]>
Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* |
3 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
4 | * operating system. INET is implemented using the BSD Socket | |
5 | * interface as the means of communication with the user level. | |
6 | * | |
7 | * Implementation of the Transmission Control Protocol(TCP). | |
8 | * | |
02c30a84 | 9 | * Authors: Ross Biro |
1da177e4 LT |
10 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
11 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
12 | * Corey Minyard <wf-rch!minyard@relay.EU.net> | |
13 | * Florian La Roche, <flla@stud.uni-sb.de> | |
14 | * Charles Hedrick, <hedrick@klinzhai.rutgers.edu> | |
15 | * Linus Torvalds, <torvalds@cs.helsinki.fi> | |
16 | * Alan Cox, <gw4pts@gw4pts.ampr.org> | |
17 | * Matthew Dillon, <dillon@apollo.west.oic.com> | |
18 | * Arnt Gulbrandsen, <agulbra@nvg.unit.no> | |
19 | * Jorge Cwik, <jorge@laser.satlink.net> | |
20 | * | |
21 | * Fixes: | |
22 | * Alan Cox : Numerous verify_area() calls | |
23 | * Alan Cox : Set the ACK bit on a reset | |
24 | * Alan Cox : Stopped it crashing if it closed while | |
25 | * sk->inuse=1 and was trying to connect | |
26 | * (tcp_err()). | |
27 | * Alan Cox : All icmp error handling was broken | |
28 | * pointers passed where wrong and the | |
29 | * socket was looked up backwards. Nobody | |
30 | * tested any icmp error code obviously. | |
31 | * Alan Cox : tcp_err() now handled properly. It | |
32 | * wakes people on errors. poll | |
33 | * behaves and the icmp error race | |
34 | * has gone by moving it into sock.c | |
35 | * Alan Cox : tcp_send_reset() fixed to work for | |
36 | * everything not just packets for | |
37 | * unknown sockets. | |
38 | * Alan Cox : tcp option processing. | |
39 | * Alan Cox : Reset tweaked (still not 100%) [Had | |
40 | * syn rule wrong] | |
41 | * Herp Rosmanith : More reset fixes | |
42 | * Alan Cox : No longer acks invalid rst frames. | |
43 | * Acking any kind of RST is right out. | |
44 | * Alan Cox : Sets an ignore me flag on an rst | |
45 | * receive otherwise odd bits of prattle | |
46 | * escape still | |
47 | * Alan Cox : Fixed another acking RST frame bug. | |
48 | * Should stop LAN workplace lockups. | |
49 | * Alan Cox : Some tidyups using the new skb list | |
50 | * facilities | |
51 | * Alan Cox : sk->keepopen now seems to work | |
52 | * Alan Cox : Pulls options out correctly on accepts | |
53 | * Alan Cox : Fixed assorted sk->rqueue->next errors | |
54 | * Alan Cox : PSH doesn't end a TCP read. Switched a | |
55 | * bit to skb ops. | |
56 | * Alan Cox : Tidied tcp_data to avoid a potential | |
57 | * nasty. | |
58 | * Alan Cox : Added some better commenting, as the | |
59 | * tcp is hard to follow | |
60 | * Alan Cox : Removed incorrect check for 20 * psh | |
61 | * Michael O'Reilly : ack < copied bug fix. | |
62 | * Johannes Stille : Misc tcp fixes (not all in yet). | |
63 | * Alan Cox : FIN with no memory -> CRASH | |
64 | * Alan Cox : Added socket option proto entries. | |
65 | * Also added awareness of them to accept. | |
66 | * Alan Cox : Added TCP options (SOL_TCP) | |
67 | * Alan Cox : Switched wakeup calls to callbacks, | |
68 | * so the kernel can layer network | |
69 | * sockets. | |
70 | * Alan Cox : Use ip_tos/ip_ttl settings. | |
71 | * Alan Cox : Handle FIN (more) properly (we hope). | |
72 | * Alan Cox : RST frames sent on unsynchronised | |
73 | * state ack error. | |
74 | * Alan Cox : Put in missing check for SYN bit. | |
75 | * Alan Cox : Added tcp_select_window() aka NET2E | |
76 | * window non shrink trick. | |
77 | * Alan Cox : Added a couple of small NET2E timer | |
78 | * fixes | |
79 | * Charles Hedrick : TCP fixes | |
80 | * Toomas Tamm : TCP window fixes | |
81 | * Alan Cox : Small URG fix to rlogin ^C ack fight | |
82 | * Charles Hedrick : Rewrote most of it to actually work | |
83 | * Linus : Rewrote tcp_read() and URG handling | |
84 | * completely | |
85 | * Gerhard Koerting: Fixed some missing timer handling | |
86 | * Matthew Dillon : Reworked TCP machine states as per RFC | |
87 | * Gerhard Koerting: PC/TCP workarounds | |
88 | * Adam Caldwell : Assorted timer/timing errors | |
89 | * Matthew Dillon : Fixed another RST bug | |
90 | * Alan Cox : Move to kernel side addressing changes. | |
91 | * Alan Cox : Beginning work on TCP fastpathing | |
92 | * (not yet usable) | |
93 | * Arnt Gulbrandsen: Turbocharged tcp_check() routine. | |
94 | * Alan Cox : TCP fast path debugging | |
95 | * Alan Cox : Window clamping | |
96 | * Michael Riepe : Bug in tcp_check() | |
97 | * Matt Dillon : More TCP improvements and RST bug fixes | |
98 | * Matt Dillon : Yet more small nasties remove from the | |
99 | * TCP code (Be very nice to this man if | |
100 | * tcp finally works 100%) 8) | |
101 | * Alan Cox : BSD accept semantics. | |
102 | * Alan Cox : Reset on closedown bug. | |
103 | * Peter De Schrijver : ENOTCONN check missing in tcp_sendto(). | |
104 | * Michael Pall : Handle poll() after URG properly in | |
105 | * all cases. | |
106 | * Michael Pall : Undo the last fix in tcp_read_urg() | |
107 | * (multi URG PUSH broke rlogin). | |
108 | * Michael Pall : Fix the multi URG PUSH problem in | |
109 | * tcp_readable(), poll() after URG | |
110 | * works now. | |
111 | * Michael Pall : recv(...,MSG_OOB) never blocks in the | |
112 | * BSD api. | |
113 | * Alan Cox : Changed the semantics of sk->socket to | |
114 | * fix a race and a signal problem with | |
115 | * accept() and async I/O. | |
116 | * Alan Cox : Relaxed the rules on tcp_sendto(). | |
117 | * Yury Shevchuk : Really fixed accept() blocking problem. | |
118 | * Craig I. Hagan : Allow for BSD compatible TIME_WAIT for | |
119 | * clients/servers which listen in on | |
120 | * fixed ports. | |
121 | * Alan Cox : Cleaned the above up and shrank it to | |
122 | * a sensible code size. | |
123 | * Alan Cox : Self connect lockup fix. | |
124 | * Alan Cox : No connect to multicast. | |
125 | * Ross Biro : Close unaccepted children on master | |
126 | * socket close. | |
127 | * Alan Cox : Reset tracing code. | |
128 | * Alan Cox : Spurious resets on shutdown. | |
129 | * Alan Cox : Giant 15 minute/60 second timer error | |
130 | * Alan Cox : Small whoops in polling before an | |
131 | * accept. | |
132 | * Alan Cox : Kept the state trace facility since | |
133 | * it's handy for debugging. | |
134 | * Alan Cox : More reset handler fixes. | |
135 | * Alan Cox : Started rewriting the code based on | |
136 | * the RFC's for other useful protocol | |
137 | * references see: Comer, KA9Q NOS, and | |
138 | * for a reference on the difference | |
139 | * between specifications and how BSD | |
140 | * works see the 4.4lite source. | |
141 | * A.N.Kuznetsov : Don't time wait on completion of tidy | |
142 | * close. | |
143 | * Linus Torvalds : Fin/Shutdown & copied_seq changes. | |
144 | * Linus Torvalds : Fixed BSD port reuse to work first syn | |
145 | * Alan Cox : Reimplemented timers as per the RFC | |
146 | * and using multiple timers for sanity. | |
147 | * Alan Cox : Small bug fixes, and a lot of new | |
148 | * comments. | |
149 | * Alan Cox : Fixed dual reader crash by locking | |
150 | * the buffers (much like datagram.c) | |
151 | * Alan Cox : Fixed stuck sockets in probe. A probe | |
152 | * now gets fed up of retrying without | |
153 | * (even a no space) answer. | |
154 | * Alan Cox : Extracted closing code better | |
155 | * Alan Cox : Fixed the closing state machine to | |
156 | * resemble the RFC. | |
157 | * Alan Cox : More 'per spec' fixes. | |
158 | * Jorge Cwik : Even faster checksumming. | |
159 | * Alan Cox : tcp_data() doesn't ack illegal PSH | |
160 | * only frames. At least one pc tcp stack | |
161 | * generates them. | |
162 | * Alan Cox : Cache last socket. | |
163 | * Alan Cox : Per route irtt. | |
164 | * Matt Day : poll()->select() match BSD precisely on error | |
165 | * Alan Cox : New buffers | |
166 | * Marc Tamsky : Various sk->prot->retransmits and | |
167 | * sk->retransmits misupdating fixed. | |
168 | * Fixed tcp_write_timeout: stuck close, | |
169 | * and TCP syn retries gets used now. | |
170 | * Mark Yarvis : In tcp_read_wakeup(), don't send an | |
171 | * ack if state is TCP_CLOSED. | |
172 | * Alan Cox : Look up device on a retransmit - routes may | |
173 | * change. Doesn't yet cope with MSS shrink right | |
174 | * but it's a start! | |
175 | * Marc Tamsky : Closing in closing fixes. | |
176 | * Mike Shaver : RFC1122 verifications. | |
177 | * Alan Cox : rcv_saddr errors. | |
178 | * Alan Cox : Block double connect(). | |
179 | * Alan Cox : Small hooks for enSKIP. | |
180 | * Alexey Kuznetsov: Path MTU discovery. | |
181 | * Alan Cox : Support soft errors. | |
182 | * Alan Cox : Fix MTU discovery pathological case | |
183 | * when the remote claims no mtu! | |
184 | * Marc Tamsky : TCP_CLOSE fix. | |
185 | * Colin (G3TNE) : Send a reset on syn ack replies in | |
186 | * window but wrong (fixes NT lpd problems) | |
187 | * Pedro Roque : Better TCP window handling, delayed ack. | |
188 | * Joerg Reuter : No modification of locked buffers in | |
189 | * tcp_do_retransmit() | |
190 | * Eric Schenk : Changed receiver side silly window | |
191 | * avoidance algorithm to BSD style | |
192 | * algorithm. This doubles throughput | |
193 | * against machines running Solaris, | |
194 | * and seems to result in general | |
195 | * improvement. | |
196 | * Stefan Magdalinski : adjusted tcp_readable() to fix FIONREAD | |
197 | * Willy Konynenberg : Transparent proxying support. | |
198 | * Mike McLagan : Routing by source | |
199 | * Keith Owens : Do proper merging with partial SKB's in | |
200 | * tcp_do_sendmsg to avoid burstiness. | |
201 | * Eric Schenk : Fix fast close down bug with | |
202 | * shutdown() followed by close(). | |
203 | * Andi Kleen : Make poll agree with SIGIO | |
204 | * Salvatore Sanfilippo : Support SO_LINGER with linger == 1 and | |
205 | * lingertime == 0 (RFC 793 ABORT Call) | |
206 | * Hirokazu Takahashi : Use copy_from_user() instead of | |
207 | * csum_and_copy_from_user() if possible. | |
208 | * | |
1da177e4 LT |
209 | * Description of States: |
210 | * | |
211 | * TCP_SYN_SENT sent a connection request, waiting for ack | |
212 | * | |
213 | * TCP_SYN_RECV received a connection request, sent ack, | |
214 | * waiting for final ack in three-way handshake. | |
215 | * | |
216 | * TCP_ESTABLISHED connection established | |
217 | * | |
218 | * TCP_FIN_WAIT1 our side has shutdown, waiting to complete | |
219 | * transmission of remaining buffered data | |
220 | * | |
221 | * TCP_FIN_WAIT2 all buffered data sent, waiting for remote | |
222 | * to shutdown | |
223 | * | |
224 | * TCP_CLOSING both sides have shutdown but we still have | |
225 | * data we have to finish sending | |
226 | * | |
227 | * TCP_TIME_WAIT timeout to catch resent junk before entering | |
228 | * closed, can only be entered from FIN_WAIT2 | |
229 | * or CLOSING. Required because the other end | |
230 | * may not have gotten our last ACK causing it | |
231 | * to retransmit the data packet (which we ignore) | |
232 | * | |
233 | * TCP_CLOSE_WAIT remote side has shutdown and is waiting for | |
234 | * us to finish writing our data and to shutdown | |
235 | * (we have to close() to move on to LAST_ACK) | |
236 | * | |
237 | * TCP_LAST_ACK out side has shutdown after remote has | |
238 | * shutdown. There may still be data in our | |
239 | * buffer that we have to finish sending | |
240 | * | |
241 | * TCP_CLOSE socket is finished | |
242 | */ | |
243 | ||
afd46503 JP |
244 | #define pr_fmt(fmt) "TCP: " fmt |
245 | ||
cf80e0e4 | 246 | #include <crypto/hash.h> |
172589cc | 247 | #include <linux/kernel.h> |
1da177e4 LT |
248 | #include <linux/module.h> |
249 | #include <linux/types.h> | |
250 | #include <linux/fcntl.h> | |
251 | #include <linux/poll.h> | |
6e9250f5 | 252 | #include <linux/inet_diag.h> |
1da177e4 | 253 | #include <linux/init.h> |
1da177e4 | 254 | #include <linux/fs.h> |
9c55e01c | 255 | #include <linux/skbuff.h> |
81b23b4a | 256 | #include <linux/scatterlist.h> |
9c55e01c JA |
257 | #include <linux/splice.h> |
258 | #include <linux/net.h> | |
259 | #include <linux/socket.h> | |
1da177e4 | 260 | #include <linux/random.h> |
57c8a661 | 261 | #include <linux/memblock.h> |
57413ebc MS |
262 | #include <linux/highmem.h> |
263 | #include <linux/swap.h> | |
b8059ead | 264 | #include <linux/cache.h> |
f4c50d99 | 265 | #include <linux/err.h> |
da5c78c8 | 266 | #include <linux/time.h> |
5a0e3ad6 | 267 | #include <linux/slab.h> |
98aaa913 | 268 | #include <linux/errqueue.h> |
60e2a778 | 269 | #include <linux/static_key.h> |
97a19caf | 270 | #include <linux/btf.h> |
1da177e4 LT |
271 | |
272 | #include <net/icmp.h> | |
cf60af03 | 273 | #include <net/inet_common.h> |
1da177e4 | 274 | #include <net/tcp.h> |
f870fa0b | 275 | #include <net/mptcp.h> |
1da177e4 LT |
276 | #include <net/xfrm.h> |
277 | #include <net/ip.h> | |
9c55e01c | 278 | #include <net/sock.h> |
1da177e4 | 279 | |
7c0f6ba6 | 280 | #include <linux/uaccess.h> |
1da177e4 | 281 | #include <asm/ioctls.h> |
076bb0c8 | 282 | #include <net/busy_poll.h> |
1da177e4 | 283 | |
925bba24 AR |
284 | /* Track pending CMSGs. */ |
285 | enum { | |
286 | TCP_CMSG_INQ = 1, | |
287 | TCP_CMSG_TS = 2 | |
288 | }; | |
289 | ||
e40ec5b7 ED |
290 | DEFINE_PER_CPU(unsigned int, tcp_orphan_count); |
291 | EXPORT_PER_CPU_SYMBOL_GPL(tcp_orphan_count); | |
0a5578cf | 292 | |
a4fe34bf | 293 | long sysctl_tcp_mem[3] __read_mostly; |
a4fe34bf | 294 | EXPORT_SYMBOL(sysctl_tcp_mem); |
1da177e4 | 295 | |
8d987e5c | 296 | atomic_long_t tcp_memory_allocated; /* Current allocated memory. */ |
1da177e4 | 297 | EXPORT_SYMBOL(tcp_memory_allocated); |
1748376b | 298 | |
60e2a778 UB |
299 | #if IS_ENABLED(CONFIG_SMC) |
300 | DEFINE_STATIC_KEY_FALSE(tcp_have_smc); | |
301 | EXPORT_SYMBOL(tcp_have_smc); | |
302 | #endif | |
303 | ||
1748376b ED |
304 | /* |
305 | * Current number of TCP sockets. | |
306 | */ | |
307 | struct percpu_counter tcp_sockets_allocated; | |
1da177e4 LT |
308 | EXPORT_SYMBOL(tcp_sockets_allocated); |
309 | ||
9c55e01c JA |
310 | /* |
311 | * TCP splice context | |
312 | */ | |
313 | struct tcp_splice_state { | |
314 | struct pipe_inode_info *pipe; | |
315 | size_t len; | |
316 | unsigned int flags; | |
317 | }; | |
318 | ||
1da177e4 LT |
319 | /* |
320 | * Pressure flag: try to collapse. | |
321 | * Technical note: it is used by multiple contexts non atomically. | |
3ab224be | 322 | * All the __sk_mem_schedule() is of this nature: accounting |
1da177e4 LT |
323 | * is strict, actions are advisory and have some latency. |
324 | */ | |
06044751 ED |
325 | unsigned long tcp_memory_pressure __read_mostly; |
326 | EXPORT_SYMBOL_GPL(tcp_memory_pressure); | |
1da177e4 | 327 | |
2e05fcae ED |
328 | DEFINE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key); |
329 | EXPORT_SYMBOL(tcp_rx_skb_cache_key); | |
330 | ||
331 | DEFINE_STATIC_KEY_FALSE(tcp_tx_skb_cache_key); | |
332 | ||
5c52ba17 | 333 | void tcp_enter_memory_pressure(struct sock *sk) |
1da177e4 | 334 | { |
06044751 ED |
335 | unsigned long val; |
336 | ||
1f142c17 | 337 | if (READ_ONCE(tcp_memory_pressure)) |
06044751 ED |
338 | return; |
339 | val = jiffies; | |
340 | ||
341 | if (!val) | |
342 | val--; | |
343 | if (!cmpxchg(&tcp_memory_pressure, 0, val)) | |
4e673444 | 344 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES); |
1da177e4 | 345 | } |
06044751 ED |
346 | EXPORT_SYMBOL_GPL(tcp_enter_memory_pressure); |
347 | ||
348 | void tcp_leave_memory_pressure(struct sock *sk) | |
349 | { | |
350 | unsigned long val; | |
351 | ||
1f142c17 | 352 | if (!READ_ONCE(tcp_memory_pressure)) |
06044751 ED |
353 | return; |
354 | val = xchg(&tcp_memory_pressure, 0); | |
355 | if (val) | |
356 | NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURESCHRONO, | |
357 | jiffies_to_msecs(jiffies - val)); | |
358 | } | |
359 | EXPORT_SYMBOL_GPL(tcp_leave_memory_pressure); | |
1da177e4 | 360 | |
b103cf34 JA |
361 | /* Convert seconds to retransmits based on initial and max timeout */ |
362 | static u8 secs_to_retrans(int seconds, int timeout, int rto_max) | |
363 | { | |
364 | u8 res = 0; | |
365 | ||
366 | if (seconds > 0) { | |
367 | int period = timeout; | |
368 | ||
369 | res = 1; | |
370 | while (seconds > period && res < 255) { | |
371 | res++; | |
372 | timeout <<= 1; | |
373 | if (timeout > rto_max) | |
374 | timeout = rto_max; | |
375 | period += timeout; | |
376 | } | |
377 | } | |
378 | return res; | |
379 | } | |
380 | ||
381 | /* Convert retransmits to seconds based on initial and max timeout */ | |
382 | static int retrans_to_secs(u8 retrans, int timeout, int rto_max) | |
383 | { | |
384 | int period = 0; | |
385 | ||
386 | if (retrans > 0) { | |
387 | period = timeout; | |
388 | while (--retrans) { | |
389 | timeout <<= 1; | |
390 | if (timeout > rto_max) | |
391 | timeout = rto_max; | |
392 | period += timeout; | |
393 | } | |
394 | } | |
395 | return period; | |
396 | } | |
397 | ||
0263598c WW |
398 | static u64 tcp_compute_delivery_rate(const struct tcp_sock *tp) |
399 | { | |
400 | u32 rate = READ_ONCE(tp->rate_delivered); | |
401 | u32 intv = READ_ONCE(tp->rate_interval_us); | |
402 | u64 rate64 = 0; | |
403 | ||
404 | if (rate && intv) { | |
405 | rate64 = (u64)rate * tp->mss_cache * USEC_PER_SEC; | |
406 | do_div(rate64, intv); | |
407 | } | |
408 | return rate64; | |
409 | } | |
410 | ||
900f65d3 NC |
411 | /* Address-family independent initialization for a tcp_sock. |
412 | * | |
413 | * NOTE: A lot of things set to zero explicitly by call to | |
414 | * sk_alloc() so need not be done here. | |
415 | */ | |
416 | void tcp_init_sock(struct sock *sk) | |
417 | { | |
418 | struct inet_connection_sock *icsk = inet_csk(sk); | |
419 | struct tcp_sock *tp = tcp_sk(sk); | |
420 | ||
9f5afeae | 421 | tp->out_of_order_queue = RB_ROOT; |
75c119af | 422 | sk->tcp_rtx_queue = RB_ROOT; |
900f65d3 | 423 | tcp_init_xmit_timers(sk); |
46d3ceab | 424 | INIT_LIST_HEAD(&tp->tsq_node); |
e2080072 | 425 | INIT_LIST_HEAD(&tp->tsorted_sent_queue); |
900f65d3 NC |
426 | |
427 | icsk->icsk_rto = TCP_TIMEOUT_INIT; | |
ca584ba0 | 428 | icsk->icsk_rto_min = TCP_RTO_MIN; |
2b8ee4f0 | 429 | icsk->icsk_delack_max = TCP_DELACK_MAX; |
740b0f18 | 430 | tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT); |
ac9517fc | 431 | minmax_reset(&tp->rtt_min, tcp_jiffies32, ~0U); |
900f65d3 NC |
432 | |
433 | /* So many TCP implementations out there (incorrectly) count the | |
434 | * initial SYN frame in their delayed-ACK and congestion control | |
435 | * algorithms that we must have the following bandaid to talk | |
436 | * efficiently to them. -DaveM | |
437 | */ | |
0d24ea1d | 438 | tcp_snd_cwnd_set(tp, TCP_INIT_CWND); |
900f65d3 | 439 | |
d7722e85 SHY |
440 | /* There's a bubble in the pipe until at least the first ACK. */ |
441 | tp->app_limited = ~0U; | |
442 | ||
900f65d3 NC |
443 | /* See draft-stevens-tcpca-spec-01 for discussion of the |
444 | * initialization of these values. | |
445 | */ | |
446 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; | |
447 | tp->snd_cwnd_clamp = ~0; | |
448 | tp->mss_cache = TCP_MSS_DEFAULT; | |
449 | ||
1043e25f | 450 | tp->reordering = sock_net(sk)->ipv4.sysctl_tcp_reordering; |
55d8694f | 451 | tcp_assign_congestion_control(sk); |
900f65d3 | 452 | |
ceaa1fef | 453 | tp->tsoffset = 0; |
1f255691 | 454 | tp->rack.reo_wnd_steps = 1; |
ceaa1fef | 455 | |
900f65d3 NC |
456 | sk->sk_write_space = sk_stream_write_space; |
457 | sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); | |
458 | ||
459 | icsk->icsk_sync_mss = tcp_sync_mss; | |
460 | ||
e292f05e | 461 | WRITE_ONCE(sk->sk_sndbuf, sock_net(sk)->ipv4.sysctl_tcp_wmem[1]); |
ebb3b78d | 462 | WRITE_ONCE(sk->sk_rcvbuf, sock_net(sk)->ipv4.sysctl_tcp_rmem[1]); |
900f65d3 | 463 | |
900f65d3 | 464 | sk_sockets_allocated_inc(sk); |
0a6b2a1d | 465 | sk->sk_route_forced_caps = NETIF_F_GSO; |
900f65d3 NC |
466 | } |
467 | EXPORT_SYMBOL(tcp_init_sock); | |
468 | ||
4e8cc228 | 469 | static void tcp_tx_timestamp(struct sock *sk, u16 tsflags) |
4ed2d765 | 470 | { |
4e8cc228 ED |
471 | struct sk_buff *skb = tcp_write_queue_tail(sk); |
472 | ||
ad02c4f5 | 473 | if (tsflags && skb) { |
f066e2b0 | 474 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
6b084928 | 475 | struct tcp_skb_cb *tcb = TCP_SKB_CB(skb); |
4ed2d765 | 476 | |
c14ac945 | 477 | sock_tx_timestamp(sk, tsflags, &shinfo->tx_flags); |
0a2cf20c SHY |
478 | if (tsflags & SOF_TIMESTAMPING_TX_ACK) |
479 | tcb->txstamp_ack = 1; | |
480 | if (tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) | |
f066e2b0 WB |
481 | shinfo->tskey = TCP_SKB_CB(skb)->seq + skb->len - 1; |
482 | } | |
4ed2d765 WB |
483 | } |
484 | ||
05dc72ab | 485 | static bool tcp_stream_is_readable(struct sock *sk, int target) |
8934ce2f | 486 | { |
05dc72ab ED |
487 | if (tcp_epollin_ready(sk, target)) |
488 | return true; | |
7b50ecfc | 489 | return sk_is_readable(sk); |
8934ce2f JF |
490 | } |
491 | ||
1da177e4 | 492 | /* |
a11e1d43 LT |
493 | * Wait for a TCP event. |
494 | * | |
495 | * Note that we don't need to lock the socket, as the upper poll layers | |
496 | * take care of normal races (between the test and the event) and we don't | |
497 | * go look at any of the socket buffers directly. | |
1da177e4 | 498 | */ |
a11e1d43 | 499 | __poll_t tcp_poll(struct file *file, struct socket *sock, poll_table *wait) |
1da177e4 | 500 | { |
a11e1d43 | 501 | __poll_t mask; |
1da177e4 | 502 | struct sock *sk = sock->sk; |
cf533ea5 | 503 | const struct tcp_sock *tp = tcp_sk(sk); |
00fd38d9 | 504 | int state; |
1da177e4 | 505 | |
89ab066d | 506 | sock_poll_wait(file, sock, wait); |
a11e1d43 | 507 | |
986ffdfd | 508 | state = inet_sk_state_load(sk); |
00fd38d9 | 509 | if (state == TCP_LISTEN) |
dc40c7bc | 510 | return inet_csk_listen_poll(sk); |
1da177e4 | 511 | |
a11e1d43 LT |
512 | /* Socket is not locked. We are protected from async events |
513 | * by poll logic and correct handling of state changes | |
514 | * made by other threads is impossible in any case. | |
515 | */ | |
516 | ||
517 | mask = 0; | |
518 | ||
1da177e4 | 519 | /* |
a9a08845 | 520 | * EPOLLHUP is certainly not done right. But poll() doesn't |
1da177e4 LT |
521 | * have a notion of HUP in just one direction, and for a |
522 | * socket the read side is more interesting. | |
523 | * | |
a9a08845 LT |
524 | * Some poll() documentation says that EPOLLHUP is incompatible |
525 | * with the EPOLLOUT/POLLWR flags, so somebody should check this | |
1da177e4 LT |
526 | * all. But careful, it tends to be safer to return too many |
527 | * bits than too few, and you can easily break real applications | |
528 | * if you don't tell them that something has hung up! | |
529 | * | |
530 | * Check-me. | |
531 | * | |
a9a08845 | 532 | * Check number 1. EPOLLHUP is _UNMASKABLE_ event (see UNIX98 and |
1da177e4 LT |
533 | * our fs/select.c). It means that after we received EOF, |
534 | * poll always returns immediately, making impossible poll() on write() | |
a9a08845 | 535 | * in state CLOSE_WAIT. One solution is evident --- to set EPOLLHUP |
1da177e4 LT |
536 | * if and only if shutdown has been made in both directions. |
537 | * Actually, it is interesting to look how Solaris and DUX | |
a9a08845 | 538 | * solve this dilemma. I would prefer, if EPOLLHUP were maskable, |
1da177e4 LT |
539 | * then we could set it on SND_SHUTDOWN. BTW examples given |
540 | * in Stevens' books assume exactly this behaviour, it explains | |
a9a08845 | 541 | * why EPOLLHUP is incompatible with EPOLLOUT. --ANK |
1da177e4 LT |
542 | * |
543 | * NOTE. Check for TCP_CLOSE is added. The goal is to prevent | |
544 | * blocking on fresh not-connected or disconnected socket. --ANK | |
545 | */ | |
00fd38d9 | 546 | if (sk->sk_shutdown == SHUTDOWN_MASK || state == TCP_CLOSE) |
a9a08845 | 547 | mask |= EPOLLHUP; |
1da177e4 | 548 | if (sk->sk_shutdown & RCV_SHUTDOWN) |
a9a08845 | 549 | mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; |
1da177e4 | 550 | |
8336886f | 551 | /* Connected or passive Fast Open socket? */ |
00fd38d9 | 552 | if (state != TCP_SYN_SENT && |
d983ea6f | 553 | (state != TCP_SYN_RECV || rcu_access_pointer(tp->fastopen_rsk))) { |
c7004482 DM |
554 | int target = sock_rcvlowat(sk, 0, INT_MAX); |
555 | ||
d9b55bf7 | 556 | if (READ_ONCE(tp->urg_seq) == READ_ONCE(tp->copied_seq) && |
c7004482 DM |
557 | !sock_flag(sk, SOCK_URGINLINE) && |
558 | tp->urg_data) | |
b634f875 | 559 | target++; |
c7004482 | 560 | |
05dc72ab | 561 | if (tcp_stream_is_readable(sk, target)) |
a9a08845 | 562 | mask |= EPOLLIN | EPOLLRDNORM; |
1da177e4 LT |
563 | |
564 | if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { | |
8ba3c9d1 | 565 | if (__sk_stream_is_writeable(sk, 1)) { |
a9a08845 | 566 | mask |= EPOLLOUT | EPOLLWRNORM; |
1da177e4 | 567 | } else { /* send SIGIO later */ |
9cd3e072 | 568 | sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 LT |
569 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
570 | ||
571 | /* Race breaker. If space is freed after | |
572 | * wspace test but before the flags are set, | |
3c715127 | 573 | * IO signal will be lost. Memory barrier |
574 | * pairs with the input side. | |
1da177e4 | 575 | */ |
3c715127 | 576 | smp_mb__after_atomic(); |
8ba3c9d1 | 577 | if (__sk_stream_is_writeable(sk, 1)) |
a9a08845 | 578 | mask |= EPOLLOUT | EPOLLWRNORM; |
1da177e4 | 579 | } |
d84ba638 | 580 | } else |
a9a08845 | 581 | mask |= EPOLLOUT | EPOLLWRNORM; |
1da177e4 LT |
582 | |
583 | if (tp->urg_data & TCP_URG_VALID) | |
a9a08845 | 584 | mask |= EPOLLPRI; |
d68be71e | 585 | } else if (state == TCP_SYN_SENT && inet_sk(sk)->defer_connect) { |
19f6d3f3 | 586 | /* Active TCP fastopen socket with defer_connect |
a9a08845 | 587 | * Return EPOLLOUT so application can call write() |
19f6d3f3 WW |
588 | * in order for kernel to generate SYN+data |
589 | */ | |
a9a08845 | 590 | mask |= EPOLLOUT | EPOLLWRNORM; |
1da177e4 | 591 | } |
a4d25803 TM |
592 | /* This barrier is coupled with smp_wmb() in tcp_reset() */ |
593 | smp_rmb(); | |
3ef7cf57 | 594 | if (sk->sk_err || !skb_queue_empty_lockless(&sk->sk_error_queue)) |
a9a08845 | 595 | mask |= EPOLLERR; |
a4d25803 | 596 | |
1da177e4 LT |
597 | return mask; |
598 | } | |
a11e1d43 | 599 | EXPORT_SYMBOL(tcp_poll); |
1da177e4 LT |
600 | |
601 | int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg) | |
602 | { | |
603 | struct tcp_sock *tp = tcp_sk(sk); | |
604 | int answ; | |
0e71c55c | 605 | bool slow; |
1da177e4 LT |
606 | |
607 | switch (cmd) { | |
608 | case SIOCINQ: | |
609 | if (sk->sk_state == TCP_LISTEN) | |
610 | return -EINVAL; | |
611 | ||
0e71c55c | 612 | slow = lock_sock_fast(sk); |
473bd239 | 613 | answ = tcp_inq(sk); |
0e71c55c | 614 | unlock_sock_fast(sk, slow); |
1da177e4 LT |
615 | break; |
616 | case SIOCATMARK: | |
d9b55bf7 ED |
617 | answ = tp->urg_data && |
618 | READ_ONCE(tp->urg_seq) == READ_ONCE(tp->copied_seq); | |
1da177e4 LT |
619 | break; |
620 | case SIOCOUTQ: | |
621 | if (sk->sk_state == TCP_LISTEN) | |
622 | return -EINVAL; | |
623 | ||
624 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
625 | answ = 0; | |
626 | else | |
0f317464 | 627 | answ = READ_ONCE(tp->write_seq) - tp->snd_una; |
1da177e4 | 628 | break; |
2f4e1b39 MS |
629 | case SIOCOUTQNSD: |
630 | if (sk->sk_state == TCP_LISTEN) | |
631 | return -EINVAL; | |
632 | ||
633 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
634 | answ = 0; | |
635 | else | |
e0d694d6 ED |
636 | answ = READ_ONCE(tp->write_seq) - |
637 | READ_ONCE(tp->snd_nxt); | |
2f4e1b39 | 638 | break; |
1da177e4 LT |
639 | default: |
640 | return -ENOIOCTLCMD; | |
3ff50b79 | 641 | } |
1da177e4 LT |
642 | |
643 | return put_user(answ, (int __user *)arg); | |
644 | } | |
4bc2f18b | 645 | EXPORT_SYMBOL(tcp_ioctl); |
1da177e4 | 646 | |
1da177e4 LT |
647 | static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb) |
648 | { | |
4de075e0 | 649 | TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH; |
1da177e4 LT |
650 | tp->pushed_seq = tp->write_seq; |
651 | } | |
652 | ||
a2a385d6 | 653 | static inline bool forced_push(const struct tcp_sock *tp) |
1da177e4 LT |
654 | { |
655 | return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1)); | |
656 | } | |
657 | ||
f4a775d1 | 658 | static void skb_entail(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 659 | { |
9e412ba7 | 660 | struct tcp_sock *tp = tcp_sk(sk); |
352d4800 ACM |
661 | struct tcp_skb_cb *tcb = TCP_SKB_CB(skb); |
662 | ||
663 | skb->csum = 0; | |
664 | tcb->seq = tcb->end_seq = tp->write_seq; | |
4de075e0 | 665 | tcb->tcp_flags = TCPHDR_ACK; |
352d4800 | 666 | tcb->sacked = 0; |
f4a775d1 | 667 | __skb_header_release(skb); |
fe067e8a | 668 | tcp_add_write_queue_tail(sk, skb); |
ab4e846a | 669 | sk_wmem_queued_add(sk, skb->truesize); |
3ab224be | 670 | sk_mem_charge(sk, skb->truesize); |
89ebd197 | 671 | if (tp->nonagle & TCP_NAGLE_PUSH) |
e905a9ed | 672 | tp->nonagle &= ~TCP_NAGLE_PUSH; |
6f021c62 ED |
673 | |
674 | tcp_slow_start_after_idle_check(sk); | |
1da177e4 LT |
675 | } |
676 | ||
afeca340 | 677 | static inline void tcp_mark_urg(struct tcp_sock *tp, int flags) |
1da177e4 | 678 | { |
33f5f57e | 679 | if (flags & MSG_OOB) |
1da177e4 | 680 | tp->snd_up = tp->write_seq; |
1da177e4 LT |
681 | } |
682 | ||
f54b3111 | 683 | /* If a not yet filled skb is pushed, do not send it if |
a181ceb5 | 684 | * we have data packets in Qdisc or NIC queues : |
f54b3111 ED |
685 | * Because TX completion will happen shortly, it gives a chance |
686 | * to coalesce future sendmsg() payload into this skb, without | |
687 | * need for a timer, and with no latency trade off. | |
688 | * As packets containing data payload have a bigger truesize | |
a181ceb5 ED |
689 | * than pure acks (dataless) packets, the last checks prevent |
690 | * autocorking if we only have an ACK in Qdisc/NIC queues, | |
691 | * or if TX completion was delayed after we processed ACK packet. | |
f54b3111 ED |
692 | */ |
693 | static bool tcp_should_autocork(struct sock *sk, struct sk_buff *skb, | |
694 | int size_goal) | |
1da177e4 | 695 | { |
f54b3111 | 696 | return skb->len < size_goal && |
790f00e1 | 697 | sock_net(sk)->ipv4.sysctl_tcp_autocorking && |
114f39fe | 698 | !tcp_rtx_queue_empty(sk) && |
14afee4b | 699 | refcount_read(&sk->sk_wmem_alloc) > skb->truesize; |
f54b3111 ED |
700 | } |
701 | ||
35b2c321 MM |
702 | void tcp_push(struct sock *sk, int flags, int mss_now, |
703 | int nonagle, int size_goal) | |
f54b3111 ED |
704 | { |
705 | struct tcp_sock *tp = tcp_sk(sk); | |
706 | struct sk_buff *skb; | |
afeca340 | 707 | |
f54b3111 | 708 | skb = tcp_write_queue_tail(sk); |
75c119af ED |
709 | if (!skb) |
710 | return; | |
f54b3111 ED |
711 | if (!(flags & MSG_MORE) || forced_push(tp)) |
712 | tcp_mark_push(tp, skb); | |
713 | ||
714 | tcp_mark_urg(tp, flags); | |
715 | ||
716 | if (tcp_should_autocork(sk, skb, size_goal)) { | |
717 | ||
718 | /* avoid atomic op if TSQ_THROTTLED bit is already set */ | |
7aa5470c | 719 | if (!test_bit(TSQ_THROTTLED, &sk->sk_tsq_flags)) { |
f54b3111 | 720 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAUTOCORKING); |
7aa5470c | 721 | set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags); |
f54b3111 | 722 | } |
a181ceb5 ED |
723 | /* It is possible TX completion already happened |
724 | * before we set TSQ_THROTTLED. | |
725 | */ | |
14afee4b | 726 | if (refcount_read(&sk->sk_wmem_alloc) > skb->truesize) |
a181ceb5 | 727 | return; |
1da177e4 | 728 | } |
f54b3111 ED |
729 | |
730 | if (flags & MSG_MORE) | |
731 | nonagle = TCP_NAGLE_CORK; | |
732 | ||
733 | __tcp_push_pending_frames(sk, mss_now, nonagle); | |
1da177e4 LT |
734 | } |
735 | ||
6ff7751d AB |
736 | static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, |
737 | unsigned int offset, size_t len) | |
9c55e01c JA |
738 | { |
739 | struct tcp_splice_state *tss = rd_desc->arg.data; | |
33966dd0 | 740 | int ret; |
9c55e01c | 741 | |
a60e3cc7 | 742 | ret = skb_splice_bits(skb, skb->sk, offset, tss->pipe, |
25869262 | 743 | min(rd_desc->count, len), tss->flags); |
33966dd0 WT |
744 | if (ret > 0) |
745 | rd_desc->count -= ret; | |
746 | return ret; | |
9c55e01c JA |
747 | } |
748 | ||
749 | static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss) | |
750 | { | |
751 | /* Store TCP splice context information in read_descriptor_t. */ | |
752 | read_descriptor_t rd_desc = { | |
753 | .arg.data = tss, | |
33966dd0 | 754 | .count = tss->len, |
9c55e01c JA |
755 | }; |
756 | ||
757 | return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv); | |
758 | } | |
759 | ||
760 | /** | |
761 | * tcp_splice_read - splice data from TCP socket to a pipe | |
762 | * @sock: socket to splice from | |
763 | * @ppos: position (not valid) | |
764 | * @pipe: pipe to splice to | |
765 | * @len: number of bytes to splice | |
766 | * @flags: splice modifier flags | |
767 | * | |
768 | * Description: | |
769 | * Will read pages from given socket and fill them into a pipe. | |
770 | * | |
771 | **/ | |
772 | ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos, | |
773 | struct pipe_inode_info *pipe, size_t len, | |
774 | unsigned int flags) | |
775 | { | |
776 | struct sock *sk = sock->sk; | |
777 | struct tcp_splice_state tss = { | |
778 | .pipe = pipe, | |
779 | .len = len, | |
780 | .flags = flags, | |
781 | }; | |
782 | long timeo; | |
783 | ssize_t spliced; | |
784 | int ret; | |
785 | ||
3a047bf8 | 786 | sock_rps_record_flow(sk); |
9c55e01c JA |
787 | /* |
788 | * We can't seek on a socket input | |
789 | */ | |
790 | if (unlikely(*ppos)) | |
791 | return -ESPIPE; | |
792 | ||
793 | ret = spliced = 0; | |
794 | ||
795 | lock_sock(sk); | |
796 | ||
42324c62 | 797 | timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK); |
9c55e01c JA |
798 | while (tss.len) { |
799 | ret = __tcp_splice_read(sk, &tss); | |
800 | if (ret < 0) | |
801 | break; | |
802 | else if (!ret) { | |
803 | if (spliced) | |
804 | break; | |
9c55e01c JA |
805 | if (sock_flag(sk, SOCK_DONE)) |
806 | break; | |
807 | if (sk->sk_err) { | |
808 | ret = sock_error(sk); | |
809 | break; | |
810 | } | |
811 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
812 | break; | |
813 | if (sk->sk_state == TCP_CLOSE) { | |
814 | /* | |
815 | * This occurs when user tries to read | |
816 | * from never connected socket. | |
817 | */ | |
c47078d6 | 818 | ret = -ENOTCONN; |
9c55e01c JA |
819 | break; |
820 | } | |
821 | if (!timeo) { | |
822 | ret = -EAGAIN; | |
823 | break; | |
824 | } | |
ccf7abb9 ED |
825 | /* if __tcp_splice_read() got nothing while we have |
826 | * an skb in receive queue, we do not want to loop. | |
827 | * This might happen with URG data. | |
828 | */ | |
829 | if (!skb_queue_empty(&sk->sk_receive_queue)) | |
830 | break; | |
dfbafc99 | 831 | sk_wait_data(sk, &timeo, NULL); |
9c55e01c JA |
832 | if (signal_pending(current)) { |
833 | ret = sock_intr_errno(timeo); | |
834 | break; | |
835 | } | |
836 | continue; | |
837 | } | |
838 | tss.len -= ret; | |
839 | spliced += ret; | |
840 | ||
33966dd0 WT |
841 | if (!timeo) |
842 | break; | |
9c55e01c JA |
843 | release_sock(sk); |
844 | lock_sock(sk); | |
845 | ||
846 | if (sk->sk_err || sk->sk_state == TCP_CLOSE || | |
33966dd0 | 847 | (sk->sk_shutdown & RCV_SHUTDOWN) || |
9c55e01c JA |
848 | signal_pending(current)) |
849 | break; | |
850 | } | |
851 | ||
852 | release_sock(sk); | |
853 | ||
854 | if (spliced) | |
855 | return spliced; | |
856 | ||
857 | return ret; | |
858 | } | |
4bc2f18b | 859 | EXPORT_SYMBOL(tcp_splice_read); |
9c55e01c | 860 | |
eb934478 ED |
861 | struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, |
862 | bool force_schedule) | |
f561d0f2 PE |
863 | { |
864 | struct sk_buff *skb; | |
865 | ||
4f661542 ED |
866 | if (likely(!size)) { |
867 | skb = sk->sk_tx_skb_cache; | |
858f5017 | 868 | if (skb) { |
d7cc399e | 869 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
472c2e07 ED |
870 | sk->sk_tx_skb_cache = NULL; |
871 | pskb_trim(skb, 0); | |
872 | INIT_LIST_HEAD(&skb->tcp_tsorted_anchor); | |
eb70a1ae ED |
873 | skb_shinfo(skb)->tx_flags = 0; |
874 | memset(TCP_SKB_CB(skb), 0, sizeof(struct tcp_skb_cb)); | |
472c2e07 ED |
875 | return skb; |
876 | } | |
877 | } | |
f561d0f2 PE |
878 | /* The TCP header must be at least 32-bit aligned. */ |
879 | size = ALIGN(size, 4); | |
880 | ||
8e4d980a ED |
881 | if (unlikely(tcp_under_memory_pressure(sk))) |
882 | sk_mem_reclaim_partial(sk); | |
883 | ||
f561d0f2 | 884 | skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp); |
8e4d980a | 885 | if (likely(skb)) { |
eb934478 | 886 | bool mem_scheduled; |
8e4d980a | 887 | |
eb934478 ED |
888 | if (force_schedule) { |
889 | mem_scheduled = true; | |
8e4d980a ED |
890 | sk_forced_mem_schedule(sk, skb->truesize); |
891 | } else { | |
eb934478 | 892 | mem_scheduled = sk_wmem_schedule(sk, skb->truesize); |
8e4d980a | 893 | } |
eb934478 | 894 | if (likely(mem_scheduled)) { |
a21d4572 | 895 | skb_reserve(skb, sk->sk_prot->max_header); |
f561d0f2 PE |
896 | /* |
897 | * Make sure that we have exactly size bytes | |
898 | * available to the caller, no more, no less. | |
899 | */ | |
16fad69c | 900 | skb->reserved_tailroom = skb->end - skb->tail - size; |
e2080072 | 901 | INIT_LIST_HEAD(&skb->tcp_tsorted_anchor); |
f561d0f2 PE |
902 | return skb; |
903 | } | |
904 | __kfree_skb(skb); | |
905 | } else { | |
5c52ba17 | 906 | sk->sk_prot->enter_memory_pressure(sk); |
f561d0f2 PE |
907 | sk_stream_moderate_sndbuf(sk); |
908 | } | |
909 | return NULL; | |
910 | } | |
911 | ||
0c54b85f IJ |
912 | static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now, |
913 | int large_allowed) | |
914 | { | |
915 | struct tcp_sock *tp = tcp_sk(sk); | |
6c09fa09 | 916 | u32 new_size_goal, size_goal; |
605ad7f1 | 917 | |
74d4a8f8 | 918 | if (!large_allowed) |
605ad7f1 ED |
919 | return mss_now; |
920 | ||
6c09fa09 ED |
921 | /* Note : tcp_tso_autosize() will eventually split this later */ |
922 | new_size_goal = sk->sk_gso_max_size - 1 - MAX_TCP_HEADER; | |
605ad7f1 ED |
923 | new_size_goal = tcp_bound_to_half_wnd(tp, new_size_goal); |
924 | ||
925 | /* We try hard to avoid divides here */ | |
926 | size_goal = tp->gso_segs * mss_now; | |
927 | if (unlikely(new_size_goal < size_goal || | |
928 | new_size_goal >= size_goal + mss_now)) { | |
929 | tp->gso_segs = min_t(u16, new_size_goal / mss_now, | |
930 | sk->sk_gso_max_segs); | |
931 | size_goal = tp->gso_segs * mss_now; | |
0c54b85f IJ |
932 | } |
933 | ||
605ad7f1 | 934 | return max(size_goal, mss_now); |
0c54b85f IJ |
935 | } |
936 | ||
35b2c321 | 937 | int tcp_send_mss(struct sock *sk, int *size_goal, int flags) |
0c54b85f IJ |
938 | { |
939 | int mss_now; | |
940 | ||
941 | mss_now = tcp_current_mss(sk); | |
942 | *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB)); | |
943 | ||
944 | return mss_now; | |
945 | } | |
946 | ||
fdfc5c85 ED |
947 | /* In some cases, both sendpage() and sendmsg() could have added |
948 | * an skb to the write queue, but failed adding payload on it. | |
949 | * We need to remove it to consume less memory, but more | |
950 | * importantly be able to generate EPOLLOUT for Edge Trigger epoll() | |
951 | * users. | |
952 | */ | |
b796d04b | 953 | void tcp_remove_empty_skb(struct sock *sk, struct sk_buff *skb) |
fdfc5c85 | 954 | { |
285fabe6 | 955 | if (skb && TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) { |
fdfc5c85 ED |
956 | tcp_unlink_write_queue(skb, sk); |
957 | if (tcp_write_queue_empty(sk)) | |
958 | tcp_chrono_stop(sk, TCP_CHRONO_BUSY); | |
959 | sk_wmem_free_skb(sk, skb); | |
960 | } | |
961 | } | |
962 | ||
b796d04b PA |
963 | struct sk_buff *tcp_build_frag(struct sock *sk, int size_goal, int flags, |
964 | struct page *page, int offset, size_t *size) | |
965 | { | |
966 | struct sk_buff *skb = tcp_write_queue_tail(sk); | |
967 | struct tcp_sock *tp = tcp_sk(sk); | |
968 | bool can_coalesce; | |
969 | int copy, i; | |
970 | ||
971 | if (!skb || (copy = size_goal - skb->len) <= 0 || | |
972 | !tcp_skb_can_collapse_to(skb)) { | |
973 | new_segment: | |
974 | if (!sk_stream_memory_free(sk)) | |
975 | return NULL; | |
976 | ||
977 | skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation, | |
978 | tcp_rtx_and_write_queues_empty(sk)); | |
979 | if (!skb) | |
980 | return NULL; | |
981 | ||
982 | #ifdef CONFIG_TLS_DEVICE | |
983 | skb->decrypted = !!(flags & MSG_SENDPAGE_DECRYPTED); | |
984 | #endif | |
985 | skb_entail(sk, skb); | |
986 | copy = size_goal; | |
987 | } | |
988 | ||
989 | if (copy > *size) | |
990 | copy = *size; | |
991 | ||
992 | i = skb_shinfo(skb)->nr_frags; | |
993 | can_coalesce = skb_can_coalesce(skb, i, page, offset); | |
994 | if (!can_coalesce && i >= sysctl_max_skb_frags) { | |
995 | tcp_mark_push(tp, skb); | |
996 | goto new_segment; | |
997 | } | |
998 | if (!sk_wmem_schedule(sk, copy)) | |
999 | return NULL; | |
1000 | ||
1001 | if (can_coalesce) { | |
1002 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); | |
1003 | } else { | |
1004 | get_page(page); | |
1005 | skb_fill_page_desc(skb, i, page, offset, copy); | |
1006 | } | |
1007 | ||
1008 | if (!(flags & MSG_NO_SHARED_FRAGS)) | |
06b4feb3 | 1009 | skb_shinfo(skb)->flags |= SKBFL_SHARED_FRAG; |
b796d04b PA |
1010 | |
1011 | skb->len += copy; | |
1012 | skb->data_len += copy; | |
1013 | skb->truesize += copy; | |
1014 | sk_wmem_queued_add(sk, copy); | |
1015 | sk_mem_charge(sk, copy); | |
1016 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1017 | WRITE_ONCE(tp->write_seq, tp->write_seq + copy); | |
1018 | TCP_SKB_CB(skb)->end_seq += copy; | |
1019 | tcp_skb_pcount_set(skb, 0); | |
1020 | ||
1021 | *size = copy; | |
1022 | return skb; | |
1023 | } | |
1024 | ||
e3b5616a DW |
1025 | ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, |
1026 | size_t size, int flags) | |
1da177e4 LT |
1027 | { |
1028 | struct tcp_sock *tp = tcp_sk(sk); | |
c1b4a7e6 | 1029 | int mss_now, size_goal; |
1da177e4 LT |
1030 | int err; |
1031 | ssize_t copied; | |
1032 | long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); | |
1033 | ||
a10674bf | 1034 | if (IS_ENABLED(CONFIG_DEBUG_VM) && |
cf83a17e CL |
1035 | WARN_ONCE(!sendpage_ok(page), |
1036 | "page must not be a Slab one and have page_count > 0")) | |
a10674bf VA |
1037 | return -EINVAL; |
1038 | ||
8336886f JC |
1039 | /* Wait for a connection to finish. One exception is TCP Fast Open |
1040 | * (passive side) where data is allowed to be sent before a connection | |
1041 | * is fully established. | |
1042 | */ | |
1043 | if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) && | |
1044 | !tcp_passive_fastopen(sk)) { | |
686a5624 YM |
1045 | err = sk_stream_wait_connect(sk, &timeo); |
1046 | if (err != 0) | |
1da177e4 | 1047 | goto out_err; |
8336886f | 1048 | } |
1da177e4 | 1049 | |
9cd3e072 | 1050 | sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 | 1051 | |
0c54b85f | 1052 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
1053 | copied = 0; |
1054 | ||
1055 | err = -EPIPE; | |
1056 | if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) | |
0d6a775e | 1057 | goto out_err; |
1da177e4 | 1058 | |
64022d0b | 1059 | while (size > 0) { |
b796d04b PA |
1060 | struct sk_buff *skb; |
1061 | size_t copy = size; | |
1da177e4 | 1062 | |
b796d04b PA |
1063 | skb = tcp_build_frag(sk, size_goal, flags, page, offset, ©); |
1064 | if (!skb) | |
afb83012 | 1065 | goto wait_for_space; |
e905a9ed | 1066 | |
1da177e4 | 1067 | if (!copied) |
4de075e0 | 1068 | TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH; |
1da177e4 LT |
1069 | |
1070 | copied += copy; | |
64022d0b | 1071 | offset += copy; |
686a5624 | 1072 | size -= copy; |
ad02c4f5 | 1073 | if (!size) |
1da177e4 LT |
1074 | goto out; |
1075 | ||
69d15067 | 1076 | if (skb->len < size_goal || (flags & MSG_OOB)) |
1da177e4 LT |
1077 | continue; |
1078 | ||
1079 | if (forced_push(tp)) { | |
1080 | tcp_mark_push(tp, skb); | |
9e412ba7 | 1081 | __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH); |
fe067e8a | 1082 | } else if (skb == tcp_send_head(sk)) |
1da177e4 LT |
1083 | tcp_push_one(sk, mss_now); |
1084 | continue; | |
1085 | ||
afb83012 | 1086 | wait_for_space: |
1da177e4 | 1087 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
f54b3111 ED |
1088 | tcp_push(sk, flags & ~MSG_MORE, mss_now, |
1089 | TCP_NAGLE_PUSH, size_goal); | |
1da177e4 | 1090 | |
686a5624 YM |
1091 | err = sk_stream_wait_memory(sk, &timeo); |
1092 | if (err != 0) | |
1da177e4 LT |
1093 | goto do_error; |
1094 | ||
0c54b85f | 1095 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
1096 | } |
1097 | ||
1098 | out: | |
ad02c4f5 | 1099 | if (copied) { |
4e8cc228 | 1100 | tcp_tx_timestamp(sk, sk->sk_tsflags); |
ad02c4f5 SHY |
1101 | if (!(flags & MSG_SENDPAGE_NOTLAST)) |
1102 | tcp_push(sk, flags, mss_now, tp->nonagle, size_goal); | |
1103 | } | |
1da177e4 LT |
1104 | return copied; |
1105 | ||
1106 | do_error: | |
fdfc5c85 | 1107 | tcp_remove_empty_skb(sk, tcp_write_queue_tail(sk)); |
1da177e4 LT |
1108 | if (copied) |
1109 | goto out; | |
1110 | out_err: | |
ce5ec440 | 1111 | /* make sure we wake any epoll edge trigger waiter */ |
216808c6 | 1112 | if (unlikely(tcp_rtx_and_write_queues_empty(sk) && err == -EAGAIN)) { |
ce5ec440 | 1113 | sk->sk_write_space(sk); |
b0f71bd3 FY |
1114 | tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED); |
1115 | } | |
1da177e4 LT |
1116 | return sk_stream_error(sk, flags, err); |
1117 | } | |
e3b5616a | 1118 | EXPORT_SYMBOL_GPL(do_tcp_sendpages); |
1da177e4 | 1119 | |
306b13eb TH |
1120 | int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, |
1121 | size_t size, int flags) | |
1da177e4 | 1122 | { |
dead7cdb | 1123 | if (!(sk->sk_route_caps & NETIF_F_SG)) |
bd9dfc54 | 1124 | return sock_no_sendpage_locked(sk, page, offset, size, flags); |
1da177e4 | 1125 | |
d7722e85 SHY |
1126 | tcp_rate_check_app_limited(sk); /* is sending application-limited? */ |
1127 | ||
306b13eb TH |
1128 | return do_tcp_sendpages(sk, page, offset, size, flags); |
1129 | } | |
774c4673 | 1130 | EXPORT_SYMBOL_GPL(tcp_sendpage_locked); |
306b13eb TH |
1131 | |
1132 | int tcp_sendpage(struct sock *sk, struct page *page, int offset, | |
1133 | size_t size, int flags) | |
1134 | { | |
1135 | int ret; | |
1136 | ||
1137 | lock_sock(sk); | |
1138 | ret = tcp_sendpage_locked(sk, page, offset, size, flags); | |
1da177e4 | 1139 | release_sock(sk); |
306b13eb TH |
1140 | |
1141 | return ret; | |
1da177e4 | 1142 | } |
4bc2f18b | 1143 | EXPORT_SYMBOL(tcp_sendpage); |
1da177e4 | 1144 | |
cf60af03 YC |
1145 | void tcp_free_fastopen_req(struct tcp_sock *tp) |
1146 | { | |
00db4124 | 1147 | if (tp->fastopen_req) { |
cf60af03 YC |
1148 | kfree(tp->fastopen_req); |
1149 | tp->fastopen_req = NULL; | |
1150 | } | |
1151 | } | |
1152 | ||
f5ddcbbb | 1153 | static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, |
f859a448 WB |
1154 | int *copied, size_t size, |
1155 | struct ubuf_info *uarg) | |
cf60af03 YC |
1156 | { |
1157 | struct tcp_sock *tp = tcp_sk(sk); | |
19f6d3f3 | 1158 | struct inet_sock *inet = inet_sk(sk); |
ba615f67 | 1159 | struct sockaddr *uaddr = msg->msg_name; |
cf60af03 YC |
1160 | int err, flags; |
1161 | ||
e1cfcbe8 | 1162 | if (!(sock_net(sk)->ipv4.sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) || |
ba615f67 WW |
1163 | (uaddr && msg->msg_namelen >= sizeof(uaddr->sa_family) && |
1164 | uaddr->sa_family == AF_UNSPEC)) | |
cf60af03 | 1165 | return -EOPNOTSUPP; |
00db4124 | 1166 | if (tp->fastopen_req) |
cf60af03 YC |
1167 | return -EALREADY; /* Another Fast Open is in progress */ |
1168 | ||
1169 | tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request), | |
1170 | sk->sk_allocation); | |
51456b29 | 1171 | if (unlikely(!tp->fastopen_req)) |
cf60af03 YC |
1172 | return -ENOBUFS; |
1173 | tp->fastopen_req->data = msg; | |
f5ddcbbb | 1174 | tp->fastopen_req->size = size; |
f859a448 | 1175 | tp->fastopen_req->uarg = uarg; |
cf60af03 | 1176 | |
19f6d3f3 WW |
1177 | if (inet->defer_connect) { |
1178 | err = tcp_connect(sk); | |
1179 | /* Same failure procedure as in tcp_v4/6_connect */ | |
1180 | if (err) { | |
1181 | tcp_set_state(sk, TCP_CLOSE); | |
1182 | inet->inet_dport = 0; | |
1183 | sk->sk_route_caps = 0; | |
1184 | } | |
1185 | } | |
cf60af03 | 1186 | flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0; |
ba615f67 | 1187 | err = __inet_stream_connect(sk->sk_socket, uaddr, |
3979ad7e | 1188 | msg->msg_namelen, flags, 1); |
7db92362 WW |
1189 | /* fastopen_req could already be freed in __inet_stream_connect |
1190 | * if the connection times out or gets rst | |
1191 | */ | |
1192 | if (tp->fastopen_req) { | |
1193 | *copied = tp->fastopen_req->copied; | |
1194 | tcp_free_fastopen_req(tp); | |
1195 | inet->defer_connect = 0; | |
1196 | } | |
cf60af03 YC |
1197 | return err; |
1198 | } | |
1199 | ||
306b13eb | 1200 | int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size) |
1da177e4 | 1201 | { |
1da177e4 | 1202 | struct tcp_sock *tp = tcp_sk(sk); |
f214f915 | 1203 | struct ubuf_info *uarg = NULL; |
1da177e4 | 1204 | struct sk_buff *skb; |
c14ac945 | 1205 | struct sockcm_cookie sockc; |
57be5bda AV |
1206 | int flags, err, copied = 0; |
1207 | int mss_now = 0, size_goal, copied_syn = 0; | |
1a991488 | 1208 | int process_backlog = 0; |
74d4a8f8 | 1209 | bool zc = false; |
1da177e4 LT |
1210 | long timeo; |
1211 | ||
1da177e4 | 1212 | flags = msg->msg_flags; |
f214f915 | 1213 | |
5cf4a853 | 1214 | if (flags & MSG_ZEROCOPY && size && sock_flag(sk, SOCK_ZEROCOPY)) { |
75c119af | 1215 | skb = tcp_write_queue_tail(sk); |
8c793822 | 1216 | uarg = msg_zerocopy_realloc(sk, size, skb_zcopy(skb)); |
f214f915 WB |
1217 | if (!uarg) { |
1218 | err = -ENOBUFS; | |
1219 | goto out_err; | |
1220 | } | |
1221 | ||
dead7cdb | 1222 | zc = sk->sk_route_caps & NETIF_F_SG; |
02583ade | 1223 | if (!zc) |
f214f915 WB |
1224 | uarg->zerocopy = 0; |
1225 | } | |
1226 | ||
16ae6aa1 YC |
1227 | if (unlikely(flags & MSG_FASTOPEN || inet_sk(sk)->defer_connect) && |
1228 | !tp->repair) { | |
f859a448 | 1229 | err = tcp_sendmsg_fastopen(sk, msg, &copied_syn, size, uarg); |
cf60af03 YC |
1230 | if (err == -EINPROGRESS && copied_syn > 0) |
1231 | goto out; | |
1232 | else if (err) | |
1233 | goto out_err; | |
cf60af03 YC |
1234 | } |
1235 | ||
1da177e4 LT |
1236 | timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); |
1237 | ||
d7722e85 SHY |
1238 | tcp_rate_check_app_limited(sk); /* is sending application-limited? */ |
1239 | ||
8336886f JC |
1240 | /* Wait for a connection to finish. One exception is TCP Fast Open |
1241 | * (passive side) where data is allowed to be sent before a connection | |
1242 | * is fully established. | |
1243 | */ | |
1244 | if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) && | |
1245 | !tcp_passive_fastopen(sk)) { | |
686a5624 YM |
1246 | err = sk_stream_wait_connect(sk, &timeo); |
1247 | if (err != 0) | |
cf60af03 | 1248 | goto do_error; |
8336886f | 1249 | } |
1da177e4 | 1250 | |
c0e88ff0 PE |
1251 | if (unlikely(tp->repair)) { |
1252 | if (tp->repair_queue == TCP_RECV_QUEUE) { | |
1253 | copied = tcp_send_rcvq(sk, msg, size); | |
5924f17a | 1254 | goto out_nopush; |
c0e88ff0 PE |
1255 | } |
1256 | ||
1257 | err = -EINVAL; | |
1258 | if (tp->repair_queue == TCP_NO_QUEUE) | |
1259 | goto out_err; | |
1260 | ||
1261 | /* 'common' sending to sendq */ | |
1262 | } | |
1263 | ||
657a0667 | 1264 | sockcm_init(&sockc, sk); |
c14ac945 SHY |
1265 | if (msg->msg_controllen) { |
1266 | err = sock_cmsg_send(sk, msg, &sockc); | |
1267 | if (unlikely(err)) { | |
1268 | err = -EINVAL; | |
1269 | goto out_err; | |
1270 | } | |
1271 | } | |
1272 | ||
1da177e4 | 1273 | /* This should be in poll */ |
9cd3e072 | 1274 | sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 | 1275 | |
1da177e4 | 1276 | /* Ok commence sending. */ |
1da177e4 LT |
1277 | copied = 0; |
1278 | ||
d41a69f1 ED |
1279 | restart: |
1280 | mss_now = tcp_send_mss(sk, &size_goal, flags); | |
1281 | ||
1da177e4 LT |
1282 | err = -EPIPE; |
1283 | if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) | |
79d8665b | 1284 | goto do_error; |
1da177e4 | 1285 | |
01e97e65 | 1286 | while (msg_data_left(msg)) { |
57be5bda | 1287 | int copy = 0; |
1da177e4 | 1288 | |
57be5bda | 1289 | skb = tcp_write_queue_tail(sk); |
65ec6097 ED |
1290 | if (skb) |
1291 | copy = size_goal - skb->len; | |
1da177e4 | 1292 | |
c134ecb8 | 1293 | if (copy <= 0 || !tcp_skb_can_collapse_to(skb)) { |
3613b3db ED |
1294 | bool first_skb; |
1295 | ||
1da177e4 | 1296 | new_segment: |
57be5bda | 1297 | if (!sk_stream_memory_free(sk)) |
afb83012 | 1298 | goto wait_for_space; |
1da177e4 | 1299 | |
1a991488 ED |
1300 | if (unlikely(process_backlog >= 16)) { |
1301 | process_backlog = 0; | |
1302 | if (sk_flush_backlog(sk)) | |
1303 | goto restart; | |
d4011239 | 1304 | } |
75c119af | 1305 | first_skb = tcp_rtx_and_write_queues_empty(sk); |
472c2e07 | 1306 | skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation, |
3613b3db | 1307 | first_skb); |
57be5bda | 1308 | if (!skb) |
afb83012 | 1309 | goto wait_for_space; |
1da177e4 | 1310 | |
1a991488 | 1311 | process_backlog++; |
dead7cdb | 1312 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 | 1313 | |
57be5bda AV |
1314 | skb_entail(sk, skb); |
1315 | copy = size_goal; | |
9d186cac | 1316 | |
57be5bda | 1317 | /* All packets are restored as if they have |
d3edd06e | 1318 | * already been sent. skb_mstamp_ns isn't set to |
57be5bda AV |
1319 | * avoid wrong rtt estimation. |
1320 | */ | |
1321 | if (tp->repair) | |
1322 | TCP_SKB_CB(skb)->sacked |= TCPCB_REPAIRED; | |
1323 | } | |
1da177e4 | 1324 | |
57be5bda | 1325 | /* Try to append data to the end of skb. */ |
01e97e65 AV |
1326 | if (copy > msg_data_left(msg)) |
1327 | copy = msg_data_left(msg); | |
57be5bda AV |
1328 | |
1329 | /* Where to copy to? */ | |
02583ade | 1330 | if (skb_availroom(skb) > 0 && !zc) { |
57be5bda AV |
1331 | /* We have some space in skb head. Superb! */ |
1332 | copy = min_t(int, copy, skb_availroom(skb)); | |
1333 | err = skb_add_data_nocache(sk, skb, &msg->msg_iter, copy); | |
1334 | if (err) | |
1335 | goto do_fault; | |
02583ade | 1336 | } else if (!zc) { |
57be5bda AV |
1337 | bool merge = true; |
1338 | int i = skb_shinfo(skb)->nr_frags; | |
1339 | struct page_frag *pfrag = sk_page_frag(sk); | |
1340 | ||
1341 | if (!sk_page_frag_refill(sk, pfrag)) | |
afb83012 | 1342 | goto wait_for_space; |
ef015786 | 1343 | |
57be5bda AV |
1344 | if (!skb_can_coalesce(skb, i, pfrag->page, |
1345 | pfrag->offset)) { | |
74d4a8f8 | 1346 | if (i >= sysctl_max_skb_frags) { |
57be5bda AV |
1347 | tcp_mark_push(tp, skb); |
1348 | goto new_segment; | |
1da177e4 | 1349 | } |
57be5bda | 1350 | merge = false; |
1da177e4 LT |
1351 | } |
1352 | ||
57be5bda AV |
1353 | copy = min_t(int, copy, pfrag->size - pfrag->offset); |
1354 | ||
1355 | if (!sk_wmem_schedule(sk, copy)) | |
afb83012 | 1356 | goto wait_for_space; |
1da177e4 | 1357 | |
57be5bda AV |
1358 | err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb, |
1359 | pfrag->page, | |
1360 | pfrag->offset, | |
1361 | copy); | |
1362 | if (err) | |
1363 | goto do_error; | |
1da177e4 | 1364 | |
57be5bda AV |
1365 | /* Update the skb. */ |
1366 | if (merge) { | |
1367 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); | |
1368 | } else { | |
1369 | skb_fill_page_desc(skb, i, pfrag->page, | |
1370 | pfrag->offset, copy); | |
4e33e346 | 1371 | page_ref_inc(pfrag->page); |
4ed2d765 | 1372 | } |
57be5bda | 1373 | pfrag->offset += copy; |
f214f915 | 1374 | } else { |
358ed624 TA |
1375 | if (!sk_wmem_schedule(sk, copy)) |
1376 | goto wait_for_space; | |
1377 | ||
f214f915 | 1378 | err = skb_zerocopy_iter_stream(sk, skb, msg, copy, uarg); |
111856c7 WB |
1379 | if (err == -EMSGSIZE || err == -EEXIST) { |
1380 | tcp_mark_push(tp, skb); | |
f214f915 | 1381 | goto new_segment; |
111856c7 | 1382 | } |
f214f915 WB |
1383 | if (err < 0) |
1384 | goto do_error; | |
1385 | copy = err; | |
57be5bda AV |
1386 | } |
1387 | ||
1388 | if (!copied) | |
1389 | TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH; | |
1390 | ||
0f317464 | 1391 | WRITE_ONCE(tp->write_seq, tp->write_seq + copy); |
57be5bda AV |
1392 | TCP_SKB_CB(skb)->end_seq += copy; |
1393 | tcp_skb_pcount_set(skb, 0); | |
1da177e4 | 1394 | |
57be5bda | 1395 | copied += copy; |
01e97e65 | 1396 | if (!msg_data_left(msg)) { |
c134ecb8 MKL |
1397 | if (unlikely(flags & MSG_EOR)) |
1398 | TCP_SKB_CB(skb)->eor = 1; | |
57be5bda AV |
1399 | goto out; |
1400 | } | |
1da177e4 | 1401 | |
65ec6097 | 1402 | if (skb->len < size_goal || (flags & MSG_OOB) || unlikely(tp->repair)) |
1da177e4 LT |
1403 | continue; |
1404 | ||
57be5bda AV |
1405 | if (forced_push(tp)) { |
1406 | tcp_mark_push(tp, skb); | |
1407 | __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH); | |
1408 | } else if (skb == tcp_send_head(sk)) | |
1409 | tcp_push_one(sk, mss_now); | |
1410 | continue; | |
1411 | ||
afb83012 | 1412 | wait_for_space: |
57be5bda | 1413 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
57be5bda AV |
1414 | if (copied) |
1415 | tcp_push(sk, flags & ~MSG_MORE, mss_now, | |
1416 | TCP_NAGLE_PUSH, size_goal); | |
1da177e4 | 1417 | |
686a5624 YM |
1418 | err = sk_stream_wait_memory(sk, &timeo); |
1419 | if (err != 0) | |
57be5bda | 1420 | goto do_error; |
1da177e4 | 1421 | |
57be5bda | 1422 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
1423 | } |
1424 | ||
1425 | out: | |
ad02c4f5 | 1426 | if (copied) { |
4e8cc228 | 1427 | tcp_tx_timestamp(sk, sockc.tsflags); |
f54b3111 | 1428 | tcp_push(sk, flags, mss_now, tp->nonagle, size_goal); |
ad02c4f5 | 1429 | } |
5924f17a | 1430 | out_nopush: |
8e044917 | 1431 | net_zcopy_put(uarg); |
cf60af03 | 1432 | return copied + copied_syn; |
1da177e4 | 1433 | |
fdfc5c85 ED |
1434 | do_error: |
1435 | skb = tcp_write_queue_tail(sk); | |
1da177e4 | 1436 | do_fault: |
fdfc5c85 | 1437 | tcp_remove_empty_skb(sk, skb); |
1da177e4 | 1438 | |
cf60af03 | 1439 | if (copied + copied_syn) |
1da177e4 LT |
1440 | goto out; |
1441 | out_err: | |
8e044917 | 1442 | net_zcopy_put_abort(uarg, true); |
1da177e4 | 1443 | err = sk_stream_error(sk, flags, err); |
ce5ec440 | 1444 | /* make sure we wake any epoll edge trigger waiter */ |
216808c6 | 1445 | if (unlikely(tcp_rtx_and_write_queues_empty(sk) && err == -EAGAIN)) { |
ce5ec440 | 1446 | sk->sk_write_space(sk); |
b0f71bd3 FY |
1447 | tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED); |
1448 | } | |
1da177e4 LT |
1449 | return err; |
1450 | } | |
774c4673 | 1451 | EXPORT_SYMBOL_GPL(tcp_sendmsg_locked); |
306b13eb TH |
1452 | |
1453 | int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size) | |
1454 | { | |
1455 | int ret; | |
1456 | ||
1457 | lock_sock(sk); | |
1458 | ret = tcp_sendmsg_locked(sk, msg, size); | |
1459 | release_sock(sk); | |
1460 | ||
1461 | return ret; | |
1462 | } | |
4bc2f18b | 1463 | EXPORT_SYMBOL(tcp_sendmsg); |
1da177e4 LT |
1464 | |
1465 | /* | |
1466 | * Handle reading urgent data. BSD has very simple semantics for | |
1467 | * this, no blocking and very strange errors 8) | |
1468 | */ | |
1469 | ||
377f0a08 | 1470 | static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags) |
1da177e4 LT |
1471 | { |
1472 | struct tcp_sock *tp = tcp_sk(sk); | |
1473 | ||
1474 | /* No URG data to read. */ | |
1475 | if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data || | |
1476 | tp->urg_data == TCP_URG_READ) | |
1477 | return -EINVAL; /* Yes this is right ! */ | |
1478 | ||
1479 | if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE)) | |
1480 | return -ENOTCONN; | |
1481 | ||
1482 | if (tp->urg_data & TCP_URG_VALID) { | |
1483 | int err = 0; | |
1484 | char c = tp->urg_data; | |
1485 | ||
1486 | if (!(flags & MSG_PEEK)) | |
1487 | tp->urg_data = TCP_URG_READ; | |
1488 | ||
1489 | /* Read urgent data. */ | |
1490 | msg->msg_flags |= MSG_OOB; | |
1491 | ||
1492 | if (len > 0) { | |
1493 | if (!(flags & MSG_TRUNC)) | |
7eab8d9e | 1494 | err = memcpy_to_msg(msg, &c, 1); |
1da177e4 LT |
1495 | len = 1; |
1496 | } else | |
1497 | msg->msg_flags |= MSG_TRUNC; | |
1498 | ||
1499 | return err ? -EFAULT : len; | |
1500 | } | |
1501 | ||
1502 | if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN)) | |
1503 | return 0; | |
1504 | ||
1505 | /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and | |
1506 | * the available implementations agree in this case: | |
1507 | * this call should never block, independent of the | |
1508 | * blocking state of the socket. | |
1509 | * Mike <pall@rz.uni-karlsruhe.de> | |
1510 | */ | |
1511 | return -EAGAIN; | |
1512 | } | |
1513 | ||
c0e88ff0 PE |
1514 | static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len) |
1515 | { | |
1516 | struct sk_buff *skb; | |
1517 | int copied = 0, err = 0; | |
1518 | ||
1519 | /* XXX -- need to support SO_PEEK_OFF */ | |
1520 | ||
75c119af ED |
1521 | skb_rbtree_walk(skb, &sk->tcp_rtx_queue) { |
1522 | err = skb_copy_datagram_msg(skb, 0, msg, skb->len); | |
1523 | if (err) | |
1524 | return err; | |
1525 | copied += skb->len; | |
1526 | } | |
1527 | ||
c0e88ff0 | 1528 | skb_queue_walk(&sk->sk_write_queue, skb) { |
51f3d02b | 1529 | err = skb_copy_datagram_msg(skb, 0, msg, skb->len); |
c0e88ff0 PE |
1530 | if (err) |
1531 | break; | |
1532 | ||
1533 | copied += skb->len; | |
1534 | } | |
1535 | ||
1536 | return err ?: copied; | |
1537 | } | |
1538 | ||
1da177e4 LT |
1539 | /* Clean up the receive buffer for full frames taken by the user, |
1540 | * then send an ACK if necessary. COPIED is the number of bytes | |
1541 | * tcp_recvmsg has given to the user so far, it speeds up the | |
1542 | * calculation of whether or not we must ACK for the sake of | |
1543 | * a window update. | |
1544 | */ | |
c76c6956 | 1545 | void tcp_cleanup_rbuf(struct sock *sk, int copied) |
1da177e4 LT |
1546 | { |
1547 | struct tcp_sock *tp = tcp_sk(sk); | |
a2a385d6 | 1548 | bool time_to_ack = false; |
1da177e4 | 1549 | |
1da177e4 LT |
1550 | struct sk_buff *skb = skb_peek(&sk->sk_receive_queue); |
1551 | ||
d792c100 | 1552 | WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq), |
2af6fd8b | 1553 | "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n", |
d792c100 | 1554 | tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt); |
1da177e4 | 1555 | |
463c84b9 ACM |
1556 | if (inet_csk_ack_scheduled(sk)) { |
1557 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
b6b6d653 ED |
1558 | |
1559 | if (/* Once-per-two-segments ACK was not sent by tcp_input.c */ | |
463c84b9 | 1560 | tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss || |
1da177e4 LT |
1561 | /* |
1562 | * If this read emptied read buffer, we send ACK, if | |
1563 | * connection is not bidirectional, user drained | |
1564 | * receive buffer and there was a small segment | |
1565 | * in queue. | |
1566 | */ | |
1ef9696c AK |
1567 | (copied > 0 && |
1568 | ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) || | |
1569 | ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) && | |
31954cd8 | 1570 | !inet_csk_in_pingpong_mode(sk))) && |
1ef9696c | 1571 | !atomic_read(&sk->sk_rmem_alloc))) |
a2a385d6 | 1572 | time_to_ack = true; |
1da177e4 LT |
1573 | } |
1574 | ||
1575 | /* We send an ACK if we can now advertise a non-zero window | |
1576 | * which has been raised "significantly". | |
1577 | * | |
1578 | * Even if window raised up to infinity, do not send window open ACK | |
1579 | * in states, where we will not receive more. It is useless. | |
1580 | */ | |
1581 | if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) { | |
1582 | __u32 rcv_window_now = tcp_receive_window(tp); | |
1583 | ||
1584 | /* Optimize, __tcp_select_window() is not cheap. */ | |
1585 | if (2*rcv_window_now <= tp->window_clamp) { | |
1586 | __u32 new_window = __tcp_select_window(sk); | |
1587 | ||
1588 | /* Send ACK now, if this read freed lots of space | |
1589 | * in our buffer. Certainly, new_window is new window. | |
1590 | * We can advertise it now, if it is not less than current one. | |
1591 | * "Lots" means "at least twice" here. | |
1592 | */ | |
1593 | if (new_window && new_window >= 2 * rcv_window_now) | |
a2a385d6 | 1594 | time_to_ack = true; |
1da177e4 LT |
1595 | } |
1596 | } | |
1597 | if (time_to_ack) | |
1598 | tcp_send_ack(sk); | |
1599 | } | |
1600 | ||
f26845b4 | 1601 | static struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off) |
1da177e4 LT |
1602 | { |
1603 | struct sk_buff *skb; | |
1604 | u32 offset; | |
1605 | ||
f26845b4 | 1606 | while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) { |
1da177e4 | 1607 | offset = seq - TCP_SKB_CB(skb)->seq; |
9d691539 ED |
1608 | if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { |
1609 | pr_err_once("%s: found a SYN, please report !\n", __func__); | |
1da177e4 | 1610 | offset--; |
9d691539 | 1611 | } |
e11ecddf | 1612 | if (offset < skb->len || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) { |
1da177e4 LT |
1613 | *off = offset; |
1614 | return skb; | |
1615 | } | |
f26845b4 ED |
1616 | /* This looks weird, but this can happen if TCP collapsing |
1617 | * splitted a fat GRO packet, while we released socket lock | |
1618 | * in skb_splice_bits() | |
1619 | */ | |
7bced397 | 1620 | sk_eat_skb(sk, skb); |
1da177e4 LT |
1621 | } |
1622 | return NULL; | |
1623 | } | |
1624 | ||
1625 | /* | |
1626 | * This routine provides an alternative to tcp_recvmsg() for routines | |
1627 | * that would like to handle copying from skbuffs directly in 'sendfile' | |
1628 | * fashion. | |
1629 | * Note: | |
1630 | * - It is assumed that the socket was locked by the caller. | |
1631 | * - The routine does not block. | |
1632 | * - At present, there is no support for reading OOB data | |
1633 | * or for 'peeking' the socket using this routine | |
1634 | * (although both would be easy to implement). | |
1635 | */ | |
1636 | int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, | |
1637 | sk_read_actor_t recv_actor) | |
1638 | { | |
1639 | struct sk_buff *skb; | |
1640 | struct tcp_sock *tp = tcp_sk(sk); | |
1641 | u32 seq = tp->copied_seq; | |
1642 | u32 offset; | |
1643 | int copied = 0; | |
1644 | ||
1645 | if (sk->sk_state == TCP_LISTEN) | |
1646 | return -ENOTCONN; | |
1647 | while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) { | |
1648 | if (offset < skb->len) { | |
374e7b59 OP |
1649 | int used; |
1650 | size_t len; | |
1da177e4 LT |
1651 | |
1652 | len = skb->len - offset; | |
1653 | /* Stop reading if we hit a patch of urgent data */ | |
1654 | if (tp->urg_data) { | |
1655 | u32 urg_offset = tp->urg_seq - seq; | |
1656 | if (urg_offset < len) | |
1657 | len = urg_offset; | |
1658 | if (!len) | |
1659 | break; | |
1660 | } | |
1661 | used = recv_actor(desc, skb, offset, len); | |
ff905b1e | 1662 | if (used <= 0) { |
ddb61a57 JA |
1663 | if (!copied) |
1664 | copied = used; | |
1665 | break; | |
1da177e4 | 1666 | } |
1d53ad7c ED |
1667 | if (WARN_ON_ONCE(used > len)) |
1668 | used = len; | |
1669 | seq += used; | |
1670 | copied += used; | |
1671 | offset += used; | |
1672 | ||
02275a2e | 1673 | /* If recv_actor drops the lock (e.g. TCP splice |
293ad604 OP |
1674 | * receive) the skb pointer might be invalid when |
1675 | * getting here: tcp_collapse might have deleted it | |
1676 | * while aggregating skbs from the socket queue. | |
1677 | */ | |
02275a2e WT |
1678 | skb = tcp_recv_skb(sk, seq - 1, &offset); |
1679 | if (!skb) | |
1da177e4 | 1680 | break; |
02275a2e WT |
1681 | /* TCP coalescing might have appended data to the skb. |
1682 | * Try to splice more frags | |
1683 | */ | |
1684 | if (offset + 1 != skb->len) | |
1685 | continue; | |
1da177e4 | 1686 | } |
e11ecddf | 1687 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) { |
7bced397 | 1688 | sk_eat_skb(sk, skb); |
1da177e4 LT |
1689 | ++seq; |
1690 | break; | |
1691 | } | |
7bced397 | 1692 | sk_eat_skb(sk, skb); |
1da177e4 LT |
1693 | if (!desc->count) |
1694 | break; | |
7db48e98 | 1695 | WRITE_ONCE(tp->copied_seq, seq); |
1da177e4 | 1696 | } |
7db48e98 | 1697 | WRITE_ONCE(tp->copied_seq, seq); |
1da177e4 LT |
1698 | |
1699 | tcp_rcv_space_adjust(sk); | |
1700 | ||
1701 | /* Clean up data we have read: This will do ACK frames. */ | |
f26845b4 ED |
1702 | if (copied > 0) { |
1703 | tcp_recv_skb(sk, seq, &offset); | |
0e4b4992 | 1704 | tcp_cleanup_rbuf(sk, copied); |
f26845b4 | 1705 | } |
1da177e4 LT |
1706 | return copied; |
1707 | } | |
4bc2f18b | 1708 | EXPORT_SYMBOL(tcp_read_sock); |
1da177e4 | 1709 | |
32035585 TH |
1710 | int tcp_peek_len(struct socket *sock) |
1711 | { | |
1712 | return tcp_inq(sock->sk); | |
1713 | } | |
1714 | EXPORT_SYMBOL(tcp_peek_len); | |
1715 | ||
d1361840 ED |
1716 | /* Make sure sk_rcvbuf is big enough to satisfy SO_RCVLOWAT hint */ |
1717 | int tcp_set_rcvlowat(struct sock *sk, int val) | |
1718 | { | |
867f816b SHY |
1719 | int cap; |
1720 | ||
1721 | if (sk->sk_userlocks & SOCK_RCVBUF_LOCK) | |
1722 | cap = sk->sk_rcvbuf >> 1; | |
1723 | else | |
1724 | cap = sock_net(sk)->ipv4.sysctl_tcp_rmem[2] >> 1; | |
1725 | val = min(val, cap); | |
eac66402 | 1726 | WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); |
03f45c88 ED |
1727 | |
1728 | /* Check if we need to signal EPOLLIN right now */ | |
1729 | tcp_data_ready(sk); | |
1730 | ||
d1361840 ED |
1731 | if (sk->sk_userlocks & SOCK_RCVBUF_LOCK) |
1732 | return 0; | |
1733 | ||
d1361840 | 1734 | val <<= 1; |
d1361840 | 1735 | if (val > sk->sk_rcvbuf) { |
ebb3b78d | 1736 | WRITE_ONCE(sk->sk_rcvbuf, val); |
d1361840 ED |
1737 | tcp_sk(sk)->window_clamp = tcp_win_from_space(sk, val); |
1738 | } | |
1739 | return 0; | |
1740 | } | |
1741 | EXPORT_SYMBOL(tcp_set_rcvlowat); | |
1742 | ||
892bfd3d FW |
1743 | void tcp_update_recv_tstamps(struct sk_buff *skb, |
1744 | struct scm_timestamping_internal *tss) | |
7eeba170 AR |
1745 | { |
1746 | if (skb->tstamp) | |
1747 | tss->ts[0] = ktime_to_timespec64(skb->tstamp); | |
1748 | else | |
1749 | tss->ts[0] = (struct timespec64) {0}; | |
1750 | ||
1751 | if (skb_hwtstamps(skb)->hwtstamp) | |
1752 | tss->ts[2] = ktime_to_timespec64(skb_hwtstamps(skb)->hwtstamp); | |
1753 | else | |
1754 | tss->ts[2] = (struct timespec64) {0}; | |
1755 | } | |
1756 | ||
05255b82 ED |
1757 | #ifdef CONFIG_MMU |
1758 | static const struct vm_operations_struct tcp_vm_ops = { | |
1759 | }; | |
1760 | ||
93ab6cc6 ED |
1761 | int tcp_mmap(struct file *file, struct socket *sock, |
1762 | struct vm_area_struct *vma) | |
1763 | { | |
05255b82 ED |
1764 | if (vma->vm_flags & (VM_WRITE | VM_EXEC)) |
1765 | return -EPERM; | |
1766 | vma->vm_flags &= ~(VM_MAYWRITE | VM_MAYEXEC); | |
1767 | ||
3e4e28c5 | 1768 | /* Instruct vm_insert_page() to not mmap_read_lock(mm) */ |
05255b82 ED |
1769 | vma->vm_flags |= VM_MIXEDMAP; |
1770 | ||
1771 | vma->vm_ops = &tcp_vm_ops; | |
1772 | return 0; | |
1773 | } | |
1774 | EXPORT_SYMBOL(tcp_mmap); | |
1775 | ||
7fba5309 AR |
1776 | static skb_frag_t *skb_advance_to_frag(struct sk_buff *skb, u32 offset_skb, |
1777 | u32 *offset_frag) | |
1778 | { | |
1779 | skb_frag_t *frag; | |
1780 | ||
ab73b188 AR |
1781 | if (unlikely(offset_skb >= skb->len)) |
1782 | return NULL; | |
1783 | ||
7fba5309 AR |
1784 | offset_skb -= skb_headlen(skb); |
1785 | if ((int)offset_skb < 0 || skb_has_frag_list(skb)) | |
1786 | return NULL; | |
1787 | ||
1788 | frag = skb_shinfo(skb)->frags; | |
1789 | while (offset_skb) { | |
1790 | if (skb_frag_size(frag) > offset_skb) { | |
1791 | *offset_frag = offset_skb; | |
1792 | return frag; | |
1793 | } | |
1794 | offset_skb -= skb_frag_size(frag); | |
1795 | ++frag; | |
1796 | } | |
1797 | *offset_frag = 0; | |
1798 | return frag; | |
1799 | } | |
1800 | ||
98917cf0 AR |
1801 | static bool can_map_frag(const skb_frag_t *frag) |
1802 | { | |
1803 | return skb_frag_size(frag) == PAGE_SIZE && !skb_frag_off(frag); | |
1804 | } | |
1805 | ||
1806 | static int find_next_mappable_frag(const skb_frag_t *frag, | |
1807 | int remaining_in_skb) | |
1808 | { | |
1809 | int offset = 0; | |
1810 | ||
1811 | if (likely(can_map_frag(frag))) | |
1812 | return 0; | |
1813 | ||
1814 | while (offset < remaining_in_skb && !can_map_frag(frag)) { | |
1815 | offset += skb_frag_size(frag); | |
1816 | ++frag; | |
1817 | } | |
1818 | return offset; | |
1819 | } | |
1820 | ||
0c3936d3 AR |
1821 | static void tcp_zerocopy_set_hint_for_skb(struct sock *sk, |
1822 | struct tcp_zerocopy_receive *zc, | |
1823 | struct sk_buff *skb, u32 offset) | |
1824 | { | |
1825 | u32 frag_offset, partial_frag_remainder = 0; | |
1826 | int mappable_offset; | |
1827 | skb_frag_t *frag; | |
1828 | ||
1829 | /* worst case: skip to next skb. try to improve on this case below */ | |
1830 | zc->recv_skip_hint = skb->len - offset; | |
1831 | ||
1832 | /* Find the frag containing this offset (and how far into that frag) */ | |
1833 | frag = skb_advance_to_frag(skb, offset, &frag_offset); | |
1834 | if (!frag) | |
1835 | return; | |
1836 | ||
1837 | if (frag_offset) { | |
1838 | struct skb_shared_info *info = skb_shinfo(skb); | |
1839 | ||
1840 | /* We read part of the last frag, must recvmsg() rest of skb. */ | |
1841 | if (frag == &info->frags[info->nr_frags - 1]) | |
1842 | return; | |
1843 | ||
1844 | /* Else, we must at least read the remainder in this frag. */ | |
1845 | partial_frag_remainder = skb_frag_size(frag) - frag_offset; | |
1846 | zc->recv_skip_hint -= partial_frag_remainder; | |
1847 | ++frag; | |
1848 | } | |
1849 | ||
1850 | /* partial_frag_remainder: If part way through a frag, must read rest. | |
1851 | * mappable_offset: Bytes till next mappable frag, *not* counting bytes | |
1852 | * in partial_frag_remainder. | |
1853 | */ | |
1854 | mappable_offset = find_next_mappable_frag(frag, zc->recv_skip_hint); | |
1855 | zc->recv_skip_hint = mappable_offset + partial_frag_remainder; | |
1856 | } | |
1857 | ||
f21a3c48 AR |
1858 | static int tcp_recvmsg_locked(struct sock *sk, struct msghdr *msg, size_t len, |
1859 | int nonblock, int flags, | |
1860 | struct scm_timestamping_internal *tss, | |
1861 | int *cmsg_flags); | |
1862 | static int receive_fallback_to_copy(struct sock *sk, | |
7eeba170 AR |
1863 | struct tcp_zerocopy_receive *zc, int inq, |
1864 | struct scm_timestamping_internal *tss) | |
f21a3c48 AR |
1865 | { |
1866 | unsigned long copy_address = (unsigned long)zc->copybuf_address; | |
f21a3c48 AR |
1867 | struct msghdr msg = {}; |
1868 | struct iovec iov; | |
7eeba170 | 1869 | int err; |
f21a3c48 AR |
1870 | |
1871 | zc->length = 0; | |
1872 | zc->recv_skip_hint = 0; | |
1873 | ||
1874 | if (copy_address != zc->copybuf_address) | |
1875 | return -EINVAL; | |
1876 | ||
1877 | err = import_single_range(READ, (void __user *)copy_address, | |
1878 | inq, &iov, &msg.msg_iter); | |
1879 | if (err) | |
1880 | return err; | |
1881 | ||
1882 | err = tcp_recvmsg_locked(sk, &msg, inq, /*nonblock=*/1, /*flags=*/0, | |
7eeba170 | 1883 | tss, &zc->msg_flags); |
f21a3c48 AR |
1884 | if (err < 0) |
1885 | return err; | |
1886 | ||
1887 | zc->copybuf_len = err; | |
0c3936d3 AR |
1888 | if (likely(zc->copybuf_len)) { |
1889 | struct sk_buff *skb; | |
1890 | u32 offset; | |
1891 | ||
1892 | skb = tcp_recv_skb(sk, tcp_sk(sk)->copied_seq, &offset); | |
1893 | if (skb) | |
1894 | tcp_zerocopy_set_hint_for_skb(sk, zc, skb, offset); | |
1895 | } | |
f21a3c48 AR |
1896 | return 0; |
1897 | } | |
1898 | ||
18fb76ed AR |
1899 | static int tcp_copy_straggler_data(struct tcp_zerocopy_receive *zc, |
1900 | struct sk_buff *skb, u32 copylen, | |
1901 | u32 *offset, u32 *seq) | |
1902 | { | |
1903 | unsigned long copy_address = (unsigned long)zc->copybuf_address; | |
1904 | struct msghdr msg = {}; | |
1905 | struct iovec iov; | |
1906 | int err; | |
1907 | ||
1908 | if (copy_address != zc->copybuf_address) | |
1909 | return -EINVAL; | |
1910 | ||
1911 | err = import_single_range(READ, (void __user *)copy_address, | |
1912 | copylen, &iov, &msg.msg_iter); | |
1913 | if (err) | |
1914 | return err; | |
1915 | err = skb_copy_datagram_msg(skb, *offset, &msg, copylen); | |
1916 | if (err) | |
1917 | return err; | |
1918 | zc->recv_skip_hint -= copylen; | |
1919 | *offset += copylen; | |
1920 | *seq += copylen; | |
1921 | return (__s32)copylen; | |
1922 | } | |
1923 | ||
7eeba170 AR |
1924 | static int tcp_zc_handle_leftover(struct tcp_zerocopy_receive *zc, |
1925 | struct sock *sk, | |
1926 | struct sk_buff *skb, | |
1927 | u32 *seq, | |
1928 | s32 copybuf_len, | |
1929 | struct scm_timestamping_internal *tss) | |
18fb76ed AR |
1930 | { |
1931 | u32 offset, copylen = min_t(u32, copybuf_len, zc->recv_skip_hint); | |
1932 | ||
1933 | if (!copylen) | |
1934 | return 0; | |
1935 | /* skb is null if inq < PAGE_SIZE. */ | |
7eeba170 | 1936 | if (skb) { |
18fb76ed | 1937 | offset = *seq - TCP_SKB_CB(skb)->seq; |
7eeba170 | 1938 | } else { |
18fb76ed | 1939 | skb = tcp_recv_skb(sk, *seq, &offset); |
7eeba170 AR |
1940 | if (TCP_SKB_CB(skb)->has_rxtstamp) { |
1941 | tcp_update_recv_tstamps(skb, tss); | |
1942 | zc->msg_flags |= TCP_CMSG_TS; | |
1943 | } | |
1944 | } | |
18fb76ed AR |
1945 | |
1946 | zc->copybuf_len = tcp_copy_straggler_data(zc, skb, copylen, &offset, | |
1947 | seq); | |
1948 | return zc->copybuf_len < 0 ? 0 : copylen; | |
1949 | } | |
1950 | ||
94ab9eb9 AR |
1951 | static int tcp_zerocopy_vm_insert_batch_error(struct vm_area_struct *vma, |
1952 | struct page **pending_pages, | |
1953 | unsigned long pages_remaining, | |
1954 | unsigned long *address, | |
1955 | u32 *length, | |
1956 | u32 *seq, | |
1957 | struct tcp_zerocopy_receive *zc, | |
1958 | u32 total_bytes_to_map, | |
1959 | int err) | |
1960 | { | |
1961 | /* At least one page did not map. Try zapping if we skipped earlier. */ | |
1962 | if (err == -EBUSY && | |
1963 | zc->flags & TCP_RECEIVE_ZEROCOPY_FLAG_TLB_CLEAN_HINT) { | |
1964 | u32 maybe_zap_len; | |
1965 | ||
1966 | maybe_zap_len = total_bytes_to_map - /* All bytes to map */ | |
1967 | *length + /* Mapped or pending */ | |
1968 | (pages_remaining * PAGE_SIZE); /* Failed map. */ | |
1969 | zap_page_range(vma, *address, maybe_zap_len); | |
1970 | err = 0; | |
1971 | } | |
1972 | ||
1973 | if (!err) { | |
1974 | unsigned long leftover_pages = pages_remaining; | |
1975 | int bytes_mapped; | |
1976 | ||
1977 | /* We called zap_page_range, try to reinsert. */ | |
1978 | err = vm_insert_pages(vma, *address, | |
1979 | pending_pages, | |
1980 | &pages_remaining); | |
1981 | bytes_mapped = PAGE_SIZE * (leftover_pages - pages_remaining); | |
1982 | *seq += bytes_mapped; | |
1983 | *address += bytes_mapped; | |
1984 | } | |
1985 | if (err) { | |
1986 | /* Either we were unable to zap, OR we zapped, retried an | |
1987 | * insert, and still had an issue. Either ways, pages_remaining | |
1988 | * is the number of pages we were unable to map, and we unroll | |
1989 | * some state we speculatively touched before. | |
1990 | */ | |
1991 | const int bytes_not_mapped = PAGE_SIZE * pages_remaining; | |
1992 | ||
1993 | *length -= bytes_not_mapped; | |
1994 | zc->recv_skip_hint += bytes_not_mapped; | |
1995 | } | |
1996 | return err; | |
1997 | } | |
1998 | ||
3763a24c AR |
1999 | static int tcp_zerocopy_vm_insert_batch(struct vm_area_struct *vma, |
2000 | struct page **pages, | |
94ab9eb9 AR |
2001 | unsigned int pages_to_map, |
2002 | unsigned long *address, | |
2003 | u32 *length, | |
3763a24c | 2004 | u32 *seq, |
94ab9eb9 AR |
2005 | struct tcp_zerocopy_receive *zc, |
2006 | u32 total_bytes_to_map) | |
3763a24c AR |
2007 | { |
2008 | unsigned long pages_remaining = pages_to_map; | |
94ab9eb9 AR |
2009 | unsigned int pages_mapped; |
2010 | unsigned int bytes_mapped; | |
2011 | int err; | |
3763a24c | 2012 | |
94ab9eb9 AR |
2013 | err = vm_insert_pages(vma, *address, pages, &pages_remaining); |
2014 | pages_mapped = pages_to_map - (unsigned int)pages_remaining; | |
2015 | bytes_mapped = PAGE_SIZE * pages_mapped; | |
3763a24c AR |
2016 | /* Even if vm_insert_pages fails, it may have partially succeeded in |
2017 | * mapping (some but not all of the pages). | |
2018 | */ | |
2019 | *seq += bytes_mapped; | |
94ab9eb9 AR |
2020 | *address += bytes_mapped; |
2021 | ||
2022 | if (likely(!err)) | |
2023 | return 0; | |
2024 | ||
2025 | /* Error: maybe zap and retry + rollback state for failed inserts. */ | |
2026 | return tcp_zerocopy_vm_insert_batch_error(vma, pages + pages_mapped, | |
2027 | pages_remaining, address, length, seq, zc, total_bytes_to_map, | |
2028 | err); | |
3763a24c AR |
2029 | } |
2030 | ||
3c5a2fd0 | 2031 | #define TCP_VALID_ZC_MSG_FLAGS (TCP_CMSG_TS) |
7eeba170 AR |
2032 | static void tcp_zc_finalize_rx_tstamp(struct sock *sk, |
2033 | struct tcp_zerocopy_receive *zc, | |
2034 | struct scm_timestamping_internal *tss) | |
2035 | { | |
2036 | unsigned long msg_control_addr; | |
2037 | struct msghdr cmsg_dummy; | |
2038 | ||
2039 | msg_control_addr = (unsigned long)zc->msg_control; | |
2040 | cmsg_dummy.msg_control = (void *)msg_control_addr; | |
2041 | cmsg_dummy.msg_controllen = | |
2042 | (__kernel_size_t)zc->msg_controllen; | |
2043 | cmsg_dummy.msg_flags = in_compat_syscall() | |
2044 | ? MSG_CMSG_COMPAT : 0; | |
a6f8ee58 | 2045 | cmsg_dummy.msg_control_is_user = true; |
7eeba170 AR |
2046 | zc->msg_flags = 0; |
2047 | if (zc->msg_control == msg_control_addr && | |
2048 | zc->msg_controllen == cmsg_dummy.msg_controllen) { | |
2049 | tcp_recv_timestamp(&cmsg_dummy, sk, tss); | |
2050 | zc->msg_control = (__u64) | |
2051 | ((uintptr_t)cmsg_dummy.msg_control); | |
2052 | zc->msg_controllen = | |
2053 | (__u64)cmsg_dummy.msg_controllen; | |
2054 | zc->msg_flags = (__u32)cmsg_dummy.msg_flags; | |
2055 | } | |
2056 | } | |
2057 | ||
94ab9eb9 | 2058 | #define TCP_ZEROCOPY_PAGE_BATCH_SIZE 32 |
05255b82 | 2059 | static int tcp_zerocopy_receive(struct sock *sk, |
7eeba170 AR |
2060 | struct tcp_zerocopy_receive *zc, |
2061 | struct scm_timestamping_internal *tss) | |
05255b82 | 2062 | { |
94ab9eb9 | 2063 | u32 length = 0, offset, vma_len, avail_len, copylen = 0; |
05255b82 | 2064 | unsigned long address = (unsigned long)zc->address; |
94ab9eb9 | 2065 | struct page *pages[TCP_ZEROCOPY_PAGE_BATCH_SIZE]; |
18fb76ed AR |
2066 | s32 copybuf_len = zc->copybuf_len; |
2067 | struct tcp_sock *tp = tcp_sk(sk); | |
05255b82 | 2068 | const skb_frag_t *frags = NULL; |
94ab9eb9 | 2069 | unsigned int pages_to_map = 0; |
05255b82 ED |
2070 | struct vm_area_struct *vma; |
2071 | struct sk_buff *skb = NULL; | |
18fb76ed | 2072 | u32 seq = tp->copied_seq; |
94ab9eb9 | 2073 | u32 total_bytes_to_map; |
18fb76ed | 2074 | int inq = tcp_inq(sk); |
93ab6cc6 ED |
2075 | int ret; |
2076 | ||
18fb76ed | 2077 | zc->copybuf_len = 0; |
7eeba170 | 2078 | zc->msg_flags = 0; |
18fb76ed | 2079 | |
05255b82 | 2080 | if (address & (PAGE_SIZE - 1) || address != zc->address) |
93ab6cc6 ED |
2081 | return -EINVAL; |
2082 | ||
93ab6cc6 | 2083 | if (sk->sk_state == TCP_LISTEN) |
05255b82 | 2084 | return -ENOTCONN; |
93ab6cc6 ED |
2085 | |
2086 | sock_rps_record_flow(sk); | |
2087 | ||
f21a3c48 | 2088 | if (inq && inq <= copybuf_len) |
7eeba170 | 2089 | return receive_fallback_to_copy(sk, zc, inq, tss); |
f21a3c48 | 2090 | |
936ced41 AR |
2091 | if (inq < PAGE_SIZE) { |
2092 | zc->length = 0; | |
2093 | zc->recv_skip_hint = inq; | |
2094 | if (!inq && sock_flag(sk, SOCK_DONE)) | |
2095 | return -EIO; | |
2096 | return 0; | |
2097 | } | |
2098 | ||
d8ed45c5 | 2099 | mmap_read_lock(current->mm); |
05255b82 | 2100 | |
47bdd1db LH |
2101 | vma = vma_lookup(current->mm, address); |
2102 | if (!vma || vma->vm_ops != &tcp_vm_ops) { | |
d8ed45c5 | 2103 | mmap_read_unlock(current->mm); |
e776af60 ED |
2104 | return -EINVAL; |
2105 | } | |
18fb76ed AR |
2106 | vma_len = min_t(unsigned long, zc->length, vma->vm_end - address); |
2107 | avail_len = min_t(u32, vma_len, inq); | |
94ab9eb9 AR |
2108 | total_bytes_to_map = avail_len & ~(PAGE_SIZE - 1); |
2109 | if (total_bytes_to_map) { | |
2110 | if (!(zc->flags & TCP_RECEIVE_ZEROCOPY_FLAG_TLB_CLEAN_HINT)) | |
2111 | zap_page_range(vma, address, total_bytes_to_map); | |
2112 | zc->length = total_bytes_to_map; | |
8f2b0293 SHY |
2113 | zc->recv_skip_hint = 0; |
2114 | } else { | |
18fb76ed AR |
2115 | zc->length = avail_len; |
2116 | zc->recv_skip_hint = avail_len; | |
8f2b0293 | 2117 | } |
05255b82 ED |
2118 | ret = 0; |
2119 | while (length + PAGE_SIZE <= zc->length) { | |
98917cf0 | 2120 | int mappable_offset; |
94ab9eb9 | 2121 | struct page *page; |
98917cf0 | 2122 | |
05255b82 | 2123 | if (zc->recv_skip_hint < PAGE_SIZE) { |
7fba5309 AR |
2124 | u32 offset_frag; |
2125 | ||
05255b82 | 2126 | if (skb) { |
0e627190 AR |
2127 | if (zc->recv_skip_hint > 0) |
2128 | break; | |
05255b82 ED |
2129 | skb = skb->next; |
2130 | offset = seq - TCP_SKB_CB(skb)->seq; | |
2131 | } else { | |
2132 | skb = tcp_recv_skb(sk, seq, &offset); | |
2133 | } | |
7eeba170 AR |
2134 | |
2135 | if (TCP_SKB_CB(skb)->has_rxtstamp) { | |
2136 | tcp_update_recv_tstamps(skb, tss); | |
2137 | zc->msg_flags |= TCP_CMSG_TS; | |
2138 | } | |
05255b82 | 2139 | zc->recv_skip_hint = skb->len - offset; |
7fba5309 AR |
2140 | frags = skb_advance_to_frag(skb, offset, &offset_frag); |
2141 | if (!frags || offset_frag) | |
05255b82 | 2142 | break; |
93ab6cc6 | 2143 | } |
789762ce | 2144 | |
98917cf0 AR |
2145 | mappable_offset = find_next_mappable_frag(frags, |
2146 | zc->recv_skip_hint); | |
2147 | if (mappable_offset) { | |
2148 | zc->recv_skip_hint = mappable_offset; | |
05255b82 | 2149 | break; |
789762ce | 2150 | } |
94ab9eb9 AR |
2151 | page = skb_frag_page(frags); |
2152 | prefetchw(page); | |
2153 | pages[pages_to_map++] = page; | |
05255b82 | 2154 | length += PAGE_SIZE; |
05255b82 ED |
2155 | zc->recv_skip_hint -= PAGE_SIZE; |
2156 | frags++; | |
94ab9eb9 AR |
2157 | if (pages_to_map == TCP_ZEROCOPY_PAGE_BATCH_SIZE || |
2158 | zc->recv_skip_hint < PAGE_SIZE) { | |
2159 | /* Either full batch, or we're about to go to next skb | |
2160 | * (and we cannot unroll failed ops across skbs). | |
2161 | */ | |
2162 | ret = tcp_zerocopy_vm_insert_batch(vma, pages, | |
2163 | pages_to_map, | |
2164 | &address, &length, | |
2165 | &seq, zc, | |
2166 | total_bytes_to_map); | |
3763a24c AR |
2167 | if (ret) |
2168 | goto out; | |
94ab9eb9 | 2169 | pages_to_map = 0; |
3763a24c AR |
2170 | } |
2171 | } | |
94ab9eb9 AR |
2172 | if (pages_to_map) { |
2173 | ret = tcp_zerocopy_vm_insert_batch(vma, pages, pages_to_map, | |
2174 | &address, &length, &seq, | |
2175 | zc, total_bytes_to_map); | |
93ab6cc6 | 2176 | } |
93ab6cc6 | 2177 | out: |
d8ed45c5 | 2178 | mmap_read_unlock(current->mm); |
18fb76ed AR |
2179 | /* Try to copy straggler data. */ |
2180 | if (!ret) | |
7eeba170 | 2181 | copylen = tcp_zc_handle_leftover(zc, sk, skb, &seq, copybuf_len, tss); |
18fb76ed AR |
2182 | |
2183 | if (length + copylen) { | |
7db48e98 | 2184 | WRITE_ONCE(tp->copied_seq, seq); |
05255b82 ED |
2185 | tcp_rcv_space_adjust(sk); |
2186 | ||
2187 | /* Clean up data we have read: This will do ACK frames. */ | |
2188 | tcp_recv_skb(sk, seq, &offset); | |
18fb76ed | 2189 | tcp_cleanup_rbuf(sk, length + copylen); |
05255b82 ED |
2190 | ret = 0; |
2191 | if (length == zc->length) | |
2192 | zc->recv_skip_hint = 0; | |
2193 | } else { | |
2194 | if (!zc->recv_skip_hint && sock_flag(sk, SOCK_DONE)) | |
2195 | ret = -EIO; | |
2196 | } | |
2197 | zc->length = length; | |
93ab6cc6 ED |
2198 | return ret; |
2199 | } | |
05255b82 | 2200 | #endif |
93ab6cc6 | 2201 | |
98aaa913 | 2202 | /* Similar to __sock_recv_timestamp, but does not require an skb */ |
892bfd3d FW |
2203 | void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, |
2204 | struct scm_timestamping_internal *tss) | |
98aaa913 | 2205 | { |
887feae3 | 2206 | int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW); |
98aaa913 MM |
2207 | bool has_timestamping = false; |
2208 | ||
2209 | if (tss->ts[0].tv_sec || tss->ts[0].tv_nsec) { | |
2210 | if (sock_flag(sk, SOCK_RCVTSTAMP)) { | |
2211 | if (sock_flag(sk, SOCK_RCVTSTAMPNS)) { | |
887feae3 | 2212 | if (new_tstamp) { |
df1b4ba9 AB |
2213 | struct __kernel_timespec kts = { |
2214 | .tv_sec = tss->ts[0].tv_sec, | |
2215 | .tv_nsec = tss->ts[0].tv_nsec, | |
2216 | }; | |
887feae3 DD |
2217 | put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW, |
2218 | sizeof(kts), &kts); | |
2219 | } else { | |
df1b4ba9 AB |
2220 | struct __kernel_old_timespec ts_old = { |
2221 | .tv_sec = tss->ts[0].tv_sec, | |
2222 | .tv_nsec = tss->ts[0].tv_nsec, | |
2223 | }; | |
887feae3 | 2224 | put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD, |
9718475e | 2225 | sizeof(ts_old), &ts_old); |
887feae3 | 2226 | } |
98aaa913 | 2227 | } else { |
887feae3 | 2228 | if (new_tstamp) { |
df1b4ba9 AB |
2229 | struct __kernel_sock_timeval stv = { |
2230 | .tv_sec = tss->ts[0].tv_sec, | |
2231 | .tv_usec = tss->ts[0].tv_nsec / 1000, | |
2232 | }; | |
887feae3 DD |
2233 | put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW, |
2234 | sizeof(stv), &stv); | |
2235 | } else { | |
df1b4ba9 AB |
2236 | struct __kernel_old_timeval tv = { |
2237 | .tv_sec = tss->ts[0].tv_sec, | |
2238 | .tv_usec = tss->ts[0].tv_nsec / 1000, | |
2239 | }; | |
887feae3 DD |
2240 | put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD, |
2241 | sizeof(tv), &tv); | |
2242 | } | |
98aaa913 MM |
2243 | } |
2244 | } | |
2245 | ||
2246 | if (sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) | |
2247 | has_timestamping = true; | |
2248 | else | |
9718475e | 2249 | tss->ts[0] = (struct timespec64) {0}; |
98aaa913 MM |
2250 | } |
2251 | ||
2252 | if (tss->ts[2].tv_sec || tss->ts[2].tv_nsec) { | |
2253 | if (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) | |
2254 | has_timestamping = true; | |
2255 | else | |
9718475e | 2256 | tss->ts[2] = (struct timespec64) {0}; |
98aaa913 MM |
2257 | } |
2258 | ||
2259 | if (has_timestamping) { | |
9718475e DD |
2260 | tss->ts[1] = (struct timespec64) {0}; |
2261 | if (sock_flag(sk, SOCK_TSTAMP_NEW)) | |
2262 | put_cmsg_scm_timestamping64(msg, tss); | |
2263 | else | |
2264 | put_cmsg_scm_timestamping(msg, tss); | |
98aaa913 MM |
2265 | } |
2266 | } | |
2267 | ||
b75eba76 SHY |
2268 | static int tcp_inq_hint(struct sock *sk) |
2269 | { | |
2270 | const struct tcp_sock *tp = tcp_sk(sk); | |
2271 | u32 copied_seq = READ_ONCE(tp->copied_seq); | |
2272 | u32 rcv_nxt = READ_ONCE(tp->rcv_nxt); | |
2273 | int inq; | |
2274 | ||
2275 | inq = rcv_nxt - copied_seq; | |
2276 | if (unlikely(inq < 0 || copied_seq != READ_ONCE(tp->copied_seq))) { | |
2277 | lock_sock(sk); | |
2278 | inq = tp->rcv_nxt - tp->copied_seq; | |
2279 | release_sock(sk); | |
2280 | } | |
6466e715 SHY |
2281 | /* After receiving a FIN, tell the user-space to continue reading |
2282 | * by returning a non-zero inq. | |
2283 | */ | |
2284 | if (inq == 0 && sock_flag(sk, SOCK_DONE)) | |
2285 | inq = 1; | |
b75eba76 SHY |
2286 | return inq; |
2287 | } | |
2288 | ||
1da177e4 LT |
2289 | /* |
2290 | * This routine copies from a sock struct into the user buffer. | |
2291 | * | |
2292 | * Technical note: in 2.3 we work on _locked_ socket, so that | |
2293 | * tricks with *seq access order and skb->users are not required. | |
2294 | * Probably, code can be easily improved even more. | |
2295 | */ | |
2296 | ||
2cd81161 AR |
2297 | static int tcp_recvmsg_locked(struct sock *sk, struct msghdr *msg, size_t len, |
2298 | int nonblock, int flags, | |
2299 | struct scm_timestamping_internal *tss, | |
2300 | int *cmsg_flags) | |
1da177e4 LT |
2301 | { |
2302 | struct tcp_sock *tp = tcp_sk(sk); | |
2303 | int copied = 0; | |
2304 | u32 peek_seq; | |
2305 | u32 *seq; | |
2306 | unsigned long used; | |
2cd81161 | 2307 | int err; |
1da177e4 LT |
2308 | int target; /* Read at least this many bytes */ |
2309 | long timeo; | |
dfbafc99 | 2310 | struct sk_buff *skb, *last; |
77527313 | 2311 | u32 urg_hole = 0; |
1da177e4 | 2312 | |
1da177e4 LT |
2313 | err = -ENOTCONN; |
2314 | if (sk->sk_state == TCP_LISTEN) | |
2315 | goto out; | |
2316 | ||
2cd81161 | 2317 | if (tp->recvmsg_inq) |
925bba24 | 2318 | *cmsg_flags = TCP_CMSG_INQ; |
1da177e4 LT |
2319 | timeo = sock_rcvtimeo(sk, nonblock); |
2320 | ||
2321 | /* Urgent data needs to be handled specially. */ | |
2322 | if (flags & MSG_OOB) | |
2323 | goto recv_urg; | |
2324 | ||
c0e88ff0 PE |
2325 | if (unlikely(tp->repair)) { |
2326 | err = -EPERM; | |
2327 | if (!(flags & MSG_PEEK)) | |
2328 | goto out; | |
2329 | ||
2330 | if (tp->repair_queue == TCP_SEND_QUEUE) | |
2331 | goto recv_sndq; | |
2332 | ||
2333 | err = -EINVAL; | |
2334 | if (tp->repair_queue == TCP_NO_QUEUE) | |
2335 | goto out; | |
2336 | ||
2337 | /* 'common' recv queue MSG_PEEK-ing */ | |
2338 | } | |
2339 | ||
1da177e4 LT |
2340 | seq = &tp->copied_seq; |
2341 | if (flags & MSG_PEEK) { | |
2342 | peek_seq = tp->copied_seq; | |
2343 | seq = &peek_seq; | |
2344 | } | |
2345 | ||
2346 | target = sock_rcvlowat(sk, flags & MSG_WAITALL, len); | |
2347 | ||
2348 | do { | |
1da177e4 LT |
2349 | u32 offset; |
2350 | ||
2351 | /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */ | |
2352 | if (tp->urg_data && tp->urg_seq == *seq) { | |
2353 | if (copied) | |
2354 | break; | |
2355 | if (signal_pending(current)) { | |
2356 | copied = timeo ? sock_intr_errno(timeo) : -EAGAIN; | |
2357 | break; | |
2358 | } | |
2359 | } | |
2360 | ||
2361 | /* Next get a buffer. */ | |
2362 | ||
dfbafc99 | 2363 | last = skb_peek_tail(&sk->sk_receive_queue); |
91521944 | 2364 | skb_queue_walk(&sk->sk_receive_queue, skb) { |
dfbafc99 | 2365 | last = skb; |
1da177e4 LT |
2366 | /* Now that we have two receive queues this |
2367 | * shouldn't happen. | |
2368 | */ | |
d792c100 | 2369 | if (WARN(before(*seq, TCP_SKB_CB(skb)->seq), |
e56b8ce3 | 2370 | "TCP recvmsg seq # bug: copied %X, seq %X, rcvnxt %X, fl %X\n", |
2af6fd8b JP |
2371 | *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, |
2372 | flags)) | |
1da177e4 | 2373 | break; |
d792c100 | 2374 | |
1da177e4 | 2375 | offset = *seq - TCP_SKB_CB(skb)->seq; |
9d691539 ED |
2376 | if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { |
2377 | pr_err_once("%s: found a SYN, please report !\n", __func__); | |
1da177e4 | 2378 | offset--; |
9d691539 | 2379 | } |
1da177e4 LT |
2380 | if (offset < skb->len) |
2381 | goto found_ok_skb; | |
e11ecddf | 2382 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) |
1da177e4 | 2383 | goto found_fin_ok; |
2af6fd8b | 2384 | WARN(!(flags & MSG_PEEK), |
e56b8ce3 | 2385 | "TCP recvmsg seq # bug 2: copied %X, seq %X, rcvnxt %X, fl %X\n", |
2af6fd8b | 2386 | *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags); |
91521944 | 2387 | } |
1da177e4 LT |
2388 | |
2389 | /* Well, if we have backlog, try to process it now yet. */ | |
2390 | ||
9ed498c6 | 2391 | if (copied >= target && !READ_ONCE(sk->sk_backlog.tail)) |
1da177e4 LT |
2392 | break; |
2393 | ||
2394 | if (copied) { | |
2395 | if (sk->sk_err || | |
2396 | sk->sk_state == TCP_CLOSE || | |
2397 | (sk->sk_shutdown & RCV_SHUTDOWN) || | |
2398 | !timeo || | |
518a09ef | 2399 | signal_pending(current)) |
1da177e4 LT |
2400 | break; |
2401 | } else { | |
2402 | if (sock_flag(sk, SOCK_DONE)) | |
2403 | break; | |
2404 | ||
2405 | if (sk->sk_err) { | |
2406 | copied = sock_error(sk); | |
2407 | break; | |
2408 | } | |
2409 | ||
2410 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
2411 | break; | |
2412 | ||
2413 | if (sk->sk_state == TCP_CLOSE) { | |
c47078d6 ED |
2414 | /* This occurs when user tries to read |
2415 | * from never connected socket. | |
2416 | */ | |
2417 | copied = -ENOTCONN; | |
1da177e4 LT |
2418 | break; |
2419 | } | |
2420 | ||
2421 | if (!timeo) { | |
2422 | copied = -EAGAIN; | |
2423 | break; | |
2424 | } | |
2425 | ||
2426 | if (signal_pending(current)) { | |
2427 | copied = sock_intr_errno(timeo); | |
2428 | break; | |
2429 | } | |
2430 | } | |
2431 | ||
0e4b4992 | 2432 | tcp_cleanup_rbuf(sk, copied); |
1da177e4 | 2433 | |
1da177e4 LT |
2434 | if (copied >= target) { |
2435 | /* Do not sleep, just process backlog. */ | |
2436 | release_sock(sk); | |
2437 | lock_sock(sk); | |
dfbafc99 SD |
2438 | } else { |
2439 | sk_wait_data(sk, &timeo, last); | |
2440 | } | |
1da177e4 | 2441 | |
77527313 IJ |
2442 | if ((flags & MSG_PEEK) && |
2443 | (peek_seq - copied - urg_hole != tp->copied_seq)) { | |
e87cc472 JP |
2444 | net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n", |
2445 | current->comm, | |
2446 | task_pid_nr(current)); | |
1da177e4 LT |
2447 | peek_seq = tp->copied_seq; |
2448 | } | |
2449 | continue; | |
2450 | ||
fdb8b298 | 2451 | found_ok_skb: |
1da177e4 LT |
2452 | /* Ok so how much can we use? */ |
2453 | used = skb->len - offset; | |
2454 | if (len < used) | |
2455 | used = len; | |
2456 | ||
2457 | /* Do we have urgent data here? */ | |
2458 | if (tp->urg_data) { | |
2459 | u32 urg_offset = tp->urg_seq - *seq; | |
2460 | if (urg_offset < used) { | |
2461 | if (!urg_offset) { | |
2462 | if (!sock_flag(sk, SOCK_URGINLINE)) { | |
7db48e98 | 2463 | WRITE_ONCE(*seq, *seq + 1); |
77527313 | 2464 | urg_hole++; |
1da177e4 LT |
2465 | offset++; |
2466 | used--; | |
2467 | if (!used) | |
2468 | goto skip_copy; | |
2469 | } | |
2470 | } else | |
2471 | used = urg_offset; | |
2472 | } | |
2473 | } | |
2474 | ||
2475 | if (!(flags & MSG_TRUNC)) { | |
51f3d02b | 2476 | err = skb_copy_datagram_msg(skb, offset, msg, used); |
7bced397 DW |
2477 | if (err) { |
2478 | /* Exception. Bailout! */ | |
2479 | if (!copied) | |
2480 | copied = -EFAULT; | |
2481 | break; | |
1da177e4 LT |
2482 | } |
2483 | } | |
2484 | ||
7db48e98 | 2485 | WRITE_ONCE(*seq, *seq + used); |
1da177e4 LT |
2486 | copied += used; |
2487 | len -= used; | |
2488 | ||
2489 | tcp_rcv_space_adjust(sk); | |
2490 | ||
2491 | skip_copy: | |
31770e34 | 2492 | if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) { |
1da177e4 | 2493 | tp->urg_data = 0; |
31770e34 FW |
2494 | tcp_fast_path_check(sk); |
2495 | } | |
1da177e4 | 2496 | |
98aaa913 | 2497 | if (TCP_SKB_CB(skb)->has_rxtstamp) { |
2cd81161 | 2498 | tcp_update_recv_tstamps(skb, tss); |
925bba24 | 2499 | *cmsg_flags |= TCP_CMSG_TS; |
98aaa913 | 2500 | } |
cc4de047 KL |
2501 | |
2502 | if (used + offset < skb->len) | |
2503 | continue; | |
2504 | ||
e11ecddf | 2505 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) |
1da177e4 | 2506 | goto found_fin_ok; |
7bced397 DW |
2507 | if (!(flags & MSG_PEEK)) |
2508 | sk_eat_skb(sk, skb); | |
1da177e4 LT |
2509 | continue; |
2510 | ||
fdb8b298 | 2511 | found_fin_ok: |
1da177e4 | 2512 | /* Process the FIN. */ |
7db48e98 | 2513 | WRITE_ONCE(*seq, *seq + 1); |
7bced397 DW |
2514 | if (!(flags & MSG_PEEK)) |
2515 | sk_eat_skb(sk, skb); | |
1da177e4 LT |
2516 | break; |
2517 | } while (len > 0); | |
2518 | ||
1da177e4 LT |
2519 | /* According to UNIX98, msg_name/msg_namelen are ignored |
2520 | * on connected socket. I was just happy when found this 8) --ANK | |
2521 | */ | |
2522 | ||
2523 | /* Clean up data we have read: This will do ACK frames. */ | |
0e4b4992 | 2524 | tcp_cleanup_rbuf(sk, copied); |
1da177e4 LT |
2525 | return copied; |
2526 | ||
2527 | out: | |
1da177e4 LT |
2528 | return err; |
2529 | ||
2530 | recv_urg: | |
377f0a08 | 2531 | err = tcp_recv_urg(sk, msg, len, flags); |
1da177e4 | 2532 | goto out; |
c0e88ff0 PE |
2533 | |
2534 | recv_sndq: | |
2535 | err = tcp_peek_sndq(sk, msg, len); | |
2536 | goto out; | |
1da177e4 | 2537 | } |
2cd81161 AR |
2538 | |
2539 | int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, | |
2540 | int flags, int *addr_len) | |
2541 | { | |
2542 | int cmsg_flags = 0, ret, inq; | |
2543 | struct scm_timestamping_internal tss; | |
2544 | ||
2545 | if (unlikely(flags & MSG_ERRQUEUE)) | |
2546 | return inet_recv_error(sk, msg, len, addr_len); | |
2547 | ||
2548 | if (sk_can_busy_loop(sk) && | |
2549 | skb_queue_empty_lockless(&sk->sk_receive_queue) && | |
2550 | sk->sk_state == TCP_ESTABLISHED) | |
2551 | sk_busy_loop(sk, nonblock); | |
2552 | ||
2553 | lock_sock(sk); | |
2554 | ret = tcp_recvmsg_locked(sk, msg, len, nonblock, flags, &tss, | |
2555 | &cmsg_flags); | |
2556 | release_sock(sk); | |
2557 | ||
2558 | if (cmsg_flags && ret >= 0) { | |
925bba24 | 2559 | if (cmsg_flags & TCP_CMSG_TS) |
2cd81161 | 2560 | tcp_recv_timestamp(msg, sk, &tss); |
925bba24 | 2561 | if (cmsg_flags & TCP_CMSG_INQ) { |
2cd81161 AR |
2562 | inq = tcp_inq_hint(sk); |
2563 | put_cmsg(msg, SOL_TCP, TCP_CM_INQ, sizeof(inq), &inq); | |
2564 | } | |
2565 | } | |
2566 | return ret; | |
2567 | } | |
4bc2f18b | 2568 | EXPORT_SYMBOL(tcp_recvmsg); |
1da177e4 | 2569 | |
490d5046 IJ |
2570 | void tcp_set_state(struct sock *sk, int state) |
2571 | { | |
2572 | int oldstate = sk->sk_state; | |
2573 | ||
d4487491 LB |
2574 | /* We defined a new enum for TCP states that are exported in BPF |
2575 | * so as not force the internal TCP states to be frozen. The | |
2576 | * following checks will detect if an internal state value ever | |
2577 | * differs from the BPF value. If this ever happens, then we will | |
2578 | * need to remap the internal value to the BPF value before calling | |
2579 | * tcp_call_bpf_2arg. | |
2580 | */ | |
2581 | BUILD_BUG_ON((int)BPF_TCP_ESTABLISHED != (int)TCP_ESTABLISHED); | |
2582 | BUILD_BUG_ON((int)BPF_TCP_SYN_SENT != (int)TCP_SYN_SENT); | |
2583 | BUILD_BUG_ON((int)BPF_TCP_SYN_RECV != (int)TCP_SYN_RECV); | |
2584 | BUILD_BUG_ON((int)BPF_TCP_FIN_WAIT1 != (int)TCP_FIN_WAIT1); | |
2585 | BUILD_BUG_ON((int)BPF_TCP_FIN_WAIT2 != (int)TCP_FIN_WAIT2); | |
2586 | BUILD_BUG_ON((int)BPF_TCP_TIME_WAIT != (int)TCP_TIME_WAIT); | |
2587 | BUILD_BUG_ON((int)BPF_TCP_CLOSE != (int)TCP_CLOSE); | |
2588 | BUILD_BUG_ON((int)BPF_TCP_CLOSE_WAIT != (int)TCP_CLOSE_WAIT); | |
2589 | BUILD_BUG_ON((int)BPF_TCP_LAST_ACK != (int)TCP_LAST_ACK); | |
2590 | BUILD_BUG_ON((int)BPF_TCP_LISTEN != (int)TCP_LISTEN); | |
2591 | BUILD_BUG_ON((int)BPF_TCP_CLOSING != (int)TCP_CLOSING); | |
2592 | BUILD_BUG_ON((int)BPF_TCP_NEW_SYN_RECV != (int)TCP_NEW_SYN_RECV); | |
2593 | BUILD_BUG_ON((int)BPF_TCP_MAX_STATES != (int)TCP_MAX_STATES); | |
2594 | ||
97a19caf YS |
2595 | /* bpf uapi header bpf.h defines an anonymous enum with values |
2596 | * BPF_TCP_* used by bpf programs. Currently gcc built vmlinux | |
2597 | * is able to emit this enum in DWARF due to the above BUILD_BUG_ON. | |
2598 | * But clang built vmlinux does not have this enum in DWARF | |
2599 | * since clang removes the above code before generating IR/debuginfo. | |
2600 | * Let us explicitly emit the type debuginfo to ensure the | |
2601 | * above-mentioned anonymous enum in the vmlinux DWARF and hence BTF | |
2602 | * regardless of which compiler is used. | |
2603 | */ | |
2604 | BTF_TYPE_EMIT_ENUM(BPF_TCP_ESTABLISHED); | |
2605 | ||
d4487491 LB |
2606 | if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_STATE_CB_FLAG)) |
2607 | tcp_call_bpf_2arg(sk, BPF_SOCK_OPS_STATE_CB, oldstate, state); | |
e8fce239 | 2608 | |
490d5046 IJ |
2609 | switch (state) { |
2610 | case TCP_ESTABLISHED: | |
2611 | if (oldstate != TCP_ESTABLISHED) | |
81cc8a75 | 2612 | TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB); |
490d5046 IJ |
2613 | break; |
2614 | ||
2615 | case TCP_CLOSE: | |
2616 | if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED) | |
81cc8a75 | 2617 | TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS); |
490d5046 IJ |
2618 | |
2619 | sk->sk_prot->unhash(sk); | |
2620 | if (inet_csk(sk)->icsk_bind_hash && | |
2621 | !(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) | |
ab1e0a13 | 2622 | inet_put_port(sk); |
a8eceea8 | 2623 | fallthrough; |
490d5046 | 2624 | default: |
5a5f3a8d | 2625 | if (oldstate == TCP_ESTABLISHED) |
74688e48 | 2626 | TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB); |
490d5046 IJ |
2627 | } |
2628 | ||
2629 | /* Change state AFTER socket is unhashed to avoid closed | |
2630 | * socket sitting in hash tables. | |
2631 | */ | |
563e0bb0 | 2632 | inet_sk_state_store(sk, state); |
490d5046 IJ |
2633 | } |
2634 | EXPORT_SYMBOL_GPL(tcp_set_state); | |
2635 | ||
1da177e4 LT |
2636 | /* |
2637 | * State processing on a close. This implements the state shift for | |
2638 | * sending our FIN frame. Note that we only send a FIN for some | |
2639 | * states. A shutdown() may have already sent the FIN, or we may be | |
2640 | * closed. | |
2641 | */ | |
2642 | ||
9b5b5cff | 2643 | static const unsigned char new_state[16] = { |
1da177e4 | 2644 | /* current state: new state: action: */ |
0980c1e3 ED |
2645 | [0 /* (Invalid) */] = TCP_CLOSE, |
2646 | [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN, | |
2647 | [TCP_SYN_SENT] = TCP_CLOSE, | |
2648 | [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN, | |
2649 | [TCP_FIN_WAIT1] = TCP_FIN_WAIT1, | |
2650 | [TCP_FIN_WAIT2] = TCP_FIN_WAIT2, | |
2651 | [TCP_TIME_WAIT] = TCP_CLOSE, | |
2652 | [TCP_CLOSE] = TCP_CLOSE, | |
2653 | [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN, | |
2654 | [TCP_LAST_ACK] = TCP_LAST_ACK, | |
2655 | [TCP_LISTEN] = TCP_CLOSE, | |
2656 | [TCP_CLOSING] = TCP_CLOSING, | |
2657 | [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */ | |
1da177e4 LT |
2658 | }; |
2659 | ||
2660 | static int tcp_close_state(struct sock *sk) | |
2661 | { | |
2662 | int next = (int)new_state[sk->sk_state]; | |
2663 | int ns = next & TCP_STATE_MASK; | |
2664 | ||
2665 | tcp_set_state(sk, ns); | |
2666 | ||
2667 | return next & TCP_ACTION_FIN; | |
2668 | } | |
2669 | ||
2670 | /* | |
2671 | * Shutdown the sending side of a connection. Much like close except | |
1f29b058 | 2672 | * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD). |
1da177e4 LT |
2673 | */ |
2674 | ||
2675 | void tcp_shutdown(struct sock *sk, int how) | |
2676 | { | |
2677 | /* We need to grab some memory, and put together a FIN, | |
2678 | * and then put it into the queue to be sent. | |
2679 | * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92. | |
2680 | */ | |
2681 | if (!(how & SEND_SHUTDOWN)) | |
2682 | return; | |
2683 | ||
2684 | /* If we've already sent a FIN, or it's a closed state, skip this. */ | |
2685 | if ((1 << sk->sk_state) & | |
2686 | (TCPF_ESTABLISHED | TCPF_SYN_SENT | | |
2687 | TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) { | |
2688 | /* Clear out any half completed packets. FIN if needed. */ | |
2689 | if (tcp_close_state(sk)) | |
2690 | tcp_send_fin(sk); | |
2691 | } | |
2692 | } | |
4bc2f18b | 2693 | EXPORT_SYMBOL(tcp_shutdown); |
1da177e4 | 2694 | |
e40ec5b7 ED |
2695 | int tcp_orphan_count_sum(void) |
2696 | { | |
2697 | int i, total = 0; | |
2698 | ||
2699 | for_each_possible_cpu(i) | |
2700 | total += per_cpu(tcp_orphan_count, i); | |
2701 | ||
2702 | return max(total, 0); | |
2703 | } | |
2704 | ||
2705 | static int tcp_orphan_cache; | |
2706 | static struct timer_list tcp_orphan_timer; | |
2707 | #define TCP_ORPHAN_TIMER_PERIOD msecs_to_jiffies(100) | |
2708 | ||
2709 | static void tcp_orphan_update(struct timer_list *unused) | |
2710 | { | |
2711 | WRITE_ONCE(tcp_orphan_cache, tcp_orphan_count_sum()); | |
2712 | mod_timer(&tcp_orphan_timer, jiffies + TCP_ORPHAN_TIMER_PERIOD); | |
2713 | } | |
2714 | ||
2715 | static bool tcp_too_many_orphans(int shift) | |
2716 | { | |
2717 | return READ_ONCE(tcp_orphan_cache) << shift > sysctl_tcp_max_orphans; | |
2718 | } | |
2719 | ||
efcdbf24 AS |
2720 | bool tcp_check_oom(struct sock *sk, int shift) |
2721 | { | |
2722 | bool too_many_orphans, out_of_socket_memory; | |
2723 | ||
e40ec5b7 | 2724 | too_many_orphans = tcp_too_many_orphans(shift); |
efcdbf24 AS |
2725 | out_of_socket_memory = tcp_out_of_memory(sk); |
2726 | ||
e87cc472 JP |
2727 | if (too_many_orphans) |
2728 | net_info_ratelimited("too many orphaned sockets\n"); | |
2729 | if (out_of_socket_memory) | |
2730 | net_info_ratelimited("out of memory -- consider tuning tcp_mem\n"); | |
efcdbf24 AS |
2731 | return too_many_orphans || out_of_socket_memory; |
2732 | } | |
2733 | ||
77c3c956 | 2734 | void __tcp_close(struct sock *sk, long timeout) |
1da177e4 LT |
2735 | { |
2736 | struct sk_buff *skb; | |
2737 | int data_was_unread = 0; | |
75c2d907 | 2738 | int state; |
1da177e4 | 2739 | |
1da177e4 LT |
2740 | sk->sk_shutdown = SHUTDOWN_MASK; |
2741 | ||
2742 | if (sk->sk_state == TCP_LISTEN) { | |
2743 | tcp_set_state(sk, TCP_CLOSE); | |
2744 | ||
2745 | /* Special case. */ | |
0a5578cf | 2746 | inet_csk_listen_stop(sk); |
1da177e4 LT |
2747 | |
2748 | goto adjudge_to_death; | |
2749 | } | |
2750 | ||
2751 | /* We need to flush the recv. buffs. We do this only on the | |
2752 | * descriptor close, not protocol-sourced closes, because the | |
2753 | * reader process may not have drained the data yet! | |
2754 | */ | |
2755 | while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) { | |
e11ecddf ED |
2756 | u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq; |
2757 | ||
2758 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) | |
2759 | len--; | |
1da177e4 LT |
2760 | data_was_unread += len; |
2761 | __kfree_skb(skb); | |
2762 | } | |
2763 | ||
3ab224be | 2764 | sk_mem_reclaim(sk); |
1da177e4 | 2765 | |
565b7b2d KK |
2766 | /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */ |
2767 | if (sk->sk_state == TCP_CLOSE) | |
2768 | goto adjudge_to_death; | |
2769 | ||
65bb723c GR |
2770 | /* As outlined in RFC 2525, section 2.17, we send a RST here because |
2771 | * data was lost. To witness the awful effects of the old behavior of | |
2772 | * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk | |
2773 | * GET in an FTP client, suspend the process, wait for the client to | |
2774 | * advertise a zero window, then kill -9 the FTP client, wheee... | |
2775 | * Note: timeout is always zero in such a case. | |
1da177e4 | 2776 | */ |
ee995283 PE |
2777 | if (unlikely(tcp_sk(sk)->repair)) { |
2778 | sk->sk_prot->disconnect(sk, 0); | |
2779 | } else if (data_was_unread) { | |
1da177e4 | 2780 | /* Unread data was tossed, zap the connection. */ |
6aef70a8 | 2781 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE); |
1da177e4 | 2782 | tcp_set_state(sk, TCP_CLOSE); |
aa133076 | 2783 | tcp_send_active_reset(sk, sk->sk_allocation); |
1da177e4 LT |
2784 | } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) { |
2785 | /* Check zero linger _after_ checking for unread data. */ | |
2786 | sk->sk_prot->disconnect(sk, 0); | |
6aef70a8 | 2787 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONDATA); |
1da177e4 LT |
2788 | } else if (tcp_close_state(sk)) { |
2789 | /* We FIN if the application ate all the data before | |
2790 | * zapping the connection. | |
2791 | */ | |
2792 | ||
2793 | /* RED-PEN. Formally speaking, we have broken TCP state | |
2794 | * machine. State transitions: | |
2795 | * | |
2796 | * TCP_ESTABLISHED -> TCP_FIN_WAIT1 | |
2797 | * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible) | |
2798 | * TCP_CLOSE_WAIT -> TCP_LAST_ACK | |
2799 | * | |
2800 | * are legal only when FIN has been sent (i.e. in window), | |
2801 | * rather than queued out of window. Purists blame. | |
2802 | * | |
2803 | * F.e. "RFC state" is ESTABLISHED, | |
2804 | * if Linux state is FIN-WAIT-1, but FIN is still not sent. | |
2805 | * | |
2806 | * The visible declinations are that sometimes | |
2807 | * we enter time-wait state, when it is not required really | |
2808 | * (harmless), do not send active resets, when they are | |
2809 | * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when | |
2810 | * they look as CLOSING or LAST_ACK for Linux) | |
2811 | * Probably, I missed some more holelets. | |
2812 | * --ANK | |
8336886f JC |
2813 | * XXX (TFO) - To start off we don't support SYN+ACK+FIN |
2814 | * in a single packet! (May consider it later but will | |
2815 | * probably need API support or TCP_CORK SYN-ACK until | |
2816 | * data is written and socket is closed.) | |
1da177e4 LT |
2817 | */ |
2818 | tcp_send_fin(sk); | |
2819 | } | |
2820 | ||
2821 | sk_stream_wait_close(sk, timeout); | |
2822 | ||
2823 | adjudge_to_death: | |
75c2d907 HX |
2824 | state = sk->sk_state; |
2825 | sock_hold(sk); | |
2826 | sock_orphan(sk); | |
75c2d907 | 2827 | |
1da177e4 LT |
2828 | local_bh_disable(); |
2829 | bh_lock_sock(sk); | |
8873c064 ED |
2830 | /* remove backlog if any, without releasing ownership. */ |
2831 | __release_sock(sk); | |
1da177e4 | 2832 | |
e40ec5b7 | 2833 | this_cpu_inc(tcp_orphan_count); |
eb4dea58 | 2834 | |
75c2d907 HX |
2835 | /* Have we already been destroyed by a softirq or backlog? */ |
2836 | if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE) | |
2837 | goto out; | |
1da177e4 LT |
2838 | |
2839 | /* This is a (useful) BSD violating of the RFC. There is a | |
2840 | * problem with TCP as specified in that the other end could | |
2841 | * keep a socket open forever with no application left this end. | |
b10bd54c | 2842 | * We use a 1 minute timeout (about the same as BSD) then kill |
1da177e4 LT |
2843 | * our end. If they send after that then tough - BUT: long enough |
2844 | * that we won't make the old 4*rto = almost no time - whoops | |
2845 | * reset mistake. | |
2846 | * | |
2847 | * Nope, it was not mistake. It is really desired behaviour | |
2848 | * f.e. on http servers, when such sockets are useless, but | |
2849 | * consume significant resources. Let's do it with special | |
2850 | * linger2 option. --ANK | |
2851 | */ | |
2852 | ||
2853 | if (sk->sk_state == TCP_FIN_WAIT2) { | |
2854 | struct tcp_sock *tp = tcp_sk(sk); | |
2855 | if (tp->linger2 < 0) { | |
2856 | tcp_set_state(sk, TCP_CLOSE); | |
2857 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
02a1d6e7 | 2858 | __NET_INC_STATS(sock_net(sk), |
de0744af | 2859 | LINUX_MIB_TCPABORTONLINGER); |
1da177e4 | 2860 | } else { |
463c84b9 | 2861 | const int tmo = tcp_fin_time(sk); |
1da177e4 LT |
2862 | |
2863 | if (tmo > TCP_TIMEWAIT_LEN) { | |
52499afe DM |
2864 | inet_csk_reset_keepalive_timer(sk, |
2865 | tmo - TCP_TIMEWAIT_LEN); | |
1da177e4 | 2866 | } else { |
1da177e4 LT |
2867 | tcp_time_wait(sk, TCP_FIN_WAIT2, tmo); |
2868 | goto out; | |
2869 | } | |
2870 | } | |
2871 | } | |
2872 | if (sk->sk_state != TCP_CLOSE) { | |
3ab224be | 2873 | sk_mem_reclaim(sk); |
efcdbf24 | 2874 | if (tcp_check_oom(sk, 0)) { |
1da177e4 LT |
2875 | tcp_set_state(sk, TCP_CLOSE); |
2876 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
02a1d6e7 | 2877 | __NET_INC_STATS(sock_net(sk), |
de0744af | 2878 | LINUX_MIB_TCPABORTONMEMORY); |
4ee806d5 DS |
2879 | } else if (!check_net(sock_net(sk))) { |
2880 | /* Not possible to send reset; just close */ | |
2881 | tcp_set_state(sk, TCP_CLOSE); | |
1da177e4 LT |
2882 | } |
2883 | } | |
1da177e4 | 2884 | |
8336886f | 2885 | if (sk->sk_state == TCP_CLOSE) { |
d983ea6f ED |
2886 | struct request_sock *req; |
2887 | ||
2888 | req = rcu_dereference_protected(tcp_sk(sk)->fastopen_rsk, | |
2889 | lockdep_sock_is_held(sk)); | |
8336886f JC |
2890 | /* We could get here with a non-NULL req if the socket is |
2891 | * aborted (e.g., closed with unread data) before 3WHS | |
2892 | * finishes. | |
2893 | */ | |
00db4124 | 2894 | if (req) |
8336886f | 2895 | reqsk_fastopen_remove(sk, req, false); |
0a5578cf | 2896 | inet_csk_destroy_sock(sk); |
8336886f | 2897 | } |
1da177e4 LT |
2898 | /* Otherwise, socket is reprieved until protocol close. */ |
2899 | ||
2900 | out: | |
2901 | bh_unlock_sock(sk); | |
2902 | local_bh_enable(); | |
77c3c956 PA |
2903 | } |
2904 | ||
2905 | void tcp_close(struct sock *sk, long timeout) | |
2906 | { | |
2907 | lock_sock(sk); | |
2908 | __tcp_close(sk, timeout); | |
8873c064 | 2909 | release_sock(sk); |
1da177e4 LT |
2910 | sock_put(sk); |
2911 | } | |
4bc2f18b | 2912 | EXPORT_SYMBOL(tcp_close); |
1da177e4 LT |
2913 | |
2914 | /* These states need RST on ABORT according to RFC793 */ | |
2915 | ||
a2a385d6 | 2916 | static inline bool tcp_need_reset(int state) |
1da177e4 LT |
2917 | { |
2918 | return (1 << state) & | |
2919 | (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 | | |
a7150e38 | 2920 | TCPF_FIN_WAIT2 | TCPF_SYN_RECV); |
1da177e4 LT |
2921 | } |
2922 | ||
75c119af ED |
2923 | static void tcp_rtx_queue_purge(struct sock *sk) |
2924 | { | |
2925 | struct rb_node *p = rb_first(&sk->tcp_rtx_queue); | |
2926 | ||
2bec445f | 2927 | tcp_sk(sk)->highest_sack = NULL; |
75c119af ED |
2928 | while (p) { |
2929 | struct sk_buff *skb = rb_to_skb(p); | |
2930 | ||
2931 | p = rb_next(p); | |
2932 | /* Since we are deleting whole queue, no need to | |
2933 | * list_del(&skb->tcp_tsorted_anchor) | |
2934 | */ | |
2935 | tcp_rtx_queue_unlink(skb, sk); | |
2936 | sk_wmem_free_skb(sk, skb); | |
2937 | } | |
2938 | } | |
2939 | ||
ac3f09ba ED |
2940 | void tcp_write_queue_purge(struct sock *sk) |
2941 | { | |
2942 | struct sk_buff *skb; | |
2943 | ||
2944 | tcp_chrono_stop(sk, TCP_CHRONO_BUSY); | |
2945 | while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) { | |
2946 | tcp_skb_tsorted_anchor_cleanup(skb); | |
2947 | sk_wmem_free_skb(sk, skb); | |
2948 | } | |
75c119af | 2949 | tcp_rtx_queue_purge(sk); |
472c2e07 ED |
2950 | skb = sk->sk_tx_skb_cache; |
2951 | if (skb) { | |
472c2e07 ED |
2952 | __kfree_skb(skb); |
2953 | sk->sk_tx_skb_cache = NULL; | |
2954 | } | |
ac3f09ba ED |
2955 | INIT_LIST_HEAD(&tcp_sk(sk)->tsorted_sent_queue); |
2956 | sk_mem_reclaim(sk); | |
2957 | tcp_clear_all_retrans_hints(tcp_sk(sk)); | |
bffd168c | 2958 | tcp_sk(sk)->packets_out = 0; |
04c03114 | 2959 | inet_csk(sk)->icsk_backoff = 0; |
ac3f09ba ED |
2960 | } |
2961 | ||
1da177e4 LT |
2962 | int tcp_disconnect(struct sock *sk, int flags) |
2963 | { | |
2964 | struct inet_sock *inet = inet_sk(sk); | |
463c84b9 | 2965 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 2966 | struct tcp_sock *tp = tcp_sk(sk); |
1da177e4 | 2967 | int old_state = sk->sk_state; |
0f317464 | 2968 | u32 seq; |
1da177e4 LT |
2969 | |
2970 | if (old_state != TCP_CLOSE) | |
2971 | tcp_set_state(sk, TCP_CLOSE); | |
2972 | ||
2973 | /* ABORT function of RFC793 */ | |
2974 | if (old_state == TCP_LISTEN) { | |
0a5578cf | 2975 | inet_csk_listen_stop(sk); |
ee995283 PE |
2976 | } else if (unlikely(tp->repair)) { |
2977 | sk->sk_err = ECONNABORTED; | |
1da177e4 LT |
2978 | } else if (tcp_need_reset(old_state) || |
2979 | (tp->snd_nxt != tp->write_seq && | |
2980 | (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) { | |
caa20d9a | 2981 | /* The last check adjusts for discrepancy of Linux wrt. RFC |
1da177e4 LT |
2982 | * states |
2983 | */ | |
2984 | tcp_send_active_reset(sk, gfp_any()); | |
2985 | sk->sk_err = ECONNRESET; | |
a7150e38 ED |
2986 | } else if (old_state == TCP_SYN_SENT) |
2987 | sk->sk_err = ECONNRESET; | |
1da177e4 LT |
2988 | |
2989 | tcp_clear_xmit_timers(sk); | |
2990 | __skb_queue_purge(&sk->sk_receive_queue); | |
8b27dae5 ED |
2991 | if (sk->sk_rx_skb_cache) { |
2992 | __kfree_skb(sk->sk_rx_skb_cache); | |
2993 | sk->sk_rx_skb_cache = NULL; | |
2994 | } | |
7db48e98 | 2995 | WRITE_ONCE(tp->copied_seq, tp->rcv_nxt); |
6508b678 | 2996 | tp->urg_data = 0; |
fe067e8a | 2997 | tcp_write_queue_purge(sk); |
cf1ef3f0 | 2998 | tcp_fastopen_active_disable_ofo_check(sk); |
9f5afeae | 2999 | skb_rbtree_purge(&tp->out_of_order_queue); |
1da177e4 | 3000 | |
c720c7e8 | 3001 | inet->inet_dport = 0; |
1da177e4 LT |
3002 | |
3003 | if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) | |
3004 | inet_reset_saddr(sk); | |
3005 | ||
3006 | sk->sk_shutdown = 0; | |
3007 | sock_reset_flag(sk, SOCK_DONE); | |
740b0f18 | 3008 | tp->srtt_us = 0; |
b9e2e689 | 3009 | tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT); |
3f6c65d6 | 3010 | tp->rcv_rtt_last_tsecr = 0; |
0f317464 ED |
3011 | |
3012 | seq = tp->write_seq + tp->max_window + 2; | |
3013 | if (!seq) | |
3014 | seq = 1; | |
3015 | WRITE_ONCE(tp->write_seq, seq); | |
3016 | ||
463c84b9 | 3017 | icsk->icsk_backoff = 0; |
6687e988 | 3018 | icsk->icsk_probes_out = 0; |
9d9b1ee0 | 3019 | icsk->icsk_probes_tstamp = 0; |
6a408147 | 3020 | icsk->icsk_rto = TCP_TIMEOUT_INIT; |
ca584ba0 | 3021 | icsk->icsk_rto_min = TCP_RTO_MIN; |
2b8ee4f0 | 3022 | icsk->icsk_delack_max = TCP_DELACK_MAX; |
0b6a05c1 | 3023 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; |
0d24ea1d | 3024 | tcp_snd_cwnd_set(tp, TCP_INIT_CWND); |
1da177e4 | 3025 | tp->snd_cwnd_cnt = 0; |
1fdf475a | 3026 | tp->window_clamp = 0; |
2fbdd562 | 3027 | tp->delivered = 0; |
e21db6f6 | 3028 | tp->delivered_ce = 0; |
ce69e563 CP |
3029 | if (icsk->icsk_ca_ops->release) |
3030 | icsk->icsk_ca_ops->release(sk); | |
3031 | memset(icsk->icsk_ca_priv, 0, sizeof(icsk->icsk_ca_priv)); | |
8919a9b3 | 3032 | icsk->icsk_ca_initialized = 0; |
6687e988 | 3033 | tcp_set_ca_state(sk, TCP_CA_Open); |
d4761754 | 3034 | tp->is_sack_reneg = 0; |
1da177e4 | 3035 | tcp_clear_retrans(tp); |
c13c48c0 | 3036 | tp->total_retrans = 0; |
463c84b9 | 3037 | inet_csk_delack_init(sk); |
499350a5 WW |
3038 | /* Initialize rcv_mss to TCP_MIN_MSS to avoid division by 0 |
3039 | * issue in __tcp_select_window() | |
3040 | */ | |
3041 | icsk->icsk_ack.rcv_mss = TCP_MIN_MSS; | |
b40b4f79 | 3042 | memset(&tp->rx_opt, 0, sizeof(tp->rx_opt)); |
1da177e4 | 3043 | __sk_dst_reset(sk); |
a63fd42e | 3044 | dst_release(xchg((__force struct dst_entry **)&sk->sk_rx_dst, NULL)); |
17c3060b | 3045 | tcp_saved_syn_free(tp); |
5d9f4262 | 3046 | tp->compressed_ack = 0; |
784f8344 ED |
3047 | tp->segs_in = 0; |
3048 | tp->segs_out = 0; | |
ba113c3a | 3049 | tp->bytes_sent = 0; |
e858faf5 CP |
3050 | tp->bytes_acked = 0; |
3051 | tp->bytes_received = 0; | |
fb31c9b9 | 3052 | tp->bytes_retrans = 0; |
db7ffee6 ED |
3053 | tp->data_segs_in = 0; |
3054 | tp->data_segs_out = 0; | |
7788174e YC |
3055 | tp->duplicate_sack[0].start_seq = 0; |
3056 | tp->duplicate_sack[0].end_seq = 0; | |
7e10b655 | 3057 | tp->dsack_dups = 0; |
7ec65372 | 3058 | tp->reord_seen = 0; |
5c701549 ED |
3059 | tp->retrans_out = 0; |
3060 | tp->sacked_out = 0; | |
3061 | tp->tlp_high_seq = 0; | |
3062 | tp->last_oow_ack_time = 0; | |
6cda8b74 ED |
3063 | /* There's a bubble in the pipe until at least the first ACK. */ |
3064 | tp->app_limited = ~0U; | |
792c4354 ED |
3065 | tp->rack.mstamp = 0; |
3066 | tp->rack.advanced = 0; | |
3067 | tp->rack.reo_wnd_steps = 1; | |
3068 | tp->rack.last_delivered = 0; | |
3069 | tp->rack.reo_wnd_persist = 0; | |
3070 | tp->rack.dsack_seen = 0; | |
6bcdc40d ED |
3071 | tp->syn_data_acked = 0; |
3072 | tp->rx_opt.saw_tstamp = 0; | |
3073 | tp->rx_opt.dsack = 0; | |
3074 | tp->rx_opt.num_sacks = 0; | |
f9af2dbb | 3075 | tp->rcv_ooopack = 0; |
6cda8b74 | 3076 | |
1da177e4 | 3077 | |
7db92362 WW |
3078 | /* Clean up fastopen related fields */ |
3079 | tcp_free_fastopen_req(tp); | |
3080 | inet->defer_connect = 0; | |
48027478 | 3081 | tp->fastopen_client_fail = 0; |
7db92362 | 3082 | |
c720c7e8 | 3083 | WARN_ON(inet->inet_num && !icsk->icsk_bind_hash); |
1da177e4 | 3084 | |
9b42d55a LR |
3085 | if (sk->sk_frag.page) { |
3086 | put_page(sk->sk_frag.page); | |
3087 | sk->sk_frag.page = NULL; | |
3088 | sk->sk_frag.offset = 0; | |
3089 | } | |
3090 | ||
e3ae2365 | 3091 | sk_error_report(sk); |
a01512b1 | 3092 | return 0; |
1da177e4 | 3093 | } |
4bc2f18b | 3094 | EXPORT_SYMBOL(tcp_disconnect); |
1da177e4 | 3095 | |
a2a385d6 | 3096 | static inline bool tcp_can_repair_sock(const struct sock *sk) |
ee995283 | 3097 | { |
52e804c6 | 3098 | return ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) && |
319b0534 | 3099 | (sk->sk_state != TCP_LISTEN); |
ee995283 PE |
3100 | } |
3101 | ||
d38d2b00 | 3102 | static int tcp_repair_set_window(struct tcp_sock *tp, sockptr_t optbuf, int len) |
b1ed4c4f AV |
3103 | { |
3104 | struct tcp_repair_window opt; | |
3105 | ||
3106 | if (!tp->repair) | |
3107 | return -EPERM; | |
3108 | ||
3109 | if (len != sizeof(opt)) | |
3110 | return -EINVAL; | |
3111 | ||
d38d2b00 | 3112 | if (copy_from_sockptr(&opt, optbuf, sizeof(opt))) |
b1ed4c4f AV |
3113 | return -EFAULT; |
3114 | ||
3115 | if (opt.max_window < opt.snd_wnd) | |
3116 | return -EINVAL; | |
3117 | ||
3118 | if (after(opt.snd_wl1, tp->rcv_nxt + opt.rcv_wnd)) | |
3119 | return -EINVAL; | |
3120 | ||
3121 | if (after(opt.rcv_wup, tp->rcv_nxt)) | |
3122 | return -EINVAL; | |
3123 | ||
3124 | tp->snd_wl1 = opt.snd_wl1; | |
3125 | tp->snd_wnd = opt.snd_wnd; | |
3126 | tp->max_window = opt.max_window; | |
3127 | ||
3128 | tp->rcv_wnd = opt.rcv_wnd; | |
3129 | tp->rcv_wup = opt.rcv_wup; | |
3130 | ||
3131 | return 0; | |
3132 | } | |
3133 | ||
d38d2b00 CH |
3134 | static int tcp_repair_options_est(struct sock *sk, sockptr_t optbuf, |
3135 | unsigned int len) | |
b139ba4e | 3136 | { |
15e56515 | 3137 | struct tcp_sock *tp = tcp_sk(sk); |
de248a75 | 3138 | struct tcp_repair_opt opt; |
d3c48151 | 3139 | size_t offset = 0; |
b139ba4e | 3140 | |
de248a75 | 3141 | while (len >= sizeof(opt)) { |
d3c48151 | 3142 | if (copy_from_sockptr_offset(&opt, optbuf, offset, sizeof(opt))) |
b139ba4e PE |
3143 | return -EFAULT; |
3144 | ||
d3c48151 | 3145 | offset += sizeof(opt); |
de248a75 | 3146 | len -= sizeof(opt); |
b139ba4e | 3147 | |
de248a75 PE |
3148 | switch (opt.opt_code) { |
3149 | case TCPOPT_MSS: | |
3150 | tp->rx_opt.mss_clamp = opt.opt_val; | |
15e56515 | 3151 | tcp_mtup_init(sk); |
b139ba4e | 3152 | break; |
de248a75 | 3153 | case TCPOPT_WINDOW: |
bc26ccd8 AV |
3154 | { |
3155 | u16 snd_wscale = opt.opt_val & 0xFFFF; | |
3156 | u16 rcv_wscale = opt.opt_val >> 16; | |
3157 | ||
589c49cb | 3158 | if (snd_wscale > TCP_MAX_WSCALE || rcv_wscale > TCP_MAX_WSCALE) |
bc26ccd8 | 3159 | return -EFBIG; |
b139ba4e | 3160 | |
bc26ccd8 AV |
3161 | tp->rx_opt.snd_wscale = snd_wscale; |
3162 | tp->rx_opt.rcv_wscale = rcv_wscale; | |
3163 | tp->rx_opt.wscale_ok = 1; | |
3164 | } | |
b139ba4e | 3165 | break; |
b139ba4e | 3166 | case TCPOPT_SACK_PERM: |
de248a75 PE |
3167 | if (opt.opt_val != 0) |
3168 | return -EINVAL; | |
3169 | ||
b139ba4e | 3170 | tp->rx_opt.sack_ok |= TCP_SACK_SEEN; |
b139ba4e PE |
3171 | break; |
3172 | case TCPOPT_TIMESTAMP: | |
de248a75 PE |
3173 | if (opt.opt_val != 0) |
3174 | return -EINVAL; | |
3175 | ||
b139ba4e PE |
3176 | tp->rx_opt.tstamp_ok = 1; |
3177 | break; | |
3178 | } | |
3179 | } | |
3180 | ||
3181 | return 0; | |
3182 | } | |
3183 | ||
a842fe14 ED |
3184 | DEFINE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); |
3185 | EXPORT_SYMBOL(tcp_tx_delay_enabled); | |
3186 | ||
3187 | static void tcp_enable_tx_delay(void) | |
3188 | { | |
3189 | if (!static_branch_unlikely(&tcp_tx_delay_enabled)) { | |
3190 | static int __tcp_tx_delay_enabled = 0; | |
3191 | ||
3192 | if (cmpxchg(&__tcp_tx_delay_enabled, 0, 1) == 0) { | |
3193 | static_branch_enable(&tcp_tx_delay_enabled); | |
3194 | pr_info("TCP_TX_DELAY enabled\n"); | |
3195 | } | |
3196 | } | |
3197 | } | |
3198 | ||
db10538a CH |
3199 | /* When set indicates to always queue non-full frames. Later the user clears |
3200 | * this option and we transmit any pending partial frames in the queue. This is | |
3201 | * meant to be used alongside sendfile() to get properly filled frames when the | |
3202 | * user (for example) must write out headers with a write() call first and then | |
3203 | * use sendfile to send out the data parts. | |
3204 | * | |
3205 | * TCP_CORK can be set together with TCP_NODELAY and it is stronger than | |
3206 | * TCP_NODELAY. | |
3207 | */ | |
3208 | static void __tcp_sock_set_cork(struct sock *sk, bool on) | |
3209 | { | |
3210 | struct tcp_sock *tp = tcp_sk(sk); | |
3211 | ||
3212 | if (on) { | |
3213 | tp->nonagle |= TCP_NAGLE_CORK; | |
3214 | } else { | |
3215 | tp->nonagle &= ~TCP_NAGLE_CORK; | |
3216 | if (tp->nonagle & TCP_NAGLE_OFF) | |
3217 | tp->nonagle |= TCP_NAGLE_PUSH; | |
3218 | tcp_push_pending_frames(sk); | |
3219 | } | |
3220 | } | |
3221 | ||
3222 | void tcp_sock_set_cork(struct sock *sk, bool on) | |
3223 | { | |
3224 | lock_sock(sk); | |
3225 | __tcp_sock_set_cork(sk, on); | |
3226 | release_sock(sk); | |
3227 | } | |
3228 | EXPORT_SYMBOL(tcp_sock_set_cork); | |
3229 | ||
12abc5ee CH |
3230 | /* TCP_NODELAY is weaker than TCP_CORK, so that this option on corked socket is |
3231 | * remembered, but it is not activated until cork is cleared. | |
3232 | * | |
3233 | * However, when TCP_NODELAY is set we make an explicit push, which overrides | |
3234 | * even TCP_CORK for currently queued segments. | |
3235 | */ | |
3236 | static void __tcp_sock_set_nodelay(struct sock *sk, bool on) | |
3237 | { | |
3238 | if (on) { | |
3239 | tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH; | |
3240 | tcp_push_pending_frames(sk); | |
3241 | } else { | |
3242 | tcp_sk(sk)->nonagle &= ~TCP_NAGLE_OFF; | |
3243 | } | |
3244 | } | |
3245 | ||
3246 | void tcp_sock_set_nodelay(struct sock *sk) | |
3247 | { | |
3248 | lock_sock(sk); | |
3249 | __tcp_sock_set_nodelay(sk, true); | |
3250 | release_sock(sk); | |
3251 | } | |
3252 | EXPORT_SYMBOL(tcp_sock_set_nodelay); | |
3253 | ||
ddd061b8 CH |
3254 | static void __tcp_sock_set_quickack(struct sock *sk, int val) |
3255 | { | |
3256 | if (!val) { | |
3257 | inet_csk_enter_pingpong_mode(sk); | |
3258 | return; | |
3259 | } | |
3260 | ||
3261 | inet_csk_exit_pingpong_mode(sk); | |
3262 | if ((1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) && | |
3263 | inet_csk_ack_scheduled(sk)) { | |
3264 | inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_PUSHED; | |
3265 | tcp_cleanup_rbuf(sk, 1); | |
3266 | if (!(val & 1)) | |
3267 | inet_csk_enter_pingpong_mode(sk); | |
3268 | } | |
3269 | } | |
3270 | ||
3271 | void tcp_sock_set_quickack(struct sock *sk, int val) | |
3272 | { | |
3273 | lock_sock(sk); | |
3274 | __tcp_sock_set_quickack(sk, val); | |
3275 | release_sock(sk); | |
3276 | } | |
3277 | EXPORT_SYMBOL(tcp_sock_set_quickack); | |
3278 | ||
557eadfc CH |
3279 | int tcp_sock_set_syncnt(struct sock *sk, int val) |
3280 | { | |
3281 | if (val < 1 || val > MAX_TCP_SYNCNT) | |
3282 | return -EINVAL; | |
3283 | ||
3284 | lock_sock(sk); | |
3285 | inet_csk(sk)->icsk_syn_retries = val; | |
3286 | release_sock(sk); | |
3287 | return 0; | |
3288 | } | |
3289 | EXPORT_SYMBOL(tcp_sock_set_syncnt); | |
3290 | ||
c488aead CH |
3291 | void tcp_sock_set_user_timeout(struct sock *sk, u32 val) |
3292 | { | |
3293 | lock_sock(sk); | |
3294 | inet_csk(sk)->icsk_user_timeout = val; | |
3295 | release_sock(sk); | |
3296 | } | |
3297 | EXPORT_SYMBOL(tcp_sock_set_user_timeout); | |
3298 | ||
aad4a0a9 | 3299 | int tcp_sock_set_keepidle_locked(struct sock *sk, int val) |
71c48eb8 CH |
3300 | { |
3301 | struct tcp_sock *tp = tcp_sk(sk); | |
3302 | ||
3303 | if (val < 1 || val > MAX_TCP_KEEPIDLE) | |
3304 | return -EINVAL; | |
3305 | ||
3306 | tp->keepalive_time = val * HZ; | |
3307 | if (sock_flag(sk, SOCK_KEEPOPEN) && | |
3308 | !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) { | |
3309 | u32 elapsed = keepalive_time_elapsed(tp); | |
3310 | ||
3311 | if (tp->keepalive_time > elapsed) | |
3312 | elapsed = tp->keepalive_time - elapsed; | |
3313 | else | |
3314 | elapsed = 0; | |
3315 | inet_csk_reset_keepalive_timer(sk, elapsed); | |
3316 | } | |
3317 | ||
3318 | return 0; | |
3319 | } | |
3320 | ||
3321 | int tcp_sock_set_keepidle(struct sock *sk, int val) | |
3322 | { | |
3323 | int err; | |
3324 | ||
3325 | lock_sock(sk); | |
aad4a0a9 | 3326 | err = tcp_sock_set_keepidle_locked(sk, val); |
71c48eb8 CH |
3327 | release_sock(sk); |
3328 | return err; | |
3329 | } | |
3330 | EXPORT_SYMBOL(tcp_sock_set_keepidle); | |
3331 | ||
d41ecaac CH |
3332 | int tcp_sock_set_keepintvl(struct sock *sk, int val) |
3333 | { | |
3334 | if (val < 1 || val > MAX_TCP_KEEPINTVL) | |
3335 | return -EINVAL; | |
3336 | ||
3337 | lock_sock(sk); | |
3338 | tcp_sk(sk)->keepalive_intvl = val * HZ; | |
3339 | release_sock(sk); | |
3340 | return 0; | |
3341 | } | |
3342 | EXPORT_SYMBOL(tcp_sock_set_keepintvl); | |
3343 | ||
480aeb96 CH |
3344 | int tcp_sock_set_keepcnt(struct sock *sk, int val) |
3345 | { | |
3346 | if (val < 1 || val > MAX_TCP_KEEPCNT) | |
3347 | return -EINVAL; | |
3348 | ||
3349 | lock_sock(sk); | |
3350 | tcp_sk(sk)->keepalive_probes = val; | |
3351 | release_sock(sk); | |
3352 | return 0; | |
3353 | } | |
3354 | EXPORT_SYMBOL(tcp_sock_set_keepcnt); | |
3355 | ||
cb811109 P |
3356 | int tcp_set_window_clamp(struct sock *sk, int val) |
3357 | { | |
3358 | struct tcp_sock *tp = tcp_sk(sk); | |
3359 | ||
3360 | if (!val) { | |
3361 | if (sk->sk_state != TCP_CLOSE) | |
3362 | return -EINVAL; | |
3363 | tp->window_clamp = 0; | |
3364 | } else { | |
3365 | tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ? | |
3366 | SOCK_MIN_RCVBUF / 2 : val; | |
3aa7857f | 3367 | tp->rcv_ssthresh = min(tp->rcv_wnd, tp->window_clamp); |
cb811109 P |
3368 | } |
3369 | return 0; | |
3370 | } | |
3371 | ||
1da177e4 LT |
3372 | /* |
3373 | * Socket option code for TCP. | |
3374 | */ | |
d38d2b00 CH |
3375 | static int do_tcp_setsockopt(struct sock *sk, int level, int optname, |
3376 | sockptr_t optval, unsigned int optlen) | |
1da177e4 LT |
3377 | { |
3378 | struct tcp_sock *tp = tcp_sk(sk); | |
463c84b9 | 3379 | struct inet_connection_sock *icsk = inet_csk(sk); |
1e579caa | 3380 | struct net *net = sock_net(sk); |
1da177e4 LT |
3381 | int val; |
3382 | int err = 0; | |
3383 | ||
e56fb50f WAS |
3384 | /* These are data/string values, all the others are ints */ |
3385 | switch (optname) { | |
3386 | case TCP_CONGESTION: { | |
5f8ef48d SH |
3387 | char name[TCP_CA_NAME_MAX]; |
3388 | ||
3389 | if (optlen < 1) | |
3390 | return -EINVAL; | |
3391 | ||
d38d2b00 | 3392 | val = strncpy_from_sockptr(name, optval, |
4fdb78d3 | 3393 | min_t(long, TCP_CA_NAME_MAX-1, optlen)); |
5f8ef48d SH |
3394 | if (val < 0) |
3395 | return -EFAULT; | |
3396 | name[val] = 0; | |
3397 | ||
3398 | lock_sock(sk); | |
29a94932 | 3399 | err = tcp_set_congestion_control(sk, name, true, |
8d650cde ED |
3400 | ns_capable(sock_net(sk)->user_ns, |
3401 | CAP_NET_ADMIN)); | |
5f8ef48d SH |
3402 | release_sock(sk); |
3403 | return err; | |
3404 | } | |
734942cc DW |
3405 | case TCP_ULP: { |
3406 | char name[TCP_ULP_NAME_MAX]; | |
3407 | ||
3408 | if (optlen < 1) | |
3409 | return -EINVAL; | |
3410 | ||
d38d2b00 | 3411 | val = strncpy_from_sockptr(name, optval, |
734942cc DW |
3412 | min_t(long, TCP_ULP_NAME_MAX - 1, |
3413 | optlen)); | |
3414 | if (val < 0) | |
3415 | return -EFAULT; | |
3416 | name[val] = 0; | |
3417 | ||
3418 | lock_sock(sk); | |
3419 | err = tcp_set_ulp(sk, name); | |
3420 | release_sock(sk); | |
3421 | return err; | |
3422 | } | |
1fba70e5 | 3423 | case TCP_FASTOPEN_KEY: { |
0f1ce023 JB |
3424 | __u8 key[TCP_FASTOPEN_KEY_BUF_LENGTH]; |
3425 | __u8 *backup_key = NULL; | |
1fba70e5 | 3426 | |
0f1ce023 JB |
3427 | /* Allow a backup key as well to facilitate key rotation |
3428 | * First key is the active one. | |
3429 | */ | |
3430 | if (optlen != TCP_FASTOPEN_KEY_LENGTH && | |
3431 | optlen != TCP_FASTOPEN_KEY_BUF_LENGTH) | |
1fba70e5 YC |
3432 | return -EINVAL; |
3433 | ||
d38d2b00 | 3434 | if (copy_from_sockptr(key, optval, optlen)) |
1fba70e5 YC |
3435 | return -EFAULT; |
3436 | ||
0f1ce023 JB |
3437 | if (optlen == TCP_FASTOPEN_KEY_BUF_LENGTH) |
3438 | backup_key = key + TCP_FASTOPEN_KEY_LENGTH; | |
3439 | ||
438ac880 | 3440 | return tcp_fastopen_reset_cipher(net, sk, key, backup_key); |
1fba70e5 | 3441 | } |
e56fb50f WAS |
3442 | default: |
3443 | /* fallthru */ | |
3444 | break; | |
ccbd6a5a | 3445 | } |
5f8ef48d | 3446 | |
1da177e4 LT |
3447 | if (optlen < sizeof(int)) |
3448 | return -EINVAL; | |
3449 | ||
d38d2b00 | 3450 | if (copy_from_sockptr(&val, optval, sizeof(val))) |
1da177e4 LT |
3451 | return -EFAULT; |
3452 | ||
3453 | lock_sock(sk); | |
3454 | ||
3455 | switch (optname) { | |
3456 | case TCP_MAXSEG: | |
3457 | /* Values greater than interface MTU won't take effect. However | |
3458 | * at the point when this call is done we typically don't yet | |
a777f715 RC |
3459 | * know which interface is going to be used |
3460 | */ | |
cfc62d87 | 3461 | if (val && (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW)) { |
1da177e4 LT |
3462 | err = -EINVAL; |
3463 | break; | |
3464 | } | |
3465 | tp->rx_opt.user_mss = val; | |
3466 | break; | |
3467 | ||
3468 | case TCP_NODELAY: | |
12abc5ee | 3469 | __tcp_sock_set_nodelay(sk, val); |
1da177e4 LT |
3470 | break; |
3471 | ||
36e31b0a AP |
3472 | case TCP_THIN_LINEAR_TIMEOUTS: |
3473 | if (val < 0 || val > 1) | |
3474 | err = -EINVAL; | |
3475 | else | |
3476 | tp->thin_lto = val; | |
3477 | break; | |
3478 | ||
7e380175 AP |
3479 | case TCP_THIN_DUPACK: |
3480 | if (val < 0 || val > 1) | |
3481 | err = -EINVAL; | |
7e380175 AP |
3482 | break; |
3483 | ||
ee995283 PE |
3484 | case TCP_REPAIR: |
3485 | if (!tcp_can_repair_sock(sk)) | |
3486 | err = -EPERM; | |
31048d7a | 3487 | else if (val == TCP_REPAIR_ON) { |
ee995283 PE |
3488 | tp->repair = 1; |
3489 | sk->sk_reuse = SK_FORCE_REUSE; | |
3490 | tp->repair_queue = TCP_NO_QUEUE; | |
31048d7a | 3491 | } else if (val == TCP_REPAIR_OFF) { |
ee995283 PE |
3492 | tp->repair = 0; |
3493 | sk->sk_reuse = SK_NO_REUSE; | |
3494 | tcp_send_window_probe(sk); | |
31048d7a | 3495 | } else if (val == TCP_REPAIR_OFF_NO_WP) { |
ee995283 PE |
3496 | tp->repair = 0; |
3497 | sk->sk_reuse = SK_NO_REUSE; | |
ee995283 PE |
3498 | } else |
3499 | err = -EINVAL; | |
3500 | ||
3501 | break; | |
3502 | ||
3503 | case TCP_REPAIR_QUEUE: | |
3504 | if (!tp->repair) | |
3505 | err = -EPERM; | |
bf2acc94 | 3506 | else if ((unsigned int)val < TCP_QUEUES_NR) |
ee995283 PE |
3507 | tp->repair_queue = val; |
3508 | else | |
3509 | err = -EINVAL; | |
3510 | break; | |
3511 | ||
3512 | case TCP_QUEUE_SEQ: | |
8811f4a9 | 3513 | if (sk->sk_state != TCP_CLOSE) { |
ee995283 | 3514 | err = -EPERM; |
8811f4a9 ED |
3515 | } else if (tp->repair_queue == TCP_SEND_QUEUE) { |
3516 | if (!tcp_rtx_queue_empty(sk)) | |
3517 | err = -EPERM; | |
3518 | else | |
3519 | WRITE_ONCE(tp->write_seq, val); | |
3520 | } else if (tp->repair_queue == TCP_RECV_QUEUE) { | |
3521 | if (tp->rcv_nxt != tp->copied_seq) { | |
3522 | err = -EPERM; | |
3523 | } else { | |
3524 | WRITE_ONCE(tp->rcv_nxt, val); | |
3525 | WRITE_ONCE(tp->copied_seq, val); | |
3526 | } | |
3527 | } else { | |
ee995283 | 3528 | err = -EINVAL; |
8811f4a9 | 3529 | } |
ee995283 PE |
3530 | break; |
3531 | ||
b139ba4e PE |
3532 | case TCP_REPAIR_OPTIONS: |
3533 | if (!tp->repair) | |
3534 | err = -EINVAL; | |
3535 | else if (sk->sk_state == TCP_ESTABLISHED) | |
d38d2b00 | 3536 | err = tcp_repair_options_est(sk, optval, optlen); |
b139ba4e PE |
3537 | else |
3538 | err = -EPERM; | |
3539 | break; | |
3540 | ||
1da177e4 | 3541 | case TCP_CORK: |
db10538a | 3542 | __tcp_sock_set_cork(sk, val); |
1da177e4 LT |
3543 | break; |
3544 | ||
3545 | case TCP_KEEPIDLE: | |
aad4a0a9 | 3546 | err = tcp_sock_set_keepidle_locked(sk, val); |
1da177e4 LT |
3547 | break; |
3548 | case TCP_KEEPINTVL: | |
3549 | if (val < 1 || val > MAX_TCP_KEEPINTVL) | |
3550 | err = -EINVAL; | |
3551 | else | |
3552 | tp->keepalive_intvl = val * HZ; | |
3553 | break; | |
3554 | case TCP_KEEPCNT: | |
3555 | if (val < 1 || val > MAX_TCP_KEEPCNT) | |
3556 | err = -EINVAL; | |
3557 | else | |
3558 | tp->keepalive_probes = val; | |
3559 | break; | |
3560 | case TCP_SYNCNT: | |
3561 | if (val < 1 || val > MAX_TCP_SYNCNT) | |
3562 | err = -EINVAL; | |
3563 | else | |
463c84b9 | 3564 | icsk->icsk_syn_retries = val; |
1da177e4 LT |
3565 | break; |
3566 | ||
cd8ae852 | 3567 | case TCP_SAVE_SYN: |
267cf9fa MKL |
3568 | /* 0: disable, 1: enable, 2: start from ether_header */ |
3569 | if (val < 0 || val > 2) | |
cd8ae852 ED |
3570 | err = -EINVAL; |
3571 | else | |
3572 | tp->save_syn = val; | |
3573 | break; | |
3574 | ||
1da177e4 LT |
3575 | case TCP_LINGER2: |
3576 | if (val < 0) | |
3577 | tp->linger2 = -1; | |
f0628c52 CZ |
3578 | else if (val > TCP_FIN_TIMEOUT_MAX / HZ) |
3579 | tp->linger2 = TCP_FIN_TIMEOUT_MAX; | |
1da177e4 LT |
3580 | else |
3581 | tp->linger2 = val * HZ; | |
3582 | break; | |
3583 | ||
3584 | case TCP_DEFER_ACCEPT: | |
b103cf34 JA |
3585 | /* Translate value in seconds to number of retransmits */ |
3586 | icsk->icsk_accept_queue.rskq_defer_accept = | |
3587 | secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ, | |
3588 | TCP_RTO_MAX / HZ); | |
1da177e4 LT |
3589 | break; |
3590 | ||
3591 | case TCP_WINDOW_CLAMP: | |
cb811109 | 3592 | err = tcp_set_window_clamp(sk, val); |
1da177e4 LT |
3593 | break; |
3594 | ||
3595 | case TCP_QUICKACK: | |
ddd061b8 | 3596 | __tcp_sock_set_quickack(sk, val); |
1da177e4 LT |
3597 | break; |
3598 | ||
cfb6eeb4 YH |
3599 | #ifdef CONFIG_TCP_MD5SIG |
3600 | case TCP_MD5SIG: | |
8917a777 | 3601 | case TCP_MD5SIG_EXT: |
d38d2b00 | 3602 | err = tp->af_specific->md5_parse(sk, optname, optval, optlen); |
cfb6eeb4 YH |
3603 | break; |
3604 | #endif | |
dca43c75 | 3605 | case TCP_USER_TIMEOUT: |
b248230c | 3606 | /* Cap the max time in ms TCP will retry or probe the window |
dca43c75 JC |
3607 | * before giving up and aborting (ETIMEDOUT) a connection. |
3608 | */ | |
42493570 HL |
3609 | if (val < 0) |
3610 | err = -EINVAL; | |
3611 | else | |
9bcc66e1 | 3612 | icsk->icsk_user_timeout = val; |
dca43c75 | 3613 | break; |
8336886f JC |
3614 | |
3615 | case TCP_FASTOPEN: | |
3616 | if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE | | |
dfea2aa6 | 3617 | TCPF_LISTEN))) { |
43713848 | 3618 | tcp_fastopen_init_key_once(net); |
dfea2aa6 | 3619 | |
0536fcc0 | 3620 | fastopen_queue_tune(sk, val); |
dfea2aa6 | 3621 | } else { |
8336886f | 3622 | err = -EINVAL; |
dfea2aa6 | 3623 | } |
8336886f | 3624 | break; |
19f6d3f3 WW |
3625 | case TCP_FASTOPEN_CONNECT: |
3626 | if (val > 1 || val < 0) { | |
3627 | err = -EINVAL; | |
e1cfcbe8 | 3628 | } else if (net->ipv4.sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) { |
19f6d3f3 WW |
3629 | if (sk->sk_state == TCP_CLOSE) |
3630 | tp->fastopen_connect = val; | |
3631 | else | |
3632 | err = -EINVAL; | |
3633 | } else { | |
3634 | err = -EOPNOTSUPP; | |
3635 | } | |
3636 | break; | |
71c02379 CP |
3637 | case TCP_FASTOPEN_NO_COOKIE: |
3638 | if (val > 1 || val < 0) | |
3639 | err = -EINVAL; | |
3640 | else if (!((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) | |
3641 | err = -EINVAL; | |
3642 | else | |
3643 | tp->fastopen_no_cookie = val; | |
3644 | break; | |
93be6ce0 AV |
3645 | case TCP_TIMESTAMP: |
3646 | if (!tp->repair) | |
3647 | err = -EPERM; | |
3648 | else | |
9a568de4 | 3649 | tp->tsoffset = val - tcp_time_stamp_raw(); |
93be6ce0 | 3650 | break; |
b1ed4c4f AV |
3651 | case TCP_REPAIR_WINDOW: |
3652 | err = tcp_repair_set_window(tp, optval, optlen); | |
3653 | break; | |
c9bee3b7 ED |
3654 | case TCP_NOTSENT_LOWAT: |
3655 | tp->notsent_lowat = val; | |
3656 | sk->sk_write_space(sk); | |
3657 | break; | |
b75eba76 SHY |
3658 | case TCP_INQ: |
3659 | if (val > 1 || val < 0) | |
3660 | err = -EINVAL; | |
3661 | else | |
3662 | tp->recvmsg_inq = val; | |
3663 | break; | |
a842fe14 ED |
3664 | case TCP_TX_DELAY: |
3665 | if (val) | |
3666 | tcp_enable_tx_delay(); | |
3667 | tp->tcp_tx_delay = val; | |
3668 | break; | |
1da177e4 LT |
3669 | default: |
3670 | err = -ENOPROTOOPT; | |
3671 | break; | |
3ff50b79 SH |
3672 | } |
3673 | ||
1da177e4 LT |
3674 | release_sock(sk); |
3675 | return err; | |
3676 | } | |
3677 | ||
a7b75c5a | 3678 | int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, |
b7058842 | 3679 | unsigned int optlen) |
3fdadf7d | 3680 | { |
cf533ea5 | 3681 | const struct inet_connection_sock *icsk = inet_csk(sk); |
3fdadf7d DM |
3682 | |
3683 | if (level != SOL_TCP) | |
3684 | return icsk->icsk_af_ops->setsockopt(sk, level, optname, | |
3685 | optval, optlen); | |
a7b75c5a | 3686 | return do_tcp_setsockopt(sk, level, optname, optval, optlen); |
3fdadf7d | 3687 | } |
4bc2f18b | 3688 | EXPORT_SYMBOL(tcp_setsockopt); |
3fdadf7d | 3689 | |
efd90174 FY |
3690 | static void tcp_get_info_chrono_stats(const struct tcp_sock *tp, |
3691 | struct tcp_info *info) | |
3692 | { | |
3693 | u64 stats[__TCP_CHRONO_MAX], total = 0; | |
3694 | enum tcp_chrono i; | |
3695 | ||
3696 | for (i = TCP_CHRONO_BUSY; i < __TCP_CHRONO_MAX; ++i) { | |
3697 | stats[i] = tp->chrono_stat[i - 1]; | |
3698 | if (i == tp->chrono_type) | |
628174cc | 3699 | stats[i] += tcp_jiffies32 - tp->chrono_start; |
efd90174 FY |
3700 | stats[i] *= USEC_PER_SEC / HZ; |
3701 | total += stats[i]; | |
3702 | } | |
3703 | ||
3704 | info->tcpi_busy_time = total; | |
3705 | info->tcpi_rwnd_limited = stats[TCP_CHRONO_RWND_LIMITED]; | |
3706 | info->tcpi_sndbuf_limited = stats[TCP_CHRONO_SNDBUF_LIMITED]; | |
3707 | } | |
3708 | ||
1da177e4 | 3709 | /* Return information about state of tcp endpoint in API format. */ |
0df48c26 | 3710 | void tcp_get_info(struct sock *sk, struct tcp_info *info) |
1da177e4 | 3711 | { |
35ac838a | 3712 | const struct tcp_sock *tp = tcp_sk(sk); /* iff sk_type == SOCK_STREAM */ |
463c84b9 | 3713 | const struct inet_connection_sock *icsk = inet_csk(sk); |
76a9ebe8 | 3714 | unsigned long rate; |
0263598c | 3715 | u32 now; |
ff5d7497 | 3716 | u64 rate64; |
67db3e4b | 3717 | bool slow; |
1da177e4 LT |
3718 | |
3719 | memset(info, 0, sizeof(*info)); | |
35ac838a CG |
3720 | if (sk->sk_type != SOCK_STREAM) |
3721 | return; | |
1da177e4 | 3722 | |
986ffdfd | 3723 | info->tcpi_state = inet_sk_state_load(sk); |
00fd38d9 | 3724 | |
ccbf3bfa ED |
3725 | /* Report meaningful fields for all TCP states, including listeners */ |
3726 | rate = READ_ONCE(sk->sk_pacing_rate); | |
76a9ebe8 | 3727 | rate64 = (rate != ~0UL) ? rate : ~0ULL; |
f522a5fc | 3728 | info->tcpi_pacing_rate = rate64; |
ccbf3bfa ED |
3729 | |
3730 | rate = READ_ONCE(sk->sk_max_pacing_rate); | |
76a9ebe8 | 3731 | rate64 = (rate != ~0UL) ? rate : ~0ULL; |
f522a5fc | 3732 | info->tcpi_max_pacing_rate = rate64; |
ccbf3bfa ED |
3733 | |
3734 | info->tcpi_reordering = tp->reordering; | |
0d24ea1d | 3735 | info->tcpi_snd_cwnd = tcp_snd_cwnd(tp); |
ccbf3bfa ED |
3736 | |
3737 | if (info->tcpi_state == TCP_LISTEN) { | |
3738 | /* listeners aliased fields : | |
3739 | * tcpi_unacked -> Number of children ready for accept() | |
3740 | * tcpi_sacked -> max backlog | |
3741 | */ | |
288efe86 | 3742 | info->tcpi_unacked = READ_ONCE(sk->sk_ack_backlog); |
099ecf59 | 3743 | info->tcpi_sacked = READ_ONCE(sk->sk_max_ack_backlog); |
ccbf3bfa ED |
3744 | return; |
3745 | } | |
b369e7fd ED |
3746 | |
3747 | slow = lock_sock_fast(sk); | |
3748 | ||
6687e988 | 3749 | info->tcpi_ca_state = icsk->icsk_ca_state; |
463c84b9 | 3750 | info->tcpi_retransmits = icsk->icsk_retransmits; |
6687e988 | 3751 | info->tcpi_probes = icsk->icsk_probes_out; |
463c84b9 | 3752 | info->tcpi_backoff = icsk->icsk_backoff; |
1da177e4 LT |
3753 | |
3754 | if (tp->rx_opt.tstamp_ok) | |
3755 | info->tcpi_options |= TCPI_OPT_TIMESTAMPS; | |
e60402d0 | 3756 | if (tcp_is_sack(tp)) |
1da177e4 LT |
3757 | info->tcpi_options |= TCPI_OPT_SACK; |
3758 | if (tp->rx_opt.wscale_ok) { | |
3759 | info->tcpi_options |= TCPI_OPT_WSCALE; | |
3760 | info->tcpi_snd_wscale = tp->rx_opt.snd_wscale; | |
3761 | info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale; | |
e905a9ed | 3762 | } |
1da177e4 | 3763 | |
b5c5693b | 3764 | if (tp->ecn_flags & TCP_ECN_OK) |
1da177e4 | 3765 | info->tcpi_options |= TCPI_OPT_ECN; |
b5c5693b ED |
3766 | if (tp->ecn_flags & TCP_ECN_SEEN) |
3767 | info->tcpi_options |= TCPI_OPT_ECN_SEEN; | |
6f73601e YC |
3768 | if (tp->syn_data_acked) |
3769 | info->tcpi_options |= TCPI_OPT_SYN_DATA; | |
1da177e4 | 3770 | |
463c84b9 ACM |
3771 | info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto); |
3772 | info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato); | |
c1b4a7e6 | 3773 | info->tcpi_snd_mss = tp->mss_cache; |
463c84b9 | 3774 | info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss; |
1da177e4 | 3775 | |
ccbf3bfa ED |
3776 | info->tcpi_unacked = tp->packets_out; |
3777 | info->tcpi_sacked = tp->sacked_out; | |
3778 | ||
1da177e4 LT |
3779 | info->tcpi_lost = tp->lost_out; |
3780 | info->tcpi_retrans = tp->retrans_out; | |
1da177e4 | 3781 | |
d635fbe2 | 3782 | now = tcp_jiffies32; |
1da177e4 | 3783 | info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime); |
463c84b9 | 3784 | info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime); |
1da177e4 LT |
3785 | info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp); |
3786 | ||
d83d8461 | 3787 | info->tcpi_pmtu = icsk->icsk_pmtu_cookie; |
1da177e4 | 3788 | info->tcpi_rcv_ssthresh = tp->rcv_ssthresh; |
740b0f18 ED |
3789 | info->tcpi_rtt = tp->srtt_us >> 3; |
3790 | info->tcpi_rttvar = tp->mdev_us >> 2; | |
1da177e4 | 3791 | info->tcpi_snd_ssthresh = tp->snd_ssthresh; |
1da177e4 | 3792 | info->tcpi_advmss = tp->advmss; |
1da177e4 | 3793 | |
645f4c6f | 3794 | info->tcpi_rcv_rtt = tp->rcv_rtt_est.rtt_us >> 3; |
1da177e4 LT |
3795 | info->tcpi_rcv_space = tp->rcvq_space.space; |
3796 | ||
3797 | info->tcpi_total_retrans = tp->total_retrans; | |
977cb0ec | 3798 | |
f522a5fc ED |
3799 | info->tcpi_bytes_acked = tp->bytes_acked; |
3800 | info->tcpi_bytes_received = tp->bytes_received; | |
67db3e4b | 3801 | info->tcpi_notsent_bytes = max_t(int, 0, tp->write_seq - tp->snd_nxt); |
efd90174 | 3802 | tcp_get_info_chrono_stats(tp, info); |
67db3e4b | 3803 | |
2efd055c MRL |
3804 | info->tcpi_segs_out = tp->segs_out; |
3805 | info->tcpi_segs_in = tp->segs_in; | |
cd9b2660 | 3806 | |
cd9b2660 | 3807 | info->tcpi_min_rtt = tcp_min_rtt(tp); |
a44d6eac MKL |
3808 | info->tcpi_data_segs_in = tp->data_segs_in; |
3809 | info->tcpi_data_segs_out = tp->data_segs_out; | |
eb8329e0 YC |
3810 | |
3811 | info->tcpi_delivery_rate_app_limited = tp->rate_app_limited ? 1 : 0; | |
0263598c WW |
3812 | rate64 = tcp_compute_delivery_rate(tp); |
3813 | if (rate64) | |
f522a5fc | 3814 | info->tcpi_delivery_rate = rate64; |
feb5f2ec YC |
3815 | info->tcpi_delivered = tp->delivered; |
3816 | info->tcpi_delivered_ce = tp->delivered_ce; | |
ba113c3a | 3817 | info->tcpi_bytes_sent = tp->bytes_sent; |
fb31c9b9 | 3818 | info->tcpi_bytes_retrans = tp->bytes_retrans; |
7e10b655 | 3819 | info->tcpi_dsack_dups = tp->dsack_dups; |
7ec65372 | 3820 | info->tcpi_reord_seen = tp->reord_seen; |
f9af2dbb | 3821 | info->tcpi_rcv_ooopack = tp->rcv_ooopack; |
8f7baad7 | 3822 | info->tcpi_snd_wnd = tp->snd_wnd; |
48027478 | 3823 | info->tcpi_fastopen_client_fail = tp->fastopen_client_fail; |
b369e7fd | 3824 | unlock_sock_fast(sk, slow); |
1da177e4 | 3825 | } |
1da177e4 LT |
3826 | EXPORT_SYMBOL_GPL(tcp_get_info); |
3827 | ||
984988aa WW |
3828 | static size_t tcp_opt_stats_get_size(void) |
3829 | { | |
3830 | return | |
3831 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_BUSY */ | |
3832 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_RWND_LIMITED */ | |
3833 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_SNDBUF_LIMITED */ | |
3834 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_DATA_SEGS_OUT */ | |
3835 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_TOTAL_RETRANS */ | |
3836 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_PACING_RATE */ | |
3837 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_DELIVERY_RATE */ | |
3838 | nla_total_size(sizeof(u32)) + /* TCP_NLA_SND_CWND */ | |
3839 | nla_total_size(sizeof(u32)) + /* TCP_NLA_REORDERING */ | |
3840 | nla_total_size(sizeof(u32)) + /* TCP_NLA_MIN_RTT */ | |
3841 | nla_total_size(sizeof(u8)) + /* TCP_NLA_RECUR_RETRANS */ | |
3842 | nla_total_size(sizeof(u8)) + /* TCP_NLA_DELIVERY_RATE_APP_LMT */ | |
3843 | nla_total_size(sizeof(u32)) + /* TCP_NLA_SNDQ_SIZE */ | |
3844 | nla_total_size(sizeof(u8)) + /* TCP_NLA_CA_STATE */ | |
3845 | nla_total_size(sizeof(u32)) + /* TCP_NLA_SND_SSTHRESH */ | |
3846 | nla_total_size(sizeof(u32)) + /* TCP_NLA_DELIVERED */ | |
3847 | nla_total_size(sizeof(u32)) + /* TCP_NLA_DELIVERED_CE */ | |
ba113c3a | 3848 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_BYTES_SENT */ |
fb31c9b9 | 3849 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_BYTES_RETRANS */ |
7e10b655 | 3850 | nla_total_size(sizeof(u32)) + /* TCP_NLA_DSACK_DUPS */ |
7ec65372 | 3851 | nla_total_size(sizeof(u32)) + /* TCP_NLA_REORD_SEEN */ |
e8bd8fca | 3852 | nla_total_size(sizeof(u32)) + /* TCP_NLA_SRTT */ |
32efcc06 | 3853 | nla_total_size(sizeof(u16)) + /* TCP_NLA_TIMEOUT_REHASH */ |
e08ab0b3 | 3854 | nla_total_size(sizeof(u32)) + /* TCP_NLA_BYTES_NOTSENT */ |
48040793 | 3855 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_EDT */ |
e7ed11ee | 3856 | nla_total_size(sizeof(u8)) + /* TCP_NLA_TTL */ |
984988aa WW |
3857 | 0; |
3858 | } | |
3859 | ||
e7ed11ee YS |
3860 | /* Returns TTL or hop limit of an incoming packet from skb. */ |
3861 | static u8 tcp_skb_ttl_or_hop_limit(const struct sk_buff *skb) | |
3862 | { | |
3863 | if (skb->protocol == htons(ETH_P_IP)) | |
3864 | return ip_hdr(skb)->ttl; | |
3865 | else if (skb->protocol == htons(ETH_P_IPV6)) | |
3866 | return ipv6_hdr(skb)->hop_limit; | |
3867 | else | |
3868 | return 0; | |
3869 | } | |
3870 | ||
48040793 | 3871 | struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk, |
e7ed11ee YS |
3872 | const struct sk_buff *orig_skb, |
3873 | const struct sk_buff *ack_skb) | |
1c885808 FY |
3874 | { |
3875 | const struct tcp_sock *tp = tcp_sk(sk); | |
3876 | struct sk_buff *stats; | |
3877 | struct tcp_info info; | |
76a9ebe8 | 3878 | unsigned long rate; |
bb7c19f9 | 3879 | u64 rate64; |
1c885808 | 3880 | |
984988aa | 3881 | stats = alloc_skb(tcp_opt_stats_get_size(), GFP_ATOMIC); |
1c885808 FY |
3882 | if (!stats) |
3883 | return NULL; | |
3884 | ||
3885 | tcp_get_info_chrono_stats(tp, &info); | |
3886 | nla_put_u64_64bit(stats, TCP_NLA_BUSY, | |
3887 | info.tcpi_busy_time, TCP_NLA_PAD); | |
3888 | nla_put_u64_64bit(stats, TCP_NLA_RWND_LIMITED, | |
3889 | info.tcpi_rwnd_limited, TCP_NLA_PAD); | |
3890 | nla_put_u64_64bit(stats, TCP_NLA_SNDBUF_LIMITED, | |
3891 | info.tcpi_sndbuf_limited, TCP_NLA_PAD); | |
7e98102f YC |
3892 | nla_put_u64_64bit(stats, TCP_NLA_DATA_SEGS_OUT, |
3893 | tp->data_segs_out, TCP_NLA_PAD); | |
3894 | nla_put_u64_64bit(stats, TCP_NLA_TOTAL_RETRANS, | |
3895 | tp->total_retrans, TCP_NLA_PAD); | |
bb7c19f9 WW |
3896 | |
3897 | rate = READ_ONCE(sk->sk_pacing_rate); | |
76a9ebe8 | 3898 | rate64 = (rate != ~0UL) ? rate : ~0ULL; |
bb7c19f9 WW |
3899 | nla_put_u64_64bit(stats, TCP_NLA_PACING_RATE, rate64, TCP_NLA_PAD); |
3900 | ||
3901 | rate64 = tcp_compute_delivery_rate(tp); | |
3902 | nla_put_u64_64bit(stats, TCP_NLA_DELIVERY_RATE, rate64, TCP_NLA_PAD); | |
3903 | ||
0d24ea1d | 3904 | nla_put_u32(stats, TCP_NLA_SND_CWND, tcp_snd_cwnd(tp)); |
bb7c19f9 WW |
3905 | nla_put_u32(stats, TCP_NLA_REORDERING, tp->reordering); |
3906 | nla_put_u32(stats, TCP_NLA_MIN_RTT, tcp_min_rtt(tp)); | |
3907 | ||
3908 | nla_put_u8(stats, TCP_NLA_RECUR_RETRANS, inet_csk(sk)->icsk_retransmits); | |
3909 | nla_put_u8(stats, TCP_NLA_DELIVERY_RATE_APP_LMT, !!tp->rate_app_limited); | |
7156d194 | 3910 | nla_put_u32(stats, TCP_NLA_SND_SSTHRESH, tp->snd_ssthresh); |
feb5f2ec YC |
3911 | nla_put_u32(stats, TCP_NLA_DELIVERED, tp->delivered); |
3912 | nla_put_u32(stats, TCP_NLA_DELIVERED_CE, tp->delivered_ce); | |
87ecc95d PJ |
3913 | |
3914 | nla_put_u32(stats, TCP_NLA_SNDQ_SIZE, tp->write_seq - tp->snd_una); | |
be631892 | 3915 | nla_put_u8(stats, TCP_NLA_CA_STATE, inet_csk(sk)->icsk_ca_state); |
feb5f2ec | 3916 | |
ba113c3a WW |
3917 | nla_put_u64_64bit(stats, TCP_NLA_BYTES_SENT, tp->bytes_sent, |
3918 | TCP_NLA_PAD); | |
fb31c9b9 WW |
3919 | nla_put_u64_64bit(stats, TCP_NLA_BYTES_RETRANS, tp->bytes_retrans, |
3920 | TCP_NLA_PAD); | |
7e10b655 | 3921 | nla_put_u32(stats, TCP_NLA_DSACK_DUPS, tp->dsack_dups); |
7ec65372 | 3922 | nla_put_u32(stats, TCP_NLA_REORD_SEEN, tp->reord_seen); |
e8bd8fca | 3923 | nla_put_u32(stats, TCP_NLA_SRTT, tp->srtt_us >> 3); |
32efcc06 | 3924 | nla_put_u16(stats, TCP_NLA_TIMEOUT_REHASH, tp->timeout_rehash); |
e08ab0b3 YS |
3925 | nla_put_u32(stats, TCP_NLA_BYTES_NOTSENT, |
3926 | max_t(int, 0, tp->write_seq - tp->snd_nxt)); | |
48040793 YS |
3927 | nla_put_u64_64bit(stats, TCP_NLA_EDT, orig_skb->skb_mstamp_ns, |
3928 | TCP_NLA_PAD); | |
e7ed11ee YS |
3929 | if (ack_skb) |
3930 | nla_put_u8(stats, TCP_NLA_TTL, | |
3931 | tcp_skb_ttl_or_hop_limit(ack_skb)); | |
ba113c3a | 3932 | |
1c885808 FY |
3933 | return stats; |
3934 | } | |
3935 | ||
3fdadf7d DM |
3936 | static int do_tcp_getsockopt(struct sock *sk, int level, |
3937 | int optname, char __user *optval, int __user *optlen) | |
1da177e4 | 3938 | { |
295f7324 | 3939 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 3940 | struct tcp_sock *tp = tcp_sk(sk); |
6fa25166 | 3941 | struct net *net = sock_net(sk); |
1da177e4 LT |
3942 | int val, len; |
3943 | ||
1da177e4 LT |
3944 | if (get_user(len, optlen)) |
3945 | return -EFAULT; | |
3946 | ||
3947 | len = min_t(unsigned int, len, sizeof(int)); | |
3948 | ||
3949 | if (len < 0) | |
3950 | return -EINVAL; | |
3951 | ||
3952 | switch (optname) { | |
3953 | case TCP_MAXSEG: | |
c1b4a7e6 | 3954 | val = tp->mss_cache; |
1da177e4 LT |
3955 | if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) |
3956 | val = tp->rx_opt.user_mss; | |
5e6a3ce6 PE |
3957 | if (tp->repair) |
3958 | val = tp->rx_opt.mss_clamp; | |
1da177e4 LT |
3959 | break; |
3960 | case TCP_NODELAY: | |
3961 | val = !!(tp->nonagle&TCP_NAGLE_OFF); | |
3962 | break; | |
3963 | case TCP_CORK: | |
3964 | val = !!(tp->nonagle&TCP_NAGLE_CORK); | |
3965 | break; | |
3966 | case TCP_KEEPIDLE: | |
df19a626 | 3967 | val = keepalive_time_when(tp) / HZ; |
1da177e4 LT |
3968 | break; |
3969 | case TCP_KEEPINTVL: | |
df19a626 | 3970 | val = keepalive_intvl_when(tp) / HZ; |
1da177e4 LT |
3971 | break; |
3972 | case TCP_KEEPCNT: | |
df19a626 | 3973 | val = keepalive_probes(tp); |
1da177e4 LT |
3974 | break; |
3975 | case TCP_SYNCNT: | |
6fa25166 | 3976 | val = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries; |
1da177e4 LT |
3977 | break; |
3978 | case TCP_LINGER2: | |
3979 | val = tp->linger2; | |
3980 | if (val >= 0) | |
1e579caa | 3981 | val = (val ? : net->ipv4.sysctl_tcp_fin_timeout) / HZ; |
1da177e4 LT |
3982 | break; |
3983 | case TCP_DEFER_ACCEPT: | |
b103cf34 JA |
3984 | val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept, |
3985 | TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ); | |
1da177e4 LT |
3986 | break; |
3987 | case TCP_WINDOW_CLAMP: | |
3988 | val = tp->window_clamp; | |
3989 | break; | |
3990 | case TCP_INFO: { | |
3991 | struct tcp_info info; | |
3992 | ||
3993 | if (get_user(len, optlen)) | |
3994 | return -EFAULT; | |
3995 | ||
3996 | tcp_get_info(sk, &info); | |
3997 | ||
3998 | len = min_t(unsigned int, len, sizeof(info)); | |
3999 | if (put_user(len, optlen)) | |
4000 | return -EFAULT; | |
4001 | if (copy_to_user(optval, &info, len)) | |
4002 | return -EFAULT; | |
4003 | return 0; | |
4004 | } | |
6e9250f5 ED |
4005 | case TCP_CC_INFO: { |
4006 | const struct tcp_congestion_ops *ca_ops; | |
4007 | union tcp_cc_info info; | |
4008 | size_t sz = 0; | |
4009 | int attr; | |
4010 | ||
4011 | if (get_user(len, optlen)) | |
4012 | return -EFAULT; | |
4013 | ||
4014 | ca_ops = icsk->icsk_ca_ops; | |
4015 | if (ca_ops && ca_ops->get_info) | |
4016 | sz = ca_ops->get_info(sk, ~0U, &attr, &info); | |
4017 | ||
4018 | len = min_t(unsigned int, len, sz); | |
4019 | if (put_user(len, optlen)) | |
4020 | return -EFAULT; | |
4021 | if (copy_to_user(optval, &info, len)) | |
4022 | return -EFAULT; | |
4023 | return 0; | |
4024 | } | |
1da177e4 | 4025 | case TCP_QUICKACK: |
31954cd8 | 4026 | val = !inet_csk_in_pingpong_mode(sk); |
1da177e4 | 4027 | break; |
5f8ef48d SH |
4028 | |
4029 | case TCP_CONGESTION: | |
4030 | if (get_user(len, optlen)) | |
4031 | return -EFAULT; | |
4032 | len = min_t(unsigned int, len, TCP_CA_NAME_MAX); | |
4033 | if (put_user(len, optlen)) | |
4034 | return -EFAULT; | |
6687e988 | 4035 | if (copy_to_user(optval, icsk->icsk_ca_ops->name, len)) |
5f8ef48d SH |
4036 | return -EFAULT; |
4037 | return 0; | |
e56fb50f | 4038 | |
734942cc DW |
4039 | case TCP_ULP: |
4040 | if (get_user(len, optlen)) | |
4041 | return -EFAULT; | |
4042 | len = min_t(unsigned int, len, TCP_ULP_NAME_MAX); | |
d97af30f DW |
4043 | if (!icsk->icsk_ulp_ops) { |
4044 | if (put_user(0, optlen)) | |
4045 | return -EFAULT; | |
4046 | return 0; | |
4047 | } | |
734942cc DW |
4048 | if (put_user(len, optlen)) |
4049 | return -EFAULT; | |
4050 | if (copy_to_user(optval, icsk->icsk_ulp_ops->name, len)) | |
4051 | return -EFAULT; | |
4052 | return 0; | |
4053 | ||
1fba70e5 | 4054 | case TCP_FASTOPEN_KEY: { |
f19008e6 JB |
4055 | u64 key[TCP_FASTOPEN_KEY_BUF_LENGTH / sizeof(u64)]; |
4056 | unsigned int key_len; | |
1fba70e5 YC |
4057 | |
4058 | if (get_user(len, optlen)) | |
4059 | return -EFAULT; | |
4060 | ||
f19008e6 JB |
4061 | key_len = tcp_fastopen_get_cipher(net, icsk, key) * |
4062 | TCP_FASTOPEN_KEY_LENGTH; | |
0f1ce023 | 4063 | len = min_t(unsigned int, len, key_len); |
1fba70e5 YC |
4064 | if (put_user(len, optlen)) |
4065 | return -EFAULT; | |
4066 | if (copy_to_user(optval, key, len)) | |
4067 | return -EFAULT; | |
4068 | return 0; | |
4069 | } | |
3c0fef0b JH |
4070 | case TCP_THIN_LINEAR_TIMEOUTS: |
4071 | val = tp->thin_lto; | |
4072 | break; | |
4a7f6009 | 4073 | |
3c0fef0b | 4074 | case TCP_THIN_DUPACK: |
4a7f6009 | 4075 | val = 0; |
3c0fef0b | 4076 | break; |
dca43c75 | 4077 | |
ee995283 PE |
4078 | case TCP_REPAIR: |
4079 | val = tp->repair; | |
4080 | break; | |
4081 | ||
4082 | case TCP_REPAIR_QUEUE: | |
4083 | if (tp->repair) | |
4084 | val = tp->repair_queue; | |
4085 | else | |
4086 | return -EINVAL; | |
4087 | break; | |
4088 | ||
b1ed4c4f AV |
4089 | case TCP_REPAIR_WINDOW: { |
4090 | struct tcp_repair_window opt; | |
4091 | ||
4092 | if (get_user(len, optlen)) | |
4093 | return -EFAULT; | |
4094 | ||
4095 | if (len != sizeof(opt)) | |
4096 | return -EINVAL; | |
4097 | ||
4098 | if (!tp->repair) | |
4099 | return -EPERM; | |
4100 | ||
4101 | opt.snd_wl1 = tp->snd_wl1; | |
4102 | opt.snd_wnd = tp->snd_wnd; | |
4103 | opt.max_window = tp->max_window; | |
4104 | opt.rcv_wnd = tp->rcv_wnd; | |
4105 | opt.rcv_wup = tp->rcv_wup; | |
4106 | ||
4107 | if (copy_to_user(optval, &opt, len)) | |
4108 | return -EFAULT; | |
4109 | return 0; | |
4110 | } | |
ee995283 PE |
4111 | case TCP_QUEUE_SEQ: |
4112 | if (tp->repair_queue == TCP_SEND_QUEUE) | |
4113 | val = tp->write_seq; | |
4114 | else if (tp->repair_queue == TCP_RECV_QUEUE) | |
4115 | val = tp->rcv_nxt; | |
4116 | else | |
4117 | return -EINVAL; | |
4118 | break; | |
4119 | ||
dca43c75 | 4120 | case TCP_USER_TIMEOUT: |
9bcc66e1 | 4121 | val = icsk->icsk_user_timeout; |
dca43c75 | 4122 | break; |
1536e285 KN |
4123 | |
4124 | case TCP_FASTOPEN: | |
0536fcc0 | 4125 | val = icsk->icsk_accept_queue.fastopenq.max_qlen; |
1536e285 KN |
4126 | break; |
4127 | ||
19f6d3f3 WW |
4128 | case TCP_FASTOPEN_CONNECT: |
4129 | val = tp->fastopen_connect; | |
4130 | break; | |
4131 | ||
71c02379 CP |
4132 | case TCP_FASTOPEN_NO_COOKIE: |
4133 | val = tp->fastopen_no_cookie; | |
4134 | break; | |
4135 | ||
a842fe14 ED |
4136 | case TCP_TX_DELAY: |
4137 | val = tp->tcp_tx_delay; | |
4138 | break; | |
4139 | ||
93be6ce0 | 4140 | case TCP_TIMESTAMP: |
9a568de4 | 4141 | val = tcp_time_stamp_raw() + tp->tsoffset; |
93be6ce0 | 4142 | break; |
c9bee3b7 ED |
4143 | case TCP_NOTSENT_LOWAT: |
4144 | val = tp->notsent_lowat; | |
4145 | break; | |
b75eba76 SHY |
4146 | case TCP_INQ: |
4147 | val = tp->recvmsg_inq; | |
4148 | break; | |
cd8ae852 ED |
4149 | case TCP_SAVE_SYN: |
4150 | val = tp->save_syn; | |
4151 | break; | |
4152 | case TCP_SAVED_SYN: { | |
4153 | if (get_user(len, optlen)) | |
4154 | return -EFAULT; | |
4155 | ||
4156 | lock_sock(sk); | |
4157 | if (tp->saved_syn) { | |
70a217f1 MKL |
4158 | if (len < tcp_saved_syn_len(tp->saved_syn)) { |
4159 | if (put_user(tcp_saved_syn_len(tp->saved_syn), | |
4160 | optlen)) { | |
aea0929e EM |
4161 | release_sock(sk); |
4162 | return -EFAULT; | |
4163 | } | |
4164 | release_sock(sk); | |
4165 | return -EINVAL; | |
4166 | } | |
70a217f1 | 4167 | len = tcp_saved_syn_len(tp->saved_syn); |
cd8ae852 ED |
4168 | if (put_user(len, optlen)) { |
4169 | release_sock(sk); | |
4170 | return -EFAULT; | |
4171 | } | |
70a217f1 | 4172 | if (copy_to_user(optval, tp->saved_syn->data, len)) { |
cd8ae852 ED |
4173 | release_sock(sk); |
4174 | return -EFAULT; | |
4175 | } | |
4176 | tcp_saved_syn_free(tp); | |
4177 | release_sock(sk); | |
4178 | } else { | |
4179 | release_sock(sk); | |
4180 | len = 0; | |
4181 | if (put_user(len, optlen)) | |
4182 | return -EFAULT; | |
4183 | } | |
4184 | return 0; | |
4185 | } | |
05255b82 ED |
4186 | #ifdef CONFIG_MMU |
4187 | case TCP_ZEROCOPY_RECEIVE: { | |
7eeba170 | 4188 | struct scm_timestamping_internal tss; |
e0fecb28 | 4189 | struct tcp_zerocopy_receive zc = {}; |
05255b82 ED |
4190 | int err; |
4191 | ||
4192 | if (get_user(len, optlen)) | |
4193 | return -EFAULT; | |
2107d45f AR |
4194 | if (len < 0 || |
4195 | len < offsetofend(struct tcp_zerocopy_receive, length)) | |
05255b82 | 4196 | return -EINVAL; |
3c5a2fd0 AR |
4197 | if (unlikely(len > sizeof(zc))) { |
4198 | err = check_zeroed_user(optval + sizeof(zc), | |
4199 | len - sizeof(zc)); | |
4200 | if (err < 1) | |
4201 | return err == 0 ? -EINVAL : err; | |
c8856c05 | 4202 | len = sizeof(zc); |
0b7f41f6 AR |
4203 | if (put_user(len, optlen)) |
4204 | return -EFAULT; | |
4205 | } | |
05255b82 ED |
4206 | if (copy_from_user(&zc, optval, len)) |
4207 | return -EFAULT; | |
3c5a2fd0 AR |
4208 | if (zc.reserved) |
4209 | return -EINVAL; | |
4210 | if (zc.msg_flags & ~(TCP_VALID_ZC_MSG_FLAGS)) | |
4211 | return -EINVAL; | |
05255b82 | 4212 | lock_sock(sk); |
7eeba170 | 4213 | err = tcp_zerocopy_receive(sk, &zc, &tss); |
9cacf81f SF |
4214 | err = BPF_CGROUP_RUN_PROG_GETSOCKOPT_KERN(sk, level, optname, |
4215 | &zc, &len, err); | |
05255b82 | 4216 | release_sock(sk); |
7eeba170 AR |
4217 | if (len >= offsetofend(struct tcp_zerocopy_receive, msg_flags)) |
4218 | goto zerocopy_rcv_cmsg; | |
c8856c05 | 4219 | switch (len) { |
7eeba170 AR |
4220 | case offsetofend(struct tcp_zerocopy_receive, msg_flags): |
4221 | goto zerocopy_rcv_cmsg; | |
4222 | case offsetofend(struct tcp_zerocopy_receive, msg_controllen): | |
4223 | case offsetofend(struct tcp_zerocopy_receive, msg_control): | |
4224 | case offsetofend(struct tcp_zerocopy_receive, flags): | |
4225 | case offsetofend(struct tcp_zerocopy_receive, copybuf_len): | |
4226 | case offsetofend(struct tcp_zerocopy_receive, copybuf_address): | |
33946518 AR |
4227 | case offsetofend(struct tcp_zerocopy_receive, err): |
4228 | goto zerocopy_rcv_sk_err; | |
c8856c05 AR |
4229 | case offsetofend(struct tcp_zerocopy_receive, inq): |
4230 | goto zerocopy_rcv_inq; | |
4231 | case offsetofend(struct tcp_zerocopy_receive, length): | |
4232 | default: | |
4233 | goto zerocopy_rcv_out; | |
4234 | } | |
7eeba170 AR |
4235 | zerocopy_rcv_cmsg: |
4236 | if (zc.msg_flags & TCP_CMSG_TS) | |
4237 | tcp_zc_finalize_rx_tstamp(sk, &zc, &tss); | |
4238 | else | |
4239 | zc.msg_flags = 0; | |
33946518 AR |
4240 | zerocopy_rcv_sk_err: |
4241 | if (!err) | |
4242 | zc.err = sock_error(sk); | |
c8856c05 AR |
4243 | zerocopy_rcv_inq: |
4244 | zc.inq = tcp_inq_hint(sk); | |
4245 | zerocopy_rcv_out: | |
05255b82 ED |
4246 | if (!err && copy_to_user(optval, &zc, len)) |
4247 | err = -EFAULT; | |
4248 | return err; | |
4249 | } | |
4250 | #endif | |
1da177e4 LT |
4251 | default: |
4252 | return -ENOPROTOOPT; | |
3ff50b79 | 4253 | } |
1da177e4 LT |
4254 | |
4255 | if (put_user(len, optlen)) | |
4256 | return -EFAULT; | |
4257 | if (copy_to_user(optval, &val, len)) | |
4258 | return -EFAULT; | |
4259 | return 0; | |
4260 | } | |
4261 | ||
9cacf81f SF |
4262 | bool tcp_bpf_bypass_getsockopt(int level, int optname) |
4263 | { | |
4264 | /* TCP do_tcp_getsockopt has optimized getsockopt implementation | |
4265 | * to avoid extra socket lock for TCP_ZEROCOPY_RECEIVE. | |
4266 | */ | |
4267 | if (level == SOL_TCP && optname == TCP_ZEROCOPY_RECEIVE) | |
4268 | return true; | |
4269 | ||
4270 | return false; | |
4271 | } | |
4272 | EXPORT_SYMBOL(tcp_bpf_bypass_getsockopt); | |
4273 | ||
3fdadf7d DM |
4274 | int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval, |
4275 | int __user *optlen) | |
4276 | { | |
4277 | struct inet_connection_sock *icsk = inet_csk(sk); | |
4278 | ||
4279 | if (level != SOL_TCP) | |
4280 | return icsk->icsk_af_ops->getsockopt(sk, level, optname, | |
4281 | optval, optlen); | |
4282 | return do_tcp_getsockopt(sk, level, optname, optval, optlen); | |
4283 | } | |
4bc2f18b | 4284 | EXPORT_SYMBOL(tcp_getsockopt); |
3fdadf7d | 4285 | |
cfb6eeb4 | 4286 | #ifdef CONFIG_TCP_MD5SIG |
349ce993 | 4287 | static DEFINE_PER_CPU(struct tcp_md5sig_pool, tcp_md5sig_pool); |
71cea17e | 4288 | static DEFINE_MUTEX(tcp_md5sig_mutex); |
349ce993 | 4289 | static bool tcp_md5sig_pool_populated = false; |
cfb6eeb4 | 4290 | |
71cea17e | 4291 | static void __tcp_alloc_md5sig_pool(void) |
cfb6eeb4 | 4292 | { |
cf80e0e4 | 4293 | struct crypto_ahash *hash; |
cfb6eeb4 | 4294 | int cpu; |
cfb6eeb4 | 4295 | |
cf80e0e4 | 4296 | hash = crypto_alloc_ahash("md5", 0, CRYPTO_ALG_ASYNC); |
70477371 | 4297 | if (IS_ERR(hash)) |
cf80e0e4 HX |
4298 | return; |
4299 | ||
cfb6eeb4 | 4300 | for_each_possible_cpu(cpu) { |
19689e38 | 4301 | void *scratch = per_cpu(tcp_md5sig_pool, cpu).scratch; |
cf80e0e4 | 4302 | struct ahash_request *req; |
cfb6eeb4 | 4303 | |
19689e38 ED |
4304 | if (!scratch) { |
4305 | scratch = kmalloc_node(sizeof(union tcp_md5sum_block) + | |
4306 | sizeof(struct tcphdr), | |
4307 | GFP_KERNEL, | |
4308 | cpu_to_node(cpu)); | |
4309 | if (!scratch) | |
4310 | return; | |
4311 | per_cpu(tcp_md5sig_pool, cpu).scratch = scratch; | |
4312 | } | |
cf80e0e4 HX |
4313 | if (per_cpu(tcp_md5sig_pool, cpu).md5_req) |
4314 | continue; | |
4315 | ||
4316 | req = ahash_request_alloc(hash, GFP_KERNEL); | |
4317 | if (!req) | |
4318 | return; | |
4319 | ||
4320 | ahash_request_set_callback(req, 0, NULL, NULL); | |
4321 | ||
4322 | per_cpu(tcp_md5sig_pool, cpu).md5_req = req; | |
cfb6eeb4 | 4323 | } |
349ce993 ED |
4324 | /* before setting tcp_md5sig_pool_populated, we must commit all writes |
4325 | * to memory. See smp_rmb() in tcp_get_md5sig_pool() | |
71cea17e ED |
4326 | */ |
4327 | smp_wmb(); | |
349ce993 | 4328 | tcp_md5sig_pool_populated = true; |
cfb6eeb4 YH |
4329 | } |
4330 | ||
71cea17e | 4331 | bool tcp_alloc_md5sig_pool(void) |
cfb6eeb4 | 4332 | { |
349ce993 | 4333 | if (unlikely(!tcp_md5sig_pool_populated)) { |
71cea17e ED |
4334 | mutex_lock(&tcp_md5sig_mutex); |
4335 | ||
6015c71e | 4336 | if (!tcp_md5sig_pool_populated) { |
71cea17e | 4337 | __tcp_alloc_md5sig_pool(); |
6015c71e | 4338 | if (tcp_md5sig_pool_populated) |
921f9a0f | 4339 | static_branch_inc(&tcp_md5_needed); |
6015c71e | 4340 | } |
71cea17e ED |
4341 | |
4342 | mutex_unlock(&tcp_md5sig_mutex); | |
cfb6eeb4 | 4343 | } |
349ce993 | 4344 | return tcp_md5sig_pool_populated; |
cfb6eeb4 | 4345 | } |
cfb6eeb4 YH |
4346 | EXPORT_SYMBOL(tcp_alloc_md5sig_pool); |
4347 | ||
35790c04 ED |
4348 | |
4349 | /** | |
4350 | * tcp_get_md5sig_pool - get md5sig_pool for this user | |
4351 | * | |
4352 | * We use percpu structure, so if we succeed, we exit with preemption | |
4353 | * and BH disabled, to make sure another thread or softirq handling | |
4354 | * wont try to get same context. | |
4355 | */ | |
4356 | struct tcp_md5sig_pool *tcp_get_md5sig_pool(void) | |
cfb6eeb4 | 4357 | { |
35790c04 | 4358 | local_bh_disable(); |
cfb6eeb4 | 4359 | |
349ce993 ED |
4360 | if (tcp_md5sig_pool_populated) { |
4361 | /* coupled with smp_wmb() in __tcp_alloc_md5sig_pool() */ | |
4362 | smp_rmb(); | |
4363 | return this_cpu_ptr(&tcp_md5sig_pool); | |
4364 | } | |
35790c04 ED |
4365 | local_bh_enable(); |
4366 | return NULL; | |
4367 | } | |
4368 | EXPORT_SYMBOL(tcp_get_md5sig_pool); | |
cfb6eeb4 | 4369 | |
49a72dfb | 4370 | int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp, |
cf533ea5 | 4371 | const struct sk_buff *skb, unsigned int header_len) |
49a72dfb AL |
4372 | { |
4373 | struct scatterlist sg; | |
4374 | const struct tcphdr *tp = tcp_hdr(skb); | |
cf80e0e4 | 4375 | struct ahash_request *req = hp->md5_req; |
95c96174 ED |
4376 | unsigned int i; |
4377 | const unsigned int head_data_len = skb_headlen(skb) > header_len ? | |
4378 | skb_headlen(skb) - header_len : 0; | |
49a72dfb | 4379 | const struct skb_shared_info *shi = skb_shinfo(skb); |
d7fd1b57 | 4380 | struct sk_buff *frag_iter; |
49a72dfb AL |
4381 | |
4382 | sg_init_table(&sg, 1); | |
4383 | ||
4384 | sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len); | |
cf80e0e4 HX |
4385 | ahash_request_set_crypt(req, &sg, NULL, head_data_len); |
4386 | if (crypto_ahash_update(req)) | |
49a72dfb AL |
4387 | return 1; |
4388 | ||
4389 | for (i = 0; i < shi->nr_frags; ++i) { | |
d8e18a51 | 4390 | const skb_frag_t *f = &shi->frags[i]; |
b54c9d5b | 4391 | unsigned int offset = skb_frag_off(f); |
54d27fcb ED |
4392 | struct page *page = skb_frag_page(f) + (offset >> PAGE_SHIFT); |
4393 | ||
4394 | sg_set_page(&sg, page, skb_frag_size(f), | |
4395 | offset_in_page(offset)); | |
cf80e0e4 HX |
4396 | ahash_request_set_crypt(req, &sg, NULL, skb_frag_size(f)); |
4397 | if (crypto_ahash_update(req)) | |
49a72dfb AL |
4398 | return 1; |
4399 | } | |
4400 | ||
d7fd1b57 ED |
4401 | skb_walk_frags(skb, frag_iter) |
4402 | if (tcp_md5_hash_skb_data(hp, frag_iter, 0)) | |
4403 | return 1; | |
4404 | ||
49a72dfb AL |
4405 | return 0; |
4406 | } | |
49a72dfb AL |
4407 | EXPORT_SYMBOL(tcp_md5_hash_skb_data); |
4408 | ||
cf533ea5 | 4409 | int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key) |
49a72dfb | 4410 | { |
e6ced831 | 4411 | u8 keylen = READ_ONCE(key->keylen); /* paired with WRITE_ONCE() in tcp_md5_do_add */ |
49a72dfb AL |
4412 | struct scatterlist sg; |
4413 | ||
6a2febec ED |
4414 | sg_init_one(&sg, key->key, keylen); |
4415 | ahash_request_set_crypt(hp->md5_req, &sg, NULL, keylen); | |
e6ced831 ED |
4416 | |
4417 | /* We use data_race() because tcp_md5_do_add() might change key->key under us */ | |
4418 | return data_race(crypto_ahash_update(hp->md5_req)); | |
49a72dfb | 4419 | } |
49a72dfb AL |
4420 | EXPORT_SYMBOL(tcp_md5_hash_key); |
4421 | ||
cfb6eeb4 YH |
4422 | #endif |
4423 | ||
4ac02bab AK |
4424 | void tcp_done(struct sock *sk) |
4425 | { | |
d983ea6f | 4426 | struct request_sock *req; |
8336886f | 4427 | |
cab209e5 ED |
4428 | /* We might be called with a new socket, after |
4429 | * inet_csk_prepare_forced_close() has been called | |
4430 | * so we can not use lockdep_sock_is_held(sk) | |
4431 | */ | |
4432 | req = rcu_dereference_protected(tcp_sk(sk)->fastopen_rsk, 1); | |
8336886f | 4433 | |
5a5f3a8d | 4434 | if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV) |
c10d9310 | 4435 | TCP_INC_STATS(sock_net(sk), TCP_MIB_ATTEMPTFAILS); |
4ac02bab AK |
4436 | |
4437 | tcp_set_state(sk, TCP_CLOSE); | |
4438 | tcp_clear_xmit_timers(sk); | |
00db4124 | 4439 | if (req) |
8336886f | 4440 | reqsk_fastopen_remove(sk, req, false); |
4ac02bab AK |
4441 | |
4442 | sk->sk_shutdown = SHUTDOWN_MASK; | |
4443 | ||
4444 | if (!sock_flag(sk, SOCK_DEAD)) | |
4445 | sk->sk_state_change(sk); | |
4446 | else | |
4447 | inet_csk_destroy_sock(sk); | |
4448 | } | |
4449 | EXPORT_SYMBOL_GPL(tcp_done); | |
4450 | ||
c1e64e29 LC |
4451 | int tcp_abort(struct sock *sk, int err) |
4452 | { | |
4453 | if (!sk_fullsock(sk)) { | |
07f6f4a3 ED |
4454 | if (sk->sk_state == TCP_NEW_SYN_RECV) { |
4455 | struct request_sock *req = inet_reqsk(sk); | |
4456 | ||
4457 | local_bh_disable(); | |
acc2cf4e | 4458 | inet_csk_reqsk_queue_drop(req->rsk_listener, req); |
07f6f4a3 ED |
4459 | local_bh_enable(); |
4460 | return 0; | |
4461 | } | |
c1e64e29 LC |
4462 | return -EOPNOTSUPP; |
4463 | } | |
4464 | ||
4465 | /* Don't race with userspace socket closes such as tcp_close. */ | |
4466 | lock_sock(sk); | |
4467 | ||
2010b93e LC |
4468 | if (sk->sk_state == TCP_LISTEN) { |
4469 | tcp_set_state(sk, TCP_CLOSE); | |
4470 | inet_csk_listen_stop(sk); | |
4471 | } | |
4472 | ||
c1e64e29 LC |
4473 | /* Don't race with BH socket closes such as inet_csk_listen_stop. */ |
4474 | local_bh_disable(); | |
4475 | bh_lock_sock(sk); | |
4476 | ||
4477 | if (!sock_flag(sk, SOCK_DEAD)) { | |
4478 | sk->sk_err = err; | |
4479 | /* This barrier is coupled with smp_rmb() in tcp_poll() */ | |
4480 | smp_wmb(); | |
e3ae2365 | 4481 | sk_error_report(sk); |
c1e64e29 LC |
4482 | if (tcp_need_reset(sk->sk_state)) |
4483 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
4484 | tcp_done(sk); | |
4485 | } | |
4486 | ||
4487 | bh_unlock_sock(sk); | |
4488 | local_bh_enable(); | |
e05836ac | 4489 | tcp_write_queue_purge(sk); |
c1e64e29 | 4490 | release_sock(sk); |
c1e64e29 LC |
4491 | return 0; |
4492 | } | |
4493 | EXPORT_SYMBOL_GPL(tcp_abort); | |
4494 | ||
5f8ef48d | 4495 | extern struct tcp_congestion_ops tcp_reno; |
1da177e4 LT |
4496 | |
4497 | static __initdata unsigned long thash_entries; | |
4498 | static int __init set_thash_entries(char *str) | |
4499 | { | |
413c27d8 EZ |
4500 | ssize_t ret; |
4501 | ||
1da177e4 LT |
4502 | if (!str) |
4503 | return 0; | |
413c27d8 EZ |
4504 | |
4505 | ret = kstrtoul(str, 0, &thash_entries); | |
4506 | if (ret) | |
4507 | return 0; | |
4508 | ||
1da177e4 LT |
4509 | return 1; |
4510 | } | |
4511 | __setup("thash_entries=", set_thash_entries); | |
4512 | ||
47d7a88c | 4513 | static void __init tcp_init_mem(void) |
4acb4190 | 4514 | { |
b66e91cc ED |
4515 | unsigned long limit = nr_free_buffer_pages() / 16; |
4516 | ||
4acb4190 | 4517 | limit = max(limit, 128UL); |
b66e91cc ED |
4518 | sysctl_tcp_mem[0] = limit / 4 * 3; /* 4.68 % */ |
4519 | sysctl_tcp_mem[1] = limit; /* 6.25 % */ | |
4520 | sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2; /* 9.37 % */ | |
4acb4190 GC |
4521 | } |
4522 | ||
1da177e4 LT |
4523 | void __init tcp_init(void) |
4524 | { | |
b49960a0 | 4525 | int max_rshare, max_wshare, cnt; |
b2d3ea4a | 4526 | unsigned long limit; |
074b8517 | 4527 | unsigned int i; |
1da177e4 | 4528 | |
3b4929f6 | 4529 | BUILD_BUG_ON(TCP_MIN_SND_MSS <= MAX_TCP_OPTION_SPACE); |
b2d3ea4a | 4530 | BUILD_BUG_ON(sizeof(struct tcp_skb_cb) > |
c593642c | 4531 | sizeof_field(struct sk_buff, cb)); |
1da177e4 | 4532 | |
908c7f19 | 4533 | percpu_counter_init(&tcp_sockets_allocated, 0, GFP_KERNEL); |
e40ec5b7 ED |
4534 | |
4535 | timer_setup(&tcp_orphan_timer, tcp_orphan_update, TIMER_DEFERRABLE); | |
4536 | mod_timer(&tcp_orphan_timer, jiffies + TCP_ORPHAN_TIMER_PERIOD); | |
4537 | ||
1946e672 | 4538 | inet_hashinfo_init(&tcp_hashinfo); |
27da6d37 MKL |
4539 | inet_hashinfo2_init(&tcp_hashinfo, "tcp_listen_portaddr_hash", |
4540 | thash_entries, 21, /* one slot per 2 MB*/ | |
4541 | 0, 64 * 1024); | |
6e04e021 ACM |
4542 | tcp_hashinfo.bind_bucket_cachep = |
4543 | kmem_cache_create("tcp_bind_bucket", | |
4544 | sizeof(struct inet_bind_bucket), 0, | |
990c74e3 VA |
4545 | SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
4546 | SLAB_ACCOUNT, | |
4547 | NULL); | |
1da177e4 | 4548 | |
1da177e4 LT |
4549 | /* Size and allocate the main established and bind bucket |
4550 | * hash tables. | |
4551 | * | |
4552 | * The methodology is similar to that of the buffer cache. | |
4553 | */ | |
6e04e021 | 4554 | tcp_hashinfo.ehash = |
1da177e4 | 4555 | alloc_large_system_hash("TCP established", |
0f7ff927 | 4556 | sizeof(struct inet_ehash_bucket), |
1da177e4 | 4557 | thash_entries, |
fd90b29d | 4558 | 17, /* one slot per 128 KB of memory */ |
9e950efa | 4559 | 0, |
1da177e4 | 4560 | NULL, |
f373b53b | 4561 | &tcp_hashinfo.ehash_mask, |
31fe62b9 | 4562 | 0, |
0ccfe618 | 4563 | thash_entries ? 0 : 512 * 1024); |
05dbc7b5 | 4564 | for (i = 0; i <= tcp_hashinfo.ehash_mask; i++) |
3ab5aee7 | 4565 | INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i); |
05dbc7b5 | 4566 | |
230140cf ED |
4567 | if (inet_ehash_locks_alloc(&tcp_hashinfo)) |
4568 | panic("TCP: failed to alloc ehash_locks"); | |
6e04e021 | 4569 | tcp_hashinfo.bhash = |
1da177e4 | 4570 | alloc_large_system_hash("TCP bind", |
0f7ff927 | 4571 | sizeof(struct inet_bind_hashbucket), |
f373b53b | 4572 | tcp_hashinfo.ehash_mask + 1, |
fd90b29d | 4573 | 17, /* one slot per 128 KB of memory */ |
9e950efa | 4574 | 0, |
6e04e021 | 4575 | &tcp_hashinfo.bhash_size, |
1da177e4 | 4576 | NULL, |
31fe62b9 | 4577 | 0, |
1da177e4 | 4578 | 64 * 1024); |
074b8517 | 4579 | tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size; |
6e04e021 ACM |
4580 | for (i = 0; i < tcp_hashinfo.bhash_size; i++) { |
4581 | spin_lock_init(&tcp_hashinfo.bhash[i].lock); | |
4582 | INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain); | |
1da177e4 LT |
4583 | } |
4584 | ||
c5ed63d6 ED |
4585 | |
4586 | cnt = tcp_hashinfo.ehash_mask + 1; | |
c5ed63d6 | 4587 | sysctl_tcp_max_orphans = cnt / 2; |
1da177e4 | 4588 | |
a4fe34bf | 4589 | tcp_init_mem(); |
c43b874d | 4590 | /* Set per-socket limits to no more than 1/128 the pressure threshold */ |
5fb84b14 | 4591 | limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7); |
b49960a0 ED |
4592 | max_wshare = min(4UL*1024*1024, limit); |
4593 | max_rshare = min(6UL*1024*1024, limit); | |
7b4f4b5e | 4594 | |
356d1833 ED |
4595 | init_net.ipv4.sysctl_tcp_wmem[0] = SK_MEM_QUANTUM; |
4596 | init_net.ipv4.sysctl_tcp_wmem[1] = 16*1024; | |
4597 | init_net.ipv4.sysctl_tcp_wmem[2] = max(64*1024, max_wshare); | |
7b4f4b5e | 4598 | |
356d1833 | 4599 | init_net.ipv4.sysctl_tcp_rmem[0] = SK_MEM_QUANTUM; |
a337531b YC |
4600 | init_net.ipv4.sysctl_tcp_rmem[1] = 131072; |
4601 | init_net.ipv4.sysctl_tcp_rmem[2] = max(131072, max_rshare); | |
1da177e4 | 4602 | |
afd46503 | 4603 | pr_info("Hash tables configured (established %u bind %u)\n", |
058bd4d2 | 4604 | tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size); |
317a76f9 | 4605 | |
1946e672 | 4606 | tcp_v4_init(); |
51c5d0c4 | 4607 | tcp_metrics_init(); |
55d8694f | 4608 | BUG_ON(tcp_register_congestion_control(&tcp_reno) != 0); |
46d3ceab | 4609 | tcp_tasklet_init(); |
f870fa0b | 4610 | mptcp_init(); |
1da177e4 | 4611 | } |