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 | { | |
353 | skb->csum = 0; | |
354 | ||
355 | TCP_SKB_CB(skb)->flags = flags; | |
356 | TCP_SKB_CB(skb)->sacked = 0; | |
357 | ||
358 | skb_shinfo(skb)->gso_segs = 1; | |
359 | skb_shinfo(skb)->gso_size = 0; | |
360 | skb_shinfo(skb)->gso_type = 0; | |
361 | ||
362 | TCP_SKB_CB(skb)->seq = seq; | |
363 | if (flags & (TCPCB_FLAG_SYN | TCPCB_FLAG_FIN)) | |
364 | seq++; | |
365 | TCP_SKB_CB(skb)->end_seq = seq; | |
366 | } | |
367 | ||
33f5f57e IJ |
368 | static inline int tcp_urg_mode(const struct tcp_sock *tp) |
369 | { | |
370 | return tp->snd_una != tp->snd_up; | |
371 | } | |
372 | ||
33ad798c AL |
373 | #define OPTION_SACK_ADVERTISE (1 << 0) |
374 | #define OPTION_TS (1 << 1) | |
375 | #define OPTION_MD5 (1 << 2) | |
89e95a61 | 376 | #define OPTION_WSCALE (1 << 3) |
bd0388ae | 377 | #define OPTION_COOKIE_EXTENSION (1 << 4) |
33ad798c AL |
378 | |
379 | struct tcp_out_options { | |
380 | u8 options; /* bit field of OPTION_* */ | |
381 | u8 ws; /* window scale, 0 to disable */ | |
382 | u8 num_sack_blocks; /* number of SACK blocks to include */ | |
bd0388ae | 383 | u8 hash_size; /* bytes in hash_location */ |
33ad798c AL |
384 | u16 mss; /* 0 to disable */ |
385 | __u32 tsval, tsecr; /* need to include OPTION_TS */ | |
bd0388ae | 386 | __u8 *hash_location; /* temporary pointer, overloaded */ |
33ad798c AL |
387 | }; |
388 | ||
bd0388ae WAS |
389 | /* The sysctl int routines are generic, so check consistency here. |
390 | */ | |
391 | static u8 tcp_cookie_size_check(u8 desired) | |
392 | { | |
393 | if (desired > 0) { | |
394 | /* previously specified */ | |
395 | return desired; | |
396 | } | |
397 | if (sysctl_tcp_cookie_size <= 0) { | |
398 | /* no default specified */ | |
399 | return 0; | |
400 | } | |
401 | if (sysctl_tcp_cookie_size <= TCP_COOKIE_MIN) { | |
402 | /* value too small, specify minimum */ | |
403 | return TCP_COOKIE_MIN; | |
404 | } | |
405 | if (sysctl_tcp_cookie_size >= TCP_COOKIE_MAX) { | |
406 | /* value too large, specify maximum */ | |
407 | return TCP_COOKIE_MAX; | |
408 | } | |
409 | if (0x1 & sysctl_tcp_cookie_size) { | |
410 | /* 8-bit multiple, illegal, fix it */ | |
411 | return (u8)(sysctl_tcp_cookie_size + 0x1); | |
412 | } | |
413 | return (u8)sysctl_tcp_cookie_size; | |
414 | } | |
415 | ||
67edfef7 AK |
416 | /* Write previously computed TCP options to the packet. |
417 | * | |
418 | * Beware: Something in the Internet is very sensitive to the ordering of | |
fd6149d3 IJ |
419 | * TCP options, we learned this through the hard way, so be careful here. |
420 | * Luckily we can at least blame others for their non-compliance but from | |
421 | * inter-operatibility perspective it seems that we're somewhat stuck with | |
422 | * the ordering which we have been using if we want to keep working with | |
423 | * those broken things (not that it currently hurts anybody as there isn't | |
424 | * particular reason why the ordering would need to be changed). | |
425 | * | |
426 | * At least SACK_PERM as the first option is known to lead to a disaster | |
427 | * (but it may well be that other scenarios fail similarly). | |
428 | */ | |
33ad798c | 429 | static void tcp_options_write(__be32 *ptr, struct tcp_sock *tp, |
bd0388ae WAS |
430 | struct tcp_out_options *opts) |
431 | { | |
432 | u8 options = opts->options; /* mungable copy */ | |
433 | ||
434 | /* Having both authentication and cookies for security is redundant, | |
435 | * and there's certainly not enough room. Instead, the cookie-less | |
436 | * extension variant is proposed. | |
437 | * | |
438 | * Consider the pessimal case with authentication. The options | |
439 | * could look like: | |
440 | * COOKIE|MD5(20) + MSS(4) + SACK|TS(12) + WSCALE(4) == 40 | |
441 | */ | |
442 | if (unlikely(OPTION_MD5 & options)) { | |
443 | if (unlikely(OPTION_COOKIE_EXTENSION & options)) { | |
444 | *ptr++ = htonl((TCPOPT_COOKIE << 24) | | |
445 | (TCPOLEN_COOKIE_BASE << 16) | | |
446 | (TCPOPT_MD5SIG << 8) | | |
447 | TCPOLEN_MD5SIG); | |
448 | } else { | |
449 | *ptr++ = htonl((TCPOPT_NOP << 24) | | |
450 | (TCPOPT_NOP << 16) | | |
451 | (TCPOPT_MD5SIG << 8) | | |
452 | TCPOLEN_MD5SIG); | |
453 | } | |
454 | options &= ~OPTION_COOKIE_EXTENSION; | |
455 | /* overload cookie hash location */ | |
456 | opts->hash_location = (__u8 *)ptr; | |
33ad798c | 457 | ptr += 4; |
40efc6fa | 458 | } |
33ad798c | 459 | |
fd6149d3 IJ |
460 | if (unlikely(opts->mss)) { |
461 | *ptr++ = htonl((TCPOPT_MSS << 24) | | |
462 | (TCPOLEN_MSS << 16) | | |
463 | opts->mss); | |
464 | } | |
465 | ||
bd0388ae WAS |
466 | if (likely(OPTION_TS & options)) { |
467 | if (unlikely(OPTION_SACK_ADVERTISE & options)) { | |
33ad798c AL |
468 | *ptr++ = htonl((TCPOPT_SACK_PERM << 24) | |
469 | (TCPOLEN_SACK_PERM << 16) | | |
470 | (TCPOPT_TIMESTAMP << 8) | | |
471 | TCPOLEN_TIMESTAMP); | |
bd0388ae | 472 | options &= ~OPTION_SACK_ADVERTISE; |
33ad798c AL |
473 | } else { |
474 | *ptr++ = htonl((TCPOPT_NOP << 24) | | |
475 | (TCPOPT_NOP << 16) | | |
476 | (TCPOPT_TIMESTAMP << 8) | | |
477 | TCPOLEN_TIMESTAMP); | |
478 | } | |
479 | *ptr++ = htonl(opts->tsval); | |
480 | *ptr++ = htonl(opts->tsecr); | |
481 | } | |
482 | ||
bd0388ae WAS |
483 | /* Specification requires after timestamp, so do it now. |
484 | * | |
485 | * Consider the pessimal case without authentication. The options | |
486 | * could look like: | |
487 | * MSS(4) + SACK|TS(12) + COOKIE(20) + WSCALE(4) == 40 | |
488 | */ | |
489 | if (unlikely(OPTION_COOKIE_EXTENSION & options)) { | |
490 | __u8 *cookie_copy = opts->hash_location; | |
491 | u8 cookie_size = opts->hash_size; | |
492 | ||
493 | /* 8-bit multiple handled in tcp_cookie_size_check() above, | |
494 | * and elsewhere. | |
495 | */ | |
496 | if (0x2 & cookie_size) { | |
497 | __u8 *p = (__u8 *)ptr; | |
498 | ||
499 | /* 16-bit multiple */ | |
500 | *p++ = TCPOPT_COOKIE; | |
501 | *p++ = TCPOLEN_COOKIE_BASE + cookie_size; | |
502 | *p++ = *cookie_copy++; | |
503 | *p++ = *cookie_copy++; | |
504 | ptr++; | |
505 | cookie_size -= 2; | |
506 | } else { | |
507 | /* 32-bit multiple */ | |
508 | *ptr++ = htonl(((TCPOPT_NOP << 24) | | |
509 | (TCPOPT_NOP << 16) | | |
510 | (TCPOPT_COOKIE << 8) | | |
511 | TCPOLEN_COOKIE_BASE) + | |
512 | cookie_size); | |
513 | } | |
514 | ||
515 | if (cookie_size > 0) { | |
516 | memcpy(ptr, cookie_copy, cookie_size); | |
517 | ptr += (cookie_size / 4); | |
518 | } | |
519 | } | |
520 | ||
521 | if (unlikely(OPTION_SACK_ADVERTISE & options)) { | |
33ad798c AL |
522 | *ptr++ = htonl((TCPOPT_NOP << 24) | |
523 | (TCPOPT_NOP << 16) | | |
524 | (TCPOPT_SACK_PERM << 8) | | |
525 | TCPOLEN_SACK_PERM); | |
526 | } | |
527 | ||
bd0388ae | 528 | if (unlikely(OPTION_WSCALE & options)) { |
33ad798c AL |
529 | *ptr++ = htonl((TCPOPT_NOP << 24) | |
530 | (TCPOPT_WINDOW << 16) | | |
531 | (TCPOLEN_WINDOW << 8) | | |
532 | opts->ws); | |
533 | } | |
534 | ||
535 | if (unlikely(opts->num_sack_blocks)) { | |
536 | struct tcp_sack_block *sp = tp->rx_opt.dsack ? | |
537 | tp->duplicate_sack : tp->selective_acks; | |
40efc6fa SH |
538 | int this_sack; |
539 | ||
540 | *ptr++ = htonl((TCPOPT_NOP << 24) | | |
541 | (TCPOPT_NOP << 16) | | |
542 | (TCPOPT_SACK << 8) | | |
33ad798c | 543 | (TCPOLEN_SACK_BASE + (opts->num_sack_blocks * |
40efc6fa | 544 | TCPOLEN_SACK_PERBLOCK))); |
2de979bd | 545 | |
33ad798c AL |
546 | for (this_sack = 0; this_sack < opts->num_sack_blocks; |
547 | ++this_sack) { | |
40efc6fa SH |
548 | *ptr++ = htonl(sp[this_sack].start_seq); |
549 | *ptr++ = htonl(sp[this_sack].end_seq); | |
550 | } | |
2de979bd | 551 | |
5861f8e5 | 552 | tp->rx_opt.dsack = 0; |
40efc6fa | 553 | } |
33ad798c AL |
554 | } |
555 | ||
67edfef7 AK |
556 | /* Compute TCP options for SYN packets. This is not the final |
557 | * network wire format yet. | |
558 | */ | |
33ad798c AL |
559 | static unsigned tcp_syn_options(struct sock *sk, struct sk_buff *skb, |
560 | struct tcp_out_options *opts, | |
561 | struct tcp_md5sig_key **md5) { | |
562 | struct tcp_sock *tp = tcp_sk(sk); | |
bd0388ae | 563 | struct tcp_cookie_values *cvp = tp->cookie_values; |
bd0388ae WAS |
564 | unsigned remaining = MAX_TCP_OPTION_SPACE; |
565 | u8 cookie_size = (!tp->rx_opt.cookie_out_never && cvp != NULL) ? | |
566 | tcp_cookie_size_check(cvp->cookie_desired) : | |
567 | 0; | |
33ad798c | 568 | |
cfb6eeb4 | 569 | #ifdef CONFIG_TCP_MD5SIG |
33ad798c AL |
570 | *md5 = tp->af_specific->md5_lookup(sk, sk); |
571 | if (*md5) { | |
572 | opts->options |= OPTION_MD5; | |
bd0388ae | 573 | remaining -= TCPOLEN_MD5SIG_ALIGNED; |
cfb6eeb4 | 574 | } |
33ad798c AL |
575 | #else |
576 | *md5 = NULL; | |
cfb6eeb4 | 577 | #endif |
33ad798c AL |
578 | |
579 | /* We always get an MSS option. The option bytes which will be seen in | |
580 | * normal data packets should timestamps be used, must be in the MSS | |
581 | * advertised. But we subtract them from tp->mss_cache so that | |
582 | * calculations in tcp_sendmsg are simpler etc. So account for this | |
583 | * fact here if necessary. If we don't do this correctly, as a | |
584 | * receiver we won't recognize data packets as being full sized when we | |
585 | * should, and thus we won't abide by the delayed ACK rules correctly. | |
586 | * SACKs don't matter, we never delay an ACK when we have any of those | |
587 | * going out. */ | |
588 | opts->mss = tcp_advertise_mss(sk); | |
bd0388ae | 589 | remaining -= TCPOLEN_MSS_ALIGNED; |
33ad798c | 590 | |
bb5b7c11 | 591 | if (likely(sysctl_tcp_timestamps && *md5 == NULL)) { |
33ad798c AL |
592 | opts->options |= OPTION_TS; |
593 | opts->tsval = TCP_SKB_CB(skb)->when; | |
594 | opts->tsecr = tp->rx_opt.ts_recent; | |
bd0388ae | 595 | remaining -= TCPOLEN_TSTAMP_ALIGNED; |
33ad798c | 596 | } |
bb5b7c11 | 597 | if (likely(sysctl_tcp_window_scaling)) { |
33ad798c | 598 | opts->ws = tp->rx_opt.rcv_wscale; |
89e95a61 | 599 | opts->options |= OPTION_WSCALE; |
bd0388ae | 600 | remaining -= TCPOLEN_WSCALE_ALIGNED; |
33ad798c | 601 | } |
bb5b7c11 | 602 | if (likely(sysctl_tcp_sack)) { |
33ad798c | 603 | opts->options |= OPTION_SACK_ADVERTISE; |
b32d1310 | 604 | if (unlikely(!(OPTION_TS & opts->options))) |
bd0388ae | 605 | remaining -= TCPOLEN_SACKPERM_ALIGNED; |
33ad798c AL |
606 | } |
607 | ||
bd0388ae WAS |
608 | /* Note that timestamps are required by the specification. |
609 | * | |
610 | * Odd numbers of bytes are prohibited by the specification, ensuring | |
611 | * that the cookie is 16-bit aligned, and the resulting cookie pair is | |
612 | * 32-bit aligned. | |
613 | */ | |
614 | if (*md5 == NULL && | |
615 | (OPTION_TS & opts->options) && | |
616 | cookie_size > 0) { | |
617 | int need = TCPOLEN_COOKIE_BASE + cookie_size; | |
618 | ||
619 | if (0x2 & need) { | |
620 | /* 32-bit multiple */ | |
621 | need += 2; /* NOPs */ | |
622 | ||
623 | if (need > remaining) { | |
624 | /* try shrinking cookie to fit */ | |
625 | cookie_size -= 2; | |
626 | need -= 4; | |
627 | } | |
628 | } | |
629 | while (need > remaining && TCP_COOKIE_MIN <= cookie_size) { | |
630 | cookie_size -= 4; | |
631 | need -= 4; | |
632 | } | |
633 | if (TCP_COOKIE_MIN <= cookie_size) { | |
634 | opts->options |= OPTION_COOKIE_EXTENSION; | |
635 | opts->hash_location = (__u8 *)&cvp->cookie_pair[0]; | |
636 | opts->hash_size = cookie_size; | |
637 | ||
638 | /* Remember for future incarnations. */ | |
639 | cvp->cookie_desired = cookie_size; | |
640 | ||
641 | if (cvp->cookie_desired != cvp->cookie_pair_size) { | |
642 | /* Currently use random bytes as a nonce, | |
643 | * assuming these are completely unpredictable | |
644 | * by hostile users of the same system. | |
645 | */ | |
646 | get_random_bytes(&cvp->cookie_pair[0], | |
647 | cookie_size); | |
648 | cvp->cookie_pair_size = cookie_size; | |
649 | } | |
650 | ||
651 | remaining -= need; | |
652 | } | |
653 | } | |
654 | return MAX_TCP_OPTION_SPACE - remaining; | |
40efc6fa SH |
655 | } |
656 | ||
67edfef7 | 657 | /* Set up TCP options for SYN-ACKs. */ |
33ad798c AL |
658 | static unsigned tcp_synack_options(struct sock *sk, |
659 | struct request_sock *req, | |
660 | unsigned mss, struct sk_buff *skb, | |
661 | struct tcp_out_options *opts, | |
4957faad WAS |
662 | struct tcp_md5sig_key **md5, |
663 | struct tcp_extend_values *xvp) | |
664 | { | |
33ad798c | 665 | struct inet_request_sock *ireq = inet_rsk(req); |
4957faad WAS |
666 | unsigned remaining = MAX_TCP_OPTION_SPACE; |
667 | u8 cookie_plus = (xvp != NULL && !xvp->cookie_out_never) ? | |
668 | xvp->cookie_plus : | |
669 | 0; | |
670 | bool doing_ts = ireq->tstamp_ok; | |
33ad798c | 671 | |
cfb6eeb4 | 672 | #ifdef CONFIG_TCP_MD5SIG |
33ad798c AL |
673 | *md5 = tcp_rsk(req)->af_specific->md5_lookup(sk, req); |
674 | if (*md5) { | |
675 | opts->options |= OPTION_MD5; | |
4957faad WAS |
676 | remaining -= TCPOLEN_MD5SIG_ALIGNED; |
677 | ||
678 | /* We can't fit any SACK blocks in a packet with MD5 + TS | |
679 | * options. There was discussion about disabling SACK | |
680 | * rather than TS in order to fit in better with old, | |
681 | * buggy kernels, but that was deemed to be unnecessary. | |
682 | */ | |
683 | doing_ts &= !ireq->sack_ok; | |
cfb6eeb4 | 684 | } |
33ad798c AL |
685 | #else |
686 | *md5 = NULL; | |
cfb6eeb4 | 687 | #endif |
33ad798c | 688 | |
4957faad | 689 | /* We always send an MSS option. */ |
33ad798c | 690 | opts->mss = mss; |
4957faad | 691 | remaining -= TCPOLEN_MSS_ALIGNED; |
33ad798c AL |
692 | |
693 | if (likely(ireq->wscale_ok)) { | |
694 | opts->ws = ireq->rcv_wscale; | |
89e95a61 | 695 | opts->options |= OPTION_WSCALE; |
4957faad | 696 | remaining -= TCPOLEN_WSCALE_ALIGNED; |
33ad798c AL |
697 | } |
698 | if (likely(doing_ts)) { | |
699 | opts->options |= OPTION_TS; | |
700 | opts->tsval = TCP_SKB_CB(skb)->when; | |
701 | opts->tsecr = req->ts_recent; | |
4957faad | 702 | remaining -= TCPOLEN_TSTAMP_ALIGNED; |
33ad798c AL |
703 | } |
704 | if (likely(ireq->sack_ok)) { | |
705 | opts->options |= OPTION_SACK_ADVERTISE; | |
706 | if (unlikely(!doing_ts)) | |
4957faad | 707 | remaining -= TCPOLEN_SACKPERM_ALIGNED; |
33ad798c AL |
708 | } |
709 | ||
4957faad WAS |
710 | /* Similar rationale to tcp_syn_options() applies here, too. |
711 | * If the <SYN> options fit, the same options should fit now! | |
712 | */ | |
713 | if (*md5 == NULL && | |
714 | doing_ts && | |
715 | cookie_plus > TCPOLEN_COOKIE_BASE) { | |
716 | int need = cookie_plus; /* has TCPOLEN_COOKIE_BASE */ | |
717 | ||
718 | if (0x2 & need) { | |
719 | /* 32-bit multiple */ | |
720 | need += 2; /* NOPs */ | |
721 | } | |
722 | if (need <= remaining) { | |
723 | opts->options |= OPTION_COOKIE_EXTENSION; | |
724 | opts->hash_size = cookie_plus - TCPOLEN_COOKIE_BASE; | |
725 | remaining -= need; | |
726 | } else { | |
727 | /* There's no error return, so flag it. */ | |
728 | xvp->cookie_out_never = 1; /* true */ | |
729 | opts->hash_size = 0; | |
730 | } | |
731 | } | |
732 | return MAX_TCP_OPTION_SPACE - remaining; | |
33ad798c AL |
733 | } |
734 | ||
67edfef7 AK |
735 | /* Compute TCP options for ESTABLISHED sockets. This is not the |
736 | * final wire format yet. | |
737 | */ | |
33ad798c AL |
738 | static unsigned tcp_established_options(struct sock *sk, struct sk_buff *skb, |
739 | struct tcp_out_options *opts, | |
740 | struct tcp_md5sig_key **md5) { | |
741 | struct tcp_skb_cb *tcb = skb ? TCP_SKB_CB(skb) : NULL; | |
742 | struct tcp_sock *tp = tcp_sk(sk); | |
743 | unsigned size = 0; | |
cabeccbd | 744 | unsigned int eff_sacks; |
33ad798c AL |
745 | |
746 | #ifdef CONFIG_TCP_MD5SIG | |
747 | *md5 = tp->af_specific->md5_lookup(sk, sk); | |
748 | if (unlikely(*md5)) { | |
749 | opts->options |= OPTION_MD5; | |
750 | size += TCPOLEN_MD5SIG_ALIGNED; | |
751 | } | |
752 | #else | |
753 | *md5 = NULL; | |
754 | #endif | |
755 | ||
756 | if (likely(tp->rx_opt.tstamp_ok)) { | |
757 | opts->options |= OPTION_TS; | |
758 | opts->tsval = tcb ? tcb->when : 0; | |
759 | opts->tsecr = tp->rx_opt.ts_recent; | |
760 | size += TCPOLEN_TSTAMP_ALIGNED; | |
761 | } | |
762 | ||
cabeccbd IJ |
763 | eff_sacks = tp->rx_opt.num_sacks + tp->rx_opt.dsack; |
764 | if (unlikely(eff_sacks)) { | |
33ad798c AL |
765 | const unsigned remaining = MAX_TCP_OPTION_SPACE - size; |
766 | opts->num_sack_blocks = | |
cabeccbd | 767 | min_t(unsigned, eff_sacks, |
33ad798c AL |
768 | (remaining - TCPOLEN_SACK_BASE_ALIGNED) / |
769 | TCPOLEN_SACK_PERBLOCK); | |
770 | size += TCPOLEN_SACK_BASE_ALIGNED + | |
771 | opts->num_sack_blocks * TCPOLEN_SACK_PERBLOCK; | |
772 | } | |
773 | ||
774 | return size; | |
40efc6fa | 775 | } |
1da177e4 LT |
776 | |
777 | /* This routine actually transmits TCP packets queued in by | |
778 | * tcp_do_sendmsg(). This is used by both the initial | |
779 | * transmission and possible later retransmissions. | |
780 | * All SKB's seen here are completely headerless. It is our | |
781 | * job to build the TCP header, and pass the packet down to | |
782 | * IP so it can do the same plus pass the packet off to the | |
783 | * device. | |
784 | * | |
785 | * We are working here with either a clone of the original | |
786 | * SKB, or a fresh unique copy made by the retransmit engine. | |
787 | */ | |
056834d9 IJ |
788 | static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it, |
789 | gfp_t gfp_mask) | |
1da177e4 | 790 | { |
dfb4b9dc DM |
791 | const struct inet_connection_sock *icsk = inet_csk(sk); |
792 | struct inet_sock *inet; | |
793 | struct tcp_sock *tp; | |
794 | struct tcp_skb_cb *tcb; | |
33ad798c AL |
795 | struct tcp_out_options opts; |
796 | unsigned tcp_options_size, tcp_header_size; | |
cfb6eeb4 | 797 | struct tcp_md5sig_key *md5; |
dfb4b9dc | 798 | struct tcphdr *th; |
dfb4b9dc DM |
799 | int err; |
800 | ||
801 | BUG_ON(!skb || !tcp_skb_pcount(skb)); | |
802 | ||
803 | /* If congestion control is doing timestamping, we must | |
804 | * take such a timestamp before we potentially clone/copy. | |
805 | */ | |
164891aa | 806 | if (icsk->icsk_ca_ops->flags & TCP_CONG_RTT_STAMP) |
dfb4b9dc DM |
807 | __net_timestamp(skb); |
808 | ||
809 | if (likely(clone_it)) { | |
810 | if (unlikely(skb_cloned(skb))) | |
811 | skb = pskb_copy(skb, gfp_mask); | |
812 | else | |
813 | skb = skb_clone(skb, gfp_mask); | |
814 | if (unlikely(!skb)) | |
815 | return -ENOBUFS; | |
816 | } | |
1da177e4 | 817 | |
dfb4b9dc DM |
818 | inet = inet_sk(sk); |
819 | tp = tcp_sk(sk); | |
820 | tcb = TCP_SKB_CB(skb); | |
33ad798c | 821 | memset(&opts, 0, sizeof(opts)); |
1da177e4 | 822 | |
33ad798c AL |
823 | if (unlikely(tcb->flags & TCPCB_FLAG_SYN)) |
824 | tcp_options_size = tcp_syn_options(sk, skb, &opts, &md5); | |
825 | else | |
826 | tcp_options_size = tcp_established_options(sk, skb, &opts, | |
827 | &md5); | |
828 | tcp_header_size = tcp_options_size + sizeof(struct tcphdr); | |
e905a9ed | 829 | |
dfb4b9dc DM |
830 | if (tcp_packets_in_flight(tp) == 0) |
831 | tcp_ca_event(sk, CA_EVENT_TX_START); | |
832 | ||
aa8223c7 ACM |
833 | skb_push(skb, tcp_header_size); |
834 | skb_reset_transport_header(skb); | |
e89862f4 | 835 | skb_set_owner_w(skb, sk); |
dfb4b9dc DM |
836 | |
837 | /* Build TCP header and checksum it. */ | |
aa8223c7 | 838 | th = tcp_hdr(skb); |
c720c7e8 ED |
839 | th->source = inet->inet_sport; |
840 | th->dest = inet->inet_dport; | |
dfb4b9dc DM |
841 | th->seq = htonl(tcb->seq); |
842 | th->ack_seq = htonl(tp->rcv_nxt); | |
df7a3b07 | 843 | *(((__be16 *)th) + 6) = htons(((tcp_header_size >> 2) << 12) | |
dfb4b9dc DM |
844 | tcb->flags); |
845 | ||
846 | if (unlikely(tcb->flags & TCPCB_FLAG_SYN)) { | |
847 | /* RFC1323: The window in SYN & SYN/ACK segments | |
848 | * is never scaled. | |
849 | */ | |
600ff0c2 | 850 | th->window = htons(min(tp->rcv_wnd, 65535U)); |
dfb4b9dc DM |
851 | } else { |
852 | th->window = htons(tcp_select_window(sk)); | |
853 | } | |
854 | th->check = 0; | |
855 | th->urg_ptr = 0; | |
1da177e4 | 856 | |
33f5f57e | 857 | /* The urg_mode check is necessary during a below snd_una win probe */ |
7691367d HX |
858 | if (unlikely(tcp_urg_mode(tp) && before(tcb->seq, tp->snd_up))) { |
859 | if (before(tp->snd_up, tcb->seq + 0x10000)) { | |
860 | th->urg_ptr = htons(tp->snd_up - tcb->seq); | |
861 | th->urg = 1; | |
862 | } else if (after(tcb->seq + 0xFFFF, tp->snd_nxt)) { | |
863 | th->urg_ptr = 0xFFFF; | |
864 | th->urg = 1; | |
865 | } | |
dfb4b9dc | 866 | } |
1da177e4 | 867 | |
bd0388ae | 868 | tcp_options_write((__be32 *)(th + 1), tp, &opts); |
33ad798c | 869 | if (likely((tcb->flags & TCPCB_FLAG_SYN) == 0)) |
9e412ba7 | 870 | TCP_ECN_send(sk, skb, tcp_header_size); |
1da177e4 | 871 | |
cfb6eeb4 YH |
872 | #ifdef CONFIG_TCP_MD5SIG |
873 | /* Calculate the MD5 hash, as we have all we need now */ | |
874 | if (md5) { | |
33ad798c | 875 | sk->sk_route_caps &= ~NETIF_F_GSO_MASK; |
bd0388ae | 876 | tp->af_specific->calc_md5_hash(opts.hash_location, |
49a72dfb | 877 | md5, sk, NULL, skb); |
cfb6eeb4 YH |
878 | } |
879 | #endif | |
880 | ||
8292a17a | 881 | icsk->icsk_af_ops->send_check(sk, skb->len, skb); |
1da177e4 | 882 | |
dfb4b9dc DM |
883 | if (likely(tcb->flags & TCPCB_FLAG_ACK)) |
884 | tcp_event_ack_sent(sk, tcp_skb_pcount(skb)); | |
1da177e4 | 885 | |
dfb4b9dc DM |
886 | if (skb->len != tcp_header_size) |
887 | tcp_event_data_sent(tp, skb, sk); | |
1da177e4 | 888 | |
bd37a088 | 889 | if (after(tcb->end_seq, tp->snd_nxt) || tcb->seq == tcb->end_seq) |
81cc8a75 | 890 | TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTSEGS); |
1da177e4 | 891 | |
e89862f4 | 892 | err = icsk->icsk_af_ops->queue_xmit(skb, 0); |
83de47cd | 893 | if (likely(err <= 0)) |
dfb4b9dc DM |
894 | return err; |
895 | ||
3cfe3baa | 896 | tcp_enter_cwr(sk, 1); |
dfb4b9dc | 897 | |
b9df3cb8 | 898 | return net_xmit_eval(err); |
1da177e4 LT |
899 | } |
900 | ||
67edfef7 | 901 | /* This routine just queues the buffer for sending. |
1da177e4 LT |
902 | * |
903 | * NOTE: probe0 timer is not checked, do not forget tcp_push_pending_frames, | |
904 | * otherwise socket can stall. | |
905 | */ | |
906 | static void tcp_queue_skb(struct sock *sk, struct sk_buff *skb) | |
907 | { | |
908 | struct tcp_sock *tp = tcp_sk(sk); | |
909 | ||
910 | /* Advance write_seq and place onto the write_queue. */ | |
911 | tp->write_seq = TCP_SKB_CB(skb)->end_seq; | |
912 | skb_header_release(skb); | |
fe067e8a | 913 | tcp_add_write_queue_tail(sk, skb); |
3ab224be HA |
914 | sk->sk_wmem_queued += skb->truesize; |
915 | sk_mem_charge(sk, skb->truesize); | |
1da177e4 LT |
916 | } |
917 | ||
67edfef7 | 918 | /* Initialize TSO segments for a packet. */ |
056834d9 IJ |
919 | static void tcp_set_skb_tso_segs(struct sock *sk, struct sk_buff *skb, |
920 | unsigned int mss_now) | |
f6302d1d | 921 | { |
8e5b9dda HX |
922 | if (skb->len <= mss_now || !sk_can_gso(sk) || |
923 | skb->ip_summed == CHECKSUM_NONE) { | |
f6302d1d DM |
924 | /* Avoid the costly divide in the normal |
925 | * non-TSO case. | |
926 | */ | |
7967168c HX |
927 | skb_shinfo(skb)->gso_segs = 1; |
928 | skb_shinfo(skb)->gso_size = 0; | |
929 | skb_shinfo(skb)->gso_type = 0; | |
f6302d1d | 930 | } else { |
356f89e1 | 931 | skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss_now); |
7967168c | 932 | skb_shinfo(skb)->gso_size = mss_now; |
bcd76111 | 933 | skb_shinfo(skb)->gso_type = sk->sk_gso_type; |
1da177e4 LT |
934 | } |
935 | } | |
936 | ||
91fed7a1 | 937 | /* When a modification to fackets out becomes necessary, we need to check |
68f8353b | 938 | * skb is counted to fackets_out or not. |
91fed7a1 | 939 | */ |
a47e5a98 | 940 | static void tcp_adjust_fackets_out(struct sock *sk, struct sk_buff *skb, |
91fed7a1 IJ |
941 | int decr) |
942 | { | |
a47e5a98 IJ |
943 | struct tcp_sock *tp = tcp_sk(sk); |
944 | ||
dc86967b | 945 | if (!tp->sacked_out || tcp_is_reno(tp)) |
91fed7a1 IJ |
946 | return; |
947 | ||
6859d494 | 948 | if (after(tcp_highest_sack_seq(tp), TCP_SKB_CB(skb)->seq)) |
91fed7a1 | 949 | tp->fackets_out -= decr; |
91fed7a1 IJ |
950 | } |
951 | ||
797108d1 IJ |
952 | /* Pcount in the middle of the write queue got changed, we need to do various |
953 | * tweaks to fix counters | |
954 | */ | |
955 | static void tcp_adjust_pcount(struct sock *sk, struct sk_buff *skb, int decr) | |
956 | { | |
957 | struct tcp_sock *tp = tcp_sk(sk); | |
958 | ||
959 | tp->packets_out -= decr; | |
960 | ||
961 | if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) | |
962 | tp->sacked_out -= decr; | |
963 | if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) | |
964 | tp->retrans_out -= decr; | |
965 | if (TCP_SKB_CB(skb)->sacked & TCPCB_LOST) | |
966 | tp->lost_out -= decr; | |
967 | ||
968 | /* Reno case is special. Sigh... */ | |
969 | if (tcp_is_reno(tp) && decr > 0) | |
970 | tp->sacked_out -= min_t(u32, tp->sacked_out, decr); | |
971 | ||
972 | tcp_adjust_fackets_out(sk, skb, decr); | |
973 | ||
974 | if (tp->lost_skb_hint && | |
975 | before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(tp->lost_skb_hint)->seq) && | |
52cf3cc8 | 976 | (tcp_is_fack(tp) || (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED))) |
797108d1 IJ |
977 | tp->lost_cnt_hint -= decr; |
978 | ||
979 | tcp_verify_left_out(tp); | |
980 | } | |
981 | ||
1da177e4 LT |
982 | /* Function to create two new TCP segments. Shrinks the given segment |
983 | * to the specified size and appends a new segment with the rest of the | |
e905a9ed | 984 | * packet to the list. This won't be called frequently, I hope. |
1da177e4 LT |
985 | * Remember, these are still headerless SKBs at this point. |
986 | */ | |
056834d9 IJ |
987 | int tcp_fragment(struct sock *sk, struct sk_buff *skb, u32 len, |
988 | unsigned int mss_now) | |
1da177e4 LT |
989 | { |
990 | struct tcp_sock *tp = tcp_sk(sk); | |
991 | struct sk_buff *buff; | |
6475be16 | 992 | int nsize, old_factor; |
b60b49ea | 993 | int nlen; |
9ce01461 | 994 | u8 flags; |
1da177e4 | 995 | |
b2cc99f0 | 996 | BUG_ON(len > skb->len); |
6a438bbe | 997 | |
1da177e4 LT |
998 | nsize = skb_headlen(skb) - len; |
999 | if (nsize < 0) | |
1000 | nsize = 0; | |
1001 | ||
1002 | if (skb_cloned(skb) && | |
1003 | skb_is_nonlinear(skb) && | |
1004 | pskb_expand_head(skb, 0, 0, GFP_ATOMIC)) | |
1005 | return -ENOMEM; | |
1006 | ||
1007 | /* Get a new skb... force flag on. */ | |
1008 | buff = sk_stream_alloc_skb(sk, nsize, GFP_ATOMIC); | |
1009 | if (buff == NULL) | |
1010 | return -ENOMEM; /* We'll just try again later. */ | |
ef5cb973 | 1011 | |
3ab224be HA |
1012 | sk->sk_wmem_queued += buff->truesize; |
1013 | sk_mem_charge(sk, buff->truesize); | |
b60b49ea HX |
1014 | nlen = skb->len - len - nsize; |
1015 | buff->truesize += nlen; | |
1016 | skb->truesize -= nlen; | |
1da177e4 LT |
1017 | |
1018 | /* Correct the sequence numbers. */ | |
1019 | TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len; | |
1020 | TCP_SKB_CB(buff)->end_seq = TCP_SKB_CB(skb)->end_seq; | |
1021 | TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(buff)->seq; | |
1022 | ||
1023 | /* PSH and FIN should only be set in the second packet. */ | |
1024 | flags = TCP_SKB_CB(skb)->flags; | |
056834d9 | 1025 | TCP_SKB_CB(skb)->flags = flags & ~(TCPCB_FLAG_FIN | TCPCB_FLAG_PSH); |
1da177e4 | 1026 | TCP_SKB_CB(buff)->flags = flags; |
e14c3caf | 1027 | TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked; |
1da177e4 | 1028 | |
84fa7933 | 1029 | if (!skb_shinfo(skb)->nr_frags && skb->ip_summed != CHECKSUM_PARTIAL) { |
1da177e4 | 1030 | /* Copy and checksum data tail into the new buffer. */ |
056834d9 IJ |
1031 | buff->csum = csum_partial_copy_nocheck(skb->data + len, |
1032 | skb_put(buff, nsize), | |
1da177e4 LT |
1033 | nsize, 0); |
1034 | ||
1035 | skb_trim(skb, len); | |
1036 | ||
1037 | skb->csum = csum_block_sub(skb->csum, buff->csum, len); | |
1038 | } else { | |
84fa7933 | 1039 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 LT |
1040 | skb_split(skb, buff, len); |
1041 | } | |
1042 | ||
1043 | buff->ip_summed = skb->ip_summed; | |
1044 | ||
1045 | /* Looks stupid, but our code really uses when of | |
1046 | * skbs, which it never sent before. --ANK | |
1047 | */ | |
1048 | TCP_SKB_CB(buff)->when = TCP_SKB_CB(skb)->when; | |
a61bbcf2 | 1049 | buff->tstamp = skb->tstamp; |
1da177e4 | 1050 | |
6475be16 DM |
1051 | old_factor = tcp_skb_pcount(skb); |
1052 | ||
1da177e4 | 1053 | /* Fix up tso_factor for both original and new SKB. */ |
846998ae DM |
1054 | tcp_set_skb_tso_segs(sk, skb, mss_now); |
1055 | tcp_set_skb_tso_segs(sk, buff, mss_now); | |
1da177e4 | 1056 | |
6475be16 DM |
1057 | /* If this packet has been sent out already, we must |
1058 | * adjust the various packet counters. | |
1059 | */ | |
cf0b450c | 1060 | if (!before(tp->snd_nxt, TCP_SKB_CB(buff)->end_seq)) { |
6475be16 DM |
1061 | int diff = old_factor - tcp_skb_pcount(skb) - |
1062 | tcp_skb_pcount(buff); | |
1da177e4 | 1063 | |
797108d1 IJ |
1064 | if (diff) |
1065 | tcp_adjust_pcount(sk, skb, diff); | |
1da177e4 LT |
1066 | } |
1067 | ||
1068 | /* Link BUFF into the send queue. */ | |
f44b5271 | 1069 | skb_header_release(buff); |
fe067e8a | 1070 | tcp_insert_write_queue_after(skb, buff, sk); |
1da177e4 LT |
1071 | |
1072 | return 0; | |
1073 | } | |
1074 | ||
1075 | /* This is similar to __pskb_pull_head() (it will go to core/skbuff.c | |
1076 | * eventually). The difference is that pulled data not copied, but | |
1077 | * immediately discarded. | |
1078 | */ | |
f2911969 | 1079 | static void __pskb_trim_head(struct sk_buff *skb, int len) |
1da177e4 LT |
1080 | { |
1081 | int i, k, eat; | |
1082 | ||
1083 | eat = len; | |
1084 | k = 0; | |
056834d9 | 1085 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
1da177e4 LT |
1086 | if (skb_shinfo(skb)->frags[i].size <= eat) { |
1087 | put_page(skb_shinfo(skb)->frags[i].page); | |
1088 | eat -= skb_shinfo(skb)->frags[i].size; | |
1089 | } else { | |
1090 | skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; | |
1091 | if (eat) { | |
1092 | skb_shinfo(skb)->frags[k].page_offset += eat; | |
1093 | skb_shinfo(skb)->frags[k].size -= eat; | |
1094 | eat = 0; | |
1095 | } | |
1096 | k++; | |
1097 | } | |
1098 | } | |
1099 | skb_shinfo(skb)->nr_frags = k; | |
1100 | ||
27a884dc | 1101 | skb_reset_tail_pointer(skb); |
1da177e4 LT |
1102 | skb->data_len -= len; |
1103 | skb->len = skb->data_len; | |
1da177e4 LT |
1104 | } |
1105 | ||
67edfef7 | 1106 | /* Remove acked data from a packet in the transmit queue. */ |
1da177e4 LT |
1107 | int tcp_trim_head(struct sock *sk, struct sk_buff *skb, u32 len) |
1108 | { | |
056834d9 | 1109 | if (skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC)) |
1da177e4 LT |
1110 | return -ENOMEM; |
1111 | ||
f2911969 HXP |
1112 | /* If len == headlen, we avoid __skb_pull to preserve alignment. */ |
1113 | if (unlikely(len < skb_headlen(skb))) | |
1da177e4 | 1114 | __skb_pull(skb, len); |
f2911969 HXP |
1115 | else |
1116 | __pskb_trim_head(skb, len - skb_headlen(skb)); | |
1da177e4 LT |
1117 | |
1118 | TCP_SKB_CB(skb)->seq += len; | |
84fa7933 | 1119 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 LT |
1120 | |
1121 | skb->truesize -= len; | |
1122 | sk->sk_wmem_queued -= len; | |
3ab224be | 1123 | sk_mem_uncharge(sk, len); |
1da177e4 LT |
1124 | sock_set_flag(sk, SOCK_QUEUE_SHRUNK); |
1125 | ||
1126 | /* Any change of skb->len requires recalculation of tso | |
1127 | * factor and mss. | |
1128 | */ | |
1129 | if (tcp_skb_pcount(skb) > 1) | |
0c54b85f | 1130 | tcp_set_skb_tso_segs(sk, skb, tcp_current_mss(sk)); |
1da177e4 LT |
1131 | |
1132 | return 0; | |
1133 | } | |
1134 | ||
67edfef7 | 1135 | /* Calculate MSS. Not accounting for SACKs here. */ |
5d424d5a JH |
1136 | int tcp_mtu_to_mss(struct sock *sk, int pmtu) |
1137 | { | |
1138 | struct tcp_sock *tp = tcp_sk(sk); | |
1139 | struct inet_connection_sock *icsk = inet_csk(sk); | |
1140 | int mss_now; | |
1141 | ||
1142 | /* Calculate base mss without TCP options: | |
1143 | It is MMS_S - sizeof(tcphdr) of rfc1122 | |
1144 | */ | |
1145 | mss_now = pmtu - icsk->icsk_af_ops->net_header_len - sizeof(struct tcphdr); | |
1146 | ||
1147 | /* Clamp it (mss_clamp does not include tcp options) */ | |
1148 | if (mss_now > tp->rx_opt.mss_clamp) | |
1149 | mss_now = tp->rx_opt.mss_clamp; | |
1150 | ||
1151 | /* Now subtract optional transport overhead */ | |
1152 | mss_now -= icsk->icsk_ext_hdr_len; | |
1153 | ||
1154 | /* Then reserve room for full set of TCP options and 8 bytes of data */ | |
1155 | if (mss_now < 48) | |
1156 | mss_now = 48; | |
1157 | ||
1158 | /* Now subtract TCP options size, not including SACKs */ | |
1159 | mss_now -= tp->tcp_header_len - sizeof(struct tcphdr); | |
1160 | ||
1161 | return mss_now; | |
1162 | } | |
1163 | ||
1164 | /* Inverse of above */ | |
1165 | int tcp_mss_to_mtu(struct sock *sk, int mss) | |
1166 | { | |
1167 | struct tcp_sock *tp = tcp_sk(sk); | |
1168 | struct inet_connection_sock *icsk = inet_csk(sk); | |
1169 | int mtu; | |
1170 | ||
1171 | mtu = mss + | |
1172 | tp->tcp_header_len + | |
1173 | icsk->icsk_ext_hdr_len + | |
1174 | icsk->icsk_af_ops->net_header_len; | |
1175 | ||
1176 | return mtu; | |
1177 | } | |
1178 | ||
67edfef7 | 1179 | /* MTU probing init per socket */ |
5d424d5a JH |
1180 | void tcp_mtup_init(struct sock *sk) |
1181 | { | |
1182 | struct tcp_sock *tp = tcp_sk(sk); | |
1183 | struct inet_connection_sock *icsk = inet_csk(sk); | |
1184 | ||
1185 | icsk->icsk_mtup.enabled = sysctl_tcp_mtu_probing > 1; | |
1186 | icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + sizeof(struct tcphdr) + | |
e905a9ed | 1187 | icsk->icsk_af_ops->net_header_len; |
5d424d5a JH |
1188 | icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, sysctl_tcp_base_mss); |
1189 | icsk->icsk_mtup.probe_size = 0; | |
1190 | } | |
1191 | ||
1da177e4 LT |
1192 | /* This function synchronize snd mss to current pmtu/exthdr set. |
1193 | ||
1194 | tp->rx_opt.user_mss is mss set by user by TCP_MAXSEG. It does NOT counts | |
1195 | for TCP options, but includes only bare TCP header. | |
1196 | ||
1197 | tp->rx_opt.mss_clamp is mss negotiated at connection setup. | |
caa20d9a | 1198 | It is minimum of user_mss and mss received with SYN. |
1da177e4 LT |
1199 | It also does not include TCP options. |
1200 | ||
d83d8461 | 1201 | inet_csk(sk)->icsk_pmtu_cookie is last pmtu, seen by this function. |
1da177e4 LT |
1202 | |
1203 | tp->mss_cache is current effective sending mss, including | |
1204 | all tcp options except for SACKs. It is evaluated, | |
1205 | taking into account current pmtu, but never exceeds | |
1206 | tp->rx_opt.mss_clamp. | |
1207 | ||
1208 | NOTE1. rfc1122 clearly states that advertised MSS | |
1209 | DOES NOT include either tcp or ip options. | |
1210 | ||
d83d8461 ACM |
1211 | NOTE2. inet_csk(sk)->icsk_pmtu_cookie and tp->mss_cache |
1212 | are READ ONLY outside this function. --ANK (980731) | |
1da177e4 | 1213 | */ |
1da177e4 LT |
1214 | unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu) |
1215 | { | |
1216 | struct tcp_sock *tp = tcp_sk(sk); | |
d83d8461 | 1217 | struct inet_connection_sock *icsk = inet_csk(sk); |
5d424d5a | 1218 | int mss_now; |
1da177e4 | 1219 | |
5d424d5a JH |
1220 | if (icsk->icsk_mtup.search_high > pmtu) |
1221 | icsk->icsk_mtup.search_high = pmtu; | |
1da177e4 | 1222 | |
5d424d5a | 1223 | mss_now = tcp_mtu_to_mss(sk, pmtu); |
409d22b4 | 1224 | mss_now = tcp_bound_to_half_wnd(tp, mss_now); |
1da177e4 LT |
1225 | |
1226 | /* And store cached results */ | |
d83d8461 | 1227 | icsk->icsk_pmtu_cookie = pmtu; |
5d424d5a JH |
1228 | if (icsk->icsk_mtup.enabled) |
1229 | mss_now = min(mss_now, tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low)); | |
c1b4a7e6 | 1230 | tp->mss_cache = mss_now; |
1da177e4 LT |
1231 | |
1232 | return mss_now; | |
1233 | } | |
1234 | ||
1235 | /* Compute the current effective MSS, taking SACKs and IP options, | |
1236 | * and even PMTU discovery events into account. | |
1da177e4 | 1237 | */ |
0c54b85f | 1238 | unsigned int tcp_current_mss(struct sock *sk) |
1da177e4 LT |
1239 | { |
1240 | struct tcp_sock *tp = tcp_sk(sk); | |
1241 | struct dst_entry *dst = __sk_dst_get(sk); | |
c1b4a7e6 | 1242 | u32 mss_now; |
33ad798c AL |
1243 | unsigned header_len; |
1244 | struct tcp_out_options opts; | |
1245 | struct tcp_md5sig_key *md5; | |
c1b4a7e6 DM |
1246 | |
1247 | mss_now = tp->mss_cache; | |
1248 | ||
1da177e4 LT |
1249 | if (dst) { |
1250 | u32 mtu = dst_mtu(dst); | |
d83d8461 | 1251 | if (mtu != inet_csk(sk)->icsk_pmtu_cookie) |
1da177e4 LT |
1252 | mss_now = tcp_sync_mss(sk, mtu); |
1253 | } | |
1254 | ||
33ad798c AL |
1255 | header_len = tcp_established_options(sk, NULL, &opts, &md5) + |
1256 | sizeof(struct tcphdr); | |
1257 | /* The mss_cache is sized based on tp->tcp_header_len, which assumes | |
1258 | * some common options. If this is an odd packet (because we have SACK | |
1259 | * blocks etc) then our calculated header_len will be different, and | |
1260 | * we have to adjust mss_now correspondingly */ | |
1261 | if (header_len != tp->tcp_header_len) { | |
1262 | int delta = (int) header_len - tp->tcp_header_len; | |
1263 | mss_now -= delta; | |
1264 | } | |
cfb6eeb4 | 1265 | |
1da177e4 LT |
1266 | return mss_now; |
1267 | } | |
1268 | ||
a762a980 | 1269 | /* Congestion window validation. (RFC2861) */ |
9e412ba7 | 1270 | static void tcp_cwnd_validate(struct sock *sk) |
a762a980 | 1271 | { |
9e412ba7 | 1272 | struct tcp_sock *tp = tcp_sk(sk); |
a762a980 | 1273 | |
d436d686 | 1274 | if (tp->packets_out >= tp->snd_cwnd) { |
a762a980 DM |
1275 | /* Network is feed fully. */ |
1276 | tp->snd_cwnd_used = 0; | |
1277 | tp->snd_cwnd_stamp = tcp_time_stamp; | |
1278 | } else { | |
1279 | /* Network starves. */ | |
1280 | if (tp->packets_out > tp->snd_cwnd_used) | |
1281 | tp->snd_cwnd_used = tp->packets_out; | |
1282 | ||
15d33c07 DM |
1283 | if (sysctl_tcp_slow_start_after_idle && |
1284 | (s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= inet_csk(sk)->icsk_rto) | |
a762a980 DM |
1285 | tcp_cwnd_application_limited(sk); |
1286 | } | |
1287 | } | |
1288 | ||
0e3a4803 IJ |
1289 | /* Returns the portion of skb which can be sent right away without |
1290 | * introducing MSS oddities to segment boundaries. In rare cases where | |
1291 | * mss_now != mss_cache, we will request caller to create a small skb | |
1292 | * per input skb which could be mostly avoided here (if desired). | |
5ea3a748 IJ |
1293 | * |
1294 | * We explicitly want to create a request for splitting write queue tail | |
1295 | * to a small skb for Nagle purposes while avoiding unnecessary modulos, | |
1296 | * thus all the complexity (cwnd_len is always MSS multiple which we | |
1297 | * return whenever allowed by the other factors). Basically we need the | |
1298 | * modulo only when the receiver window alone is the limiting factor or | |
1299 | * when we would be allowed to send the split-due-to-Nagle skb fully. | |
0e3a4803 IJ |
1300 | */ |
1301 | static unsigned int tcp_mss_split_point(struct sock *sk, struct sk_buff *skb, | |
056834d9 | 1302 | unsigned int mss_now, unsigned int cwnd) |
c1b4a7e6 | 1303 | { |
0e3a4803 IJ |
1304 | struct tcp_sock *tp = tcp_sk(sk); |
1305 | u32 needed, window, cwnd_len; | |
c1b4a7e6 | 1306 | |
90840def | 1307 | window = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; |
c1b4a7e6 | 1308 | cwnd_len = mss_now * cwnd; |
0e3a4803 IJ |
1309 | |
1310 | if (likely(cwnd_len <= window && skb != tcp_write_queue_tail(sk))) | |
1311 | return cwnd_len; | |
1312 | ||
5ea3a748 IJ |
1313 | needed = min(skb->len, window); |
1314 | ||
17515408 | 1315 | if (cwnd_len <= needed) |
0e3a4803 IJ |
1316 | return cwnd_len; |
1317 | ||
0e3a4803 | 1318 | return needed - needed % mss_now; |
c1b4a7e6 DM |
1319 | } |
1320 | ||
1321 | /* Can at least one segment of SKB be sent right now, according to the | |
1322 | * congestion window rules? If so, return how many segments are allowed. | |
1323 | */ | |
056834d9 IJ |
1324 | static inline unsigned int tcp_cwnd_test(struct tcp_sock *tp, |
1325 | struct sk_buff *skb) | |
c1b4a7e6 DM |
1326 | { |
1327 | u32 in_flight, cwnd; | |
1328 | ||
1329 | /* Don't be strict about the congestion window for the final FIN. */ | |
104439a8 JH |
1330 | if ((TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN) && |
1331 | tcp_skb_pcount(skb) == 1) | |
c1b4a7e6 DM |
1332 | return 1; |
1333 | ||
1334 | in_flight = tcp_packets_in_flight(tp); | |
1335 | cwnd = tp->snd_cwnd; | |
1336 | if (in_flight < cwnd) | |
1337 | return (cwnd - in_flight); | |
1338 | ||
1339 | return 0; | |
1340 | } | |
1341 | ||
67edfef7 AK |
1342 | /* Intialize TSO state of a skb. |
1343 | * This must be invoked the first time we consider transmitting | |
c1b4a7e6 DM |
1344 | * SKB onto the wire. |
1345 | */ | |
056834d9 IJ |
1346 | static int tcp_init_tso_segs(struct sock *sk, struct sk_buff *skb, |
1347 | unsigned int mss_now) | |
c1b4a7e6 DM |
1348 | { |
1349 | int tso_segs = tcp_skb_pcount(skb); | |
1350 | ||
f8269a49 | 1351 | if (!tso_segs || (tso_segs > 1 && tcp_skb_mss(skb) != mss_now)) { |
846998ae | 1352 | tcp_set_skb_tso_segs(sk, skb, mss_now); |
c1b4a7e6 DM |
1353 | tso_segs = tcp_skb_pcount(skb); |
1354 | } | |
1355 | return tso_segs; | |
1356 | } | |
1357 | ||
67edfef7 | 1358 | /* Minshall's variant of the Nagle send check. */ |
c1b4a7e6 DM |
1359 | static inline int tcp_minshall_check(const struct tcp_sock *tp) |
1360 | { | |
09cb105e | 1361 | return after(tp->snd_sml, tp->snd_una) && |
c1b4a7e6 DM |
1362 | !after(tp->snd_sml, tp->snd_nxt); |
1363 | } | |
1364 | ||
1365 | /* Return 0, if packet can be sent now without violation Nagle's rules: | |
1366 | * 1. It is full sized. | |
1367 | * 2. Or it contains FIN. (already checked by caller) | |
1368 | * 3. Or TCP_NODELAY was set. | |
1369 | * 4. Or TCP_CORK is not set, and all sent packets are ACKed. | |
1370 | * With Minshall's modification: all sent small packets are ACKed. | |
1371 | */ | |
c1b4a7e6 | 1372 | static inline int tcp_nagle_check(const struct tcp_sock *tp, |
e905a9ed | 1373 | const struct sk_buff *skb, |
c1b4a7e6 DM |
1374 | unsigned mss_now, int nonagle) |
1375 | { | |
1376 | return (skb->len < mss_now && | |
056834d9 IJ |
1377 | ((nonagle & TCP_NAGLE_CORK) || |
1378 | (!nonagle && tp->packets_out && tcp_minshall_check(tp)))); | |
c1b4a7e6 DM |
1379 | } |
1380 | ||
1381 | /* Return non-zero if the Nagle test allows this packet to be | |
1382 | * sent now. | |
1383 | */ | |
1384 | static inline int tcp_nagle_test(struct tcp_sock *tp, struct sk_buff *skb, | |
1385 | unsigned int cur_mss, int nonagle) | |
1386 | { | |
1387 | /* Nagle rule does not apply to frames, which sit in the middle of the | |
1388 | * write_queue (they have no chances to get new data). | |
1389 | * | |
1390 | * This is implemented in the callers, where they modify the 'nonagle' | |
1391 | * argument based upon the location of SKB in the send queue. | |
1392 | */ | |
1393 | if (nonagle & TCP_NAGLE_PUSH) | |
1394 | return 1; | |
1395 | ||
d551e454 IJ |
1396 | /* Don't use the nagle rule for urgent data (or for the final FIN). |
1397 | * Nagle can be ignored during F-RTO too (see RFC4138). | |
1398 | */ | |
33f5f57e | 1399 | if (tcp_urg_mode(tp) || (tp->frto_counter == 2) || |
c1b4a7e6 DM |
1400 | (TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN)) |
1401 | return 1; | |
1402 | ||
1403 | if (!tcp_nagle_check(tp, skb, cur_mss, nonagle)) | |
1404 | return 1; | |
1405 | ||
1406 | return 0; | |
1407 | } | |
1408 | ||
1409 | /* Does at least the first segment of SKB fit into the send window? */ | |
056834d9 IJ |
1410 | static inline int tcp_snd_wnd_test(struct tcp_sock *tp, struct sk_buff *skb, |
1411 | unsigned int cur_mss) | |
c1b4a7e6 DM |
1412 | { |
1413 | u32 end_seq = TCP_SKB_CB(skb)->end_seq; | |
1414 | ||
1415 | if (skb->len > cur_mss) | |
1416 | end_seq = TCP_SKB_CB(skb)->seq + cur_mss; | |
1417 | ||
90840def | 1418 | return !after(end_seq, tcp_wnd_end(tp)); |
c1b4a7e6 DM |
1419 | } |
1420 | ||
fe067e8a | 1421 | /* This checks if the data bearing packet SKB (usually tcp_send_head(sk)) |
c1b4a7e6 DM |
1422 | * should be put on the wire right now. If so, it returns the number of |
1423 | * packets allowed by the congestion window. | |
1424 | */ | |
1425 | static unsigned int tcp_snd_test(struct sock *sk, struct sk_buff *skb, | |
1426 | unsigned int cur_mss, int nonagle) | |
1427 | { | |
1428 | struct tcp_sock *tp = tcp_sk(sk); | |
1429 | unsigned int cwnd_quota; | |
1430 | ||
846998ae | 1431 | tcp_init_tso_segs(sk, skb, cur_mss); |
c1b4a7e6 DM |
1432 | |
1433 | if (!tcp_nagle_test(tp, skb, cur_mss, nonagle)) | |
1434 | return 0; | |
1435 | ||
1436 | cwnd_quota = tcp_cwnd_test(tp, skb); | |
056834d9 | 1437 | if (cwnd_quota && !tcp_snd_wnd_test(tp, skb, cur_mss)) |
c1b4a7e6 DM |
1438 | cwnd_quota = 0; |
1439 | ||
1440 | return cwnd_quota; | |
1441 | } | |
1442 | ||
67edfef7 | 1443 | /* Test if sending is allowed right now. */ |
9e412ba7 | 1444 | int tcp_may_send_now(struct sock *sk) |
c1b4a7e6 | 1445 | { |
9e412ba7 | 1446 | struct tcp_sock *tp = tcp_sk(sk); |
fe067e8a | 1447 | struct sk_buff *skb = tcp_send_head(sk); |
c1b4a7e6 DM |
1448 | |
1449 | return (skb && | |
0c54b85f | 1450 | tcp_snd_test(sk, skb, tcp_current_mss(sk), |
c1b4a7e6 | 1451 | (tcp_skb_is_last(sk, skb) ? |