN_SACKPERM_ALIGNED; |
33ad798c AL |
607 | } |
608 | ||
bd0388ae WAS |
609 | /* Note that timestamps are required by the specification. |
610 | * | |
611 | * Odd numbers of bytes are prohibited by the specification, ensuring | |
612 | * that the cookie is 16-bit aligned, and the resulting cookie pair is | |
613 | * 32-bit aligned. | |
614 | */ | |
615 | if (*md5 == NULL && | |
616 | (OPTION_TS & opts->options) && | |
617 | cookie_size > 0) { | |
618 | int need = TCPOLEN_COOKIE_BASE + cookie_size; | |
619 | ||
620 | if (0x2 & need) { | |
621 | /* 32-bit multiple */ | |
622 | need += 2; /* NOPs */ | |
623 | ||
624 | if (need > remaining) { | |
625 | /* try shrinking cookie to fit */ | |
626 | cookie_size -= 2; | |
627 | need -= 4; | |
628 | } | |
629 | } | |
630 | while (need > remaining && TCP_COOKIE_MIN <= cookie_size) { | |
631 | cookie_size -= 4; | |
632 | need -= 4; | |
633 | } | |
634 | if (TCP_COOKIE_MIN <= cookie_size) { | |
635 | opts->options |= OPTION_COOKIE_EXTENSION; | |
636 | opts->hash_location = (__u8 *)&cvp->cookie_pair[0]; | |
637 | opts->hash_size = cookie_size; | |
638 | ||
639 | /* Remember for future incarnations. */ | |
640 | cvp->cookie_desired = cookie_size; | |
641 | ||
642 | if (cvp->cookie_desired != cvp->cookie_pair_size) { | |
643 | /* Currently use random bytes as a nonce, | |
644 | * assuming these are completely unpredictable | |
645 | * by hostile users of the same system. | |
646 | */ | |
647 | get_random_bytes(&cvp->cookie_pair[0], | |
648 | cookie_size); | |
649 | cvp->cookie_pair_size = cookie_size; | |
650 | } | |
651 | ||
652 | remaining -= need; | |
653 | } | |
654 | } | |
655 | return MAX_TCP_OPTION_SPACE - remaining; | |
40efc6fa SH |
656 | } |
657 | ||
67edfef7 | 658 | /* Set up TCP options for SYN-ACKs. */ |
33ad798c AL |
659 | static unsigned tcp_synack_options(struct sock *sk, |
660 | struct request_sock *req, | |
661 | unsigned mss, struct sk_buff *skb, | |
662 | struct tcp_out_options *opts, | |
4957faad WAS |
663 | struct tcp_md5sig_key **md5, |
664 | struct tcp_extend_values *xvp) | |
665 | { | |
33ad798c | 666 | struct inet_request_sock *ireq = inet_rsk(req); |
4957faad WAS |
667 | unsigned remaining = MAX_TCP_OPTION_SPACE; |
668 | u8 cookie_plus = (xvp != NULL && !xvp->cookie_out_never) ? | |
669 | xvp->cookie_plus : | |
670 | 0; | |
671 | bool doing_ts = ireq->tstamp_ok; | |
33ad798c | 672 | |
cfb6eeb4 | 673 | #ifdef CONFIG_TCP_MD5SIG |
33ad798c AL |
674 | *md5 = tcp_rsk(req)->af_specific->md5_lookup(sk, req); |
675 | if (*md5) { | |
676 | opts->options |= OPTION_MD5; | |
4957faad WAS |
677 | remaining -= TCPOLEN_MD5SIG_ALIGNED; |
678 | ||
679 | /* We can't fit any SACK blocks in a packet with MD5 + TS | |
680 | * options. There was discussion about disabling SACK | |
681 | * rather than TS in order to fit in better with old, | |
682 | * buggy kernels, but that was deemed to be unnecessary. | |
683 | */ | |
684 | doing_ts &= !ireq->sack_ok; | |
cfb6eeb4 | 685 | } |
33ad798c AL |
686 | #else |
687 | *md5 = NULL; | |
cfb6eeb4 | 688 | #endif |
33ad798c | 689 | |
4957faad | 690 | /* We always send an MSS option. */ |
33ad798c | 691 | opts->mss = mss; |
4957faad | 692 | remaining -= TCPOLEN_MSS_ALIGNED; |
33ad798c AL |
693 | |
694 | if (likely(ireq->wscale_ok)) { | |
695 | opts->ws = ireq->rcv_wscale; | |
89e95a61 | 696 | opts->options |= OPTION_WSCALE; |
4957faad | 697 | remaining -= TCPOLEN_WSCALE_ALIGNED; |
33ad798c AL |
698 | } |
699 | if (likely(doing_ts)) { | |
700 | opts->options |= OPTION_TS; | |
701 | opts->tsval = TCP_SKB_CB(skb)->when; | |
702 | opts->tsecr = req->ts_recent; | |
4957faad | 703 | remaining -= TCPOLEN_TSTAMP_ALIGNED; |
33ad798c AL |
704 | } |
705 | if (likely(ireq->sack_ok)) { | |
706 | opts->options |= OPTION_SACK_ADVERTISE; | |
707 | if (unlikely(!doing_ts)) | |
4957faad | 708 | remaining -= TCPOLEN_SACKPERM_ALIGNED; |
33ad798c AL |
709 | } |
710 | ||
4957faad WAS |
711 | /* Similar rationale to tcp_syn_options() applies here, too. |
712 | * If the <SYN> options fit, the same options should fit now! | |
713 | */ | |
714 | if (*md5 == NULL && | |
715 | doing_ts && | |
716 | cookie_plus > TCPOLEN_COOKIE_BASE) { | |
717 | int need = cookie_plus; /* has TCPOLEN_COOKIE_BASE */ | |
718 | ||
719 | if (0x2 & need) { | |
720 | /* 32-bit multiple */ | |
721 | need += 2; /* NOPs */ | |
722 | } | |
723 | if (need <= remaining) { | |
724 | opts->options |= OPTION_COOKIE_EXTENSION; | |
725 | opts->hash_size = cookie_plus - TCPOLEN_COOKIE_BASE; | |
726 | remaining -= need; | |
727 | } else { | |
728 | /* There's no error return, so flag it. */ | |
729 | xvp->cookie_out_never = 1; /* true */ | |
730 | opts->hash_size = 0; | |
731 | } | |
732 | } | |
733 | return MAX_TCP_OPTION_SPACE - remaining; | |
33ad798c AL |
734 | } |
735 | ||
67edfef7 AK |
736 | /* Compute TCP options for ESTABLISHED sockets. This is not the |
737 | * final wire format yet. | |
738 | */ | |
33ad798c AL |
739 | static unsigned tcp_established_options(struct sock *sk, struct sk_buff *skb, |
740 | struct tcp_out_options *opts, | |
741 | struct tcp_md5sig_key **md5) { | |
742 | struct tcp_skb_cb *tcb = skb ? TCP_SKB_CB(skb) : NULL; | |
743 | struct tcp_sock *tp = tcp_sk(sk); | |
744 | unsigned size = 0; | |
cabeccbd | 745 | unsigned int eff_sacks; |
33ad798c AL |
746 | |
747 | #ifdef CONFIG_TCP_MD5SIG | |
748 | *md5 = tp->af_specific->md5_lookup(sk, sk); | |
749 | if (unlikely(*md5)) { | |
750 | opts->options |= OPTION_MD5; | |
751 | size += TCPOLEN_MD5SIG_ALIGNED; | |
752 | } | |
753 | #else | |
754 | *md5 = NULL; | |
755 | #endif | |
756 | ||
757 | if (likely(tp->rx_opt.tstamp_ok)) { | |
758 | opts->options |= OPTION_TS; | |
759 | opts->tsval = tcb ? tcb->when : 0; | |
760 | opts->tsecr = tp->rx_opt.ts_recent; | |
761 | size += TCPOLEN_TSTAMP_ALIGNED; | |
762 | } | |
763 | ||
cabeccbd IJ |
764 | eff_sacks = tp->rx_opt.num_sacks + tp->rx_opt.dsack; |
765 | if (unlikely(eff_sacks)) { | |
33ad798c AL |
766 | const unsigned remaining = MAX_TCP_OPTION_SPACE - size; |
767 | opts->num_sack_blocks = | |
cabeccbd | 768 | min_t(unsigned, eff_sacks, |
33ad798c AL |
769 | (remaining - TCPOLEN_SACK_BASE_ALIGNED) / |
770 | TCPOLEN_SACK_PERBLOCK); | |
771 | size += TCPOLEN_SACK_BASE_ALIGNED + | |
772 | opts->num_sack_blocks * TCPOLEN_SACK_PERBLOCK; | |
773 | } | |
774 | ||
775 | return size; | |
40efc6fa | 776 | } |
1da177e4 LT |
777 | |
778 | /* This routine actually transmits TCP packets queued in by | |
779 | * tcp_do_sendmsg(). This is used by both the initial | |
780 | * transmission and possible later retransmissions. | |
781 | * All SKB's seen here are completely headerless. It is our | |
782 | * job to build the TCP header, and pass the packet down to | |
783 | * IP so it can do the same plus pass the packet off to the | |
784 | * device. | |
785 | * | |
786 | * We are working here with either a clone of the original | |
787 | * SKB, or a fresh unique copy made by the retransmit engine. | |
788 | */ | |
056834d9 IJ |
789 | static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it, |
790 | gfp_t gfp_mask) | |
1da177e4 | 791 | { |
dfb4b9dc DM |
792 | const struct inet_connection_sock *icsk = inet_csk(sk); |
793 | struct inet_sock *inet; | |
794 | struct tcp_sock *tp; | |
795 | struct tcp_skb_cb *tcb; | |
33ad798c AL |
796 | struct tcp_out_options opts; |
797 | unsigned tcp_options_size, tcp_header_size; | |
cfb6eeb4 | 798 | struct tcp_md5sig_key *md5; |
dfb4b9dc | 799 | struct tcphdr *th; |
dfb4b9dc DM |
800 | int err; |
801 | ||
802 | BUG_ON(!skb || !tcp_skb_pcount(skb)); | |
803 | ||
804 | /* If congestion control is doing timestamping, we must | |
805 | * take such a timestamp before we potentially clone/copy. | |
806 | */ | |
164891aa | 807 | if (icsk->icsk_ca_ops->flags & TCP_CONG_RTT_STAMP) |
dfb4b9dc DM |
808 | __net_timestamp(skb); |
809 | ||
810 | if (likely(clone_it)) { | |
811 | if (unlikely(skb_cloned(skb))) | |
812 | skb = pskb_copy(skb, gfp_mask); | |
813 | else | |
814 | skb = skb_clone(skb, gfp_mask); | |
815 | if (unlikely(!skb)) | |
816 | return -ENOBUFS; | |
817 | } | |
1da177e4 | 818 | |
dfb4b9dc DM |
819 | inet = inet_sk(sk); |
820 | tp = tcp_sk(sk); | |
821 | tcb = TCP_SKB_CB(skb); | |
33ad798c | 822 | memset(&opts, 0, sizeof(opts)); |
1da177e4 | 823 | |
33ad798c AL |
824 | if (unlikely(tcb->flags & TCPCB_FLAG_SYN)) |
825 | tcp_options_size = tcp_syn_options(sk, skb, &opts, &md5); | |
826 | else | |
827 | tcp_options_size = tcp_established_options(sk, skb, &opts, | |
828 | &md5); | |
829 | tcp_header_size = tcp_options_size + sizeof(struct tcphdr); | |
e905a9ed | 830 | |
dfb4b9dc DM |
831 | if (tcp_packets_in_flight(tp) == 0) |
832 | tcp_ca_event(sk, CA_EVENT_TX_START); | |
833 | ||
aa8223c7 ACM |
834 | skb_push(skb, tcp_header_size); |
835 | skb_reset_transport_header(skb); | |
e89862f4 | 836 | skb_set_owner_w(skb, sk); |
dfb4b9dc DM |
837 | |
838 | /* Build TCP header and checksum it. */ | |
aa8223c7 | 839 | th = tcp_hdr(skb); |
c720c7e8 ED |
840 | th->source = inet->inet_sport; |
841 | th->dest = inet->inet_dport; | |
dfb4b9dc DM |
842 | th->seq = htonl(tcb->seq); |
843 | th->ack_seq = htonl(tp->rcv_nxt); | |
df7a3b07 | 844 | *(((__be16 *)th) + 6) = htons(((tcp_header_size >> 2) << 12) | |
dfb4b9dc DM |
845 | tcb->flags); |
846 | ||
847 | if (unlikely(tcb->flags & TCPCB_FLAG_SYN)) { | |
848 | /* RFC1323: The window in SYN & SYN/ACK segments | |
849 | * is never scaled. | |
850 | */ | |
600ff0c2 | 851 | th->window = htons(min(tp->rcv_wnd, 65535U)); |
dfb4b9dc DM |
852 | } else { |
853 | th->window = htons(tcp_select_window(sk)); | |
854 | } | |
855 | th->check = 0; | |
856 | th->urg_ptr = 0; | |
1da177e4 | 857 | |
33f5f57e | 858 | /* The urg_mode check is necessary during a below snd_una win probe */ |
7691367d HX |
859 | if (unlikely(tcp_urg_mode(tp) && before(tcb->seq, tp->snd_up))) { |
860 | if (before(tp->snd_up, tcb->seq + 0x10000)) { | |
861 | th->urg_ptr = htons(tp->snd_up - tcb->seq); | |
862 | th->urg = 1; | |
863 | } else if (after(tcb->seq + 0xFFFF, tp->snd_nxt)) { | |
0eae88f3 | 864 | th->urg_ptr = htons(0xFFFF); |
7691367d HX |
865 | th->urg = 1; |
866 | } | |
dfb4b9dc | 867 | } |
1da177e4 | 868 | |
bd0388ae | 869 | tcp_options_write((__be32 *)(th + 1), tp, &opts); |
33ad798c | 870 | if (likely((tcb->flags & TCPCB_FLAG_SYN) == 0)) |
9e412ba7 | 871 | TCP_ECN_send(sk, skb, tcp_header_size); |
1da177e4 | 872 | |
cfb6eeb4 YH |
873 | #ifdef CONFIG_TCP_MD5SIG |
874 | /* Calculate the MD5 hash, as we have all we need now */ | |
875 | if (md5) { | |
33ad798c | 876 | sk->sk_route_caps &= ~NETIF_F_GSO_MASK; |
bd0388ae | 877 | tp->af_specific->calc_md5_hash(opts.hash_location, |
49a72dfb | 878 | md5, sk, NULL, skb); |
cfb6eeb4 YH |
879 | } |
880 | #endif | |
881 | ||
bb296246 | 882 | icsk->icsk_af_ops->send_check(sk, skb); |
1da177e4 | 883 | |
dfb4b9dc DM |
884 | if (likely(tcb->flags & TCPCB_FLAG_ACK)) |
885 | tcp_event_ack_sent(sk, tcp_skb_pcount(skb)); | |
1da177e4 | 886 | |
dfb4b9dc DM |
887 | if (skb->len != tcp_header_size) |
888 | tcp_event_data_sent(tp, skb, sk); | |
1da177e4 | 889 | |
bd37a088 | 890 | if (after(tcb->end_seq, tp->snd_nxt) || tcb->seq == tcb->end_seq) |
81cc8a75 | 891 | TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTSEGS); |
1da177e4 | 892 | |
4e15ed4d | 893 | err = icsk->icsk_af_ops->queue_xmit(skb); |
83de47cd | 894 | if (likely(err <= 0)) |
dfb4b9dc DM |
895 | return err; |
896 | ||
3cfe3baa | 897 | tcp_enter_cwr(sk, 1); |
dfb4b9dc | 898 | |
b9df3cb8 | 899 | return net_xmit_eval(err); |
1da177e4 LT |
900 | } |
901 | ||
67edfef7 | 902 | /* This routine just queues the buffer for sending. |
1da177e4 LT |
903 | * |
904 | * NOTE: probe0 timer is not checked, do not forget tcp_push_pending_frames, | |
905 | * otherwise socket can stall. | |
906 | */ | |
907 | static void tcp_queue_skb(struct sock *sk, struct sk_buff *skb) | |
908 | { | |
909 | struct tcp_sock *tp = tcp_sk(sk); | |
910 | ||
911 | /* Advance write_seq and place onto the write_queue. */ | |
912 | tp->write_seq = TCP_SKB_CB(skb)->end_seq; | |
913 | skb_header_release(skb); | |
fe067e8a | 914 | tcp_add_write_queue_tail(sk, skb); |
3ab224be HA |
915 | sk->sk_wmem_queued += skb->truesize; |
916 | sk_mem_charge(sk, skb->truesize); | |
1da177e4 LT |
917 | } |
918 | ||
67edfef7 | 919 | /* Initialize TSO segments for a packet. */ |
056834d9 IJ |
920 | static void tcp_set_skb_tso_segs(struct sock *sk, struct sk_buff *skb, |
921 | unsigned int mss_now) | |
f6302d1d | 922 | { |
8e5b9dda HX |
923 | if (skb->len <= mss_now || !sk_can_gso(sk) || |
924 | skb->ip_summed == CHECKSUM_NONE) { | |
f6302d1d DM |
925 | /* Avoid the costly divide in the normal |
926 | * non-TSO case. | |
927 | */ | |
7967168c HX |
928 | skb_shinfo(skb)->gso_segs = 1; |
929 | skb_shinfo(skb)->gso_size = 0; | |
930 | skb_shinfo(skb)->gso_type = 0; | |
f6302d1d | 931 | } else { |
356f89e1 | 932 | skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss_now); |
7967168c | 933 | skb_shinfo(skb)->gso_size = mss_now; |
bcd76111 | 934 | skb_shinfo(skb)->gso_type = sk->sk_gso_type; |
1da177e4 LT |
935 | } |
936 | } | |
937 | ||
91fed7a1 | 938 | /* When a modification to fackets out becomes necessary, we need to check |
68f8353b | 939 | * skb is counted to fackets_out or not. |
91fed7a1 | 940 | */ |
a47e5a98 | 941 | static void tcp_adjust_fackets_out(struct sock *sk, struct sk_buff *skb, |
91fed7a1 IJ |
942 | int decr) |
943 | { | |
a47e5a98 IJ |
944 | struct tcp_sock *tp = tcp_sk(sk); |
945 | ||
dc86967b | 946 | if (!tp->sacked_out || tcp_is_reno(tp)) |
91fed7a1 IJ |
947 | return; |
948 | ||
6859d494 | 949 | if (after(tcp_highest_sack_seq(tp), TCP_SKB_CB(skb)->seq)) |
91fed7a1 | 950 | tp->fackets_out -= decr; |
91fed7a1 IJ |
951 | } |
952 | ||
797108d1 IJ |
953 | /* Pcount in the middle of the write queue got changed, we need to do various |
954 | * tweaks to fix counters | |
955 | */ | |
956 | static void tcp_adjust_pcount(struct sock *sk, struct sk_buff *skb, int decr) | |
957 | { | |
958 | struct tcp_sock *tp = tcp_sk(sk); | |
959 | ||
960 | tp->packets_out -= decr; | |
961 | ||
962 | if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) | |
963 | tp->sacked_out -= decr; | |
964 | if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) | |
965 | tp->retrans_out -= decr; | |
966 | if (TCP_SKB_CB(skb)->sacked & TCPCB_LOST) | |
967 | tp->lost_out -= decr; | |
968 | ||
969 | /* Reno case is special. Sigh... */ | |
970 | if (tcp_is_reno(tp) && decr > 0) | |
971 | tp->sacked_out -= min_t(u32, tp->sacked_out, decr); | |
972 | ||
973 | tcp_adjust_fackets_out(sk, skb, decr); | |
974 | ||
975 | if (tp->lost_skb_hint && | |
976 | before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(tp->lost_skb_hint)->seq) && | |
52cf3cc8 | 977 | (tcp_is_fack(tp) || (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED))) |
797108d1 IJ |
978 | tp->lost_cnt_hint -= decr; |
979 | ||
980 | tcp_verify_left_out(tp); | |
981 | } | |
982 | ||
1da177e4 LT |
983 | /* Function to create two new TCP segments. Shrinks the given segment |
984 | * to the specified size and appends a new segment with the rest of the | |
e905a9ed | 985 | * packet to the list. This won't be called frequently, I hope. |
1da177e4 LT |
986 | * Remember, these are still headerless SKBs at this point. |
987 | */ | |
056834d9 IJ |
988 | int tcp_fragment(struct sock *sk, struct sk_buff *skb, u32 len, |
989 | unsigned int mss_now) | |
1da177e4 LT |
990 | { |
991 | struct tcp_sock *tp = tcp_sk(sk); | |
992 | struct sk_buff *buff; | |
6475be16 | 993 | int nsize, old_factor; |
b60b49ea | 994 | int nlen; |
9ce01461 | 995 | u8 flags; |
1da177e4 | 996 | |
b2cc99f0 | 997 | BUG_ON(len > skb->len); |
6a438bbe | 998 | |
1da177e4 LT |
999 | nsize = skb_headlen(skb) - len; |
1000 | if (nsize < 0) | |
1001 | nsize = 0; | |
1002 | ||
1003 | if (skb_cloned(skb) && | |
1004 | skb_is_nonlinear(skb) && | |
1005 | pskb_expand_head(skb, 0, 0, GFP_ATOMIC)) | |
1006 | return -ENOMEM; | |
1007 | ||
1008 | /* Get a new skb... force flag on. */ | |
1009 | buff = sk_stream_alloc_skb(sk, nsize, GFP_ATOMIC); | |
1010 | if (buff == NULL) | |
1011 | return -ENOMEM; /* We'll just try again later. */ | |
ef5cb973 | 1012 | |
3ab224be HA |
1013 | sk->sk_wmem_queued += buff->truesize; |
1014 | sk_mem_charge(sk, buff->truesize); | |
b60b49ea HX |
1015 | nlen = skb->len - len - nsize; |
1016 | buff->truesize += nlen; | |
1017 | skb->truesize -= nlen; | |
1da177e4 LT |
1018 | |
1019 | /* Correct the sequence numbers. */ | |
1020 | TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len; | |
1021 | TCP_SKB_CB(buff)->end_seq = TCP_SKB_CB(skb)->end_seq; | |
1022 | TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(buff)->seq; | |
1023 | ||
1024 | /* PSH and FIN should only be set in the second packet. */ | |
1025 | flags = TCP_SKB_CB(skb)->flags; | |
056834d9 | 1026 | TCP_SKB_CB(skb)->flags = flags & ~(TCPCB_FLAG_FIN | TCPCB_FLAG_PSH); |
1da177e4 | 1027 | TCP_SKB_CB(buff)->flags = flags; |
e14c3caf | 1028 | TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked; |
1da177e4 | 1029 | |
84fa7933 | 1030 | if (!skb_shinfo(skb)->nr_frags && skb->ip_summed != CHECKSUM_PARTIAL) { |
1da177e4 | 1031 | /* Copy and checksum data tail into the new buffer. */ |
056834d9 IJ |
1032 | buff->csum = csum_partial_copy_nocheck(skb->data + len, |
1033 | skb_put(buff, nsize), | |
1da177e4 LT |
1034 | nsize, 0); |
1035 | ||
1036 | skb_trim(skb, len); | |
1037 | ||
1038 | skb->csum = csum_block_sub(skb->csum, buff->csum, len); | |
1039 | } else { | |
84fa7933 | 1040 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 LT |
1041 | skb_split(skb, buff, len); |
1042 | } | |
1043 | ||
1044 | buff->ip_summed = skb->ip_summed; | |
1045 | ||
1046 | /* Looks stupid, but our code really uses when of | |
1047 | * skbs, which it never sent before. --ANK | |
1048 | */ | |
1049 | TCP_SKB_CB(buff)->when = TCP_SKB_CB(skb)->when; | |
a61bbcf2 | 1050 | buff->tstamp = skb->tstamp; |
1da177e4 | 1051 | |
6475be16 DM |
1052 | old_factor = tcp_skb_pcount(skb); |
1053 | ||
1da177e4 | 1054 | /* Fix up tso_factor for both original and new SKB. */ |
846998ae DM |
1055 | tcp_set_skb_tso_segs(sk, skb, mss_now); |
1056 | tcp_set_skb_tso_segs(sk, buff, mss_now); | |
1da177e4 | 1057 | |
6475be16 DM |
1058 | /* If this packet has been sent out already, we must |
1059 | * adjust the various packet counters. | |
1060 | */ | |
cf0b450c | 1061 | if (!before(tp->snd_nxt, TCP_SKB_CB(buff)->end_seq)) { |
6475be16 DM |
1062 | int diff = old_factor - tcp_skb_pcount(skb) - |
1063 | tcp_skb_pcount(buff); | |
1da177e4 | 1064 | |
797108d1 IJ |
1065 | if (diff) |
1066 | tcp_adjust_pcount(sk, skb, diff); | |
1da177e4 LT |
1067 | } |
1068 | ||
1069 | /* Link BUFF into the send queue. */ | |
f44b5271 | 1070 | skb_header_release(buff); |
fe067e8a | 1071 | tcp_insert_write_queue_after(skb, buff, sk); |
1da177e4 LT |
1072 | |
1073 | return 0; | |
1074 | } | |
1075 | ||
1076 | /* This is similar to __pskb_pull_head() (it will go to core/skbuff.c | |
1077 | * eventually). The difference is that pulled data not copied, but | |
1078 | * immediately discarded. | |
1079 | */ | |
f2911969 | 1080 | static void __pskb_trim_head(struct sk_buff *skb, int len) |
1da177e4 LT |
1081 | { |
1082 | int i, k, eat; | |
1083 | ||
1084 | eat = len; | |
1085 | k = 0; | |
056834d9 | 1086 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
1da177e4 LT |
1087 | if (skb_shinfo(skb)->frags[i].size <= eat) { |
1088 | put_page(skb_shinfo(skb)->frags[i].page); | |
1089 | eat -= skb_shinfo(skb)->frags[i].size; | |
1090 | } else { | |
1091 | skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; | |
1092 | if (eat) { | |
1093 | skb_shinfo(skb)->frags[k].page_offset += eat; | |
1094 | skb_shinfo(skb)->frags[k].size -= eat; | |
1095 | eat = 0; | |
1096 | } | |
1097 | k++; | |
1098 | } | |
1099 | } | |
1100 | skb_shinfo(skb)->nr_frags = k; | |
1101 | ||
27a884dc | 1102 | skb_reset_tail_pointer(skb); |
1da177e4 LT |
1103 | skb->data_len -= len; |
1104 | skb->len = skb->data_len; | |
1da177e4 LT |
1105 | } |
1106 | ||
67edfef7 | 1107 | /* Remove acked data from a packet in the transmit queue. */ |
1da177e4 LT |
1108 | int tcp_trim_head(struct sock *sk, struct sk_buff *skb, u32 len) |
1109 | { | |
056834d9 | 1110 | if (skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC)) |
1da177e4 LT |
1111 | return -ENOMEM; |
1112 | ||
f2911969 HXP |
1113 | /* If len == headlen, we avoid __skb_pull to preserve alignment. */ |
1114 | if (unlikely(len < skb_headlen(skb))) | |
1da177e4 | 1115 | __skb_pull(skb, len); |
f2911969 HXP |
1116 | else |
1117 | __pskb_trim_head(skb, len - skb_headlen(skb)); | |
1da177e4 LT |
1118 | |
1119 | TCP_SKB_CB(skb)->seq += len; | |
84fa7933 | 1120 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 LT |
1121 | |
1122 | skb->truesize -= len; | |
1123 | sk->sk_wmem_queued -= len; | |
3ab224be | 1124 | sk_mem_uncharge(sk, len); |
1da177e4 LT |
1125 | sock_set_flag(sk, SOCK_QUEUE_SHRUNK); |
1126 | ||
1127 | /* Any change of skb->len requires recalculation of tso | |
1128 | * factor and mss. | |
1129 | */ | |
1130 | if (tcp_skb_pcount(skb) > 1) | |
0c54b85f | 1131 | tcp_set_skb_tso_segs(sk, skb, tcp_current_mss(sk)); |
1da177e4 LT |
1132 | |
1133 | return 0; | |
1134 | } | |
1135 | ||
67edfef7 | 1136 | /* Calculate MSS. Not accounting for SACKs here. */ |
5d424d5a JH |
1137 | int tcp_mtu_to_mss(struct sock *sk, int pmtu) |
1138 | { | |
1139 | struct tcp_sock *tp = tcp_sk(sk); | |
1140 | struct inet_connection_sock *icsk = inet_csk(sk); | |
1141 | int mss_now; | |
1142 | ||
1143 | /* Calculate base mss without TCP options: | |
1144 | It is MMS_S - sizeof(tcphdr) of rfc1122 | |
1145 | */ | |
1146 | mss_now = pmtu - icsk->icsk_af_ops->net_header_len - sizeof(struct tcphdr); | |
1147 | ||
1148 | /* Clamp it (mss_clamp does not include tcp options) */ | |
1149 | if (mss_now > tp->rx_opt.mss_clamp) | |
1150 | mss_now = tp->rx_opt.mss_clamp; | |
1151 | ||
1152 | /* Now subtract optional transport overhead */ | |
1153 | mss_now -= icsk->icsk_ext_hdr_len; | |
1154 | ||
1155 | /* Then reserve room for full set of TCP options and 8 bytes of data */ | |
1156 | if (mss_now < 48) | |
1157 | mss_now = 48; | |
1158 | ||
1159 | /* Now subtract TCP options size, not including SACKs */ | |
1160 | mss_now -= tp->tcp_header_len - sizeof(struct tcphdr); | |
1161 | ||
1162 | return mss_now; | |
1163 | } | |
1164 | ||
1165 | /* Inverse of above */ | |
1166 | int tcp_mss_to_mtu(struct sock *sk, int mss) | |
1167 | { | |
1168 | struct tcp_sock *tp = tcp_sk(sk); | |
1169 | struct inet_connection_sock *icsk = inet_csk(sk); | |
1170 | int mtu; | |
1171 | ||
1172 | mtu = mss + | |
1173 | tp->tcp_header_len + | |
1174 | icsk->icsk_ext_hdr_len + | |
1175 | icsk->icsk_af_ops->net_header_len; | |
1176 | ||
1177 | return mtu; | |
1178 | } | |
1179 | ||
67edfef7 | 1180 | /* MTU probing init per socket */ |
5d424d5a JH |
1181 | void tcp_mtup_init(struct sock *sk) |
1182 | { | |
1183 | struct tcp_sock *tp = tcp_sk(sk); | |
1184 | struct inet_connection_sock *icsk = inet_csk(sk); | |
1185 | ||
1186 | icsk->icsk_mtup.enabled = sysctl_tcp_mtu_probing > 1; | |
1187 | icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + sizeof(struct tcphdr) + | |
e905a9ed | 1188 | icsk->icsk_af_ops->net_header_len; |
5d424d5a JH |
1189 | icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, sysctl_tcp_base_mss); |
1190 | icsk->icsk_mtup.probe_size = 0; | |
1191 | } | |
1192 | ||
1da177e4 LT |
1193 | /* This function synchronize snd mss to current pmtu/exthdr set. |
1194 | ||
1195 | tp->rx_opt.user_mss is mss set by user by TCP_MAXSEG. It does NOT counts | |
1196 | for TCP options, but includes only bare TCP header. | |
1197 | ||
1198 | tp->rx_opt.mss_clamp is mss negotiated at connection setup. | |
caa20d9a | 1199 | It is minimum of user_mss and mss received with SYN. |
1da177e4 LT |
1200 | It also does not include TCP options. |
1201 | ||
d83d8461 | 1202 | inet_csk(sk)->icsk_pmtu_cookie is last pmtu, seen by this function. |
1da177e4 LT |
1203 | |
1204 | tp->mss_cache is current effective sending mss, including | |
1205 | all tcp options except for SACKs. It is evaluated, | |
1206 | taking into account current pmtu, but never exceeds | |
1207 | tp->rx_opt.mss_clamp. | |
1208 | ||
1209 | NOTE1. rfc1122 clearly states that advertised MSS | |
1210 | DOES NOT include either tcp or ip options. | |
1211 | ||
d83d8461 ACM |
1212 | NOTE2. inet_csk(sk)->icsk_pmtu_cookie and tp->mss_cache |
1213 | are READ ONLY outside this function. --ANK (980731) | |
1da177e4 | 1214 | */ |
1da177e4 LT |
1215 | unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu) |
1216 | { | |
1217 | struct tcp_sock *tp = tcp_sk(sk); | |
d83d8461 | 1218 | struct inet_connection_sock *icsk = inet_csk(sk); |
5d424d5a | 1219 | int mss_now; |
1da177e4 | 1220 | |
5d424d5a JH |
1221 | if (icsk->icsk_mtup.search_high > pmtu) |
1222 | icsk->icsk_mtup.search_high = pmtu; | |
1da177e4 | 1223 | |
5d424d5a | 1224 | mss_now = tcp_mtu_to_mss(sk, pmtu); |
409d22b4 | 1225 | mss_now = tcp_bound_to_half_wnd(tp, mss_now); |
1da177e4 LT |
1226 | |
1227 | /* And store cached results */ | |
d83d8461 | 1228 | icsk->icsk_pmtu_cookie = pmtu; |
5d424d5a JH |
1229 | if (icsk->icsk_mtup.enabled) |
1230 | mss_now = min(mss_now, tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low)); | |
c1b4a7e6 | 1231 | tp->mss_cache = mss_now; |
1da177e4 LT |
1232 | |
1233 | return mss_now; | |
1234 | } | |
1235 | ||
1236 | /* Compute the current effective MSS, taking SACKs and IP options, | |
1237 | * and even PMTU discovery events into account. | |
1da177e4 | 1238 | */ |
0c54b85f | 1239 | unsigned int tcp_current_mss(struct sock *sk) |
1da177e4 LT |
1240 | { |
1241 | struct tcp_sock *tp = tcp_sk(sk); | |
1242 | struct dst_entry *dst = __sk_dst_get(sk); | |
c1b4a7e6 | 1243 | u32 mss_now; |
33ad798c AL |
1244 | unsigned header_len; |
1245 | struct tcp_out_options opts; | |
1246 | struct tcp_md5sig_key *md5; | |
c1b4a7e6 DM |
1247 | |
1248 | mss_now = tp->mss_cache; | |
1249 | ||
1da177e4 LT |
1250 | if (dst) { |
1251 | u32 mtu = dst_mtu(dst); | |
d83d8461 | 1252 | if (mtu != inet_csk(sk)->icsk_pmtu_cookie) |
1da177e4 LT |
1253 | mss_now = tcp_sync_mss(sk, mtu); |
1254 | } | |
1255 | ||
33ad798c AL |
1256 | header_len = tcp_established_options(sk, NULL, &opts, &md5) + |
1257 | sizeof(struct tcphdr); | |
1258 | /* The mss_cache is sized based on tp->tcp_header_len, which assumes | |
1259 | * some common options. If this is an odd packet (because we have SACK | |
1260 | * blocks etc) then our calculated header_len will be different, and | |
1261 | * we have to adjust mss_now correspondingly */ | |
1262 | if (header_len != tp->tcp_header_len) { | |
1263 | int delta = (int) header_len - tp->tcp_header_len; | |
1264 | mss_now -= delta; | |
1265 | } | |
cfb6eeb4 | 1266 | |
1da177e4 LT |
1267 | return mss_now; |
1268 | } | |
1269 | ||
a762a980 | 1270 | /* Congestion window validation. (RFC2861) */ |
9e412ba7 | 1271 | static void tcp_cwnd_validate(struct sock *sk) |
a762a980 | 1272 | { |
9e412ba7 | 1273 | struct tcp_sock *tp = tcp_sk(sk); |
a762a980 | 1274 | |
d436d686 | 1275 | if (tp->packets_out >= tp->snd_cwnd) { |
a762a980 DM |
1276 | /* Network is feed fully. */ |
1277 | tp->snd_cwnd_used = 0; | |
1278 | tp->snd_cwnd_stamp = tcp_time_stamp; | |
1279 | } else { | |
1280 | /* Network starves. */ | |
1281 | if (tp->packets_out > tp->snd_cwnd_used) | |
1282 | tp->snd_cwnd_used = tp->packets_out; | |
1283 | ||
15d33c07 DM |
1284 | if (sysctl_tcp_slow_start_after_idle && |
1285 | (s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= inet_csk(sk)->icsk_rto) | |
a762a980 DM |
1286 | tcp_cwnd_application_limited(sk); |
1287 | } | |
1288 | } | |
1289 | ||
0e3a4803 IJ |
1290 | /* Returns the portion of skb which can be sent right away without |
1291 | * introducing MSS oddities to segment boundaries. In rare cases where | |
1292 | * mss_now != mss_cache, we will request caller to create a small skb | |
1293 | * per input skb which could be mostly avoided here (if desired). | |
5ea3a748 IJ |
1294 | * |
1295 | * We explicitly want to create a request for splitting write queue tail | |
1296 | * to a small skb for Nagle purposes while avoiding unnecessary modulos, | |
1297 | * thus all the complexity (cwnd_len is always MSS multiple which we | |
1298 | * return whenever allowed by the other factors). Basically we need the | |
1299 | * modulo only when the receiver window alone is the limiting factor or | |
1300 | * when we would be allowed to send the split-due-to-Nagle skb fully. | |
0e3a4803 IJ |
1301 | */ |
1302 | static unsigned int tcp_mss_split_point(struct sock *sk, struct sk_buff *skb, | |
056834d9 | 1303 | unsigned int mss_now, unsigned int cwnd) |
c1b4a7e6 | 1304 | { |
0e3a4803 IJ |
1305 | struct tcp_sock *tp = tcp_sk(sk); |
1306 | u32 needed, window, cwnd_len; | |
c1b4a7e6 | 1307 | |
90840def | 1308 | window = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; |
c1b4a7e6 | 1309 | cwnd_len = mss_now * cwnd; |
0e3a4803 IJ |
1310 | |
1311 | if (likely(cwnd_len <= window && skb != tcp_write_queue_tail(sk))) | |
1312 | return cwnd_len; | |
1313 | ||
5ea3a748 IJ |
1314 | needed = min(skb->len, window); |
1315 | ||
17515408 | 1316 | if (cwnd_len <= needed) |
0e3a4803 IJ |
1317 | return cwnd_len; |
1318 | ||
0e3a4803 | 1319 | return needed - needed % mss_now; |
c1b4a7e6 DM |
1320 | } |
1321 | ||
1322 | /* Can at least one segment of SKB be sent right now, according to the | |
1323 | * congestion window rules? If so, return how many segments are allowed. | |
1324 | */ | |
056834d9 IJ |
1325 | static inline unsigned int tcp_cwnd_test(struct tcp_sock *tp, |
1326 | struct sk_buff *skb) | |
c1b4a7e6 DM |
1327 | { |
1328 | u32 in_flight, cwnd; | |
1329 | ||
1330 | /* Don't be strict about the congestion window for the final FIN. */ | |
104439a8 JH |
1331 | if ((TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN) && |
1332 | tcp_skb_pcount(skb) == 1) | |
c1b4a7e6 DM |
1333 | return 1; |
1334 | ||
1335 | in_flight = tcp_packets_in_flight(tp); | |
1336 | cwnd = tp->snd_cwnd; | |
1337 | if (in_flight < cwnd) | |
1338 | return (cwnd - in_flight); | |
1339 | ||
1340 | return 0; | |
1341 | } | |
1342 | ||
67edfef7 AK |
1343 | /* Intialize TSO state of a skb. |
1344 | * This must be invoked the first time we consider transmitting | |
c1b4a7e6 DM |
1345 | * SKB onto the wire. |
1346 | */ | |
056834d9 IJ |
1347 | static int tcp_init_tso_segs(struct sock *sk, struct sk_buff *skb, |
1348 | unsigned int mss_now) | |
c1b4a7e6 DM |
1349 | { |
1350 | int tso_segs = tcp_skb_pcount(skb); | |
1351 | ||
f8269a49 | 1352 | if (!tso_segs || (tso_segs > 1 && tcp_skb_mss(skb) != mss_now)) { |
846998ae | 1353 | tcp_set_skb_tso_segs(sk, skb, mss_now); |
c1b4a7e6 DM |
1354 | tso_segs = tcp_skb_pcount(skb); |
1355 | } | |
1356 | return tso_segs; | |
1357 | } | |
1358 | ||
67edfef7 | 1359 | /* Minshall's variant of the Nagle send check. */ |
c1b4a7e6 DM |
1360 | static inline int tcp_minshall_check(const struct tcp_sock *tp) |
1361 | { | |
09cb105e | 1362 | return after(tp->snd_sml, tp->snd_una) && |
c1b4a7e6 DM |
1363 | !after(tp->snd_sml, tp->snd_nxt); |
1364 | } | |
1365 | ||
1366 | /* Return 0, if packet can be sent now without violation Nagle's rules: | |
1367 | * 1. It is full sized. | |
1368 | * 2. Or it contains FIN. (already checked by caller) | |
1369 | * 3. Or TCP_NODELAY was set. | |
1370 | * 4. Or TCP_CORK is not set, and all sent packets are ACKed. | |
1371 | * With Minshall's modification: all sent small packets are ACKed. | |
1372 | */ | |
c1b4a7e6 | 1373 | static inline int tcp_nagle_check(const struct tcp_sock *tp, |
e905a9ed | 1374 | const struct sk_buff *skb, |
c1b4a7e6 DM |
1375 | unsigned mss_now, int nonagle) |
1376 | { | |
1377 | return (skb->len < mss_now && | |
056834d9 IJ |
1378 | ((nonagle & TCP_NAGLE_CORK) || |
1379 | (!nonagle && tp->packets_out && tcp_minshall_check(tp)))); | |
c1b4a7e6 DM |
1380 | } |
1381 | ||
1382 | /* Return non-zero if the Nagle test allows this packet to be | |
1383 | * sent now. | |
1384 | */ | |
1385 | static inline int tcp_nagle_test(struct tcp_sock *tp, struct sk_buff *skb, | |
1386 | unsigned int cur_mss, int nonagle) | |
1387 | { | |
1388 | /* Nagle rule does not apply to frames, which sit in the middle of the | |
1389 | * write_queue (they have no chances to get new data). | |
1390 | * | |
1391 | * This is implemented in the callers, where they modify the 'nonagle' | |
1392 | * argument based upon the location of SKB in the send queue. | |
1393 | */ | |
1394 | if (nonagle & TCP_NAGLE_PUSH) | |
1395 | return 1; | |
1396 | ||
d551e454 IJ |
1397 | /* Don't use the nagle rule for urgent data (or for the final FIN). |
1398 | * Nagle can be ignored during F-RTO too (see RFC4138). | |
1399 | */ | |
33f5f57e | 1400 | if (tcp_urg_mode(tp) || (tp->frto_counter == 2) || |
c1b4a7e6 DM |
1401 | (TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN)) |
1402 | return 1; | |
1403 | ||
1404 | if (!tcp_nagle_check(tp, skb, cur_mss, nonagle)) | |
1405 | return 1; | |
1406 | ||
1407 | return 0; | |
1408 | } | |
1409 | ||
1410 | /* Does at least the first segment of SKB fit into the send window? */ | |
056834d9 IJ |
1411 | static inline int tcp_snd_wnd_test(struct tcp_sock *tp, struct sk_buff *skb, |
1412 | unsigned int cur_mss) | |
c1b4a7e6 DM |
1413 | { |
1414 | u32 end_seq = TCP_SKB_CB(skb)->end_seq; | |
1415 | ||
1416 | if (skb->len > cur_mss) | |
1417 | end_seq = TCP_SKB_CB(skb)->seq + cur_mss; | |
1418 | ||
90840def | 1419 | return !after(end_seq, tcp_wnd_end(tp)); |
c1b4a7e6 DM |
1420 | } |
1421 | ||
fe067e8a | 1422 | /* This checks if the data bearing packet SKB (usually tcp_send_head(sk)) |
c1b4a7e6 DM |
1423 | * should be put on the wire right now. If so, it returns the number of |
1424 | * packets allowed by the congestion window. | |
1425 | */ | |
1426 | static unsigned int tcp_snd_test(struct sock *sk, struct sk_buff *skb, | |
1427 | unsigned int cur_mss, int nonagle) | |
1428 | { | |
1429 | struct tcp_sock *tp = tcp_sk(sk); | |
1430 | unsigned int cwnd_quota; | |
1431 | ||
846998ae | 1432 | tcp_init_tso_segs(sk, skb, cur_mss); |
c1b4a7e6 DM |
1433 | |
1434 | if (!tcp_nagle_test(tp, skb, cur_mss, nonagle)) | |
1435 | return 0; | |
1436 | ||
1437 | cwnd_quota = tcp_cwnd_test(tp, skb); | |
056834d9 | 1438 | if (cwnd_quota && !tcp_snd_wnd_test(tp, skb, cur_mss)) |
c1b4a7e6 DM |
1439 | cwnd_quota = 0; |
1440 | ||
1441 | return cwnd_quota; | |
1442 | } | |
1443 | ||
67edfef7 | 1444 | /* Test if sending is allowed right now. */ |
9e412ba7 | 1445 | int tcp_may_send_now(struct sock *sk) |
c1b4a7e6 | 1446 | { |
9e412ba7 | 1447 | struct tcp_sock *tp = tcp_sk(sk); |
fe067e8a | 1448 | struct sk_buff *skb = tcp_send_head(sk); |
c1b4a7e6 DM |
1449 | |
1450 | return (skb && | |
0c54b85f | 1451 | tcp_snd_test(sk, skb, tcp_current_mss(sk), |
c1b4a7e6 | 1452 | (tcp_skb_is_last(sk, skb) ? |