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 | 262 | #include <linux/highmem.h> |
b8059ead | 263 | #include <linux/cache.h> |
f4c50d99 | 264 | #include <linux/err.h> |
da5c78c8 | 265 | #include <linux/time.h> |
5a0e3ad6 | 266 | #include <linux/slab.h> |
98aaa913 | 267 | #include <linux/errqueue.h> |
60e2a778 | 268 | #include <linux/static_key.h> |
97a19caf | 269 | #include <linux/btf.h> |
1da177e4 LT |
270 | |
271 | #include <net/icmp.h> | |
cf60af03 | 272 | #include <net/inet_common.h> |
1da177e4 | 273 | #include <net/tcp.h> |
f870fa0b | 274 | #include <net/mptcp.h> |
1da177e4 LT |
275 | #include <net/xfrm.h> |
276 | #include <net/ip.h> | |
9c55e01c | 277 | #include <net/sock.h> |
1da177e4 | 278 | |
7c0f6ba6 | 279 | #include <linux/uaccess.h> |
1da177e4 | 280 | #include <asm/ioctls.h> |
076bb0c8 | 281 | #include <net/busy_poll.h> |
1da177e4 | 282 | |
925bba24 AR |
283 | /* Track pending CMSGs. */ |
284 | enum { | |
285 | TCP_CMSG_INQ = 1, | |
286 | TCP_CMSG_TS = 2 | |
287 | }; | |
288 | ||
19757ceb ED |
289 | DEFINE_PER_CPU(unsigned int, tcp_orphan_count); |
290 | EXPORT_PER_CPU_SYMBOL_GPL(tcp_orphan_count); | |
0a5578cf | 291 | |
a4fe34bf | 292 | long sysctl_tcp_mem[3] __read_mostly; |
a4fe34bf | 293 | EXPORT_SYMBOL(sysctl_tcp_mem); |
1da177e4 | 294 | |
91b6d325 | 295 | atomic_long_t tcp_memory_allocated ____cacheline_aligned_in_smp; /* Current allocated memory. */ |
1da177e4 | 296 | EXPORT_SYMBOL(tcp_memory_allocated); |
0defbb0a ED |
297 | DEFINE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc); |
298 | EXPORT_PER_CPU_SYMBOL_GPL(tcp_memory_per_cpu_fw_alloc); | |
1748376b | 299 | |
60e2a778 UB |
300 | #if IS_ENABLED(CONFIG_SMC) |
301 | DEFINE_STATIC_KEY_FALSE(tcp_have_smc); | |
302 | EXPORT_SYMBOL(tcp_have_smc); | |
303 | #endif | |
304 | ||
1748376b ED |
305 | /* |
306 | * Current number of TCP sockets. | |
307 | */ | |
91b6d325 | 308 | struct percpu_counter tcp_sockets_allocated ____cacheline_aligned_in_smp; |
1da177e4 LT |
309 | EXPORT_SYMBOL(tcp_sockets_allocated); |
310 | ||
9c55e01c JA |
311 | /* |
312 | * TCP splice context | |
313 | */ | |
314 | struct tcp_splice_state { | |
315 | struct pipe_inode_info *pipe; | |
316 | size_t len; | |
317 | unsigned int flags; | |
318 | }; | |
319 | ||
1da177e4 LT |
320 | /* |
321 | * Pressure flag: try to collapse. | |
322 | * Technical note: it is used by multiple contexts non atomically. | |
3ab224be | 323 | * All the __sk_mem_schedule() is of this nature: accounting |
1da177e4 LT |
324 | * is strict, actions are advisory and have some latency. |
325 | */ | |
06044751 ED |
326 | unsigned long tcp_memory_pressure __read_mostly; |
327 | EXPORT_SYMBOL_GPL(tcp_memory_pressure); | |
1da177e4 | 328 | |
5c52ba17 | 329 | void tcp_enter_memory_pressure(struct sock *sk) |
1da177e4 | 330 | { |
06044751 ED |
331 | unsigned long val; |
332 | ||
1f142c17 | 333 | if (READ_ONCE(tcp_memory_pressure)) |
06044751 ED |
334 | return; |
335 | val = jiffies; | |
336 | ||
337 | if (!val) | |
338 | val--; | |
339 | if (!cmpxchg(&tcp_memory_pressure, 0, val)) | |
4e673444 | 340 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES); |
1da177e4 | 341 | } |
06044751 ED |
342 | EXPORT_SYMBOL_GPL(tcp_enter_memory_pressure); |
343 | ||
344 | void tcp_leave_memory_pressure(struct sock *sk) | |
345 | { | |
346 | unsigned long val; | |
347 | ||
1f142c17 | 348 | if (!READ_ONCE(tcp_memory_pressure)) |
06044751 ED |
349 | return; |
350 | val = xchg(&tcp_memory_pressure, 0); | |
351 | if (val) | |
352 | NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURESCHRONO, | |
353 | jiffies_to_msecs(jiffies - val)); | |
354 | } | |
355 | EXPORT_SYMBOL_GPL(tcp_leave_memory_pressure); | |
1da177e4 | 356 | |
b103cf34 JA |
357 | /* Convert seconds to retransmits based on initial and max timeout */ |
358 | static u8 secs_to_retrans(int seconds, int timeout, int rto_max) | |
359 | { | |
360 | u8 res = 0; | |
361 | ||
362 | if (seconds > 0) { | |
363 | int period = timeout; | |
364 | ||
365 | res = 1; | |
366 | while (seconds > period && res < 255) { | |
367 | res++; | |
368 | timeout <<= 1; | |
369 | if (timeout > rto_max) | |
370 | timeout = rto_max; | |
371 | period += timeout; | |
372 | } | |
373 | } | |
374 | return res; | |
375 | } | |
376 | ||
377 | /* Convert retransmits to seconds based on initial and max timeout */ | |
378 | static int retrans_to_secs(u8 retrans, int timeout, int rto_max) | |
379 | { | |
380 | int period = 0; | |
381 | ||
382 | if (retrans > 0) { | |
383 | period = timeout; | |
384 | while (--retrans) { | |
385 | timeout <<= 1; | |
386 | if (timeout > rto_max) | |
387 | timeout = rto_max; | |
388 | period += timeout; | |
389 | } | |
390 | } | |
391 | return period; | |
392 | } | |
393 | ||
0263598c WW |
394 | static u64 tcp_compute_delivery_rate(const struct tcp_sock *tp) |
395 | { | |
396 | u32 rate = READ_ONCE(tp->rate_delivered); | |
397 | u32 intv = READ_ONCE(tp->rate_interval_us); | |
398 | u64 rate64 = 0; | |
399 | ||
400 | if (rate && intv) { | |
401 | rate64 = (u64)rate * tp->mss_cache * USEC_PER_SEC; | |
402 | do_div(rate64, intv); | |
403 | } | |
404 | return rate64; | |
405 | } | |
406 | ||
900f65d3 NC |
407 | /* Address-family independent initialization for a tcp_sock. |
408 | * | |
409 | * NOTE: A lot of things set to zero explicitly by call to | |
410 | * sk_alloc() so need not be done here. | |
411 | */ | |
412 | void tcp_init_sock(struct sock *sk) | |
413 | { | |
414 | struct inet_connection_sock *icsk = inet_csk(sk); | |
415 | struct tcp_sock *tp = tcp_sk(sk); | |
416 | ||
9f5afeae | 417 | tp->out_of_order_queue = RB_ROOT; |
75c119af | 418 | sk->tcp_rtx_queue = RB_ROOT; |
900f65d3 | 419 | tcp_init_xmit_timers(sk); |
46d3ceab | 420 | INIT_LIST_HEAD(&tp->tsq_node); |
e2080072 | 421 | INIT_LIST_HEAD(&tp->tsorted_sent_queue); |
900f65d3 NC |
422 | |
423 | icsk->icsk_rto = TCP_TIMEOUT_INIT; | |
ca584ba0 | 424 | icsk->icsk_rto_min = TCP_RTO_MIN; |
2b8ee4f0 | 425 | icsk->icsk_delack_max = TCP_DELACK_MAX; |
740b0f18 | 426 | tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT); |
ac9517fc | 427 | minmax_reset(&tp->rtt_min, tcp_jiffies32, ~0U); |
900f65d3 NC |
428 | |
429 | /* So many TCP implementations out there (incorrectly) count the | |
430 | * initial SYN frame in their delayed-ACK and congestion control | |
431 | * algorithms that we must have the following bandaid to talk | |
432 | * efficiently to them. -DaveM | |
433 | */ | |
40570375 | 434 | tcp_snd_cwnd_set(tp, TCP_INIT_CWND); |
900f65d3 | 435 | |
d7722e85 SHY |
436 | /* There's a bubble in the pipe until at least the first ACK. */ |
437 | tp->app_limited = ~0U; | |
300b655d | 438 | tp->rate_app_limited = 1; |
d7722e85 | 439 | |
900f65d3 NC |
440 | /* See draft-stevens-tcpca-spec-01 for discussion of the |
441 | * initialization of these values. | |
442 | */ | |
443 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; | |
444 | tp->snd_cwnd_clamp = ~0; | |
445 | tp->mss_cache = TCP_MSS_DEFAULT; | |
446 | ||
46778cd1 | 447 | tp->reordering = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reordering); |
55d8694f | 448 | tcp_assign_congestion_control(sk); |
900f65d3 | 449 | |
ceaa1fef | 450 | tp->tsoffset = 0; |
1f255691 | 451 | tp->rack.reo_wnd_steps = 1; |
ceaa1fef | 452 | |
900f65d3 NC |
453 | sk->sk_write_space = sk_stream_write_space; |
454 | sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); | |
455 | ||
456 | icsk->icsk_sync_mss = tcp_sync_mss; | |
457 | ||
02739545 KI |
458 | WRITE_ONCE(sk->sk_sndbuf, READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_wmem[1])); |
459 | WRITE_ONCE(sk->sk_rcvbuf, READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[1])); | |
dfa2f048 | 460 | tcp_scaling_ratio_init(sk); |
900f65d3 | 461 | |
e993ffe3 | 462 | set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); |
900f65d3 | 463 | sk_sockets_allocated_inc(sk); |
900f65d3 NC |
464 | } |
465 | EXPORT_SYMBOL(tcp_init_sock); | |
466 | ||
4e8cc228 | 467 | static void tcp_tx_timestamp(struct sock *sk, u16 tsflags) |
4ed2d765 | 468 | { |
4e8cc228 ED |
469 | struct sk_buff *skb = tcp_write_queue_tail(sk); |
470 | ||
ad02c4f5 | 471 | if (tsflags && skb) { |
f066e2b0 | 472 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
6b084928 | 473 | struct tcp_skb_cb *tcb = TCP_SKB_CB(skb); |
4ed2d765 | 474 | |
c14ac945 | 475 | sock_tx_timestamp(sk, tsflags, &shinfo->tx_flags); |
0a2cf20c SHY |
476 | if (tsflags & SOF_TIMESTAMPING_TX_ACK) |
477 | tcb->txstamp_ack = 1; | |
478 | if (tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) | |
f066e2b0 WB |
479 | shinfo->tskey = TCP_SKB_CB(skb)->seq + skb->len - 1; |
480 | } | |
4ed2d765 WB |
481 | } |
482 | ||
05dc72ab | 483 | static bool tcp_stream_is_readable(struct sock *sk, int target) |
8934ce2f | 484 | { |
05dc72ab ED |
485 | if (tcp_epollin_ready(sk, target)) |
486 | return true; | |
7b50ecfc | 487 | return sk_is_readable(sk); |
8934ce2f JF |
488 | } |
489 | ||
1da177e4 | 490 | /* |
a11e1d43 LT |
491 | * Wait for a TCP event. |
492 | * | |
493 | * Note that we don't need to lock the socket, as the upper poll layers | |
494 | * take care of normal races (between the test and the event) and we don't | |
495 | * go look at any of the socket buffers directly. | |
1da177e4 | 496 | */ |
a11e1d43 | 497 | __poll_t tcp_poll(struct file *file, struct socket *sock, poll_table *wait) |
1da177e4 | 498 | { |
a11e1d43 | 499 | __poll_t mask; |
1da177e4 | 500 | struct sock *sk = sock->sk; |
cf533ea5 | 501 | const struct tcp_sock *tp = tcp_sk(sk); |
e14cadfd | 502 | u8 shutdown; |
00fd38d9 | 503 | int state; |
1da177e4 | 504 | |
89ab066d | 505 | sock_poll_wait(file, sock, wait); |
a11e1d43 | 506 | |
986ffdfd | 507 | state = inet_sk_state_load(sk); |
00fd38d9 | 508 | if (state == TCP_LISTEN) |
dc40c7bc | 509 | return inet_csk_listen_poll(sk); |
1da177e4 | 510 | |
a11e1d43 LT |
511 | /* Socket is not locked. We are protected from async events |
512 | * by poll logic and correct handling of state changes | |
513 | * made by other threads is impossible in any case. | |
514 | */ | |
515 | ||
516 | mask = 0; | |
517 | ||
1da177e4 | 518 | /* |
a9a08845 | 519 | * EPOLLHUP is certainly not done right. But poll() doesn't |
1da177e4 LT |
520 | * have a notion of HUP in just one direction, and for a |
521 | * socket the read side is more interesting. | |
522 | * | |
a9a08845 LT |
523 | * Some poll() documentation says that EPOLLHUP is incompatible |
524 | * with the EPOLLOUT/POLLWR flags, so somebody should check this | |
1da177e4 LT |
525 | * all. But careful, it tends to be safer to return too many |
526 | * bits than too few, and you can easily break real applications | |
527 | * if you don't tell them that something has hung up! | |
528 | * | |
529 | * Check-me. | |
530 | * | |
a9a08845 | 531 | * Check number 1. EPOLLHUP is _UNMASKABLE_ event (see UNIX98 and |
1da177e4 LT |
532 | * our fs/select.c). It means that after we received EOF, |
533 | * poll always returns immediately, making impossible poll() on write() | |
a9a08845 | 534 | * in state CLOSE_WAIT. One solution is evident --- to set EPOLLHUP |
1da177e4 LT |
535 | * if and only if shutdown has been made in both directions. |
536 | * Actually, it is interesting to look how Solaris and DUX | |
a9a08845 | 537 | * solve this dilemma. I would prefer, if EPOLLHUP were maskable, |
1da177e4 LT |
538 | * then we could set it on SND_SHUTDOWN. BTW examples given |
539 | * in Stevens' books assume exactly this behaviour, it explains | |
a9a08845 | 540 | * why EPOLLHUP is incompatible with EPOLLOUT. --ANK |
1da177e4 LT |
541 | * |
542 | * NOTE. Check for TCP_CLOSE is added. The goal is to prevent | |
543 | * blocking on fresh not-connected or disconnected socket. --ANK | |
544 | */ | |
e14cadfd ED |
545 | shutdown = READ_ONCE(sk->sk_shutdown); |
546 | if (shutdown == SHUTDOWN_MASK || state == TCP_CLOSE) | |
a9a08845 | 547 | mask |= EPOLLHUP; |
e14cadfd | 548 | if (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 | 554 | int target = sock_rcvlowat(sk, 0, INT_MAX); |
7b6a893a | 555 | u16 urg_data = READ_ONCE(tp->urg_data); |
c7004482 | 556 | |
b96c51bd | 557 | if (unlikely(urg_data) && |
7b6a893a ED |
558 | READ_ONCE(tp->urg_seq) == READ_ONCE(tp->copied_seq) && |
559 | !sock_flag(sk, SOCK_URGINLINE)) | |
b634f875 | 560 | target++; |
c7004482 | 561 | |
05dc72ab | 562 | if (tcp_stream_is_readable(sk, target)) |
a9a08845 | 563 | mask |= EPOLLIN | EPOLLRDNORM; |
1da177e4 | 564 | |
e14cadfd | 565 | if (!(shutdown & SEND_SHUTDOWN)) { |
8ba3c9d1 | 566 | if (__sk_stream_is_writeable(sk, 1)) { |
a9a08845 | 567 | mask |= EPOLLOUT | EPOLLWRNORM; |
1da177e4 | 568 | } else { /* send SIGIO later */ |
9cd3e072 | 569 | sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 LT |
570 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
571 | ||
572 | /* Race breaker. If space is freed after | |
573 | * wspace test but before the flags are set, | |
3c715127 | 574 | * IO signal will be lost. Memory barrier |
575 | * pairs with the input side. | |
1da177e4 | 576 | */ |
3c715127 | 577 | smp_mb__after_atomic(); |
8ba3c9d1 | 578 | if (__sk_stream_is_writeable(sk, 1)) |
a9a08845 | 579 | mask |= EPOLLOUT | EPOLLWRNORM; |
1da177e4 | 580 | } |
d84ba638 | 581 | } else |
a9a08845 | 582 | mask |= EPOLLOUT | EPOLLWRNORM; |
1da177e4 | 583 | |
7b6a893a | 584 | if (urg_data & TCP_URG_VALID) |
a9a08845 | 585 | mask |= EPOLLPRI; |
08e39c0d ED |
586 | } else if (state == TCP_SYN_SENT && |
587 | inet_test_bit(DEFER_CONNECT, sk)) { | |
19f6d3f3 | 588 | /* Active TCP fastopen socket with defer_connect |
a9a08845 | 589 | * Return EPOLLOUT so application can call write() |
19f6d3f3 WW |
590 | * in order for kernel to generate SYN+data |
591 | */ | |
a9a08845 | 592 | mask |= EPOLLOUT | EPOLLWRNORM; |
1da177e4 | 593 | } |
a4d25803 TM |
594 | /* This barrier is coupled with smp_wmb() in tcp_reset() */ |
595 | smp_rmb(); | |
e13ec3da ED |
596 | if (READ_ONCE(sk->sk_err) || |
597 | !skb_queue_empty_lockless(&sk->sk_error_queue)) | |
a9a08845 | 598 | mask |= EPOLLERR; |
a4d25803 | 599 | |
1da177e4 LT |
600 | return mask; |
601 | } | |
a11e1d43 | 602 | EXPORT_SYMBOL(tcp_poll); |
1da177e4 | 603 | |
e1d001fa | 604 | int tcp_ioctl(struct sock *sk, int cmd, int *karg) |
1da177e4 LT |
605 | { |
606 | struct tcp_sock *tp = tcp_sk(sk); | |
607 | int answ; | |
0e71c55c | 608 | bool slow; |
1da177e4 LT |
609 | |
610 | switch (cmd) { | |
611 | case SIOCINQ: | |
612 | if (sk->sk_state == TCP_LISTEN) | |
613 | return -EINVAL; | |
614 | ||
0e71c55c | 615 | slow = lock_sock_fast(sk); |
473bd239 | 616 | answ = tcp_inq(sk); |
0e71c55c | 617 | unlock_sock_fast(sk, slow); |
1da177e4 LT |
618 | break; |
619 | case SIOCATMARK: | |
7b6a893a | 620 | answ = READ_ONCE(tp->urg_data) && |
d9b55bf7 | 621 | READ_ONCE(tp->urg_seq) == READ_ONCE(tp->copied_seq); |
1da177e4 LT |
622 | break; |
623 | case SIOCOUTQ: | |
624 | if (sk->sk_state == TCP_LISTEN) | |
625 | return -EINVAL; | |
626 | ||
627 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
628 | answ = 0; | |
629 | else | |
0f317464 | 630 | answ = READ_ONCE(tp->write_seq) - tp->snd_una; |
1da177e4 | 631 | break; |
2f4e1b39 MS |
632 | case SIOCOUTQNSD: |
633 | if (sk->sk_state == TCP_LISTEN) | |
634 | return -EINVAL; | |
635 | ||
636 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
637 | answ = 0; | |
638 | else | |
e0d694d6 ED |
639 | answ = READ_ONCE(tp->write_seq) - |
640 | READ_ONCE(tp->snd_nxt); | |
2f4e1b39 | 641 | break; |
1da177e4 LT |
642 | default: |
643 | return -ENOIOCTLCMD; | |
3ff50b79 | 644 | } |
1da177e4 | 645 | |
e1d001fa BL |
646 | *karg = answ; |
647 | return 0; | |
1da177e4 | 648 | } |
4bc2f18b | 649 | EXPORT_SYMBOL(tcp_ioctl); |
1da177e4 | 650 | |
04d8825c | 651 | void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb) |
1da177e4 | 652 | { |
4de075e0 | 653 | TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH; |
1da177e4 LT |
654 | tp->pushed_seq = tp->write_seq; |
655 | } | |
656 | ||
a2a385d6 | 657 | static inline bool forced_push(const struct tcp_sock *tp) |
1da177e4 LT |
658 | { |
659 | return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1)); | |
660 | } | |
661 | ||
04d8825c | 662 | void tcp_skb_entail(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 663 | { |
9e412ba7 | 664 | struct tcp_sock *tp = tcp_sk(sk); |
352d4800 ACM |
665 | struct tcp_skb_cb *tcb = TCP_SKB_CB(skb); |
666 | ||
352d4800 | 667 | tcb->seq = tcb->end_seq = tp->write_seq; |
4de075e0 | 668 | tcb->tcp_flags = TCPHDR_ACK; |
f4a775d1 | 669 | __skb_header_release(skb); |
fe067e8a | 670 | tcp_add_write_queue_tail(sk, skb); |
ab4e846a | 671 | sk_wmem_queued_add(sk, skb->truesize); |
3ab224be | 672 | sk_mem_charge(sk, skb->truesize); |
89ebd197 | 673 | if (tp->nonagle & TCP_NAGLE_PUSH) |
e905a9ed | 674 | tp->nonagle &= ~TCP_NAGLE_PUSH; |
6f021c62 ED |
675 | |
676 | tcp_slow_start_after_idle_check(sk); | |
1da177e4 LT |
677 | } |
678 | ||
afeca340 | 679 | static inline void tcp_mark_urg(struct tcp_sock *tp, int flags) |
1da177e4 | 680 | { |
33f5f57e | 681 | if (flags & MSG_OOB) |
1da177e4 | 682 | tp->snd_up = tp->write_seq; |
1da177e4 LT |
683 | } |
684 | ||
f54b3111 | 685 | /* If a not yet filled skb is pushed, do not send it if |
a181ceb5 | 686 | * we have data packets in Qdisc or NIC queues : |
f54b3111 ED |
687 | * Because TX completion will happen shortly, it gives a chance |
688 | * to coalesce future sendmsg() payload into this skb, without | |
689 | * need for a timer, and with no latency trade off. | |
690 | * As packets containing data payload have a bigger truesize | |
a181ceb5 ED |
691 | * than pure acks (dataless) packets, the last checks prevent |
692 | * autocorking if we only have an ACK in Qdisc/NIC queues, | |
693 | * or if TX completion was delayed after we processed ACK packet. | |
f54b3111 ED |
694 | */ |
695 | static bool tcp_should_autocork(struct sock *sk, struct sk_buff *skb, | |
696 | int size_goal) | |
1da177e4 | 697 | { |
f54b3111 | 698 | return skb->len < size_goal && |
85225e6f | 699 | READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_autocorking) && |
114f39fe | 700 | !tcp_rtx_queue_empty(sk) && |
b0de0cf4 ED |
701 | refcount_read(&sk->sk_wmem_alloc) > skb->truesize && |
702 | tcp_skb_can_collapse_to(skb); | |
f54b3111 ED |
703 | } |
704 | ||
35b2c321 MM |
705 | void tcp_push(struct sock *sk, int flags, int mss_now, |
706 | int nonagle, int size_goal) | |
f54b3111 ED |
707 | { |
708 | struct tcp_sock *tp = tcp_sk(sk); | |
709 | struct sk_buff *skb; | |
afeca340 | 710 | |
f54b3111 | 711 | skb = tcp_write_queue_tail(sk); |
75c119af ED |
712 | if (!skb) |
713 | return; | |
f54b3111 ED |
714 | if (!(flags & MSG_MORE) || forced_push(tp)) |
715 | tcp_mark_push(tp, skb); | |
716 | ||
717 | tcp_mark_urg(tp, flags); | |
718 | ||
719 | if (tcp_should_autocork(sk, skb, size_goal)) { | |
720 | ||
721 | /* avoid atomic op if TSQ_THROTTLED bit is already set */ | |
7aa5470c | 722 | if (!test_bit(TSQ_THROTTLED, &sk->sk_tsq_flags)) { |
f54b3111 | 723 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAUTOCORKING); |
7aa5470c | 724 | set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags); |
f54b3111 | 725 | } |
a181ceb5 ED |
726 | /* It is possible TX completion already happened |
727 | * before we set TSQ_THROTTLED. | |
728 | */ | |
14afee4b | 729 | if (refcount_read(&sk->sk_wmem_alloc) > skb->truesize) |
a181ceb5 | 730 | return; |
1da177e4 | 731 | } |
f54b3111 ED |
732 | |
733 | if (flags & MSG_MORE) | |
734 | nonagle = TCP_NAGLE_CORK; | |
735 | ||
736 | __tcp_push_pending_frames(sk, mss_now, nonagle); | |
1da177e4 LT |
737 | } |
738 | ||
6ff7751d AB |
739 | static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, |
740 | unsigned int offset, size_t len) | |
9c55e01c JA |
741 | { |
742 | struct tcp_splice_state *tss = rd_desc->arg.data; | |
33966dd0 | 743 | int ret; |
9c55e01c | 744 | |
a60e3cc7 | 745 | ret = skb_splice_bits(skb, skb->sk, offset, tss->pipe, |
25869262 | 746 | min(rd_desc->count, len), tss->flags); |
33966dd0 WT |
747 | if (ret > 0) |
748 | rd_desc->count -= ret; | |
749 | return ret; | |
9c55e01c JA |
750 | } |
751 | ||
752 | static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss) | |
753 | { | |
754 | /* Store TCP splice context information in read_descriptor_t. */ | |
755 | read_descriptor_t rd_desc = { | |
756 | .arg.data = tss, | |
33966dd0 | 757 | .count = tss->len, |
9c55e01c JA |
758 | }; |
759 | ||
760 | return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv); | |
761 | } | |
762 | ||
763 | /** | |
764 | * tcp_splice_read - splice data from TCP socket to a pipe | |
765 | * @sock: socket to splice from | |
766 | * @ppos: position (not valid) | |
767 | * @pipe: pipe to splice to | |
768 | * @len: number of bytes to splice | |
769 | * @flags: splice modifier flags | |
770 | * | |
771 | * Description: | |
772 | * Will read pages from given socket and fill them into a pipe. | |
773 | * | |
774 | **/ | |
775 | ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos, | |
776 | struct pipe_inode_info *pipe, size_t len, | |
777 | unsigned int flags) | |
778 | { | |
779 | struct sock *sk = sock->sk; | |
780 | struct tcp_splice_state tss = { | |
781 | .pipe = pipe, | |
782 | .len = len, | |
783 | .flags = flags, | |
784 | }; | |
785 | long timeo; | |
786 | ssize_t spliced; | |
787 | int ret; | |
788 | ||
3a047bf8 | 789 | sock_rps_record_flow(sk); |
9c55e01c JA |
790 | /* |
791 | * We can't seek on a socket input | |
792 | */ | |
793 | if (unlikely(*ppos)) | |
794 | return -ESPIPE; | |
795 | ||
796 | ret = spliced = 0; | |
797 | ||
798 | lock_sock(sk); | |
799 | ||
42324c62 | 800 | timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK); |
9c55e01c JA |
801 | while (tss.len) { |
802 | ret = __tcp_splice_read(sk, &tss); | |
803 | if (ret < 0) | |
804 | break; | |
805 | else if (!ret) { | |
806 | if (spliced) | |
807 | break; | |
9c55e01c JA |
808 | if (sock_flag(sk, SOCK_DONE)) |
809 | break; | |
810 | if (sk->sk_err) { | |
811 | ret = sock_error(sk); | |
812 | break; | |
813 | } | |
814 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
815 | break; | |
816 | if (sk->sk_state == TCP_CLOSE) { | |
817 | /* | |
818 | * This occurs when user tries to read | |
819 | * from never connected socket. | |
820 | */ | |
c47078d6 | 821 | ret = -ENOTCONN; |
9c55e01c JA |
822 | break; |
823 | } | |
824 | if (!timeo) { | |
825 | ret = -EAGAIN; | |
826 | break; | |
827 | } | |
ccf7abb9 ED |
828 | /* if __tcp_splice_read() got nothing while we have |
829 | * an skb in receive queue, we do not want to loop. | |
830 | * This might happen with URG data. | |
831 | */ | |
832 | if (!skb_queue_empty(&sk->sk_receive_queue)) | |
833 | break; | |
dfbafc99 | 834 | sk_wait_data(sk, &timeo, NULL); |
9c55e01c JA |
835 | if (signal_pending(current)) { |
836 | ret = sock_intr_errno(timeo); | |
837 | break; | |
838 | } | |
839 | continue; | |
840 | } | |
841 | tss.len -= ret; | |
842 | spliced += ret; | |
843 | ||
2fe11c9d | 844 | if (!tss.len || !timeo) |
33966dd0 | 845 | break; |
9c55e01c JA |
846 | release_sock(sk); |
847 | lock_sock(sk); | |
848 | ||
849 | if (sk->sk_err || sk->sk_state == TCP_CLOSE || | |
33966dd0 | 850 | (sk->sk_shutdown & RCV_SHUTDOWN) || |
9c55e01c JA |
851 | signal_pending(current)) |
852 | break; | |
853 | } | |
854 | ||
855 | release_sock(sk); | |
856 | ||
857 | if (spliced) | |
858 | return spliced; | |
859 | ||
860 | return ret; | |
861 | } | |
4bc2f18b | 862 | EXPORT_SYMBOL(tcp_splice_read); |
9c55e01c | 863 | |
5882efff | 864 | struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, gfp_t gfp, |
f8dd3b8d | 865 | bool force_schedule) |
f561d0f2 PE |
866 | { |
867 | struct sk_buff *skb; | |
868 | ||
5882efff | 869 | skb = alloc_skb_fclone(MAX_TCP_HEADER, gfp); |
8e4d980a | 870 | if (likely(skb)) { |
eb934478 | 871 | bool mem_scheduled; |
8e4d980a | 872 | |
9b65b17d | 873 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
eb934478 ED |
874 | if (force_schedule) { |
875 | mem_scheduled = true; | |
8e4d980a ED |
876 | sk_forced_mem_schedule(sk, skb->truesize); |
877 | } else { | |
eb934478 | 878 | mem_scheduled = sk_wmem_schedule(sk, skb->truesize); |
8e4d980a | 879 | } |
eb934478 | 880 | if (likely(mem_scheduled)) { |
8a794df6 | 881 | skb_reserve(skb, MAX_TCP_HEADER); |
a52fe46e | 882 | skb->ip_summed = CHECKSUM_PARTIAL; |
e2080072 | 883 | INIT_LIST_HEAD(&skb->tcp_tsorted_anchor); |
f561d0f2 PE |
884 | return skb; |
885 | } | |
886 | __kfree_skb(skb); | |
887 | } else { | |
5c52ba17 | 888 | sk->sk_prot->enter_memory_pressure(sk); |
f561d0f2 PE |
889 | sk_stream_moderate_sndbuf(sk); |
890 | } | |
891 | return NULL; | |
892 | } | |
893 | ||
0c54b85f IJ |
894 | static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now, |
895 | int large_allowed) | |
896 | { | |
897 | struct tcp_sock *tp = tcp_sk(sk); | |
6c09fa09 | 898 | u32 new_size_goal, size_goal; |
605ad7f1 | 899 | |
74d4a8f8 | 900 | if (!large_allowed) |
605ad7f1 ED |
901 | return mss_now; |
902 | ||
6c09fa09 | 903 | /* Note : tcp_tso_autosize() will eventually split this later */ |
ab14f180 | 904 | new_size_goal = tcp_bound_to_half_wnd(tp, sk->sk_gso_max_size); |
605ad7f1 ED |
905 | |
906 | /* We try hard to avoid divides here */ | |
907 | size_goal = tp->gso_segs * mss_now; | |
908 | if (unlikely(new_size_goal < size_goal || | |
909 | new_size_goal >= size_goal + mss_now)) { | |
910 | tp->gso_segs = min_t(u16, new_size_goal / mss_now, | |
911 | sk->sk_gso_max_segs); | |
912 | size_goal = tp->gso_segs * mss_now; | |
0c54b85f IJ |
913 | } |
914 | ||
605ad7f1 | 915 | return max(size_goal, mss_now); |
0c54b85f IJ |
916 | } |
917 | ||
35b2c321 | 918 | int tcp_send_mss(struct sock *sk, int *size_goal, int flags) |
0c54b85f IJ |
919 | { |
920 | int mss_now; | |
921 | ||
922 | mss_now = tcp_current_mss(sk); | |
923 | *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB)); | |
924 | ||
925 | return mss_now; | |
926 | } | |
927 | ||
dc97391e DH |
928 | /* In some cases, both sendmsg() could have added an skb to the write queue, |
929 | * but failed adding payload on it. We need to remove it to consume less | |
930 | * memory, but more importantly be able to generate EPOLLOUT for Edge Trigger | |
931 | * epoll() users. | |
fdfc5c85 | 932 | */ |
27728ba8 | 933 | void tcp_remove_empty_skb(struct sock *sk) |
fdfc5c85 | 934 | { |
27728ba8 ED |
935 | struct sk_buff *skb = tcp_write_queue_tail(sk); |
936 | ||
cf12e6f9 | 937 | if (skb && TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) { |
fdfc5c85 ED |
938 | tcp_unlink_write_queue(skb, sk); |
939 | if (tcp_write_queue_empty(sk)) | |
940 | tcp_chrono_stop(sk, TCP_CHRONO_BUSY); | |
03271f3a | 941 | tcp_wmem_free_skb(sk, skb); |
fdfc5c85 ED |
942 | } |
943 | } | |
944 | ||
f8d9d938 ED |
945 | /* skb changing from pure zc to mixed, must charge zc */ |
946 | static int tcp_downgrade_zcopy_pure(struct sock *sk, struct sk_buff *skb) | |
947 | { | |
948 | if (unlikely(skb_zcopy_pure(skb))) { | |
949 | u32 extra = skb->truesize - | |
950 | SKB_TRUESIZE(skb_end_offset(skb)); | |
951 | ||
952 | if (!sk_wmem_schedule(sk, extra)) | |
953 | return -ENOMEM; | |
954 | ||
955 | sk_mem_charge(sk, extra); | |
956 | skb_shinfo(skb)->flags &= ~SKBFL_PURE_ZEROCOPY; | |
957 | } | |
958 | return 0; | |
959 | } | |
960 | ||
f54755f6 | 961 | |
fbf93406 | 962 | int tcp_wmem_schedule(struct sock *sk, int copy) |
849b425c ED |
963 | { |
964 | int left; | |
965 | ||
966 | if (likely(sk_wmem_schedule(sk, copy))) | |
967 | return copy; | |
968 | ||
969 | /* We could be in trouble if we have nothing queued. | |
970 | * Use whatever is left in sk->sk_forward_alloc and tcp_wmem[0] | |
971 | * to guarantee some progress. | |
972 | */ | |
973 | left = sock_net(sk)->ipv4.sysctl_tcp_wmem[0] - sk->sk_wmem_queued; | |
974 | if (left > 0) | |
975 | sk_forced_mem_schedule(sk, min(left, copy)); | |
976 | return min(copy, sk->sk_forward_alloc); | |
977 | } | |
978 | ||
cf60af03 YC |
979 | void tcp_free_fastopen_req(struct tcp_sock *tp) |
980 | { | |
00db4124 | 981 | if (tp->fastopen_req) { |
cf60af03 YC |
982 | kfree(tp->fastopen_req); |
983 | tp->fastopen_req = NULL; | |
984 | } | |
985 | } | |
986 | ||
3242abeb BH |
987 | int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied, |
988 | size_t size, struct ubuf_info *uarg) | |
cf60af03 YC |
989 | { |
990 | struct tcp_sock *tp = tcp_sk(sk); | |
19f6d3f3 | 991 | struct inet_sock *inet = inet_sk(sk); |
ba615f67 | 992 | struct sockaddr *uaddr = msg->msg_name; |
cf60af03 YC |
993 | int err, flags; |
994 | ||
5a542133 KI |
995 | if (!(READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen) & |
996 | TFO_CLIENT_ENABLE) || | |
ba615f67 WW |
997 | (uaddr && msg->msg_namelen >= sizeof(uaddr->sa_family) && |
998 | uaddr->sa_family == AF_UNSPEC)) | |
cf60af03 | 999 | return -EOPNOTSUPP; |
00db4124 | 1000 | if (tp->fastopen_req) |
cf60af03 YC |
1001 | return -EALREADY; /* Another Fast Open is in progress */ |
1002 | ||
1003 | tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request), | |
1004 | sk->sk_allocation); | |
51456b29 | 1005 | if (unlikely(!tp->fastopen_req)) |
cf60af03 YC |
1006 | return -ENOBUFS; |
1007 | tp->fastopen_req->data = msg; | |
f5ddcbbb | 1008 | tp->fastopen_req->size = size; |
f859a448 | 1009 | tp->fastopen_req->uarg = uarg; |
cf60af03 | 1010 | |
08e39c0d | 1011 | if (inet_test_bit(DEFER_CONNECT, sk)) { |
19f6d3f3 WW |
1012 | err = tcp_connect(sk); |
1013 | /* Same failure procedure as in tcp_v4/6_connect */ | |
1014 | if (err) { | |
1015 | tcp_set_state(sk, TCP_CLOSE); | |
1016 | inet->inet_dport = 0; | |
1017 | sk->sk_route_caps = 0; | |
1018 | } | |
1019 | } | |
cf60af03 | 1020 | flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0; |
ba615f67 | 1021 | err = __inet_stream_connect(sk->sk_socket, uaddr, |
3979ad7e | 1022 | msg->msg_namelen, flags, 1); |
7db92362 WW |
1023 | /* fastopen_req could already be freed in __inet_stream_connect |
1024 | * if the connection times out or gets rst | |
1025 | */ | |
1026 | if (tp->fastopen_req) { | |
1027 | *copied = tp->fastopen_req->copied; | |
1028 | tcp_free_fastopen_req(tp); | |
08e39c0d | 1029 | inet_clear_bit(DEFER_CONNECT, sk); |
7db92362 | 1030 | } |
cf60af03 YC |
1031 | return err; |
1032 | } | |
1033 | ||
306b13eb | 1034 | int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size) |
1da177e4 | 1035 | { |
1da177e4 | 1036 | struct tcp_sock *tp = tcp_sk(sk); |
f214f915 | 1037 | struct ubuf_info *uarg = NULL; |
1da177e4 | 1038 | struct sk_buff *skb; |
c14ac945 | 1039 | struct sockcm_cookie sockc; |
57be5bda AV |
1040 | int flags, err, copied = 0; |
1041 | int mss_now = 0, size_goal, copied_syn = 0; | |
1a991488 | 1042 | int process_backlog = 0; |
270a1c3d | 1043 | int zc = 0; |
1da177e4 LT |
1044 | long timeo; |
1045 | ||
1da177e4 | 1046 | flags = msg->msg_flags; |
f214f915 | 1047 | |
eb315a7d | 1048 | if ((flags & MSG_ZEROCOPY) && size) { |
eb315a7d PB |
1049 | if (msg->msg_ubuf) { |
1050 | uarg = msg->msg_ubuf; | |
270a1c3d DH |
1051 | if (sk->sk_route_caps & NETIF_F_SG) |
1052 | zc = MSG_ZEROCOPY; | |
eb315a7d | 1053 | } else if (sock_flag(sk, SOCK_ZEROCOPY)) { |
eea96a3e | 1054 | skb = tcp_write_queue_tail(sk); |
eb315a7d PB |
1055 | uarg = msg_zerocopy_realloc(sk, size, skb_zcopy(skb)); |
1056 | if (!uarg) { | |
1057 | err = -ENOBUFS; | |
1058 | goto out_err; | |
1059 | } | |
270a1c3d DH |
1060 | if (sk->sk_route_caps & NETIF_F_SG) |
1061 | zc = MSG_ZEROCOPY; | |
1062 | else | |
e7d2b510 | 1063 | uarg_to_msgzc(uarg)->zerocopy = 0; |
eb315a7d | 1064 | } |
270a1c3d DH |
1065 | } else if (unlikely(msg->msg_flags & MSG_SPLICE_PAGES) && size) { |
1066 | if (sk->sk_route_caps & NETIF_F_SG) | |
1067 | zc = MSG_SPLICE_PAGES; | |
f214f915 WB |
1068 | } |
1069 | ||
08e39c0d ED |
1070 | if (unlikely(flags & MSG_FASTOPEN || |
1071 | inet_test_bit(DEFER_CONNECT, sk)) && | |
16ae6aa1 | 1072 | !tp->repair) { |
f859a448 | 1073 | err = tcp_sendmsg_fastopen(sk, msg, &copied_syn, size, uarg); |
cf60af03 YC |
1074 | if (err == -EINPROGRESS && copied_syn > 0) |
1075 | goto out; | |
1076 | else if (err) | |
1077 | goto out_err; | |
cf60af03 YC |
1078 | } |
1079 | ||
1da177e4 LT |
1080 | timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); |
1081 | ||
d7722e85 SHY |
1082 | tcp_rate_check_app_limited(sk); /* is sending application-limited? */ |
1083 | ||
8336886f JC |
1084 | /* Wait for a connection to finish. One exception is TCP Fast Open |
1085 | * (passive side) where data is allowed to be sent before a connection | |
1086 | * is fully established. | |
1087 | */ | |
1088 | if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) && | |
1089 | !tcp_passive_fastopen(sk)) { | |
686a5624 YM |
1090 | err = sk_stream_wait_connect(sk, &timeo); |
1091 | if (err != 0) | |
cf60af03 | 1092 | goto do_error; |
8336886f | 1093 | } |
1da177e4 | 1094 | |
c0e88ff0 PE |
1095 | if (unlikely(tp->repair)) { |
1096 | if (tp->repair_queue == TCP_RECV_QUEUE) { | |
1097 | copied = tcp_send_rcvq(sk, msg, size); | |
5924f17a | 1098 | goto out_nopush; |
c0e88ff0 PE |
1099 | } |
1100 | ||
1101 | err = -EINVAL; | |
1102 | if (tp->repair_queue == TCP_NO_QUEUE) | |
1103 | goto out_err; | |
1104 | ||
1105 | /* 'common' sending to sendq */ | |
1106 | } | |
1107 | ||
657a0667 | 1108 | sockcm_init(&sockc, sk); |
c14ac945 SHY |
1109 | if (msg->msg_controllen) { |
1110 | err = sock_cmsg_send(sk, msg, &sockc); | |
1111 | if (unlikely(err)) { | |
1112 | err = -EINVAL; | |
1113 | goto out_err; | |
1114 | } | |
1115 | } | |
1116 | ||
1da177e4 | 1117 | /* This should be in poll */ |
9cd3e072 | 1118 | sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 | 1119 | |
1da177e4 | 1120 | /* Ok commence sending. */ |
1da177e4 LT |
1121 | copied = 0; |
1122 | ||
d41a69f1 ED |
1123 | restart: |
1124 | mss_now = tcp_send_mss(sk, &size_goal, flags); | |
1125 | ||
1da177e4 LT |
1126 | err = -EPIPE; |
1127 | if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) | |
79d8665b | 1128 | goto do_error; |
1da177e4 | 1129 | |
01e97e65 | 1130 | while (msg_data_left(msg)) { |
270a1c3d | 1131 | ssize_t copy = 0; |
1da177e4 | 1132 | |
57be5bda | 1133 | skb = tcp_write_queue_tail(sk); |
65ec6097 ED |
1134 | if (skb) |
1135 | copy = size_goal - skb->len; | |
1da177e4 | 1136 | |
c134ecb8 | 1137 | if (copy <= 0 || !tcp_skb_can_collapse_to(skb)) { |
3613b3db ED |
1138 | bool first_skb; |
1139 | ||
1da177e4 | 1140 | new_segment: |
57be5bda | 1141 | if (!sk_stream_memory_free(sk)) |
afb83012 | 1142 | goto wait_for_space; |
1da177e4 | 1143 | |
1a991488 ED |
1144 | if (unlikely(process_backlog >= 16)) { |
1145 | process_backlog = 0; | |
1146 | if (sk_flush_backlog(sk)) | |
1147 | goto restart; | |
d4011239 | 1148 | } |
75c119af | 1149 | first_skb = tcp_rtx_and_write_queues_empty(sk); |
5882efff | 1150 | skb = tcp_stream_alloc_skb(sk, sk->sk_allocation, |
f8dd3b8d | 1151 | first_skb); |
57be5bda | 1152 | if (!skb) |
afb83012 | 1153 | goto wait_for_space; |
1da177e4 | 1154 | |
1a991488 | 1155 | process_backlog++; |
1da177e4 | 1156 | |
04d8825c | 1157 | tcp_skb_entail(sk, skb); |
57be5bda | 1158 | copy = size_goal; |
9d186cac | 1159 | |
57be5bda | 1160 | /* All packets are restored as if they have |
d3edd06e | 1161 | * already been sent. skb_mstamp_ns isn't set to |
57be5bda AV |
1162 | * avoid wrong rtt estimation. |
1163 | */ | |
1164 | if (tp->repair) | |
1165 | TCP_SKB_CB(skb)->sacked |= TCPCB_REPAIRED; | |
1166 | } | |
1da177e4 | 1167 | |
57be5bda | 1168 | /* Try to append data to the end of skb. */ |
01e97e65 AV |
1169 | if (copy > msg_data_left(msg)) |
1170 | copy = msg_data_left(msg); | |
57be5bda | 1171 | |
270a1c3d | 1172 | if (zc == 0) { |
57be5bda AV |
1173 | bool merge = true; |
1174 | int i = skb_shinfo(skb)->nr_frags; | |
1175 | struct page_frag *pfrag = sk_page_frag(sk); | |
1176 | ||
1177 | if (!sk_page_frag_refill(sk, pfrag)) | |
afb83012 | 1178 | goto wait_for_space; |
ef015786 | 1179 | |
57be5bda AV |
1180 | if (!skb_can_coalesce(skb, i, pfrag->page, |
1181 | pfrag->offset)) { | |
657b991a | 1182 | if (i >= READ_ONCE(sysctl_max_skb_frags)) { |
57be5bda AV |
1183 | tcp_mark_push(tp, skb); |
1184 | goto new_segment; | |
1da177e4 | 1185 | } |
57be5bda | 1186 | merge = false; |
1da177e4 LT |
1187 | } |
1188 | ||
57be5bda AV |
1189 | copy = min_t(int, copy, pfrag->size - pfrag->offset); |
1190 | ||
eb315a7d PB |
1191 | if (unlikely(skb_zcopy_pure(skb) || skb_zcopy_managed(skb))) { |
1192 | if (tcp_downgrade_zcopy_pure(sk, skb)) | |
1193 | goto wait_for_space; | |
1194 | skb_zcopy_downgrade_managed(skb); | |
1195 | } | |
849b425c ED |
1196 | |
1197 | copy = tcp_wmem_schedule(sk, copy); | |
1198 | if (!copy) | |
afb83012 | 1199 | goto wait_for_space; |
1da177e4 | 1200 | |
57be5bda AV |
1201 | err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb, |
1202 | pfrag->page, | |
1203 | pfrag->offset, | |
1204 | copy); | |
1205 | if (err) | |
1206 | goto do_error; | |
1da177e4 | 1207 | |
57be5bda AV |
1208 | /* Update the skb. */ |
1209 | if (merge) { | |
1210 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); | |
1211 | } else { | |
1212 | skb_fill_page_desc(skb, i, pfrag->page, | |
1213 | pfrag->offset, copy); | |
4e33e346 | 1214 | page_ref_inc(pfrag->page); |
4ed2d765 | 1215 | } |
57be5bda | 1216 | pfrag->offset += copy; |
270a1c3d | 1217 | } else if (zc == MSG_ZEROCOPY) { |
9b65b17d TA |
1218 | /* First append to a fragless skb builds initial |
1219 | * pure zerocopy skb | |
1220 | */ | |
1221 | if (!skb->len) | |
1222 | skb_shinfo(skb)->flags |= SKBFL_PURE_ZEROCOPY; | |
1223 | ||
1224 | if (!skb_zcopy_pure(skb)) { | |
849b425c ED |
1225 | copy = tcp_wmem_schedule(sk, copy); |
1226 | if (!copy) | |
9b65b17d TA |
1227 | goto wait_for_space; |
1228 | } | |
358ed624 | 1229 | |
f214f915 | 1230 | err = skb_zerocopy_iter_stream(sk, skb, msg, copy, uarg); |
111856c7 WB |
1231 | if (err == -EMSGSIZE || err == -EEXIST) { |
1232 | tcp_mark_push(tp, skb); | |
f214f915 | 1233 | goto new_segment; |
111856c7 | 1234 | } |
f214f915 WB |
1235 | if (err < 0) |
1236 | goto do_error; | |
1237 | copy = err; | |
270a1c3d DH |
1238 | } else if (zc == MSG_SPLICE_PAGES) { |
1239 | /* Splice in data if we can; copy if we can't. */ | |
1240 | if (tcp_downgrade_zcopy_pure(sk, skb)) | |
1241 | goto wait_for_space; | |
1242 | copy = tcp_wmem_schedule(sk, copy); | |
1243 | if (!copy) | |
1244 | goto wait_for_space; | |
1245 | ||
1246 | err = skb_splice_from_iter(skb, &msg->msg_iter, copy, | |
1247 | sk->sk_allocation); | |
1248 | if (err < 0) { | |
1249 | if (err == -EMSGSIZE) { | |
1250 | tcp_mark_push(tp, skb); | |
1251 | goto new_segment; | |
1252 | } | |
1253 | goto do_error; | |
1254 | } | |
1255 | copy = err; | |
1256 | ||
1257 | if (!(flags & MSG_NO_SHARED_FRAGS)) | |
1258 | skb_shinfo(skb)->flags |= SKBFL_SHARED_FRAG; | |
1259 | ||
1260 | sk_wmem_queued_add(sk, copy); | |
1261 | sk_mem_charge(sk, copy); | |
57be5bda AV |
1262 | } |
1263 | ||
1264 | if (!copied) | |
1265 | TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH; | |
1266 | ||
0f317464 | 1267 | WRITE_ONCE(tp->write_seq, tp->write_seq + copy); |
57be5bda AV |
1268 | TCP_SKB_CB(skb)->end_seq += copy; |
1269 | tcp_skb_pcount_set(skb, 0); | |
1da177e4 | 1270 | |
57be5bda | 1271 | copied += copy; |
01e97e65 | 1272 | if (!msg_data_left(msg)) { |
c134ecb8 MKL |
1273 | if (unlikely(flags & MSG_EOR)) |
1274 | TCP_SKB_CB(skb)->eor = 1; | |
57be5bda AV |
1275 | goto out; |
1276 | } | |
1da177e4 | 1277 | |
65ec6097 | 1278 | if (skb->len < size_goal || (flags & MSG_OOB) || unlikely(tp->repair)) |
1da177e4 LT |
1279 | continue; |
1280 | ||
57be5bda AV |
1281 | if (forced_push(tp)) { |
1282 | tcp_mark_push(tp, skb); | |
1283 | __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH); | |
1284 | } else if (skb == tcp_send_head(sk)) | |
1285 | tcp_push_one(sk, mss_now); | |
1286 | continue; | |
1287 | ||
afb83012 | 1288 | wait_for_space: |
57be5bda | 1289 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
57be5bda AV |
1290 | if (copied) |
1291 | tcp_push(sk, flags & ~MSG_MORE, mss_now, | |
1292 | TCP_NAGLE_PUSH, size_goal); | |
1da177e4 | 1293 | |
686a5624 YM |
1294 | err = sk_stream_wait_memory(sk, &timeo); |
1295 | if (err != 0) | |
57be5bda | 1296 | goto do_error; |
1da177e4 | 1297 | |
57be5bda | 1298 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
1299 | } |
1300 | ||
1301 | out: | |
ad02c4f5 | 1302 | if (copied) { |
4e8cc228 | 1303 | tcp_tx_timestamp(sk, sockc.tsflags); |
f54b3111 | 1304 | tcp_push(sk, flags, mss_now, tp->nonagle, size_goal); |
ad02c4f5 | 1305 | } |
5924f17a | 1306 | out_nopush: |
a7533584 PB |
1307 | /* msg->msg_ubuf is pinned by the caller so we don't take extra refs */ |
1308 | if (uarg && !msg->msg_ubuf) | |
1309 | net_zcopy_put(uarg); | |
cf60af03 | 1310 | return copied + copied_syn; |
1da177e4 | 1311 | |
fdfc5c85 | 1312 | do_error: |
27728ba8 | 1313 | tcp_remove_empty_skb(sk); |
1da177e4 | 1314 | |
cf60af03 | 1315 | if (copied + copied_syn) |
1da177e4 LT |
1316 | goto out; |
1317 | out_err: | |
a7533584 PB |
1318 | /* msg->msg_ubuf is pinned by the caller so we don't take extra refs */ |
1319 | if (uarg && !msg->msg_ubuf) | |
1320 | net_zcopy_put_abort(uarg, true); | |
1da177e4 | 1321 | err = sk_stream_error(sk, flags, err); |
ce5ec440 | 1322 | /* make sure we wake any epoll edge trigger waiter */ |
216808c6 | 1323 | if (unlikely(tcp_rtx_and_write_queues_empty(sk) && err == -EAGAIN)) { |
ce5ec440 | 1324 | sk->sk_write_space(sk); |
b0f71bd3 FY |
1325 | tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED); |
1326 | } | |
1da177e4 LT |
1327 | return err; |
1328 | } | |
774c4673 | 1329 | EXPORT_SYMBOL_GPL(tcp_sendmsg_locked); |
306b13eb TH |
1330 | |
1331 | int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size) | |
1332 | { | |
1333 | int ret; | |
1334 | ||
1335 | lock_sock(sk); | |
1336 | ret = tcp_sendmsg_locked(sk, msg, size); | |
1337 | release_sock(sk); | |
1338 | ||
1339 | return ret; | |
1340 | } | |
4bc2f18b | 1341 | EXPORT_SYMBOL(tcp_sendmsg); |
1da177e4 | 1342 | |
1d7e4538 DH |
1343 | void tcp_splice_eof(struct socket *sock) |
1344 | { | |
1345 | struct sock *sk = sock->sk; | |
1346 | struct tcp_sock *tp = tcp_sk(sk); | |
1347 | int mss_now, size_goal; | |
1348 | ||
1349 | if (!tcp_write_queue_tail(sk)) | |
1350 | return; | |
1351 | ||
1352 | lock_sock(sk); | |
1353 | mss_now = tcp_send_mss(sk, &size_goal, 0); | |
1354 | tcp_push(sk, 0, mss_now, tp->nonagle, size_goal); | |
1355 | release_sock(sk); | |
1356 | } | |
1357 | EXPORT_SYMBOL_GPL(tcp_splice_eof); | |
1358 | ||
1da177e4 LT |
1359 | /* |
1360 | * Handle reading urgent data. BSD has very simple semantics for | |
1361 | * this, no blocking and very strange errors 8) | |
1362 | */ | |
1363 | ||
377f0a08 | 1364 | static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags) |
1da177e4 LT |
1365 | { |
1366 | struct tcp_sock *tp = tcp_sk(sk); | |
1367 | ||
1368 | /* No URG data to read. */ | |
1369 | if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data || | |
1370 | tp->urg_data == TCP_URG_READ) | |
1371 | return -EINVAL; /* Yes this is right ! */ | |
1372 | ||
1373 | if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE)) | |
1374 | return -ENOTCONN; | |
1375 | ||
1376 | if (tp->urg_data & TCP_URG_VALID) { | |
1377 | int err = 0; | |
1378 | char c = tp->urg_data; | |
1379 | ||
1380 | if (!(flags & MSG_PEEK)) | |
7b6a893a | 1381 | WRITE_ONCE(tp->urg_data, TCP_URG_READ); |
1da177e4 LT |
1382 | |
1383 | /* Read urgent data. */ | |
1384 | msg->msg_flags |= MSG_OOB; | |
1385 | ||
1386 | if (len > 0) { | |
1387 | if (!(flags & MSG_TRUNC)) | |
7eab8d9e | 1388 | err = memcpy_to_msg(msg, &c, 1); |
1da177e4 LT |
1389 | len = 1; |
1390 | } else | |
1391 | msg->msg_flags |= MSG_TRUNC; | |
1392 | ||
1393 | return err ? -EFAULT : len; | |
1394 | } | |
1395 | ||
1396 | if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN)) | |
1397 | return 0; | |
1398 | ||
1399 | /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and | |
1400 | * the available implementations agree in this case: | |
1401 | * this call should never block, independent of the | |
1402 | * blocking state of the socket. | |
1403 | * Mike <pall@rz.uni-karlsruhe.de> | |
1404 | */ | |
1405 | return -EAGAIN; | |
1406 | } | |
1407 | ||
c0e88ff0 PE |
1408 | static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len) |
1409 | { | |
1410 | struct sk_buff *skb; | |
1411 | int copied = 0, err = 0; | |
1412 | ||
1413 | /* XXX -- need to support SO_PEEK_OFF */ | |
1414 | ||
75c119af ED |
1415 | skb_rbtree_walk(skb, &sk->tcp_rtx_queue) { |
1416 | err = skb_copy_datagram_msg(skb, 0, msg, skb->len); | |
1417 | if (err) | |
1418 | return err; | |
1419 | copied += skb->len; | |
1420 | } | |
1421 | ||
c0e88ff0 | 1422 | skb_queue_walk(&sk->sk_write_queue, skb) { |
51f3d02b | 1423 | err = skb_copy_datagram_msg(skb, 0, msg, skb->len); |
c0e88ff0 PE |
1424 | if (err) |
1425 | break; | |
1426 | ||
1427 | copied += skb->len; | |
1428 | } | |
1429 | ||
1430 | return err ?: copied; | |
1431 | } | |
1432 | ||
1da177e4 LT |
1433 | /* Clean up the receive buffer for full frames taken by the user, |
1434 | * then send an ACK if necessary. COPIED is the number of bytes | |
1435 | * tcp_recvmsg has given to the user so far, it speeds up the | |
1436 | * calculation of whether or not we must ACK for the sake of | |
1437 | * a window update. | |
1438 | */ | |
e5c6de5f | 1439 | void __tcp_cleanup_rbuf(struct sock *sk, int copied) |
1da177e4 LT |
1440 | { |
1441 | struct tcp_sock *tp = tcp_sk(sk); | |
a2a385d6 | 1442 | bool time_to_ack = false; |
1da177e4 | 1443 | |
463c84b9 ACM |
1444 | if (inet_csk_ack_scheduled(sk)) { |
1445 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
b6b6d653 ED |
1446 | |
1447 | if (/* Once-per-two-segments ACK was not sent by tcp_input.c */ | |
463c84b9 | 1448 | tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss || |
1da177e4 LT |
1449 | /* |
1450 | * If this read emptied read buffer, we send ACK, if | |
1451 | * connection is not bidirectional, user drained | |
1452 | * receive buffer and there was a small segment | |
1453 | * in queue. | |
1454 | */ | |
1ef9696c AK |
1455 | (copied > 0 && |
1456 | ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) || | |
1457 | ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) && | |
31954cd8 | 1458 | !inet_csk_in_pingpong_mode(sk))) && |
1ef9696c | 1459 | !atomic_read(&sk->sk_rmem_alloc))) |
a2a385d6 | 1460 | time_to_ack = true; |
1da177e4 LT |
1461 | } |
1462 | ||
1463 | /* We send an ACK if we can now advertise a non-zero window | |
1464 | * which has been raised "significantly". | |
1465 | * | |
1466 | * Even if window raised up to infinity, do not send window open ACK | |
1467 | * in states, where we will not receive more. It is useless. | |
1468 | */ | |
1469 | if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) { | |
1470 | __u32 rcv_window_now = tcp_receive_window(tp); | |
1471 | ||
1472 | /* Optimize, __tcp_select_window() is not cheap. */ | |
1473 | if (2*rcv_window_now <= tp->window_clamp) { | |
1474 | __u32 new_window = __tcp_select_window(sk); | |
1475 | ||
1476 | /* Send ACK now, if this read freed lots of space | |
1477 | * in our buffer. Certainly, new_window is new window. | |
1478 | * We can advertise it now, if it is not less than current one. | |
1479 | * "Lots" means "at least twice" here. | |
1480 | */ | |
1481 | if (new_window && new_window >= 2 * rcv_window_now) | |
a2a385d6 | 1482 | time_to_ack = true; |
1da177e4 LT |
1483 | } |
1484 | } | |
1485 | if (time_to_ack) | |
1486 | tcp_send_ack(sk); | |
1487 | } | |
1488 | ||
c457985a CW |
1489 | void tcp_cleanup_rbuf(struct sock *sk, int copied) |
1490 | { | |
1491 | struct sk_buff *skb = skb_peek(&sk->sk_receive_queue); | |
1492 | struct tcp_sock *tp = tcp_sk(sk); | |
1493 | ||
1494 | WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq), | |
1495 | "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n", | |
1496 | tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt); | |
1497 | __tcp_cleanup_rbuf(sk, copied); | |
1498 | } | |
1499 | ||
3df684c1 ED |
1500 | static void tcp_eat_recv_skb(struct sock *sk, struct sk_buff *skb) |
1501 | { | |
f35f8219 | 1502 | __skb_unlink(skb, &sk->sk_receive_queue); |
3df684c1 ED |
1503 | if (likely(skb->destructor == sock_rfree)) { |
1504 | sock_rfree(skb); | |
1505 | skb->destructor = NULL; | |
1506 | skb->sk = NULL; | |
68822bdf | 1507 | return skb_attempt_defer_free(skb); |
3df684c1 | 1508 | } |
f35f8219 | 1509 | __kfree_skb(skb); |
3df684c1 ED |
1510 | } |
1511 | ||
3f92a64e | 1512 | struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off) |
1da177e4 LT |
1513 | { |
1514 | struct sk_buff *skb; | |
1515 | u32 offset; | |
1516 | ||
f26845b4 | 1517 | while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) { |
1da177e4 | 1518 | offset = seq - TCP_SKB_CB(skb)->seq; |
9d691539 ED |
1519 | if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { |
1520 | pr_err_once("%s: found a SYN, please report !\n", __func__); | |
1da177e4 | 1521 | offset--; |
9d691539 | 1522 | } |
e11ecddf | 1523 | if (offset < skb->len || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) { |
1da177e4 LT |
1524 | *off = offset; |
1525 | return skb; | |
1526 | } | |
f26845b4 ED |
1527 | /* This looks weird, but this can happen if TCP collapsing |
1528 | * splitted a fat GRO packet, while we released socket lock | |
1529 | * in skb_splice_bits() | |
1530 | */ | |
3df684c1 | 1531 | tcp_eat_recv_skb(sk, skb); |
1da177e4 LT |
1532 | } |
1533 | return NULL; | |
1534 | } | |
3f92a64e | 1535 | EXPORT_SYMBOL(tcp_recv_skb); |
1da177e4 LT |
1536 | |
1537 | /* | |
1538 | * This routine provides an alternative to tcp_recvmsg() for routines | |
1539 | * that would like to handle copying from skbuffs directly in 'sendfile' | |
1540 | * fashion. | |
1541 | * Note: | |
1542 | * - It is assumed that the socket was locked by the caller. | |
1543 | * - The routine does not block. | |
1544 | * - At present, there is no support for reading OOB data | |
1545 | * or for 'peeking' the socket using this routine | |
1546 | * (although both would be easy to implement). | |
1547 | */ | |
1548 | int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, | |
1549 | sk_read_actor_t recv_actor) | |
1550 | { | |
1551 | struct sk_buff *skb; | |
1552 | struct tcp_sock *tp = tcp_sk(sk); | |
1553 | u32 seq = tp->copied_seq; | |
1554 | u32 offset; | |
1555 | int copied = 0; | |
1556 | ||
1557 | if (sk->sk_state == TCP_LISTEN) | |
1558 | return -ENOTCONN; | |
1559 | while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) { | |
1560 | if (offset < skb->len) { | |
374e7b59 OP |
1561 | int used; |
1562 | size_t len; | |
1da177e4 LT |
1563 | |
1564 | len = skb->len - offset; | |
1565 | /* Stop reading if we hit a patch of urgent data */ | |
b96c51bd | 1566 | if (unlikely(tp->urg_data)) { |
1da177e4 LT |
1567 | u32 urg_offset = tp->urg_seq - seq; |
1568 | if (urg_offset < len) | |
1569 | len = urg_offset; | |
1570 | if (!len) | |
1571 | break; | |
1572 | } | |
1573 | used = recv_actor(desc, skb, offset, len); | |
ff905b1e | 1574 | if (used <= 0) { |
ddb61a57 JA |
1575 | if (!copied) |
1576 | copied = used; | |
1577 | break; | |
1da177e4 | 1578 | } |
e3d5ea2c ED |
1579 | if (WARN_ON_ONCE(used > len)) |
1580 | used = len; | |
1581 | seq += used; | |
1582 | copied += used; | |
1583 | offset += used; | |
1584 | ||
02275a2e | 1585 | /* If recv_actor drops the lock (e.g. TCP splice |
293ad604 OP |
1586 | * receive) the skb pointer might be invalid when |
1587 | * getting here: tcp_collapse might have deleted it | |
1588 | * while aggregating skbs from the socket queue. | |
1589 | */ | |
02275a2e WT |
1590 | skb = tcp_recv_skb(sk, seq - 1, &offset); |
1591 | if (!skb) | |
1da177e4 | 1592 | break; |
02275a2e WT |
1593 | /* TCP coalescing might have appended data to the skb. |
1594 | * Try to splice more frags | |
1595 | */ | |
1596 | if (offset + 1 != skb->len) | |
1597 | continue; | |
1da177e4 | 1598 | } |
e11ecddf | 1599 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) { |
3df684c1 | 1600 | tcp_eat_recv_skb(sk, skb); |
1da177e4 LT |
1601 | ++seq; |
1602 | break; | |
1603 | } | |
3df684c1 | 1604 | tcp_eat_recv_skb(sk, skb); |
1da177e4 LT |
1605 | if (!desc->count) |
1606 | break; | |
7db48e98 | 1607 | WRITE_ONCE(tp->copied_seq, seq); |
1da177e4 | 1608 | } |
7db48e98 | 1609 | WRITE_ONCE(tp->copied_seq, seq); |
1da177e4 LT |
1610 | |
1611 | tcp_rcv_space_adjust(sk); | |
1612 | ||
1613 | /* Clean up data we have read: This will do ACK frames. */ | |
f26845b4 ED |
1614 | if (copied > 0) { |
1615 | tcp_recv_skb(sk, seq, &offset); | |
0e4b4992 | 1616 | tcp_cleanup_rbuf(sk, copied); |
f26845b4 | 1617 | } |
1da177e4 LT |
1618 | return copied; |
1619 | } | |
4bc2f18b | 1620 | EXPORT_SYMBOL(tcp_read_sock); |
1da177e4 | 1621 | |
965b57b4 | 1622 | int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor) |
04919bed CW |
1623 | { |
1624 | struct tcp_sock *tp = tcp_sk(sk); | |
1625 | u32 seq = tp->copied_seq; | |
1626 | struct sk_buff *skb; | |
1627 | int copied = 0; | |
1628 | u32 offset; | |
1629 | ||
1630 | if (sk->sk_state == TCP_LISTEN) | |
1631 | return -ENOTCONN; | |
1632 | ||
db4192a7 CW |
1633 | while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) { |
1634 | u8 tcp_flags; | |
1635 | int used; | |
04919bed | 1636 | |
db4192a7 CW |
1637 | __skb_unlink(skb, &sk->sk_receive_queue); |
1638 | WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk)); | |
1639 | tcp_flags = TCP_SKB_CB(skb)->tcp_flags; | |
1640 | used = recv_actor(sk, skb); | |
db4192a7 CW |
1641 | if (used < 0) { |
1642 | if (!copied) | |
1643 | copied = used; | |
1644 | break; | |
1645 | } | |
1646 | seq += used; | |
1647 | copied += used; | |
04919bed | 1648 | |
db4192a7 | 1649 | if (tcp_flags & TCPHDR_FIN) { |
04919bed | 1650 | ++seq; |
db4192a7 CW |
1651 | break; |
1652 | } | |
04919bed | 1653 | } |
04919bed CW |
1654 | return copied; |
1655 | } | |
1656 | EXPORT_SYMBOL(tcp_read_skb); | |
1657 | ||
3f92a64e JK |
1658 | void tcp_read_done(struct sock *sk, size_t len) |
1659 | { | |
1660 | struct tcp_sock *tp = tcp_sk(sk); | |
1661 | u32 seq = tp->copied_seq; | |
1662 | struct sk_buff *skb; | |
1663 | size_t left; | |
1664 | u32 offset; | |
1665 | ||
1666 | if (sk->sk_state == TCP_LISTEN) | |
1667 | return; | |
1668 | ||
1669 | left = len; | |
1670 | while (left && (skb = tcp_recv_skb(sk, seq, &offset)) != NULL) { | |
1671 | int used; | |
1672 | ||
1673 | used = min_t(size_t, skb->len - offset, left); | |
1674 | seq += used; | |
1675 | left -= used; | |
1676 | ||
1677 | if (skb->len > offset + used) | |
1678 | break; | |
1679 | ||
1680 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) { | |
1681 | tcp_eat_recv_skb(sk, skb); | |
1682 | ++seq; | |
1683 | break; | |
1684 | } | |
1685 | tcp_eat_recv_skb(sk, skb); | |
1686 | } | |
1687 | WRITE_ONCE(tp->copied_seq, seq); | |
1688 | ||
1689 | tcp_rcv_space_adjust(sk); | |
1690 | ||
1691 | /* Clean up data we have read: This will do ACK frames. */ | |
1692 | if (left != len) | |
1693 | tcp_cleanup_rbuf(sk, len - left); | |
1694 | } | |
1695 | EXPORT_SYMBOL(tcp_read_done); | |
1696 | ||
32035585 TH |
1697 | int tcp_peek_len(struct socket *sock) |
1698 | { | |
1699 | return tcp_inq(sock->sk); | |
1700 | } | |
1701 | EXPORT_SYMBOL(tcp_peek_len); | |
1702 | ||
d1361840 ED |
1703 | /* Make sure sk_rcvbuf is big enough to satisfy SO_RCVLOWAT hint */ |
1704 | int tcp_set_rcvlowat(struct sock *sk, int val) | |
1705 | { | |
dfa2f048 | 1706 | int space, cap; |
867f816b SHY |
1707 | |
1708 | if (sk->sk_userlocks & SOCK_RCVBUF_LOCK) | |
1709 | cap = sk->sk_rcvbuf >> 1; | |
1710 | else | |
02739545 | 1711 | cap = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2]) >> 1; |
867f816b | 1712 | val = min(val, cap); |
eac66402 | 1713 | WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); |
03f45c88 ED |
1714 | |
1715 | /* Check if we need to signal EPOLLIN right now */ | |
1716 | tcp_data_ready(sk); | |
1717 | ||
d1361840 ED |
1718 | if (sk->sk_userlocks & SOCK_RCVBUF_LOCK) |
1719 | return 0; | |
1720 | ||
dfa2f048 ED |
1721 | space = tcp_space_from_win(sk, val); |
1722 | if (space > sk->sk_rcvbuf) { | |
1723 | WRITE_ONCE(sk->sk_rcvbuf, space); | |
1724 | tcp_sk(sk)->window_clamp = val; | |
d1361840 ED |
1725 | } |
1726 | return 0; | |
1727 | } | |
1728 | EXPORT_SYMBOL(tcp_set_rcvlowat); | |
1729 | ||
892bfd3d FW |
1730 | void tcp_update_recv_tstamps(struct sk_buff *skb, |
1731 | struct scm_timestamping_internal *tss) | |
7eeba170 AR |
1732 | { |
1733 | if (skb->tstamp) | |
1734 | tss->ts[0] = ktime_to_timespec64(skb->tstamp); | |
1735 | else | |
1736 | tss->ts[0] = (struct timespec64) {0}; | |
1737 | ||
1738 | if (skb_hwtstamps(skb)->hwtstamp) | |
1739 | tss->ts[2] = ktime_to_timespec64(skb_hwtstamps(skb)->hwtstamp); | |
1740 | else | |
1741 | tss->ts[2] = (struct timespec64) {0}; | |
1742 | } | |
1743 | ||
05255b82 | 1744 | #ifdef CONFIG_MMU |
350f6bbc | 1745 | static const struct vm_operations_struct tcp_vm_ops = { |
05255b82 ED |
1746 | }; |
1747 | ||
93ab6cc6 ED |
1748 | int tcp_mmap(struct file *file, struct socket *sock, |
1749 | struct vm_area_struct *vma) | |
1750 | { | |
05255b82 ED |
1751 | if (vma->vm_flags & (VM_WRITE | VM_EXEC)) |
1752 | return -EPERM; | |
1c71222e | 1753 | vm_flags_clear(vma, VM_MAYWRITE | VM_MAYEXEC); |
05255b82 | 1754 | |
3e4e28c5 | 1755 | /* Instruct vm_insert_page() to not mmap_read_lock(mm) */ |
1c71222e | 1756 | vm_flags_set(vma, VM_MIXEDMAP); |
05255b82 ED |
1757 | |
1758 | vma->vm_ops = &tcp_vm_ops; | |
1759 | return 0; | |
1760 | } | |
1761 | EXPORT_SYMBOL(tcp_mmap); | |
1762 | ||
7fba5309 AR |
1763 | static skb_frag_t *skb_advance_to_frag(struct sk_buff *skb, u32 offset_skb, |
1764 | u32 *offset_frag) | |
1765 | { | |
1766 | skb_frag_t *frag; | |
1767 | ||
70701b83 AR |
1768 | if (unlikely(offset_skb >= skb->len)) |
1769 | return NULL; | |
1770 | ||
7fba5309 AR |
1771 | offset_skb -= skb_headlen(skb); |
1772 | if ((int)offset_skb < 0 || skb_has_frag_list(skb)) | |
1773 | return NULL; | |
1774 | ||
1775 | frag = skb_shinfo(skb)->frags; | |
1776 | while (offset_skb) { | |
1777 | if (skb_frag_size(frag) > offset_skb) { | |
1778 | *offset_frag = offset_skb; | |
1779 | return frag; | |
1780 | } | |
1781 | offset_skb -= skb_frag_size(frag); | |
1782 | ++frag; | |
1783 | } | |
1784 | *offset_frag = 0; | |
1785 | return frag; | |
1786 | } | |
1787 | ||
98917cf0 AR |
1788 | static bool can_map_frag(const skb_frag_t *frag) |
1789 | { | |
1790 | return skb_frag_size(frag) == PAGE_SIZE && !skb_frag_off(frag); | |
1791 | } | |
1792 | ||
1793 | static int find_next_mappable_frag(const skb_frag_t *frag, | |
1794 | int remaining_in_skb) | |
1795 | { | |
1796 | int offset = 0; | |
1797 | ||
1798 | if (likely(can_map_frag(frag))) | |
1799 | return 0; | |
1800 | ||
1801 | while (offset < remaining_in_skb && !can_map_frag(frag)) { | |
1802 | offset += skb_frag_size(frag); | |
1803 | ++frag; | |
1804 | } | |
1805 | return offset; | |
1806 | } | |
1807 | ||
0c3936d3 AR |
1808 | static void tcp_zerocopy_set_hint_for_skb(struct sock *sk, |
1809 | struct tcp_zerocopy_receive *zc, | |
1810 | struct sk_buff *skb, u32 offset) | |
1811 | { | |
1812 | u32 frag_offset, partial_frag_remainder = 0; | |
1813 | int mappable_offset; | |
1814 | skb_frag_t *frag; | |
1815 | ||
1816 | /* worst case: skip to next skb. try to improve on this case below */ | |
1817 | zc->recv_skip_hint = skb->len - offset; | |
1818 | ||
1819 | /* Find the frag containing this offset (and how far into that frag) */ | |
1820 | frag = skb_advance_to_frag(skb, offset, &frag_offset); | |
1821 | if (!frag) | |
1822 | return; | |
1823 | ||
1824 | if (frag_offset) { | |
1825 | struct skb_shared_info *info = skb_shinfo(skb); | |
1826 | ||
1827 | /* We read part of the last frag, must recvmsg() rest of skb. */ | |
1828 | if (frag == &info->frags[info->nr_frags - 1]) | |
1829 | return; | |
1830 | ||
1831 | /* Else, we must at least read the remainder in this frag. */ | |
1832 | partial_frag_remainder = skb_frag_size(frag) - frag_offset; | |
1833 | zc->recv_skip_hint -= partial_frag_remainder; | |
1834 | ++frag; | |
1835 | } | |
1836 | ||
1837 | /* partial_frag_remainder: If part way through a frag, must read rest. | |
1838 | * mappable_offset: Bytes till next mappable frag, *not* counting bytes | |
1839 | * in partial_frag_remainder. | |
1840 | */ | |
1841 | mappable_offset = find_next_mappable_frag(frag, zc->recv_skip_hint); | |
1842 | zc->recv_skip_hint = mappable_offset + partial_frag_remainder; | |
1843 | } | |
1844 | ||
f21a3c48 | 1845 | static int tcp_recvmsg_locked(struct sock *sk, struct msghdr *msg, size_t len, |
ec095263 | 1846 | int flags, struct scm_timestamping_internal *tss, |
f21a3c48 AR |
1847 | int *cmsg_flags); |
1848 | static int receive_fallback_to_copy(struct sock *sk, | |
7eeba170 AR |
1849 | struct tcp_zerocopy_receive *zc, int inq, |
1850 | struct scm_timestamping_internal *tss) | |
f21a3c48 AR |
1851 | { |
1852 | unsigned long copy_address = (unsigned long)zc->copybuf_address; | |
f21a3c48 AR |
1853 | struct msghdr msg = {}; |
1854 | struct iovec iov; | |
7eeba170 | 1855 | int err; |
f21a3c48 AR |
1856 | |
1857 | zc->length = 0; | |
1858 | zc->recv_skip_hint = 0; | |
1859 | ||
1860 | if (copy_address != zc->copybuf_address) | |
1861 | return -EINVAL; | |
1862 | ||
de4eda9d | 1863 | err = import_single_range(ITER_DEST, (void __user *)copy_address, |
f21a3c48 AR |
1864 | inq, &iov, &msg.msg_iter); |
1865 | if (err) | |
1866 | return err; | |
1867 | ||
ec095263 | 1868 | err = tcp_recvmsg_locked(sk, &msg, inq, MSG_DONTWAIT, |
7eeba170 | 1869 | tss, &zc->msg_flags); |
f21a3c48 AR |
1870 | if (err < 0) |
1871 | return err; | |
1872 | ||
1873 | zc->copybuf_len = err; | |
0c3936d3 AR |
1874 | if (likely(zc->copybuf_len)) { |
1875 | struct sk_buff *skb; | |
1876 | u32 offset; | |
1877 | ||
1878 | skb = tcp_recv_skb(sk, tcp_sk(sk)->copied_seq, &offset); | |
1879 | if (skb) | |
1880 | tcp_zerocopy_set_hint_for_skb(sk, zc, skb, offset); | |
1881 | } | |
f21a3c48 AR |
1882 | return 0; |
1883 | } | |
1884 | ||
18fb76ed AR |
1885 | static int tcp_copy_straggler_data(struct tcp_zerocopy_receive *zc, |
1886 | struct sk_buff *skb, u32 copylen, | |
1887 | u32 *offset, u32 *seq) | |
1888 | { | |
1889 | unsigned long copy_address = (unsigned long)zc->copybuf_address; | |
1890 | struct msghdr msg = {}; | |
1891 | struct iovec iov; | |
1892 | int err; | |
1893 | ||
1894 | if (copy_address != zc->copybuf_address) | |
1895 | return -EINVAL; | |
1896 | ||
de4eda9d | 1897 | err = import_single_range(ITER_DEST, (void __user *)copy_address, |
18fb76ed AR |
1898 | copylen, &iov, &msg.msg_iter); |
1899 | if (err) | |
1900 | return err; | |
1901 | err = skb_copy_datagram_msg(skb, *offset, &msg, copylen); | |
1902 | if (err) | |
1903 | return err; | |
1904 | zc->recv_skip_hint -= copylen; | |
1905 | *offset += copylen; | |
1906 | *seq += copylen; | |
1907 | return (__s32)copylen; | |
1908 | } | |
1909 | ||
7eeba170 AR |
1910 | static int tcp_zc_handle_leftover(struct tcp_zerocopy_receive *zc, |
1911 | struct sock *sk, | |
1912 | struct sk_buff *skb, | |
1913 | u32 *seq, | |
1914 | s32 copybuf_len, | |
1915 | struct scm_timestamping_internal *tss) | |
18fb76ed AR |
1916 | { |
1917 | u32 offset, copylen = min_t(u32, copybuf_len, zc->recv_skip_hint); | |
1918 | ||
1919 | if (!copylen) | |
1920 | return 0; | |
1921 | /* skb is null if inq < PAGE_SIZE. */ | |
7eeba170 | 1922 | if (skb) { |
18fb76ed | 1923 | offset = *seq - TCP_SKB_CB(skb)->seq; |
7eeba170 | 1924 | } else { |
18fb76ed | 1925 | skb = tcp_recv_skb(sk, *seq, &offset); |
7eeba170 AR |
1926 | if (TCP_SKB_CB(skb)->has_rxtstamp) { |
1927 | tcp_update_recv_tstamps(skb, tss); | |
1928 | zc->msg_flags |= TCP_CMSG_TS; | |
1929 | } | |
1930 | } | |
18fb76ed AR |
1931 | |
1932 | zc->copybuf_len = tcp_copy_straggler_data(zc, skb, copylen, &offset, | |
1933 | seq); | |
1934 | return zc->copybuf_len < 0 ? 0 : copylen; | |
1935 | } | |
1936 | ||
94ab9eb9 AR |
1937 | static int tcp_zerocopy_vm_insert_batch_error(struct vm_area_struct *vma, |
1938 | struct page **pending_pages, | |
1939 | unsigned long pages_remaining, | |
1940 | unsigned long *address, | |
1941 | u32 *length, | |
1942 | u32 *seq, | |
1943 | struct tcp_zerocopy_receive *zc, | |
1944 | u32 total_bytes_to_map, | |
1945 | int err) | |
1946 | { | |
1947 | /* At least one page did not map. Try zapping if we skipped earlier. */ | |
1948 | if (err == -EBUSY && | |
1949 | zc->flags & TCP_RECEIVE_ZEROCOPY_FLAG_TLB_CLEAN_HINT) { | |
1950 | u32 maybe_zap_len; | |
1951 | ||
1952 | maybe_zap_len = total_bytes_to_map - /* All bytes to map */ | |
1953 | *length + /* Mapped or pending */ | |
1954 | (pages_remaining * PAGE_SIZE); /* Failed map. */ | |
e9adcfec | 1955 | zap_page_range_single(vma, *address, maybe_zap_len, NULL); |
94ab9eb9 AR |
1956 | err = 0; |
1957 | } | |
1958 | ||
1959 | if (!err) { | |
1960 | unsigned long leftover_pages = pages_remaining; | |
1961 | int bytes_mapped; | |
1962 | ||
e9adcfec | 1963 | /* We called zap_page_range_single, try to reinsert. */ |
94ab9eb9 AR |
1964 | err = vm_insert_pages(vma, *address, |
1965 | pending_pages, | |
1966 | &pages_remaining); | |
1967 | bytes_mapped = PAGE_SIZE * (leftover_pages - pages_remaining); | |
1968 | *seq += bytes_mapped; | |
1969 | *address += bytes_mapped; | |
1970 | } | |
1971 | if (err) { | |
1972 | /* Either we were unable to zap, OR we zapped, retried an | |
1973 | * insert, and still had an issue. Either ways, pages_remaining | |
1974 | * is the number of pages we were unable to map, and we unroll | |
1975 | * some state we speculatively touched before. | |
1976 | */ | |
1977 | const int bytes_not_mapped = PAGE_SIZE * pages_remaining; | |
1978 | ||
1979 | *length -= bytes_not_mapped; | |
1980 | zc->recv_skip_hint += bytes_not_mapped; | |
1981 | } | |
1982 | return err; | |
1983 | } | |
1984 | ||
3763a24c AR |
1985 | static int tcp_zerocopy_vm_insert_batch(struct vm_area_struct *vma, |
1986 | struct page **pages, | |
94ab9eb9 AR |
1987 | unsigned int pages_to_map, |
1988 | unsigned long *address, | |
1989 | u32 *length, | |
3763a24c | 1990 | u32 *seq, |
94ab9eb9 AR |
1991 | struct tcp_zerocopy_receive *zc, |
1992 | u32 total_bytes_to_map) | |
3763a24c AR |
1993 | { |
1994 | unsigned long pages_remaining = pages_to_map; | |
94ab9eb9 AR |
1995 | unsigned int pages_mapped; |
1996 | unsigned int bytes_mapped; | |
1997 | int err; | |
3763a24c | 1998 | |
94ab9eb9 AR |
1999 | err = vm_insert_pages(vma, *address, pages, &pages_remaining); |
2000 | pages_mapped = pages_to_map - (unsigned int)pages_remaining; | |
2001 | bytes_mapped = PAGE_SIZE * pages_mapped; | |
3763a24c AR |
2002 | /* Even if vm_insert_pages fails, it may have partially succeeded in |
2003 | * mapping (some but not all of the pages). | |
2004 | */ | |
2005 | *seq += bytes_mapped; | |
94ab9eb9 AR |
2006 | *address += bytes_mapped; |
2007 | ||
2008 | if (likely(!err)) | |
2009 | return 0; | |
2010 | ||
2011 | /* Error: maybe zap and retry + rollback state for failed inserts. */ | |
2012 | return tcp_zerocopy_vm_insert_batch_error(vma, pages + pages_mapped, | |
2013 | pages_remaining, address, length, seq, zc, total_bytes_to_map, | |
2014 | err); | |
3763a24c AR |
2015 | } |
2016 | ||
3c5a2fd0 | 2017 | #define TCP_VALID_ZC_MSG_FLAGS (TCP_CMSG_TS) |
7eeba170 AR |
2018 | static void tcp_zc_finalize_rx_tstamp(struct sock *sk, |
2019 | struct tcp_zerocopy_receive *zc, | |
2020 | struct scm_timestamping_internal *tss) | |
2021 | { | |
2022 | unsigned long msg_control_addr; | |
2023 | struct msghdr cmsg_dummy; | |
2024 | ||
2025 | msg_control_addr = (unsigned long)zc->msg_control; | |
c39ef213 | 2026 | cmsg_dummy.msg_control_user = (void __user *)msg_control_addr; |
7eeba170 AR |
2027 | cmsg_dummy.msg_controllen = |
2028 | (__kernel_size_t)zc->msg_controllen; | |
2029 | cmsg_dummy.msg_flags = in_compat_syscall() | |
2030 | ? MSG_CMSG_COMPAT : 0; | |
a6f8ee58 | 2031 | cmsg_dummy.msg_control_is_user = true; |
7eeba170 AR |
2032 | zc->msg_flags = 0; |
2033 | if (zc->msg_control == msg_control_addr && | |
2034 | zc->msg_controllen == cmsg_dummy.msg_controllen) { | |
2035 | tcp_recv_timestamp(&cmsg_dummy, sk, tss); | |
2036 | zc->msg_control = (__u64) | |
c39ef213 | 2037 | ((uintptr_t)cmsg_dummy.msg_control_user); |
7eeba170 AR |
2038 | zc->msg_controllen = |
2039 | (__u64)cmsg_dummy.msg_controllen; | |
2040 | zc->msg_flags = (__u32)cmsg_dummy.msg_flags; | |
2041 | } | |
2042 | } | |
2043 | ||
7a7f0946 AR |
2044 | static struct vm_area_struct *find_tcp_vma(struct mm_struct *mm, |
2045 | unsigned long address, | |
2046 | bool *mmap_locked) | |
2047 | { | |
350f6bbc | 2048 | struct vm_area_struct *vma = lock_vma_under_rcu(mm, address); |
7a7f0946 | 2049 | |
7a7f0946 | 2050 | if (vma) { |
350f6bbc | 2051 | if (vma->vm_ops != &tcp_vm_ops) { |
7a7f0946 AR |
2052 | vma_end_read(vma); |
2053 | return NULL; | |
2054 | } | |
2055 | *mmap_locked = false; | |
2056 | return vma; | |
2057 | } | |
2058 | ||
2059 | mmap_read_lock(mm); | |
2060 | vma = vma_lookup(mm, address); | |
350f6bbc | 2061 | if (!vma || vma->vm_ops != &tcp_vm_ops) { |
7a7f0946 AR |
2062 | mmap_read_unlock(mm); |
2063 | return NULL; | |
2064 | } | |
2065 | *mmap_locked = true; | |
2066 | return vma; | |
2067 | } | |
2068 | ||
94ab9eb9 | 2069 | #define TCP_ZEROCOPY_PAGE_BATCH_SIZE 32 |
05255b82 | 2070 | static int tcp_zerocopy_receive(struct sock *sk, |
7eeba170 AR |
2071 | struct tcp_zerocopy_receive *zc, |
2072 | struct scm_timestamping_internal *tss) | |
05255b82 | 2073 | { |
94ab9eb9 | 2074 | u32 length = 0, offset, vma_len, avail_len, copylen = 0; |
05255b82 | 2075 | unsigned long address = (unsigned long)zc->address; |
94ab9eb9 | 2076 | struct page *pages[TCP_ZEROCOPY_PAGE_BATCH_SIZE]; |
18fb76ed AR |
2077 | s32 copybuf_len = zc->copybuf_len; |
2078 | struct tcp_sock *tp = tcp_sk(sk); | |
05255b82 | 2079 | const skb_frag_t *frags = NULL; |
94ab9eb9 | 2080 | unsigned int pages_to_map = 0; |
05255b82 ED |
2081 | struct vm_area_struct *vma; |
2082 | struct sk_buff *skb = NULL; | |
18fb76ed | 2083 | u32 seq = tp->copied_seq; |
94ab9eb9 | 2084 | u32 total_bytes_to_map; |
18fb76ed | 2085 | int inq = tcp_inq(sk); |
7a7f0946 | 2086 | bool mmap_locked; |
93ab6cc6 ED |
2087 | int ret; |
2088 | ||
18fb76ed | 2089 | zc->copybuf_len = 0; |
7eeba170 | 2090 | zc->msg_flags = 0; |
18fb76ed | 2091 | |
05255b82 | 2092 | if (address & (PAGE_SIZE - 1) || address != zc->address) |
93ab6cc6 ED |
2093 | return -EINVAL; |
2094 | ||
93ab6cc6 | 2095 | if (sk->sk_state == TCP_LISTEN) |
05255b82 | 2096 | return -ENOTCONN; |
93ab6cc6 ED |
2097 | |
2098 | sock_rps_record_flow(sk); | |
2099 | ||
f21a3c48 | 2100 | if (inq && inq <= copybuf_len) |
7eeba170 | 2101 | return receive_fallback_to_copy(sk, zc, inq, tss); |
f21a3c48 | 2102 | |
936ced41 AR |
2103 | if (inq < PAGE_SIZE) { |
2104 | zc->length = 0; | |
2105 | zc->recv_skip_hint = inq; | |
2106 | if (!inq && sock_flag(sk, SOCK_DONE)) | |
2107 | return -EIO; | |
2108 | return 0; | |
2109 | } | |
2110 | ||
7a7f0946 AR |
2111 | vma = find_tcp_vma(current->mm, address, &mmap_locked); |
2112 | if (!vma) | |
e776af60 | 2113 | return -EINVAL; |
7a7f0946 | 2114 | |
18fb76ed AR |
2115 | vma_len = min_t(unsigned long, zc->length, vma->vm_end - address); |
2116 | avail_len = min_t(u32, vma_len, inq); | |
94ab9eb9 AR |
2117 | total_bytes_to_map = avail_len & ~(PAGE_SIZE - 1); |
2118 | if (total_bytes_to_map) { | |
2119 | if (!(zc->flags & TCP_RECEIVE_ZEROCOPY_FLAG_TLB_CLEAN_HINT)) | |
e9adcfec MK |
2120 | zap_page_range_single(vma, address, total_bytes_to_map, |
2121 | NULL); | |
94ab9eb9 | 2122 | zc->length = total_bytes_to_map; |
8f2b0293 SHY |
2123 | zc->recv_skip_hint = 0; |
2124 | } else { | |
18fb76ed AR |
2125 | zc->length = avail_len; |
2126 | zc->recv_skip_hint = avail_len; | |
8f2b0293 | 2127 | } |
05255b82 ED |
2128 | ret = 0; |
2129 | while (length + PAGE_SIZE <= zc->length) { | |
98917cf0 | 2130 | int mappable_offset; |
94ab9eb9 | 2131 | struct page *page; |
98917cf0 | 2132 | |
05255b82 | 2133 | if (zc->recv_skip_hint < PAGE_SIZE) { |
7fba5309 AR |
2134 | u32 offset_frag; |
2135 | ||
05255b82 | 2136 | if (skb) { |
0e627190 AR |
2137 | if (zc->recv_skip_hint > 0) |
2138 | break; | |
05255b82 ED |
2139 | skb = skb->next; |
2140 | offset = seq - TCP_SKB_CB(skb)->seq; | |
2141 | } else { | |
2142 | skb = tcp_recv_skb(sk, seq, &offset); | |
2143 | } | |
7eeba170 AR |
2144 | |
2145 | if (TCP_SKB_CB(skb)->has_rxtstamp) { | |
2146 | tcp_update_recv_tstamps(skb, tss); | |
2147 | zc->msg_flags |= TCP_CMSG_TS; | |
2148 | } | |
05255b82 | 2149 | zc->recv_skip_hint = skb->len - offset; |
7fba5309 AR |
2150 | frags = skb_advance_to_frag(skb, offset, &offset_frag); |
2151 | if (!frags || offset_frag) | |
05255b82 | 2152 | break; |
93ab6cc6 | 2153 | } |
789762ce | 2154 | |
98917cf0 AR |
2155 | mappable_offset = find_next_mappable_frag(frags, |
2156 | zc->recv_skip_hint); | |
2157 | if (mappable_offset) { | |
2158 | zc->recv_skip_hint = mappable_offset; | |
05255b82 | 2159 | break; |
789762ce | 2160 | } |
94ab9eb9 AR |
2161 | page = skb_frag_page(frags); |
2162 | prefetchw(page); | |
2163 | pages[pages_to_map++] = page; | |
05255b82 | 2164 | length += PAGE_SIZE; |
05255b82 ED |
2165 | zc->recv_skip_hint -= PAGE_SIZE; |
2166 | frags++; | |
94ab9eb9 AR |
2167 | if (pages_to_map == TCP_ZEROCOPY_PAGE_BATCH_SIZE || |
2168 | zc->recv_skip_hint < PAGE_SIZE) { | |
2169 | /* Either full batch, or we're about to go to next skb | |
2170 | * (and we cannot unroll failed ops across skbs). | |
2171 | */ | |
2172 | ret = tcp_zerocopy_vm_insert_batch(vma, pages, | |
2173 | pages_to_map, | |
2174 | &address, &length, | |
2175 | &seq, zc, | |
2176 | total_bytes_to_map); | |
3763a24c AR |
2177 | if (ret) |
2178 | goto out; | |
94ab9eb9 | 2179 | pages_to_map = 0; |
3763a24c AR |
2180 | } |
2181 | } | |
94ab9eb9 AR |
2182 | if (pages_to_map) { |
2183 | ret = tcp_zerocopy_vm_insert_batch(vma, pages, pages_to_map, | |
2184 | &address, &length, &seq, | |
2185 | zc, total_bytes_to_map); | |
93ab6cc6 | 2186 | } |
93ab6cc6 | 2187 | out: |
7a7f0946 AR |
2188 | if (mmap_locked) |
2189 | mmap_read_unlock(current->mm); | |
2190 | else | |
2191 | vma_end_read(vma); | |
18fb76ed AR |
2192 | /* Try to copy straggler data. */ |
2193 | if (!ret) | |
7eeba170 | 2194 | copylen = tcp_zc_handle_leftover(zc, sk, skb, &seq, copybuf_len, tss); |
18fb76ed AR |
2195 | |
2196 | if (length + copylen) { | |
7db48e98 | 2197 | WRITE_ONCE(tp->copied_seq, seq); |
05255b82 ED |
2198 | tcp_rcv_space_adjust(sk); |
2199 | ||
2200 | /* Clean up data we have read: This will do ACK frames. */ | |
2201 | tcp_recv_skb(sk, seq, &offset); | |
18fb76ed | 2202 | tcp_cleanup_rbuf(sk, length + copylen); |
05255b82 ED |
2203 | ret = 0; |
2204 | if (length == zc->length) | |
2205 | zc->recv_skip_hint = 0; | |
2206 | } else { | |
2207 | if (!zc->recv_skip_hint && sock_flag(sk, SOCK_DONE)) | |
2208 | ret = -EIO; | |
2209 | } | |
2210 | zc->length = length; | |
93ab6cc6 ED |
2211 | return ret; |
2212 | } | |
05255b82 | 2213 | #endif |
93ab6cc6 | 2214 | |
98aaa913 | 2215 | /* Similar to __sock_recv_timestamp, but does not require an skb */ |
892bfd3d FW |
2216 | void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, |
2217 | struct scm_timestamping_internal *tss) | |
98aaa913 | 2218 | { |
887feae3 | 2219 | int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW); |
98aaa913 MM |
2220 | bool has_timestamping = false; |
2221 | ||
2222 | if (tss->ts[0].tv_sec || tss->ts[0].tv_nsec) { | |
2223 | if (sock_flag(sk, SOCK_RCVTSTAMP)) { | |
2224 | if (sock_flag(sk, SOCK_RCVTSTAMPNS)) { | |
887feae3 | 2225 | if (new_tstamp) { |
df1b4ba9 AB |
2226 | struct __kernel_timespec kts = { |
2227 | .tv_sec = tss->ts[0].tv_sec, | |
2228 | .tv_nsec = tss->ts[0].tv_nsec, | |
2229 | }; | |
887feae3 DD |
2230 | put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW, |
2231 | sizeof(kts), &kts); | |
2232 | } else { | |
df1b4ba9 AB |
2233 | struct __kernel_old_timespec ts_old = { |
2234 | .tv_sec = tss->ts[0].tv_sec, | |
2235 | .tv_nsec = tss->ts[0].tv_nsec, | |
2236 | }; | |
887feae3 | 2237 | put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD, |
9718475e | 2238 | sizeof(ts_old), &ts_old); |
887feae3 | 2239 | } |
98aaa913 | 2240 | } else { |
887feae3 | 2241 | if (new_tstamp) { |
df1b4ba9 AB |
2242 | struct __kernel_sock_timeval stv = { |
2243 | .tv_sec = tss->ts[0].tv_sec, | |
2244 | .tv_usec = tss->ts[0].tv_nsec / 1000, | |
2245 | }; | |
887feae3 DD |
2246 | put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW, |
2247 | sizeof(stv), &stv); | |
2248 | } else { | |
df1b4ba9 AB |
2249 | struct __kernel_old_timeval tv = { |
2250 | .tv_sec = tss->ts[0].tv_sec, | |
2251 | .tv_usec = tss->ts[0].tv_nsec / 1000, | |
2252 | }; | |
887feae3 DD |
2253 | put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD, |
2254 | sizeof(tv), &tv); | |
2255 | } | |
98aaa913 MM |
2256 | } |
2257 | } | |
2258 | ||
e3390b30 | 2259 | if (READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_SOFTWARE) |
98aaa913 MM |
2260 | has_timestamping = true; |
2261 | else | |
9718475e | 2262 | tss->ts[0] = (struct timespec64) {0}; |
98aaa913 MM |
2263 | } |
2264 | ||
2265 | if (tss->ts[2].tv_sec || tss->ts[2].tv_nsec) { | |
e3390b30 | 2266 | if (READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_RAW_HARDWARE) |
98aaa913 MM |
2267 | has_timestamping = true; |
2268 | else | |
9718475e | 2269 | tss->ts[2] = (struct timespec64) {0}; |
98aaa913 MM |
2270 | } |
2271 | ||
2272 | if (has_timestamping) { | |
9718475e DD |
2273 | tss->ts[1] = (struct timespec64) {0}; |
2274 | if (sock_flag(sk, SOCK_TSTAMP_NEW)) | |
2275 | put_cmsg_scm_timestamping64(msg, tss); | |
2276 | else | |
2277 | put_cmsg_scm_timestamping(msg, tss); | |
98aaa913 MM |
2278 | } |
2279 | } | |
2280 | ||
b75eba76 SHY |
2281 | static int tcp_inq_hint(struct sock *sk) |
2282 | { | |
2283 | const struct tcp_sock *tp = tcp_sk(sk); | |
2284 | u32 copied_seq = READ_ONCE(tp->copied_seq); | |
2285 | u32 rcv_nxt = READ_ONCE(tp->rcv_nxt); | |
2286 | int inq; | |
2287 | ||
2288 | inq = rcv_nxt - copied_seq; | |
2289 | if (unlikely(inq < 0 || copied_seq != READ_ONCE(tp->copied_seq))) { | |
2290 | lock_sock(sk); | |
2291 | inq = tp->rcv_nxt - tp->copied_seq; | |
2292 | release_sock(sk); | |
2293 | } | |
6466e715 SHY |
2294 | /* After receiving a FIN, tell the user-space to continue reading |
2295 | * by returning a non-zero inq. | |
2296 | */ | |
2297 | if (inq == 0 && sock_flag(sk, SOCK_DONE)) | |
2298 | inq = 1; | |
b75eba76 SHY |
2299 | return inq; |
2300 | } | |
2301 | ||
1da177e4 LT |
2302 | /* |
2303 | * This routine copies from a sock struct into the user buffer. | |
2304 | * | |
2305 | * Technical note: in 2.3 we work on _locked_ socket, so that | |
2306 | * tricks with *seq access order and skb->users are not required. | |
2307 | * Probably, code can be easily improved even more. | |
2308 | */ | |
2309 | ||
2cd81161 | 2310 | static int tcp_recvmsg_locked(struct sock *sk, struct msghdr *msg, size_t len, |
ec095263 | 2311 | int flags, struct scm_timestamping_internal *tss, |
2cd81161 | 2312 | int *cmsg_flags) |
1da177e4 LT |
2313 | { |
2314 | struct tcp_sock *tp = tcp_sk(sk); | |
2315 | int copied = 0; | |
2316 | u32 peek_seq; | |
2317 | u32 *seq; | |
2318 | unsigned long used; | |
2cd81161 | 2319 | int err; |
1da177e4 LT |
2320 | int target; /* Read at least this many bytes */ |
2321 | long timeo; | |
dfbafc99 | 2322 | struct sk_buff *skb, *last; |
77527313 | 2323 | u32 urg_hole = 0; |
1da177e4 | 2324 | |
1da177e4 LT |
2325 | err = -ENOTCONN; |
2326 | if (sk->sk_state == TCP_LISTEN) | |
2327 | goto out; | |
2328 | ||
f94fd25c | 2329 | if (tp->recvmsg_inq) { |
925bba24 | 2330 | *cmsg_flags = TCP_CMSG_INQ; |
f94fd25c JA |
2331 | msg->msg_get_inq = 1; |
2332 | } | |
ec095263 | 2333 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); |
1da177e4 LT |
2334 | |
2335 | /* Urgent data needs to be handled specially. */ | |
2336 | if (flags & MSG_OOB) | |
2337 | goto recv_urg; | |
2338 | ||
c0e88ff0 PE |
2339 | if (unlikely(tp->repair)) { |
2340 | err = -EPERM; | |
2341 | if (!(flags & MSG_PEEK)) | |
2342 | goto out; | |
2343 | ||
2344 | if (tp->repair_queue == TCP_SEND_QUEUE) | |
2345 | goto recv_sndq; | |
2346 | ||
2347 | err = -EINVAL; | |
2348 | if (tp->repair_queue == TCP_NO_QUEUE) | |
2349 | goto out; | |
2350 | ||
2351 | /* 'common' recv queue MSG_PEEK-ing */ | |
2352 | } | |
2353 | ||
1da177e4 LT |
2354 | seq = &tp->copied_seq; |
2355 | if (flags & MSG_PEEK) { | |
2356 | peek_seq = tp->copied_seq; | |
2357 | seq = &peek_seq; | |
2358 | } | |
2359 | ||
2360 | target = sock_rcvlowat(sk, flags & MSG_WAITALL, len); | |
2361 | ||
2362 | do { | |
1da177e4 LT |
2363 | u32 offset; |
2364 | ||
2365 | /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */ | |
b96c51bd | 2366 | if (unlikely(tp->urg_data) && tp->urg_seq == *seq) { |
1da177e4 LT |
2367 | if (copied) |
2368 | break; | |
2369 | if (signal_pending(current)) { | |
2370 | copied = timeo ? sock_intr_errno(timeo) : -EAGAIN; | |
2371 | break; | |
2372 | } | |
2373 | } | |
2374 | ||
2375 | /* Next get a buffer. */ | |
2376 | ||
dfbafc99 | 2377 | last = skb_peek_tail(&sk->sk_receive_queue); |
91521944 | 2378 | skb_queue_walk(&sk->sk_receive_queue, skb) { |
dfbafc99 | 2379 | last = skb; |
1da177e4 LT |
2380 | /* Now that we have two receive queues this |
2381 | * shouldn't happen. | |
2382 | */ | |
d792c100 | 2383 | if (WARN(before(*seq, TCP_SKB_CB(skb)->seq), |
e56b8ce3 | 2384 | "TCP recvmsg seq # bug: copied %X, seq %X, rcvnxt %X, fl %X\n", |
2af6fd8b JP |
2385 | *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, |
2386 | flags)) | |
1da177e4 | 2387 | break; |
d792c100 | 2388 | |
1da177e4 | 2389 | offset = *seq - TCP_SKB_CB(skb)->seq; |
9d691539 ED |
2390 | if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { |
2391 | pr_err_once("%s: found a SYN, please report !\n", __func__); | |
1da177e4 | 2392 | offset--; |
9d691539 | 2393 | } |
1da177e4 LT |
2394 | if (offset < skb->len) |
2395 | goto found_ok_skb; | |
e11ecddf | 2396 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) |
1da177e4 | 2397 | goto found_fin_ok; |
2af6fd8b | 2398 | WARN(!(flags & MSG_PEEK), |
e56b8ce3 | 2399 | "TCP recvmsg seq # bug 2: copied %X, seq %X, rcvnxt %X, fl %X\n", |
2af6fd8b | 2400 | *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags); |
91521944 | 2401 | } |
1da177e4 LT |
2402 | |
2403 | /* Well, if we have backlog, try to process it now yet. */ | |
2404 | ||
9ed498c6 | 2405 | if (copied >= target && !READ_ONCE(sk->sk_backlog.tail)) |
1da177e4 LT |
2406 | break; |
2407 | ||
2408 | if (copied) { | |
8bd172b7 ED |
2409 | if (!timeo || |
2410 | sk->sk_err || | |
1da177e4 LT |
2411 | sk->sk_state == TCP_CLOSE || |
2412 | (sk->sk_shutdown & RCV_SHUTDOWN) || | |
518a09ef | 2413 | signal_pending(current)) |
1da177e4 LT |
2414 | break; |
2415 | } else { | |
2416 | if (sock_flag(sk, SOCK_DONE)) | |
2417 | break; | |
2418 | ||
2419 | if (sk->sk_err) { | |
2420 | copied = sock_error(sk); | |
2421 | break; | |
2422 | } | |
2423 | ||
2424 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
2425 | break; | |
2426 | ||
2427 | if (sk->sk_state == TCP_CLOSE) { | |
c47078d6 ED |
2428 | /* This occurs when user tries to read |
2429 | * from never connected socket. | |
2430 | */ | |
2431 | copied = -ENOTCONN; | |
1da177e4 LT |
2432 | break; |
2433 | } | |
2434 | ||
2435 | if (!timeo) { | |
2436 | copied = -EAGAIN; | |
2437 | break; | |
2438 | } | |
2439 | ||
2440 | if (signal_pending(current)) { | |
2441 | copied = sock_intr_errno(timeo); | |
2442 | break; | |
2443 | } | |
2444 | } | |
2445 | ||
1da177e4 LT |
2446 | if (copied >= target) { |
2447 | /* Do not sleep, just process backlog. */ | |
93afcfd1 | 2448 | __sk_flush_backlog(sk); |
dfbafc99 | 2449 | } else { |
29fbc26e | 2450 | tcp_cleanup_rbuf(sk, copied); |
dfbafc99 SD |
2451 | sk_wait_data(sk, &timeo, last); |
2452 | } | |
1da177e4 | 2453 | |
77527313 IJ |
2454 | if ((flags & MSG_PEEK) && |
2455 | (peek_seq - copied - urg_hole != tp->copied_seq)) { | |
e87cc472 JP |
2456 | net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n", |
2457 | current->comm, | |
2458 | task_pid_nr(current)); | |
1da177e4 LT |
2459 | peek_seq = tp->copied_seq; |
2460 | } | |
2461 | continue; | |
2462 | ||
fdb8b298 | 2463 | found_ok_skb: |
1da177e4 LT |
2464 | /* Ok so how much can we use? */ |
2465 | used = skb->len - offset; | |
2466 | if (len < used) | |
2467 | used = len; | |
2468 | ||
2469 | /* Do we have urgent data here? */ | |
b96c51bd | 2470 | if (unlikely(tp->urg_data)) { |
1da177e4 LT |
2471 | u32 urg_offset = tp->urg_seq - *seq; |
2472 | if (urg_offset < used) { | |
2473 | if (!urg_offset) { | |
2474 | if (!sock_flag(sk, SOCK_URGINLINE)) { | |
7db48e98 | 2475 | WRITE_ONCE(*seq, *seq + 1); |
77527313 | 2476 | urg_hole++; |
1da177e4 LT |
2477 | offset++; |
2478 | used--; | |
2479 | if (!used) | |
2480 | goto skip_copy; | |
2481 | } | |
2482 | } else | |
2483 | used = urg_offset; | |
2484 | } | |
2485 | } | |
2486 | ||
2487 | if (!(flags & MSG_TRUNC)) { | |
51f3d02b | 2488 | err = skb_copy_datagram_msg(skb, offset, msg, used); |
7bced397 DW |
2489 | if (err) { |
2490 | /* Exception. Bailout! */ | |
2491 | if (!copied) | |
2492 | copied = -EFAULT; | |
2493 | break; | |
1da177e4 LT |
2494 | } |
2495 | } | |
2496 | ||
7db48e98 | 2497 | WRITE_ONCE(*seq, *seq + used); |
1da177e4 LT |
2498 | copied += used; |
2499 | len -= used; | |
2500 | ||
2501 | tcp_rcv_space_adjust(sk); | |
2502 | ||
2503 | skip_copy: | |
b96c51bd | 2504 | if (unlikely(tp->urg_data) && after(tp->copied_seq, tp->urg_seq)) { |
7b6a893a | 2505 | WRITE_ONCE(tp->urg_data, 0); |
31770e34 FW |
2506 | tcp_fast_path_check(sk); |
2507 | } | |
1da177e4 | 2508 | |
98aaa913 | 2509 | if (TCP_SKB_CB(skb)->has_rxtstamp) { |
2cd81161 | 2510 | tcp_update_recv_tstamps(skb, tss); |
925bba24 | 2511 | *cmsg_flags |= TCP_CMSG_TS; |
98aaa913 | 2512 | } |
cc4de047 KL |
2513 | |
2514 | if (used + offset < skb->len) | |
2515 | continue; | |
2516 | ||
e11ecddf | 2517 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) |
1da177e4 | 2518 | goto found_fin_ok; |
7bced397 | 2519 | if (!(flags & MSG_PEEK)) |
3df684c1 | 2520 | tcp_eat_recv_skb(sk, skb); |
1da177e4 LT |
2521 | continue; |
2522 | ||
fdb8b298 | 2523 | found_fin_ok: |
1da177e4 | 2524 | /* Process the FIN. */ |
7db48e98 | 2525 | WRITE_ONCE(*seq, *seq + 1); |
7bced397 | 2526 | if (!(flags & MSG_PEEK)) |
3df684c1 | 2527 | tcp_eat_recv_skb(sk, skb); |
1da177e4 LT |
2528 | break; |
2529 | } while (len > 0); | |
2530 | ||
1da177e4 LT |
2531 | /* According to UNIX98, msg_name/msg_namelen are ignored |
2532 | * on connected socket. I was just happy when found this 8) --ANK | |
2533 | */ | |
2534 | ||
2535 | /* Clean up data we have read: This will do ACK frames. */ | |
0e4b4992 | 2536 | tcp_cleanup_rbuf(sk, copied); |
1da177e4 LT |
2537 | return copied; |
2538 | ||
2539 | out: | |
1da177e4 LT |
2540 | return err; |
2541 | ||
2542 | recv_urg: | |
377f0a08 | 2543 | err = tcp_recv_urg(sk, msg, len, flags); |
1da177e4 | 2544 | goto out; |
c0e88ff0 PE |
2545 | |
2546 | recv_sndq: | |
2547 | err = tcp_peek_sndq(sk, msg, len); | |
2548 | goto out; | |
1da177e4 | 2549 | } |
2cd81161 | 2550 | |
ec095263 OH |
2551 | int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags, |
2552 | int *addr_len) | |
2cd81161 | 2553 | { |
f94fd25c | 2554 | int cmsg_flags = 0, ret; |
2cd81161 AR |
2555 | struct scm_timestamping_internal tss; |
2556 | ||
2557 | if (unlikely(flags & MSG_ERRQUEUE)) | |
2558 | return inet_recv_error(sk, msg, len, addr_len); | |
2559 | ||
2560 | if (sk_can_busy_loop(sk) && | |
2561 | skb_queue_empty_lockless(&sk->sk_receive_queue) && | |
2562 | sk->sk_state == TCP_ESTABLISHED) | |
ec095263 | 2563 | sk_busy_loop(sk, flags & MSG_DONTWAIT); |
2cd81161 AR |
2564 | |
2565 | lock_sock(sk); | |
ec095263 | 2566 | ret = tcp_recvmsg_locked(sk, msg, len, flags, &tss, &cmsg_flags); |
2cd81161 AR |
2567 | release_sock(sk); |
2568 | ||
f94fd25c | 2569 | if ((cmsg_flags || msg->msg_get_inq) && ret >= 0) { |
925bba24 | 2570 | if (cmsg_flags & TCP_CMSG_TS) |
2cd81161 | 2571 | tcp_recv_timestamp(msg, sk, &tss); |
f94fd25c JA |
2572 | if (msg->msg_get_inq) { |
2573 | msg->msg_inq = tcp_inq_hint(sk); | |
2574 | if (cmsg_flags & TCP_CMSG_INQ) | |
2575 | put_cmsg(msg, SOL_TCP, TCP_CM_INQ, | |
2576 | sizeof(msg->msg_inq), &msg->msg_inq); | |
2cd81161 AR |
2577 | } |
2578 | } | |
2579 | return ret; | |
2580 | } | |
4bc2f18b | 2581 | EXPORT_SYMBOL(tcp_recvmsg); |
1da177e4 | 2582 | |
490d5046 IJ |
2583 | void tcp_set_state(struct sock *sk, int state) |
2584 | { | |
2585 | int oldstate = sk->sk_state; | |
2586 | ||
d4487491 LB |
2587 | /* We defined a new enum for TCP states that are exported in BPF |
2588 | * so as not force the internal TCP states to be frozen. The | |
2589 | * following checks will detect if an internal state value ever | |
2590 | * differs from the BPF value. If this ever happens, then we will | |
2591 | * need to remap the internal value to the BPF value before calling | |
2592 | * tcp_call_bpf_2arg. | |
2593 | */ | |
2594 | BUILD_BUG_ON((int)BPF_TCP_ESTABLISHED != (int)TCP_ESTABLISHED); | |
2595 | BUILD_BUG_ON((int)BPF_TCP_SYN_SENT != (int)TCP_SYN_SENT); | |
2596 | BUILD_BUG_ON((int)BPF_TCP_SYN_RECV != (int)TCP_SYN_RECV); | |
2597 | BUILD_BUG_ON((int)BPF_TCP_FIN_WAIT1 != (int)TCP_FIN_WAIT1); | |
2598 | BUILD_BUG_ON((int)BPF_TCP_FIN_WAIT2 != (int)TCP_FIN_WAIT2); | |
2599 | BUILD_BUG_ON((int)BPF_TCP_TIME_WAIT != (int)TCP_TIME_WAIT); | |
2600 | BUILD_BUG_ON((int)BPF_TCP_CLOSE != (int)TCP_CLOSE); | |
2601 | BUILD_BUG_ON((int)BPF_TCP_CLOSE_WAIT != (int)TCP_CLOSE_WAIT); | |
2602 | BUILD_BUG_ON((int)BPF_TCP_LAST_ACK != (int)TCP_LAST_ACK); | |
2603 | BUILD_BUG_ON((int)BPF_TCP_LISTEN != (int)TCP_LISTEN); | |
2604 | BUILD_BUG_ON((int)BPF_TCP_CLOSING != (int)TCP_CLOSING); | |
2605 | BUILD_BUG_ON((int)BPF_TCP_NEW_SYN_RECV != (int)TCP_NEW_SYN_RECV); | |
2606 | BUILD_BUG_ON((int)BPF_TCP_MAX_STATES != (int)TCP_MAX_STATES); | |
2607 | ||
97a19caf YS |
2608 | /* bpf uapi header bpf.h defines an anonymous enum with values |
2609 | * BPF_TCP_* used by bpf programs. Currently gcc built vmlinux | |
2610 | * is able to emit this enum in DWARF due to the above BUILD_BUG_ON. | |
2611 | * But clang built vmlinux does not have this enum in DWARF | |
2612 | * since clang removes the above code before generating IR/debuginfo. | |
2613 | * Let us explicitly emit the type debuginfo to ensure the | |
2614 | * above-mentioned anonymous enum in the vmlinux DWARF and hence BTF | |
2615 | * regardless of which compiler is used. | |
2616 | */ | |
2617 | BTF_TYPE_EMIT_ENUM(BPF_TCP_ESTABLISHED); | |
2618 | ||
d4487491 LB |
2619 | if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_STATE_CB_FLAG)) |
2620 | tcp_call_bpf_2arg(sk, BPF_SOCK_OPS_STATE_CB, oldstate, state); | |
e8fce239 | 2621 | |
490d5046 IJ |
2622 | switch (state) { |
2623 | case TCP_ESTABLISHED: | |
2624 | if (oldstate != TCP_ESTABLISHED) | |
81cc8a75 | 2625 | TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB); |
490d5046 IJ |
2626 | break; |
2627 | ||
2628 | case TCP_CLOSE: | |
2629 | if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED) | |
81cc8a75 | 2630 | TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS); |
490d5046 IJ |
2631 | |
2632 | sk->sk_prot->unhash(sk); | |
2633 | if (inet_csk(sk)->icsk_bind_hash && | |
2634 | !(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) | |
ab1e0a13 | 2635 | inet_put_port(sk); |
a8eceea8 | 2636 | fallthrough; |
490d5046 | 2637 | default: |
5a5f3a8d | 2638 | if (oldstate == TCP_ESTABLISHED) |
74688e48 | 2639 | TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB); |
490d5046 IJ |
2640 | } |
2641 | ||
2642 | /* Change state AFTER socket is unhashed to avoid closed | |
2643 | * socket sitting in hash tables. | |
2644 | */ | |
563e0bb0 | 2645 | inet_sk_state_store(sk, state); |
490d5046 IJ |
2646 | } |
2647 | EXPORT_SYMBOL_GPL(tcp_set_state); | |
2648 | ||
1da177e4 LT |
2649 | /* |
2650 | * State processing on a close. This implements the state shift for | |
2651 | * sending our FIN frame. Note that we only send a FIN for some | |
2652 | * states. A shutdown() may have already sent the FIN, or we may be | |
2653 | * closed. | |
2654 | */ | |
2655 | ||
9b5b5cff | 2656 | static const unsigned char new_state[16] = { |
1da177e4 | 2657 | /* current state: new state: action: */ |
0980c1e3 ED |
2658 | [0 /* (Invalid) */] = TCP_CLOSE, |
2659 | [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN, | |
2660 | [TCP_SYN_SENT] = TCP_CLOSE, | |
2661 | [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN, | |
2662 | [TCP_FIN_WAIT1] = TCP_FIN_WAIT1, | |
2663 | [TCP_FIN_WAIT2] = TCP_FIN_WAIT2, | |
2664 | [TCP_TIME_WAIT] = TCP_CLOSE, | |
2665 | [TCP_CLOSE] = TCP_CLOSE, | |
2666 | [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN, | |
2667 | [TCP_LAST_ACK] = TCP_LAST_ACK, | |
2668 | [TCP_LISTEN] = TCP_CLOSE, | |
2669 | [TCP_CLOSING] = TCP_CLOSING, | |
2670 | [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */ | |
1da177e4 LT |
2671 | }; |
2672 | ||
2673 | static int tcp_close_state(struct sock *sk) | |
2674 | { | |
2675 | int next = (int)new_state[sk->sk_state]; | |
2676 | int ns = next & TCP_STATE_MASK; | |
2677 | ||
2678 | tcp_set_state(sk, ns); | |
2679 | ||
2680 | return next & TCP_ACTION_FIN; | |
2681 | } | |
2682 | ||
2683 | /* | |
2684 | * Shutdown the sending side of a connection. Much like close except | |
1f29b058 | 2685 | * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD). |
1da177e4 LT |
2686 | */ |
2687 | ||
2688 | void tcp_shutdown(struct sock *sk, int how) | |
2689 | { | |
2690 | /* We need to grab some memory, and put together a FIN, | |
2691 | * and then put it into the queue to be sent. | |
2692 | * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92. | |
2693 | */ | |
2694 | if (!(how & SEND_SHUTDOWN)) | |
2695 | return; | |
2696 | ||
2697 | /* If we've already sent a FIN, or it's a closed state, skip this. */ | |
2698 | if ((1 << sk->sk_state) & | |
2699 | (TCPF_ESTABLISHED | TCPF_SYN_SENT | | |
2700 | TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) { | |
2701 | /* Clear out any half completed packets. FIN if needed. */ | |
2702 | if (tcp_close_state(sk)) | |
2703 | tcp_send_fin(sk); | |
2704 | } | |
2705 | } | |
4bc2f18b | 2706 | EXPORT_SYMBOL(tcp_shutdown); |
1da177e4 | 2707 | |
19757ceb ED |
2708 | int tcp_orphan_count_sum(void) |
2709 | { | |
2710 | int i, total = 0; | |
2711 | ||
2712 | for_each_possible_cpu(i) | |
2713 | total += per_cpu(tcp_orphan_count, i); | |
2714 | ||
2715 | return max(total, 0); | |
2716 | } | |
2717 | ||
2718 | static int tcp_orphan_cache; | |
2719 | static struct timer_list tcp_orphan_timer; | |
2720 | #define TCP_ORPHAN_TIMER_PERIOD msecs_to_jiffies(100) | |
2721 | ||
2722 | static void tcp_orphan_update(struct timer_list *unused) | |
2723 | { | |
2724 | WRITE_ONCE(tcp_orphan_cache, tcp_orphan_count_sum()); | |
2725 | mod_timer(&tcp_orphan_timer, jiffies + TCP_ORPHAN_TIMER_PERIOD); | |
2726 | } | |
2727 | ||
2728 | static bool tcp_too_many_orphans(int shift) | |
2729 | { | |
47e6ab24 KI |
2730 | return READ_ONCE(tcp_orphan_cache) << shift > |
2731 | READ_ONCE(sysctl_tcp_max_orphans); | |
19757ceb ED |
2732 | } |
2733 | ||
efcdbf24 AS |
2734 | bool tcp_check_oom(struct sock *sk, int shift) |
2735 | { | |
2736 | bool too_many_orphans, out_of_socket_memory; | |
2737 | ||
19757ceb | 2738 | too_many_orphans = tcp_too_many_orphans(shift); |
efcdbf24 AS |
2739 | out_of_socket_memory = tcp_out_of_memory(sk); |
2740 | ||
e87cc472 JP |
2741 | if (too_many_orphans) |
2742 | net_info_ratelimited("too many orphaned sockets\n"); | |
2743 | if (out_of_socket_memory) | |
2744 | net_info_ratelimited("out of memory -- consider tuning tcp_mem\n"); | |
efcdbf24 AS |
2745 | return too_many_orphans || out_of_socket_memory; |
2746 | } | |
2747 | ||
77c3c956 | 2748 | void __tcp_close(struct sock *sk, long timeout) |
1da177e4 LT |
2749 | { |
2750 | struct sk_buff *skb; | |
2751 | int data_was_unread = 0; | |
75c2d907 | 2752 | int state; |
1da177e4 | 2753 | |
e14cadfd | 2754 | WRITE_ONCE(sk->sk_shutdown, SHUTDOWN_MASK); |
1da177e4 LT |
2755 | |
2756 | if (sk->sk_state == TCP_LISTEN) { | |
2757 | tcp_set_state(sk, TCP_CLOSE); | |
2758 | ||
2759 | /* Special case. */ | |
0a5578cf | 2760 | inet_csk_listen_stop(sk); |
1da177e4 LT |
2761 | |
2762 | goto adjudge_to_death; | |
2763 | } | |
2764 | ||
2765 | /* We need to flush the recv. buffs. We do this only on the | |
2766 | * descriptor close, not protocol-sourced closes, because the | |
2767 | * reader process may not have drained the data yet! | |
2768 | */ | |
2769 | while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) { | |
e11ecddf ED |
2770 | u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq; |
2771 | ||
2772 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) | |
2773 | len--; | |
1da177e4 LT |
2774 | data_was_unread += len; |
2775 | __kfree_skb(skb); | |
2776 | } | |
2777 | ||
565b7b2d KK |
2778 | /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */ |
2779 | if (sk->sk_state == TCP_CLOSE) | |
2780 | goto adjudge_to_death; | |
2781 | ||
65bb723c GR |
2782 | /* As outlined in RFC 2525, section 2.17, we send a RST here because |
2783 | * data was lost. To witness the awful effects of the old behavior of | |
2784 | * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk | |
2785 | * GET in an FTP client, suspend the process, wait for the client to | |
2786 | * advertise a zero window, then kill -9 the FTP client, wheee... | |
2787 | * Note: timeout is always zero in such a case. | |
1da177e4 | 2788 | */ |
ee995283 PE |
2789 | if (unlikely(tcp_sk(sk)->repair)) { |
2790 | sk->sk_prot->disconnect(sk, 0); | |
2791 | } else if (data_was_unread) { | |
1da177e4 | 2792 | /* Unread data was tossed, zap the connection. */ |
6aef70a8 | 2793 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE); |
1da177e4 | 2794 | tcp_set_state(sk, TCP_CLOSE); |
aa133076 | 2795 | tcp_send_active_reset(sk, sk->sk_allocation); |
1da177e4 LT |
2796 | } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) { |
2797 | /* Check zero linger _after_ checking for unread data. */ | |
2798 | sk->sk_prot->disconnect(sk, 0); | |
6aef70a8 | 2799 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONDATA); |
1da177e4 LT |
2800 | } else if (tcp_close_state(sk)) { |
2801 | /* We FIN if the application ate all the data before | |
2802 | * zapping the connection. | |
2803 | */ | |
2804 | ||
2805 | /* RED-PEN. Formally speaking, we have broken TCP state | |
2806 | * machine. State transitions: | |
2807 | * | |
2808 | * TCP_ESTABLISHED -> TCP_FIN_WAIT1 | |
2809 | * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible) | |
2810 | * TCP_CLOSE_WAIT -> TCP_LAST_ACK | |
2811 | * | |
2812 | * are legal only when FIN has been sent (i.e. in window), | |
2813 | * rather than queued out of window. Purists blame. | |
2814 | * | |
2815 | * F.e. "RFC state" is ESTABLISHED, | |
2816 | * if Linux state is FIN-WAIT-1, but FIN is still not sent. | |
2817 | * | |
2818 | * The visible declinations are that sometimes | |
2819 | * we enter time-wait state, when it is not required really | |
2820 | * (harmless), do not send active resets, when they are | |
2821 | * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when | |
2822 | * they look as CLOSING or LAST_ACK for Linux) | |
2823 | * Probably, I missed some more holelets. | |
2824 | * --ANK | |
8336886f JC |
2825 | * XXX (TFO) - To start off we don't support SYN+ACK+FIN |
2826 | * in a single packet! (May consider it later but will | |
2827 | * probably need API support or TCP_CORK SYN-ACK until | |
2828 | * data is written and socket is closed.) | |
1da177e4 LT |
2829 | */ |
2830 | tcp_send_fin(sk); | |
2831 | } | |
2832 | ||
2833 | sk_stream_wait_close(sk, timeout); | |
2834 | ||
2835 | adjudge_to_death: | |
75c2d907 HX |
2836 | state = sk->sk_state; |
2837 | sock_hold(sk); | |
2838 | sock_orphan(sk); | |
75c2d907 | 2839 | |
1da177e4 LT |
2840 | local_bh_disable(); |
2841 | bh_lock_sock(sk); | |
8873c064 ED |
2842 | /* remove backlog if any, without releasing ownership. */ |
2843 | __release_sock(sk); | |
1da177e4 | 2844 | |
19757ceb | 2845 | this_cpu_inc(tcp_orphan_count); |
eb4dea58 | 2846 | |
75c2d907 HX |
2847 | /* Have we already been destroyed by a softirq or backlog? */ |
2848 | if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE) | |
2849 | goto out; | |
1da177e4 LT |
2850 | |
2851 | /* This is a (useful) BSD violating of the RFC. There is a | |
2852 | * problem with TCP as specified in that the other end could | |
2853 | * keep a socket open forever with no application left this end. | |
b10bd54c | 2854 | * We use a 1 minute timeout (about the same as BSD) then kill |
1da177e4 LT |
2855 | * our end. If they send after that then tough - BUT: long enough |
2856 | * that we won't make the old 4*rto = almost no time - whoops | |
2857 | * reset mistake. | |
2858 | * | |
2859 | * Nope, it was not mistake. It is really desired behaviour | |
2860 | * f.e. on http servers, when such sockets are useless, but | |
2861 | * consume significant resources. Let's do it with special | |
2862 | * linger2 option. --ANK | |
2863 | */ | |
2864 | ||
2865 | if (sk->sk_state == TCP_FIN_WAIT2) { | |
2866 | struct tcp_sock *tp = tcp_sk(sk); | |
a81722dd | 2867 | if (READ_ONCE(tp->linger2) < 0) { |
1da177e4 LT |
2868 | tcp_set_state(sk, TCP_CLOSE); |
2869 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
02a1d6e7 | 2870 | __NET_INC_STATS(sock_net(sk), |
de0744af | 2871 | LINUX_MIB_TCPABORTONLINGER); |
1da177e4 | 2872 | } else { |
463c84b9 | 2873 | const int tmo = tcp_fin_time(sk); |
1da177e4 LT |
2874 | |
2875 | if (tmo > TCP_TIMEWAIT_LEN) { | |
52499afe DM |
2876 | inet_csk_reset_keepalive_timer(sk, |
2877 | tmo - TCP_TIMEWAIT_LEN); | |
1da177e4 | 2878 | } else { |
1da177e4 LT |
2879 | tcp_time_wait(sk, TCP_FIN_WAIT2, tmo); |
2880 | goto out; | |
2881 | } | |
2882 | } | |
2883 | } | |
2884 | if (sk->sk_state != TCP_CLOSE) { | |
efcdbf24 | 2885 | if (tcp_check_oom(sk, 0)) { |
1da177e4 LT |
2886 | tcp_set_state(sk, TCP_CLOSE); |
2887 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
02a1d6e7 | 2888 | __NET_INC_STATS(sock_net(sk), |
de0744af | 2889 | LINUX_MIB_TCPABORTONMEMORY); |
4ee806d5 DS |
2890 | } else if (!check_net(sock_net(sk))) { |
2891 | /* Not possible to send reset; just close */ | |
2892 | tcp_set_state(sk, TCP_CLOSE); | |
1da177e4 LT |
2893 | } |
2894 | } | |
1da177e4 | 2895 | |
8336886f | 2896 | if (sk->sk_state == TCP_CLOSE) { |
d983ea6f ED |
2897 | struct request_sock *req; |
2898 | ||
2899 | req = rcu_dereference_protected(tcp_sk(sk)->fastopen_rsk, | |
2900 | lockdep_sock_is_held(sk)); | |
8336886f JC |
2901 | /* We could get here with a non-NULL req if the socket is |
2902 | * aborted (e.g., closed with unread data) before 3WHS | |
2903 | * finishes. | |
2904 | */ | |
00db4124 | 2905 | if (req) |
8336886f | 2906 | reqsk_fastopen_remove(sk, req, false); |
0a5578cf | 2907 | inet_csk_destroy_sock(sk); |
8336886f | 2908 | } |
1da177e4 LT |
2909 | /* Otherwise, socket is reprieved until protocol close. */ |
2910 | ||
2911 | out: | |
2912 | bh_unlock_sock(sk); | |
2913 | local_bh_enable(); | |
77c3c956 PA |
2914 | } |
2915 | ||
2916 | void tcp_close(struct sock *sk, long timeout) | |
2917 | { | |
2918 | lock_sock(sk); | |
2919 | __tcp_close(sk, timeout); | |
8873c064 | 2920 | release_sock(sk); |
1da177e4 LT |
2921 | sock_put(sk); |
2922 | } | |
4bc2f18b | 2923 | EXPORT_SYMBOL(tcp_close); |
1da177e4 LT |
2924 | |
2925 | /* These states need RST on ABORT according to RFC793 */ | |
2926 | ||
a2a385d6 | 2927 | static inline bool tcp_need_reset(int state) |
1da177e4 LT |
2928 | { |
2929 | return (1 << state) & | |
2930 | (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 | | |
a7150e38 | 2931 | TCPF_FIN_WAIT2 | TCPF_SYN_RECV); |
1da177e4 LT |
2932 | } |
2933 | ||
75c119af ED |
2934 | static void tcp_rtx_queue_purge(struct sock *sk) |
2935 | { | |
2936 | struct rb_node *p = rb_first(&sk->tcp_rtx_queue); | |
2937 | ||
2bec445f | 2938 | tcp_sk(sk)->highest_sack = NULL; |
75c119af ED |
2939 | while (p) { |
2940 | struct sk_buff *skb = rb_to_skb(p); | |
2941 | ||
2942 | p = rb_next(p); | |
2943 | /* Since we are deleting whole queue, no need to | |
2944 | * list_del(&skb->tcp_tsorted_anchor) | |
2945 | */ | |
2946 | tcp_rtx_queue_unlink(skb, sk); | |
03271f3a | 2947 | tcp_wmem_free_skb(sk, skb); |
75c119af ED |
2948 | } |
2949 | } | |
2950 | ||
ac3f09ba ED |
2951 | void tcp_write_queue_purge(struct sock *sk) |
2952 | { | |
2953 | struct sk_buff *skb; | |
2954 | ||
2955 | tcp_chrono_stop(sk, TCP_CHRONO_BUSY); | |
2956 | while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) { | |
2957 | tcp_skb_tsorted_anchor_cleanup(skb); | |
03271f3a | 2958 | tcp_wmem_free_skb(sk, skb); |
ac3f09ba | 2959 | } |
75c119af | 2960 | tcp_rtx_queue_purge(sk); |
ac3f09ba | 2961 | INIT_LIST_HEAD(&tcp_sk(sk)->tsorted_sent_queue); |
ac3f09ba | 2962 | tcp_clear_all_retrans_hints(tcp_sk(sk)); |
bffd168c | 2963 | tcp_sk(sk)->packets_out = 0; |
04c03114 | 2964 | inet_csk(sk)->icsk_backoff = 0; |
ac3f09ba ED |
2965 | } |
2966 | ||
1da177e4 LT |
2967 | int tcp_disconnect(struct sock *sk, int flags) |
2968 | { | |
2969 | struct inet_sock *inet = inet_sk(sk); | |
463c84b9 | 2970 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 2971 | struct tcp_sock *tp = tcp_sk(sk); |
1da177e4 | 2972 | int old_state = sk->sk_state; |
0f317464 | 2973 | u32 seq; |
1da177e4 | 2974 | |
4faeee0c ED |
2975 | /* Deny disconnect if other threads are blocked in sk_wait_event() |
2976 | * or inet_wait_for_connect(). | |
2977 | */ | |
2978 | if (sk->sk_wait_pending) | |
2979 | return -EBUSY; | |
2980 | ||
1da177e4 LT |
2981 | if (old_state != TCP_CLOSE) |
2982 | tcp_set_state(sk, TCP_CLOSE); | |
2983 | ||
2984 | /* ABORT function of RFC793 */ | |
2985 | if (old_state == TCP_LISTEN) { | |
0a5578cf | 2986 | inet_csk_listen_stop(sk); |
ee995283 | 2987 | } else if (unlikely(tp->repair)) { |
e13ec3da | 2988 | WRITE_ONCE(sk->sk_err, ECONNABORTED); |
1da177e4 LT |
2989 | } else if (tcp_need_reset(old_state) || |
2990 | (tp->snd_nxt != tp->write_seq && | |
2991 | (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) { | |
caa20d9a | 2992 | /* The last check adjusts for discrepancy of Linux wrt. RFC |
1da177e4 LT |
2993 | * states |
2994 | */ | |
2995 | tcp_send_active_reset(sk, gfp_any()); | |
e13ec3da | 2996 | WRITE_ONCE(sk->sk_err, ECONNRESET); |
a7150e38 | 2997 | } else if (old_state == TCP_SYN_SENT) |
e13ec3da | 2998 | WRITE_ONCE(sk->sk_err, ECONNRESET); |
1da177e4 LT |
2999 | |
3000 | tcp_clear_xmit_timers(sk); | |
3001 | __skb_queue_purge(&sk->sk_receive_queue); | |
7db48e98 | 3002 | WRITE_ONCE(tp->copied_seq, tp->rcv_nxt); |
7b6a893a | 3003 | WRITE_ONCE(tp->urg_data, 0); |
fe067e8a | 3004 | tcp_write_queue_purge(sk); |
cf1ef3f0 | 3005 | tcp_fastopen_active_disable_ofo_check(sk); |
9f5afeae | 3006 | skb_rbtree_purge(&tp->out_of_order_queue); |
1da177e4 | 3007 | |
c720c7e8 | 3008 | inet->inet_dport = 0; |
1da177e4 | 3009 | |
e0833d1f | 3010 | inet_bhash2_reset_saddr(sk); |
1da177e4 | 3011 | |
e14cadfd | 3012 | WRITE_ONCE(sk->sk_shutdown, 0); |
1da177e4 | 3013 | sock_reset_flag(sk, SOCK_DONE); |
740b0f18 | 3014 | tp->srtt_us = 0; |
b9e2e689 | 3015 | tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT); |
3f6c65d6 | 3016 | tp->rcv_rtt_last_tsecr = 0; |
0f317464 ED |
3017 | |
3018 | seq = tp->write_seq + tp->max_window + 2; | |
3019 | if (!seq) | |
3020 | seq = 1; | |
3021 | WRITE_ONCE(tp->write_seq, seq); | |
3022 | ||
463c84b9 | 3023 | icsk->icsk_backoff = 0; |
6687e988 | 3024 | icsk->icsk_probes_out = 0; |
9d9b1ee0 | 3025 | icsk->icsk_probes_tstamp = 0; |
6a408147 | 3026 | icsk->icsk_rto = TCP_TIMEOUT_INIT; |
ca584ba0 | 3027 | icsk->icsk_rto_min = TCP_RTO_MIN; |
2b8ee4f0 | 3028 | icsk->icsk_delack_max = TCP_DELACK_MAX; |
0b6a05c1 | 3029 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; |
40570375 | 3030 | tcp_snd_cwnd_set(tp, TCP_INIT_CWND); |
1da177e4 | 3031 | tp->snd_cwnd_cnt = 0; |
f4ce91ce NC |
3032 | tp->is_cwnd_limited = 0; |
3033 | tp->max_packets_out = 0; | |
1fdf475a | 3034 | tp->window_clamp = 0; |
2fbdd562 | 3035 | tp->delivered = 0; |
e21db6f6 | 3036 | tp->delivered_ce = 0; |
ce69e563 CP |
3037 | if (icsk->icsk_ca_ops->release) |
3038 | icsk->icsk_ca_ops->release(sk); | |
3039 | memset(icsk->icsk_ca_priv, 0, sizeof(icsk->icsk_ca_priv)); | |
8919a9b3 | 3040 | icsk->icsk_ca_initialized = 0; |
6687e988 | 3041 | tcp_set_ca_state(sk, TCP_CA_Open); |
d4761754 | 3042 | tp->is_sack_reneg = 0; |
1da177e4 | 3043 | tcp_clear_retrans(tp); |
c13c48c0 | 3044 | tp->total_retrans = 0; |
463c84b9 | 3045 | inet_csk_delack_init(sk); |
499350a5 WW |
3046 | /* Initialize rcv_mss to TCP_MIN_MSS to avoid division by 0 |
3047 | * issue in __tcp_select_window() | |
3048 | */ | |
3049 | icsk->icsk_ack.rcv_mss = TCP_MIN_MSS; | |
b40b4f79 | 3050 | memset(&tp->rx_opt, 0, sizeof(tp->rx_opt)); |
1da177e4 | 3051 | __sk_dst_reset(sk); |
8f905c0e | 3052 | dst_release(xchg((__force struct dst_entry **)&sk->sk_rx_dst, NULL)); |
17c3060b | 3053 | tcp_saved_syn_free(tp); |
5d9f4262 | 3054 | tp->compressed_ack = 0; |
784f8344 ED |
3055 | tp->segs_in = 0; |
3056 | tp->segs_out = 0; | |
ba113c3a | 3057 | tp->bytes_sent = 0; |
e858faf5 CP |
3058 | tp->bytes_acked = 0; |
3059 | tp->bytes_received = 0; | |
fb31c9b9 | 3060 | tp->bytes_retrans = 0; |
db7ffee6 ED |
3061 | tp->data_segs_in = 0; |
3062 | tp->data_segs_out = 0; | |
7788174e YC |
3063 | tp->duplicate_sack[0].start_seq = 0; |
3064 | tp->duplicate_sack[0].end_seq = 0; | |
7e10b655 | 3065 | tp->dsack_dups = 0; |
7ec65372 | 3066 | tp->reord_seen = 0; |
5c701549 ED |
3067 | tp->retrans_out = 0; |
3068 | tp->sacked_out = 0; | |
3069 | tp->tlp_high_seq = 0; | |
3070 | tp->last_oow_ack_time = 0; | |
29c1c446 | 3071 | tp->plb_rehash = 0; |
6cda8b74 ED |
3072 | /* There's a bubble in the pipe until at least the first ACK. */ |
3073 | tp->app_limited = ~0U; | |
300b655d | 3074 | tp->rate_app_limited = 1; |
792c4354 ED |
3075 | tp->rack.mstamp = 0; |
3076 | tp->rack.advanced = 0; | |
3077 | tp->rack.reo_wnd_steps = 1; | |
3078 | tp->rack.last_delivered = 0; | |
3079 | tp->rack.reo_wnd_persist = 0; | |
3080 | tp->rack.dsack_seen = 0; | |
6bcdc40d ED |
3081 | tp->syn_data_acked = 0; |
3082 | tp->rx_opt.saw_tstamp = 0; | |
3083 | tp->rx_opt.dsack = 0; | |
3084 | tp->rx_opt.num_sacks = 0; | |
f9af2dbb | 3085 | tp->rcv_ooopack = 0; |
6cda8b74 | 3086 | |
1da177e4 | 3087 | |
7db92362 WW |
3088 | /* Clean up fastopen related fields */ |
3089 | tcp_free_fastopen_req(tp); | |
08e39c0d | 3090 | inet_clear_bit(DEFER_CONNECT, sk); |
48027478 | 3091 | tp->fastopen_client_fail = 0; |
7db92362 | 3092 | |
c720c7e8 | 3093 | WARN_ON(inet->inet_num && !icsk->icsk_bind_hash); |
1da177e4 | 3094 | |
9b42d55a LR |
3095 | if (sk->sk_frag.page) { |
3096 | put_page(sk->sk_frag.page); | |
3097 | sk->sk_frag.page = NULL; | |
3098 | sk->sk_frag.offset = 0; | |
3099 | } | |
e3ae2365 | 3100 | sk_error_report(sk); |
a01512b1 | 3101 | return 0; |
1da177e4 | 3102 | } |
4bc2f18b | 3103 | EXPORT_SYMBOL(tcp_disconnect); |
1da177e4 | 3104 | |
a2a385d6 | 3105 | static inline bool tcp_can_repair_sock(const struct sock *sk) |
ee995283 | 3106 | { |
cb388e7e | 3107 | return sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) && |
319b0534 | 3108 | (sk->sk_state != TCP_LISTEN); |
ee995283 PE |
3109 | } |
3110 | ||
d38d2b00 | 3111 | static int tcp_repair_set_window(struct tcp_sock *tp, sockptr_t optbuf, int len) |
b1ed4c4f AV |
3112 | { |
3113 | struct tcp_repair_window opt; | |
3114 | ||
3115 | if (!tp->repair) | |
3116 | return -EPERM; | |
3117 | ||
3118 | if (len != sizeof(opt)) | |
3119 | return -EINVAL; | |
3120 | ||
d38d2b00 | 3121 | if (copy_from_sockptr(&opt, optbuf, sizeof(opt))) |
b1ed4c4f AV |
3122 | return -EFAULT; |
3123 | ||
3124 | if (opt.max_window < opt.snd_wnd) | |
3125 | return -EINVAL; | |
3126 | ||
3127 | if (after(opt.snd_wl1, tp->rcv_nxt + opt.rcv_wnd)) | |
3128 | return -EINVAL; | |
3129 | ||
3130 | if (after(opt.rcv_wup, tp->rcv_nxt)) | |
3131 | return -EINVAL; | |
3132 | ||
3133 | tp->snd_wl1 = opt.snd_wl1; | |
3134 | tp->snd_wnd = opt.snd_wnd; | |
3135 | tp->max_window = opt.max_window; | |
3136 | ||
3137 | tp->rcv_wnd = opt.rcv_wnd; | |
3138 | tp->rcv_wup = opt.rcv_wup; | |
3139 | ||
3140 | return 0; | |
3141 | } | |
3142 | ||
d38d2b00 CH |
3143 | static int tcp_repair_options_est(struct sock *sk, sockptr_t optbuf, |
3144 | unsigned int len) | |
b139ba4e | 3145 | { |
15e56515 | 3146 | struct tcp_sock *tp = tcp_sk(sk); |
de248a75 | 3147 | struct tcp_repair_opt opt; |
d3c48151 | 3148 | size_t offset = 0; |
b139ba4e | 3149 | |
de248a75 | 3150 | while (len >= sizeof(opt)) { |
d3c48151 | 3151 | if (copy_from_sockptr_offset(&opt, optbuf, offset, sizeof(opt))) |
b139ba4e PE |
3152 | return -EFAULT; |
3153 | ||
d3c48151 | 3154 | offset += sizeof(opt); |
de248a75 | 3155 | len -= sizeof(opt); |
b139ba4e | 3156 | |
de248a75 PE |
3157 | switch (opt.opt_code) { |
3158 | case TCPOPT_MSS: | |
3159 | tp->rx_opt.mss_clamp = opt.opt_val; | |
15e56515 | 3160 | tcp_mtup_init(sk); |
b139ba4e | 3161 | break; |
de248a75 | 3162 | case TCPOPT_WINDOW: |
bc26ccd8 AV |
3163 | { |
3164 | u16 snd_wscale = opt.opt_val & 0xFFFF; | |
3165 | u16 rcv_wscale = opt.opt_val >> 16; | |
3166 | ||
589c49cb | 3167 | if (snd_wscale > TCP_MAX_WSCALE || rcv_wscale > TCP_MAX_WSCALE) |
bc26ccd8 | 3168 | return -EFBIG; |
b139ba4e | 3169 | |
bc26ccd8 AV |
3170 | tp->rx_opt.snd_wscale = snd_wscale; |
3171 | tp->rx_opt.rcv_wscale = rcv_wscale; | |
3172 | tp->rx_opt.wscale_ok = 1; | |
3173 | } | |
b139ba4e | 3174 | break; |
b139ba4e | 3175 | case TCPOPT_SACK_PERM: |
de248a75 PE |
3176 | if (opt.opt_val != 0) |
3177 | return -EINVAL; | |
3178 | ||
b139ba4e | 3179 | tp->rx_opt.sack_ok |= TCP_SACK_SEEN; |
b139ba4e PE |
3180 | break; |
3181 | case TCPOPT_TIMESTAMP: | |
de248a75 PE |
3182 | if (opt.opt_val != 0) |
3183 | return -EINVAL; | |
3184 | ||
b139ba4e PE |
3185 | tp->rx_opt.tstamp_ok = 1; |
3186 | break; | |
3187 | } | |
3188 | } | |
3189 | ||
3190 | return 0; | |
3191 | } | |
3192 | ||
a842fe14 ED |
3193 | DEFINE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); |
3194 | EXPORT_SYMBOL(tcp_tx_delay_enabled); | |
3195 | ||
3196 | static void tcp_enable_tx_delay(void) | |
3197 | { | |
3198 | if (!static_branch_unlikely(&tcp_tx_delay_enabled)) { | |
3199 | static int __tcp_tx_delay_enabled = 0; | |
3200 | ||
3201 | if (cmpxchg(&__tcp_tx_delay_enabled, 0, 1) == 0) { | |
3202 | static_branch_enable(&tcp_tx_delay_enabled); | |
3203 | pr_info("TCP_TX_DELAY enabled\n"); | |
3204 | } | |
3205 | } | |
3206 | } | |
3207 | ||
db10538a CH |
3208 | /* When set indicates to always queue non-full frames. Later the user clears |
3209 | * this option and we transmit any pending partial frames in the queue. This is | |
3210 | * meant to be used alongside sendfile() to get properly filled frames when the | |
3211 | * user (for example) must write out headers with a write() call first and then | |
3212 | * use sendfile to send out the data parts. | |
3213 | * | |
3214 | * TCP_CORK can be set together with TCP_NODELAY and it is stronger than | |
3215 | * TCP_NODELAY. | |
3216 | */ | |
6fadaa56 | 3217 | void __tcp_sock_set_cork(struct sock *sk, bool on) |
db10538a CH |
3218 | { |
3219 | struct tcp_sock *tp = tcp_sk(sk); | |
3220 | ||
3221 | if (on) { | |
3222 | tp->nonagle |= TCP_NAGLE_CORK; | |
3223 | } else { | |
3224 | tp->nonagle &= ~TCP_NAGLE_CORK; | |
3225 | if (tp->nonagle & TCP_NAGLE_OFF) | |
3226 | tp->nonagle |= TCP_NAGLE_PUSH; | |
3227 | tcp_push_pending_frames(sk); | |
3228 | } | |
3229 | } | |
3230 | ||
3231 | void tcp_sock_set_cork(struct sock *sk, bool on) | |
3232 | { | |
3233 | lock_sock(sk); | |
3234 | __tcp_sock_set_cork(sk, on); | |
3235 | release_sock(sk); | |
3236 | } | |
3237 | EXPORT_SYMBOL(tcp_sock_set_cork); | |
3238 | ||
12abc5ee CH |
3239 | /* TCP_NODELAY is weaker than TCP_CORK, so that this option on corked socket is |
3240 | * remembered, but it is not activated until cork is cleared. | |
3241 | * | |
3242 | * However, when TCP_NODELAY is set we make an explicit push, which overrides | |
3243 | * even TCP_CORK for currently queued segments. | |
3244 | */ | |
6fadaa56 | 3245 | void __tcp_sock_set_nodelay(struct sock *sk, bool on) |
12abc5ee CH |
3246 | { |
3247 | if (on) { | |
3248 | tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH; | |
3249 | tcp_push_pending_frames(sk); | |
3250 | } else { | |
3251 | tcp_sk(sk)->nonagle &= ~TCP_NAGLE_OFF; | |
3252 | } | |
3253 | } | |
3254 | ||
3255 | void tcp_sock_set_nodelay(struct sock *sk) | |
3256 | { | |
3257 | lock_sock(sk); | |
3258 | __tcp_sock_set_nodelay(sk, true); | |
3259 | release_sock(sk); | |
3260 | } | |
3261 | EXPORT_SYMBOL(tcp_sock_set_nodelay); | |
3262 | ||
ddd061b8 CH |
3263 | static void __tcp_sock_set_quickack(struct sock *sk, int val) |
3264 | { | |
3265 | if (!val) { | |
3266 | inet_csk_enter_pingpong_mode(sk); | |
3267 | return; | |
3268 | } | |
3269 | ||
3270 | inet_csk_exit_pingpong_mode(sk); | |
3271 | if ((1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) && | |
3272 | inet_csk_ack_scheduled(sk)) { | |
3273 | inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_PUSHED; | |
3274 | tcp_cleanup_rbuf(sk, 1); | |
3275 | if (!(val & 1)) | |
3276 | inet_csk_enter_pingpong_mode(sk); | |
3277 | } | |
3278 | } | |
3279 | ||
3280 | void tcp_sock_set_quickack(struct sock *sk, int val) | |
3281 | { | |
3282 | lock_sock(sk); | |
3283 | __tcp_sock_set_quickack(sk, val); | |
3284 | release_sock(sk); | |
3285 | } | |
3286 | EXPORT_SYMBOL(tcp_sock_set_quickack); | |
3287 | ||
557eadfc CH |
3288 | int tcp_sock_set_syncnt(struct sock *sk, int val) |
3289 | { | |
3290 | if (val < 1 || val > MAX_TCP_SYNCNT) | |
3291 | return -EINVAL; | |
3292 | ||
3a037f0f | 3293 | WRITE_ONCE(inet_csk(sk)->icsk_syn_retries, val); |
557eadfc CH |
3294 | return 0; |
3295 | } | |
3296 | EXPORT_SYMBOL(tcp_sock_set_syncnt); | |
3297 | ||
d58f2e15 | 3298 | int tcp_sock_set_user_timeout(struct sock *sk, int val) |
c488aead | 3299 | { |
d58f2e15 ED |
3300 | /* Cap the max time in ms TCP will retry or probe the window |
3301 | * before giving up and aborting (ETIMEDOUT) a connection. | |
3302 | */ | |
3303 | if (val < 0) | |
3304 | return -EINVAL; | |
3305 | ||
26023e91 | 3306 | WRITE_ONCE(inet_csk(sk)->icsk_user_timeout, val); |
d58f2e15 | 3307 | return 0; |
c488aead CH |
3308 | } |
3309 | EXPORT_SYMBOL(tcp_sock_set_user_timeout); | |
3310 | ||
aad4a0a9 | 3311 | int tcp_sock_set_keepidle_locked(struct sock *sk, int val) |
71c48eb8 CH |
3312 | { |
3313 | struct tcp_sock *tp = tcp_sk(sk); | |
3314 | ||
3315 | if (val < 1 || val > MAX_TCP_KEEPIDLE) | |
3316 | return -EINVAL; | |
3317 | ||
4164245c ED |
3318 | /* Paired with WRITE_ONCE() in keepalive_time_when() */ |
3319 | WRITE_ONCE(tp->keepalive_time, val * HZ); | |
71c48eb8 CH |
3320 | if (sock_flag(sk, SOCK_KEEPOPEN) && |
3321 | !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) { | |
3322 | u32 elapsed = keepalive_time_elapsed(tp); | |
3323 | ||
3324 | if (tp->keepalive_time > elapsed) | |
3325 | elapsed = tp->keepalive_time - elapsed; | |
3326 | else | |
3327 | elapsed = 0; | |
3328 | inet_csk_reset_keepalive_timer(sk, elapsed); | |
3329 | } | |
3330 | ||
3331 | return 0; | |
3332 | } | |
3333 | ||
3334 | int tcp_sock_set_keepidle(struct sock *sk, int val) | |
3335 | { | |
3336 | int err; | |
3337 | ||
3338 | lock_sock(sk); | |
aad4a0a9 | 3339 | err = tcp_sock_set_keepidle_locked(sk, val); |
71c48eb8 CH |
3340 | release_sock(sk); |
3341 | return err; | |
3342 | } | |
3343 | EXPORT_SYMBOL(tcp_sock_set_keepidle); | |
3344 | ||
d41ecaac CH |
3345 | int tcp_sock_set_keepintvl(struct sock *sk, int val) |
3346 | { | |
3347 | if (val < 1 || val > MAX_TCP_KEEPINTVL) | |
3348 | return -EINVAL; | |
3349 | ||
5ecf9d4f | 3350 | WRITE_ONCE(tcp_sk(sk)->keepalive_intvl, val * HZ); |
d41ecaac CH |
3351 | return 0; |
3352 | } | |
3353 | EXPORT_SYMBOL(tcp_sock_set_keepintvl); | |
3354 | ||
480aeb96 CH |
3355 | int tcp_sock_set_keepcnt(struct sock *sk, int val) |
3356 | { | |
3357 | if (val < 1 || val > MAX_TCP_KEEPCNT) | |
3358 | return -EINVAL; | |
3359 | ||
6e5e1de6 ED |
3360 | /* Paired with READ_ONCE() in keepalive_probes() */ |
3361 | WRITE_ONCE(tcp_sk(sk)->keepalive_probes, val); | |
480aeb96 CH |
3362 | return 0; |
3363 | } | |
3364 | EXPORT_SYMBOL(tcp_sock_set_keepcnt); | |
3365 | ||
cb811109 P |
3366 | int tcp_set_window_clamp(struct sock *sk, int val) |
3367 | { | |
3368 | struct tcp_sock *tp = tcp_sk(sk); | |
3369 | ||
3370 | if (!val) { | |
3371 | if (sk->sk_state != TCP_CLOSE) | |
3372 | return -EINVAL; | |
3373 | tp->window_clamp = 0; | |
3374 | } else { | |
3375 | tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ? | |
3376 | SOCK_MIN_RCVBUF / 2 : val; | |
3aa7857f | 3377 | tp->rcv_ssthresh = min(tp->rcv_wnd, tp->window_clamp); |
cb811109 P |
3378 | } |
3379 | return 0; | |
3380 | } | |
3381 | ||
1da177e4 LT |
3382 | /* |
3383 | * Socket option code for TCP. | |
3384 | */ | |
0c751f70 MKL |
3385 | int do_tcp_setsockopt(struct sock *sk, int level, int optname, |
3386 | sockptr_t optval, unsigned int optlen) | |
1da177e4 LT |
3387 | { |
3388 | struct tcp_sock *tp = tcp_sk(sk); | |
463c84b9 | 3389 | struct inet_connection_sock *icsk = inet_csk(sk); |
1e579caa | 3390 | struct net *net = sock_net(sk); |
1da177e4 LT |
3391 | int val; |
3392 | int err = 0; | |
3393 | ||
e56fb50f WAS |
3394 | /* These are data/string values, all the others are ints */ |
3395 | switch (optname) { | |
3396 | case TCP_CONGESTION: { | |
5f8ef48d SH |
3397 | char name[TCP_CA_NAME_MAX]; |
3398 | ||
3399 | if (optlen < 1) | |
3400 | return -EINVAL; | |
3401 | ||
d38d2b00 | 3402 | val = strncpy_from_sockptr(name, optval, |
4fdb78d3 | 3403 | min_t(long, TCP_CA_NAME_MAX-1, optlen)); |
5f8ef48d SH |
3404 | if (val < 0) |
3405 | return -EFAULT; | |
3406 | name[val] = 0; | |
3407 | ||
cb388e7e | 3408 | sockopt_lock_sock(sk); |
84e5a0f2 | 3409 | err = tcp_set_congestion_control(sk, name, !has_current_bpf_ctx(), |
cb388e7e MKL |
3410 | sockopt_ns_capable(sock_net(sk)->user_ns, |
3411 | CAP_NET_ADMIN)); | |
3412 | sockopt_release_sock(sk); | |
5f8ef48d SH |
3413 | return err; |
3414 | } | |
734942cc DW |
3415 | case TCP_ULP: { |
3416 | char name[TCP_ULP_NAME_MAX]; | |
3417 | ||
3418 | if (optlen < 1) | |
3419 | return -EINVAL; | |
3420 | ||
d38d2b00 | 3421 | val = strncpy_from_sockptr(name, optval, |
734942cc DW |
3422 | min_t(long, TCP_ULP_NAME_MAX - 1, |
3423 | optlen)); | |
3424 | if (val < 0) | |
3425 | return -EFAULT; | |
3426 | name[val] = 0; | |
3427 | ||
cb388e7e | 3428 | sockopt_lock_sock(sk); |
734942cc | 3429 | err = tcp_set_ulp(sk, name); |
cb388e7e | 3430 | sockopt_release_sock(sk); |
734942cc DW |
3431 | return err; |
3432 | } | |
1fba70e5 | 3433 | case TCP_FASTOPEN_KEY: { |
0f1ce023 JB |
3434 | __u8 key[TCP_FASTOPEN_KEY_BUF_LENGTH]; |
3435 | __u8 *backup_key = NULL; | |
1fba70e5 | 3436 | |
0f1ce023 JB |
3437 | /* Allow a backup key as well to facilitate key rotation |
3438 | * First key is the active one. | |
3439 | */ | |
3440 | if (optlen != TCP_FASTOPEN_KEY_LENGTH && | |
3441 | optlen != TCP_FASTOPEN_KEY_BUF_LENGTH) | |
1fba70e5 YC |
3442 | return -EINVAL; |
3443 | ||
d38d2b00 | 3444 | if (copy_from_sockptr(key, optval, optlen)) |
1fba70e5 YC |
3445 | return -EFAULT; |
3446 | ||
0f1ce023 JB |
3447 | if (optlen == TCP_FASTOPEN_KEY_BUF_LENGTH) |
3448 | backup_key = key + TCP_FASTOPEN_KEY_LENGTH; | |
3449 | ||
438ac880 | 3450 | return tcp_fastopen_reset_cipher(net, sk, key, backup_key); |
1fba70e5 | 3451 | } |
e56fb50f WAS |
3452 | default: |
3453 | /* fallthru */ | |
3454 | break; | |
ccbd6a5a | 3455 | } |
5f8ef48d | 3456 | |
1da177e4 LT |
3457 | if (optlen < sizeof(int)) |
3458 | return -EINVAL; | |
3459 | ||
d38d2b00 | 3460 | if (copy_from_sockptr(&val, optval, sizeof(val))) |
1da177e4 LT |
3461 | return -EFAULT; |
3462 | ||
d44fd4a7 ED |
3463 | /* Handle options that can be set without locking the socket. */ |
3464 | switch (optname) { | |
3465 | case TCP_SYNCNT: | |
3466 | return tcp_sock_set_syncnt(sk, val); | |
d58f2e15 ED |
3467 | case TCP_USER_TIMEOUT: |
3468 | return tcp_sock_set_user_timeout(sk, val); | |
6fd70a6b ED |
3469 | case TCP_KEEPINTVL: |
3470 | return tcp_sock_set_keepintvl(sk, val); | |
84485080 ED |
3471 | case TCP_KEEPCNT: |
3472 | return tcp_sock_set_keepcnt(sk, val); | |
a81722dd ED |
3473 | case TCP_LINGER2: |
3474 | if (val < 0) | |
3475 | WRITE_ONCE(tp->linger2, -1); | |
3476 | else if (val > TCP_FIN_TIMEOUT_MAX / HZ) | |
3477 | WRITE_ONCE(tp->linger2, TCP_FIN_TIMEOUT_MAX); | |
3478 | else | |
3479 | WRITE_ONCE(tp->linger2, val * HZ); | |
3480 | return 0; | |
6e97ba55 ED |
3481 | case TCP_DEFER_ACCEPT: |
3482 | /* Translate value in seconds to number of retransmits */ | |
3483 | WRITE_ONCE(icsk->icsk_accept_queue.rskq_defer_accept, | |
3484 | secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ, | |
3485 | TCP_RTO_MAX / HZ)); | |
3486 | return 0; | |
d44fd4a7 ED |
3487 | } |
3488 | ||
cb388e7e | 3489 | sockopt_lock_sock(sk); |
1da177e4 LT |
3490 | |
3491 | switch (optname) { | |
3492 | case TCP_MAXSEG: | |
3493 | /* Values greater than interface MTU won't take effect. However | |
3494 | * at the point when this call is done we typically don't yet | |
a777f715 RC |
3495 | * know which interface is going to be used |
3496 | */ | |
cfc62d87 | 3497 | if (val && (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW)) { |
1da177e4 LT |
3498 | err = -EINVAL; |
3499 | break; | |
3500 | } | |
3501 | tp->rx_opt.user_mss = val; | |
3502 | break; | |
3503 | ||
3504 | case TCP_NODELAY: | |
12abc5ee | 3505 | __tcp_sock_set_nodelay(sk, val); |
1da177e4 LT |
3506 | break; |
3507 | ||
36e31b0a AP |
3508 | case TCP_THIN_LINEAR_TIMEOUTS: |
3509 | if (val < 0 || val > 1) | |
3510 | err = -EINVAL; | |
3511 | else | |
3512 | tp->thin_lto = val; | |
3513 | break; | |
3514 | ||
7e380175 AP |
3515 | case TCP_THIN_DUPACK: |
3516 | if (val < 0 || val > 1) | |
3517 | err = -EINVAL; | |
7e380175 AP |
3518 | break; |
3519 | ||
ee995283 PE |
3520 | case TCP_REPAIR: |
3521 | if (!tcp_can_repair_sock(sk)) | |
3522 | err = -EPERM; | |
31048d7a | 3523 | else if (val == TCP_REPAIR_ON) { |
ee995283 PE |
3524 | tp->repair = 1; |
3525 | sk->sk_reuse = SK_FORCE_REUSE; | |
3526 | tp->repair_queue = TCP_NO_QUEUE; | |
31048d7a | 3527 | } else if (val == TCP_REPAIR_OFF) { |
ee995283 PE |
3528 | tp->repair = 0; |
3529 | sk->sk_reuse = SK_NO_REUSE; | |
3530 | tcp_send_window_probe(sk); | |
31048d7a | 3531 | } else if (val == TCP_REPAIR_OFF_NO_WP) { |
ee995283 PE |
3532 | tp->repair = 0; |
3533 | sk->sk_reuse = SK_NO_REUSE; | |
ee995283 PE |
3534 | } else |
3535 | err = -EINVAL; | |
3536 | ||
3537 | break; | |
3538 | ||
3539 | case TCP_REPAIR_QUEUE: | |
3540 | if (!tp->repair) | |
3541 | err = -EPERM; | |
bf2acc94 | 3542 | else if ((unsigned int)val < TCP_QUEUES_NR) |
ee995283 PE |
3543 | tp->repair_queue = val; |
3544 | else | |
3545 | err = -EINVAL; | |
3546 | break; | |
3547 | ||
3548 | case TCP_QUEUE_SEQ: | |
8811f4a9 | 3549 | if (sk->sk_state != TCP_CLOSE) { |
ee995283 | 3550 | err = -EPERM; |
8811f4a9 ED |
3551 | } else if (tp->repair_queue == TCP_SEND_QUEUE) { |
3552 | if (!tcp_rtx_queue_empty(sk)) | |
3553 | err = -EPERM; | |
3554 | else | |
3555 | WRITE_ONCE(tp->write_seq, val); | |
3556 | } else if (tp->repair_queue == TCP_RECV_QUEUE) { | |
3557 | if (tp->rcv_nxt != tp->copied_seq) { | |
3558 | err = -EPERM; | |
3559 | } else { | |
3560 | WRITE_ONCE(tp->rcv_nxt, val); | |
3561 | WRITE_ONCE(tp->copied_seq, val); | |
3562 | } | |
3563 | } else { | |
ee995283 | 3564 | err = -EINVAL; |
8811f4a9 | 3565 | } |
ee995283 PE |
3566 | break; |
3567 | ||
b139ba4e PE |
3568 | case TCP_REPAIR_OPTIONS: |
3569 | if (!tp->repair) | |
3570 | err = -EINVAL; | |
0c175da7 | 3571 | else if (sk->sk_state == TCP_ESTABLISHED && !tp->bytes_sent) |
d38d2b00 | 3572 | err = tcp_repair_options_est(sk, optval, optlen); |
b139ba4e PE |
3573 | else |
3574 | err = -EPERM; | |
3575 | break; | |
3576 | ||
1da177e4 | 3577 | case TCP_CORK: |
db10538a | 3578 | __tcp_sock_set_cork(sk, val); |
1da177e4 LT |
3579 | break; |
3580 | ||
3581 | case TCP_KEEPIDLE: | |
aad4a0a9 | 3582 | err = tcp_sock_set_keepidle_locked(sk, val); |
1da177e4 | 3583 | break; |
cd8ae852 | 3584 | case TCP_SAVE_SYN: |
267cf9fa MKL |
3585 | /* 0: disable, 1: enable, 2: start from ether_header */ |
3586 | if (val < 0 || val > 2) | |
cd8ae852 ED |
3587 | err = -EINVAL; |
3588 | else | |
3589 | tp->save_syn = val; | |
3590 | break; | |
3591 | ||
1da177e4 | 3592 | case TCP_WINDOW_CLAMP: |
cb811109 | 3593 | err = tcp_set_window_clamp(sk, val); |
1da177e4 LT |
3594 | break; |
3595 | ||
3596 | case TCP_QUICKACK: | |
ddd061b8 | 3597 | __tcp_sock_set_quickack(sk, val); |
1da177e4 LT |
3598 | break; |
3599 | ||
cfb6eeb4 YH |
3600 | #ifdef CONFIG_TCP_MD5SIG |
3601 | case TCP_MD5SIG: | |
8917a777 | 3602 | case TCP_MD5SIG_EXT: |
d38d2b00 | 3603 | err = tp->af_specific->md5_parse(sk, optname, optval, optlen); |
cfb6eeb4 YH |
3604 | break; |
3605 | #endif | |
8336886f JC |
3606 | case TCP_FASTOPEN: |
3607 | if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE | | |
dfea2aa6 | 3608 | TCPF_LISTEN))) { |
43713848 | 3609 | tcp_fastopen_init_key_once(net); |
dfea2aa6 | 3610 | |
0536fcc0 | 3611 | fastopen_queue_tune(sk, val); |
dfea2aa6 | 3612 | } else { |
8336886f | 3613 | err = -EINVAL; |
dfea2aa6 | 3614 | } |
8336886f | 3615 | break; |
19f6d3f3 WW |
3616 | case TCP_FASTOPEN_CONNECT: |
3617 | if (val > 1 || val < 0) { | |
3618 | err = -EINVAL; | |
5a542133 KI |
3619 | } else if (READ_ONCE(net->ipv4.sysctl_tcp_fastopen) & |
3620 | TFO_CLIENT_ENABLE) { | |
19f6d3f3 WW |
3621 | if (sk->sk_state == TCP_CLOSE) |
3622 | tp->fastopen_connect = val; | |
3623 | else | |
3624 | err = -EINVAL; | |
3625 | } else { | |
3626 | err = -EOPNOTSUPP; | |
3627 | } | |
3628 | break; | |
71c02379 CP |
3629 | case TCP_FASTOPEN_NO_COOKIE: |
3630 | if (val > 1 || val < 0) | |
3631 | err = -EINVAL; | |
3632 | else if (!((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) | |
3633 | err = -EINVAL; | |
3634 | else | |
3635 | tp->fastopen_no_cookie = val; | |
3636 | break; | |
93be6ce0 AV |
3637 | case TCP_TIMESTAMP: |
3638 | if (!tp->repair) | |
3639 | err = -EPERM; | |
3640 | else | |
dd23c9f1 | 3641 | WRITE_ONCE(tp->tsoffset, val - tcp_time_stamp_raw()); |
93be6ce0 | 3642 | break; |
b1ed4c4f AV |
3643 | case TCP_REPAIR_WINDOW: |
3644 | err = tcp_repair_set_window(tp, optval, optlen); | |
3645 | break; | |
c9bee3b7 | 3646 | case TCP_NOTSENT_LOWAT: |
1aeb87bc | 3647 | WRITE_ONCE(tp->notsent_lowat, val); |
c9bee3b7 ED |
3648 | sk->sk_write_space(sk); |
3649 | break; | |
b75eba76 SHY |
3650 | case TCP_INQ: |
3651 | if (val > 1 || val < 0) | |
3652 | err = -EINVAL; | |
3653 | else | |
3654 | tp->recvmsg_inq = val; | |
3655 | break; | |
a842fe14 ED |
3656 | case TCP_TX_DELAY: |
3657 | if (val) | |
3658 | tcp_enable_tx_delay(); | |
348b81b6 | 3659 | WRITE_ONCE(tp->tcp_tx_delay, val); |
a842fe14 | 3660 | break; |
1da177e4 LT |
3661 | default: |
3662 | err = -ENOPROTOOPT; | |
3663 | break; | |
3ff50b79 SH |
3664 | } |
3665 | ||
cb388e7e | 3666 | sockopt_release_sock(sk); |
1da177e4 LT |
3667 | return err; |
3668 | } | |
3669 | ||
a7b75c5a | 3670 | int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, |
b7058842 | 3671 | unsigned int optlen) |
3fdadf7d | 3672 | { |
cf533ea5 | 3673 | const struct inet_connection_sock *icsk = inet_csk(sk); |
3fdadf7d DM |
3674 | |
3675 | if (level != SOL_TCP) | |
f49cd2f4 KI |
3676 | /* Paired with WRITE_ONCE() in do_ipv6_setsockopt() and tcp_v6_connect() */ |
3677 | return READ_ONCE(icsk->icsk_af_ops)->setsockopt(sk, level, optname, | |
3678 | optval, optlen); | |
a7b75c5a | 3679 | return do_tcp_setsockopt(sk, level, optname, optval, optlen); |
3fdadf7d | 3680 | } |
4bc2f18b | 3681 | EXPORT_SYMBOL(tcp_setsockopt); |
3fdadf7d | 3682 | |
efd90174 FY |
3683 | static void tcp_get_info_chrono_stats(const struct tcp_sock *tp, |
3684 | struct tcp_info *info) | |
3685 | { | |
3686 | u64 stats[__TCP_CHRONO_MAX], total = 0; | |
3687 | enum tcp_chrono i; | |
3688 | ||
3689 | for (i = TCP_CHRONO_BUSY; i < __TCP_CHRONO_MAX; ++i) { | |
3690 | stats[i] = tp->chrono_stat[i - 1]; | |
3691 | if (i == tp->chrono_type) | |
628174cc | 3692 | stats[i] += tcp_jiffies32 - tp->chrono_start; |
efd90174 FY |
3693 | stats[i] *= USEC_PER_SEC / HZ; |
3694 | total += stats[i]; | |
3695 | } | |
3696 | ||
3697 | info->tcpi_busy_time = total; | |
3698 | info->tcpi_rwnd_limited = stats[TCP_CHRONO_RWND_LIMITED]; | |
3699 | info->tcpi_sndbuf_limited = stats[TCP_CHRONO_SNDBUF_LIMITED]; | |
3700 | } | |
3701 | ||
1da177e4 | 3702 | /* Return information about state of tcp endpoint in API format. */ |
0df48c26 | 3703 | void tcp_get_info(struct sock *sk, struct tcp_info *info) |
1da177e4 | 3704 | { |
35ac838a | 3705 | const struct tcp_sock *tp = tcp_sk(sk); /* iff sk_type == SOCK_STREAM */ |
463c84b9 | 3706 | const struct inet_connection_sock *icsk = inet_csk(sk); |
76a9ebe8 | 3707 | unsigned long rate; |
0263598c | 3708 | u32 now; |
ff5d7497 | 3709 | u64 rate64; |
67db3e4b | 3710 | bool slow; |
1da177e4 LT |
3711 | |
3712 | memset(info, 0, sizeof(*info)); | |
35ac838a CG |
3713 | if (sk->sk_type != SOCK_STREAM) |
3714 | return; | |
1da177e4 | 3715 | |
986ffdfd | 3716 | info->tcpi_state = inet_sk_state_load(sk); |
00fd38d9 | 3717 | |
ccbf3bfa ED |
3718 | /* Report meaningful fields for all TCP states, including listeners */ |
3719 | rate = READ_ONCE(sk->sk_pacing_rate); | |
76a9ebe8 | 3720 | rate64 = (rate != ~0UL) ? rate : ~0ULL; |
f522a5fc | 3721 | info->tcpi_pacing_rate = rate64; |
ccbf3bfa ED |
3722 | |
3723 | rate = READ_ONCE(sk->sk_max_pacing_rate); | |
76a9ebe8 | 3724 | rate64 = (rate != ~0UL) ? rate : ~0ULL; |
f522a5fc | 3725 | info->tcpi_max_pacing_rate = rate64; |
ccbf3bfa ED |
3726 | |
3727 | info->tcpi_reordering = tp->reordering; | |
40570375 | 3728 | info->tcpi_snd_cwnd = tcp_snd_cwnd(tp); |
ccbf3bfa ED |
3729 | |
3730 | if (info->tcpi_state == TCP_LISTEN) { | |
3731 | /* listeners aliased fields : | |
3732 | * tcpi_unacked -> Number of children ready for accept() | |
3733 | * tcpi_sacked -> max backlog | |
3734 | */ | |
288efe86 | 3735 | info->tcpi_unacked = READ_ONCE(sk->sk_ack_backlog); |
099ecf59 | 3736 | info->tcpi_sacked = READ_ONCE(sk->sk_max_ack_backlog); |
ccbf3bfa ED |
3737 | return; |
3738 | } | |
b369e7fd ED |
3739 | |
3740 | slow = lock_sock_fast(sk); | |
3741 | ||
6687e988 | 3742 | info->tcpi_ca_state = icsk->icsk_ca_state; |
463c84b9 | 3743 | info->tcpi_retransmits = icsk->icsk_retransmits; |
6687e988 | 3744 | info->tcpi_probes = icsk->icsk_probes_out; |
463c84b9 | 3745 | info->tcpi_backoff = icsk->icsk_backoff; |
1da177e4 LT |
3746 | |
3747 | if (tp->rx_opt.tstamp_ok) | |
3748 | info->tcpi_options |= TCPI_OPT_TIMESTAMPS; | |
e60402d0 | 3749 | if (tcp_is_sack(tp)) |
1da177e4 LT |
3750 | info->tcpi_options |= TCPI_OPT_SACK; |
3751 | if (tp->rx_opt.wscale_ok) { | |
3752 | info->tcpi_options |= TCPI_OPT_WSCALE; | |
3753 | info->tcpi_snd_wscale = tp->rx_opt.snd_wscale; | |
3754 | info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale; | |
e905a9ed | 3755 | } |
1da177e4 | 3756 | |
b5c5693b | 3757 | if (tp->ecn_flags & TCP_ECN_OK) |
1da177e4 | 3758 | info->tcpi_options |= TCPI_OPT_ECN; |
b5c5693b ED |
3759 | if (tp->ecn_flags & TCP_ECN_SEEN) |
3760 | info->tcpi_options |= TCPI_OPT_ECN_SEEN; | |
6f73601e YC |
3761 | if (tp->syn_data_acked) |
3762 | info->tcpi_options |= TCPI_OPT_SYN_DATA; | |
1da177e4 | 3763 | |
463c84b9 ACM |
3764 | info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto); |
3765 | info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato); | |
c1b4a7e6 | 3766 | info->tcpi_snd_mss = tp->mss_cache; |
463c84b9 | 3767 | info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss; |
1da177e4 | 3768 | |
ccbf3bfa ED |
3769 | info->tcpi_unacked = tp->packets_out; |
3770 | info->tcpi_sacked = tp->sacked_out; | |
3771 | ||
1da177e4 LT |
3772 | info->tcpi_lost = tp->lost_out; |
3773 | info->tcpi_retrans = tp->retrans_out; | |
1da177e4 | 3774 | |
d635fbe2 | 3775 | now = tcp_jiffies32; |
1da177e4 | 3776 | info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime); |
463c84b9 | 3777 | info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime); |
1da177e4 LT |
3778 | info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp); |
3779 | ||
d83d8461 | 3780 | info->tcpi_pmtu = icsk->icsk_pmtu_cookie; |
1da177e4 | 3781 | info->tcpi_rcv_ssthresh = tp->rcv_ssthresh; |
740b0f18 ED |
3782 | info->tcpi_rtt = tp->srtt_us >> 3; |
3783 | info->tcpi_rttvar = tp->mdev_us >> 2; | |
1da177e4 | 3784 | info->tcpi_snd_ssthresh = tp->snd_ssthresh; |
1da177e4 | 3785 | info->tcpi_advmss = tp->advmss; |
1da177e4 | 3786 | |
645f4c6f | 3787 | info->tcpi_rcv_rtt = tp->rcv_rtt_est.rtt_us >> 3; |
1da177e4 LT |
3788 | info->tcpi_rcv_space = tp->rcvq_space.space; |
3789 | ||
3790 | info->tcpi_total_retrans = tp->total_retrans; | |
977cb0ec | 3791 | |
f522a5fc ED |
3792 | info->tcpi_bytes_acked = tp->bytes_acked; |
3793 | info->tcpi_bytes_received = tp->bytes_received; | |
67db3e4b | 3794 | info->tcpi_notsent_bytes = max_t(int, 0, tp->write_seq - tp->snd_nxt); |
efd90174 | 3795 | tcp_get_info_chrono_stats(tp, info); |
67db3e4b | 3796 | |
2efd055c | 3797 | info->tcpi_segs_out = tp->segs_out; |
0307a0b7 ED |
3798 | |
3799 | /* segs_in and data_segs_in can be updated from tcp_segs_in() from BH */ | |
3800 | info->tcpi_segs_in = READ_ONCE(tp->segs_in); | |
3801 | info->tcpi_data_segs_in = READ_ONCE(tp->data_segs_in); | |
cd9b2660 | 3802 | |
cd9b2660 | 3803 | info->tcpi_min_rtt = tcp_min_rtt(tp); |
a44d6eac | 3804 | info->tcpi_data_segs_out = tp->data_segs_out; |
eb8329e0 YC |
3805 | |
3806 | info->tcpi_delivery_rate_app_limited = tp->rate_app_limited ? 1 : 0; | |
0263598c WW |
3807 | rate64 = tcp_compute_delivery_rate(tp); |
3808 | if (rate64) | |
f522a5fc | 3809 | info->tcpi_delivery_rate = rate64; |
feb5f2ec YC |
3810 | info->tcpi_delivered = tp->delivered; |
3811 | info->tcpi_delivered_ce = tp->delivered_ce; | |
ba113c3a | 3812 | info->tcpi_bytes_sent = tp->bytes_sent; |
fb31c9b9 | 3813 | info->tcpi_bytes_retrans = tp->bytes_retrans; |
7e10b655 | 3814 | info->tcpi_dsack_dups = tp->dsack_dups; |
7ec65372 | 3815 | info->tcpi_reord_seen = tp->reord_seen; |
f9af2dbb | 3816 | info->tcpi_rcv_ooopack = tp->rcv_ooopack; |
8f7baad7 | 3817 | info->tcpi_snd_wnd = tp->snd_wnd; |
71fc7047 MAQ |
3818 | info->tcpi_rcv_wnd = tp->rcv_wnd; |
3819 | info->tcpi_rehash = tp->plb_rehash + tp->timeout_rehash; | |
48027478 | 3820 | info->tcpi_fastopen_client_fail = tp->fastopen_client_fail; |
b369e7fd | 3821 | unlock_sock_fast(sk, slow); |
1da177e4 | 3822 | } |
1da177e4 LT |
3823 | EXPORT_SYMBOL_GPL(tcp_get_info); |
3824 | ||
984988aa WW |
3825 | static size_t tcp_opt_stats_get_size(void) |
3826 | { | |
3827 | return | |
3828 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_BUSY */ | |
3829 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_RWND_LIMITED */ | |
3830 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_SNDBUF_LIMITED */ | |
3831 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_DATA_SEGS_OUT */ | |
3832 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_TOTAL_RETRANS */ | |
3833 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_PACING_RATE */ | |
3834 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_DELIVERY_RATE */ | |
3835 | nla_total_size(sizeof(u32)) + /* TCP_NLA_SND_CWND */ | |
3836 | nla_total_size(sizeof(u32)) + /* TCP_NLA_REORDERING */ | |
3837 | nla_total_size(sizeof(u32)) + /* TCP_NLA_MIN_RTT */ | |
3838 | nla_total_size(sizeof(u8)) + /* TCP_NLA_RECUR_RETRANS */ | |
3839 | nla_total_size(sizeof(u8)) + /* TCP_NLA_DELIVERY_RATE_APP_LMT */ | |
3840 | nla_total_size(sizeof(u32)) + /* TCP_NLA_SNDQ_SIZE */ | |
3841 | nla_total_size(sizeof(u8)) + /* TCP_NLA_CA_STATE */ | |
3842 | nla_total_size(sizeof(u32)) + /* TCP_NLA_SND_SSTHRESH */ | |
3843 | nla_total_size(sizeof(u32)) + /* TCP_NLA_DELIVERED */ | |
3844 | nla_total_size(sizeof(u32)) + /* TCP_NLA_DELIVERED_CE */ | |
ba113c3a | 3845 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_BYTES_SENT */ |
fb31c9b9 | 3846 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_BYTES_RETRANS */ |
7e10b655 | 3847 | nla_total_size(sizeof(u32)) + /* TCP_NLA_DSACK_DUPS */ |
7ec65372 | 3848 | nla_total_size(sizeof(u32)) + /* TCP_NLA_REORD_SEEN */ |
e8bd8fca | 3849 | nla_total_size(sizeof(u32)) + /* TCP_NLA_SRTT */ |
32efcc06 | 3850 | nla_total_size(sizeof(u16)) + /* TCP_NLA_TIMEOUT_REHASH */ |
e08ab0b3 | 3851 | nla_total_size(sizeof(u32)) + /* TCP_NLA_BYTES_NOTSENT */ |
48040793 | 3852 | nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_EDT */ |
e7ed11ee | 3853 | nla_total_size(sizeof(u8)) + /* TCP_NLA_TTL */ |
29c1c446 | 3854 | nla_total_size(sizeof(u32)) + /* TCP_NLA_REHASH */ |
984988aa WW |
3855 | 0; |
3856 | } | |
3857 | ||
e7ed11ee YS |
3858 | /* Returns TTL or hop limit of an incoming packet from skb. */ |
3859 | static u8 tcp_skb_ttl_or_hop_limit(const struct sk_buff *skb) | |
3860 | { | |
3861 | if (skb->protocol == htons(ETH_P_IP)) | |
3862 | return ip_hdr(skb)->ttl; | |
3863 | else if (skb->protocol == htons(ETH_P_IPV6)) | |
3864 | return ipv6_hdr(skb)->hop_limit; | |
3865 | else | |
3866 | return 0; | |
3867 | } | |
3868 | ||
48040793 | 3869 | struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk, |
e7ed11ee YS |
3870 | const struct sk_buff *orig_skb, |
3871 | const struct sk_buff *ack_skb) | |
1c885808 FY |
3872 | { |
3873 | const struct tcp_sock *tp = tcp_sk(sk); | |
3874 | struct sk_buff *stats; | |
3875 | struct tcp_info info; | |
76a9ebe8 | 3876 | unsigned long rate; |
bb7c19f9 | 3877 | u64 rate64; |
1c885808 | 3878 | |
984988aa | 3879 | stats = alloc_skb(tcp_opt_stats_get_size(), GFP_ATOMIC); |
1c885808 FY |
3880 | if (!stats) |
3881 | return NULL; | |
3882 | ||
3883 | tcp_get_info_chrono_stats(tp, &info); | |
3884 | nla_put_u64_64bit(stats, TCP_NLA_BUSY, | |
3885 | info.tcpi_busy_time, TCP_NLA_PAD); | |
3886 | nla_put_u64_64bit(stats, TCP_NLA_RWND_LIMITED, | |
3887 | info.tcpi_rwnd_limited, TCP_NLA_PAD); | |
3888 | nla_put_u64_64bit(stats, TCP_NLA_SNDBUF_LIMITED, | |
3889 | info.tcpi_sndbuf_limited, TCP_NLA_PAD); | |
7e98102f YC |
3890 | nla_put_u64_64bit(stats, TCP_NLA_DATA_SEGS_OUT, |
3891 | tp->data_segs_out, TCP_NLA_PAD); | |
3892 | nla_put_u64_64bit(stats, TCP_NLA_TOTAL_RETRANS, | |
3893 | tp->total_retrans, TCP_NLA_PAD); | |
bb7c19f9 WW |
3894 | |
3895 | rate = READ_ONCE(sk->sk_pacing_rate); | |
76a9ebe8 | 3896 | rate64 = (rate != ~0UL) ? rate : ~0ULL; |
bb7c19f9 WW |
3897 | nla_put_u64_64bit(stats, TCP_NLA_PACING_RATE, rate64, TCP_NLA_PAD); |
3898 | ||
3899 | rate64 = tcp_compute_delivery_rate(tp); | |
3900 | nla_put_u64_64bit(stats, TCP_NLA_DELIVERY_RATE, rate64, TCP_NLA_PAD); | |
3901 | ||
40570375 | 3902 | nla_put_u32(stats, TCP_NLA_SND_CWND, tcp_snd_cwnd(tp)); |
bb7c19f9 WW |
3903 | nla_put_u32(stats, TCP_NLA_REORDERING, tp->reordering); |
3904 | nla_put_u32(stats, TCP_NLA_MIN_RTT, tcp_min_rtt(tp)); | |
3905 | ||
3906 | nla_put_u8(stats, TCP_NLA_RECUR_RETRANS, inet_csk(sk)->icsk_retransmits); | |
3907 | nla_put_u8(stats, TCP_NLA_DELIVERY_RATE_APP_LMT, !!tp->rate_app_limited); | |
7156d194 | 3908 | nla_put_u32(stats, TCP_NLA_SND_SSTHRESH, tp->snd_ssthresh); |
feb5f2ec YC |
3909 | nla_put_u32(stats, TCP_NLA_DELIVERED, tp->delivered); |
3910 | nla_put_u32(stats, TCP_NLA_DELIVERED_CE, tp->delivered_ce); | |
87ecc95d PJ |
3911 | |
3912 | nla_put_u32(stats, TCP_NLA_SNDQ_SIZE, tp->write_seq - tp->snd_una); | |
be631892 | 3913 | nla_put_u8(stats, TCP_NLA_CA_STATE, inet_csk(sk)->icsk_ca_state); |
feb5f2ec | 3914 | |
ba113c3a WW |
3915 | nla_put_u64_64bit(stats, TCP_NLA_BYTES_SENT, tp->bytes_sent, |
3916 | TCP_NLA_PAD); | |
fb31c9b9 WW |
3917 | nla_put_u64_64bit(stats, TCP_NLA_BYTES_RETRANS, tp->bytes_retrans, |
3918 | TCP_NLA_PAD); | |
7e10b655 | 3919 | nla_put_u32(stats, TCP_NLA_DSACK_DUPS, tp->dsack_dups); |
7ec65372 | 3920 | nla_put_u32(stats, TCP_NLA_REORD_SEEN, tp->reord_seen); |
e8bd8fca | 3921 | nla_put_u32(stats, TCP_NLA_SRTT, tp->srtt_us >> 3); |
32efcc06 | 3922 | nla_put_u16(stats, TCP_NLA_TIMEOUT_REHASH, tp->timeout_rehash); |
e08ab0b3 YS |
3923 | nla_put_u32(stats, TCP_NLA_BYTES_NOTSENT, |
3924 | max_t(int, 0, tp->write_seq - tp->snd_nxt)); | |
48040793 YS |
3925 | nla_put_u64_64bit(stats, TCP_NLA_EDT, orig_skb->skb_mstamp_ns, |
3926 | TCP_NLA_PAD); | |
e7ed11ee YS |
3927 | if (ack_skb) |
3928 | nla_put_u8(stats, TCP_NLA_TTL, | |
3929 | tcp_skb_ttl_or_hop_limit(ack_skb)); | |
ba113c3a | 3930 | |
29c1c446 | 3931 | nla_put_u32(stats, TCP_NLA_REHASH, tp->plb_rehash + tp->timeout_rehash); |
1c885808 FY |
3932 | return stats; |
3933 | } | |
3934 | ||
273b7f0f MKL |
3935 | int do_tcp_getsockopt(struct sock *sk, int level, |
3936 | int optname, sockptr_t optval, sockptr_t optlen) | |
1da177e4 | 3937 | { |
295f7324 | 3938 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 3939 | struct tcp_sock *tp = tcp_sk(sk); |
6fa25166 | 3940 | struct net *net = sock_net(sk); |
1da177e4 LT |
3941 | int val, len; |
3942 | ||
34704ef0 | 3943 | if (copy_from_sockptr(&len, optlen, sizeof(int))) |
1da177e4 LT |
3944 | return -EFAULT; |
3945 | ||
3946 | len = min_t(unsigned int, len, sizeof(int)); | |
3947 | ||
3948 | if (len < 0) | |
3949 | return -EINVAL; | |
3950 | ||
3951 | switch (optname) { | |
3952 | case TCP_MAXSEG: | |
c1b4a7e6 | 3953 | val = tp->mss_cache; |
34dfde4a CZ |
3954 | if (tp->rx_opt.user_mss && |
3955 | ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) | |
1da177e4 | 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: | |
3a037f0f | 3976 | val = READ_ONCE(icsk->icsk_syn_retries) ? : |
20a3b1c0 | 3977 | READ_ONCE(net->ipv4.sysctl_tcp_syn_retries); |
1da177e4 LT |
3978 | break; |
3979 | case TCP_LINGER2: | |
9df5335c | 3980 | val = READ_ONCE(tp->linger2); |
1da177e4 | 3981 | if (val >= 0) |
39e24435 | 3982 | val = (val ? : READ_ONCE(net->ipv4.sysctl_tcp_fin_timeout)) / HZ; |
1da177e4 LT |
3983 | break; |
3984 | case TCP_DEFER_ACCEPT: | |
ae488c74 ED |
3985 | val = READ_ONCE(icsk->icsk_accept_queue.rskq_defer_accept); |
3986 | val = retrans_to_secs(val, TCP_TIMEOUT_INIT / HZ, | |
3987 | TCP_RTO_MAX / HZ); | |
1da177e4 LT |
3988 | break; |
3989 | case TCP_WINDOW_CLAMP: | |
3990 | val = tp->window_clamp; | |
3991 | break; | |
3992 | case TCP_INFO: { | |
3993 | struct tcp_info info; | |
3994 | ||
34704ef0 | 3995 | if (copy_from_sockptr(&len, optlen, sizeof(int))) |
1da177e4 LT |
3996 | return -EFAULT; |
3997 | ||
3998 | tcp_get_info(sk, &info); | |
3999 | ||
4000 | len = min_t(unsigned int, len, sizeof(info)); | |
34704ef0 | 4001 | if (copy_to_sockptr(optlen, &len, sizeof(int))) |
1da177e4 | 4002 | return -EFAULT; |
34704ef0 | 4003 | if (copy_to_sockptr(optval, &info, len)) |
1da177e4 LT |
4004 | return -EFAULT; |
4005 | return 0; | |
4006 | } | |
6e9250f5 ED |
4007 | case TCP_CC_INFO: { |
4008 | const struct tcp_congestion_ops *ca_ops; | |
4009 | union tcp_cc_info info; | |
4010 | size_t sz = 0; | |
4011 | int attr; | |
4012 | ||
34704ef0 | 4013 | if (copy_from_sockptr(&len, optlen, sizeof(int))) |
6e9250f5 ED |
4014 | return -EFAULT; |
4015 | ||
4016 | ca_ops = icsk->icsk_ca_ops; | |
4017 | if (ca_ops && ca_ops->get_info) | |
4018 | sz = ca_ops->get_info(sk, ~0U, &attr, &info); | |
4019 | ||
4020 | len = min_t(unsigned int, len, sz); | |
34704ef0 | 4021 | if (copy_to_sockptr(optlen, &len, sizeof(int))) |
6e9250f5 | 4022 | return -EFAULT; |
34704ef0 | 4023 | if (copy_to_sockptr(optval, &info, len)) |
6e9250f5 ED |
4024 | return -EFAULT; |
4025 | return 0; | |
4026 | } | |
1da177e4 | 4027 | case TCP_QUICKACK: |
31954cd8 | 4028 | val = !inet_csk_in_pingpong_mode(sk); |
1da177e4 | 4029 | break; |
5f8ef48d SH |
4030 | |
4031 | case TCP_CONGESTION: | |
34704ef0 | 4032 | if (copy_from_sockptr(&len, optlen, sizeof(int))) |
5f8ef48d SH |
4033 | return -EFAULT; |
4034 | len = min_t(unsigned int, len, TCP_CA_NAME_MAX); | |
34704ef0 | 4035 | if (copy_to_sockptr(optlen, &len, sizeof(int))) |
5f8ef48d | 4036 | return -EFAULT; |
34704ef0 | 4037 | if (copy_to_sockptr(optval, icsk->icsk_ca_ops->name, len)) |
5f8ef48d SH |
4038 | return -EFAULT; |
4039 | return 0; | |
e56fb50f | 4040 | |
734942cc | 4041 | case TCP_ULP: |
34704ef0 | 4042 | if (copy_from_sockptr(&len, optlen, sizeof(int))) |
734942cc DW |
4043 | return -EFAULT; |
4044 | len = min_t(unsigned int, len, TCP_ULP_NAME_MAX); | |
d97af30f | 4045 | if (!icsk->icsk_ulp_ops) { |
34704ef0 MKL |
4046 | len = 0; |
4047 | if (copy_to_sockptr(optlen, &len, sizeof(int))) | |
d97af30f DW |
4048 | return -EFAULT; |
4049 | return 0; | |
4050 | } | |
34704ef0 | 4051 | if (copy_to_sockptr(optlen, &len, sizeof(int))) |
734942cc | 4052 | return -EFAULT; |
34704ef0 | 4053 | if (copy_to_sockptr(optval, icsk->icsk_ulp_ops->name, len)) |
734942cc DW |
4054 | return -EFAULT; |
4055 | return 0; | |
4056 | ||
1fba70e5 | 4057 | case TCP_FASTOPEN_KEY: { |
f19008e6 JB |
4058 | u64 key[TCP_FASTOPEN_KEY_BUF_LENGTH / sizeof(u64)]; |
4059 | unsigned int key_len; | |
1fba70e5 | 4060 | |
34704ef0 | 4061 | if (copy_from_sockptr(&len, optlen, sizeof(int))) |
1fba70e5 YC |
4062 | return -EFAULT; |
4063 | ||
f19008e6 JB |
4064 | key_len = tcp_fastopen_get_cipher(net, icsk, key) * |
4065 | TCP_FASTOPEN_KEY_LENGTH; | |
0f1ce023 | 4066 | len = min_t(unsigned int, len, key_len); |
34704ef0 | 4067 | if (copy_to_sockptr(optlen, &len, sizeof(int))) |
1fba70e5 | 4068 | return -EFAULT; |
34704ef0 | 4069 | if (copy_to_sockptr(optval, key, len)) |
1fba70e5 YC |
4070 | return -EFAULT; |
4071 | return 0; | |
4072 | } | |
3c0fef0b JH |
4073 | case TCP_THIN_LINEAR_TIMEOUTS: |
4074 | val = tp->thin_lto; | |
4075 | break; | |
4a7f6009 | 4076 | |
3c0fef0b | 4077 | case TCP_THIN_DUPACK: |
4a7f6009 | 4078 | val = 0; |
3c0fef0b | 4079 | break; |
dca43c75 | 4080 | |
ee995283 PE |
4081 | case TCP_REPAIR: |
4082 | val = tp->repair; | |
4083 | break; | |
4084 | ||
4085 | case TCP_REPAIR_QUEUE: | |
4086 | if (tp->repair) | |
4087 | val = tp->repair_queue; | |
4088 | else | |
4089 | return -EINVAL; | |
4090 | break; | |
4091 | ||
b1ed4c4f AV |
4092 | case TCP_REPAIR_WINDOW: { |
4093 | struct tcp_repair_window opt; | |
4094 | ||
34704ef0 | 4095 | if (copy_from_sockptr(&len, optlen, sizeof(int))) |
b1ed4c4f AV |
4096 | return -EFAULT; |
4097 | ||
4098 | if (len != sizeof(opt)) | |
4099 | return -EINVAL; | |
4100 | ||
4101 | if (!tp->repair) | |
4102 | return -EPERM; | |
4103 | ||
4104 | opt.snd_wl1 = tp->snd_wl1; | |
4105 | opt.snd_wnd = tp->snd_wnd; | |
4106 | opt.max_window = tp->max_window; | |
4107 | opt.rcv_wnd = tp->rcv_wnd; | |
4108 | opt.rcv_wup = tp->rcv_wup; | |
4109 | ||
34704ef0 | 4110 | if (copy_to_sockptr(optval, &opt, len)) |
b1ed4c4f AV |
4111 | return -EFAULT; |
4112 | return 0; | |
4113 | } | |
ee995283 PE |
4114 | case TCP_QUEUE_SEQ: |
4115 | if (tp->repair_queue == TCP_SEND_QUEUE) | |
4116 | val = tp->write_seq; | |
4117 | else if (tp->repair_queue == TCP_RECV_QUEUE) | |
4118 | val = tp->rcv_nxt; | |
4119 | else | |
4120 | return -EINVAL; | |
4121 | break; | |
4122 | ||
dca43c75 | 4123 | case TCP_USER_TIMEOUT: |
26023e91 | 4124 | val = READ_ONCE(icsk->icsk_user_timeout); |
dca43c75 | 4125 | break; |
1536e285 KN |
4126 | |
4127 | case TCP_FASTOPEN: | |
70f360dd | 4128 | val = READ_ONCE(icsk->icsk_accept_queue.fastopenq.max_qlen); |
1536e285 KN |
4129 | break; |
4130 | ||
19f6d3f3 WW |
4131 | case TCP_FASTOPEN_CONNECT: |
4132 | val = tp->fastopen_connect; | |
4133 | break; | |
4134 | ||
71c02379 CP |
4135 | case TCP_FASTOPEN_NO_COOKIE: |
4136 | val = tp->fastopen_no_cookie; | |
4137 | break; | |
4138 | ||
a842fe14 | 4139 | case TCP_TX_DELAY: |
348b81b6 | 4140 | val = READ_ONCE(tp->tcp_tx_delay); |
a842fe14 ED |
4141 | break; |
4142 | ||
93be6ce0 | 4143 | case TCP_TIMESTAMP: |
dd23c9f1 | 4144 | val = tcp_time_stamp_raw() + READ_ONCE(tp->tsoffset); |
93be6ce0 | 4145 | break; |
c9bee3b7 | 4146 | case TCP_NOTSENT_LOWAT: |
1aeb87bc | 4147 | val = READ_ONCE(tp->notsent_lowat); |
c9bee3b7 | 4148 | break; |
b75eba76 SHY |
4149 | case TCP_INQ: |
4150 | val = tp->recvmsg_inq; | |
4151 | break; | |
cd8ae852 ED |
4152 | case TCP_SAVE_SYN: |
4153 | val = tp->save_syn; | |
4154 | break; | |
4155 | case TCP_SAVED_SYN: { | |
34704ef0 | 4156 | if (copy_from_sockptr(&len, optlen, sizeof(int))) |
cd8ae852 ED |
4157 | return -EFAULT; |
4158 | ||
d51bbff2 | 4159 | sockopt_lock_sock(sk); |
cd8ae852 | 4160 | if (tp->saved_syn) { |
70a217f1 | 4161 | if (len < tcp_saved_syn_len(tp->saved_syn)) { |
34704ef0 MKL |
4162 | len = tcp_saved_syn_len(tp->saved_syn); |
4163 | if (copy_to_sockptr(optlen, &len, sizeof(int))) { | |
d51bbff2 | 4164 | sockopt_release_sock(sk); |
aea0929e EM |
4165 | return -EFAULT; |
4166 | } | |
d51bbff2 | 4167 | sockopt_release_sock(sk); |
aea0929e EM |
4168 | return -EINVAL; |
4169 | } | |
70a217f1 | 4170 | len = tcp_saved_syn_len(tp->saved_syn); |
34704ef0 | 4171 | if (copy_to_sockptr(optlen, &len, sizeof(int))) { |
d51bbff2 | 4172 | sockopt_release_sock(sk); |
cd8ae852 ED |
4173 | return -EFAULT; |
4174 | } | |
34704ef0 | 4175 | if (copy_to_sockptr(optval, tp->saved_syn->data, len)) { |
d51bbff2 | 4176 | sockopt_release_sock(sk); |
cd8ae852 ED |
4177 | return -EFAULT; |
4178 | } | |
4179 | tcp_saved_syn_free(tp); | |
d51bbff2 | 4180 | sockopt_release_sock(sk); |
cd8ae852 | 4181 | } else { |
d51bbff2 | 4182 | sockopt_release_sock(sk); |
cd8ae852 | 4183 | len = 0; |
34704ef0 | 4184 | if (copy_to_sockptr(optlen, &len, sizeof(int))) |
cd8ae852 ED |
4185 | return -EFAULT; |
4186 | } | |
4187 | return 0; | |
4188 | } | |
05255b82 ED |
4189 | #ifdef CONFIG_MMU |
4190 | case TCP_ZEROCOPY_RECEIVE: { | |
7eeba170 | 4191 | struct scm_timestamping_internal tss; |
e0fecb28 | 4192 | struct tcp_zerocopy_receive zc = {}; |
05255b82 ED |
4193 | int err; |
4194 | ||
34704ef0 | 4195 | if (copy_from_sockptr(&len, optlen, sizeof(int))) |
05255b82 | 4196 | return -EFAULT; |
2107d45f AR |
4197 | if (len < 0 || |
4198 | len < offsetofend(struct tcp_zerocopy_receive, length)) | |
05255b82 | 4199 | return -EINVAL; |
3c5a2fd0 | 4200 | if (unlikely(len > sizeof(zc))) { |
34704ef0 MKL |
4201 | err = check_zeroed_sockptr(optval, sizeof(zc), |
4202 | len - sizeof(zc)); | |
3c5a2fd0 AR |
4203 | if (err < 1) |
4204 | return err == 0 ? -EINVAL : err; | |
c8856c05 | 4205 | len = sizeof(zc); |
34704ef0 | 4206 | if (copy_to_sockptr(optlen, &len, sizeof(int))) |
0b7f41f6 AR |
4207 | return -EFAULT; |
4208 | } | |
34704ef0 | 4209 | if (copy_from_sockptr(&zc, optval, len)) |
05255b82 | 4210 | return -EFAULT; |
3c5a2fd0 AR |
4211 | if (zc.reserved) |
4212 | return -EINVAL; | |
4213 | if (zc.msg_flags & ~(TCP_VALID_ZC_MSG_FLAGS)) | |
4214 | return -EINVAL; | |
d51bbff2 | 4215 | sockopt_lock_sock(sk); |
7eeba170 | 4216 | err = tcp_zerocopy_receive(sk, &zc, &tss); |
9cacf81f SF |
4217 | err = BPF_CGROUP_RUN_PROG_GETSOCKOPT_KERN(sk, level, optname, |
4218 | &zc, &len, err); | |
d51bbff2 | 4219 | sockopt_release_sock(sk); |
7eeba170 AR |
4220 | if (len >= offsetofend(struct tcp_zerocopy_receive, msg_flags)) |
4221 | goto zerocopy_rcv_cmsg; | |
c8856c05 | 4222 | switch (len) { |
7eeba170 AR |
4223 | case offsetofend(struct tcp_zerocopy_receive, msg_flags): |
4224 | goto zerocopy_rcv_cmsg; | |
4225 | case offsetofend(struct tcp_zerocopy_receive, msg_controllen): | |
4226 | case offsetofend(struct tcp_zerocopy_receive, msg_control): | |
4227 | case offsetofend(struct tcp_zerocopy_receive, flags): | |
4228 | case offsetofend(struct tcp_zerocopy_receive, copybuf_len): | |
4229 | case offsetofend(struct tcp_zerocopy_receive, copybuf_address): | |
33946518 AR |
4230 | case offsetofend(struct tcp_zerocopy_receive, err): |
4231 | goto zerocopy_rcv_sk_err; | |
c8856c05 AR |
4232 | case offsetofend(struct tcp_zerocopy_receive, inq): |
4233 | goto zerocopy_rcv_inq; | |
4234 | case offsetofend(struct tcp_zerocopy_receive, length): | |
4235 | default: | |
4236 | goto zerocopy_rcv_out; | |
4237 | } | |
7eeba170 AR |
4238 | zerocopy_rcv_cmsg: |
4239 | if (zc.msg_flags & TCP_CMSG_TS) | |
4240 | tcp_zc_finalize_rx_tstamp(sk, &zc, &tss); | |
4241 | else | |
4242 | zc.msg_flags = 0; | |
33946518 AR |
4243 | zerocopy_rcv_sk_err: |
4244 | if (!err) | |
4245 | zc.err = sock_error(sk); | |
c8856c05 AR |
4246 | zerocopy_rcv_inq: |
4247 | zc.inq = tcp_inq_hint(sk); | |
4248 | zerocopy_rcv_out: | |
34704ef0 | 4249 | if (!err && copy_to_sockptr(optval, &zc, len)) |
05255b82 ED |
4250 | err = -EFAULT; |
4251 | return err; | |
4252 | } | |
4253 | #endif | |
1da177e4 LT |
4254 | default: |
4255 | return -ENOPROTOOPT; | |
3ff50b79 | 4256 | } |
1da177e4 | 4257 | |
34704ef0 | 4258 | if (copy_to_sockptr(optlen, &len, sizeof(int))) |
1da177e4 | 4259 | return -EFAULT; |
34704ef0 | 4260 | if (copy_to_sockptr(optval, &val, len)) |
1da177e4 LT |
4261 | return -EFAULT; |
4262 | return 0; | |
4263 | } | |
4264 | ||
9cacf81f SF |
4265 | bool tcp_bpf_bypass_getsockopt(int level, int optname) |
4266 | { | |
4267 | /* TCP do_tcp_getsockopt has optimized getsockopt implementation | |
4268 | * to avoid extra socket lock for TCP_ZEROCOPY_RECEIVE. | |
4269 | */ | |
4270 | if (level == SOL_TCP && optname == TCP_ZEROCOPY_RECEIVE) | |
4271 | return true; | |
4272 | ||
4273 | return false; | |
4274 | } | |
4275 | EXPORT_SYMBOL(tcp_bpf_bypass_getsockopt); | |
4276 | ||
3fdadf7d DM |
4277 | int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval, |
4278 | int __user *optlen) | |
4279 | { | |
4280 | struct inet_connection_sock *icsk = inet_csk(sk); | |
4281 | ||
4282 | if (level != SOL_TCP) | |
f49cd2f4 KI |
4283 | /* Paired with WRITE_ONCE() in do_ipv6_setsockopt() and tcp_v6_connect() */ |
4284 | return READ_ONCE(icsk->icsk_af_ops)->getsockopt(sk, level, optname, | |
4285 | optval, optlen); | |
34704ef0 MKL |
4286 | return do_tcp_getsockopt(sk, level, optname, USER_SOCKPTR(optval), |
4287 | USER_SOCKPTR(optlen)); | |
3fdadf7d | 4288 | } |
4bc2f18b | 4289 | EXPORT_SYMBOL(tcp_getsockopt); |
3fdadf7d | 4290 | |
cfb6eeb4 | 4291 | #ifdef CONFIG_TCP_MD5SIG |
349ce993 | 4292 | static DEFINE_PER_CPU(struct tcp_md5sig_pool, tcp_md5sig_pool); |
71cea17e | 4293 | static DEFINE_MUTEX(tcp_md5sig_mutex); |
349ce993 | 4294 | static bool tcp_md5sig_pool_populated = false; |
cfb6eeb4 | 4295 | |
71cea17e | 4296 | static void __tcp_alloc_md5sig_pool(void) |
cfb6eeb4 | 4297 | { |
cf80e0e4 | 4298 | struct crypto_ahash *hash; |
cfb6eeb4 | 4299 | int cpu; |
cfb6eeb4 | 4300 | |
cf80e0e4 | 4301 | hash = crypto_alloc_ahash("md5", 0, CRYPTO_ALG_ASYNC); |
70477371 | 4302 | if (IS_ERR(hash)) |
cf80e0e4 HX |
4303 | return; |
4304 | ||
cfb6eeb4 | 4305 | for_each_possible_cpu(cpu) { |
19689e38 | 4306 | void *scratch = per_cpu(tcp_md5sig_pool, cpu).scratch; |
cf80e0e4 | 4307 | struct ahash_request *req; |
cfb6eeb4 | 4308 | |
19689e38 ED |
4309 | if (!scratch) { |
4310 | scratch = kmalloc_node(sizeof(union tcp_md5sum_block) + | |
4311 | sizeof(struct tcphdr), | |
4312 | GFP_KERNEL, | |
4313 | cpu_to_node(cpu)); | |
4314 | if (!scratch) | |
4315 | return; | |
4316 | per_cpu(tcp_md5sig_pool, cpu).scratch = scratch; | |
4317 | } | |
cf80e0e4 HX |
4318 | if (per_cpu(tcp_md5sig_pool, cpu).md5_req) |
4319 | continue; | |
4320 | ||
4321 | req = ahash_request_alloc(hash, GFP_KERNEL); | |
4322 | if (!req) | |
4323 | return; | |
4324 | ||
4325 | ahash_request_set_callback(req, 0, NULL, NULL); | |
4326 | ||
4327 | per_cpu(tcp_md5sig_pool, cpu).md5_req = req; | |
cfb6eeb4 | 4328 | } |
349ce993 ED |
4329 | /* before setting tcp_md5sig_pool_populated, we must commit all writes |
4330 | * to memory. See smp_rmb() in tcp_get_md5sig_pool() | |
71cea17e ED |
4331 | */ |
4332 | smp_wmb(); | |
aacd467c ED |
4333 | /* Paired with READ_ONCE() from tcp_alloc_md5sig_pool() |
4334 | * and tcp_get_md5sig_pool(). | |
4335 | */ | |
4336 | WRITE_ONCE(tcp_md5sig_pool_populated, true); | |
cfb6eeb4 YH |
4337 | } |
4338 | ||
71cea17e | 4339 | bool tcp_alloc_md5sig_pool(void) |
cfb6eeb4 | 4340 | { |
aacd467c ED |
4341 | /* Paired with WRITE_ONCE() from __tcp_alloc_md5sig_pool() */ |
4342 | if (unlikely(!READ_ONCE(tcp_md5sig_pool_populated))) { | |
71cea17e ED |
4343 | mutex_lock(&tcp_md5sig_mutex); |
4344 | ||
459837b5 | 4345 | if (!tcp_md5sig_pool_populated) |
71cea17e ED |
4346 | __tcp_alloc_md5sig_pool(); |
4347 | ||
4348 | mutex_unlock(&tcp_md5sig_mutex); | |
cfb6eeb4 | 4349 | } |
aacd467c ED |
4350 | /* Paired with WRITE_ONCE() from __tcp_alloc_md5sig_pool() */ |
4351 | return READ_ONCE(tcp_md5sig_pool_populated); | |
cfb6eeb4 | 4352 | } |
cfb6eeb4 YH |
4353 | EXPORT_SYMBOL(tcp_alloc_md5sig_pool); |
4354 | ||
35790c04 ED |
4355 | |
4356 | /** | |
4357 | * tcp_get_md5sig_pool - get md5sig_pool for this user | |
4358 | * | |
4359 | * We use percpu structure, so if we succeed, we exit with preemption | |
4360 | * and BH disabled, to make sure another thread or softirq handling | |
4361 | * wont try to get same context. | |
4362 | */ | |
4363 | struct tcp_md5sig_pool *tcp_get_md5sig_pool(void) | |
cfb6eeb4 | 4364 | { |
35790c04 | 4365 | local_bh_disable(); |
cfb6eeb4 | 4366 | |
aacd467c ED |
4367 | /* Paired with WRITE_ONCE() from __tcp_alloc_md5sig_pool() */ |
4368 | if (READ_ONCE(tcp_md5sig_pool_populated)) { | |
349ce993 ED |
4369 | /* coupled with smp_wmb() in __tcp_alloc_md5sig_pool() */ |
4370 | smp_rmb(); | |
4371 | return this_cpu_ptr(&tcp_md5sig_pool); | |
4372 | } | |
35790c04 ED |
4373 | local_bh_enable(); |
4374 | return NULL; | |
4375 | } | |
4376 | EXPORT_SYMBOL(tcp_get_md5sig_pool); | |
cfb6eeb4 | 4377 | |
49a72dfb | 4378 | int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp, |
cf533ea5 | 4379 | const struct sk_buff *skb, unsigned int header_len) |
49a72dfb AL |
4380 | { |
4381 | struct scatterlist sg; | |
4382 | const struct tcphdr *tp = tcp_hdr(skb); | |
cf80e0e4 | 4383 | struct ahash_request *req = hp->md5_req; |
95c96174 ED |
4384 | unsigned int i; |
4385 | const unsigned int head_data_len = skb_headlen(skb) > header_len ? | |
4386 | skb_headlen(skb) - header_len : 0; | |
49a72dfb | 4387 | const struct skb_shared_info *shi = skb_shinfo(skb); |
d7fd1b57 | 4388 | struct sk_buff *frag_iter; |
49a72dfb AL |
4389 | |
4390 | sg_init_table(&sg, 1); | |
4391 | ||
4392 | sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len); | |
cf80e0e4 HX |
4393 | ahash_request_set_crypt(req, &sg, NULL, head_data_len); |
4394 | if (crypto_ahash_update(req)) | |
49a72dfb AL |
4395 | return 1; |
4396 | ||
4397 | for (i = 0; i < shi->nr_frags; ++i) { | |
d8e18a51 | 4398 | const skb_frag_t *f = &shi->frags[i]; |
b54c9d5b | 4399 | unsigned int offset = skb_frag_off(f); |
54d27fcb ED |
4400 | struct page *page = skb_frag_page(f) + (offset >> PAGE_SHIFT); |
4401 | ||
4402 | sg_set_page(&sg, page, skb_frag_size(f), | |
4403 | offset_in_page(offset)); | |
cf80e0e4 HX |
4404 | ahash_request_set_crypt(req, &sg, NULL, skb_frag_size(f)); |
4405 | if (crypto_ahash_update(req)) | |
49a72dfb AL |
4406 | return 1; |
4407 | } | |
4408 | ||
d7fd1b57 ED |
4409 | skb_walk_frags(skb, frag_iter) |
4410 | if (tcp_md5_hash_skb_data(hp, frag_iter, 0)) | |
4411 | return 1; | |
4412 | ||
49a72dfb AL |
4413 | return 0; |
4414 | } | |
49a72dfb AL |
4415 | EXPORT_SYMBOL(tcp_md5_hash_skb_data); |
4416 | ||
cf533ea5 | 4417 | int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key) |
49a72dfb | 4418 | { |
e6ced831 | 4419 | u8 keylen = READ_ONCE(key->keylen); /* paired with WRITE_ONCE() in tcp_md5_do_add */ |
49a72dfb AL |
4420 | struct scatterlist sg; |
4421 | ||
6a2febec ED |
4422 | sg_init_one(&sg, key->key, keylen); |
4423 | ahash_request_set_crypt(hp->md5_req, &sg, NULL, keylen); | |
e6ced831 ED |
4424 | |
4425 | /* We use data_race() because tcp_md5_do_add() might change key->key under us */ | |
4426 | return data_race(crypto_ahash_update(hp->md5_req)); | |
49a72dfb | 4427 | } |
49a72dfb AL |
4428 | EXPORT_SYMBOL(tcp_md5_hash_key); |
4429 | ||
7bbb765b | 4430 | /* Called with rcu_read_lock() */ |
1330b6ef JK |
4431 | enum skb_drop_reason |
4432 | tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, | |
4433 | const void *saddr, const void *daddr, | |
4434 | int family, int dif, int sdif) | |
7bbb765b DS |
4435 | { |
4436 | /* | |
4437 | * This gets called for each TCP segment that arrives | |
4438 | * so we want to be efficient. | |
4439 | * We have 3 drop cases: | |
4440 | * o No MD5 hash and one expected. | |
4441 | * o MD5 hash and we're not expecting one. | |
4442 | * o MD5 hash and its wrong. | |
4443 | */ | |
4444 | const __u8 *hash_location = NULL; | |
4445 | struct tcp_md5sig_key *hash_expected; | |
4446 | const struct tcphdr *th = tcp_hdr(skb); | |
e9d9da91 | 4447 | const struct tcp_sock *tp = tcp_sk(sk); |
7bbb765b DS |
4448 | int genhash, l3index; |
4449 | u8 newhash[16]; | |
4450 | ||
4451 | /* sdif set, means packet ingressed via a device | |
4452 | * in an L3 domain and dif is set to the l3mdev | |
4453 | */ | |
4454 | l3index = sdif ? dif : 0; | |
4455 | ||
4456 | hash_expected = tcp_md5_do_lookup(sk, l3index, saddr, family); | |
4457 | hash_location = tcp_parse_md5sig_option(th); | |
4458 | ||
4459 | /* We've parsed the options - do we have a hash? */ | |
4460 | if (!hash_expected && !hash_location) | |
1330b6ef | 4461 | return SKB_NOT_DROPPED_YET; |
7bbb765b DS |
4462 | |
4463 | if (hash_expected && !hash_location) { | |
7bbb765b | 4464 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND); |
1330b6ef | 4465 | return SKB_DROP_REASON_TCP_MD5NOTFOUND; |
7bbb765b DS |
4466 | } |
4467 | ||
4468 | if (!hash_expected && hash_location) { | |
7bbb765b | 4469 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED); |
1330b6ef | 4470 | return SKB_DROP_REASON_TCP_MD5UNEXPECTED; |
7bbb765b DS |
4471 | } |
4472 | ||
e62d2e11 ED |
4473 | /* Check the signature. |
4474 | * To support dual stack listeners, we need to handle | |
4475 | * IPv4-mapped case. | |
4476 | */ | |
4477 | if (family == AF_INET) | |
4478 | genhash = tcp_v4_md5_hash_skb(newhash, | |
4479 | hash_expected, | |
4480 | NULL, skb); | |
4481 | else | |
4482 | genhash = tp->af_specific->calc_md5_hash(newhash, | |
4483 | hash_expected, | |
4484 | NULL, skb); | |
7bbb765b DS |
4485 | |
4486 | if (genhash || memcmp(hash_location, newhash, 16) != 0) { | |
7bbb765b DS |
4487 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE); |
4488 | if (family == AF_INET) { | |
4489 | net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s L3 index %d\n", | |
4490 | saddr, ntohs(th->source), | |
4491 | daddr, ntohs(th->dest), | |
4492 | genhash ? " tcp_v4_calc_md5_hash failed" | |
4493 | : "", l3index); | |
4494 | } else { | |
4495 | net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u L3 index %d\n", | |
4496 | genhash ? "failed" : "mismatch", | |
4497 | saddr, ntohs(th->source), | |
4498 | daddr, ntohs(th->dest), l3index); | |
4499 | } | |
1330b6ef | 4500 | return SKB_DROP_REASON_TCP_MD5FAILURE; |
7bbb765b | 4501 | } |
1330b6ef | 4502 | return SKB_NOT_DROPPED_YET; |
7bbb765b DS |
4503 | } |
4504 | EXPORT_SYMBOL(tcp_inbound_md5_hash); | |
4505 | ||
cfb6eeb4 YH |
4506 | #endif |
4507 | ||
4ac02bab AK |
4508 | void tcp_done(struct sock *sk) |
4509 | { | |
d983ea6f | 4510 | struct request_sock *req; |
8336886f | 4511 | |
cab209e5 ED |
4512 | /* We might be called with a new socket, after |
4513 | * inet_csk_prepare_forced_close() has been called | |
4514 | * so we can not use lockdep_sock_is_held(sk) | |
4515 | */ | |
4516 | req = rcu_dereference_protected(tcp_sk(sk)->fastopen_rsk, 1); | |
8336886f | 4517 | |
5a5f3a8d | 4518 | if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV) |
c10d9310 | 4519 | TCP_INC_STATS(sock_net(sk), TCP_MIB_ATTEMPTFAILS); |
4ac02bab AK |
4520 | |
4521 | tcp_set_state(sk, TCP_CLOSE); | |
4522 | tcp_clear_xmit_timers(sk); | |
00db4124 | 4523 | if (req) |
8336886f | 4524 | reqsk_fastopen_remove(sk, req, false); |
4ac02bab | 4525 | |
e14cadfd | 4526 | WRITE_ONCE(sk->sk_shutdown, SHUTDOWN_MASK); |
4ac02bab AK |
4527 | |
4528 | if (!sock_flag(sk, SOCK_DEAD)) | |
4529 | sk->sk_state_change(sk); | |
4530 | else | |
4531 | inet_csk_destroy_sock(sk); | |
4532 | } | |
4533 | EXPORT_SYMBOL_GPL(tcp_done); | |
4534 | ||
c1e64e29 LC |
4535 | int tcp_abort(struct sock *sk, int err) |
4536 | { | |
af9784d0 | 4537 | int state = inet_sk_state_load(sk); |
07f6f4a3 | 4538 | |
af9784d0 ED |
4539 | if (state == TCP_NEW_SYN_RECV) { |
4540 | struct request_sock *req = inet_reqsk(sk); | |
4541 | ||
4542 | local_bh_disable(); | |
4543 | inet_csk_reqsk_queue_drop(req->rsk_listener, req); | |
4544 | local_bh_enable(); | |
4545 | return 0; | |
4546 | } | |
4547 | if (state == TCP_TIME_WAIT) { | |
4548 | struct inet_timewait_sock *tw = inet_twsk(sk); | |
4549 | ||
4550 | refcount_inc(&tw->tw_refcnt); | |
4551 | local_bh_disable(); | |
4552 | inet_twsk_deschedule_put(tw); | |
4553 | local_bh_enable(); | |
4554 | return 0; | |
c1e64e29 LC |
4555 | } |
4556 | ||
4ddbcb88 AG |
4557 | /* BPF context ensures sock locking. */ |
4558 | if (!has_current_bpf_ctx()) | |
4559 | /* Don't race with userspace socket closes such as tcp_close. */ | |
4560 | lock_sock(sk); | |
c1e64e29 | 4561 | |
2010b93e LC |
4562 | if (sk->sk_state == TCP_LISTEN) { |
4563 | tcp_set_state(sk, TCP_CLOSE); | |
4564 | inet_csk_listen_stop(sk); | |
4565 | } | |
4566 | ||
c1e64e29 LC |
4567 | /* Don't race with BH socket closes such as inet_csk_listen_stop. */ |
4568 | local_bh_disable(); | |
4569 | bh_lock_sock(sk); | |
4570 | ||
4571 | if (!sock_flag(sk, SOCK_DEAD)) { | |
e13ec3da | 4572 | WRITE_ONCE(sk->sk_err, err); |
c1e64e29 LC |
4573 | /* This barrier is coupled with smp_rmb() in tcp_poll() */ |
4574 | smp_wmb(); | |
e3ae2365 | 4575 | sk_error_report(sk); |
c1e64e29 LC |
4576 | if (tcp_need_reset(sk->sk_state)) |
4577 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
4578 | tcp_done(sk); | |
4579 | } | |
4580 | ||
4581 | bh_unlock_sock(sk); | |
4582 | local_bh_enable(); | |
e05836ac | 4583 | tcp_write_queue_purge(sk); |
4ddbcb88 AG |
4584 | if (!has_current_bpf_ctx()) |
4585 | release_sock(sk); | |
c1e64e29 LC |
4586 | return 0; |
4587 | } | |
4588 | EXPORT_SYMBOL_GPL(tcp_abort); | |
4589 | ||
5f8ef48d | 4590 | extern struct tcp_congestion_ops tcp_reno; |
1da177e4 LT |
4591 | |
4592 | static __initdata unsigned long thash_entries; | |
4593 | static int __init set_thash_entries(char *str) | |
4594 | { | |
413c27d8 EZ |
4595 | ssize_t ret; |
4596 | ||
1da177e4 LT |
4597 | if (!str) |
4598 | return 0; | |
413c27d8 EZ |
4599 | |
4600 | ret = kstrtoul(str, 0, &thash_entries); | |
4601 | if (ret) | |
4602 | return 0; | |
4603 | ||
1da177e4 LT |
4604 | return 1; |
4605 | } | |
4606 | __setup("thash_entries=", set_thash_entries); | |
4607 | ||
47d7a88c | 4608 | static void __init tcp_init_mem(void) |
4acb4190 | 4609 | { |
b66e91cc ED |
4610 | unsigned long limit = nr_free_buffer_pages() / 16; |
4611 | ||
4acb4190 | 4612 | limit = max(limit, 128UL); |
b66e91cc ED |
4613 | sysctl_tcp_mem[0] = limit / 4 * 3; /* 4.68 % */ |
4614 | sysctl_tcp_mem[1] = limit; /* 6.25 % */ | |
4615 | sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2; /* 9.37 % */ | |
4acb4190 GC |
4616 | } |
4617 | ||
1da177e4 LT |
4618 | void __init tcp_init(void) |
4619 | { | |
b49960a0 | 4620 | int max_rshare, max_wshare, cnt; |
b2d3ea4a | 4621 | unsigned long limit; |
074b8517 | 4622 | unsigned int i; |
1da177e4 | 4623 | |
3b4929f6 | 4624 | BUILD_BUG_ON(TCP_MIN_SND_MSS <= MAX_TCP_OPTION_SPACE); |
b2d3ea4a | 4625 | BUILD_BUG_ON(sizeof(struct tcp_skb_cb) > |
c593642c | 4626 | sizeof_field(struct sk_buff, cb)); |
1da177e4 | 4627 | |
908c7f19 | 4628 | percpu_counter_init(&tcp_sockets_allocated, 0, GFP_KERNEL); |
19757ceb ED |
4629 | |
4630 | timer_setup(&tcp_orphan_timer, tcp_orphan_update, TIMER_DEFERRABLE); | |
4631 | mod_timer(&tcp_orphan_timer, jiffies + TCP_ORPHAN_TIMER_PERIOD); | |
4632 | ||
27da6d37 MKL |
4633 | inet_hashinfo2_init(&tcp_hashinfo, "tcp_listen_portaddr_hash", |
4634 | thash_entries, 21, /* one slot per 2 MB*/ | |
4635 | 0, 64 * 1024); | |
6e04e021 ACM |
4636 | tcp_hashinfo.bind_bucket_cachep = |
4637 | kmem_cache_create("tcp_bind_bucket", | |
4638 | sizeof(struct inet_bind_bucket), 0, | |
990c74e3 VA |
4639 | SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
4640 | SLAB_ACCOUNT, | |
4641 | NULL); | |
28044fc1 JK |
4642 | tcp_hashinfo.bind2_bucket_cachep = |
4643 | kmem_cache_create("tcp_bind2_bucket", | |
4644 | sizeof(struct inet_bind2_bucket), 0, | |
4645 | SLAB_HWCACHE_ALIGN | SLAB_PANIC | | |
4646 | SLAB_ACCOUNT, | |
4647 | NULL); | |
1da177e4 | 4648 | |
1da177e4 LT |
4649 | /* Size and allocate the main established and bind bucket |
4650 | * hash tables. | |
4651 | * | |
4652 | * The methodology is similar to that of the buffer cache. | |
4653 | */ | |
6e04e021 | 4654 | tcp_hashinfo.ehash = |
1da177e4 | 4655 | alloc_large_system_hash("TCP established", |
0f7ff927 | 4656 | sizeof(struct inet_ehash_bucket), |
1da177e4 | 4657 | thash_entries, |
fd90b29d | 4658 | 17, /* one slot per 128 KB of memory */ |
9e950efa | 4659 | 0, |
1da177e4 | 4660 | NULL, |
f373b53b | 4661 | &tcp_hashinfo.ehash_mask, |
31fe62b9 | 4662 | 0, |
0ccfe618 | 4663 | thash_entries ? 0 : 512 * 1024); |
05dbc7b5 | 4664 | for (i = 0; i <= tcp_hashinfo.ehash_mask; i++) |
3ab5aee7 | 4665 | INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i); |
05dbc7b5 | 4666 | |
230140cf ED |
4667 | if (inet_ehash_locks_alloc(&tcp_hashinfo)) |
4668 | panic("TCP: failed to alloc ehash_locks"); | |
6e04e021 | 4669 | tcp_hashinfo.bhash = |
593d1ebe | 4670 | alloc_large_system_hash("TCP bind", |
28044fc1 | 4671 | 2 * sizeof(struct inet_bind_hashbucket), |
f373b53b | 4672 | tcp_hashinfo.ehash_mask + 1, |
fd90b29d | 4673 | 17, /* one slot per 128 KB of memory */ |
9e950efa | 4674 | 0, |
6e04e021 | 4675 | &tcp_hashinfo.bhash_size, |
1da177e4 | 4676 | NULL, |
31fe62b9 | 4677 | 0, |
1da177e4 | 4678 | 64 * 1024); |
074b8517 | 4679 | tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size; |
28044fc1 | 4680 | tcp_hashinfo.bhash2 = tcp_hashinfo.bhash + tcp_hashinfo.bhash_size; |
6e04e021 ACM |
4681 | for (i = 0; i < tcp_hashinfo.bhash_size; i++) { |
4682 | spin_lock_init(&tcp_hashinfo.bhash[i].lock); | |
4683 | INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain); | |
28044fc1 JK |
4684 | spin_lock_init(&tcp_hashinfo.bhash2[i].lock); |
4685 | INIT_HLIST_HEAD(&tcp_hashinfo.bhash2[i].chain); | |
1da177e4 LT |
4686 | } |
4687 | ||
d1e5e640 | 4688 | tcp_hashinfo.pernet = false; |
c5ed63d6 ED |
4689 | |
4690 | cnt = tcp_hashinfo.ehash_mask + 1; | |
c5ed63d6 | 4691 | sysctl_tcp_max_orphans = cnt / 2; |
1da177e4 | 4692 | |
a4fe34bf | 4693 | tcp_init_mem(); |
c43b874d | 4694 | /* Set per-socket limits to no more than 1/128 the pressure threshold */ |
5fb84b14 | 4695 | limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7); |
b49960a0 ED |
4696 | max_wshare = min(4UL*1024*1024, limit); |
4697 | max_rshare = min(6UL*1024*1024, limit); | |
7b4f4b5e | 4698 | |
100fdd1f | 4699 | init_net.ipv4.sysctl_tcp_wmem[0] = PAGE_SIZE; |
356d1833 ED |
4700 | init_net.ipv4.sysctl_tcp_wmem[1] = 16*1024; |
4701 | init_net.ipv4.sysctl_tcp_wmem[2] = max(64*1024, max_wshare); | |
7b4f4b5e | 4702 | |
100fdd1f | 4703 | init_net.ipv4.sysctl_tcp_rmem[0] = PAGE_SIZE; |
a337531b YC |
4704 | init_net.ipv4.sysctl_tcp_rmem[1] = 131072; |
4705 | init_net.ipv4.sysctl_tcp_rmem[2] = max(131072, max_rshare); | |
1da177e4 | 4706 | |
afd46503 | 4707 | pr_info("Hash tables configured (established %u bind %u)\n", |
058bd4d2 | 4708 | tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size); |
317a76f9 | 4709 | |
1946e672 | 4710 | tcp_v4_init(); |
51c5d0c4 | 4711 | tcp_metrics_init(); |
55d8694f | 4712 | BUG_ON(tcp_register_congestion_control(&tcp_reno) != 0); |
46d3ceab | 4713 | tcp_tasklet_init(); |
f870fa0b | 4714 | mptcp_init(); |
1da177e4 | 4715 | } |